Datasets:

suolyer

/

pile_pubmed-central

like 1

The content published in Cureus is the result of clinical experience and/or research by independent individuals or organizations. Cureus is not responsible for the scientific accuracy or reliability of data or conclusions published herein. All content published within Cureus is intended only for educational, research and reference purposes. Additionally, articles published within Cureus should not be deemed a suitable substitute for the advice of a qualified health care professional. Do not disregard or avoid professional medical advice due to content published within Cureus. Introduction ============ Total knee arthroplasty (TKA) is a promising treatment for end-stage osteoarthritis (OA) of the knee for alleviating pain and restoring the function of the knee. Some of the cases with bilateral TKA are symptomatic, necessitating revision arthroplasty in both the knees. A bilateral revision TKA can be done either in two stage or simultaneously as a single stage procedure. However, the decision to perform simultaneous bilateral revision TKA is debatable because of possible higher complexity and complication rate. Very few cases have been reported in the literature on this issue. There are various advantages of doing simultaneous bilateral revision TKA compared with staged bilateral revision TKA. These include single operation and single anesthesia as well as better rehabilitation of both knees, apart from a significant reduction in the hospital stay and hospital costs. Case presentation ================= A 67-year-old hypothyroid and hypertensive female presented to us with unstable and painful knees 14 years after primary bilateral TKA for advanced OA. She began developing pain in both the knees for last six months, followed by instability in both knees (right \> left). She was managed symptomatically with painkillers, bracing, and physiotherapy but her pain and instability were not relieved. On clinical examination, the active and passive knee range of motion was painful. The flexion was 0° to 100°, anterior--posterior laxity of 5--10 mm, and a mild valgus laxity. The plain radiographs showed malalignment and loosening of the implants (Figures [1](#FIG1){ref-type="fig"}-[2](#FIG2){ref-type="fig"}). The leucocyte counts, C-reactive protein, and erythrocyte sedimentation rate (ESR) were within normal limits. A three-phase bone scan was also found to be negative for infection. {#FIG1} {#FIG2} Bilateral revision TKAs were performed using modified Insall's midline approach with lateral retraction of the patella (Figure [3](#FIG3){ref-type="fig"}) \[[@REF1]\]. A joint wound swab was taken and sent for gram stain, culture, and sensitivity. It was found to be negative for any microorganisms. The original cemented TKA implants were removed carefully, preserving as much bone as possible. Revision TKA was done on both sides sequentially, under the same anesthesia, using Scorpio® Total Stabilizer (Stryker®, Mahwah, NJ) constrained implants with long femoral and tibial stems. {#FIG3} The knees were protected in hinged braces postoperatively. The drains were removed 48 hours postoperatively; continuous passive motion (CPM) and active knee flexion exercises were started on postoperative day one and gradually increased to 0°--90° of flexion (Figure [4](#FIG4){ref-type="fig"}). {#FIG4} The postoperative radiographs showed satisfactory implant positions (Figures [5](#FIG5){ref-type="fig"}-[6](#FIG6){ref-type="fig"}). The patient had no complaints and was able to flex the knee to 80° easily. The range of motion and quadriceps strengthening exercises continued without forced flexion. She gradually resumed full weight-bearing with the help of the walker. Three months after surgery, the brace was removed, and active pain-free range of motion of 0°--115° was achieved with complete stability. At four months, the patient had returned to full activity without the brace or cane. At the final follow-up of four years, the knee was fully stable, and the patient was pain-free with no loosening or wear of the implants. {#FIG5} {#FIG6} Discussion ========== Symptomatic instability and pain following primary TKA requires revision surgery. In one retrospective study of 49 TKA patients with bilateral simultaneous revision, no postoperative cardiovascular complications, stroke, or death were noted \[[@REF2]\]. The minor reported complications included transient, self-limited confusion (in three cases); pulmonary embolism (in one patient), which was treated successfully with an inferior vena cava filter and extended anticoagulation; posterior compartment syndrome (in one case), which was treated by fasciotomy; and stiff knee in one patient (that was manipulated under anesthesia at three months). In a retrospective cohort study, Carter, et al. \[[@REF3]\] found that 33 of 141 morbidly obese patients (23.4%) who had revision TKA had a complication compared to 10 of 96 patients with a BMI 18.5 - 25 (10.4%) (p = 0.011). The most common complication was wound healing. Kevin, et al. reviewed 60,355 revision TKA procedures done in the USA and noted that the most common causes of revision TKA were an infection in 25.2%, implant loosening in 16.1%, and implant failure/breakage in 9.7% cases \[[@REF4]\]. They found that revision of all the components was the most common type of procedure done (35.2%). Singh, et al. found a high prevalence (46.5%) of overall moderate to severe activity limitation at two years and 50.5% at five years following revision TKA \[[@REF5]\]. Significantly higher odds of moderate to severe overall activity limitation was noted both at two and five-year follow-ups in patients with a BMI of 40 or higher, age greater than 80 years, higher Deyo-Charlson score, and in females. Kasmire, et al. studied predictors of functional outcome after revision TKA by using various parameters, such as short-form 36 (SF-36), Western Ontario and McMaster Osteoarthritis Index (WOMAC), and Knee Society Scores (KSS) \[[@REF6]\]. The data was collected preoperatively and at two years follow-up in their 175 revision TKAs done for aseptic failure. All of the above-mentioned parameters improved significantly after revision TKA (p \< 0.001). Lower preoperative pain and higher clinical KSS were found to be predictors of a better outcome. Sheth, et al. found that the complication rates were different for bilateral TKA done simultaneously and as staged procedures \[[@REF7]\]. These authors reported aseptic revision (1.17% vs. 0.9%), septic revision (0.8% vs. 0.7%), mortality (0.28% vs. 0.1%), and adverse events (2.49% vs. 1.97%). According to Bohm, et al., simultaneous bilateral primary TKA patients required more blood transfusions, a shorter hospital stay, more transfers to a rehabilitation facility, and less frequency of knee infections than staged bilateral TKA patients \[[@REF8]\]. However, these patients had a higher rate of cardiac complications and in-hospital mortality rate. The three-year revision, however, was same in both the groups. In a meta-analysis of 14 studies, Hu, et al. showed that the prevalence of mortality immediately postoperatively, mortality at 30 days postoperatively, and neurological complications were significantly higher in simultaneous TKA compared to staged TKA patients \[[@REF9]\]. The prevalence of thromboembolic disease, infection, and cardiac complications were not significantly different between simultaneous TKA compared to staged TKA patients. According to Hersekli, et al., the amount of blood loss, intensive care unit days and perioperative complications were same between single- and two-staged operations (p \> 0.05) \[[@REF10]\]. However, hospital stay and overall cost were significantly less in single-staged operations. We faced the challenge in decision-making regarding the staging of the procedures in this reported case, where revision of the components was necessary for both knees. We could not find proper guidelines regarding bilateral revision TKA as there are only a few documented reports of simultaneous bilateral revision TKA. There is limited evidence to support the one-stage practice of doing bilateral revision TKAs, as its safety remains controversial. We chose to do a single-staged bilateral revision TKA in this case, as a two-staged procedure would have required two anesthesias, longer hospital stay, more hospital bills, and surgery-related complications, which were overcome by a single-staged procedure in this case. With the use of constrained implants and long stems of the prosthetic components, we achieved good knee stability and satisfactory range of motion immediately postoperatively and at the four year follow-up. Conclusions =========== Two-staged bilateral revision total knee replacement (TKA) has many disadvantages, such as requiring anesthesia to be given twice, a longer hospital stay, more hospital bills, and higher surgery-related complications, which can be overcome by a single stage procedure. In carefully selected patients, single-staged bilateral revision TKAs should be considered over two-staged procedures. The authors have declared that no competing interests exist. Consent was obtained by all participants in this study

J Med Radiat Sci 65 (2018) 275--281 Introduction {#jmrs290-sec-0005} ============ There is a growing interest globally in making sure that graduates emerge from higher education with the capabilities and competencies that will equip them not only to be 'work ready' on graduation but also prepared for the development of technology, new models of service delivery and advances for practice in the future.[1](#jmrs290-bib-0001){ref-type="ref"}, [2](#jmrs290-bib-0002){ref-type="ref"}, [3](#jmrs290-bib-0003){ref-type="ref"} In a profession, such as medical imaging, the health workforce needs graduates who are ready to understand and apply emerging technology alongside meeting the demands of ever changing healthcare systems.[4](#jmrs290-bib-0004){ref-type="ref"} This paper reports on the outcomes of a survey undertaken as part of preparation for the review and redesign of clinical placements in a medical imaging programme in New Zealand. The project embraced the goal of defining work ready plus graduates for the medical imaging workforce. Identification of the capabilities required of a medical imaging technologist (MIT) in their graduate years was critical for the development of the clinical experience programme, as it is clinical placement and emersion in work that is most likely to develop capability and work readiness skills in graduates. It was envisaged that by defining, for our regional context, the capabilities and work skills employers seek in our graduates, we would have the data we needed to review and if necessary rewrite the graduate profile and utilise fully and effectively the real‐life clinical experiences that support the development of these capabilities. The results are also impacting positively on lecturers teaching methods as they consider how they can develop these capabilities in students through teaching, learning, and assessment methodologies. The theoretical underpinning for this study was Scott\'s fellowship work for the Australian Teaching and Learning Council and the professional and graduate capability framework published for the Australian tertiary environment.[1](#jmrs290-bib-0001){ref-type="ref"} The Professional Capability Framework as used by Western Sydney University was used as the foundation for the development of a survey tool as it was current and had been validated in a range of disciplines that included health professions. In addition, it looks beyond graduation and standards for practice (as required by the New Zealand Medical Radiation Technologist Registration Board and the Medical Radiation Practice Board of Australia towards the generic skills graduates need to flourish in a profession in the future.[1](#jmrs290-bib-0001){ref-type="ref"}, [5](#jmrs290-bib-0005){ref-type="ref"} Hence the term work ready plus. Using a validated and comprehensive professional and graduate capability framework ensured that all potentially relevant capability options had been considered. It was deemed generalisable to the New Zealand health care environment due to the similarities between both the health and education systems. Figure [1](#jmrs290-fig-0001){ref-type="fig"} summarises the key elements of the professional capability framework. The overlapping aspects of professional capability are identified -- personal, interpersonal and cognitive which have been validated in a range of investigations, mainly focused on professional leadership.[1](#jmrs290-bib-0001){ref-type="ref"}, [5](#jmrs290-bib-0005){ref-type="ref"} These domains are underpinned by relevant role‐specific and generic competencies (the skills and knowledge found to be essential to the specific role of an MIT). The key terms "competence" and "capability" are problematic and therefore often confused. We adopted the definition that competence is the possession of the skills and knowledge necessary to perform the duties set down for a specific role. The New Zealand Medical Radiation Technologist Registration Board (MRTB) reviewed and updated their competencies for New Zealand registration in March 2017, so a list of competencies was current and available. We have adopted a definition of capability that goes beyond the skills to practice as a safe and competent practitioner, to embrace the concept of being work ready plus. Being "work ready *plus"* requires capabilities for not just today, for current practice but for the future. Capabilities include the ability to work with others from a range of professions and backgrounds, manage the unexpected, adopt new technology, to be changed implementation savvy, inventive, sustainability responsive, to learn from experience and to operate with a clear understanding of one\'s ethical position.[1](#jmrs290-bib-0001){ref-type="ref"} {ref-type="ref"} Permission was obtained to reproduce this figure.](JMRS-65-275-g001){#jmrs290-fig-0001} These capabilities require a mixture of emotional and cognitive intelligence, including the ability to determine when and when not to deploy these competences.[1](#jmrs290-bib-0001){ref-type="ref"} We believed this concept was less developed for the medical imaging profession in New Zealand. The Professional Capability Framework developed through a scholarship awarded by the Australian Teaching and Learning Council formed the basis for the development of a survey that asked practicing MITs and MIT clinical managers at the three largest placements sites in New Zealand to rate the capabilities deemed critical in a graduate to ensure they are "work ready *plus"*.[1](#jmrs290-bib-0001){ref-type="ref"} The items used in the survey fall into three domains which align with the capability domains identified in Figure [1](#jmrs290-fig-0001){ref-type="fig"}. These domains are discussed in more detail in Scott, Coates and Anderson[5](#jmrs290-bib-0005){ref-type="ref"} and Fullan and Scott.[6](#jmrs290-bib-0006){ref-type="ref"} This paper shares the results from the survey and discusses the impact these are having on curriculum review and development. Method {#jmrs290-sec-0006} ====== This study was carried within all the public (three District Health Board, which includes 2 hospitals on the Northshore, 3 Inner City and 1 in South Auckland), Radiology Services, in the Auckland Region, where Unitec Institute of Technology\'s MIT students are placed for clinical experience during their 3‐year training programme. Data collection period, April and August 2017. A prospective survey was selected as the method of data collection tool as it allowed us to collect anonymous responses from stakeholders with minimal disruption to the work environment. The survey was distributed electronically. The SurveyMonkey online tool was used to develop a rating scale questionnaire, using the statements and domains from the Australian Capability Framework.[1](#jmrs290-bib-0001){ref-type="ref"} The survey was trialled by three clinicians and during this process one question was removed that was perceived repetitive. The final questionnaire had 39 capability statements that were clustered into three domains: personal, interpersonal and cognitive. The first question of the questionnaire requested participants consent before proceeding with the survey. An open survey link was sent to MIT clinical managers for internal circulation. A participant information sheet was attached to the email invitation email. Participants were assured of the anonymity of their responses and this was achieved by using the anonymity function on SurveyMonkey. Ethics approval was granted by the Unitec Research Ethics Committee (UREC) -- No 2017--1002. Analysis design {#jmrs290-sec-0007} --------------- For the demographic variables of the survey, the data were represented either in the form of tables or graphs. Owing to the subjective nature of the data related to capabilities that participants were requested to provide in ordinal form (ranking), the average ranking measure was considered most appropriate to statistically determine which answer choice was most preferred overall. The answer choice with the largest average ranking is the most preferred choice. The calculations were conducted using Microsoft Excel. The questionnaire was organised with a total of 39 statements which were grouped into the three domain categories: personal capabilities included 15 statements, interpersonal capabilities had 11 statements and cognitive capabilities had 13 statements. Thus, the ranking for personal capabilities was from 1 to 15, interpersonal from 1 to 11 and cognitive from 1 to 13. The average ranking was calculated as follows:$${{Average}\mspace{720mu}{Ranking}} = \frac{x_{1}w_{1} + x_{2}w_{2} + \ldots + x_{n}w_{n}}{Total},$$where *w* represented the weight of ranked position and *x* represented the response count for the answer choice. Weights are applied in reverse order. The respondent\'s most preferred choice, which is ranked 1, has the largest weight and their least preferred choice has a weight of 1. In our case, the personal capabilities had 15 statements. The highest ranked statement had a weight of 15, second highest had 14, third highest had 13 and so on with the last ranked statement having a weight of 1. Similar weights, depending on the number of statements, were applied to the interpersonal and cognitive capabilities. Results {#jmrs290-sec-0008} ======= A total of 52 responses were received from a maximum sample size of 265. This indicates a response rate of 19.6%. However, it is not possible to exactly predict the size of the actual sample pool, as the surveys were distributed via the clinical managers to their staff. From the responses, 90% (47) of the respondents were female and the remaining 10% (5) were males. In terms of the position/title of the respondents, the majority of the respondents (76%) were senior qualified MITs and 15% team leader/clinical specialist. 74% (39) had over 6 years experience. Average ranking reported by domain {#jmrs290-sec-0009} ---------------------------------- Table [1](#jmrs290-tbl-0001){ref-type="table"} shows the average rankings for the domain personal capabilities with the top five clearly visible. ###### Average ranking scores for personal capabilities Statements -- personal capabilities Average ranking score ------------------------------------------------------------------------- ----------------------- Being willing to face and learn from errors 11.56 Wanting to do as good a job as possible 11.21 Understanding personal strengths and limitations 11.12 Remaining calm under pressure or when things take an unexpected turn 10.98 Having energy, passion and enthusiasm for the profession and role 10.94 Willingness to persevere when things are not working out as anticipated 8.35 Pitching in and undertaking menial tasks as required 7.44 Being true to one\'s personal values and ethics 7.27 Deferring judgment and not jumping in too quickly to resolve a problem 7.1 Maintaining a good work/life balance and keeping things in perspective 6.61 Being willing to take a hard decision 6.52 Bouncing back from adversity 6.37 Being confident to take calculated risks 5.71 Being willing to take responsibility for projects and how they turn out 5.62 Tolerating ambiguity and uncertainty 4.69 John Wiley & Sons, Ltd Table [2](#jmrs290-tbl-0002){ref-type="table"} shows the average rankings for the domain interpersonal capabilities. These rankings have a flatter profile with eight capabilities ranking higher than 5. ###### Average ranking scores for interpersonal capabilities Statements -- interpersonal capabilities Average ranking score ------------------------------------------------------------------------------------------------------------- ----------------------- Being transparent and honest in dealings with others 7.64 Empathising and working productively with people from a wide range of backgrounds 7.57 Listening to different points of view before coming to a decision 7.13 Understanding how the different groups that make up a work place operate and influence different situations 7.02 Giving and receiving constructive feedback to/from work colleagues and others 6.73 Being able to develop and contribute positively to team‐based programs 6.1 Being able to work with senior staff within and beyond the organisation without being intimidated 5.87 Motivating others to achieve positive outcomes 5.85 Being able to develop and use networks of colleagues to solve key workplace problems 4.77 Influencing people\'s behaviour and decisions in effective ways 3.96 Working constructively with people who are 'resistors' or are over‐enthusiastic 3.58 John Wiley & Sons, Ltd Table [3](#jmrs290-tbl-0003){ref-type="table"} shows the average rankings for the domain cognitive capabilities. All cognitive capabilities achieved an average ranking score of more than 5. ###### Average ranking scores for cognitive capabilities Statements -- cognitive capabilities Average ranking score ------------------------------------------------------------------------------------------------------------ ----------------------- Diagnosing the underlying causes of a problem and taking appropriate action to address it 9.2 Making sense of and learning from experience 9.19 Being able to identify the core issue from a mass of detail in any situation 8.04 Using previous experience to figure out what\'s going on when a current situation takes an unexpected turn 7.96 Having a clear, justified and achievable direction in area of responsibility 7.85 Thinking creatively and laterally 6.92 Seeing the best way to respond to a perplexing situation 6.72 Setting and justifying priorities for daily work 6.71 Adjusting a plan of action in response to problems that are identified during its implementation 6.67 Recognising patterns in a complex situation 5.94 Seeing and then acting on an opportunity for a new direction 5.88 Recognising how seemingly unconnected activities are linked 5.23 Tracing out and assessing the likely consequences of alternative courses of action 5.15 John Wiley & Sons, Ltd Discussion {#jmrs290-sec-0010} ========== Personal domain capabilities {#jmrs290-sec-0011} ---------------------------- The results from the personal domain show a clear top five capability rated highly by the respondents. Namely: remaining calm under pressure or when things take an unexpected turn; understanding personal strengths and limitations; being willing to face and learn from errors; wanting to do as good a job as possible; having energy, passion and enthusiasm for the profession and role. These five capabilities had strong face validity when presented to a meeting of national managers. They certainly provide a clear mandate as to which personal qualities should be incorporated into the graduate profile. The challenge in curriculum design will be to find ways to highlight, reinforce and role model these capabilities. The literature provides limited guidance, however Fraser & Greenhalgh[3](#jmrs290-bib-0003){ref-type="ref"} suggest that capability can be strengthened by the use of feedback, self‐reflection, and consolidation, with students following a nonlinear education model. Therefore, the incorporation of directed educator and supervisor feedback could assist in the recognition and development of this capability. To further consolidate these skills, it would be advantageous to encourage students to observe these capabilities in others and reflect as to how their own developing practice incorporates and builds this capability. Interpersonal domain capabilities {#jmrs290-sec-0012} --------------------------------- We have noted that in the interpersonal domain, the six top ranked qualities (all with a ranking above 6) have an alignment to the competencies identified for those working in interprofessional teams. Ponzer et al.[7](#jmrs290-bib-0007){ref-type="ref"} published the five core competencies that form the basis of many interprofessional education activities which have been modified for specific contexts and are frequently used to describe strong interprofessional teams.[8](#jmrs290-bib-0008){ref-type="ref"} Table [4](#jmrs290-tbl-0004){ref-type="table"} compares these. ###### Frequently used statements of attributes of effective interprofessional teams compared to the top six interprofessional capabilities in this study ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Interpersonal capabilities compared to interprofessional team attributes -------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------- Understanding how the different groups that make up a work place operate and influence different situations\ Mutual understanding of roles and recognition of difference\ Empathising and working productively with people from a wide range of backgrounds\ Good patient‐ care/co‐operation\ Being able to develop and contribute positively to team‐based programs\ Mutual trust and respect\ Giving and receiving constructive feedback to/from work colleagues and others\ The importance of good communication for teamwork\ Listening to different points of view before coming to a decision\ Assertiveness needed for effective conflict management\ Being transparent and honest in dealings Be aware of ethical issues ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- John Wiley & Sons, Ltd Following the World Health Organisation report in 2010[9](#jmrs290-bib-0009){ref-type="ref"} there is a growing international commitment to the promotion of collaborative practice in healthcare delivery supported by interprofessional education to ensure graduates have the capabilities required for collaborative roles on graduation. In 2012, the national boards and the Australian Health Ministers' Advisory Council, conducted an independent Review of the National Registration and Accreditation Scheme for health professionals.[10](#jmrs290-bib-0010){ref-type="ref"} As part of this review, the Health Practitioner Regulation Agency (AHPRA) reviewed the performance of each of these accreditation authorities against the domains of the Quality Framework for the Accreditation Function to inform the decisions on how to continue to implement the accreditation function under the National Law. Following this review process, all the current profession‐specific accreditation authorities were asked to consider opportunities to increase cross‐profession collaboration and innovation and support interprofessional learning.[10](#jmrs290-bib-0010){ref-type="ref"} There is a close working relationship between New Zealand and Australian registration and accrediting bodies and considerable influence in both directions. The growth of the Australian interprofessional agenda is likely to have a growing impact on New Zealand health professional registration and accreditation requirements. The capabilities for team work and collaborative practice recognised by practicing MITs in New Zealand appears to support the growing international agenda supporting collaborative practice models of care. There has been some substantial work in the interprofessional education and team working space around both learning and assessment methods that we can use to guide our curriculum planning.[11](#jmrs290-bib-0011){ref-type="ref"}, [12](#jmrs290-bib-0012){ref-type="ref"}, [13](#jmrs290-bib-0013){ref-type="ref"} Simulation with other professionals, interprofessional activities within the academic curriculum and opportunities to observe and engage with interprofessional teams while on placement (evidenced in a clinical portfolio) align to these capabilities. Cognitive domain capabilities {#jmrs290-sec-0013} ----------------------------- In the cognitive abilities domain diagnosing underlying causes of a problem, taking appropriate action and making sense of learning from experience are the most highly rated, followed by being able to identify the core issue from a mass of detail in any situation and using previous experience to figure out what\'s going on when a current situation takes an unexpected turn, are capabilities that aid problem solving. Overall the profile of preference in this domain is relatively flat. We note alignment to the concepts of clinical reasoning and critical thinking as it is described in the health professions. In the literature, the terms clinical reasoning, clinical judgment, problem‐solving, decision‐making and critical thinking are often used interchangeably. The term clinical reasoning is used to describe the process by which clinicians collect cues, process the information, come to an understanding of a patient problem or situation, plan and implement interventions, evaluate outcomes, and reflect on and learn from the process.[14](#jmrs290-bib-0014){ref-type="ref"}, [15](#jmrs290-bib-0015){ref-type="ref"}, [16](#jmrs290-bib-0016){ref-type="ref"} The clinical reasoning process is also described as dependent upon a critical thinking "disposition".[17](#jmrs290-bib-0017){ref-type="ref"} The American Philosophical Association defined critical thinking as purposeful, self‐regulatory judgment that uses cognitive tools such as interpretation, analysis, evaluation, inference, and explanation of the evidential, conceptual, methodological, criteriological or contextual considerations on which judgment is based.[18](#jmrs290-bib-0018){ref-type="ref"} We have noted that some students have a problem understanding how these capabilities are demonstrated in the work place. Responses in this domain are assisting us to define what capabilities are associated with critical thinking and clinical reasoning in the medical imaging profession and how they are evidenced in clinical practice. We are turning our attention to building processes to support the development of these capabilities within our class‐based learning, simulated learning, and clinical supervision. We will also incorporate post‐practicum experiences that will encourage students to appraise their experiences, seek clarification and comparisons and link their learning to the future, including securing employment.[19](#jmrs290-bib-0019){ref-type="ref"} The goal is to develop the student\'s ability to make judgements and decisions about their work experiences and learning that will position them as future critical thinkers, life longer enquirers and learners. Conclusion {#jmrs290-sec-0014} ========== Identification of the core capabilities that our stakeholder community rate highly has proved informative in assisting us to describe a "work ready *plus"* medical imaging graduate for the New Zealand context. The results have provided data to the curriculum development team allowing them to align the graduate profile to these expectations and raised awareness among academic staff of the need to include these capabilities in the curriculum. In addition, it has enabled a dialog with stakeholders about capability in the profession, refreshing and revising the involvement of the professional community in the academic programme. Scott reminds us that capability cannot be taught, people cannot be trained in it; but it can be learnt through exposure to educational experiences which entail coming to grips with real world dilemmas. Clinical placements provide this learning experience; it is here students learn what others do when the unexpected happens and develop the skills to make sense of what is unfolding to successfully resolve the situation. This naturally occurring curriculum of the workplace is often tacit and therefore not clearly visible to learners and students needs support.[19](#jmrs290-bib-0019){ref-type="ref"} These results provide a blue print for conceptualising the key opportunities a clinical placement offers beyond learning technical skills and competencies; highlighting the capabilities that can be learnt and developed on placement, bringing these learning opportunities to the attention of students and clinical supervisors alike and bringing a new clarity to the design of support for learning on placement. We now have descriptors of capability that will allow us to be more specific in our communication of the capabilities our graduates should aspire to (beyond but building on those established by the regulatory body) and we are incorporating these into the curriculum design process for both teaching and assessment purposes. They will inform clinical supervision and clinical learning, allowing clinical supervisors to focus on highlighting experiences that can develop these capabilities. This study is informing curriculum planning and energising discussions around the design of simulation, class room teaching activities and clinical placements designed to develop these capabilities. Conflict of Interest {#jmrs290-sec-0016} ==================== The authors declare no conflict of interest. We acknowledge the help and support of the Radiology Services and Research Office staff at Auckland and Waitemata District Health Boards, Auckland New Zealand. This project was kindly supported by the Unitec Strategic Fund, Unitec Research Office.

1. Introduction {#sec1-jcm-08-02039} =============== Venous thromboembolism (VTE) is a common cause of morbidity and mortality in hospitalized and non-hospitalized patients \[[@B1-jcm-08-02039]\]. The American Society of Hematology and American College of Chest Physicians guidelines recommend low-molecular-weight heparin (LMWH) as a first-line pharmacological option for most patients at risk of VTE \[[@B2-jcm-08-02039],[@B3-jcm-08-02039]\]. Several prophylactic doses and types of LMWH are used worldwide, which is reflected by differences in national summaries of product characteristics (SPCs) and dosing regimens of randomized controlled trials (RCTs). There is no high-quality evidence or guidance on the optimal prophylactic LMWH dose. Preceding systematic reviews on thrombosis prophylaxis have not specifically assessed benefits and harms associated with different LMWH doses \[[@B4-jcm-08-02039],[@B5-jcm-08-02039],[@B6-jcm-08-02039],[@B7-jcm-08-02039],[@B8-jcm-08-02039],[@B9-jcm-08-02039],[@B10-jcm-08-02039]\]. In addition, there have been very few direct comparisons of prophylactic LMWH dose regimens, and therefore indirect evidence could provide a 'second best' estimate of benefits and harms. There is no generally accepted definition of different prophylactic LMWH dose categories, which is why we previously categorized LMWH thrombosis prophylaxis regimens as either 'low-dose' or 'intermediate-dose', based on different registered doses in SPCs worldwide \[[@B11-jcm-08-02039]\]. Using this approach in a previous meta-analysis, we found that intermediate-dose LMWH, compared with placebo or no treatment, was associated with a significant decrease in symptomatic VTE, at the cost of an increase in major bleeding \[[@B11-jcm-08-02039]\]. The main objective of the current study was to perform a systematic review with meta-analysis and trial sequential analysis (TSA) comparing benefits and harms of low-dose LMWH versus placebo or no treatment for thrombosis prophylaxis in all types of patients at risk of VTE \[[@B12-jcm-08-02039]\]. 2. Materials and Methods {#sec2-jcm-08-02039} ======================== We conducted this systematic review according to a pre-published protocol on PROSPERO (<https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42019124722>) following the methodology suggested by Jakobsen et al, the Cochrane Handbook for Systematic Reviews of Interventions, the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement, and the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) recommendations \[[@B12-jcm-08-02039],[@B13-jcm-08-02039],[@B14-jcm-08-02039],[@B15-jcm-08-02039]\]. 2.1. Study Selection {#sec2dot1-jcm-08-02039} -------------------- ### 2.1.1. Patients {#sec2dot1dot1-jcm-08-02039} Studies were considered for inclusion irrespective of language, blinding, publication status, or sample size. We included RCTs with adult patients allocated to receive thrombosis prophylaxis using either low-dose LMWH, placebo, or no treatment, regardless of their underlying disease or whether they were admitted to the hospital or visited the outpatient clinic. ### 2.1.2. Interventions {#sec2dot1dot2-jcm-08-02039} The experimental intervention was low-dose LMWH, irrespective of LMWH type or duration of treatment. We a priori defined 'low dose' in our protocol according to the SPCs as approved by the US Food and Drug Administration, the European Medicines Agency, and several national authorities ([Table 1](#jcm-08-02039-t001){ref-type="table"}). If different LMWHs or (weight-adjusted) doses were used in one trial, we classified the dose according to what was used most frequently. We included trials evaluating ultra-low-molecular-weight heparins and LMWHs not listed in [Table 1](#jcm-08-02039-t001){ref-type="table"} (e.g., LMWHs we were unable to classify into a specific dose) in a sensitivity analysis. The control intervention was placebo or no treatment. Co-interventions such as mechanical compression devices were allowed if they were applied in both treatment groups. ### 2.1.3. Outcomes {#sec2dot1dot3-jcm-08-02039} Predefined co-primary outcomes were all-cause mortality, symptomatic VTE, and major bleeding. Secondary outcomes were serious adverse events (SAE), clinically relevant non-major bleeding, and any VTE (including both symptomatic and asymptomatic events). All outcomes were assessed at maximum follow-up. VTE was defined as deep vein thrombosis or pulmonary embolism, and the diagnosis was accepted when objectified by an imaging technique or autopsy. We made no distinction between distal or proximal, or lower versus upper extremity thrombosis. Major bleeding and clinically relevant non-major bleeding were defined according to trial criteria. SAE were defined according to the International Conference on Harmonisation of Good Clinical Practice definitions (ICH-GCP) \[[@B16-jcm-08-02039]\]. 2.2. Data Sources and Searches {#sec2dot2-jcm-08-02039} ------------------------------ We searched the Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library, PubMed/MEDLINE, EMBASE and Web of Science ([Table S1](#app1-jcm-08-02039){ref-type="app"}). References of identified studies were screened to identify further relevant trials. Finally, we searched the World Health Organization's International Clinical Trials Registry and ClinicalTrials.gov for ongoing trials ([Table S2](#app1-jcm-08-02039){ref-type="app"}). The search was last updated on 10 June 2019. 2.3. Data Extraction and Quality Assessment {#sec2dot3-jcm-08-02039} ------------------------------------------- Two authors (RJE, WB) independently identified trials for inclusion. Trials excluded on the basis of full text were listed with reasons for exclusion. We extracted information on characteristics (year of publication, country, numbers of sites and patients enrolled), participants (age, sex, eligibility criteria), interventions (type, dose, and duration of LMWH treatment), and outcomes. We resolved differences in opinion through discussion. Two authors (R.J.E., W.B.) independently assessed risks of bias of the included trials according to the revised Cochrane risk of bias tool version 2 \[[@B17-jcm-08-02039]\] in the following five domains: "Bias arising from the randomization process'', "Bias due to deviations from intended interventions'', "Bias due to missing outcome data'', "Bias in measurement of the outcome'', "Bias in selection of the reported result''. RCTs were classified as 'overall low risk of bias' when all bias domains were judged as 'low risk'. Conversely, trials were classified as 'overall high risk of bias' when 'some concerns' or 'high risk' was judged in one or more domains \[[@B18-jcm-08-02039]\]. Publication bias was assessed by inspecting funnel plots for signs of asymmetry when 10 or more trials were included in the analyses \[[@B12-jcm-08-02039],[@B14-jcm-08-02039]\]. 2.4. Data Synthesis and Analysis {#sec2dot4-jcm-08-02039} -------------------------------- We calculated relative risk (RR) with both conventional 95% confidence intervals (CIs) and TSA-adjusted CI if there were two or more trials for each outcome. ### 2.4.1. Assessment of Significance {#sec2dot4dot1-jcm-08-02039} We used adjusted thresholds for statistical significance to correct for multiplicity issues due to repeated testing. An alpha of 0.025 was used for the co-primary and secondary outcomes to keep the family-wise error rate at a maximum of 5% \[[@B14-jcm-08-02039]\]. In case of statistically significant RR, we calculated numbers needed to treat (NNT) or numbers needed to harm (NNH) with 97.5% CI. ### 2.4.2. Meta-Analysis {#sec2dot4dot2-jcm-08-02039} Data were pooled using both a fixed-effect and a random-effects model. In case of discrepancy between the models, we emphasized the most conservative estimate. Analyses were performed on an intention-to-treat basis whenever possible or otherwise using an 'available-case analysis'. ### 2.4.3. Trial Sequential Analysis {#sec2dot4dot3-jcm-08-02039} Conventional meta-analyses may result in type-I errors due to risks of random error when few data have been collected or due to repeated significance testing when a meta-analysis is updated with new trials \[[@B19-jcm-08-02039],[@B20-jcm-08-02039],[@B21-jcm-08-02039],[@B22-jcm-08-02039],[@B23-jcm-08-02039]\]. TSA is a sequential meta-analysis method that combines required information size estimation (i.e., the number of patients needed to detect an a priori specified relative risk reduction) with an adjusted threshold for statistical significance \[[@B21-jcm-08-02039],[@B22-jcm-08-02039]\]. This adjusted threshold is more conservative when data are sparse and becomes progressively more lenient as the accumulated sample size approaches the required information size. Accordingly, the TSA-adjusted CI is initially wider than the conventional 95% CI, but when the required information size has been reached, they become identical. The required information size is calculated on the basis of the unweighted event proportion in the control group, the assumption of a plausible relative risk reduction/increase (RRR/RRI), and the anticipated heterogeneity variance (D^2^) of the meta-analysis. We applied TSA to all outcomes, using the control event proportion from the actual meta-analyses; D^2^ as suggested by the meta-analysis; alpha of 2.5%; beta of 90%; and an anticipated RRR/RRI of 20%. ### 2.4.4. Assessment of Heterogeneity {#sec2dot4dot4-jcm-08-02039} Statistical heterogeneity I^2^ was explored by the chi-squared test with significance set at a *p*-value of 0.10. The quantity of heterogeneity was also measured by D^2^ \[[@B24-jcm-08-02039]\]. Clinical heterogeneity was explored by conducting explorative subgroup analyses. ### 2.4.5. Subgroup Analysis {#sec2dot4dot5-jcm-08-02039} We performed subgroup analyses according to overall risk of bias (low vs. high), type of patients, LMWH type, duration of the intervention (less vs. more than 30 days), and length of follow-up (less vs. more than 30 days). Statistically significant subgroup differences (test of interaction *p* \< 0.05) provided evidence of an intervention effect pending the subgroup. ### 2.4.6. Sensitivity Analysis {#sec2dot4dot6-jcm-08-02039} All analyses were re-conducted including trials that evaluated LMWH types not covered by [Table 1](#jcm-08-02039-t001){ref-type="table"}. In addition, sensitivity TSAs were conducted using an RRR as suggested by the overall low-risk-of-bias studies and using a D^2^ of 25% if the actual D^2^ was 0%. In case of rare events (\<2% in the control group), TSA was also performed using Peto's odds ratio. SAE are often inconsistently reported and, in addition to assessing SAE according to trial reporting, we estimated the number of patients with one or more SAE using two methods: (1) the highest proportion of either reported mortality, symptomatic VTE, or major bleeding in each trial and (2) all mortality, SAE, symptomatic VTE, and major bleeding events cumulated in each trial. The idea is that the 'true proportion' of SAE should lie between these two extremes. Finally, to assess the impact of attrition bias on the primary outcomes, we imputed missing outcome data in best-/worst-case and worst-/best-case scenarios \[[@B14-jcm-08-02039]\]. 2.5. GRADE {#sec2dot5-jcm-08-02039} ---------- We used GRADE to assess the quality of the body of evidence associated with each outcome \[[@B15-jcm-08-02039]\]. 3. Results {#sec3-jcm-08-02039} ========== Our search strategy identified 10,374 records. After removal of duplicates and selections based on titles and abstracts, 312 records remained. A total of 271 reports were excluded on the basis of full text, and 41 records reporting 44 RCTs with a total of 22,579 patients were included ([Figure 1](#jcm-08-02039-f001){ref-type="fig"}). 3.1. Characteristics of the Included Studies {#sec3dot1-jcm-08-02039} -------------------------------------------- Detailed characteristics of the 44 included trials are presented in [Table S3](#app1-jcm-08-02039){ref-type="app"}. The year of publication ranged from 1988 to 2018. Forty trials were in English, two in German, one in French, and one in Chinese. Three trials were published as abstracts only, and the Chinese trial was assessed as abstract only due to lacking translation capacity. There were 24 single-center and 20 multicenter trials. Nine different types of LMWH preparations were used, and several types of patients were evaluated: orthopedic or immobilized patients (16 trials), surgical patients (13 trials), ambulatory cancer patients (8 trials), acutely ill medical patients (4 trials), and neurological patients (3 trials). 3.2. Bias Risk Assessment {#sec3dot2-jcm-08-02039} ------------------------- Six trials including 8172 patients were considered at overall low risk of bias ([Table S4](#app1-jcm-08-02039){ref-type="app"}). Thirty-eight trials were classified as overall high risk of bias. We did not suspect publication bias except for the outcome any VTE, in which asymmetry in the funnel plot was observed ([Figures S1--S4](#app1-jcm-08-02039){ref-type="app"}). Sensitivity analyses of imputed missing data suggested potential for attrition bias in all primary outcomes, since the imputed effect estimates in the best/worse and worse/best scenario's suggested benefit and harm, respectively ([Table S5](#app1-jcm-08-02039){ref-type="app"}). Results of a post-hoc sensitivity analysis excluding trials published before 2005 were comparable to those of the main analyses ([Table S6](#app1-jcm-08-02039){ref-type="app"}). 3.3. Co-Primary Outcomes {#sec3dot3-jcm-08-02039} ------------------------ ### 3.3.1. All-Cause Mortality {#sec3dot3dot1-jcm-08-02039} Twenty-three trials with 15,487 patients reported data on all-cause mortality, including five trials with 4960 patients at overall low risk of bias. Mortality proportions were 8.0% in the LMWH group and 6.2% in the control group ([Figure 2](#jcm-08-02039-f002){ref-type="fig"}). Meta-analysis of low-risk-of-bias trials showed no statistically significant effect on all-cause mortality (RR 1.03; 95%CI 0.92 to 1.16; *p* = 0.60; *I^2^* = 0%; TSA-adjusted CI 0.88 to 1.20; [Table 2](#jcm-08-02039-t002){ref-type="table"}). When assessing all trials, the conventional meta-analysis results remained similar, while TSA suggested futility, rejecting a 20% RRR or RRI in mortality. All sensitivity analyses were consistent with the primary analysis ([Table 2](#jcm-08-02039-t002){ref-type="table"}, [Table S7](#app1-jcm-08-02039){ref-type="app"}). Subgroup analyses showed no statistically significant tests of interaction ([Table S8](#app1-jcm-08-02039){ref-type="app"}). The overall level of certainty of the evidence was low ([Table 2](#jcm-08-02039-t002){ref-type="table"}). ### 3.3.2. Symptomatic Venous Thromboembolism {#sec3dot3dot2-jcm-08-02039} Twenty-five trials with 15,920 patients reported data on symptomatic VTE, including five trials with 4878 patients at overall low risk of bias. Symptomatic VTE proportions were 1.1% in the LMWH group and 1.8% in the control group ([Figure 3](#jcm-08-02039-f003){ref-type="fig"} and [Figure 4](#jcm-08-02039-f004){ref-type="fig"}). Meta-analysis of low-risk-of-bias trials showed a statistically significant beneficial effect on symptomatic VTE, which was not confirmed by TSA (RR 0.65; 95%CI 0.45 to 0.94; *p* = 0.02; *I^2^* = 0%; TSA-adjusted CI 0.15 to 3.05; [Table 2](#jcm-08-02039-t002){ref-type="table"}). When including all trials, both conventional meta-analysis and TSA showed a beneficial intervention effect (RR 0.62; 95%CI 0.48 to 0.81; *p* = 0.0006; *I^2^* = 0%; TSA-adjusted CI 0.44 to 0.89; NNT 137; 97.5%CI 87 to 330; [Table 2](#jcm-08-02039-t002){ref-type="table"}; [Figure 3](#jcm-08-02039-f003){ref-type="fig"} and [Figure 4](#jcm-08-02039-f004){ref-type="fig"}). The primary analysis results were confirmed by three out of four sensitivity analyses ([Table 2](#jcm-08-02039-t002){ref-type="table"}, [Table S7](#app1-jcm-08-02039){ref-type="app"}). The direction of the intervention effect consistently suggested benefit in all subgroups, and there were no statistically significant tests of interaction ([Table S8](#app1-jcm-08-02039){ref-type="app"}). The overall level of certainty of the evidence was moderate ([Table 2](#jcm-08-02039-t002){ref-type="table"}). ### 3.3.3. Major Bleeding {#sec3dot3dot3-jcm-08-02039} Thirty-three trials with 13,091 patients reported data on major bleeding, including five trials with 4960 patients at overall low risk of bias. Major bleeding proportions were 0.9% in the LMWH group and 0.8% in the control group ([Figure 5](#jcm-08-02039-f005){ref-type="fig"}). Meta-analysis of low-risk-of-bias trials showed a non-statistically significant increase in major bleeding (RR 1.70; 95%CI 0.77 to 3.74; *p* = 0.19; *I^2^* = 0%; [Table 2](#jcm-08-02039-t002){ref-type="table"}). TSA could not be conducted, since less than 5% of the required information size was accrued. When including all trials, both conventional meta-analysis and TSA showed no statistically significant effect (RR 1.07; 95%CI RR 0.72 to 1.59; *p* = 0.74; *I ^2^*= 0%; TSA-adjusted CI 0.18 to 5.73; [Table 2](#jcm-08-02039-t002){ref-type="table"}). Sensitivity analyses were consistent with the primary analyses ([Table 2](#jcm-08-02039-t002){ref-type="table"}, [Table S7](#app1-jcm-08-02039){ref-type="app"}). Subgroup analyses showed that low-dose LMWH for more than 30 days was associated with higher risk of major bleeding as compared to shorter treatments (RR 2.20; 95% CI 1.00 to 4.82 vs RR 0.84; 95%CI 0.53 to 1.32, *p* = 0.04 for test of interaction; [Table S8](#app1-jcm-08-02039){ref-type="app"}). The overall level of certainty of the evidence was low to moderate ([Table 2](#jcm-08-02039-t002){ref-type="table"}). 3.4. Secondary Outcomes {#sec3dot4-jcm-08-02039} ----------------------- ### 3.4.1. Serious Adverse Events {#sec3dot4dot1-jcm-08-02039} Eight trials with 5180 patients reported data on SAE, although events were generally not defined according to ICH-GCP. SAE proportions were 5.4% in the LMWH group and 3.8% in the control group ([Supplement Figure S5](#app1-jcm-08-02039){ref-type="app"}). The one trial at overall low risk of bias, including 1150 patients, showed no statistically significant intervention effect on SAE (RR 0.89; 95%CI 0.68 to 1.17; *p* = 0.42; TSA-adjusted CI 0.41 to 1.96; [Table 2](#jcm-08-02039-t002){ref-type="table"}). This result was confirmed in both conventional meta-analysis and TSA of all trials regardless of bias risk (RR 0.98; 95% CI 0.78 to 1.25; *p* = 0.89; *I^2^* = 0%; TSA-adjusted CI 0.37 to 2.58; [Table 2](#jcm-08-02039-t002){ref-type="table"}). As predefined sensitivity analysis, we categorized mortality, symptomatic VTE, and major bleeding events from 37 trials as SAE and used these data to estimate the proportion of patients with one or more SAEs: the results were consistent with those of the primary analysis ([Table S9](#app1-jcm-08-02039){ref-type="app"}). Subgroup analyses showed no statistically significant tests of interaction. The overall level of certainty of the evidence was very low to low ([Table 2](#jcm-08-02039-t002){ref-type="table"}). ### 3.4.2. Clinically Relevant Non-Major Bleeding {#sec3dot4dot2-jcm-08-02039} Five trials with 3372 patients reported data on clinically relevant non-major bleeding. Clinically relevant non-major bleeding proportions were 1.0% in the LMWH group and 0.7% in the control group ([Figure S6](#app1-jcm-08-02039){ref-type="app"}). No trials were at overall low risk of bias. Meta-analysis of all trials showed no statistically significant intervention effect on clinically relevant non-major bleeding (RR 1.50; 95%CI 0.72 to 3.12; *p* = 0.28; *I^2^* = 0%; [Table 2](#jcm-08-02039-t002){ref-type="table"}), and TSA could not be conducted, since less than 5% of the required information size was accrued. Sensitivity analyses were consistent with the primary analysis ([Table 2](#jcm-08-02039-t002){ref-type="table"}, [Table S7](#app1-jcm-08-02039){ref-type="app"}). Subgroup analyses showed no statistically significant tests of interaction. The overall level of certainty of the evidence was very low ([Table 2](#jcm-08-02039-t002){ref-type="table"}). ### 3.4.3. Any Venous Thromboembolism {#sec3dot4dot3-jcm-08-02039} Thirty trials with 5849 patients reported data on any VTE, including three trials with 1254 patients at overall low risk of bias. Proportions of any VTE were 10.7% in the LMWH group and 17.6% in the control group ([Figure S7](#app1-jcm-08-02039){ref-type="app"}). Meta-analysis of the low risk of bias trials showed a statistically significant beneficial effect on any VTE, which was not confirmed by TSA (RR 0.57; 95%CI 0.38 to 0.84; *p* = 0.005; *I^2^* = 0%; TSA-adjusted CI 0.11 to 2.82; [Table 2](#jcm-08-02039-t002){ref-type="table"}). When including all trials, both conventional meta-analysis and TSA showed a beneficial intervention effect (RR 0.61; 95%CI 0.50 to 0.75; *p* \< 0.00001; *I^2^* = 47%; TSA-adjusted CI 0.49 to 0.82; NNT 15; 97.5%CI 11 to 21; [Table 2](#jcm-08-02039-t002){ref-type="table"}). The primary analysis results were confirmed by all sensitivity analyses ([Table 2](#jcm-08-02039-t002){ref-type="table"}, [Table S7](#app1-jcm-08-02039){ref-type="app"}). Subgroup analyses showed no statistically significant tests of interaction. The overall level of certainty of the evidence was low to moderate ([Table 2](#jcm-08-02039-t002){ref-type="table"}). 4. Discussion {#sec4-jcm-08-02039} ============= In this systematic review on low-dose LMWH versus placebo or no treatment, LMWH was not associated with a statistically significant intervention effect on mortality, major bleeding, clinically relevant non-major bleeding, or SAE. Conversely, we found a large beneficial intervention effect on both symptomatic VTE and on any VTE which included asymptomatic events detected through screening. These effects were consistent among subgroup and sensitivity analyses, but the effect size varied per patient type, and the quality of the evidence was moderate. In the TSAs of mortality, symptomatic VTE, and any VTE, the adjusted monitoring boundaries were crossed (respectively, for futility and for benefit), indicating a low risk of random error. The intervention effects of low-dose LMWH on SAE and bleeding events remain inconclusive, as TSA monitoring boundaries were not crossed, and quality of evidence was low. There was a suggestion of publication bias in the reporting of any VTE, and attrition bias may have influenced the primary outcomes. 4.1. Considerations on the Optimal Prophylactic Dose {#sec4dot1-jcm-08-02039} ---------------------------------------------------- Previous systematic reviews did not observe a mortality benefit for patients receiving LMWH thrombosis prophylaxis compared to patients receiving placebo or no treatment, which is confirmed by our results including TSA. Although it was previously thought that LMWHs might improve survival in cancer patients, later systematic reviews found no survival benefit in cancer patients receiving different prophylactic doses of LMWH \[[@B6-jcm-08-02039],[@B9-jcm-08-02039]\]. Additionally, we detected no beneficial effect on mortality in any patient category in a previous meta-analysis on intermediate-dose LMWH \[[@B11-jcm-08-02039]\]. Nevertheless, we cannot exclude the possibility of a smaller intervention effect than 20% RRR/RRI on mortality; this would require many more randomized patients, as we used a 20% RRR for calculating the required information size in TSA. In line with previous literature, we found a consistent beneficial intervention effect on VTE in subgroup analyses according to patient type, although effect sizes varied among subgroups. The overall incidence of symptomatic VTE was low, resulting in an NNT of 137. Effect estimates were rather similar regardless of bias risk (low risk RCTs estimated an RRR of 35%, while all RCTs combined estimated an RRR of 41%), suggesting we may base our conclusions on the more accurate estimates derived from the meta-analyses of all trials. Previous systematic reviews on thrombosis prophylaxis have found larger relative risk reductions \[[@B6-jcm-08-02039],[@B7-jcm-08-02039],[@B10-jcm-08-02039],[@B25-jcm-08-02039]\]. This could indicate that low-dose LMWH may be slightly less effective for the prevention of VTE than more frequently used higher doses. However, this indirect comparison should be viewed with caution, as differences between reviews regarding study selection criteria could also explain the difference. A direct comparison in a homogeneous patient population is required for strong inferences about the efficacy of low-dose LMWH compared to higher doses. Finally, evidence on adverse events remains inconclusive. The point-estimate of the low-risk-of-bias trials suggested a 70% RRI in major bleeding which was not statistically significant, while the estimate including all trials was neutral. This difference may relate to bias risk but could also be explained by trial characteristics: cancer and treatment duration are risk factors for major bleeding, and three out of five low-risk-of-bias trials included oncological patients who were generally treated for a longer duration \[[@B26-jcm-08-02039]\]. The increased risk of major bleeding in the subgroup of oncological patients was comparable to that reported in previous systematic reviews for this patient category \[[@B6-jcm-08-02039],[@B7-jcm-08-02039]\]. Conversely, the risk of major bleeding for other patient types was low compared to that indicated in other systematic reviews \[[@B10-jcm-08-02039],[@B11-jcm-08-02039],[@B25-jcm-08-02039]\]. This may be explained by the low LMWH dose but also by differences in included patients or co-interventions. Data on clinically relevant non-major bleeding were reported by only a few trials, and analyses were inconclusive. There was no apparent effect on SAE, confirmed by sensitivity analyses in which we incorporated data from nearly all available trials. Assessment of these two outcomes was hampered by wide variations in definitions and reporting between trials, resulting in low- to very low quality evidence and limiting inferences on the harms of low-dose LMWH. 4.2. Implications for Clinical Practice {#sec4dot2-jcm-08-02039} --------------------------------------- In general, clinicians will not prescribe thrombosis prophylaxis without considering both effectiveness and harms. This balance may differ depending on patients' characteristics such as disease type, severity of illness, or surgery. In prespecified subgroup analyses according to patient type, we found that, in surgical patients, low-dose LMWH reduced both symptomatic and any VTE, without evidence for increased major bleeding. In orthopedic patients, there was a statistically significant reduction in any, but not in symptomatic, VTE, with no evidence for increased major bleeding events. Although not statistically significant, there was a 39% RRR in symptomatic VTE, and the discrepancy may be explained by low power. In oncological patients, a beneficial effect on symptomatic VTE, but not on any VTE, was found. Additionally, the direction of the intervention effect suggested an increase in major bleeding. There were no statistically significant beneficial or harmful effects in acutely ill medical patients, suggesting either that there was a very small intervention effect with concurrent high numbers needed to treat or that a low LMWH dose is insufficient for this type of patient. Recent guidelines have recommended an individualized approach towards thrombosis prophylaxis in acutely ill medical patients \[[@B2-jcm-08-02039]\]. On the basis of our results, one could hypothesize that medical patients deemed at high risk of VTE will mainly benefit from higher doses of thrombosis prophylaxis. Finally, only very few neurological patients were included, limiting inferences for this subgroup. This systematic review provides a general overview of the effects of low-dose LMWH: although there are differences between patient subgroups, there also are many similarities in the direction of effects. Overall, we found that low-dose LMWH was most effective in surgical, orthopedic, and oncological patients, while the estimated RRI for bleeding events was low in most prespecified patient subgroups, except for oncological patients. These results should be viewed in the perspective of the limited quality of the evidence and the inherent limited power of subgroup analyses. In cases where physicians are in doubt whether a patient should receive thrombosis prophylaxis or no prophylaxis at all or when a higher prophylactic dose is deemed inappropriate with respect to bleeding risk, clinicians may consider a low-dose LMWH for thrombosis prophylaxis, especially in surgical and orthopedic patients. 4.3. Strengths and Limitations {#sec4dot3-jcm-08-02039} ------------------------------ Strengths of this review include its systematic and transparent methodology according to recommendations by the Cochrane Handbook, the PRISMA statement, and the GRADE working group. We used a prespecified protocol, a comprehensive search strategy without language restrictions, although we did assess one Chinese article as abstract only, independent data extraction and bias assessment by two authors, and incorporation of bias risk assessment in the results and conclusions. Finally, we applied TSA to all outcomes to assess the risks of random error and to estimate the required information size. Nevertheless, several important limitations apply. Our main goal was to make general inferences on the efficacy and safety of low-dose LMWH, using all available evidence. Consequently, there was a high amount of clinical heterogeneity between trials. The balance between thrombosis and bleeding may vary depending on patient subgroup characteristics: relying on overall effect estimates could obscure more subtle associations or lead to wrong inferences about a subpopulation. However, the distinction between different patient populations is somewhat arbitrary in any systematic review, and we attempted to account for clinical heterogeneity by conducting several preplanned subgroup analyses. This approach offers the benefit of increased power of the meta-analysis, and we found the direction of the intervention effects was equal in most subgroups. A second limitation concerns the inclusion of trials comparing low-dose LMWH to an inactive comparator, which led to the selection of mainly older trials or trials assessing LMWH in specific patient types or countries, limiting the generalizability of our results. In a post-hoc sensitivity analysis excluding trials published before 2005, the results remained comparable, although no inferences could be made for subgroups due to the very limited sample size. Third, to estimate the effect of low-dose LMWH on SAEs we conducted a sensitivity analysis to estimate the proportion of patients having one or more SAEs. For this purpose, we categorized mortality, symptomatic VTE, SAE, and major bleeding events from 37 trials as SAE. In reality, not all symptomatic VTE and major bleeding events are SAEs by definition (i.e., a distal leg thrombosis may be classified as adverse event, while pulmonary embolism can be a serious adverse event), but making this distinction was impossible on the basis of insufficiently detailed trial reports. Last, the best-/worst- and worst-/best-case analyses we performed to explore the influence of missing outcome data were probably overpowered to detect potential attrition bias, since the incidence of lost to follow-up was higher than the incidence of the primary outcomes. 5. Conclusions {#sec5-jcm-08-02039} ============== In a wide variety of patients at risk of VTE, there was very low to moderate-quality evidence that low-dose LMWH for thrombosis prophylaxis did not decrease all-cause mortality but reduced the incidence of symptomatic and asymptomatic VTE, while results on the intervention effects on major bleeding, clinically relevant non-major bleeding, and SAE remain inconclusive. We would like to thank S. van der Werf, medical information specialist, for her assistance with the development and execution of the search strategy. ###### Click here for additional data file. The following are available online at <https://www.mdpi.com/2077-0383/8/12/2039/s1>, Figure S1: Forest plot of SAE, stratified for patient type; Figure S2: Forest plot of clinically relevant non-major bleeding, stratified for patient type; Figure S3: Forest plot of any VTE, stratified for patient type; Figure S4: Funnel plot of all-cause mortality; Figure S5: Funnel plot of symptomatic VTE; Figure S6: Funnel plot of major bleeding; Figure S7: Funnel plot of any VTE; Table S1: Search strategy; Table S2: Characteristics of included randomized trials, stratified by patient type; Table S3: Ongoing trials; Table S4: Risk of bias assessment; Table S5: Sensitivity analysis: best-worse and worst-best case scenario's; Table S6: Sensitivity analysis: trials with publication year ≥ 2005; Table S7: Sensitivity analysis: including LMWH types not a priori defined; Table S8: Subgroup analysis: co-primary outcomes; Table S9: Sensitivity analysis: proportion of SAE and cumulative SAE. F.K. developed the original idea for this study. R.J.E., W.B., J.W., R.O.B.G., K.M., I.C.C.v.d.H., and F.K. contributed to the design of the study including the development of the protocol. R.J.E. and W.B. acquired the data. R.J.E. and F.K. performed the statistical analysis. R.J.E., W.B., J.W., R.O.B.G., K.M., I.C.C.v.d.H., and F.K contributed to interpretation of the data. R.J.E. drafted the first version of the manuscript, and all authors contributed critically to subsequent versions. R.J.E., W.B., J.W., R.O.B.G., K.M., I.C.C.v.d.H., and F.K approved the final manuscript. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. RJE is supported by a personal grant from the 'Groninger AGIKO programme', funded by the University of Groningen. This program supports young clinicians who combine medical specialist training with obtaining a PhD degree. The university had no role in the design or conduct of the study, analysis or interpretation of the data, review or approval of the manuscript, or the decision to submit the manuscript for publication. KM reports grants from Bayer, Sanquin, and Pfizer; speaker fees from Bayer, Sanquin, Boehringer Ingelheim, BMS, and Aspen; travel support from Bayer, and consulting fees from Uniqure outside the submitted work; JW is a member of the task force at the Copenhagen Trial Unit to develop theory and software of Trial Sequential Analysis; other authors have disclosed no potential conflicts of interest. {#jcm-08-02039-f001} {#jcm-08-02039-f002} {#jcm-08-02039-f003} {#jcm-08-02039-f004} {#jcm-08-02039-f005} jcm-08-02039-t001_Table 1 ###### LMWH dose definitions. LMWH Type A Priori Defined as Low-Dose LMWH Dose Used in Included Trials ------------ ----------------------------------- ------------------------------ Bemiparin \<3500 IU 2500 IU Certoparin \<5000 IU 3000 IU Dalteparin \<5000 IU 2500 IU ^a^ Enoxaparin \<40 mg 20 mg Nadroparin \<5700 IU 2850--3800 IU ^b,c^ Parnaparin \<4250 IU 3200 IU Reviparin \<3436 IU 1750 IU Tinzaparin \<4500 IU 3500 IU ^d^ IU: International Units; LMWH: low-molecular-weight heparin; mg: milligrams; ^a^ Sandset et al. used weight-dependent doses of 3000--5500 IU; ^b^ Fraisse et al. used weight-dependent doses of 3800--5700 IU; ^b^ Yoo et al. used weight-dependent doses of 2850--5700 IU; ^c^ Xiao-Li et al. used weight-dependent doses of 41--62 IU/kg; ^d^ Sorensen et al. and Lassen et al. used weight-dependent doses: 50 IU/kg. jcm-08-02039-t002_Table 2 ###### Conventional meta-analysis and trial sequential analysis outcomes. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Outcome Included Trials Trials (Patients) Conventional Meta-Analysis ^a^ Primary TSA ^a^\ Sensitivity TSA ^a^\ Sensitivity TSA ^a^\ Certainty of Evidence α 2.5%; ß 90%; RRR 20%;\ α 2.5%; ß 90%; RRR Based on Low Risk Trials; D^2^ Model Variance Based α 2.5%; ß 90%; RRR 20%; D^2^ 25% D^2^ Model Variance Based ---------------------------------------- ----------------- ------------------------ --------------------------------- --------------------------------- ------------------------------------------------------------------------ ---------------------------------- -------------------------- Mortality Low bias risk 5 (4.960) RR 1.03 (0.92 to 1.16) RR 1.03 (0.88 to 1.20) Insufficient data (\<5% of DIS) RR 1.03 (0.86 to 1.23) Low ^d,\ e,\ f^ All 23 (15.487) RR 0.99 (0.85 to 1.14) RR 1.02 (0.89 to 1.16) ^b^ Insufficient data (\<5% of DIS) RR 1.02 (0.90 to 1.15) Low ^d,\ g^ Symptomatic VTE Low bias risk 5 (4.878) RR 0.65 (0.45 to 0.94) RR 0.67 (0.15 to 3.05) RR 0.67 (0.32 to 1.38) 0.67 (0.15 to 3.05) Moderate ^e^ All 25 (15.920) RR 0.62 (0.48 to 0.81) RR 0.62 (0.44 to 0.89) ^c^ RR 0.62 (0.42 to 0.92) RR 0.62 (0.20 to 1.95) Moderate ^g^ Major bleeding Low bias risk 5 (4.960) RR 1.70 (0.77 to 3.74) Insufficient data (\<5% of DIS) Insufficient data (\<5% of DIS) Insufficient data (\<5% of DIS) Moderate ^f^ All 33 (13.091) RR 1.07 (0.72 to 1.59) RR 1.01 (0.18 to 5.73) ^c^ RR 1.01 (0.52 to 1.93) RR 1.09 (0.75 to 1.60) Low ^e,\ f,\ g^ SAE Low bias risk 1 (1.150) RR 0.89 (0.68 to 1.17) RR 0.89 (0.41 to 1.96) RR 0.89 (0.27 to 2.96) RR 0.89 (0.36 to 2.23) Low ^e,\ f,\ h^ All 8 (5.180) RR 0.98 (0.78 to 1.25) RR 0.98 (0.37 to 2.58) Insufficient data (\<5% of DIS) RR 0.98 (0.77 to 1.24) Very low ^d,\ e,\ f,\ g^ Clinically relevant non-major bleeding Low bias risk 0 (0) \- \- \- \- \- All 5 (3.372) RR 1.50 (0.72 to 3.12) Insufficient data (\<5% of DIS) Insufficient data (\<5% of DIS) Insufficient data (\<5% of DIS) Very low ^d,\ e,\ f,\ g^ Any VTE Low bias risk 3 (1.254) RR 0.57 (0.38 to 0.84) RR 0.57 (0.11 to 2.82) RR 0.57 (0.32 to 1.01) Not performed (D^2^ \>25%) Moderate ^e,\ i,\ k^ All 30 (5.849) RR 0.61 (0.50 to 0.75) RR 0.63 (0.49 to 0.82) RR 0.63 (0.50 to 0.80) Not performed (D^2^ \>25%) Low ^e,\ i,\ j,\ k^ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^ Small discrepancies of the intervention effect estimates between traditional RevMan meta-analyses and the TSA-adjusted results may occur due to different pooling methods (for example the inclusion of zero-event trials in TSA analyses); ^b^ TSA monitoring boundary for futility crossed; ^c^ sensitivity analysis using Peto's odds ratio showed similar results; ^d^ downgraded for inconsistency, since point estimates varied widely; ^e^ downgraded for imprecision, since TSA-adjusted confidence interval crossed '1'; ^f^ downgraded for imprecision, since conventional confidence interval crossed '1'; ^g^ downgraded for risk of bias, since (some) included trials were at high risk of bias; ^h^ downgraded for indirectness, since only one trial was included under assessment; ^i^ downgraded for risk of publication bias, since there was important asymmetry in the funnel plot; ^j^ downgraded for inconsistency, since point estimates varied widely and there was moderate statistical heterogeneity; ^k^ upgraded, since there was a strong association. α: two-sided significance level, ß: power; D^2^: diversity; DIS: diversity-adjusted information size; OR: odds ratio; RR: relative risk; RRR: relative risk reduction; SAE: serious adverse events; TSA: trial sequential analysis; VTE: venous thromboembolism.

INTRODUCTION {#sec1-1} ============ Bonding of composite resin to dentin has always been a challenge for the clinicians. Although there is an improvement in handling and bonding characteristics of the newer dental materials,\[[@ref1][@ref2]\] the collagen structure and stability of the dentin bond strength are still a challenge. Studies have proved that collagen in the hybrid layer is affected by enzymatic degradation, leading to bond failure over time.\[[@ref3][@ref4]\] Collagen fibrils in tissue are stabilized by lysyl oxidase-mediated covalent intermolecular cross-linking.\[[@ref5]\] The biomechanical properties of type I collagen can be improved by introducing more cross-links within and/or between the fibrils with the treatment of specific chemical agents. Grape seed extract, which is a naturally occurring cross-linker, mainly composed of proanthocyanidins (PAs), could be a good candidate to fulfill such role. It is a natural plant metabolite and showed to possess low toxicity and ability to induce exogenous cross-links.\[[@ref6]\] PA-based compounds can improve dentin collagen physical properties.\[[@ref7][@ref8]\] Another such important agent is ascorbic acid. It is an essentially required component in the synthesis of hydroxyproline and hydroxylysine in collagen. Hydroxyproline serves to stabilize the collagen triple helix. Hydroxylysine is necessary for the formation of the intermolecular crosslinks in collagen. It is believed that ascorbate modulates collagen production through its effect on prolyl hydroxylation.\[[@ref9]\] Hence, we can employ it as an effective chairside procedure to overcome the disadvantage of reduced bond strength of composite resin to dentin. The present study was carried out to evaluate if collagen cross-linking agents such as PA and sodium ascorbate can affect the shear bond strength of composite resin bonded to the dentinal surfaces of the teeth. The null hypothesis was that "there was no difference in shear bond strength of resin composite with dentin with or without treated with collagen cross-linking agent PA and sodium ascorbate." SUBJECTS AND METHODS {#sec1-2} ==================== One hundred freshly extracted human permanent molars were used in this study with the following exclusion/inclusion criteria. Inclusion criteria {#sec2-1} ------------------ Permanent molars with intact crownAll teeth should be free of caries, any visible discoloration, or crack and without any restoration. Exclusion criteria {#sec2-2} ------------------ Teeth which did not meet our inclusion criteria were excluded from the study. The roots of the specimens were mounted in self-cure acrylic resin, with occlusal surface parallel to the floor and extending 4 mm above the surface. Dentin surface was prepared with the help of slow-speed sectioning diamond disc under copious water supply and remove complete enamel portion and exposing the dentinal surface, then the surface of dentin was finished with wet 600 grit silicon carbide paper under running water to produce a standardized smear layer. After that, the dentinal surfaces of the specimens were acid etched with 35% phosphoric acid (SwissTec, COLTENE, Switzerland) as per manufacturers\' recommendation. The specimens were randomly divided into three groups based on the surface treatment of dentin as follows: Group I (*n* = 20) -- No dentin pretreatment was doneGroup II (*n* = 40) -- 10% sodium ascorbate pretreatment. This group was further divided into two subgroups based on the pretreatment time as follows:Subgroup IIa (*n* = 20) -- The etched dentin surface was treated with 10% sodium ascorbate solution for 5 min and rinsed with water, after which they were blot dried leaving a moist dentinal surface for bondingSubgroup IIb (*n* = 20) -- The etched dentin surface was treated with 10% sodium ascorbate solution for 10 min and rinsed with water, after which they were blot dried leaving a moist dentinal surface for bonding.Group III (*n* = 40) -- 6.5% PA pretreatment. This group was further subdivided into subgroups based on pretreatment time as follows:Subgroup IIIa (*n* = 20) -- The etched dentin surface was treated with 6.5% PA solution for 5 min and rinsed with water, after which they were blot dried leaving a moist dentinal surface for bondingSubgroup IIIb (*n* = 20) -- The etched dentin surface was treated with 6.5% PA solution for 10 min and rinsed with water, after which they were blot dried leaving a moist dentinal surface for bonding. Bonding for composite resin build-up {#sec2-3} ------------------------------------ Before application of bonding agent, a transparent plastic tube (3 mm in diameter and 5 mm height) was placed on the prepared occlusal surface, and the outer diameter was delineated with lead pencil as a reference mark for the application of the bonding agent. Then, plastic tube was removed, and bonding agent (One Coat Bond SL, SwissTec, COLTENE, Switzerland) was applied according to the manufacturer\'s instructions. Then, it was cured with light-emitting diode light (Woodpecker, Guilin Woodpecker Medical Instruments Co., Ltd., China) for 30 s (according to manufacturer\'s instruction). Composite resin buildup {#sec2-4} ----------------------- Transparent plastic tube was then fixed manually on the prepared surface of each tooth. A microhybrid composite resin (SwisTec, COLTENE, Switzerland) was applied in five 1 mm layers, and each layer was photopolymerized for 40 s, reaching 5 mm in total height, during which the light was moved around the tube to assure curing of the entire composite resin cylinder. The plastic tube was then removed, and excess composite resin was removed using a polishing disc (Sof-Lex™; 3M Espe, St. Paul, MN, USA). The samples were then stored in distilled water at 37°C for 48 h. Thermocycling and shear bond strength testing {#sec2-5} --------------------------------------------- After 48 h, samples were transferred from distilled water to normal water at 37°C for 24 h and then thermocycled for 500 cycles between 5°C and 55°C with a dwell time of 30 s each and transfer time between two baths was 5 and 10 s. After thermocycling, shear bond strength tests were performed on a universal testing machine (Unitek, 9450 PC, FIE, India) at a cross-head speed of 1 mm/min until the composite cylinder was dislodged from the tooth. Shear bond strength was calculated as the ratio of fracture load and bonding area, expressed in megapascals (MPa). The results were tabulated and subjected to statistical analysis. RESULTS {#sec1-3} ======= Obtained data analyzed and expressed in mean ± standard deviation (SD). Mean was compared by one-way analysis of variance and Tukey *post hoc* test *P* = 0.05 was taken as statistically significant. The results obtained through statistical analysis depicted that dentin pretreatment significantly increases the shear bond strength of the composite resin with dentin; duration of application of pretreatment materials on dentin also plays a significant role \[Tables [1](#T1){ref-type="table"}-[3](#T3){ref-type="table"}\]. ###### Shear bond strength in different groups irrespective of time of treatment  ###### Shear bond strength in different groups taking time of treatment into consideration  ###### Shear bond strength comparison between group/subgroup (Tukey\'s HSD test)  Mean difference was found to be maximum between Groups I and IIb and minimum between Groups IIIa and IIIb. All the between-group comparisons except between Groups IIIa and IIIb were significant statistically \[[Table 3](#T3){ref-type="table"}\]. On the basis of above evaluation, the following order of shear bond strength was observed in different groups/subgroups: IIb \> IIa \> IIIb \~ IIIa \> I" with "IIb \> IIa \> IIIb \> IIIa \> I". It was observed that shear bond strength values in Group I was of lower order, whereas shear bond strength values in Group II were of higher order. Shear bond strength values in Group III were of middle order. In Group I, shear bond strength ranged from 13.14 to 17.00 MPa with a mean value of 15.36 and a SD of 1.16 MPa. In Group IIa, shear bond strength ranged from 16.31 to 23.07 MPa with a mean value of 20.09 and a SD of 1.87 MPa. In Group IIb, shear bond strength ranged from 18.74 to 29.99 MPa, with a mean value of 22.55 and a SD of 2.51 MPa. In Group IIIa, shear bond strength ranged from 12.36 to 20.26 MPa, with a mean value of 17.01 and a SD of 2.10 MPa. In Group IIIb, shear bond strength ranged from 17.11 to 20.66 MPa, with a mean value of 18.46 and a SD of 0.97 MPa \[[Table 2](#T2){ref-type="table"}\]. DISCUSSION {#sec1-4} ========== The results obtained through statistical analysis depicted that sodium ascorbate and PA application before bonding significantly improve the bond strength of composite with dentin \[[Table 1](#T1){ref-type="table"}\]. The increase in bond strength in the present study may be attributed to improve dentin collagen stability, due to the higher number of collagen cross-links. Matrix metalloproteinases (MMPs) are such class of enzymes that are known to degrade collagen. These proteins are found in the dentin but more abundant on enamel--dentin junction and in the predentin.\[[@ref10]\] The MMPs converted from Pro-MMP which are trapped or bound in the dentin during its formation by lowering the pH to 4.5 or below.\[[@ref11][@ref12]\] Application of self-etching adhesives (acidic resins) to dentin increases 14-fold in MMP enzyme activity.\[[@ref13]\] Thus, adjunctive collagen pretreatment strategies have been proposed to improve dentin adhesion, through the use of agents that maintain the stability of collagen toward enzymatic degradation.\[[@ref14]\] These agents include the use of substances that are considered to be inhibitors of MMPs and cysteine cathepsins. Thus, pretreatment with collagen cross-linking agents can promise improvement in the bonding mechanism preserving the integrity of collagen hybrid layer. Various natural as well as synthetic cross-linking agents such as glutaraldehyde, tannic acid, PA, genipin, and cocoa seed extract have been used to strengthen the hybrid layer and have shown positive results in improving the bond strength to significant levels.\[[@ref15][@ref16][@ref17]\] Results of this study show that sodium ascorbate application significantly improves bond strength compared with control and PA application \[[Table 3](#T3){ref-type="table"}\]. In Group IIa (10% sodium ascorbate for 5 min), shear bond strength ranged from 16.31 to 23.07 MPa, with a mean value of 20.09 Mpa and a SD of 1.87 MPa. In Group IIb (10% sodium ascorbate for 10 min), shear bond strength ranged from 18.74 to 29.99 MPa, with a mean value of 22.55 MPa and a SD of 2.51 MPa. These results for the sodium ascorbate in the present study were found to be time dependent (*P* \< 0.001), whereas for PA difference was found to be insignificant (*P* = 0.096) \[Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}\]. The major action of the sodium ascorbate is in the stabilization of the collagen, as a cofactor of hydroxylation of proline and lysine.\[[@ref18]\] A study based on energy dispersive spectroscopy analysis has shown that calcium ion concentration after demineralization dropped from the 28.62 to 12.77% and increased to 15.99% after sodium ascorbate treatment for 10 min, With this may due to chemical interaction of sodium ascorbate with collagen fibres in dentine.\[[@ref19]\] However, shear bond strength of composite to deep dentin after treatment with 6.5% PA and 10% sodium ascorbate found the statistically significant better results with PA than the sodium ascorbate.\[[@ref20]\] PAs, which form a complex subgroup of the flavonoid compounds, have been found in a wide variety of fruits, vegetables, flowers, nuts, seeds, and bark. PA from grape seed extract has been shown to safely and effectively cross-link collagen in both *in vitro* and *in vivo* models and also inhibit MMP activity.\[[@ref21][@ref22]\] Considering its wide spectrum of benefits and high biocompatibility, we have selected this agent for this study. Studies done by Han *et al*. (2003) and Bedran-Russo *et al*. and found the improvement in ultimate tensile strength and shear bond strength, respectively, after pretreatment with PA.\[[@ref7][@ref8]\] These findings also corroborate with the results of these studies. A statistically significant improvement was seen after pretreatment with PA (*P* \< 0.001) \[[Table 1](#T1){ref-type="table"}\]. The proposed mechanisms for interaction between PA and proteins include covalent, ionic, hydrogen bonding, and hydrophobic interactions.\[[@ref7][@ref23]\] In the present study, molars were selected since they are most commonly restored, due to high incidence of caries. Therefore, validation and extension of our results await further investigations. The results of the *ex vivo* assays may not be directly comparable with the *in vivo* conditions, where all other parameters are to be considered. *In vivo* research is must to assess the clinical outcome and analysis of these agents so as to make out most of the benefits to the clinical adhesive dentistry. CONCLUSIONS {#sec1-5} =========== We can conclude that the treatment of dentinal surfaces with collagen cross-linking agent increases the shear bond strengths. Results for sodium ascorbate were found to be time dependent, whereas for PA, differences were nonsignificant. Financial support and sponsorship {#sec2-6} --------------------------------- Nil. Conflicts of interest {#sec2-7} --------------------- There are no conflicts of interest.

Introduction ============ Sickle cell disease (SCD) is a genetic disorder in which polymerization of deoxygenated sickle hemoglobin (HbS) leads to decreased deformability of the normally flexible erythrocytes. These rigid sickle-shaped red blood cells (RBC) can occlude the microvasculature leading to the sudden onset of painful vaso-occlusive episodes (VOC).^[@b1-1050083],[@b2-1050083]^ After HbS deoxygenates in the capillaries, it takes some time (seconds) for HbS polymerization and the subsequent flexible-to-rigid transformation. If the transit time of RBC through the microvasculature is longer than the polymerization time, sickled RBC will lodge in the microvasculature.^[@b3-1050083]^ Any trigger that decreases microvascular blood flow will prolong the transit time, promoting the entrapment of sickled RBC, resulting in vaso-occlusion. This physiology of SCD, described decades ago,^[@b4-1050083],[@b5-1050083]^ is fundamental to understanding the triggering of VOC. Patients report that stress, cold, and pain itself can trigger the onset of VOC^[@b6-1050083]^ but the frequency of VOC is highly variable. To date, the mechanism of how such events might trigger regional vaso-occlusion has not been fully elucidated. Psychological stress is an exacerbating factor in many chronic illnesses, such as SCD,^[@b7-1050083]--[@b10-1050083]^ coronary artery disease and myocardial ischemia.^[@b11-1050083],[@b12-1050083]^ Stress is significantly associated with increased pain intensity, reductions in social and physical activities and greater health care utilization.^[@b8-1050083],[@b13-1050083],[@b14-1050083]^ Day-to-day stressors have been associated with onset of pain and the course of VOC in SCD.^[@b9-1050083],[@b10-1050083]^ Stress is well-known to modulate autonomic nervous system (ANS) activity which in turn plays a major role in the regulation of regional blood flow.^[@b15-1050083]^ Interestingly, SCD children with greater mental-stress-induced autonomic reactivity had more severe clinical disease.^[@b16-1050083],[@b17-1050083]^ SCD subjects also have augmented ANS-mediated vasoconstriction in response to sighing, hypoxia, and pain.^[@b18-1050083]--[@b20-1050083]^ Therefore, autonomic dysregulation in SCD represents a plausible physiological link between mental stress and sickle RBC retention in the microvasculature.^[@b16-1050083],[@b18-1050083]--[@b21-1050083]^ Further understanding of this proposed mechanism of VOC triggering would not only help to predict disease manifestations, but would also open up opportunities for therapeutic intervention in disorders such as SCD in which preservation of microvascular blood flow is important.^[@b22-1050083]^ To address the role of mental stress in the physiology of SCD, we objectively quantified microvascular blood flow, measured by photoplethysmography, in response to standardized mental stress tasks in subjects with SCD and in controls. We also assessed cardiac ANS balance by analysis of heart rate variability in response to mental stress. We correlated photoplethysmogram-derived physiological indices with subjective indices of perceived stress assessed from standardized anxiety questionnaires. The aim of this study was to determine the relationship of peripheral and cardiac ANS responses with mental stress in SCD. Methods ======= The study was conducted under an institutional review board-approved protocol at the Children's Hospital Los Angeles with approved consent/assent. Twenty SCD subjects with Hb SS, S-β^0^, S-β^+^ or SC genotype and 16 age- and race-matched controls from the patients' family and friends were recruited. Experimental setup and study protocol ------------------------------------- All studies were performed in an ANS laboratory under strictly controlled settings.^[@b18-1050083]^ Neuropsychological stress was assessed at baseline using the State-Trait Anxiety Inventory (STAI) questionnaire.^[@b23-1050083]^ The STAI Y-1 and Y-2 evaluate "anxiety at this moment, aka *state anxiety*" and "how people generally feel, aka *trait anxiety*", respectively. Following 5 minutes of baseline recording, the stress induction protocol was presented through psychological software (E-prime 2.0, Psychology Software Tools, USA). The protocol consisted of a memory task (N-back)^[@b24-1050083]^ and a conflict test (Stroop),^[@b25-1050083],[@b26-1050083]^ presented in a randomized order, followed by a pain anticipation (PA) test ([Figure 1](#f1-1050083){ref-type="fig"}). During the N-back task, the subjects were asked to respond when the current letter matched the letter from n steps (n=zero, one, two, or three back) earlier in the sequence. During the Stroop task, the participants were asked to identify the font color of a word, not the written name of the word. We measured state anxiety between tasks. During the PA task, subjects read the following sentence on their computer screen: "You will receive a maximum pain stimulus in one minute. When you cannot tolerate the pain any longer, say STOP and the device will cool down to normal level immediately". However, no pain stimulus was actually applied. {#f1-1050083} Physiological measurements and analysis parameters -------------------------------------------------- All the physiological monitoring sensors were attached to the subjects' left arm. Microvascular blood flow was measured using photoplethysmography (Nonin Medical Inc., USA) and laser Doppler flowmetry (Perimed, Sweden). Respiration (thoracic and abdominal bands, zRip DuraBelt, Philips), the electrocardiogram and continuous blood pressure (Nexfin, Amsterdam) were recorded. Recorded data from all devices were exported for processing and analysis in MATLAB. The photoplethysmogram amplitude was normalized to its own 95^th^ percentile value during the full study. The average microvascular blood flow was calculated over the 5 min baseline period, the N-back, Stroop and PA tasks. The percent decrease in the amplitude of the photoplethysmogram or microvascular perfusion waveforms ([Figure 2](#f2-1050083){ref-type="fig"}; 2^nd^ and 3^rd^ signals, respectively) from the baseline mean was interpreted as a vasoconstriction response.^[@b18-1050083],[@b27-1050083]^ {#f2-1050083} Cardiac autonomic balance was assessed by analysis of the R-to-R interval and heart rate variability^[@b19-1050083],[@b28-1050083],[@b29-1050083]^ during baseline and mental stress tasks. The following power spectral indices were calculated: low frequency power, reflecting a combination of cardiac sympathetic and parasympathetic activity; high frequency power, reflecting parasympathetic activity;^[@b29-1050083],[@b30-1050083]^ and the ratio of low frequency power to high frequency power, reflecting sympathovagal balance.^[@b30-1050083]^ Percent changes in mean microvascular blood flow and mean spectral indices from baseline to tasks were calculated. The Student *t*-test (or Wilcoxon sign rank) or χ^2^ test was used to test baseline group differences and task differences. Robust regression was used to correlate vasoconstriction response and state anxiety during the PA task. Repeated measures analysis of variance was used to test differences in N-back and Stroop sublevels and accuracy scores. All statistical analyses were performed using STATA/IC 14.1 (StataCorp LP, TX, USA) with nominal significance set at *P*≤0.05. The methods are described in detail in the *Online Supplementary Methods S1*. Results ======= Data from a total of 20 SCD patients and 16 controls were analyzed. Transfused and non-transfused subjects with SCD were grouped together and healthy and sickle cell trait subjects (controls) were combined after it had been demonstrated that these factors were not statistically significant in the analyses. The percentage of HbS (HbS%) was considered to be zero in patients with sickle cell trait as the cellular distribution of HbS differs in sickle cell trait and does not contribute to sickling under the conditions of the experiments in this study, making the HbS% in sickle cell trait not comparable to that in transfused or non-trait sickle phenotypes. The subjects' characteristics are summarized in [Table 1](#t1-1050083){ref-type="table"}. Nine (45%) SCD subjects were on chronic transfusion, nine (45%) were being treated with hydroxyurea and two (10%) were not receiving either treatment. The characteristics of both groups were balanced except for hemoglobin concentration on the study day. Sixty-one percent of subjects had a level of education equivalent to high school or superior. Seventy-two percent reported that they had a high level of competitiveness on the visit screening questionnaire. ###### Population characteristics.  Vasoconstriction due to mental stress ------------------------------------- As determined from the photoplethysmogram, there was a significant drop in microvascular blood flow during both cognitive tasks (N-back and Stroop, *P*\<0.0001) and the PA task (*P*\<0.0001) ([Figure 3](#f3-1050083){ref-type="fig"}). [Figure 2](#f2-1050083){ref-type="fig"} (signal 2) shows a typical response of vasoconstriction in one subject. The drop in microvascular blood flow from baseline was greater during the PA task than during the cognitive tasks. A similar vasoconstriction response was observed when the microvascular blood flow was assessed by laser-Doppler flowmetry. Subjects had higher anxiety scores immediately after completing the tasks than at baseline (STAI Y-1, mean difference=6; *P*=0.0007). Eighty-five percent of patients with SCD and 75% of controls showed vasoconstriction compared to baseline during at least one cognitive task. Eighty-five percent of SCD patients and 87.5% of controls had decreases in mean blood flow during the PA task. There was no difference in the magnitude of responses between individuals with SCD and controls. Demographic variables such as age, gender, race, number of days from last menstruation, and laboratory values were not associated with the magnitude of the vasoconstriction response. {#f3-1050083} The Stroop test caused greater vasoconstriction than the N-back task, irrespective of the order in which the tests were presented (*P*=0.019) ([Figure 3](#f3-1050083){ref-type="fig"}). Subjects who were randomized to perform the Stroop task first had greater anxiety responses than did the subjects who performed the N-back task first (mean difference=10; *P*=0.03). Overall the accuracy score was significantly lower for the Stroop task than for the N-back task in all subjects (mean score difference=25, *P*\<0.001). The accuracy score for the Stroop and N-back tasks decreased as the difficulty increased from zero-back to three-back in the N-back task and from level one to level three in the Stroop task (*P*\<0.0001) ([Figure 4](#f4-1050083){ref-type="fig"}) but there was no further change in blood flow with increasing difficulty. Once the subjects manifested vasoconstriction, in comparison with baseline vascular tone, the vasoconstriction remained throughout the whole task regardless of the difficulty of the tasks. {#f4-1050083} Vasoconstriction response to perceived anxiety during pain anticipation ----------------------------------------------------------------------- On robust regression, the effect of state anxiety on blood flow response was greater in SCD patients than in controls (*P*=0.03 for the interaction), suggesting that higher anxiety at baseline (STAI Y-1) in SCD subjects is associated with less change in blood flow (coefficient = −1.85, *P*=0.002) in response to pain anticipation ([Figure 5](#f5-1050083){ref-type="fig"}). State anxiety had no effect on change in blood flow in control subjects. To understand why SCD subjects would have less response with high anxiety, we looked at the baseline blood flow. We found that highly anxious subjects tended to have a lower mean baseline blood flow (*Online Supplementary Figure S1*), meaning they were already vasoconstricted at baseline, limiting them from further vasoconstriction. This trend was not seen among controls. (*Online Supplementary Figure S2A*, *B*: high-anxiety SCD responder and low-anxiety SCD responder). {#f5-1050083} Cardiac autonomic response -------------------------- Since the ANS regulates blood flow and SCD subjects have dysautonomia,^[@b15-1050083],[@b28-1050083],[@b31-1050083],[@b32-1050083]^ we explored the effect of mental stress responses on cardiac autonomic balance. In com parison to the value at baseline, there was a significant decrease in R-to-R interval, signifying an increase in heart rate, during all tasks (*P*\<0.0001) ([Figure 6A](#f6-1050083){ref-type="fig"}). As for the microvascular blood flow response, the R-to-R interval was less during the Stroop task than during the N-back task (*P*=0.002). ![Autonomic nervous system responses to mental stress. (A) R-to-R interval (sec) and (B) high frequency power (sec^[@b2-1050083]^/Hz, shown on a log scale) in response to the N-Back and Stroop tasks in all subjects. There is a significant decrease in R-to-R interval and parasympathetic withdrawal during mental stress tasks compared to baseline. SE: standard error of mean; HFP: high frequency power.](10583.fig6){#f6-1050083} There was significant parasympathetic withdrawal during the N-back and Stroop tasks as reflected by the drop in high frequency power (*P*=0.002 and *P*\<0.0001, respectively) ([Figure 6B](#f6-1050083){ref-type="fig"}) The Stroop task caused stronger parasympathetic withdrawal than the N-back task (*P*\<0.0001). There was more sympathetic activation during the Stroop test (low-to-high power ratio: *P*=0.03), but not during the N-back task. We did not analyze autonomic reactivity during the PA task because the 1-minute test period was not long enough to derive spectral indices.^[@b29-1050083]^ Discussion ========== VOC is a significant complication of SCD and a major cause of morbidity and mortality.^[@b33-1050083]^ The frequency of VOC is related in part to hemoglobin-F content, white blood cell count, inflammatory status and other factors.^[@b34-1050083]--[@b36-1050083]^ However, there is still significant variability in crisis frequency among SCD subjects with otherwise similar hematologic status. Pain crises can be promoted by preceding dehydration, infection, injury, exposure to cold or emotional stress.^[@b37-1050083],[@b38-1050083]^ Much of the research in past decades has focused on adhesion and processes attributed to occlusion in the post-capillary venule, and to decreased flow due to nitric oxide depletion.^[@b39-1050083]^ While stress and cold are often mentioned, very little attention has been paid to decreased flow due to neurally induced vasoconstriction.^[@b32-1050083],[@b40-1050083]^ SCD patients undergo a tremendous amount of stress not only due to environmental challenges but also the illness-related stress of painful episodes, repeated medical procedures and life-threatening complications. Stress causes ANS hyperreactivity by enhancing the sympathetic nervous system and dampening the parasympathetic system in SCD subjects compared to non-SCD individuals.^[@b16-1050083],[@b17-1050083]^ Sympathetic and parasympathetic responses have been related to clinical vaso-occlusion in SCD, through ANS modulation of regional blood flow.^[@b15-1050083],[@b17-1050083]^ SCD is probably the best example of a disorder in which decreased microvascular perfusion can be directly related to the pathology of the disorder, because the increase in transit time from decreased blood flow promotes entrapment of rigid red cells in small vessels.^[@b3-1050083],[@b5-1050083]^ To our knowledge, this is the first study to quantify regional blood flow modulated by ANS reactivity under mental stress in SCD. Our data show that experimental mental stress caused a decrease in regional blood flow in all participants. While we thought that SCD subjects would exhibit stronger vasoconstriction because of their hyperresponsiveness to sympathetically induced stimuli, such as sighing,^[@b28-1050083]^ we did not detect a difference in stress-induced vasoconstriction between SCD patients and controls. We did find a significantly higher anxiety response score (*P*=0.03) in subjects who were exposed to the more difficult mental stress test first (Stroop). We also found that the degree of vasoconstriction was proportional to the magnitude of the stress. Subjects reported that overall the Stroop task was more stressful: accuracy scores were lower and there was also a greater decrease in blood flow with this cognitive stressor task. However, different sublevels of difficulty within a task type did not correlate with levels of vasoconstriction. This finding suggests that consecutive stressful events could make SCD patients more vulnerable to vaso-occlusion. We think that variability in the vasoconstriction response to stress may account in part for differences in clinical severity among SCD patients who have the same hemoglobin phenotype. The frequency of VOC and intensity of pain are higher among patients found to have high anxiety and stress scores on standard psychological assessments.^[@b8-1050083],[@b41-1050083],[@b42-1050083]^ We tried to correlate the vasoconstrictive response with clinical severity. As our SCD patients were either on chronic transfusion or hydroxyurea, the number of VOC was too low to detect differences in this current relatively small sample. Along with a strong vasoconstriction response, significant autonomic reactivity was seen in all subjects. The Stroop test was consistently more stressful, and induced greater vasoconstriction as well as greater autonomic reactivity. There was both sympathetic activation as well as parasympathetic withdrawal during this cognitive task. Mental stressors are known to influence autonomic function by sympathetic or parasympathetic tone alterations. Higher anxiety induces atherosclerosis via enhanced sympathetic modulation, increasing the risk of cardiovascular disease.^[@b43-1050083]^ In addition, mental stress and anxiety have been linked to impaired endothelial function via autonomic dysfunction.^[@b43-1050083]--[@b45-1050083]^ Endothelial function, quantified by flow mediated dilation, decreases as a result of mental stress tasks.^[@b46-1050083]^ Similarly, in SCD, a synergistic interaction between impaired local vascular function and the exaggerated neurally mediated vasoconstrictive response could further reduce peripheral blood flow, setting the stage for VOC. Consistent with the findings of our previous study,^[@b18-1050083]^ anticipation of pain caused significant vasoconstriction and this response was quantitatively greater than that of the calibrated experimental stress tasks ([Figure 3](#f3-1050083){ref-type="fig"}). We do not have strong evidence to conclude that the presence of SCD alone influences mental stress-induced vasoconstriction but anxiety seems to be a modifying factor. Interestingly unlike control subjects, SCD subjects who were highly anxious had less vasoconstriction during the PA task and *vice versa*. We think that this pattern of response occurred because highly anxious subjects were already vasoconstricted at baseline and this limited the magnitude of further vasoconstriction. So the fact that SCD subjects have less change in the vasoconstriction response to the stressors than controls actually reflects their chronically vasoconstricted state. Although not statistically significant, the trend of lower baseline blood flow with high anxiety in SCD can be seen in *Online Supplementary Figure S1*, which also shows the significant variability in baseline measures. Photoplethysmogram and microvascular perfusion signals from Perimed do not have absolute units, so measurements made as percent changes from baseline are more reliable, basically correcting for baseline variability and allowing detection of the differences seen in [Figure 5](#f5-1050083){ref-type="fig"}. These findings may be related to pain catastrophization and increased psychophysical pain sensitivity due to frequent pain episodes.^[@b7-1050083],[@b47-1050083],[@b48-1050083]^ Over the years, pain catastrophization may increase the frequency of pain and severity of pain crises.^[@b47-1050083],[@b49-1050083]^ From a standpoint of neural physiology, repeated acute pain creates a central neural pathological pain connectome^[@b50-1050083]^ that leads to baseline chronic pain and chronic vasoconstriction. Although baseline blood flow was not statistically significantly lower, probably due to insufficient study power, we suspect that the above-described phenomenon is the explanation for our findings and warrants further study. We showed that neurally mediated vasoconstriction is a biophysical marker of mental stress in SCD patients and controls. Mental stress has been identified as a trigger for pain crises in SCD and its connection with a decrease in microvascular perfusion seems to make a causal link to VOC. The probability of vaso-occlusion is predicted to be related to the relation between time to polymerization of deoxy HbS and microvascular flow.^[@b3-1050083],[@b5-1050083]^ Obviously, HbS is the major pathology in SCD. However, neurally mediated changes in microvascular flow certainly play a significant and unappreciated role. Individual variation in patterns of vasoconstriction with different ANS reactivity may offer a possible biological explanation for the variability in the frequency of VOC in SCD patients with similar hemoglobin phenotype. Identifying the high-risk individuals who show a phenotype of chronic vasoconstriction and repeated pain crises, and targeting them with neuro-modulatory cognitive-based therapies may improve vascular and neural physiology in SCD. In the primary stage of a crisis, implementing these learned cognitive-based therapies or distraction and relaxation techniques will help to improve the prognosis during acute pain. Microvascular flow in response to stress may also serve as an important surrogate endpoint for therapy in SCD and other diseases in which small vessel blood flow and reactivity are important. Some limitations of this study should be acknowledged. One limitation was that the small sample size did not allow us to detect a difference in the magnitude of vasoconstriction between groups and correlate it with a clinical outcome such as VOC. Since the concept that mental stress causes vasoconstriction has not been studied in SCD, prior effect size was not known to permit sample size calculation. Another reason for lack of difference between groups is that over 90% of our patients are on hydroxyurea or chronic transfusion and thus clinical crises are relatively uncommon. Any real magnitude differences would be more likely to emerge in studies with larger samples and untreated patients. However, the primary aim of this study was to understand the changes in peripheral and cardiac responses to mental stress. The fundamental study design presented here was able to detect changes in physiological signals with millisecond accuracy and clearly showed vasoconstriction responses and ANS reactivity due to mental stress in all subjects. We think that the consequences of these findings are mechanistically related to the pathophysiology of sickle cell vaso-occlusion. This work was supported by grants from the National Institutes of Health National Heart, Lung, and Blood Institute (U01 HL117718). The authors thank Justin Abbott for his contribution to the data collection. Check the online version for the most updated information on this article, online supplements, and information on authorship & disclosures: [www.haematologica.org/content/105/1/83](http://www.haematologica.org/content/105/1/83)

INTRODUCTION {#s1} ============ Upper track urothelial carcinoma (UTUC) is less common than bladder urothelial carcinoma and accounts for 5--10% of all urothelial carcinoma.^[@b1]^ The incidence of ureteral urothelial carcinoma (UUC) is approximately half that of pyelocaliceal urothelial carcinoma.^[@b2]^ UUC has a worse prognosis than pyelocaliceal urothelial carcinoma.^[@b3],[@b4]^ Owing to different anatomical considerations and oncological outcomes in UTUC, these different malignant entities must be evaluated independently.^[@b5]^ The gold standard treatment for UTUC is radical nephroureterectomy with excision of the bladder cuff, regardless of the tumour location.^[@b6]^ Recently, conservative surgery such as endoscopic ablation or segmental ureteral resection, which allows preservation of the upper urinary renal unit, has also been applied.^[@b7]^ However, preoperative histological evaluation through biopsy of the upper urinary tract is difficult, because ureteroscopy is an invasive procedure and usually requires general anaesthesia. Furthermore, the accuracy of ureteroscopic biopsy in predicting tumour stage and grade is limited, and the limitations of endoscopic biopsy must be balanced against the possible advantage of avoiding radical surgery.^[@b8]^ Thus, accurate preoperative prediction of tumour grade could be helpful in selecting more appropriate therapeutic options. CT urography (CTU) is an imaging modality with high diagnostic accuracy in the detection of UTUC and has replaced intravenous excretory urography and ultrasonography as the first-line imaging test for investigating high-risk patients.^[@b9]^ Even though several studies have investigated diffusion-weighted MRI (DW-MRI) as an imaging assessment for predicting tumour grade of UTUC,^[@b10]--[@b11]^ characteristic CTU findings that can predict tumour grade of UUC have not been identified, to the best of our knowledge. In this study, we aimed to evaluate the correlation between CTU imaging variables, including tumour size and imaging features, and histological grade of UUC, and to identify CTU imaging features that allow prediction of high-grade UUC, which should be treated by radical surgery. METHODS AND MATERIALS {#s2} ===================== Patients -------- This retrospective single-centre study was approved by the institutional review board at and written informed consent was not required. We searched institutional patient information systems to identify all consecutive patients with UUC who had undergone nephroureterectomy between January 2005 and July 2016. A total of 79 consecutive patients who underwent surgery with removal of a surgical specimen for histological analysis were registered. The inclusion criteria for this study were as follows: (i) tumours only located in the ureter, (ii) patients had undergone CTU scan prior to surgery and (iii) histologicalal confirmation of UUC with clear statement of histological grade according to the WHO 2004 classification system. Four patients were excluded because histological grade was not available in the pathological reports, and two patients did not undergo a CTU scan. Ultimately, 73 patients (52 males and 21 females; mean age, 68.92 ± 9.08 years) with 81 UUCs were included in our study. All pathological data were reviewed by a board-certificated pathologist, and all tumours were classified into low-grade and high-grade groups according to the WHO 2004 classification system and pathologic T stage of the tumours was assessed according to the TNM staging system. CTU technique ------------- All CTU examinations were performed using various CT scanners from 16-channel to 128-channel MDCT scanners (Somatom Sensation 16, Siemens Healthcare, Brilliance 64, Philips Medical Systems, Best, Netherlands or Somatom Definition Flash 128, Siemens Healthcare Forchheim, Germany). Scanning parameters of the most frequently used CT scanner (Brilliance 64, Philips Medical Systems, Best, Netherlands) were as follows: tube voltage, 120 kVp; effective tube current, 300 mAs; section thickness, 5 mm; pitch and speed, 0.891:1; rotation time, 0.75 s and collimation, 64 × 0.625 mm for 64-channel MDCT. Before acquisition of contrast-enhanced scans, simple unenhanced scans were obtained, after which 2 ml kg^--1^ non-ionic contrast material containing 300--350 mg ml^−1^ of iodine \[iomeprol (Iomeron 300, Bracco Altana Pharma, Konstanz, Germany), iopamidol (Pamiray 300, Dongkook Pharmaceutical, Seoul, Republic of Korea) or iobitridol (Xenetix 300, Guerbet, Villepinte, France)\] was intravenously administered at a rate of 3.0 ml s^−1^ using a standard power injector. For CTU, in addition to the unenhanced scan, two-phase studies were performed with combinations of corticomedullary and excretory phases at our institution. The corticomedullary phase began 30--40 s after contrast administration, and excretory phases began 300 s after contrast administration, respectively. Image analysis -------------- Two radiologists (DJS and STH with 17 and 3 years of experience, respectively, in interpreting genitourinary images) independently reviewed all CTU images on a picture archiving and communication system workstation (INFINITT PACS, INFINITT Healthcare, Seoul, Republic of Korea). The readers knew all patients had been diagnosed with UUC, but were informed of neither the histological grade nor the findings listed in the initial radiological report. They evaluated the following CTU imaging features: tumour location, tumour size, tumour enhancement value, multiplicity, periureteral infiltration, enlarged retroperitoneal lymph nodes with a short axis of more than 1 cm, and hydronephrosis grade. Tumour location was categorized into three groups (proximal, middle, and distal) according to anatomic ureteral segmentation. Tumour size was determined as the maximal length or diameter of the whole tumour presenting as ureteral soft-tissue mass or enhancing wall thickening on the axial, sagittal, or coronal CTU images. In patients with multiple lesions, the largest one was selected for size measurement. Tumour enhancement value was calculated as the difference between attenuation values in the corticomedullary phase and unenhanced phase. On corticomedullary phase images, the readers drew a circular ROI that included the enhancing solid portion of the tumour, avoiding adjacent mesenteric fat. The ROI was as large as possible to minimize noise. A ROI of the same size was placed in the corresponding location on the unenhanced scan image. The readers also reported hydronephrosis grade according to the modified version of the Society for Foetal Urology Hydronephrosis Grading System ([Table 1](#t1){ref-type="table"}). ###### Modified version of the Society for Fetal Urology Hydronephrosis Grading System Grade 0 1 2 3 4 --------------------------------- --------------- ------------------------------ -------------------------------------- -------------------------------------------------------------- -------------------------------------------------- Ureter and pelvocalyceal system No dilatation Local dilation of the ureter Ureteral and renal pelvis dilatation Ureteral and renal pelvis dilatation plus calices dilatation Further dilatation of ureter, pelvis and calices Renal parenchymal thickness Normal Normal Normal Normal Thin Statistical analysis -------------------- Descriptive statistics of means, standard deviations and frequencies were used to describe patient characteristics. Univariate logistic regression modelling, Mann--Whitney *U* tests, and *Χ*^2^ tests were used to assess the correlation between CTU imaging variables and histological tumour grade. Multiple logistic regression analysis using a backward selection method was performed to identify significantly independent CTU imaging variables that could predict high-grade tumours. Spearman correlation analysis was used to assess the correlation between tumour size and hydronephrosis grade. *Χ*^2^ test and linear-by-linear association were used to investigate the correlation of hydronephrosis grade and peritumoural intfiltration with pathologic T stage. A receiver operating characteristic (ROC) curve was constructed to identify the cut-off value of effective factors that provided the best diagnostic accuracy. Interobserver agreement was calculated using kappa statistics for nominal values, including hydronephrosis grade, peritumoural infiltration, multiplicity and presence of enlarged retroperitoneal lymph nodes. Intraclass correlation was calculated for continuous values including tumour size and contrast enhancement value. The scores were used to define agreement as follows: 0.41--0.60 denoted moderate agreement; 0.61--0.80, good agreement and greater than 0.81, excellent agreement. Statistical analysis was done using IBM SPSS Statistics version 22.0 for Windows (IBM Corp., Armonk, NY). A *p* value of less than 0.05 was considered statistically significant. RESULTS {#s3} ======= Images of the 73 patients with 81 UUCs were reviewed. The lesions were unilateral in all cases. 15 patients (20.5%) had low-grade UUCs ([Figure 1](#f1){ref-type="fig"}) and 58 patients (79.5%) had high-grade UUCs ([Figure 2](#f2){ref-type="fig"}). 22 (27.1%) lesions were located in the proximal ureter, 14 (17.2%) in the middle ureter, and 45 (55.5%) in the distal ureter. Eight (5.8%) patients had multiple lesions in the ipsilateral ureter. Clinicopathological characteristics of the patients are summarized in [Table 2](#t2){ref-type="table"}. {#f1} {#f2} ###### Clinicopathological characteristics of enrolled patients Characteristic Data -------------------------------------------------------- ----------------- Age (years)[*^a^*](#tb2fn1){ref-type="fn"} 68 ± 9 (43--86) Sex[*^b^*](#tb2fn2){ref-type="fn"}  Male 52 (71.2)  Female 21 (28.8) Hitologic grade of UTUC[*^b^*](#tb2fn2){ref-type="fn"}  High grade 58 (79.5)  Low grade 15 (20.5) Data are presented as mean (range) values. Data are presented as number (percentage) of patients. CTU imaging variables (tumour size, multiplicity, peritumoural infiltration, hydronephrosis grade, contrast enhancement value, presence of enlarged retroperitoneal lymph nodes) with respect to histological grade of UUCs are summarized in [Table 3](#t3){ref-type="table"}. The readers had excellent agreement for the other CT variables (*к* = 0.862 for hydronephrosis grade, intraclass correlation = 0.829 for tumour size, intraclass correlation = 0.892 for contrast enhancement value). In addition, there were good or moderate interobserver agreements for the other subjective assessments (*к* = 0.748 for multiplicity, *к* = 0.546 for periureteral infiltration). Tumour size was significantly larger in the high-grade group than in the low-grade group according to reader 1 (*p* = 0.028). Hydronephrosis grade was significantly higher in the high-grade group than in the low-grade group (*p* \< 0.001 for both readers). There was no significant difference in multiplicity, peritumoural infiltration, contrast enhancement value, or presence of enlarged retroperitoneal lymph nodes between the two groups. ###### Clinical characteristics of the enrolled patients according to histological grade Grade -------------------------------------- ----------------------- ---------------------- ---- ---------------------------------------- **Reader 1** Tumour size (mm) 39.7 (10--140) 23.3 (1--41) 0.028^[*b*](#tb3fn2){ref-type="fn"}^ Hydronephrosis grade \<0.001^[*c*](#tb3fn3){ref-type="fn"}^ 4 22 (37.9) 1 (6.7) 23 3 27 (46.6) 2 (13.3) 29 2 6 (10.3) 5 (33.3) 11 1 2 (3.4) 2 (13.3) 4 0 1 (1.7) 5 (33.3) 6 Enhancement value 56.4 (2--120) 51.2 (6--92) 0.508 Peritumoural infiltration 0.07^[*c*](#tb3fn3){ref-type="fn"}^ Present 17 (29.3) 1 (6.7) 18 Absent 41 (70.7) 14 (93.3) 55 Multiplicity 0.55^[*c*](#tb3fn3){ref-type="fn"}^ Present 7 (12.1) 1 (6.7) 8 Absent 51 (87.9) 14 (93.3) 65 Enlarged retroperitoneal lymph nodes 0.611^[*c*](#tb3fn3){ref-type="fn"}^ Present 11 (19.0) 2 (13.3) 13 Absent 47 (81.0) 13 (86.7) 57 Reader 2 Tumour size(mm) 43.10 (11--140) 34.50 30.14 (15--58) 27.50 0.234^[*b*](#tb3fn2){ref-type="fn"}^ Hydronephrosis grade \<0.001^[*c*](#tb3fn3){ref-type="fn"}^ 4 22 (37.9) 1 (6.7) 23 3 27 (46.6) 4 (26.7) 31 2 4 (7.0) 2 (13.3) 6 1 4 (7.0) 2 (13.3) 6 0 1 (1.7) 6 (40.0) 7 Enhancement value 55.5 (2--121) 58.9 (19--101) 0.793 Peritumoural infiltration 0.127^[*c*](#tb3fn3){ref-type="fn"}^ Present 8 (13.8) 0 (0.0) 8 Absent 50 (86.2) 15 (100.0) 65 Multiplicity 0.239^[*c*](#tb3fn3){ref-type="fn"}^ Present 5 (8.6) 0 (0.0) 5 Absent 53 (91.4) 15 (100.0) 68 Enlarged retroperitoneal lymph nodes 0.611^[*c*](#tb3fn3){ref-type="fn"}^ Present 11 (19.0) 2 (13.3) 13 Absent 47 (81.0) 13 (86.7) 57 Pathologic T stage \<0.001^[*c*](#tb3fn3){ref-type="fn"}^ Ta 4 (6.9) 6 (40.0) 10 T1 14 (24.1) 8 (53.3) 22 T2 11 (19.0) 0 (0.0) 11 T3 29 (50.0) 1 (6.7) 30 Data are presented as number (percentage) of patients. Mann--Whitney *U* test. Pearson's *Χ*^2^ test. Univariate logistic regression analysis revealed that hydronephrosis of Grade 3 or higher was significantly associated with high-grade tumour for both readers, and tumour size was significantly associated with high-grade tumour for reader 1. Multivariate logistic regression analysis using a backward selection method demonstrated that only hydronephrosis of Grade 3 or higher was a significant independent predictor of high-grade tumour for both readers ([Table 4](#t4){ref-type="table"}). Other CTU imaging variables, including tumour size, were omitted as independent variables in multivariate logistic regression analysis. In addition, there was no significant correlation between tumour size and hydronephrosis grade according to Spearman correlation analysis. ###### Results of the multivariate logistic regression analysis with backward selection of independent variables predictive of high-grade tumours Univariate logistic Multivariate logistic with variable selection -------------------------------------- ---------------------- ----------------------------------------------- -------------------- ------- **Reader 1** Tumour size (mm) 1.050 (1.006--1.096) 0.025 Grade of hydronephrosis 4 110 (5.83--2074.45) 0.002 72 (3.67--1411.89) 0.005 3 67.50 (5.09--893.63) 0.001 48 (3.48--661.60) 0.004 2 6 (0.51--69.75) 0.152 6 (0.47--75.34) 0.165 1 5 (0.27--91.51) 0.278 8 (0.31--206.37) 0.21 0 0.16 0.097 Enhancement value 1.01 (0.98--1.03) 0.536 Peritumoural infiltration 5.81 (0.71--47.69) 0.102 Multiplicity 1.92 (0.22--16.95) 0.557 Enlarged retroperitoneal lymph nodes 1.47 (0.29--7.53) 0.646 **Reader 2** Tumour size (mm) 1.027 (0.99--1.06) 0.146 Grade of hydronephrosis 4 126 (6.82--2328.09) 0.001 68 (3.46--1336.27) 0.005 3 40.5 (3.81--430.28) 0.002 24 (2.11--273.59) 0.009 2 12 (0.80--180.97) 0.073 16 (0.72--354.80) 0.08 1 12 (0.780--180.97) 0.073 16 (0.72--354.80) 0.08 0 0.01 Enhancement value 0.99 (0.97--1.02) 0.718 Peritumoural infiltration 5.22 (0.24--113.30) 0.293 Multiplicity 3.25 (0.12--81.44) 0.4737 Enlarged retroperitoneal lymph nodes 1.47 (0.29--7.53) 0.646 Values in parentheses are 95% confidence intervals. Pathologic T stage did not significantly correlate with peritumoral infiltration and hydronephrosis grade, respectively ([Table 5](#t5){ref-type="table"}). ###### Pathologic T stage correlation with periureteral infiltration and hydronephrosis grade -------------------- ------------------------------ ---------------------------------------- ----------- ------------------ ----------- ------------ ---------------------------------------------------------------------------- **Reader 1** **Peritumoral infiltration** **Hydronephrosis grade (*****n*****)** **Total** ***p*****value** Pathologic T stage Present Absent 0 1, 2 3, 4 0.194^[*a*](#tb5fn1){ref-type="fn"}^, 0.308^[*b*](#tb5fn2){ref-type="fn"}^ Ta-T1 5 (15.6) 27 (84.4) 4 (12.5) 9 (28.1) 19 (59.4) 32 (100.0) T2 4 (36.3) 7 (63.7) 0 (0.0) 4 (36.4) 7 (63.7) 11 (100.0) T3-4 9 (30.0) 21 (70.0) 2 (6.7) 2 (6.7) 26 (86.7) 30 (100.0) Total 18 (24.7) 55 (75.3) 6 (8.2) 15 (20.6) 52 (71.2) 73 (100.0) **Reader 2** **Peritumoral infiltration** **Hydronephrosisgrade (*n*)** **Total** ***p*value** Pathologic T Stage Present Absent 0 1, 2 3, 4 0.403^[*a*](#tb5fn1){ref-type="fn"}^, 0.173^[*b*](#tb5fn2){ref-type="fn"}^ Ta-T1 1 (3.1) 31 (96.9) 5 (15.6) 7 (21.9) 20 (62.5) 32 (100.0) T2 1 (9.0) 10 (91.0) 0 (0.0) 3 (27.3) 8 (72.7) 11 (100.0) T3-4 6 (20.0) 24 (80.0) 2 (6.7) 2 (6.7) 26 (86.7) 30 (100.0) Total 8 (11.0) 65 (89.0) 7 (9.7) 12 (16.6) 54 (73.7) 73 (100.0) -------------------- ------------------------------ ---------------------------------------- ----------- ------------------ ----------- ------------ ---------------------------------------------------------------------------- Data in parentheses are percentages. *p* value from the *Χ* ^2^ test for correlation of peritumoural intfiltration with pathologic T stage. *p* value from the *Χ* ^2^ test for correlation of hydronephrosis grade with pathologic T stage. ROC curve analysis showed that the best cut-off point of hydronephrosis grade was 2.5 for the prediction of high-grade tumour. The area under the curve (AUC) using the final model was 0.856 for reader 1 and 0.813 for reader 2 ([Figure 3](#f3){ref-type="fig"}). For clinical application in practice, the optimal cut-off grade of hydronephrosis was set at Grade 3, which corresponded to a prediction of high-grade UUC with an AUC of 0.830 and sensitivity and specificity of 88 and 79%, respectively, for reader 1, and AUC of 0.763 and sensitivity and specificity of 86 and 80%, respectively, for reader 2 ([Figure 4](#f4){ref-type="fig"}). {#f3} {#f4} DISCUSSION {#s4} ========== As with most other malignancies, the most accurate independent predictors of prognostic outcome in UUC are tumour stage and grade.^[@b14]^ However, preoperative tumour staging is difficult in UUC because the accuracy of imaging and endoscopic biopsy for T categorization remains unsatisfactory. It is not possible to differentiate a T1 lesion from T2 UUC on CTU, and it is also difficult to obtain representative muscularis tissue with ureteroscopic biopsy. Even though T3 lesions can be characterized by periureteral infiltration, current imaging modalities cannot reliably identify microscopic invasion. Periureteral infiltration, which represents the invasiveness of UUC on CT, can cause overstaging due to additional inflammatory changes, while understaging can occur due to microscopic invasion.^[@b15]^ In our study, there was no significant correlation between periureteral infiltration on CT and tumour grade. In addition, periureteral infiltration on CT did not significantly correlate with pathologic T stage. In clinical practice, tumour grade is a crucial factor in determining whether radical surgery or endoscopic conservative treatment is optimal for UUC, because accurate tumour staging is only available postoperatively based on the pathological evaluation of radical nephroureterectomy specimens. Ureteroscopic evaluation and biopsy definitively set up the diagnosis of UUC and provide fundamental information for risk stratification and clinical management. Several studies have reported that biopsy tumour grade accurately predicts surgical tumour grade in 78--91.6% of patients.^[@b16]--[@b18]^Contrary to these reports, it has been shown that ureteroscopic biopsy performance is inadequate in predicting final pathological grade.^[@b19],[@b20]^ Tumour grade is misinterpreted in more than one third of patients with conservatively managed UTUC,^[@b19]^ and 15% of high-grade tumours are underestimated as low-grade urothelial carcinoma.^[@b20]^ DW-MRI has shown potential as an biomarker in oncological imaging practice, and apparent diffusion coefficient (ADC) values obtained from DW-MRI may help predict tumour invasiveness and metastatic potential of UTUC.^[@b21]^ Some researchers report that high-grade UTUCs have significantly lower ADC values than low-grade tumours.^[@b10],[@b11]^ More recently, however, others have found no significant correlation between ADC value and histological grade of UTUC.^[@b12],[@b13]^ Furthermore, different imaging sequences, parameters, and MRI scanners can cause inconsistency in ADC measurement. Thus, our study aimed to determine whether CTU imaging features reproducible in routine practice could preoperatively predict the histological grade of UUC. There have been a number of studies demonstrating an association between hydronephrosis and advanced clinicopathological features and poor oncologic outcomes in UTUC.^[@b22]--[@b27]^Pyelocaliceal urothelial carcinomas usually do not result in urinary tract obstruction except in tumours involving the ureteropelvic junction. In contrast, UUCs are more likely to have hydronephrosis compared to pyelocaliceal urothelial carcinomas.^[@b4]^ A few studies that focused on UUC alone also reported a predictive role of hydronephrosis in advanced pathological features.^[@b5],[@b28]^ Cho et al found that 86% of patients with hydronephrosis of Grade 3 or 4 had an invasive tumour of T2 stage or greater.^[@b28]^ However, their research was based on various imaging assessments using CT, excretory urography, and renal ultrasonography. In our study, there was no significant correlation between hydronephrosis grade and pathologic T stage. Luo et al reported that hydronephrosis of Grade 2 or higher was associated with non-organ-confined disease,^[@b5]^ although their imaging review was not performed either in consensus or independently, and the specificity was limited to 37.3%. Chung et al assumed that hydronephrosis may cause outward expansion and longitudinal thinning of the already narrow ureter or renal pelvis wall, which may facilitate the seeding of cancer cells to regional or distant organs.^[@b22]^ Even so, the mechanism of the development of hydronephrosis and its relationship with tumour invasiveness is not fully understood.^[@b5]^ To the best of our knowledge, however, our study is the first evaluation of the association between hydronephrosis grade and tumour grade in pure UUC, and adequate diagnostic performance (sensitivity and specificity over 79%) was obtained at a cut-off point of hydronephrosis Grade 3 in the prediction of high-grade tumours. Cho et al reported that the tumour diameter of UUC correlated with pathological T stage and 80% of patients with a tumour diameter of 1.5 cm or greater had invasive UUC.^[@b28]^ In their study, however, tumour diameter was measured on axial CT images and was classified as less than 1.5 cm, greater than or equal to 1.5 cm but less than 2.5 cm, and 2.5 cm or greater. In our study, in which the largest tumour size was measured on multireconstructed images, tumour size did not independently predict tumour grade. In addition, our study showed no significant association between tumour size and hydronephrosis grade. At the time of diagnosis, patients with UTUC and a contralateral normal kidney can be classified as having low-risk UTUC or high-risk UTUC.^[@b29]^ Preoperative clinical factors associated with low-risk UTUC include low-grade ureteroscopic biopsy, low-grade cytology, tumour size \<1 cm, no invasive features on cross-sectional imaging, unifocal disease, and the availability of feasible close follow-up.^[@b29]^ According to the current European guidelines on UTUC,^[@b7]^ diagnostic ureteroscopy with biopsy should be performed in the preoperative assessment of UTUC. On the other hand, the routine use of ureteroscopy is not advocated for the confirmation of UTUC.^[@b30]^ Based on the results of our study, the need for ureteroscopy and biopsy may be obviated in patients with UUC causing hydronephrosis of Grade 3 or higher. The current study has limitations. First, the study population was relatively small due to the rarity of UUC, and because the study was conducted retrospectively at a single institution, the possibility of selection bias should be considered. Prospective multicentre studies with larger sample size are needed to validate our results. Second, the direct imaging-pathological correlation was not obtained in tumour size assessed on CTU. Consequently, tumour size could have been overestimated if there was concomitant inflammation. In conclusion, high-grade hydronephrosis on preoperative CTU was significantly associated with high-grade UUC. The results of the current study may help develop algorithms for risk stratification of patients with pure UUC. Radical surgical treatment should be considered in patients with UUC causing hydronephrosis of Grade 3 or higher regardless of tumour size and absence of peritumoural infiltration on CTU.

Introduction {#S1} ============ Despite significant improvements in outcome,([@R1]--[@R3]) relapse remains the leading cause of treatment failure for children with acute lymphoblastic leukemia (ALL) and occurred in 11 to 36% of those with high-risk B-precursor ALL.([@R4]--[@R10]) Mechanisms by which genomic variation influence relapse risk could involve somatically acquired mutations or inherited genetic variations, which could affect intrinsic resistance to chemotherapy([@R11]--[@R13]) or host pharmacokinetics of anti-leukemic agents.([@R14]--[@R16]) Some studies report that black and Hispanic children with ALL have inferior outcomes to non-Hispanic white children.([@R17]--[@R21]) Reasons for these differences are likely multifactorial, including differences in treatment adherence and access to therapy,([@R22]--[@R24]) in the incidence of favorable and unfavorable presenting features and cytogenetics,([@R25]--[@R27]) and in the frequency of genetic variants affecting pharmacokinetics and pharmacodynamics of antileukemic agents which segregate with ancestry.([@R28]) It remains uncertain whether racial disparities persist with modern intensive ALL regimens. We performed a genome wide association study (GWAS) in a large cohort of children with high-risk B-ALL to identify inherited genetic variations associated with relapse. We performed an analysis adjusting for both treatment and ancestry to identify single nucleotide polymorphisms (SNPs) which increased risk across ancestries (ancestry-agnostic SNPs). Because racial disparities in relapse persisted in this trial, we also performed analyses within each of the three largest ancestral groups (white, black, Hispanic) to identify ancestry-specific variations associated with relapse. We also interrogated relapse SNPs for associations with risk of central nervous system (CNS) relapse, relapse among patients randomized to receive either escalating-dose methotrexate and asparaginase (i.e., Capizzi regimen) or high-dose methotrexate during the first interim maintenance (IM1), and for associations with the pharmacokinetics of antileukemic agents or the intrinsic sensitivity of leukemia cells to chemotherapy. Finally, to assess robustness of relapse SNPs across different therapies, we tested for replication in an independent cohort. Methods {#S2} ======= Patients and treatment {#S3} ---------------------- For the discovery cohort, germline DNA was obtained at remission in children and young adults with newly diagnosed B-precursor ALL enrolled on COG AALL0232 (NCT00075725, <https://clinicaltrials.gov/ct2/show/NCT00075725>).([@R8]) This protocol involved a 2×2 factorial randomization for induction steroid (prednisone ×28 days vs. dexamethasone ×14 days) and interim maintenance 1 regimen (Capizzi escalating-dose methotrexate with pegylated-asparaginase vs. high-dose methotrexate). Exclusion criteria are described in [Figure 1](#F1){ref-type="fig"} and the [Supplementary Methods](#SD1){ref-type="supplementary-material"}. The replication cohort comprised children treated on prior generation protocols who would have met the eligibility criteria of AALL0232 ([Supplementary Methods and Supplementary Table 1](#SD1){ref-type="supplementary-material"}). All studies were approved by the institutional review boards of participating institutions, and all patients and/or guardians provided age appropriate consent/assent in accordance with the Declaration of Helsinki. Genotyping and genetic ancestry {#S4} ------------------------------- Genotyping and genetic imputation was performed as described in the [Supplementary Methods](#SD1){ref-type="supplementary-material"}. Genetic ancestry was defined using STRUCTURE v2.2.3.([@R29]) For categorization of patients into discrete ancestral groups, individuals were classified based on inferred genetic ancestry as white \[Northern European (CEU) \>90%\], black \[West African (YRI) \>70%\], Hispanic \[Native American([@R30]) \>10% and Native American greater than West African\], or Other, including Asian \[East Asian (CHB/JPT) \>90%\]. Quality control steps for both patients and SNPs are detailed in the [Supplementary Methods](#SD1){ref-type="supplementary-material"}. Identification of relapse associated SNPs {#S5} ----------------------------------------- The approaches to perform GWASs for relapse are detailed in the [Supplementary Methods](#SD1){ref-type="supplementary-material"}. GWASs were performed to identify SNPs using an ancestry-agnostic ([Supplementary Table 2 and Supplementary Figure 1](#SD1){ref-type="supplementary-material"}) and an ancestry-specific approach ([Supplementary Figures 2a--c](#SD1){ref-type="supplementary-material"}). Treatment arm and site specific annotation of relapse SNPs {#S6} ---------------------------------------------------------- SNPs associated with relapse were further characterized in subsets of patients based on their IM1 randomization (the Capizzi arm with escalating-dose methotrexate plus pegylated-asparaginase vs. the high-dose methotrexate arm) while adjusting for induction randomization, rapid early response, and ancestry as categorical variables. Additionally, SNPs were tested for their association with CNS relapse (isolated or combined with other sites), with isolated hematologic or other extramedullary relapse treated as competing risks. Significant association thresholds for all analyses were determined by profile information criteria (Ip),([@R31]) which balances false positives and negatives while addressing the effects of multiple testing. Association with orthogonal pharmacologic data {#S7} ---------------------------------------------- SNPs associated with relapse (ancestry-specific or ancestry-agnostic) were evaluated for association with drug resistance in HapMap cells lines (prednisone, asparaginase, mercaptopurine, methotrexate polyglutamate accumulation), primary ALL cells from newly diagnosed patients (prednisone, vincristine, mercaptopurine, asparaginase, *in vivo* leukocyte count decrease following methotrexate), or for association with increased drug clearance (asparaginase allergy, methotrexate clearance, dexamethasone clearance), as described in the [Supplementary Methods](#SD1){ref-type="supplementary-material"}. SNPs were considered supported by orthogonal data if the risk allele for relapse was associated (at P\<0.05) with *in vitro* drug resistance, decreased methotrexate polyglutamate accumulation, smaller leukocyte decrease after methotrexate, more rapid drug clearance, or greater incidence of asparaginase allergy. Evaluation of relapse-associated SNPs in replication cohort {#S8} ----------------------------------------------------------- Relapse-associated SNPs were evaluated in an independent replication cohort (n=719) for their association with relapse using a Cox proportional hazard regression, with patients censored at the time of competing events (i.e. remission death, second malignancy) or last follow-up and adjusting for treatment categorized into 6 groups ([Supplementary Table 2](#SD1){ref-type="supplementary-material"}).([@R5], [@R9], [@R32]) AALL0232 ancestry-agnostic SNPs were evaluated in all patients while adjusting for treatment and ancestry. AALL0232 ancestry-specific SNPs were evaluated in the same ancestry subset of the replication cohort while adjusting for treatment and, in blacks and Hispanics, percent ancestry. The replication cohort SNPs were evaluable if they passed quality control steps as described for the discovery cohort ([Supplementary Methods](#SD1){ref-type="supplementary-material"}). Differences in genotyping platforms between the discovery and replication cohorts, as well as the smaller size of the replication cohort, resulted in 595 of the 1,017 relapse SNPs from the discovery cohort being evaluable in the replication cohort. Validated SNPs were those associated with relapse at P\<0.05 and with identical risk alleles. Quantitative contribution of SNPs to ancestral differences in relapse {#S9} --------------------------------------------------------------------- To identify SNPs which most contributed to ancestry-associated differences in relapse risk, a classification and regression tree analysis was performed separately in blacks and Hispanics considering treatment arm and validated ancestry-agnostic and ancestry-specific SNPs as potential branches. Branches were limited to two levels with each new branch needing to contain at least 20% of the initial ancestral patient group (representing \~1% or at least 22 patients from the discovery cohort for the smallest group, those with black ancestry). The impact of these SNPs on the risk of relapse associated with black or Hispanic ancestry was then evaluated in a competing risk regression model of relapse including the SNPs, treatment, and ancestry. Statistical analysis {#S10} -------------------- Statistical and bioinformatics analyses were performed using R versions 3.2.2, including the "survival", "cmprsk", "rpart", and "forestplot" packages. Association studies of orthogonal phenotypes were performed either in R or PLINK version 1.07. Results {#S11} ======= Patient Characteristics {#S12} ----------------------- Of 3,084 children and young adults enrolled on AALL0232, germline genotype and relapse data were available for 2,652, and 2,225 were included in the GWAS for relapse ([Figure 1](#F1){ref-type="fig"}). To identify covariates to include in the GWAS, we examined the importance of treatment group and ancestry on relapse risk. Consistent with findings in the entire randomized cohort,([@R8]) patients treated with Capizzi-methotrexate had a higher relapse risk than those treated with high-dose methotrexate ([Supplementary Table 3](#SD1){ref-type="supplementary-material"}). Because patients with slow early response did not differ by their induction steroid assignment but did differ by IM1 randomization, patients with slow early response were combined for multivariable and GWAS analyses ([Supplementary Table 2](#SD1){ref-type="supplementary-material"}). Blacks \[P=2.66×10^−4^, hazard ratio (HR)=2.31\] and Hispanics (P=2.17×10^−5^, HR=1.77) had an increased relapse risk compared to whites ([Supplementary Table 3](#SD1){ref-type="supplementary-material"}). The effects of ancestry and treatment groups remained significant in multivariate analyses ([Supplementary Table 3](#SD1){ref-type="supplementary-material"}, [Figure 2](#F2){ref-type="fig"}). Blacks and Hispanics also had a higher risk of any CNS relapse than whites (P=0.016, HR=2.54 for blacks; P=0.0018, HR=2.08 for Hispanics). Association of SNPs with relapse {#S13} -------------------------------- Following quality control steps, 11,180,806 SNPs were evaluated for their association with relapse. A total of 302 SNPs representing 175 unique genetic loci (LD blocks) were associated with relapse in an analysis adjusting for both treatment and percent genetic ancestry (i.e. their association with relapse was "agnostic" to ancestry; [Supplementary Table 4, Supplementary Figure 4](#SD1){ref-type="supplementary-material"}). An additional 715 SNPs representing 424 unique genetic loci were associated with relapse in ancestry-specific analyses, with 280 SNPs (179 loci) associated with relapse in Hispanics, 258 SNPs (167 loci) in blacks, 173 SNPs (72 loci) in whites, 2 SNPs (2 loci) in both blacks and whites, and 2 SNPs (1 locus) in both blacks and Hispanics ([Supplementary Tables 5--7, Supplementary Figures 3, 5--7](#SD1){ref-type="supplementary-material"}). Of the 1,017 relapse SNPs, 192 were associated with relapse in patients treated on the Capizzi arm, 186 in patients treated on the high-dose methotrexate arm, and 18 in both treatment groups; 621 SNPs were not associated with relapse in either group alone but were associated with relapse in the combined cohort ([Supplementary Tables 4--7, Supplementary Figures 4--7](#SD1){ref-type="supplementary-material"}). Of the 302 ancestry-agnostic SNPs, 54 were also associated with an increased risk of CNS relapse ([Supplementary Table 4, Supplementary Figure 4](#SD1){ref-type="supplementary-material"}). Of these, 25 were associated with increased CNS relapse in patients treated on the Capizzi arm, 14 in patients on the high-dose methotrexate arm, and 4 in patients on both arms. Because of the association between ancestry and CNS relapse risk, we evaluated ancestry-specific SNPs for their association with CNS relapse and identified 18 SNPs associated with increased CNS relapse risk in whites, 38 SNPs in blacks, and 52 SNPs in Hispanics ([Supplementary Tables 5--7, Supplementary Figures 5--7](#SD1){ref-type="supplementary-material"}). Because of the importance of minimal residual disease (MRD) in defining high-risk patients,([@R33]) we also evaluated relapse SNPs for their adverse impact in the 1,931 patients with end of induction (day 29) MRD less than 0.1%. 617 SNPs remained significant at the previously defined significance threshold, including 209 ancestry-agnostic SNPs ([Supplementary Tables 4--7, Supplementary Figures 4--7](#SD1){ref-type="supplementary-material"}). Association of relapse SNPs with orthogonal pharmacologic data {#S14} -------------------------------------------------------------- To explore possible mechanisms underlying the 1,017 SNPs associated with relapse, we tested for their association with orthogonal phenotypes including *in vitro* resistance to chemotherapy, decreased response to methotrexate *in vivo*, increased chemotherapeutic drug clearance *in vivo*, and asparaginase allergy *in vivo.* Of the 302 ancestry-agnostic SNPs, 54 were associated with one resistance/clearance phenotype and 10 were associated with more than one such phenotype ([Supplementary Table 4](#SD1){ref-type="supplementary-material"}). Of the 715 ancestry-specific SNPs, 128 were associated with one resistance/clearance phenotype and 32 with more than one phenotype ([Supplementary Tables 5--7](#SD1){ref-type="supplementary-material"}). 36 of the 162 relapse SNPs associated with CNS relapse were associated with at least one resistance/clearance phenotype. Of the 54 relapse SNPs associated with intrinsic leukemic asparaginase resistance (N=24 SNPs) or asparaginase allergy (N=30 SNPs), 20 were associated with relapse in the Capizzi arms, which included additional doses of asparaginase, compared to only eight associated with relapse in the high-dose methotrexate arms (Fisher's P=0.015). In contrast, relapse SNPs associated with decreased intracellular methotrexate polyglutamates (N=15 SNPs), rapid methotrexate clearance (N=19 SNPs), or decreased *in vivo* response to methotrexate (N=42 SNPs) were balanced equally in their association across IM randomization arm (19 of 76 SNPs significant in the Capizzi arm, 13 of 76 significant in the high-dose arm; Fisher's P=0.32). Relapse SNPs were associated with both pharmacokinetic and pharmacodynamic phenotypes. For example, the relapse SNP rs10496350 was associated with asparaginase allergy (which results in decreased exposure to asparaginase), and patients carrying at least one copy of the C risk allele had a higher (P adjusted for treatment and ancestry =2.94×10^−5^) five-year cumulative incidence of relapse (CIR, 37.5%) than did patients with GG genotype (five-year CIR 13.3%) as well as double the risk (P=0.006) of allergy (23.3% for CC or CG genotype vs. 10.8% for GG genotype, [Figure 3](#F3){ref-type="fig"}). Relapse SNPs were also associated with resistance to chemotherapeutic agents: for example, rs743535 (intronic within *CYP2E1*) was associated with both vincristine resistance (median lethal concentration for 50% of cells 0.27 μM for GG genotype vs. 2 μM for GA/AA genotypes, P=0.016) and increased five-year CIR (12% for the GG genotype vs. 20.6% for the GA or AA genotypes, P=2.42×10^−4^, [Figure 4](#F4){ref-type="fig"}). Replication cohort {#S15} ------------------ Of the 1,017 relapse SNPs, 595 were evaluable in the independent replication cohort of 719 patients and 32 replicated (representing 19 loci). Of 138 evaluable ancestry-agnostic SNPs, seven were associated with increased relapse in the replication cohort. 25 of the 457 evaluable ancestry-specific SNPs were also associated with increased relapse in the replication cohort in the same ancestry as was identified in the AALL0232 cohort, including three which increased relapse risk in blacks, 18 in Hispanics, and four in whites ([Table 1](#T1){ref-type="table"}). Of the 32 replicated SNPs, four were associated with an increased relapse in patients treated with high-dose methotrexate, two in patients treated with Capizzi-methotrexate, and two in both cohorts. Of the seven replicated ancestry-agnostic SNPs, four were associated with at least one unfavorable pharmacological phenotype: rs41530849 in *PTPN14* with both rapid methotrexate clearance and *in vitro* asparaginase resistance, rs743535 in *CYP2E1* with *in vitro* vincristine resistance, intergenic SNP rs2463380 with rapid methotrexate clearance, and the missense SNP rs16843643 in *FARP2* with a diminished *in vivo* response to high-dose methotrexate. Additionally, four ancestry-agnostic SNPs and 12 Hispanic-specific SNPs that were associated with CNS relapse in the discovery cohort were replicated in the independent replication cohort, and 23 SNPs were significant among MRD negative patients ([Table 1](#T1){ref-type="table"}). SNP contribution to excess relapse risk in black and Hispanic patients {#S16} ---------------------------------------------------------------------- Using classification and regression trees, we identified two SNPs in blacks (rs4710143 and rs16843643), and in Hispanics (rs9325870 and rs743535) most contributing to their excess relapse risk. In a multivariate model, these four SNPs attenuated the adverse risk associated with black (P=0.79) and Hispanic (P=0.065) ancestry group status ([Figure 5a](#F5){ref-type="fig"}). Additionally, ancestry did not improve the ability to predict relapse if SNPs and treatment group were already known (ANOVA P=0.19 comparing a model with treatment and SNPs as covariates to a model with treatment, SNPs, and ancestry). Patients carrying at least one risk allele for any of the four SNPs had higher relapse than did patients without any risk alleles, regardless of their ancestry ([Figure 5b](#F5){ref-type="fig"}). These variants were less prevalent in whites, with the average white patient carrying 0.21 risk alleles (of a possible eight, range in whites 0--2) compared to a mean of 1.28 in black patients (range 0--5), 0.79 in Hispanics (range 0--4), and 0.63 in patients of other ancestry (range 0--4; Mann-Whitney P\<1×10^−15^). Discussion {#S17} ========== Relapse in high-risk B-ALL remains a significant problem, and most patients who relapse do not survive. Although evaluation of early treatment response and MRD identifies many patients at high risk for relapse, many patients who relapse do not carry these adverse features.([@R33], [@R34]) Further identification of adverse biologic features is needed to allow further refinements in therapy. In this study, we focused on three primary implications of this genetic analysis: whether host genetic variation explained ancestry-related differences in relapse, whether the importance of genetic variation differed by major treatment arms, and how genetic variations were replicated for orthogonal pharmacologic phenotypes and in an independent ALL cohort. In this GWAS, we identified 1,017 SNPs associated with increased relapse risk in children with high-risk B-ALL. We identified both SNPs associated with relapse risk regardless of patient ancestry (ancestry-agnostic) as well as SNPs associated with relapse in an ancestry-specific fashion. Of these relapse SNPs, 7 ancestry-agnostic and 25 ancestry-specific SNPs were also associated with an increased relapse risk in an independent replication cohort ([Table 1](#T1){ref-type="table"}). Importantly, we identified genetic variants associated with increased relapse risk in an ancestry-specific manner across two generations of B-ALL protocols. The identified SNPs contribute to the higher risk of relapse in blacks and Hispanics but also identify patients in each ancestral group at high risk of relapse. Using only four SNPs (rs4710143, rs16843643, rs9325870, and rs743535), we identified 73% of blacks and 57% of Hispanics at high-risk of relapse ([Figure 5b](#F5){ref-type="fig"}). These SNPs were also associated with relapse risk in whites and patients of other ancestry. However, more than 50% of blacks and Hispanics carry at least one risk allele in these SNPs compared to 20% of whites, suggesting the increased relapse risk attributable to these SNPs is disproportionately distributed to blacks and Hispanics, simply on the basis of racial differences in allele frequency. The addition of ancestry group to a model including these SNPs and treatment group failed to improve the model (ANOVA P=0.19) suggesting these SNPs attenuate the adverse impact of ancestry on relapse. These data mirror findings in other malignant([@R35], [@R36]) and non-malignant diseases([@R37]--[@R42]) in which variants strongly associated with ancestry may be the cause of discrepant disease outcomes in different ancestral populations. Such variants offer the opportunity for therapy modification and risk stratification when their effects are stable across multiple settings, as are the replicated ancestry-specific variants identified in this study ([Table 1](#T1){ref-type="table"}). One of the principles of discovery research in pharmacogenomics is that the variants identified in any study will be influenced by the therapy that has been given. Because the randomly assigned methotrexate treatment arm had a significant effect on treatment outcome in AALL0232, we had a unique opportunity to test whether some genomic variants associated with relapse were more important in those receiving one treatment arm (high-dose methotrexate) versus the other (Capizzi methotrexate plus asparaginase). Interestingly, the SNPs directly associated with methotrexate pharmacology did not differentially distribute between the two treatment arms (Fisher's P=0.32), but relapse SNPs associated with asparaginase resistance or asparaginase allergy did cluster in the Capizzi arm (Fisher's P=0.015). Those in the Capizzi arm received more asparaginase but less methotrexate than those in the high-dose methotrexate arm. The association with asparaginase resistance/allergy in the Capizzi arm suggests that asparaginase exposure was more critical to preventing relapse among the patients whose methotrexate exposure was low (Capizzi treatment), and that treatment with high-dose methotrexate diminishes the importance of maximizing asparaginase. Therapeutic and patient differences may also explain differences in the SNPs associated with relapse in this cohort compared to prior analyses. In prior GWAS of ALL relapse risk and MRD,([@R43], [@R44]) the majority of patients were NCI standard-risk, in contrast to the high-risk population studied here. Moreover, all patients in the discovery cohort of this study also received delayed intensification and MRD-directed therapy intensification, whereas many patients in the prior GWASs([@R43], [@R44]) did not. In a review of the SNPs previously associated with relapse or MRD,([@R43], [@R44]) we identified five (rs35229355, rs7517671, rs10883699, rs7350429, and rs6773449) that associated with relapse (P\<0.05) after adjusting for both treatment and ancestry in the current discovery cohort. However, these SNPs did not reach the Ip selected P value threshold, nor were they replicated at least 20 times during iterative resampling. This finding highlights the importance of population and therapeutic differences on the association of pharmacogenomic variants and outcome. It is encouraging that many of the SNPs identified in the current GWAS were associated with relapse among patients treated on both the high-dose methotrexate and the Capizzi escalating-methotrexate/asparaginase arms, suggesting that some of these variants may be prognostic across therapies. The analysis of relapse SNPs' association with orthogonal pharmacologic phenotypes suggests mechanisms through which some relapse SNPs may be exerting their effects on relapse risk. Relapse SNPs were associated with both pharmacokinetic and pharmacodynamic phenotypes. For example, rs6786341 (an intronic variant in lactoferrin) was associated with more rapid methotrexate clearance, a phenotype which has previously been associated with decreased methotrexate polyglutamate accumulation and increased relapse risk.([@R45], [@R46]) The rs743535 variant in *CYP2E1* was associated with resistance to vincristine ([Figure 4](#F4){ref-type="fig"}). Variants in this gene have previously been implicated in inferior survival in non-Hodgkin's lymphoma([@R47]) and non-small cell lung cancer,([@R48]) potentially due to resistance to chemotherapeutic agents used in those diseases. Variants near *LZTS1*, which include promoter and enhancer marks in neural tissues,([@R49]) were associated with CNS relapse in Hispanics. Suppression of this gene has previously been implicated in metastatic potential in multiple solid tumors,([@R50]--[@R52]) suggesting these variants may alter leukemic trafficing into the CNS, thereby altering CNS relapse risk. Other identified CNS relapse SNPs likely contribute to CNS relapse through alterations in leukemic drug resistance or rapid drug clearance, as 36 of 162 CNS relapse SNPs were also associated with pharmakokinetic or drug resistance phenotypes. We identified several novel inherited risk variants for relapse in a large population of children with high-risk B-precursor ALL. Several of these are associated with the increased relapse risk specific to black and Hispanic ancestry and may contribute to the adverse outcomes attributed to "race." Many of these variants are associated with "inherited" leukemic resistance or rapid clearance of chemotherapy. These findings may allow personalized therapy to further improve outcomes for children with high-risk B-ALL. Supplementary Material {#S18} ====================== **Funding/Support** The work was supported by the National Institutes of Health \[grant numbers GM 92666, GM 115279, CA142665, CA 21765, CA 36401, CA98543 (COG Chair's grant), CA98413 (COG Statistical Center), CA114766 (COG Specimen Banking), U01-HG04603, RC2- GM092618, R01-LM010685, 5T32-GM007569\]; Leukemia Lymphoma Society (grant number 6168); and by the American Lebanese Syrian Associated Charities. **Role of funding source** The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. **Original Data Statement** Drs. Mary Relling and Seth Karol had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. **Authors' contributions** MVR and JJY contributed to the conception and design of the study. EL, LBR, CAF, JRM, SWP, RJA, ELL, BD, SJ, C-HP, EAR, NJW, WLC, SPH, MLL, MD, WEE, JJY, and MVR contributed to the provision of study materials, patient recruitment, or acquisition of data. SEK, CC, XC, and MVR contributed to data analysis and interpretation. All authors contributed to the drafting and reviewing of the manuscript and gave their final approval to submit for publication. **Conflicts of interest** The authors declare no competing financial interests. **Data Availability:** Detailed information on the primary clinical trial (COG AALL0232) for the discovery cohort is available from: <https://clinicaltrials.gov/ct2/show/results/NCT00075725?term=0232&rank=3> {#F1} ![Association of non-white genetic ancestry with increased relapse risk\ Non-whites had an increased risk of relapse in the discovery cohort. The five-year cumulative incidence of relapse was higher in blacks \[23.7%, 95% confidence interval (CI) 14.7--32.7%, P=2.27×10^−4^, HR=2.32\] and Hispanics (19.3%, 95% CI 15.7--22.9%, P=8.23×10^−5^, HR=1.7) than whites (10.3%, 95% CI 8.9--12.8%). P values are adjusted for treatment.\ White: \>90% CEU; black: \>70% YRI; Hispanic: \>10% Native American and Native American \>YRI](nihms842426f2){#F2} {#F3} {#F4} ###### Relapse SNPs attenuate the adverse impact of black and Hispanic ancestry a: **Forest plot of relapse risk comparing multivariable models with and without four relapse SNPs** b: **Presence of a risk allele in any of the four SNPs confers high relapse risk regardless of ancestry** Risk alleles in any of four SNPs (rs4710143, rs16843643, rs9325870, and rs743535) confer increased relapse risk regardless of ancestry. (A) In multivariate models, these SNPs largely attenuate the adverse effect of black or Hispanic ancestry, while leaving unchanged the association between treatment arm and relapse. Treatment arms are described in [Supplementary Table 2](#SD1){ref-type="supplementary-material"}: For the rapid early response patients, Dex/Capizzi, Pred/Capizzi, Dex/HD, Pred/HD refer to the induction steroid (dex = dexamethasone, pred = prednisone) and the interim maintenance (Capizzi=escalating dose methotrexate plus asparaginase, HD = high-dose methotrexate). For the slow early response patients (SER), induction steroid groups were combined. (Hazard ratio (HR) from model without SNPs shown in blue, models with SNPs shown in red). \(B\) Whites, blacks, and Hispanics carrying risk alleles for any of these SNPs (dashed lines) have higher five-year relapse risks than do those without any risk alleles (solid lines) \[15.3% vs. 9.7% (P=0.025) for whites, 32.3% vs. 0% (P=1.28×10^−4^) for blacks, and 25.5% vs. 10.7% (P=3.72×10^−6^) for Hispanics\]. P values are adjusted for treatment.   ###### SNPs associated with relapse in discovery (n=2,225) and replication (n=719) cohorts ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- rsID Gene Risk allele RAF P: discovery cohort Hazard Ratio (95% CI): discovery cohort P: replication cohort Ancestry Additional phenotypes -------------------- ----------- ------------- ------- --------------------- ----------------------------------------- ----------------------- ---------- ------------------------------------ rs41530849 *PTPN14* T 0.006 4.26E-06 3.87\ 0.019 agnostic HD arm, MTX clearance, PPL ASP (2.17--6.89) **rs10205940** A 0.224 6.85E-06 1.42\ 0.044 agnostic HD arm, Capizzi arm, CNS (1.22--1.66) **chr23: 9863426** *SHROOM2* T 0.008 1.04E-05 2.45\ 0.049 agnostic HD arm, Capizzi arm, CNS (1.64--3.64) rs2463380 G 0.22 3.98E-05 1.52\ 0.045 agnostic HD arm, CNS, MTX clearance (1.24--1.86) **rs2710418** *NELL2* T 0.031 4.99E-05 1.98\ 0.021 agnostic HD arm (1.42--2.76) rs743535 *CYP2E1* A 0.124 5.00E-05 1.54\ 0.045 agnostic PPL Vinc (1.25--1.9) **rs16843643** *FARP2* C 0.012 0.000226 2.95\ 0.031 agnostic Capizzi arm, CNS, MTX WBC response (1.66--5.25) **rs775491** *BEST3* A 0.304 0.000265 1.61\ 0.0086 white (1.24--2.07) rs156008 *PCSK1* A 0.158 0.000297 1.65\ 0.024 white (1.26--2.16) **rs4710143** *RNASET2* G 0.074 0.000579 4.92\ 0.014 black Capizzi arm (1.98--12.2) **rs202408** C 0.144 0.000789 3.56\ 0.021 black (1.7--7.49) **rs7860525** T 0.134 0.00175 2.79\ 0.016 black (1.47--5.31) **rs9325870** *LZTS1* C 0.205 1.84E-05 2\ 0.036 Hispanic CNS (1.46--2.75) rs16999479 *DSCAM* G 0.016 0.000219 4.02\ 0.046 Hispanic CNS (1.92--8.42) **rs141707566** *GRIN2A* C 0.014 0.000289 2.76\ 0.037 Hispanic HD arm (1.59--4.77) **rs12535024** *DDC* C 0.181 0.00103 1.76\ 0.0499 Hispanic (1.26--2.47) **rs6786341** *LTF* T 0.012 0.00173 4.14\ 0.038 Hispanic MTX clearance (1.7--10.1) rs16945138 *DNAH9* T 0.007 0.00186 7.5\ 0.014 Hispanic (2.11--26.7) rs6651255 *GSDMC* C 0.425 0.00222 1.59\ 0.0029 Hispanic (1.18--2.13) ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- RAF: Risk allele frequency; CI: confidence interval Characteristics of validated SNPs are shown for the discovery cohort, with one SNP for each locus shown (SNPs removed through LD pruning are shown in [Supplementary Tables 4--7](#SD1){ref-type="supplementary-material"}). Bolded SNPs were significant at the Ip determined significance threshold when evaluated among patients who were end-induction minimal residual disease negative. SNPs are ordered by ancestry of discovery, with SNPs associated with relapse while adjusting for both treatment and ancestry (i.e. "ancestry agnostic") labeled as agnostic and ancestry-specific SNPs labeled with their associated ancestry group. Additional phenotypes include: association with relapse among patients treated on either first interim maintenance arm \[Capizzi arm, HD (high-dose methotrexate) arm\], association with CNS relapse (CNS), as well as association with *in vitro* resistance among primary patient lymphoblasts to asparaginase (PPL ASP) or vincristine (PPL Vinc), more rapid methotrexate clearance (MTX clearance), or diminished white blood cell decrease after *in vivo* methotrexate treatment (MTX WBC response).

Introduction {#S0001} ============ The rotator cuff tear is the most common tendinopathy in humans and over 200,000 cuff repairs are performed annually in the United States \[[1](#CIT0001),[2](#CIT0002)\]. The decreased morbidity associated with arthroscopic repairs has contributed to the popularity and broad indications for this surgical intervention \[[1](#CIT0001),[3](#CIT0003)\]. Tendon reattachment even if biomechanically strong at the time of repair often fails and approximately 50% of patients with full-thickness tears of the rotator cuff report symptoms at 6 months after surgery \[[1](#CIT0001),[4](#CIT0004),[5](#CIT0005)\].10.1080/21623945.2019.1609201-F0005Figure 5.Schematic representation of the changes in the number and cross-section area of fat clumps and of adipocyte number in the proximal, medial and distal SSP muscle after a complete SSP tendon detachment. IMF increased closer to the tendon tear compared to the proximal SSP muscle. Detached muscles had more clumps in the distal and medial sections and of larger size in the distal section. There were more adipocytes in the distal and medial detached SSP muscles compared to proximal and cross-sectional area was smaller in the distal SSP muscle. The fat clumps are represented by ovals and adipocytes by smaller filed black shapes. Results from the statistical analysis are indicated: 0.001 ≤ P \< 0.01 (\*), 0.0005 ≤ P \< 0.001 (\*\*), P \< 0.0005 (\*\*\*). The unsatisfactory success of rotator cuff repair surgeries has been attributed in many cases to muscle atrophy and fat accumulation both assessed by medical imaging methods \[[6](#CIT0006)--[8](#CIT0008)\]. The benefits of arthroscopy to repair the cuff and of advanced imaging methods to measure rotator cuff muscle fat content are undeniable but enhancing postoperative outcomes remain a challenge and basic knowledge on the mechanisms of intramuscular fat accumulation is needed \[[9](#CIT0009)--[11](#CIT0011)\]. Animal models of rotator cuff tendon injury and repair capture important aspects of the human disease \[[12](#CIT0012)--[16](#CIT0016)\]. Imaging of the rabbit's SSP muscle documented fat accumulations both extra- and intra-muscular, and were evident as early as 4 weeks after SSP tendon detachment and progressed up to 12 weeks \[[17](#CIT0017)\]. The fat signal increased from proximal-to-distal with the highest amount of fat detected in the distal quarter of the SSP muscle, the site closer to tendon detachment \[[17](#CIT0017)\]. Both fat accumulation and muscle atrophy were present at week 1 and 2 after immediate repair but only fat accumulation persisted at 6 weeks \[[18](#CIT0018),[19](#CIT0019)\]. In a different study, delayed tendon reattachment did not reverse SSP fat accumulation \[[20](#CIT0020)\]. The rabbit experimental model of rotator cuff tear and repair reproduced accurately the human pathology and represents a valuable avenue to decipher the pathophysiology of IMF accumulation associated with rotator cuff tear \[[12](#CIT0012),[21](#CIT0021)\]. The mechanisms for adipose tissue expansion have been studied extensively. In the context of obesity resulting from a high-fat diet, large fat accumulations are noticeable in subcutaneous and visceral deposits \[[22](#CIT0022)--[24](#CIT0024)\]. Overnutrition induced adipocyte hypertrophy in upper-body subcutaneous fat while a cycle between hypertrophy and hyperplasia characterized deposits below the waist \[[25](#CIT0025)\]. The IMF deposit, considered a small fat deposit, is made up of white adipocytes and its accumulation characterizes late stages of muscular dystrophies \[[24](#CIT0024),[26](#CIT0026)\]. The pathophysiology of adipocytes leading to IMF accumulation associated with rotator cuff tear remains unknown. We hypothesized that IMF accumulation observed after rotator cuff tears results from adipocyte hypertrophy rather than hyperplasia leading to the enlargement of resident muscle fat clumps. The purpose of the current study was to characterize, at the microscopic level and over time, the expansion of the adipose tissue in the SSP muscle of rabbits after detachment of the distal SSP tendon. Materials and methods {#S0002} ===================== Animals and surgical procedure {#S0002-S2001} ------------------------------ This study was approved by the University of Ottawa Animal Care Committee. Adult female New Zealand rabbits (n = 45) weighing 3.0 kg were purchased from Charles River, Saint-Constant, Quebec, Canada and allowed to acclimate for one week upon arrival. For the experimental group, a supraspinatus tenotomy was performed unilaterally in 30 rabbits by sectioning completely the SSP tendon from the greater tuberosity of the humerus using a surgical blade under general anaesthesia \[[14](#CIT0014)\]. Left and right shoulders were alternated. To prevent postoperative adhesions, the stump of the tendon was wrapped with a polyvinylidene membrane (5 µm, Durapore, Millipore, Bedford MA USA). Animals were housed individually, divided into three equal groups, killed at 4, 8 or 12 weeks after surgery and the operated shoulders were collected for histological analysis. For the control group, 15 unoperated rabbits were equally divided into three groups, killed at 4, 8 and 12 weeks and both shoulders were collected. The harvesting method of shoulders was described in our previous publication. Complete SSP muscles were dissected from the scapula, wrapped and frozen at −20°C until processed for histology analysis \[[17](#CIT0017)\]. Radiology and macroscopic data on this group of animals have already been reported [\[17\];](#CIT0017) the current microscopy analysis at the cellular level builds on those studies. Histology specimen preparation {#S0002-S2002} ------------------------------ Harvested SSP muscles were fixed in 4% paraformaldehyde and rinsed twice for 1 h in phosphate buffered saline to begin processing for histology. Muscle specimens were frozen to preserve fat structures during sectioning. From each muscle, three cross-section slices of 1-mm thickness were cut at the proximal quarter, middle-half, and distal quarter sites of the supraspinatus muscle. Muscle slices were stained for 2 weeks with 5% potassium dichromate and 2% osmium tetroxide followed by paraffin embedding \[[14](#CIT0014)\]. Using a microtome, 6µm-thick microscopy slides were prepared. Fixation in osmium tetroxide stained adipocytes black. Histology evaluation and microscopy image analysis {#S0002-S2003} -------------------------------------------------- A total of 180 slides from detached tendons and from unoperated tendons, at time points 4, 8 or 12 weeks, in the proximal, middle or distal quarters of the SSP muscle were analysed by light microscopy ([Table 1](#T0001)).10.1080/21623945.2019.1609201-T0001Table 1.Summary of the samples studied including numbers of rabbits, shoulders, and tissue sections for both fat clump and adipocyte analyses.SSP Muscle Quarter/WeeksDetached vs ControlRabbits (N)Shoulders\ (N)Muscle Sections (Clumps)\ (N)Muscle Sections (Cells) (N)Proximal Quarter4Detached1010810Control51010108Detached10101010Control510101012Detached1010910Control5101010Middle Quarter4Detached10101010Control5109108Detached1010810Control510101012Detached1010910Control5101010Distal Quarter4Detached10101010Control51010108Detached10101010Control510101012Detached10101010Control5101010 **Fat ClumpsAdipocytes**Total Muscle Sections/Fields AnalyzedDetached8490 (270 fields)Control8990 (270 fields)Total Fat Clumps/Adipocytes AnalyzedDetached18,54210,389Control14,3456706 Fat clumps were measured on entire SSP muscle cross-sections digitized at 6.7x magnification and backgrounds were cropped using Corel Photo-Paint 11. Images were then imported for computer-assisted quantitative image analysis using software ImageJ software (version 1.34s; National Institute of Health, Bethesda, MD, USA). Scales were set by using calliper measurements of two reference points on the slide and converted to pixels. Pictures were converted into binary black and white images (8-bit; grey scale). A fat clump was defined as an area of fat stained black, not in contact with another stained area ([Figure 1](#F0001)). The 'threshold' function was manually adjusted to select only black pixels. The 'watershed' function was used to mark the boundaries of individual fat clumps. The 'analyse particle' command was used to measure clump numbers and areas with 'cellularity' set at 0--1 and 'size' set at 0-infinity. The command 'measure all' was used to automatically generate all measurements.10.1080/21623945.2019.1609201-F0001Figure 1.Representative micrographs of IMF accumulation in the distal quarter of the SSP muscle cross-sections. (a). SSP muscle sections at 4, 8 and 12 weeks after tendon detachment. (b). SSP muscle sections in control animals at the same time points. IMF was stained using osmium tetroxide and is visible at black-stained areas. Note the higher accumulation of fat in the tendon detached group compared to controls at all time points studied. Original magnification at 6.7x. Adipocyte number per field and average cross-sectional area were measured using computer-assisted image analysis of the same microscopic slices captured at 25x magnification. Three different fields of equal and fixed areas (0.149 mm^2^ each) were chosen using the following criteria: included black staining, not contiguous with the other selected fields, included minimal empty space, and included at least one blood vessel. No field overlapped. The three different fields analysed in each 3 muscle sections (proximal, middle and distal) in 10 rabbits per time point (4, 8 and 12 weeks) group in each of detached SSP and control groups amounted to a total of 540 fields. Representative images from distal quarters at 4, 8 and 12 weeks after tenotomy and corresponding controls are presented in [Figure 2](#F0002). To measure adipocyte number and size, we again used ImageJ, images were converted to 8-bit grey-scale pictures. Default settings of the 'thresholding' function were used to select only the black-stained adipocytes. Applying the threshold converts the image to black and white, displaying only adipocytes. The 'watershed' function was used to separate individual cells, and 'analyse particles' was used to count and measure the cross-section area of adipocytes. Minimum size was set at 350 pixels (to remove artefacts originating from microtomy) and circularity of 0.5 (to remove any cells cut off at the edges of the picture). ImageJ was calibrated by using a scale bar to convert pixels into mm^2^.10.1080/21623945.2019.1609201-F0002Figure 2.Representative micrographs of adipocytes in the proximal, middle and distal quarters of the SSP muscle. (a). SSP muscle sections at 12 weeks after tendon detachment. (b). SSP muscle sections from control age-matched animals. Adipocyte vacuoles stained black using the osmium tetroxide protocol described in the Method section. Note the increased number of smaller adipocytes in the tendon detached group compared to controls. Original magnification at 25x. Data and statistical analysis {#S0002-S2004} ----------------------------- Descriptive statistics displayed the medians and interquartile ranges of the four outcomes measured; the number and cross-sectional area of both fat clumps and adipocytes. We first explored the distribution of the fat clumps and adipocytes outcomes because a skewed distribution of adipocytes diameter had previously been described \[[27](#CIT0027)\]. Our data for fat clumps and for adipocytes confirmed an asymmetrical distribution ([Figures 3](#F0003) and [4](#F0004)); median lines were off centre of the interquartile boxplots and upper and lower whiskers for the same box were of different size indicative of a skewed distribution. Skewed data distribution was log transformed to meet the normality assumptions for ANOVA and regression-based statistical analyses. In this paper, non-log-transformed data were reported as descriptive statistics whereas log-transformed data were used in the quantitative statistical analyses. Linear mixed-effects (LME) models were fitted to the data, and statistical significance for the four outcomes was evaluated by ANOVA considering the different muscle locations as a single fixed-effect factor and similarly for the three time points after detachment, but using a random effect to account for correlation between measurements taken from the same rabbit. Post hoc analysis was conducted when significant differences were observed also using LME models, and pairwise comparisons of the fixed effects were performed. The fixed effects were introduced as a single term in the equations and considered without interaction. For each of the four outcomes evaluated; fat clumps number, fat clumps cross-sectional area, adipocyte number and adipocytes cross-section area, the following equation was applied: Outcome \~ log(μ) + β~1~ time + β~2~ location + β~3~ detachment + error (1\|rabbit).10.1080/21623945.2019.1609201-F0003Figure 3.Boxplots showing the distribution of intramuscular fat clump numbers (a) and cross-sectional area (b) (mm^2^) for SSP tendons detached for 4, 8 and 12 weeks and for age-matched controls. Horizontal lines in the boxes represent the median values, limits of the boxes represent upper and lower quartiles, lines extending vertically from boxes represent variability outside the boxes and outliers are plotted as individual points. The dispersion of the number of fat clumps was similar for both detached and controls. A large variability in the fat clump cross-sectional areas was observed for the detached group in the distal quarter at 8 and 12 weeks after detachment and displayed in the large sizes of the boxes for these two groups compared to controls.10.1080/21623945.2019.1609201-F0004Figure 4.Boxplots showing the distribution of intramuscular adipocyte numbers (a) and cross-sectional area (b) (mm^2^) for SSP tendons detached for 4, 8 and 12 weeks and for age-matched controls. Horizontal lines in the boxes represent the median values, limits of the boxes represent upper and lower quartiles, lines extending vertically from boxes represent variability outside the boxes and outliers are plotted as individual points. The dispersion in the number of adipocytes was larger in the middle and distal quarters of the SSP muscle in the detached group compared to controls. The dispersion in adipocyte cross-section area was comparable in both groups. LME modelling also accounted for two characteristics of our study design with potential influence on the outcomes of the statistical analyses \[[28](#CIT0028)\]. First, the fat clumps and adipocytes were measured in three quarters of the same SSP muscle and are not independent observations. Second, number and cross-sectional area outcomes are potentially influenced by a random effect corresponding to animals and by fixed effects including SSP tendon detachment, quarter of the muscle, and time after SSP tendon detachment. P-values for the calculated coefficient of individual fixed effect estimates were used to determine their contribution to fat accumulation. Significance was determined according to P values at: 0.001 ≤ P \< 0.01 (\*), 0.0005 ≤ P \< 0.001 (\*\*), P \< 0.0005 (\*\*\*). We considered p \< 0.01 to be statistically significant because of the multiple outcomes and models analysed. All descriptive and statistical analyses were performed using the open-source programming environment R \[[29](#CIT0029)\] and the lmerTest package \[[30](#CIT0030)\]. Results {#S0003} ======= [Table 1](#T0001) describes the samples analysed including numbers of rabbits, shoulders, tissue sections and fields in detached and control SSP muscles. Seven out of 360 slides showed poor staining quality in some areas and were omitted from the low magnification microscopy analysis (fat clumps) ([Table 1](#T0001)). The total number of stained fat clumps was 18,542 for the detached groups and 14,345 for the control groups. The total number of adipocytes analysed was 10,389 for the detached group and 6706 for the control group. Representative micrographs of osmium tetroxide stained SSP muscle sections and of fat clumps and adipocytes are presented in [Figures 1](#F0001) and [2](#F0002). Descriptive statistics of fat clump numbers and areas {#S0003-S2001} ----------------------------------------------------- The average number (± standard error) of fat clumps for all quarters of all the detached SSP muscles was 223.1 ± 87.5 and for all the control muscles 160.8 ± 70.5. Average fat clump areas were 0.031 ± 0.011 mm^2^ for detached SSP muscles and 0.013 ± 0.023 mm^2^ for controls ([Figure 3](#F0003)). Quantitative analysis of fat clumps after SSP tendon detachment {#S0003-S2002} --------------------------------------------------------------- SSP tendon detachment was associated with increased fat clump numbers (P \< 0.001) and area (P \< 0.0005) in SSP muscles compared to controls ([Table 2](#T0002)). Time after SSP tendon detachment did not significantly influence fat clump numbers and area (both at P \> 0.01). The muscle location (distal, middle or proximal) was strongly associated with increases of fat clumps number (P \< 0.0005) and area (P \< 0.0005) ([Table 2](#T0002)). There were significantly more fat clumps in the distal quarter compared to the proximal quarter (P \< 0.0005) but not significantly different from the medial quarter (P \> 0.01). The proximal quarter contained fewer fat clumps compared to the medial quarter (P \< 0.0005). Fat clumps were significantly larger in the distal quarter compared to the medial (P \< 0.001) and proximal quarters (P \< 0.0005) while proximal and medial quarters were not significantly different (P \> 0.01) ([Table 2](#T0002)).10.1080/21623945.2019.1609201-T0002Table 2.Summary of ANOVA and of the post hoc linear mixed-effects model for the fat clump number and cross-section area. 0.001 ≤ P \< 0.01 (\*), 0.0005 ≤ P \< 0.001 (\*\*), P \< 0.0005 (\*\*\*).VariableANOVA (P Value)DetachmentP \< 0.001 (\*\*)WeekP \> 0.01LocationP \< 0.0005 (\*\*\*) **Fat clump number \~ log(μ) + β~1~ time + β~2~ location + β~3~ detachment + error (1\|rabbit)β coefficientStd ErrorP Value**Distal vs Medial−0.1150.077P \> 0.01Distal vs Proximal−0.7870.076P \< 0.0005 (\*\*\*)Proximal vs Medial0.6720.076P \< 0.0005 (\*\*\*)Fat Clump Cross Section AreaVariable**ANOVA (P Value)**DetachmentP \< 0.0005 (\*\*\*)WeekP \> 0.01LocationP \< 0.0005 (\*\*\*) **Fat clump area \~ log(μ) + β~1~ time + β~2~ location + β~3~ detachment + error (1\|rabbit)β coefficientStd ErrorP Value**Distal vs Medial−0.2870.084P \< 0.001 (\*\*)Distal vs Proximal−0.3990.082P \< 0.0005 (\*\*\*)Proximal vs Medial0.1110.083P \> 0.01 Descriptive statistics of adipocyte numbers and areas {#S0003-S2003} ----------------------------------------------------- Detached SSP muscles had on average 38.5 ± 11.7 adipocytes per field of view compared to 24.8 ± 4.8 for controls. Average adipocyte cross-sectional area was 0.0020 ± 0.0003 mm^2^ for detached SSP muscles compared to 0.0016 ± 0.0003 mm^2^ for controls ([Figure 4](#F0004)). Quantitative analysis of adipocytes after SSP tendon detachment {#S0003-S2004} --------------------------------------------------------------- SSP tendon detachment was associated with increased adipocyte numbers (P \< 0.0005) and cross-section area (P \< 0.01) in SSP muscles compared to controls ([Table 3](#T0003)). Time after SSP tendon detachment significantly increased adipocytes number (P \< 0.01) but had no significant influence on adipocytes cross-section area (P \> 0.01). Detached SSP muscles had significantly more adipocytes at week 12 (P \< 0.01) compared to week 4 ([Figure 4](#F0004)). The number of adipocytes was not significantly different between week 4 and week 8 (P \> 0.01) and between week 8 and week 12 (P \> 0.01). Muscle location (distal, middle or proximal) was associated with increased adipocyte numbers (P \< 0.0005) and cross-section areas (P \< 0.0005) ([Table 3](#T0003)). There were significantly more adipocytes in the distal quarter (P \< 0.0005) compared to the proximal quarter but not compared to the medial quarter (P \> 0.01). The medial quarter contained more adipocytes than the proximal quarter (P \< 0.0005). Adipocytes were significantly smaller in the distal quarter (P \< 0.01) compared to the medial and to the proximal quarters (P \< 0.0005) ([Table 3](#T0003) and [Figure 4](#F0004)). Adipocyte in the medial quarter was also smaller than in the proximal quarter (P \< 0.01).10.1080/21623945.2019.1609201-T0003Table 3.Summary of ANOVA and of the post hoc linear mixed-effects model of adipocyte number and cross-section area. 0.001 ≤ P \< 0.01 (\*), 0.0005 ≤ P \< 0.001 (\*\*), P \< 0.0005 (\*\*\*).VariableANOVA (P Value)DetachmentP \< 0.0005 (\*\*\*)WeekP \< 0.01 (\*)LocationP \< 0.0005 (\*\*\*) **Adipocyte number \~ log(μ) + β~1~ time + β~2~ location + β~3~ detachment + error (1\|rabbit)β coefficientStd ErrorP Value**Week 4 vs 80.1470.072P \> 0.01Week 4 vs 120.2390.072P \< 0.01 (\*)Week 12 vs 8−0.0910.072P \> 0.01Distal vs Medial−0.090.057P \> 0.01Distal vs Proximal−0.4930.057P \< 0.0005 (\*\*\*)Proximal vs Medial0.4030.057P \< 0.0005 (\*\*\*)Adipocyte Cross Section AreaVariable**ANOVA (P Value)**DetachmentP \< 0.01 (\*)WeekP \> 0.01LocationP \< 0.0005 (\*\*\*) **Adipocyte area \~ log(μ) + β~1~ time + β~2~ location + β~3~ detachment + error (1\|rabbit)β coefficientStd ErrorP Value**Distal vs Medial0.1170.041P \< 0.01 (\*)Distal vs Proximal0.2560.041P \< 0.0005 (\*\*\*)Proximal vs Medial−0.1380.041P \< 0.01 (\*) Discussion {#S0004} ========== We characterized intramuscular fat accumulation in the SSP muscle in the rabbit model of rotator cuff tear. SSP tendon detachment produced an increase number of larger fat clumps and an increased number of smaller adipocytes in the distal quarter of the SSP muscle near the site of tendon tear ([Figure 5](#F0005)). Time after tendon detachment significantly increased the number of adipocytes. Our hypothesis based on literature on obesity that: IMF accumulation after rotator cuff tears resulted from adipocyte hypertrophy rather than hyperplasia was infirmed. The current study established that adipocyte hyperplasia was the main contributor to fat clump enlargements and explained SSP IMF expansion up to 12 weeks after tendon detachment. Fat tissue has been described to expand via two mechanisms; adipocyte hyperplasia (the increase of the number of adipocytes) and adipocyte hypertrophy (the increase in individual adipocyte size) \[[24](#CIT0024),[31](#CIT0031)\]. Knowledge of adipocytes' behaviour originates mostly from obesity research and expansion of sub-cutaneous white fat deposits. Experiments from the 1970s showed that overfeeding combined with reduced energy expenditure over several months resulted in an important increase of adipocyte size without significant changes in the number adipocytes \[[32](#CIT0032)\]. Consistently, the turnover of human subcutaneous adipocytes each year is very low, approximately 8%, resulting in little change in adipocyte number and emphasizing the importance of adipocyte hypertrophy in the expansion of fat tissue in the context of obesity \[[33](#CIT0033)\]. There is evidence for regional differences in adipocytes behaviour in human obesity. While adipocyte hypertrophy characterizes upper body sub-cutaneous fat, adipocytes cycling between hyperplasia and hypertrophy characterized deposits below the waist as obesity progresses upon high-fat feeding \[[23](#CIT0023),[25](#CIT0025),[34](#CIT0034)\]. Our results indicate that expansion of IMF in the SSP muscle present significant similarities with subcutaneous fat deposits located below the waist; fat expansion resulted from adipocyte hyperplasia at least within the first 12 weeks after tendon detachment. Intramuscular adipocytes in the current study were approximately 0.002 mm^2^ or 25 microns in diameter (assuming circular shape of adipocytes, πr^2^). This was smaller than mature white adipocytes with approximately 110 microns in diameter (ranges from 20 to 300 microns) \[[27](#CIT0027),[35](#CIT0035)\]. Smaller adipocytes less than 10 microns in diameter were previously described in rat epididymal fat deposit \[[27](#CIT0027),[36](#CIT0036)\]. Considering the published spectrum of sizes for white adipocytes, intramuscular adipocytes were therefore characterized as small in both healthy and detached SSP muscles. Increased adipocyte cellularity is indicative of the mechanism of adipogenesis taking place in the detached SSP muscle. The observations of increased adipocyte number in combination with small cross-section areas in the distal quarter where fat accumulation was the most important suggested the presence of newly formed cells. Pre-adipocytes are smaller in size than mature adipocytes \[[37](#CIT0037)--[39](#CIT0039)\]. Newer adipocytes of smaller size driving the average adipocyte size lower are a potential explanation of the lower adipocyte size in the distal SSP muscle. Moreover, the persistence of smaller average size adipocytes 12 weeks after tendon detachment suggests that, rather than maturing and growing to reach proximal size, new adipocytes that were present 4 weeks after detachment have remained small or new adipocytes were continuously generated in the distal detached SSP muscle. The identity of the precursor cells contributing to the increased intramuscular adipocyte hyperplasia is actively investigated. Adipocytes derive from pre-adipocytes which themselves differentiate from mesenchymal precursor cells \[[39](#CIT0039)\]. Adipocytes can also originate from existing mesenchymal tissue in the muscle \[[37](#CIT0037)\]. Four candidate muscle cells able to generate adipocytes have been described; a population of fibrocyte/adipocyte progenitors, muscle satellite cells, pericytes \[[35](#CIT0035)\] and bone marrow-derived cells \[[40](#CIT0040)\]. During skeletal muscle degeneration, adipocytes were demonstrated to derive from a population of bipotent progenitors residing within muscles and different from muscle progenitors \[[38](#CIT0038),[40](#CIT0040)\]. There is experimental support in mice for these cells as the source of SSP muscle adipocytes after rotator cuff tear \[[41](#CIT0041),[42](#CIT0042)\]. Satellite cells are also a population of primary cells residing in muscles with the ability to differentiate into adipocytes *in vitro*. [\[43](#CIT0043)--[45\]](#CIT0045) Fibrocyte/adipocyte progenitor and satellite cells may be activated locally in the distal quarter of the SSP muscle to produce adipocytes. The trigger may be the absence of forces transmitted to the muscle through the intramuscular tendon fibres after tendon detachment. Altered mechanical activity at the myotendinous junction may also explain the more prominent fatty accumulation at the distal quarter of the SSP muscle\[[46](#CIT0046)\]. Pericytes physically associated with the walls of intra-adipose blood vessels showed the potential to differentiate into adipocytes *in vitro* \[[47](#CIT0047)\]. Interestingly, the habitual presence of blood vessels in the vicinity of the fat cells was used as criteria to select the fields for measurements. But the vascularization of skeletal muscles enters through the middle half of the SSP muscle and is distributed to the distal and proximal portions \[[48](#CIT0048)\]. The muscle vascular distribution is inconsistent with the IMF we observed. The identity of the precursor cell(s) differentiating into adipocytes is only speculative at this time and all four previously identified precursors are potential candidates. The direct clinical implication of the current findings of adipocyte hyperplasia as the mechanism for fatty accumulation lies in its treatment. Successful treatment of rotator cuff tear and IMF accumulation will require a strategy to reduce the number of adipocytes. This is a significant challenge since extensive literature indicates that adipocyte hypertrophy in obesity can be combatted by reducing caloric intake and increasing energy expenditure \[[22](#CIT0022)--[24](#CIT0024)\]. However, this approach is unsuccessful for adipocyte hyperplasia. Intramuscular white fat, similar to other white fat deposits is characterized by a persisting number of adipocytes. Once adipocyte number increases, they are durable and difficult to lose [\[24\];](#CIT0024) important weight loss resulted from a reduction in adipocyte volume but not overall number \[[25](#CIT0025),[32](#CIT0032)\]. This concept is consistent with the literature on reversibility of fat accumulation after rotator cuff tear. While initially believed to recover after successful tendon repair, numerous experimental as well as clinical studies have shown that fat accumulation is largely irreversible. Uhthoff et al. \[[19](#CIT0019)\] showed that animals with reattachment immediately after tear could recover muscle volume but did not reverse fat accumulation. Delayed repair also failed to reverse fat accumulation \[[18](#CIT0018)--[21](#CIT0021)\]. These four studies used precise, invasive as well as radiologic measures and followed up SSP muscles for 3 months after repair. Clinically, 38 patients showed no reversal of fatty accumulation 12--15 months postoperatively \[[47](#CIT0047),[49](#CIT0049)\], 35 patients showed progression of fat accumulation 6 months after repair\[[23](#CIT0023)\], and 47 patients followed between 60 and 133 months also showed progression of the fatty content of the rotator cuff muscles \[[50](#CIT0050),[51](#CIT0051)\]. The lack of reversibility of IMF accumulation may indicate a need for a fast intervention at repairing SSP tendon tears. Limitations of the current study include: 1) the anatomy of the rabbit rotator cuff muscles differs from human; 2) sectioned tendons were wrapped in polyvinylidine fluoride membranes to prevent the formation of adhesions and this is not the case in humans; 3) changes were studied during the first 12 weeks after tendon detachment; in clinical practice, longer delays before surgical repair of the SSP tendon tear are common; 4) tendon sectioning performed to achieve complete detachment is different than tendon tear; 5) some fat deposits in human have no precise correlates in animals and vice versa; 6) the osmium fixation method of determining adipocyte size and numbers is only possible in experimental studies. In spite of those limitations, the rabbit model of rotator cuff tear has stood out in its potential to replicate the clinical findings of fat accumulation. Conclusion {#S0005} ========== This study established adipocyte hyperplasia and increased fat clump numbers and size as the main mechanism causing fat accumulation in the SSP muscle with 12 weeks after a rotator cuff tear. The changes were predominant in the distal quarter of the SSP muscle, near the tendon tear where adipocyte number but not size increased. The trigger for adipocyte hyperplasia and the cell precursor(s) remains to be identified as a next step in the search for better SSP repair outcome. Acknowledgments =============== Funded in part by the Workplace Safety and Insurance Board of Ontario (04031) and the Canadian Institutes of Health Research (1109995). We thank Philippe Poitras for the surgeries, Ying Nie for tissue processing, Carmen Fletcher for intramuscular fat measurements. Disclosure statement {#S0006} ==================== No potential conflict of interest was reported by the authors. [^1]: This study was approved by the University of Ottawa Animal Care Committee.

1. Introduction {#sec1-membranes-08-00069} =============== Polymer electrolytes are regarded as one of the most promising candidates in advanced electrochemical applications, such as "smart" windows, displays, sensors, and more importantly, rechargeable lithium batteries \[[@B1-membranes-08-00069],[@B2-membranes-08-00069],[@B3-membranes-08-00069],[@B4-membranes-08-00069]\]. For this last one, in particular, the research has focused for decades on gel-type membrane \[[@B5-membranes-08-00069]\], generally achieved by immobilizing a liquid solution (for instance, a polar aprotic organic solvent or mixtures with a lithium salt) into a hosting polymeric matrix, such as poly(ethylene oxide) (PEO) and its derivatives (e.g., polyacrylonitrile (PAN), poly(vinylidene fluoride) (PVDF), poly(methyl methacrylate) (PMMA)) \[[@B6-membranes-08-00069],[@B7-membranes-08-00069]\]. Respect to liquid electrolytes, in fact, gel polymer electrolytes (GPEs) are able to conjugate high ion conductivities with good mechanical strength, flexible geometry, reducing of liquid leaking and, thus, higher safety \[[@B8-membranes-08-00069]\]. Owing to its ability to dissolve a large variety of salts, through interaction of its ether oxygen with cations, PEO has been one of the most extensively studied polymer used to prepare solid-state electrolytes, lighter, thinner, and safer for lithium-ion polymer batteries \[[@B9-membranes-08-00069],[@B10-membranes-08-00069]\]. Thought, the low ionic conductivities at room temperature (10^−6^--10^−8^ S cm^−1^), the Li^+^ transference number lower than 0.5 and the poor mechanical strength, still hinder the large scale diffusion of PEO-based device. Conversely, PAN ensures an ionic conductivity of circa 10^−3^ S cm^−1^, satisfactory flame and mechanical resistances, but the dimensional stability of gels is poor \[[@B11-membranes-08-00069],[@B12-membranes-08-00069]\]. After GPE preparation, in fact, a phase separation between the encapsulated electrolyte solution and the PAN matrix typically occurs, leading to a leakage problem and, thus, the passivation phenomena of the lithium electrode when in contact with the gel, as well as failure of the electrode/electrolyte contact both resulting in a dramatic reduction of the ionic conductivity. One of the strategy undertaken to bypass the drawbacks is the blending method, according to which two or more polymers are mixed to obtain a blend electrolyte. As already probed \[[@B13-membranes-08-00069],[@B14-membranes-08-00069],[@B15-membranes-08-00069],[@B16-membranes-08-00069]\] the method allows to easily control a large number of factors, directly affecting the thermal, mechanical and electrical properties of the final polymer electrolytes. By mixing PMMA and PVdF polymers, Nicotera and coworkers obtained a blend with remarkable improvement of mechanical stability respect to unblended polymers \[[@B17-membranes-08-00069]\]. Helan et al. have been reported outstanding thermal stability up to 230 °C for PAN/PMMA blends, but with quite low ionic conductivity, of the order of 2 × 10^−7^ S cm^−1^ \[[@B18-membranes-08-00069]\]. Very interesting electrical behavior and dimensional stability have been obtained by Choi et al. on PEO-PAN blend gel electrolytes, despite no evidence regarding mechanical resistance being provided \[[@B19-membranes-08-00069]\]. An alternative approach for creating gel electrolyte system with improved mechanical properties and electrochemical performances foresees the incorporation of nanoscale organic/inorganic fillers within the polymer matrix \[[@B20-membranes-08-00069]\]. The addition of SiO~2~ \[[@B21-membranes-08-00069]\], Al~2~O~3~ \[[@B22-membranes-08-00069]\], TiO~2~ \[[@B23-membranes-08-00069]\], and other metal oxides \[[@B24-membranes-08-00069],[@B25-membranes-08-00069]\] generally act as solid plasticizers, softening the polymer backbone and, thus, enhancing the segmental motion of the hosting polymer which, in turn, results in improved ion conductivity. Among inorganic fillers, layered nanoparticles based on clays have been actively investigated lately since they offer a large number of interesting properties such as high cation exchange capacity, large chemically active surface area, outstanding swelling ability, intercalation, catalytic activity, and high chemical and thermal stability. Finally, the properties of the smectite nanoclays can be tailored using simple chemical methods such as intercalation with organic or inorganic guest molecules. From the above, the dispersion of proper clay minerals within the polymer matrix could enhance the ionic conductivity improving at the same time the strength and heat resistance of the GPE. Smectite clay with different particle sizes has been effectively tested as filler for the preparation of PEO nanocomposite electrolytes, demonstrating a discrete improvement of ionic conduction \[[@B26-membranes-08-00069]\]. Kurian et al. \[[@B27-membranes-08-00069]\] have shown that the surface modification of clay by ion exchange reactions with cationic organic surfactants such as alkyl amines, enhance the chemical affinity with the polymer matrix, leading to exfoliation of the clay particles and improving the gel's strength. Organic montmorillonite (MMT) prepared by ion exchange with HTAB was dispersed in PAN polymer, obtaining a composite GPEs with improved thermal stability and ionic conductivity \[[@B28-membranes-08-00069]\]. Despite the efforts, however, there is still the need to design a gel electrolyte able to guarantee adequate electrical performance without sacrificing mechanical strength and thermal resistance. In the present study, PAN/PEO blend (80:20 weight ratio) polymers were used in order to prepare nanocomposite GPEs with an organo-modified clay. Specifically, hydrated sodium calcium aluminum magnesium silicate hydroxide (SWy-2, Nanocor) was the natural montmorillonite/smectite clay selected since it is relatively inexpensive, widely available and has small particle size as well as it shows good intercalation capability. The organo-modification of the SWy-2 (org-SWy) was achieved by ion exchange reaction with hexadecyltrimethyl ammonium bromide (CTAB). The filler loading of org-SWy in the GPE was 10 wt % with respect to the polymers PAN/PEO. For the gel preparation, a mixture of ethylene carbonate (EC) and propylene carbonate (PC), with molar ratio EC:PC 1:0.4, was used as plasticizer, while lithium trifluoromethanesulfonate (LiTr) was the salt chosen. In order to compare the effect of the clay on the gel properties, also not blended and filler-free GPE membranes were also prepared. All the GPEs were investigated by thermal (DSC), morphological (scanning electronic microscopy-SEM) and mechanical (DMA) analysis, while the ion transport studies were conducted by electrochemical impedance spectroscopy (EIS) and by multinuclear NMR spectroscopy. In particular, the ^1^H, ^7^Li, and ^19^F pulse-field-gradient (PFG) method was employed to obtain a direct measurement of the self-diffusion coefficients both of ions and solvents plasticizers (EC/PC), while the spin-lattice relaxation time (T~1~) was obtained by the inversion recovery sequence. The combination of the electrochemical and NMR data has provided a wide description of the ions dynamics inside the so complex systems, as well as information on ion associations and interactions between polymers, filler and ions. 2. Materials and Methods {#sec2-membranes-08-00069} ======================== 2.1. Materials {#sec2dot1-membranes-08-00069} -------------- Poly(ethylene oxide) (PEO, M.W. 5,000,000), polyacrylonitrile (PAN), lithium trifluoromethanesulfonate (LiCF~3~SO~3~ or LiTr, 99.95%), ethylene carbonate (EC, 98%), propylene carbonate anhydrous (PC, 99.7%), and hexadecyltrimethyl ammonium bromide (CTAB, 98%) were purchased from Sigma Aldrich, Milan, Italy and used as received. Natural smectite Wyoming montmorillonite (SWy-2) has obtained from the Source Clay Minerals Repository, University of Missouri Columbia, MO, USA. The cation exchange capacity (CEC), measured by the Co(II) procedure, is equal to 80 mequiv. per 100 g of clay, charge density 0.6 e^−1^/unit cell (the unit cell is the Si~8~O~20~ unit) and particle size around 200 nm. The structural formula is Na~0.62~\[Al~3.01~Fe(III)~0.41~Mg~0.54~Mn~0.01~Ti~0.02~\](Si~7.98~Al~0.02~) O~20~(OH)~4~. 2.2. Synthesys of Organo-Modified Clay (Org-Swy) {#sec2dot2-membranes-08-00069} ------------------------------------------------ SWy-2 were first fractioned to \<2 μm by gravity sedimentation and purified by well-established procedures in clay science \[[@B29-membranes-08-00069]\]. For the chemical modification, the cation exchange capacity of smectite clay has been exploited. CTAB (0.4 g) was dissolved in boiling deionized water until complete dissolution, then the resulting solution has been dropwise added, under vigorous stirring, to a dispersion of SWy-2 (1.0 g) in deionized water at 60 °C and left for 6 h to achieve the total cationic exchange. Finally, the mixture solution was separated by centrifugation, rinsed repeatedly with deionized water until Br− was completely removed, and dried for 24 h at 90 °C. 2.3. GPE Membrane Preparation {#sec2dot3-membranes-08-00069} ----------------------------- The solvent casting technique has been used to prepare both blended and not blended membranes, by immobilization of a lithium salt solution in a polymer matrix. The required amounts of PAN and PAN-PEO (80/20 blend ratio) were dissolved in anhydrous dimetylformammide (DMF). The solution was stirred for several hours at 60 °C, until a homogeneous mixture was obtained and, after complete dissolution, the electrolyte solution was added. For the electrolyte solution, LiCF~3~SO~3~ was dissolved in a mixture of EC and PC with a fixed molar ratio (1:0.4). The lithium content, expressed as the ratio between the number of EC-PC moles and the LiTr moles (also O/Li ratio), was 10/1. Finally, the polymers/plasticizers \[PAN:(EC-PC) and (PAN-PEO)/(EC-PC)\] weight ratio was of 26:74. For the nanocomposite GPEs, the appropriate amount of organo-modified clay has been added to DMF, mechanically stirred for 16 h and sonicated for 8 h to obtain a homogeneous dispersion. The dispersion was then added dropwise to the polymer solution, followed by further sonication and stirring. Here composite membranes with 10% of filler loading with respect to the polymer were prepared. The membranes were achieved by casting the solution on the aluminum plate at 50 °C overnight to favor the evaporation of DMF. 2.4. Characterization Techniques {#sec2dot4-membranes-08-00069} -------------------------------- NMR measurements were performed with a BRUKER AVANCE 300 Wide Bore spectrometer working at 116.6 MHz on ^7^Li, and 282.4 MHz on ^19^F, respectively. The employed probe was a Diff30 Z-diffusion 30 G/cm/A multinuclear with substitutable RF inserts. Spectra were obtained by transforming the resulting free-induction decay (FID) of single π/2 pulse sequences. The pulsed field gradient stimulated-echo (PFG-STE) method \[[@B30-membranes-08-00069]\] was used to measure the self-diffusion coefficients of lithium and triflate ions. The sequence consists of three 90° RF pulses (π/2 − τ~1~ − π/2 − τ~m~ − π/2) and two gradient pulses that are applied after the first and the third RF pulses, respectively. The echo is found at time τ = 2τ~1~ + τ~m~. Following the usual notation, the magnetic field pulses have magnitude g, duration δ, and time delay Δ. The FT echo decays were analyzed by means of the relevant Stejskal--Tanner expression: $$I = I_{0}e^{- \mathsf{\beta}D}$$ Here *I* and *I*~0~ represent the intensity/area of a selected resonance peak in the presence and in absence of gradients, respectively. β is the field gradient parameter, defined as β = \[(γgδ)\]^2^ (∆ − δ/3)\]; *D* is the measured self-diffusion coefficient. In these experiments, the used experimental parameters were: δ = 3 ms, time delay Δ = 30 ms, and the gradient amplitude varied from 350 to 1000 G cm^−1^. Based on the very low standard deviation of the fitting curve and repeatability of the measurements, the uncertainties in *D* values are estimated to about 3%. Finally, longitudinal relaxation times (T~1~) of ^7^Li and ^19^F were measured by the inversion-recovery sequence (π -- τ − π/2). All the NMR measurements were run by increasing temperature step by step from 20 to 80 °C, with steps of 10 °C, and leaving the sample to equilibrate for about 20 min at each temperature value. From *D~Li~* and *D~F~* self-diffusion coefficients, σ*~NMR~* values were calculated according with the Nernst-Einstein equation: $$\mathsf{\sigma}_{NMR} = \frac{F^{2}c_{LiTr}}{RT}\left( {D_{Li^{+}} + D_{F^{+}}} \right)$$ Here, *F* is the Faraday constant, *R* is the molar gas constant, *T* is the temperature to which *D* has been measured and *c~LiTr~* is the salt concentration. The ionic conductivity (σ, S cm^−1^) was measured by impedance spectroscopy recorded at OCV with an oscillating potential of 10 mV in the frequency range 0.1--1 × 10^6^ Hz using a PGSTAT 30 (MetrohmAutolab) potentiostat/galvanostat/FRA. GPEs were sandwiched between two disks of conductive carbon cloth, placed between two stainless steel electrodes and assembled in a homemade two-electrode cell. The impedance responses of the cell were analyzed using MetrohmAutolab NOVA software and the bulk resistance (*R~b~*) was extracted from the intercept of the low frequency signal in the Nyquist plot. The equation for calculating the conductivity is:$$\mathsf{\sigma} = \frac{l}{R_{b} \ast A}$$ where *l* is the thickness of the membrane and *A* is the area of the carbon cloth electrode. Dynamic mechanical analysis (DMA) measurements were carried out on a Metravib DMA/25 analyzer equipped with a shear jaw for film clamping. Frequency sweep experiments were collected by subjecting a rectangular film to a dynamic strain of amplitude 10^-4^ in the range between 0.2 and 20 Hz. For temperature sweep (time cure) experiments a dynamic strain of amplitude 10^−4^ at 1 Hz was applied from 25 to 160 °C with a heating rate of 2 °C/min. A periodic sinusoidal displacement was applied to the sample, and the resultant force was measured. The damping factor, tan *d*, is defined as the ratio of loss (E′′) to storage (E′) modulus. The thermal behaviors were investigated by Setaram 131 DSC. Samples were hermetically sealed and cooled from room temperature to −40 °C using liquid N~2~. Measurements were carried out from −30 °C up to 120 °C at the scan rate of 10 °C/min and purging nitrogen gas. Finally, the membrane's morphology was investigated by scanning electron microscopy (SEM, Cambridge Stereoscan 360, Santa Clara, CA, USA). To observe the membrane cross-sections, the samples were first frozen and fractured in liquid nitrogen, to guarantee a sharp fracture without modifications of the morphology, and then observed with SEM. The samples were sputter-coated with a thin gold film prior to SEM observation. 3. Results and Discussion {#sec3-membranes-08-00069} ========================= 3.1. Morphological, Thermal, and Mechanical Characterizzation of the GPEs {#sec3dot1-membranes-08-00069} ------------------------------------------------------------------------- The organo-modification of the clay's layers has as the main objective of favoring a good and homogeneous dispersion of the nanoparticles into the hosting matrix. For this purpose, hexadecyltrimethyl ammonium bromide was used as organophilic reagents: the quaternary ammonium group should allow an easy intercalation into the hydrophilic clay layers while the long alkyl chain should enhance the affinity between particles and polymer chains \[[@B31-membranes-08-00069]\]. The photos of the four gel electrolytes prepared in this study are reported in [Figure 1](#membranes-08-00069-f001){ref-type="fig"}. They all appear opalescent, while the introduction of the org-SWy causes a slight yellowing of the resulting GPEs ([Figure 1](#membranes-08-00069-f001){ref-type="fig"}b,d). However, they are very dense and homogeneous, and there is no evidence of phase segregation between PAN and PEO polymers into the blended gels, indicating that the proposed method allows to obtain a homogenous and stable mixtures of polymers. Further, no clay particles crystals were observed, confirming that the chemical modification of the layers' surface improves the clay/polymer interaction and, thus, highly homogeneous composite membranes, without formation of agglomerates or clusters, can be prepared. Scanning electron microscopy (SEM) coupled with BSE (backscattered electrons) was used to deeper investigate the morphology of the composite membranes. The BSE technique is generally used to detect contrast between areas with different chemical compositions (elements with high atomic number backscatter electrons more efficiently than light elements, appearing brighter in the image). By comparing the SEM-BSE images obtained on pristine PAN and PAN/org-SWy electrolytes, shown in [Figure 2](#membranes-08-00069-f002){ref-type="fig"}a,b, respectively, it clearly emerges that the presence of the filler particles severely affect the film morphology. The porous structure of the PAN based gel disappear in the composite gel, becoming a very dense membrane. Sporadic particle aggregations are also detectable, as expected if we take into account the large percentage of filler added into the polymer matrix (10 wt %). However, the average particles size of such aggregates is circa 500 nm, therefore, it can be stated that the nano-sized and homogeneous dispersion of clay layers was achieved in these composite GPEs. Concerning the blends ([Figure 2](#membranes-08-00069-f002){ref-type="fig"}c,d), SEM + BSE images give clear evidence that no phase separation occurs between the two polymers, as well as the presence of PEO allows the reduction of the number of nanosized aggregates in the composite blend electrolyte by virtue of a greater affinity between poly(ethylene oxide) chains and the org-SWy lamellae. The analysis of the thermal properties of the prepared electrolytes has been carried out by DSC, and the thermograms collected in the temperature range between −30 and 120 °C are showed in [Figure 3](#membranes-08-00069-f003){ref-type="fig"}. For clarity it must be noticed that, in order to highlight the peaks, an enlarged scale was used. The PAN-based gel shows two endothermic peak, the first one narrow, at about 71 °C (T~gI~), and the second broad peak at circa 100 °C (T~gII~). It was already demonstrated \[[@B32-membranes-08-00069]\] that unoriented PAN has a "two-phase" morphology consisting of laterally-ordered and amorphous domains, both in a glassy state at room temperature, thus leading to two glass transition at 100 and 150 °C, respectively. In our films, the inclusion of EC/PC plasticizer lowers both T~g~ respect to pristine PAN, as a consequence of the reduction of the crystallites size \[[@B33-membranes-08-00069]\]. The dispersion of org-SWy platelets leads to a large shift of the transitions of both laterally-ordered and amorphous domains (red line in the [Figure 3](#membranes-08-00069-f003){ref-type="fig"}), and also to a reduction of the peaks intensities, suggesting interactions between the organo-modified silicate layers and the polymer chains. It can be hypothesized that org-SWy particles increase the distance between polymer chains and, hence, diminish their capability to re-aggregate in glassy domains. Focusing on blended electrolytes, in PAN:PEO films a small peak at 44 °C appears, corresponding to the typical temperature at which PEO crystalline domains becomes rubbery amorphous phase (T~m~ PEO). Finally, the nanocomposite blend electrolyte shows a single broad peak at 45 °C ascribed to the T~gI~ of PAN while disappear the T~m~ of PEO. The result can be explained in terms of larger chemical affinity between clay platelets and PEO chains, which reduces PEO re-crystallization and, at the same time, favors the dispersion of filler's particles within the polymer matrix. Concerning the mechanical properties of the GPEs systems, the measurements were performed by dynamic mechanical analysis, by using a shear jaw for films sample holder. It is worth pointing out that, generally, oscillatory rheological tests on typical GPEs are carried out by using a plate-plate geometry, while, in this case, due to the solid-like nature of our gels, a typical DMA configuration for thin films was used. [Figure 4](#membranes-08-00069-f004){ref-type="fig"}a shows the storage modulus (E′) in the frequency range of 0.2--20 Hz measured at 25 °C: Except for one sample which will be discussed later, E′ shows values above 10^7^ Pa, significantly higher than other gels reported in the literature \[[@B34-membranes-08-00069],[@B35-membranes-08-00069],[@B36-membranes-08-00069]\], and it reaches 10^8^ Pa upon inclusion of org-SWy lamellae in the PAN matrix, indicating an increase in the rigidity of the system. Blending PAN and PEO polymers also results in an enhanced storage modulus as a consequence of the increased overall crystallinity of the polymeric matrix. However, completely unexpected is the net reduction of the storage modulus of the composite blend PAN:PEO/org-SWy electrolyte to 10^6^ Pa. This evidence can be explained by taking into account the DSC data seen above. The inclusion of the clay into the polymer matrices prevents the reorganization of PAN and PEO chains into crystalline stacks, affecting the mechanical strength of the film but, at the same time, improves the flexibility of polymer chains, with important implications on the transport properties of this electrolyte gel. However, the temperature-sweep test shown in [Figure 4](#membranes-08-00069-f004){ref-type="fig"}b demonstrates that this composite still maintains the typical strong-gel behavior, likely due to the interactions between clay platelets and polymer chains. In fact, at least up to 160 °C, the storage modulus E' exceeds significantly the loss modulus E′′, indicating that the gel responds elastically at small deformations and its microstructure is unchanged over this temperature range. The slight slope of the moduli is indicative of an evolution towards a "weak-gel" configuration, nonetheless, no crossover between the moduli occurs; therefore, the structure of the gel is preserved. 3.2. Transport Properties of Ions {#sec3dot2-membranes-08-00069} --------------------------------- The ionic conductivities of the prepared gel polymer electrolytes were investigated by EIS analysis. The impedance Nyquist plots of two representative GPEs are reported in [Figure 5](#membranes-08-00069-f005){ref-type="fig"}. The insets in each graph show an enlargement of the low resistance region, where the semicircle is achieved. In fact, the spectra show two well-defined regions: a semicircular region at high frequency range (attributed to ion conduction process in the bulk of the gel polymer electrolyte) followed by a straight line inclined at constant angle of circa 40° to the real axis at low frequency range related to the effect of blocking electrodes \[[@B37-membranes-08-00069],[@B38-membranes-08-00069]\]. By comparing the spectra of PAN gels ([Figure 5](#membranes-08-00069-f005){ref-type="fig"}a) and of PAN/org-SWy nanocomposite ([Figure 5](#membranes-08-00069-f005){ref-type="fig"}b), we can notice that the semicircle of the nanocomposite appears as depressed, i.e., it is not completed in the frequency range used, although very high (1 MHz). This indicates that multiple processes and/or mechanisms of conduction simultaneously coexist \[[@B27-membranes-08-00069]\]. A similar trend has been also observed in blended PAN:PEO/org-SWy electrolyte, even if less pronounced. From the fitting of the semicircle in the high-frequency region, the electrolyte resistance was estimated and the ionic conductivity (σ) calculated according to the formula reported in the experimental and displayed in [Figure 6](#membranes-08-00069-f006){ref-type="fig"}. It clearly emerges that PAN-SWy nanocomposite gel is the less conductive electrolyte. Such an outcome can be explained by considering the changing of the gel morphology upon addition of the clay to the polymer matrix, as discussed above, which becomes dense, as well as more rigid (higher Young's modulus). Therefore, the polymer chains experience lower flexibility, as well as a large reduction of liquid electrolyte mobility is expected by the decrease of the membrane porosity, both contributing to the reduction of the ion conduction. Similar discussion can be made on the PAN:PEO blend gel, where the enhanced membrane rigidity caused by the increased number of crystalline domains of PEO significantly affects σ compared to the unblended PAN. The best result was achieved by the addition of 10 wt % of organo-modified SWy in the PAN:PEO blend, which displays the highest ion conductivity over the whole temperature range, with a σ of almost 2.8 mS/cm at r.t. Comparing to similar GPEs reported in the literature, these conductivities are surely remarkable: e.g., they are two orders of magnitude higher than hybrid electrolytes composed of PEO and glass-ceramic particles (2.81 × 10^−2^ mS/cm) \[[@B26-membranes-08-00069]\] and three orders higher than PEO containing conductive microsized particles (1 × 10^−3^ mS/cm) \[[@B39-membranes-08-00069]\], while they are close to those reported by He et al. \[[@B31-membranes-08-00069]\] for a PAN/organic montmorillonite system (2.23 mS/cm), even if, here, an electrolyte uptake of ca. 300% was needed, resulting in deterioration of the membrane stability. Accordingly, it can be stated that the PAN:PEO composite gels are able to guarantee good polymer chain flexibility together with outstanding mechanical and thermal resistance, making these systems particularly attractive as solid electrolytes for lithium batteries. It is well known that the ionic conductivity obtained by EIS only refers to the mobility of charged species, with no possibility to distinguish between the cation and the anion. Conversely, NMR methods allow to discriminate and selectively investigate the mobility of Li^+^ and the corresponding counterion, confirming the effectiveness regarding the investigation of ions dynamics inside the complex systems, as well as information on ion associations and interactions between polymers, filler, and ions. Accordingly, NMR was used here to investigate the transport properties of both lithium cations and triflate anions, by detecting the ^7^Li and ^19^F spin-nuclei, respectively. [Figure 7](#membranes-08-00069-f007){ref-type="fig"} displays the lithium self-diffusion coefficients (*D~Li~*) measured on the GPEs' membranes, both unblended (left) and blended (right), respectively. In agreement with the conductivity seen above, the addition of org-SWy to PAN reduces the lithium mobility while it has beneficial impact in the PAN:PEO blend. However, very interesting is the bi-exponential decay of the echo-signal obtained in both composite systems, observed also for the *D~F~* (diffusion values for ^19^F, not reported in the graph). This result indicates that two different mechanism for the diffusing species coexist as a consequence of the presence of the clay lamellae. The aluminosilicate platelets possess a fixed negative charge and the quaternary ammonium group of CTAB molecules was chosen as intercalating cation. Ions are solvated both from the clay layers ("lamellae-solvation") and from the EC/PC solvents ("bulk-solvation") and, of course, the polymers play their role in such coordination. Ions involved in the "bulk-solvation" show higher mobility (*D*~1~) respect to that one involved in the "lamellae-solvation" (*D*~2~). Such a hypothesis was confirmed by the spin-lattice relaxation time (T~1~), which, compared to diffusion, reflects more localized motions, including both translation and rotation on a time scale comparable to the reciprocal of the NMR angular frequency (few nanoseconds). T~1~ quantifies the energy transfer rate from the nuclear spin system to the neighboring molecules (the lattice). The stronger the interaction, the quicker the relaxation (shorter T~1~). [Figure 8](#membranes-08-00069-f008){ref-type="fig"} reports the Arrhenius plots of T~1~ measured on the different GPEs for ^7^Li and ^19^F, respectively. It is clear that the introduction of org-SWy particles produces a decrease of T~1~, both for ^7^Li and ^19^F. This outcome can be ascribed to the stronger overall interactions of the ions with the lattice, i.e., lithium ions interact with negative charged surface of the platelets, while counterions solvate the quaternary ammonium groups of the organo-surfactant. In other words, ions experience a lower degree of freedom resulting in shorter T~1~ values. According to the Nernst-Einstein equation, conductivity values (σ*~NMR~*) were calculated from *D~Li~* and *D~F~* for the different GPEs and compared with the experimental ion conductivity (σ*~EIS~*) in [Table 1](#membranes-08-00069-t001){ref-type="table"} (for two representative gels). We need to consider that differently from σ*~EIS~*, σ*~NMR~* is affected not from the mobility of all species containing ^7^Li and ^19^F, including neutral ion pairs, and not only from the charged species. Therefore, it is not unusual for the NMR conductivity to be greater than the experimental σ, in particular when ion associations occurs. By considering the bi-exponentiality of both Li^+^ and F^−^ diffusion, and based on the hypothesis discussed above, we managed to calculate an average of *D*~1~ and *D*~2~ weighed with respect to the amount of filler added, i.e., 10 wt %. It is evident from the data reported that NMR conductivity values are always much higher than experimental ones suggesting the presence of a large number of ion pairing. This is also confirmed by the ionicity indices reported in [Table 1](#membranes-08-00069-t001){ref-type="table"} and computed as the ratio σ*~EIS~*/σ*~NMR~*. PAN gel, our reference's system, shows an ionicity close to 0.45. This suggests that 55% of Li^+^ and Tr^−^ exist as neutral ion pairs, which is typical for GPEs. The addition of filler particles into the blend increases the level of salt dissociation, likely due to the high dielectric constants of the charged organo-modified smectite clays that should also help to prevent the ionic association. Both phenomena leads to an ionicity index of 0.68 at r.t., which is a particularly high value for a double-ion solid-state electrolyte. Ionic association increases by increasing the temperature \[[@B11-membranes-08-00069],[@B39-membranes-08-00069]\], therefore, the ionicity index decreases. Finally, an important parameter for allowing a proper operation of the polymer electrolyte in real device is the lithium transport number ($t_{Li^{+}}$). It was calculated in this work according to the following equation and reported in [Table 1](#membranes-08-00069-t001){ref-type="table"}: $$t_{Li^{+}} = \frac{D_{Li^{+}}}{D_{Li^{+}} + D_{F^{-}}}$$ The PAN:PEO/org-SWy electrolyte shows a value of 0.68 at r.t., much higher than the PAN-gel and also the typical GPEs, for which values lower than 0.30 are generally reported \[[@B26-membranes-08-00069],[@B40-membranes-08-00069],[@B41-membranes-08-00069],[@B42-membranes-08-00069],[@B43-membranes-08-00069]\]. GPEs with higher lithium transport number, i.e., ca. 0.55, has also been reported, but the ion conductivities are quite low \[[@B44-membranes-08-00069]\]. The reasons of the improved $t_{Li^{+}}$ in our blend composite membrane can be multiple and synergistic: (i) the organo-clay particles have a plasticizing effect, lowering the cristallinity and, thus, improving the flexibility of polymer chains, favoring the Li^+^ transport through polymer segmental motions; and (ii) electrostatic interactions between the filler surface and lithium can create a preferential pathways for lithium conduction. 4. Conclusions {#sec4-membranes-08-00069} ============== Organo-modified smectite clay particles were prepared and dispersed into PAN and PAN:PEO blend polymers in order to prepare hybrid gel polymer electrolytes. Morphological studies proved that the procedure herein proposed allows to avoid phase separation between PAN and PEO as well as guarantee high nano-dispersion of the clay particles in the polymer matrix. The presence of the clay platelets strongly affected morphology, thermal and mechanical stability and electrochemical properties of the GPEs. In particular, outstanding behavior was displayed by the PAN:PEO/org-SWy membrane. ^7^Li and ^19^F NMR spectroscopy was successfully applied to get a complete description of the ions dynamics in so complex systems, probing as the smectite clay surfaces are able to "solvate" both lithium and triflate ions, preventing the ion pairing (as also confirmed by the high ionicity index) and creating preferential pathways for lithium conduction. The authors would like to thank Mariano Davoli, University of Calabria, for his precious support in the morphological characterization of the membranes. Conceptualization, I.N. and C.S.; Methodology, C.S., E.L. and L.C.; Validation, I.N.; Investigation, C.S. and E.L.; Resources, I.N.; Data Curation, C.S. and L.C.; Writing-Original Draft Preparation, C.S.; Writing-Review & Editing, I.N.; Supervision, I.N.; Project Administration, I.N.; Funding Acquisition, I.N. This work was supported by the European Community's Seventh Framework Program (FP7 2007-2013) through the MATERIA Project (PONa3_00370). The authors declare no conflict of interest. {#membranes-08-00069-f001} {#membranes-08-00069-f002} {#membranes-08-00069-f003} {#membranes-08-00069-f004} {#membranes-08-00069-f005} {#membranes-08-00069-f006} {#membranes-08-00069-f007} {#membranes-08-00069-f008} membranes-08-00069-t001_Table 1 ###### Comparison between σ*~EIS~* and σ*~NMR~*(in Ms cm^−1^), ionicity index and lithium transport number for PAN and PAN:PEO/org-SWy electrolytes. T (°C) PAN PAN-PEO + 10% SW -------- ------ ------------------ ------ ------ ------ ------- ------ ------ 20 1.77 3.92 0.45 0.40 2.79 4.31 0.68 0.68 30 2.08 5.02 0.41 0.41 3.22 4.74 0.65 0.67 40 2.31 6.96 0.33 0.41 3.37 7.84 0.53 0.59 50 2.56 8.19 0.31 0.43 3.17 9.07 0.45 0.56 60 2.74 9.68 0.28 0.44 3.77 9.82 0.48 0.56 70 3.03 11.1 0.27 0.43 4.05 11.20 0.42 0.57 80 3.12 11.8 0.27 0.45 4.24 12.33 0.38 0.58

1. Introduction {#sec1} =============== Acute respiratory distress syndrome (ARDS) incorporates a cluster of clinical features including shortness of breath and tachypnea and is defined by the Berlin criteria as having an acute onset with the development of hypoxemia and bilateral pulmonary opacities on radiographic imaging \[[@B1]\]. There are many causes of ARDS; however, most often, it is triggered by infections, blood transfusions, direct lung injury, and toxins \[[@B2]\]. Treatment includes removal of the inciting cause, supportive therapy, and the attainment of sufficient blood oxygenation. Adequate oxygenation in ARDS is often achieved by endotracheal intubation and mechanical ventilation. Noninvasive positive pressure ventilation (NIPPV) has been less frequently indicated as an alternative form of adequate oxygenation. Typically, its use is focused and intended on preventing complications that are associated with invasive ventilation such as barotrauma, vocal cord injury, and ventilator-associated pneumonia \[[@B3]\]. Bupropion is an atypical oral antidepressant medication commonly used to treat depression, tobacco dependence, obesity, and hypoactive sexual disorder. Its mechanism of action involves the inhibition and reuptake of norepinephrine and dopamine which can induce a state of euphoria. Because of these effects, bupropion has been reported to be abused recreationally \[[@B4], [@B5]\]. We describe a case of ARDS induced by bupropion inhalation that was treated with NIPPV. 2. Case Presentation {#sec2} ==================== A 30-year-old male with a past medical history significant for polysubstance abuse presented to the emergency department (ED) two hours after ingesting 90 mg of oxycodone and 30 mg of diazepam along with intranasal bupropion. On arrival, he complained of extreme fatigue, respiratory distress, and confusion. In the ED, he was noted to be lethargic and short of breath. Initial vital signs revealed heart rate of 104 beats per minute, respiratory rate of 35 per minute, and O~2~ saturation of 75% on room air. Respiratory exam revealed diffuse bilateral coarse breath sounds on inspiration and expiration. His laboratory results were notable for a leukocytosis of 14.1 K, creatinine of 1.25 g/dl, lactate of 4.1, and an elevated procalcitonin level. Venous blood gas analysis showed a pH of 7.18 with a pCO~2~ of 51 mmHg. Chest X-ray (CXR) demonstrated diffuse bilateral alveolar infiltrates ([Figure 1(a)](#fig1){ref-type="fig"}). Computed tomography (CT) of the chest showed diffuse bilateral airspace disease characterized by groundglass and consolidative opacities with relative peripheral lung sparing and perihilar predominance ([Figure 2](#fig2){ref-type="fig"}). He received naloxone with mild improvement in mental status and was initiated on Bilevel Positive Airway Pressure (BiPAP) with an inspiratory positive airway pressure (IPAP) of 15 cm H~2~O, expiratory positive airway pressure (EPAP) of 5 cm H~2~O with 40% fraction of inspired oxygen (FiO~2~). Empiric broad spectrum antibiotics including vancomycin, cefepime, and metronidazole were initiated along with intravenous methylprednisolone 40 mg every 12 hours. Initially while on BIPAP, he remained tachypneic and tachycardic. His respiratory rate improved over the following 6 hours. He was gradually weaned off BiPAP to nasal cannula oxygen over the course of 36 hours while receiving ongoing corticosteroid and antibiotic therapy. Repeat CXR on hospital day \#6 showed markedly improved bilateral airspace opacities ([Figure 1(b)](#fig1){ref-type="fig"}). After 6 days, he was discharged in stable condition without requiring supplemental oxygen. 3. Discussion {#sec3} ============= ARDS is associated with a high mortality that has declined from over 50% to 30% over the last four decades \[[@B6], [@B7]\]. This is primarily due to implementation of lung-protective ventilation protocols and intensified research after formation of the National Heart, Lung, and Blood Institute (NHLBI) ARDS network \[[@B8], [@B9]\]. This case highlights a mild manifestation of ARDS as a result of bupropion inhalation, an exceedingly rare etiology. The Food and Drug Administration (FDA) classifies bupropion as a psychiatric medication with low abuse potential \[[@B10]\]. However, several case reports and studies have indicated increasing recreational use of bupropion mostly intravenously or intranasally \[[@B11]--[@B13]\]. Lewis et al. conducted a review on bupropion inhalation in a total of 67 patients. Seizures were noted as a common adverse effect occurring in 30% of patients. Acute lung toxicity was not reported as a complication \[[@B14]\]. We were not able to find a second reported case of bupropion-inhalation-induced ARDS. Endotracheal intubation and mechanical ventilation are mostly required to ensure adequate oxygenation, but this was avoided in this patient as we maintained adequate oxygenation with BiPAP alone. NIPPV is advantageous in such patients as it does not expose them to the potential complications of invasive ventilation and may shorten their hospital length of stay \[[@B3]\]. There is, however, an ongoing debate concerning the most effective mode of NIPPV \[[@B15], [@B16]\]. Recent studies show that noninvasive ventilation can be used in mild cases of ARDS with acute nonhypercapnic hypoxemic respiratory failure \[[@B17]\]. In these cases, BiPAP via facemask is the most commonly used strategy \[[@B18]\]. Another approach with high-flow nasal cannula was shown to have a similar degree of treatment failure and incidence of subsequent intubation as BiPAP. High-flow nasal cannula, however, was associated with decreased 90-day mortality as compared to BiPAP \[[@B19]\]. A randomized, single-center trial by Pohlman et al. showed that delivering noninvasive ventilation via helmet instead of by facemask was associated with a significant reduction in intubations. There was also a decrease in intensive care unit length of stay and mortality at 90 days \[[@B20]\]. However, a subset analysis of the observational "LUNG-SAFE" study for patients with severe ARDS (defined as a PaO~2~/FiO~2~ ratio below 150 mmHg) showed increased mortality with noninvasive ventilation \[[@B21]\]. Per our review, data on the use of noninvasive ventilation in the management of ARDS remains inconclusive and conflicting. NIPPV may decrease the incidence of ventilator-associated complications; however, given that its efficacy is not well established, patients with ARDS should still be treated in a critical care setting for close monitoring with invasive ventilation available on standby. 4. Conclusion {#sec4} ============= Advances in the treatment of the underlying etiologies and improvement in mechanical ventilation strategies have led to a decrease in the overall mortality associated with ARDS. Despite these developments, it remains a diagnosis with an exceedingly high mortality. New emerging data indicates that NIPPV can be a beneficial and equivalent approach for a subset of patients with ARDS, although the optimal type of noninvasive ventilation and patient group that would benefit most is yet to be determined. Conflicts of Interest ===================== All authors declare no conflict of interest. {#fig1} {#fig2} [^1]: Academic Editor: Mabrouk Bahloul

Historically pulmonary emphysema was described in 1834 by Laennec on the basis of observations made on the cut surface of postmortem human lungs being the lesion attributed to the atrophy of lung tissue from pulmonary hyperinflation.^([@B1])^ Hence, emphysema was redefined as a "abnormal and permanent dilation of distal air spaces of terminal bronchiole".^([@B2])^ In addition, evidences of destruction of alveolar wall and fibrosis must not be ignored in this disease pathogenesis.^([@B3])^ These anatomopathological changes result in loss of respiratory surface and blood irrigation, decrease of elastic recognition and pulmonary hyperexpansion, and it could also affect part of acinus or its structure.^([@B4])^ Pulmonary emphysema is caused by enzymatic imbalance between proteases and anti-proteases that results in destruction of the alveolar wall due to proteolytic enzymes action, which affects the extracellular matrix (ECM)^([@B5])^ and its component integrity especially the elastic fibres.^([@B6])^ Experimental model of pulmonary emphysema is based on nebulization or instillation of proteolytic enzyme, such as panain *(Carica papaya)*,^([@B7])^ porcine pancreatic elastase,^([@B4])^ and human neutrophil elastase.^([@B8])^ This proteolytic process, associated with uniform destruction of ECM of pulmonary acinus, ends up in morphohistological and physiological changes in lungs that resemble those changes find in emphysema in humans.^([@B9],[@B10])^ Dilatation of distal air spaces of terminal bronchiole ([Figure 1](#f01){ref-type="fig"}) and reduction of area occupied by elastic fibres ([Figure 2](#f02){ref-type="fig"}) evidenced histologically the pulmonary emphysema in experimental models that use porcine pancreatic elastase. Figure 1Photomicrographs of lung parenchyma (hematoxylin-eosin) x 100 increased. (A) Naïve lung and (B) emphysematous lung showing hyperdistension of alveolar ducts associated with the rupture of alveolar septa Figure 2Photomicrographs of lung parenchyma (Verhoeff), x 400 increased. Lung naïve showing integrity of elastic component of alveolar wall, opposing to areas revealed throughout septa associated with thickening of elastic fibres in alveolar wall and decreasing of proportion of elastic fibres in emphysematous lung (B) Aprendendo Por Imagens Características histopatológicas do enfisema pulmonar em modelo experimental Petta Antonio Di 1 Universidade de São Paulo, São Paulo, SP, Brasil. Autor correspondente: Antonio Di Petta − Rua Rodolfo Marcos Teófilo, 49 − Freguesia do Ó -- CEP: 02862-100 − São Paulo, SP, Brasil − Tel.: (11) 3851-0028 − E-mail: antoniodipetta\@usp.br Historicamente, Laennec (1834) descreveu o enfisema pulmonar a partir de observações em cortes necroscópicos superficiais de pulmões humanos, atribuindo a lesão à atrofia do tecido pulmonar resultante da hiperinsuflação pulmonar.^([@B1])^ O enfisema foi, então, redefinido como uma "dilatação anormal e permanente dos espaços aéreos distais do bronquíolo terminal".^([@B2])^ Além do mais, a evidência da destruição da parede alveolar e de fibrose não deve ser ignorada na patogenia da doença.^([@B3])^ Essas alterações anatomopatológicas resultam na perda da superfície respiratória e de irrigação sanguínea, na diminuição do recolhimento elástico e na hiperexpansão pulmonar, podendo atingir parte do ácino ou toda sua estrutura.^([@B4])^ O enfisema pulmonar é obviamente uma doença causada pelo desequilíbrio enzimático existente entre proteases e antiproteases, resultando na destruição da parede alveolar ocasionada pela ação de enzimas proteolíticas, que degradam a matriz extracelular (MEC)^([@B5])^ e afetam a integridade de seus componentes, particularmente as fibras elástica.^([@B6])^ Modelos experimentais de enfisema pulmonar baseiam-se na nebulização ou instilação de enzimas proteolíticas, como papaína (*Carica papaya*),^([@B7])^ elastase pancreática de porco,^([@B4])^ e elastase neutrofílica humana.^([@B8])^ Esse processo proteolítico, associado à destruição uniforme da MEC do ácino pulmonar, resulta em alterações morfo-histológicas e fisiológicas dos pulmões equivalentes às alterações encontradas no enfisema em seres humanos.^([@B9],[@B10])^ A dilatação dos espaços aéreos distais do bronquíolo terminal ([Figura 1](#f03){ref-type="fig"}) e a redução da área ocupada pelas fibras elásticas ([Figura 2](#f04){ref-type="fig"}) evidenciam histologicamente o enfisema pulmonar em modelos experimentais instilados por elastase pancreática de porco. Figura 1Fotomicrografias do parênquima pulmonar (hematoxilina-eosina), aumento x100. (A) Pulmão *naive* e (B) pulmão enfisematoso, demonstrando hiperdistensão dos ductos alveolares com ruptura dos septos alveolares Figura 2Fotomicrografias do parênquima pulmonar (*Verhoeff*), aumento de 400x. (A) Pulmão *naive* demonstrando integridade dos componentes elásticos da parede alveolar, contrastando com áreas desnudas ao longo dos septos associadas ao adensamento de fibras elásticas na parede alveolar e diminuição da concentração de fibras elásticas no pulmão enfisematoso (B)

1. Introduction {#sec1-jintelligence-05-00001} =============== Generally, Black people, White people, and East Asian people have different average IQs, but the specific reasons for this remain controversial. Of various bodies of evidence, one has been nominated as "one of the most powerful" for explaining these racial IQ differences: the IQs of adoptees of different races raised by White adoptive parents \[[@B1-jintelligence-05-00001]\] (p. 25). Several psychologists have reviewed studies of transracial adoptees' IQs and used them as evidence for two claims. The first is that East Asian adoptees raised by Whites have higher average IQs than the general White population; that White adoptees raised by Whites in turn have higher average IQs than Black adoptees raised by Whites; and that multiracial adoptees have an expected IQ that is a weighted average of their ancestral groups' mean IQs (so that adoptees born to a White genetic parent and a Black genetic parent, for example, have an average IQ midway between those of Whites and Blacks). This would suggest that racial IQ differences persist even among children raised in nurturing adoptive homes, intimating that differences in home environments cannot explain racial IQ differences. This leads into the second claim: that most of the racial IQ differences between East Asians, Whites, and Blacks are genetic \[[@B1-jintelligence-05-00001],[@B2-jintelligence-05-00001],[@B3-jintelligence-05-00001]\]. In this paper, I call the conjunction of these two claims the *hereditarian* hypothesis or model, after Rushton and Jensen \[[@B3-jintelligence-05-00001]\] (pp. 239--240, 259). This paper reanalyzes the adoption studies cited to support these claims, and puts forth an alternative explanation: most of the average racial IQ differences in those studies are spurious, arising from methodological issues and disregard of contrary results. When all of the cited data are considered with these issues in mind, they are not compelling evidence for large and consistent IQ differences between East Asian, White, and Black adoptees raised by White parents. This paper then introduces further adoption data which have yet to be considered in the race and IQ debate. The totality of the data turn out to be at least as consonant with a *nil* hypothesis or model: the IQs of adoptees raised by Whites in comparable environments are hardly affected by the adoptees' race. 2. Methodological Issues {#sec2-jintelligence-05-00001} ======================== The innovation of this paper is that it accounts for four methodological issues which imperil analyses of transracial adoption studies of IQ. Two of the issues are examples of the generic scientific problem of confounded comparisons of groups, and the third is a type of selection bias. Firstly, if a sample of adoptees of one race or ethnicity scores above another ethnic group's general-population average, one cannot automatically attribute the above-average score to the adoptees' ethnicity. The adoptees are *adoptees*, and adoptees are typically raised in unrepresentative environments which tend to be more nurturing and high in socioeconomic status. Unusually wholesome environments could then explain the adoptees' above-average IQ, rather than the adoptees' race; race and environment would be confounded. (Indeed, the adoptees themselves are likely to be unrepresentative of their own race, and a sceptic could conceivably attribute the adoptees' above-average IQ to their very unrepresentativeness, regardless of environment.). The two most obvious ways to control away this confounding are to compare adoptees against only other adoptees, or to adjust the adoptees' observed IQs downwards to allow for the adoptees' better environments. The second methodological issue is the Flynn effect. James R. Flynn and his colleagues have documented steady rises in average IQ in several countries \[[@B4-jintelligence-05-00001],[@B5-jintelligence-05-00001],[@B6-jintelligence-05-00001],[@B7-jintelligence-05-00001],[@B8-jintelligence-05-00001]\]. These rises makes IQ test norms progressively more outdated over time, so adoptees who take an IQ test would have exaggerated IQ scores relative to people who took the same test earlier. In particular, all IQ tests must be standardized against a reference population, offen the general population of a given country, and when a group of adoptees takes the IQ test at a later date, the time lag exaggerates the adoptees' performance relative to the reference population. Had the reference population taken the test at the same time as the adoptees, the reference population would have set a higher benchmark for the adoptees. The same mechanism means it is usually illegitimate to directly compare adoptees' IQs across studies, because adoptees in different studies are usually tested in different years, and with different tests. Like environment, the year in which a test is taken and the choice of test can be confounded with race. To avoid such confounds, analysts can either subtract out the Flynn effect from each set of results, or make comparisons only of groups which took the same IQ test at approximately the same time. The third issue, attrition, is less common, affecting only longitudinal studies. Even if a longitudinal study compares adoptees against only other adoptees (eliminating the first confound) who took the same IQ test at similar times (eliminating the second confound), attrition can take place between waves. When researchers lose track of some subjects between waves of a longitudinal study, the pattern of subjects lost to follow-up can vary between subgroups of subjects, degrading the statistical comparability of those subgroups. Taking the specific case of adoptees' IQs, if a longitudinal study included e.g., White and e.g., Black adoptees, and the White adoptees lost to follow-up were disproportionately lower scorers, whereas the Black adoptees lost to follow-up were not, the White--Black IQ difference among the remaining adoptees would be inflated. Race can correlate with selection for retention in the study. One can adjust for this type of selection bias by making a counterfactual estimate of how the subgroups would have scored had no one been lost to follow-up. The fourth issue is that published reviews of transracial-adoption IQ studies have not considered all of the studies which were available and germane, and this selective treatment of the data may introduce bias. Consequently, as Rushton and Jensen put it, "\[t\]o be compelling, \[\...\] researchers must take the totality of available evidence into account" \[[@B9-jintelligence-05-00001]\] (p. 921). The remainder of this paper should give an idea of whether past writers on transracial adoption and IQ have met this standard. 3. A Hereditarian Interpretation {#sec3-jintelligence-05-00001} ================================ Before contesting the hereditarian model of transracial adoption studies, this paper must first describe it. To that end, this section takes stock of the studies already discussed in the literature, and why they might seem to support the hereditarian hypothesis. [Table 1](#jintelligence-05-00001-t001){ref-type="table"} gives a list in publication-year order, where samples of adoptees with two Black genetic parents each are labelled "Black--Black" and samples of adoptees with one White and one Black genetic parent each are labelled "Black--White". The 130 East Asian adoptees in the Winick, Meyer, and Harris \[[@B10-jintelligence-05-00001]\] and Frydman and Lynn \[[@B11-jintelligence-05-00001]\] studies were all Korean, while the 25 East Asian adoptees in the Clark and Hanisee \[[@B12-jintelligence-05-00001]\] sample comprised "12 from Vietnam, 8 from Korea, 3 from Cambodia, and 2 from Thailand" (p. 596). The racial designations may not be rigorous but the current paper uses them for argument's sake. A hereditarian interpretation of the data in [Table 1](#jintelligence-05-00001-t001){ref-type="table"} begins by "plac\[ing\] greatest weight on the Minnesota Trans-Racial Adoption Study because it is the largest and best-known of these studies and is the only one that included a longitudinal follow-up" \[[@B1-jintelligence-05-00001]\] (p. 25). Scarr and Weinberg \[[@B13-jintelligence-05-00001]\] reported results from the first wave of that study, and Weinberg, Scarr, and Waldman \[[@B14-jintelligence-05-00001]\] the follow-up results. In both waves the wholly White adoptees had a higher mean IQ than the Black--White adoptees, and the Black--White adoptees had a higher mean IQ than the Black--Black adoptees, in line with the hereditarian model. Moreover, those IQ gaps widened between the study's two waves, consistent with genes exerting greater power as the adoptees aged. The other studies of Black adoptees \[[@B15-jintelligence-05-00001],[@B16-jintelligence-05-00001]\] do not fit the hereditarian model nearly as well, as the race-IQ correlations they found were all either small or associated higher IQ with greater Black ancestry. However, one can downplay these other studies on the ground that they lacked longitudinal follow-up testing, and by arguing that genetic effects could've been masked by their subjects' lower age (since "as people age, their genes exert ever more influence" \[[@B3-jintelligence-05-00001]\] (p. 259)). This aligns with the hereditarian claim that mean IQ decreases as Black ancestry increases. As for the claim that mean IQ rises with greater East Asian ancestry, there is the fact that three studies of East Asians adopted in Western countries each reported higher mean IQs for them than the norm of 100 \[[@B3-jintelligence-05-00001]\] (pp. 259--260), as can be seen in [Table 1](#jintelligence-05-00001-t001){ref-type="table"}. The lower mean IQ for adoptees with more Black ancestry and the higher mean IQ for East Asian adoptees may be read as support for the hereditarian model. 4. A Re-Analysis of East Asian Adoptee IQ Data {#sec4-jintelligence-05-00001} ============================================== However, the high IQ of these studies' East Asian adoptees is misleading because none of the studies included adoptees of other races, so they had no direct control groups. This posed the two confounding problems introduced in [Section 2](#sec2-jintelligence-05-00001){ref-type="sec"} above: the Flynn effect and the IQ boost from the adoptive environment. Adjusting for the Flynn effect is conceptually trivial: look up which test an adoptee sample took; work out how much time elapsed between when the test was standardized and when the adoptees took the test; look up the rate of IQ gains for the general population where the adoptees were raised; finally, multiply that rate by the time elapsed between standardization and the adoptees taking the test for an estimate of how much the Flynn effect inflated the adoptees' average IQ. Subtracting that estimate from the adoptees' measured IQ then gives a truer estimate of their IQ. Adjusting for the adoptive IQ boost is harder. Theoretically one could estimate the average IQ boost due to adoption and subtract it from an adoptee sample's average IQ, but which estimate to use for the average IQ boost from adoption is unclear. The meta-analysis of van IJzendoorn et al. \[[@B17-jintelligence-05-00001]\] has estimates but there are two reasons not to use them. For one thing, the van IJzendoorn et al. meta-analysis incorporated the East Asian adoptee studies I discuss here, so correcting the East Asian IQ means with the van IJzendoorn et al. estimates would be circular reasoning. Worse, van IJzendoorn et al.'s summary of various studies had inaccuracies. For example, their [Table 1](#jintelligence-05-00001-t001){ref-type="table"} (p. 306) described the mean IQ of Korean adoptees in the Winick et al. study \[[@B10-jintelligence-05-00001]\] as matching a norm of 100 (specifically, the Cohen's *d* effect size associated with IQ was "0.00"), although one can calculate from Winick et al. \[[@B10-jintelligence-05-00001]\] that the tested adoptees' mean IQ was 106.7. As a result, I have no general estimate of adoption's positive effect on IQ, which prevents me from adjusting for the adoptive IQ boost confound. Instead, I adjust for the Flynn effect alone, and then simply ask whether any remaining East Asian IQ advantage is small enough that it could plausibly be attributed to the adoptive IQ boost. Winick et al. \[[@B10-jintelligence-05-00001]\] is the oldest of the three East Asian adoptee IQ studies. It surveyed 36 "malnourished", 38 "moderately nourished", and 37 "well-nourished" Korean girls adopted as infants by Americans. Upon taking IQ tests between 1971 and 1973, the three groups obtained mean IQs of 102, 106, and 112 respectively. Rushton and Jensen \[[@B3-jintelligence-05-00001]\] (p. 260) observed that they "exceeded the national average" in IQ. The adoptees' IQs were measured in school \[[@B10-jintelligence-05-00001]\] (p. 1173) with four different group tests: the Lorge-Thorndike Intelligence Test, the Otis-Lennon Mental Ability Test, the Cognitive Abilities Test, and the California Test of Mental Maturity (p. 1175). Unfortunately, Winick et al. did not specify which editions of each test were used, so it is impossible to correct exactly for the Flynn effect. The mental-test guide *Intelligence: Tests and Reviews* \[[@B18-jintelligence-05-00001]\], up to date through January 1974 (p. xxvi), reported that various editions of these tests were published between 1936 and 1972 (pp. 7, 10, 32, 41). The degree to which the Flynn effect inflated the adoptees' apparent IQs depends on which children took which edition of each test and when. In the US the population's IQ has risen by 0.3 IQ points per year \[[@B4-jintelligence-05-00001],[@B5-jintelligence-05-00001],[@B6-jintelligence-05-00001],[@B7-jintelligence-05-00001]\]. Hence the Flynn effect might have exaggerated the adoptees' mean IQs by as many as 11 points: multiplying 0.3 points per year by the 37 years between 1973 and 1936---when the oldest versions of the tests were published---gives an 11.1-point IQ inflation. (Were the 1936 tests standardized some time before being published the effect could be even greater.) Deducting 11 points from each group mean gives an overall mean IQ of 96, illustrating that the Flynn effect could account for all of the apparent IQ advantage in Winick et al. \[[@B10-jintelligence-05-00001]\]. At the other extreme, the adoptees' schools might have tested almost all of the adoptees in 1972 with the 1972 edition of the Cognitive Abilities Test, in which case the Flynn effect would have inflated the adoptees' IQ scores only negligibly. Given the study's age, the records of how each child was tested are likely lost, which bars an exact estimate of the Flynn effect affecting this study. Splitting the difference between the two extreme scenarios suggests a Flynn effect of 5--6 points. The studies' results are therefore consistent with a baseline IQ of 100 for Korean adoptees, exaggerated by a Flynn effect of 6 points and an adoptive IQ boost of 6 points, with the undernourished groups losing 6--10 points through malnourishment. Rushton and Jensen's contention \[[@B3-jintelligence-05-00001]\] (p. 260) that the adoptees' IQ "exceeded the national average" is quite possibly fallacious. A reviewer has suggested that the "malnourishment might be a genetic effect", because low-IQ parents "are poor and have less money for food for their children". Whether this has any effect on my argument depends on the precise comparison one has in mind. If the relevant comparison is between East Asians and Whites with identical environments, including identical levels of nutrition, then whether real-world malnutrition is linked to genes is irrelevant; one wishes to equalize the levels of nutrition between the comparison groups regardless, and may write off the lower IQ of the less nourished groups as the result of an unfair comparison. Alternatively, if one thinks the relevant comparison is between East Asians and Whites born in environments that may differ as long as the differences can be attributed to genes, then correcting for malnutrition is an overcorrection insofar as the malnutrition is genetically driven. Supposing the latter, *could* genes suffice to explain the lower IQs of the less nourished groups? The causal chain the reviewer highlights runs from parental genes to parental IQ to family income/wealth to child's malnutrition to child's IQ. To set an upper bound on the overall strength of this chain, I may take the product of the upper limits of the correlations between each consecutive pair of variables, assuming arguendo that the links between each pair are causal and linear. The first correlation is the square root of IQ's heritability among the parental population; a high heritability figure of 80% gives a correlation of 0.89. For the correlation between IQ and income I use 0.23, the highest value from the meta-analysis of Strenze \[[@B19-jintelligence-05-00001]\] (p. 412); note that this is Strenze's correlation between individual IQ and individual income, which is presumably higher than that between individual IQ and *family* income. Lacking a convenient meta-analysis or large-scale study quantifying the correlation between family affluence and malnutrition, I will be generous to the reviewer's argument and suppose the correlation is implausibly high, say 0.8. The final correlation is that between child malnutrition and child IQ, which one may estimate from the Winick et al. \[[@B10-jintelligence-05-00001]\] results themselves, using height and weight percentiles as a quantitative nutritional index. The children in that study designated "malnourished" or "well-nourished" were "below the 3rd percentile for both height and weight" and "at or above the 25th percentile for both height and weight" respectively (p. 1173), and so (making the approximation that height and weight were normally distributed in the reference population) the two groups were separated by at least 1.88 − 0.67 = 1.21 standard deviations in nutritional level. However, the two groups' mean IQs differed by only 10 points, or two thirds of a standard deviation, implying a nutrition-IQ correlation of at most (2 ÷ 3) ÷ 1.21 = 0.55. The overall chained correlation is then 0.89 × 0.23 × 0.8 × 0.55 = 0.09. To explain the two-thirds-of-a-standard-deviation discrepancy between the least and most nourished subsamples in the Winick et al. study, therefore, one would have to posit that the difference in parental IQ between the two subsamples was (2 ÷ 3) ÷ 0.09 = 7.4 standard deviations. This is implausible. The assumptions feeding into that estimate are themselves collectively implausible, but they are conservative, so substituting more plausible assumptions would not change the qualitative result: the chained effect of genes on IQ, of IQ on income, of income on nutrition, and of nutrition on IQ is too feeble to explain away the IQ differences between Winick et al.'s groups on the reviewer's genetic grounds. This frees me to move on to the most recent of the three East Asian adoptee IQ studies, namely Frydman and Lynn \[[@B11-jintelligence-05-00001]\], a short report on "19 Korean children who were orphaned or abandoned in Korea in the mid-1970s and subsequently adopted by Belgian families" (p. 1323). In 1983 the children attained a mean IQ of 118.7 on the Wechsler Intelligence Scale for Children (WISC), with a verbal mean (VIQ) of 110.6 and a performance mean (PIQ) of 123.5. Frydman and Lynn recognized that they oughtn't take these scores at face value, observing that "the French WISC was standardised in 1954, \[...\] and that the mean IQ in Belgium will have increased over the 29-year period between the standardisation and the testing of the Korean children" (p. 1324). Unfortunately, when adjusting for this, they then assumed that IQ had risen in Belgium at the same rate as in the US: "3 IQ points per decade" (p. 1324). This was almost surely an under-adjustment. Flynn \[[@B5-jintelligence-05-00001]\] (p. 185) had already published Belgian military results from samples of 18-year-old men, clocking nonverbal test gains at 7--8 points per decade and verbal test gains at 4 points per decade. Mapping the verbal gains to VIQ and the performance gains to PIQ and deducting both from the adoptees' means reduces the adoptees' mean VIQ to 99 and their mean PIQ to 100--103. Adjusting properly for the Flynn effect, then, the Frydman and Lynn adoptees scored on par with Belgian children. When one recalls that adoption should have raised the adoptees' IQs above the general population's average, this outcome is surprising; it suggests that Frydman and Lynn's subjects would have scored *below* comparable White adoptees in Belgian homes. Frydman and Lynn acknowledged that the Korean adoptees "were brought up in middle class families" and that this "would have raised their mean IQ above that of all \[sic?\] Belgian children" in itself (p. 1324). Estimating that "middle class children obtain a mean IQ of about 105", Frydman and Lynn interpreted their under-adjusted mean of 108.7 as one that "would suggest a genotypic Korean advantage" (p. 1324), but this latter conclusion was an error. After a more sensible adjustment, the Korean adoptees' true mean VIQ and PIQ were between 99 and 103, both less than 105. By Frydman and Lynn's logic this would suggest a genotypic Korean IQ *disadvantage*. A possible rebuttal to my argument is that I am over-adjusting by using Flynn's data on Belgian IQ gains. Flynn's data cover only the period of 1958 to 1967, whereas Frydman and Lynn's adoptees could have been born no earlier than 1969. Possibly Belgian IQ gains were slower while Frydman and Lynn's subjects were growing up, leading me to adjust the adoptees' averages too much. Without additional Belgian data it is impossible to decisively refute this rebuttal, but records of French IQ gains, a proxy for Belgian gains, weigh against it. Military samples of French men aged 18 to 22 revealed gains of a point a year on Raven's Progressive Matrices and 0.4 points per year for verbal and mathematical tests between 1949 and 1974, gains at least as fast as those in the Belgium data \[[@B5-jintelligence-05-00001]\] (p. 185). Flieller et al. \[[@B20-jintelligence-05-00001]\] documents a gain of 1.6 standard deviations on Gille's Mosaic test for French 8-year-olds between 1944 and 1984, the equivalent of 0.6 IQ points per year. Deducting 29 × 0.6 = 17.4 points from the adoptees' full-scale IQ mean of 118.7 brings it down to 101.3, neither practically nor statistically different from 100, and again less than 105. Flynn \[[@B5-jintelligence-05-00001]\] (pp. 184--185) also records French gains on the WISC between 1955 and 1979 among 6--15-year-olds, with PIQ rising by 0.6 points a year and VIQ by 0.1 points a year. The minuscule VIQ increase is, however, "unique" and Flynn categorizes the French WISC data as "speculative". The overall pattern of French results fits my estimates of Belgian IQ gains. All in all, it is very likely that the Korean adoptees in Frydman and Lynn \[[@B11-jintelligence-05-00001]\] would have scored little better than comparable White adoptees. This cuts against Rushton and Jensen's assertion \[[@B3-jintelligence-05-00001]\] (p. 260) that "the Korean children still had a statistically significant 10-point advantage in mean IQ over indigenous Belgian children". The allegation that "\[n\]either the social class of the adopting parents nor the number of years the child spent in the adopted family had any effect on the child's IQ" (p. 260) was also an overreach. For one thing, Frydman and Lynn merely presented bivariate correlations between IQ and whether the parents had a university degree, and between IQ and the number of years the adoptee spent with the adoptive family, and it is impossible to make a conclusive causal statement such as Rushton and Jensen's from those correlations alone. For another, whether or not parents have a university degree is a blunt measure of social class. For a third, the study's sample size was only 19, so the study would have had mediocre power to detect a correlation. This leaves the study of Clark and Hanisee \[[@B12-jintelligence-05-00001]\], in which 25 adoptees raised in the US had an average IQ-equivalent of 120 on the Peabody Picture Vocabulary Test (PPVT). *Intelligence: Tests and Reviews* \[[@B18-jintelligence-05-00001]\] (p. 930 or §7:417) shows that the PPVT dates from 1959, so the test norms were presumably 20--25 years old when the 25 adoptees were tested. I deduct 6--8 points accordingly for a Flynn-adjusted mean of 112--114. As this is far above the norm of 100, the Flynn effect does not wholly explain these adoptees' elevated IQ. Possibly the adoptees' PPVT scores were inflated further by the PPVT's unrepresentative standardization "on 4012 white children and youth in Nashville, Tennessee" \[[@B18-jintelligence-05-00001]\] (p. 752 or §6:530), but it is not obvious why that would explain all of the adoptees' extra 12--14 IQ points. One reviewer has intimated that I overcorrected for the Flynn effect in Clark and Hanisee's sample because the PPVT measured the "narrow ability" of verbal ability, while my 3-point-per-decade correction comes from broad IQ batteries. My counter-counterargument: verbal ability is broad, not narrow \[[@B21-jintelligence-05-00001]\], and even vocabulary tests can show a hefty Flynn effect. I am unaware of Flynn-effect data for the PPVT, but in the US National Longitudinal Survey of Youth, standardized scores on the PPVT-R increased by 0.41 standard deviations in 14 years, equivalent to 4.4 points per decade \[[@B22-jintelligence-05-00001]\]. Clark and Hanisee hazarded that the elevated IQ might "be the result of adoptive home environment. \[...\] Only 1 child was adopted into a family with an annual income of less than \$15,000, whereas 11 were placed in families earning more than \$25,000 per year (1978 figures). Both parents typically had college degrees" (p. 597). This would be consistent with a French study which found that abandoned children adopted into upper-middle-class families had IQs 11--16 points higher than half-siblings and full siblings who remained with their biological mother \[[@B23-jintelligence-05-00001]\]. Of course, the applicability of those results to East Asian children adopted in the US is not guaranteed, and other explanations of the Clark and Hanisee results exist. For instance, one could take the observed mean IQ of 120 and deduct only a few points each for the Flynn effect and the boost from adoption, attributing the remaining excess to a genetic advantage. One could even argue that were it not for the harsh environments the adoptees suffered early in life, their IQs would be even higher, suggesting an even greater genetic advantage. Most of the sample's Vietnamese adoptees "were evacuated from Vietnam during the last stages of United States military involvement in 1975" (p. 597), and all of the sample had lived in an orphanage, foster home, or hospital. Taking this route, Rushton and Jensen \[[@B3-jintelligence-05-00001]\] wrote that "half of the babies had required hospitalization for malnutrition" (p. 260), implying that the adoptees scored well in spite of needing hospital care. In this vein, I note that Clark and Hanisee \[[@B12-jintelligence-05-00001]\] actually report a correlation of +0.3 between hospitalization and PPVT score, and indeed a correlation of +0.7 between hospitalization and scores on another test, the Vineland Social Maturity Scale (p. 598). This completes my reexamination of these three East Asian adoptee studies. After allowing for the Flynn effect, the Clark and Hanisee sample (*N* = 25) had a mean vocabulary-as-IQ of 112--114, the Frydman and Lynn sample (*N* = 19) a mean VIQ of 99 and a mean PIQ of 100--103, and the Winick et al. sample (*N* = 111) a mean IQ of 96--107. These averages do not allow for the adoptive IQ boost, about which one may only speculate. In my judgement, allowing for the adoptive IQ boost would almost certainly bring the Frydman and Lynn averages below 100, would more likely than not bring the Winick et al. average below 100, and would shift the Clark and Hanisee average to approximately 105. Bearing the studies' sample sizes in mind, these averages imply a slight IQ *dis*advantage for East Asian transracial adoptees, although the evidence leans only modestly in this direction. Conservatively, I infer that East Asian transracial adoptees would score about as well as White adoptees raised in the same homes. 5. A Re-Analysis of Black Adoptee IQ Data {#sec5-jintelligence-05-00001} ========================================= Unlike the studies of East Asian adoptees, all of the Black adoptee studies include multiple groups with differing racial admixture, tested at about the same time on similar tests. At least in theory, this allows direct comparison of the groups within each study; since the groups all benefit from adoption's effect on IQ, and their IQs are inflated to similar degrees by the Flynn effect, there is less risk of these effects biasing the results in favour of one racial group over another. However, the biggest of the Black adoptee studies has other complications to untangle. That biggest study is the Minnesota Transracial Adoption Study (MTRAS). According to Rushton and Jensen, it is "also the only transracial adoption study \[of IQ\] that includes a longitudinal follow-up" \[[@B3-jintelligence-05-00001]\] (p. 256). For the study Sandra Scarr and colleagues located White Minnesotans who had adopted non-White children, and recorded the IQs of the adopters and their children (including the non-adopted White children). Scarr et al. measured the children's IQs in two waves, one when the children had a mean age of 7 and another when they had a mean age of 17. At both times the White adoptees scored higher than the Black--Black adoptees, and the Black--White adoptees scored between the White and the Black--Black adoptees \[[@B13-jintelligence-05-00001],[@B14-jintelligence-05-00001]\]. Not only that, but the measured interracial IQ differences grew between the two waves. Scarr and Weinberg \[[@B13-jintelligence-05-00001]\] reported differences in the first wave of 2.5 points between the White and Black--White (BW) adoptees, and 14.7 points between the White and fully Black adoptees; Weinberg et al. \[[@B14-jintelligence-05-00001]\] reported final differences of 7.1 points and 16.2 points respectively. Rushton and Jensen \[[@B3-jintelligence-05-00001]\] (p. 259) implied that this widening was a genetic effect: "although the shared-family environmental component of true-score IQ variance can be quite large at age 7, by late adolescence it is the smallest component. After that age, genetic and within-family (nonshared) environmental effects account for the largest components". To convince the reader, they pointed to their Figure 3, a plot estimating the proportions of IQ variation "attributable to genetic and environmental (shared and nonshared) effects" with respect to age (p. 252). However, as Richard Nisbett realized, that diagram indicates that "a greater genetic contribution to IQ occurs only after the age of 20" \[[@B24-jintelligence-05-00001]\] (p. 308), because it shows virtually constant heritability from age 6 to age 20. Rushton and Jensen contradicted their own cited graph. But perhaps the widening interracial differences in the MTRAS were genetically driven despite Rushton and Jensen's error? Probably not, because attrition can explain the apparent widening. A total of 25 White adoptees were in the study when it began, nine of whom were lost at follow-up. The lost adoptees had relatively low IQs, so the remaining White adoptees were unrepresentatively high in IQ, as Mackintosh observed \[[@B25-jintelligence-05-00001]\]. One can prove this by comparing the original IQs of the full sample and the subgroup who were measured at both ages 7 and 17; the latter subgroup had an initial mean IQ of 117.6 (with a minimum IQ of 92) but the full sample had an initial mean of 111.5 (minimum 62). Because initial and final IQs had a correlation of 0.63 among the White group, the elite subgroup would likely have had their final mean IQ inflated by about 0.63 × (117.6 − 111.5) = 3.8 points. Meanwhile, the BW and Black--Black adoptees lost to follow-up hardly differed in IQ from the remaining adoptees, so attrition inflated those groups' mean IQs by about only 0.2 and −0.7 points respectively. Adjusting the final mean IQs accordingly ([Table 2](#jintelligence-05-00001-t002){ref-type="table"}) implies smaller racial differences of 3.5 points (White vs. BW adoptees) and 11.7 points (White vs. Black--Black adoptees) in the study's final wave. The former is only 1 point wider than the corresponding initial difference, and the latter is 3 points narrower. Hence, allowing for attrition, the IQ differences between the White and the Black adoptees were no larger at age 17 than at age 7, a sign that the apparent enlarging was an artifact and not a genetic effect. With the widening explained, the only racial IQ differences left to comment on are those present at initial testing. The scant initial gap of 2.5 ± 3.5 points between the fully White and BW adoptees is small enough to be simple statistical noise. Only the IQ of the Black--Black adoptees, who scored 12.2 ± 2.8 points below the BW adoptees, calls for a specific explanation. Differences in home environment are one possibility. On every reported environmental variable, the Black--Black adoptees were worse off than both the BW and fully White adoptees, which I quantify by comparing the former against the BW adoptees, measuring the environmental differences in BW SDs. I use the BW adoptees as a comparison group here because Scarr and Weinberg \[[@B13-jintelligence-05-00001]\] present more data for BW adoptees than White adoptees. The Black--Black adoptees were older when adopted (by 2.1 SDs, or two years); had spent less time in their adoptive home (by 1.1 SDs); had more (by 0.4 SDs) and lower-quality (by 0.8 SDs) adoptive placements; and had adoptive parents with less education and lower mean IQ (by 0.2--0.3 SDs). Additionally, 97% of the BW adoptees had White mothers while the Black--Black adoptees all had Black mothers, with whatever prenatal environmental differences that entailed. Proponents of the hereditarian model have found the notion of confounding with home environment controversial. For instance, Lee \[[@B26-jintelligence-05-00001]\] (p. 253) found confounding "very doubtful" because "\[t\]here exists no independent evidence that variables such as age at adoption exert effects on IQ lasting until late adolescence", citing the van IJzendoorn et al. meta-analysis \[[@B17-jintelligence-05-00001]\]. However, as mentioned above, that meta-analysis erroneously summarized its studies, and so its analyses (being based on mis-estimated summary statistics) are untrustworthy. Even ignoring this problem, the meta-analysis claimed low power to detect an adoptive-age effect on IQ; the IQ differences associated with higher adoptive age had wide confidence intervals and a lot of heterogeneity (p. 311). Lee added that "the proportion of IQ variance associated with these pre-adoption variables declined over the course of the MTAS from .32 to .13". This is true, but I repeat that the only racial IQ differences in the MTRAS needing a special explanation are those measured at age 7, when the pre-adoptive variables had more explanatory power. Lee also made the reasonable if tentative argument that race and IQ themselves might "affect pre-adoption experience", in which case adjusting for pre-adoptive variables would be "perhaps overly generous towards an environmental hypothesis". He was correct, but this simply means the MTRAS results are ambiguous; making the adjustment may skew the results in favour of a non-hereditarian hypothesis, but not making the adjustment may skew the results in favour of a hereditarian hypothesis. A decisive, objective, and complete interpretation of the results is not possible. Malloy \[[@B27-jintelligence-05-00001]\] presented results from the MTRAS, writing that "no simple or plausible environmental theories \[...\] explain these kinds of findings", on the grounds that "\[s\]tudies do not support a large role for peer effects on developed intelligence" and that van IJzendoorn et al.'s meta-analysis "found that neither age at adoption or even coming from an abusive or neglectful environment had an effect on the developed IQ scores of adopted children" (p. 1088). As I do not invoke peer effects on IQ I need not comment on those, and I have already commented on the meta-analysis. I will add that the meta-analysis had poor power to detect the effect of abuse on IQ, which may explain why the abuse-associated deficit found (*d* = 0.22) was statistically insignificant. Lynn \[[@B2-jintelligence-05-00001]\] (p. 24) preempted one of Lee's comments by noting that "what appears to be an age-of-adoption effect may be only a race-differences effect" because correlations between adoptive age and IQ, and between time spent in the adoptive home and IQ, "are confounded with race differences". Again, this is possible, but simply means the study's results are ambiguous. (Below I also adduce evidence that adoptive age correlates negatively with IQ among East Asian transracial adoptees, where Lynn's proposed confounding is excluded.) Lynn makes additional arguments using results for the adoptees at age 17, but the age 7 results are again the pertinent ones. Rushton and Jensen \[[@B3-jintelligence-05-00001]\] zeroed in on one particular environmental variable: age at adoption. They referred to Jensen's 1998 book *The g factor*, which cited Fisch et al. \[[@B28-jintelligence-05-00001]\], a study supposedly "showing that age of adoption does not influence children's IQ scores after age 7" \[[@B3-jintelligence-05-00001]\] (p. 259). However, Nathan Brody \[[@B29-jintelligence-05-00001]\] (p. 403) noticed that this is a "somewhat tendentious interpretation" of Fisch et al.'s work. Briefly, Fisch et al. compared the IQs of 7-year-olds adopted by their first birthday and 7-year-olds who had been adopted later, discovering a statistically insignificant 4.4-point difference. However, it is unsurprising that this difference was statistically insignificant because "the small sample of \[seventeen\] adoptees older than 1 renders the power of the statistical test of the difference weak" \[[@B29-jintelligence-05-00001]\] (p. 402). Rushton and Jensen's inference that "age of adoption does not influence children's IQ scores after age 7" stands a good chance of having been a type II error. The next sentence of Rushton and Jensen's review was similarly tendentious: "Studies of severely malnourished, late-adopted, East Asian children (see below) provide substantial evidence that age of adoption does not adversely influence IQ in transracial adoptions" \[[@B3-jintelligence-05-00001]\] (p. 259). The East Asian adoptee studies they referred to are the three I discuss above, yet the adoptees in those studies were *not* "late-adopted" relative to the Black adoptees in the MTRAS, who were adopted at 18 months on average \[[@B13-jintelligence-05-00001]\] (p. 730). The Winick et al. \[[@B10-jintelligence-05-00001]\] (p. 1175) adoptees had a mean age at adoption of 18 months and the Frydman and Lynn \[[@B11-jintelligence-05-00001]\] (p. 1323) adoptees had a mean age at adoption of 19 months. Clark and Hanisee's paper \[[@B12-jintelligence-05-00001]\] does not record an average adoptive age, but its adoptees also don't seem to have been "late-adopted", as the investigators set an upper adoptive age limit of three years (p. 596), and 10 of its 25 adoptees "were relinquished at birth to adoption agencies" (p. 598). Rushton and Jensen also omitted mention of the negative correlations between adoptive age and IQ documented in Frydman and Lynn \[[@B11-jintelligence-05-00001]\] and Clark and Hanisee \[[@B12-jintelligence-05-00001]\]. Winick et al. \[[@B10-jintelligence-05-00001]\], which paid less attention to adoptive age, does not record an age-IQ correlation, but the follow-up study Lien et al. \[[@B30-jintelligence-05-00001]\] found a statistically significant negative relationship between academic achievement and age of arrival in the US for Korean adoptees. There are no features of the Lien et al. study which explain Rushton and Jensen's omission of it. Lien et al. \[[@B30-jintelligence-05-00001]\] is a study of Korean adoptees raised in the US with extremely similar design to that of Winick et al., the key difference being that the Lien et al. adoptees were at least two years old when adopted while the Winick et al. adoptees were adopted by age 3. Comparing mean IQs across the studies shows that this adoptive age difference was associated with a 5--7 point IQ deficit for Lien et al.'s later adoptees, regardless of nutritional status. All in all, confounding of adoptee race with environmental variables is a threat to the MTRAS results. Still another factor complicating the interpretation of the MTRAS results is a hard-to-predict Flynn effect, which seems to be caused by the use of different IQ tests for adoptees of different ages \[[@B13-jintelligence-05-00001],[@B14-jintelligence-05-00001]\] and the different age distributions of the White and the Black adoptees \[[@B13-jintelligence-05-00001]\] (p. 730). Loehlin \[[@B31-jintelligence-05-00001]\] (p. 185) presented mean IQs for the study's groups, "adjusted for norm shifts over time", but his tabulation of the data is too meagre to permit detailed analysis. The original data, which I analyze here, may be skewed by this Flynn effect. Correcting for it could conceivably eliminate the attrition effect while restoring the widening of racial IQ gaps over time, but there is little a priori reason to expect that. I have had to dwell on the MTRAS at length, but there are two more oft-cited Black adoptee studies. One is Tizard \[[@B15-jintelligence-05-00001]\], a one-page report spun off from a language-acquisition study. In that study 64 4½-year-olds took the Wechsler Pre-school and Primary Scale of Intelligence (WPPSI) IQ test, of whom 24 "had been adopted into white families at a mean age of 3.1 yr" (p. 316). A total of 17 adoptees were White and had a mean IQ of 113.0, and seven were BW and had a mean IQ of 119.9. The superior IQ of the adoptees with more Black ancestry reverses the main result of Scarr and Weinberg. In Scarr and Weinberg \[[@B13-jintelligence-05-00001]\], the BW adoptees lagged the White adoptees by 2.5 ± 3.5 points, while in Tizard \[[@B15-jintelligence-05-00001]\], the BW adoptees outscored the White adoptees by 6.9 ± 6.6 points. The other study is Moore's \[[@B16-jintelligence-05-00001]\], which assessed 23 Black--Black and BW adoptees, raised in White families, on the WISC. Like Tizard \[[@B15-jintelligence-05-00001]\] and unlike Scarr and Weinberg \[[@B13-jintelligence-05-00001]\], the adoptees with more Black ancestry had higher IQs: nine fully Black adoptees had a mean IQ of 118.0 and 14 BW adoptees had a mean IQ of 116.5. The resulting IQ difference is 1.5 ± 4.1 points, where the standard error is approximate because the standard deviations involved are pooled estimates. Taking an inverse-variance-weighted average of results from Scarr and Weinberg and Tizard, BW adoptees lagged White adoptees by 0.4 ± 3.1 IQ points. Taking an inverse variance-weighted average of results from Scarr and Weinberg and Moore, fully Black adoptees lagged BW adoptees by 7.8 ± 2.3 points, though this estimate assumes a homogeneity of results that doesn't exist. Taken at face value these results suggest that higher Black ancestry is associated with lower IQ among Black adoptees, but not when comparing BW to White adoptees. If one forces these two conflicting results together by taking a weighted average of the two weighted averages, they suggest an IQ drop of about 5 points associated with having an additional Black biological parent, but statistical heterogeneity renders this result suspicious. Another reason for suspicion comes from Moore's work, which also studied 23 Black adoptees raised in *Black* families. Those adoptees had a mean IQ 13.5 ± 3.1 points below the mean of the Black adoptees raised in White families, evidence for the importance of adoptees' home environment rather than adoptees' ancestry. A hereditarian might invoke heterosis (hybrid vigour) as an explanation for the heterogeneity---perhaps Black ancestry lowers IQ on average, with this effect cancelled out in BW children by an IQ gain from hybrid vigour. However, heterosis has too weak an effect to explain more than a bit of the heterogeneity \[[@B32-jintelligence-05-00001]\]. The above discussion of Black adoptees' IQs made one reviewer unhappy; they felt it was "selective" because "Flynn effect corrections are applied only to the East Asian groups, never to the Blacks". However, there is a solid methodological reason for this: one *must* make Flynn effect corrections to interpret the three studies of East Asians, because those studies lacked comparison groups of adoptees of other races, forcing a comparison of the East Asian adoptees to the general population norm. At the same time, the studies with Black adoptees contained multiple groups which could be compared to *each other*, and such comparisons need no Flynn effect correction. The one possible exception is the MTRAS, afflicted by a hard-to-predict Flynn effect mentioned above. The published MTRAS reports do not have enough information to correct for that Flynn effect, so I take the published data as given while warning that a Flynn effect might have skewed them. This is better than the reviewer's defective approach of taking sample-size-weighted racial averages of the means in my [Table 1](#jintelligence-05-00001-t001){ref-type="table"} (thereby double counting some of the data, because Weinberg et al.'s sample is a subset of Scarr and Weinberg's) and indiscriminately subtracting 10 points from each average (neglecting the fact that the Flynn effect inflated IQs to different degrees in different samples). 6. Summarizing My Re-Analyses {#sec6-jintelligence-05-00001} ============================= I have now reviewed all of the transracial adoption studies cited by peer-reviewed papers as part of the race and IQ debate. Reviewing the data with a less selective eye, and making allowances for the Flynn effect, adoptive IQ boosts, and attrition in a longitudinal study, the results are equivocal. One study of East Asian adoptees documents an IQ advantage, while two more suggest IQ disadvantages. A sample-size-weighted average of the results would give a slight IQ disadvantage, though this result could be strengthened or reversed depending on what one assumes about the adoptive IQ boost in these samples. Meanwhile, one study of Black adoptees documents an IQ disadvantage, while two more suggest IQ advantages. Merging the studies' results suggests an inconsistent IQ disadvantage, but that conclusion can only be a weak one since the studies are small, statistically heterogeneous, and open to alternative interpretations. The data are too patchy and inconsistent for conclusive inferences, but taken together they are evidence against the hereditarian hypothesis. The net racial IQ differences are small and some disagree with the hereditarian model's predictions. The nil hypothesis---that adoptees of different races have similar IQs when raised in the same environment---fits the data at least as well. For East Asian adoptees, my sample-size-weighted average IQ is about 97 (which I derive from midpoints of the IQ ranges in [Table 2](#jintelligence-05-00001-t002){ref-type="table"}, subtracting 6 points for the adoptive boost), which would plausibly be a bit higher were malnutrition allowed for. Based on the studies of Black--Black and BW transracial adoptees I expect them to have mean IQs of about 92 and 100 respectively, but these estimates are shaky, undermined by heterogeneity and the Black adoptees in Scarr and Weinberg \[[@B13-jintelligence-05-00001]\] having less salubrious rearing circumstances. The East Asian and Black averages could plausibly be within a couple of points of 100 if measured with more precision, and if no Black adoptees were raised in environments measurably unequal to those of White adoptees. The studies reviewed so far have proven inconclusive. An obvious next step is to seek more data to see how well competing hypotheses accommodate them. The study list in van IJzendoorn et al.'s [Table 1](#jintelligence-05-00001-t001){ref-type="table"} \[[@B17-jintelligence-05-00001]\] contains three more studies of the IQs of transracially adopted East Asians, namely Lien et al. \[[@B30-jintelligence-05-00001]\], Wattier and Frydman \[[@B33-jintelligence-05-00001]\], and Stams et al. \[[@B34-jintelligence-05-00001]\]. My own searches turned up another study, Loman et al. \[[@B35-jintelligence-05-00001]\], as well as Dalen et al. \[[@B36-jintelligence-05-00001]\] and Lindblad et al. \[[@B37-jintelligence-05-00001]\], two papers about entire cohorts of Swedish men. I now discuss the results of all six papers. 7. Lien et al., 1977 {#sec7-jintelligence-05-00001} ==================== Lien et al. \[[@B30-jintelligence-05-00001]\] essentially replicated Winick et al. \[[@B10-jintelligence-05-00001]\], but studied Koreans adopted later. It confirmed a dose--response relationship between malnourishment and IQ deficits: 31 "severely malnourished" adoptees had a mean IQ of 95, 62 "moderately malnourished" adoptees a mean IQ of 101, and 39 "well-nourished" adoptees a mean IQ of 105 (pp. 1735--1736). Between 1973 and 1975 the children took the same IQ tests as those in Winick et al., but there is no precise breakdown of how many took each test. As such, the Flynn effect inflation of IQs could have been minute or could have been as high as 12 points. Subtracting a middling estimate of 6 points from these adoptees' IQ averages implies an overall mean IQ of 95. Subtracting the effect of the adoptive IQ boost would shrink this average further, probably below 90. 8. Wattier and Frydman, 1985 {#sec8-jintelligence-05-00001} ============================ Wattier and Frydman \[[@B33-jintelligence-05-00001]\], like Frydman and Lynn \[[@B11-jintelligence-05-00001]\], considered the IQ of a few Asian children adopted in Belgium. From the abstract: "the authors present a study made in 1983 on a sample of 28 children from South-East Asia and adopted four to twelve years earlier by Belgian francofone \[sic?\] families living in the province of Hainaut". The rest of the paper is in French, making it hard for me to read it closely, but its summary table (on page 67 and unnumbered) is clear enough: the sample's mean score on the WISC/WPPSI was 116.6 with a standard deviation of 11.04. The mean VIQ was 110.8 and the mean PIQ was 119.8. Making the same Flynn effect adjustment as for Frydman and Lynn \[[@B11-jintelligence-05-00001]\] leads to a mean VIQ of 99 and a mean PIQ of 96--100, both less than the Belgian norm. Subtracting an adoptive boost effect would bring them lower. 9. Stams et al., 2000 {#sec9-jintelligence-05-00001} ===================== The Stams et al. study \[[@B34-jintelligence-05-00001]\] appears to be unique in having used one IQ test to compare East Asian transracial adoptees of both sexes to non-East Asian transracial adoptees. Specifically, in 1995 \[[@B38-jintelligence-05-00001]\], 100 Sri Lankan adoptees, 11 Colombian adoptees and 36 South Korean adoptees, all raised by "Caucasian white" parents, took the abbreviated Revised Amsterdam Child Intelligence Test (RAKIT) at age 7, with their raw scores converted to standardized IQs "derived from a representative sample of 1415 children between age 4 and 11, drawn from the Dutch general school population in 1982" \[[@B34-jintelligence-05-00001]\] (pp. 1026--1028). The children were all adopted reasonably close to birth, with the Sri Lankans having a mean adoptive age of 7 weeks (with an SD of 3 weeks) and the Koreans and Colombians a mean adoptive age of 15 weeks (SD 4 weeks). However, the Korean and Colombian adoptees "stayed in an institution or foster home after separation from their biological mother at birth, until their adoption", while the Sri Lankan children "were in the care of their biological mother until their adoption" (p. 1026). Although "\[l\]ittle is known about the child-rearing conditions of the Sri Lankian \[sic\] infants", "\[a\]necdotal evidence from parent reports" indicated that "pre- and post-natal care for the relinquishing mother and her baby were far from optimal \[...\] and the health condition of the mother was deplorable in many cases" (p. 1027). This may explain the relatively low mean IQ of 104 for the Sri Lankan adoptees, compared to mean IQs of 112 and 115 for the Colombian and Korean adoptees respectively. The adoptee--adoptee comparisons need no Flynn effect correction because the study's adoptees were tested approximately simultaneously. However, a correction is necessary if one wishes to compare the adoptees to the RAKIT's standardization sample. Wicherts et al. \[[@B39-jintelligence-05-00001]\] (pp. 524--525) report that Dutch 5-year-olds gained 2.8 IQ points on the test from 1982 to "1992/1993", though the 1992--1993 sample was marginally above average in SES; extrapolating that rate of gain across the 1982--1995 period, the Stams et al. adoptees' IQs were inflated by 3--4 IQ points. This just about explains the elevated mean IQ of the Sri Lankan adoptees, but not the higher means of the Colombian and Korean adoptees. 10. Loman et al., 2009 {#sec10-jintelligence-05-00001} ====================== Loman et al. \[[@B35-jintelligence-05-00001]\] wished to observe how post-institutionalized adoptees from outside the US ("who were adopted at 12 months of age or older and spent 75% or more of their preadoptive lives in institutional care") developed differently to "children internationally adopted early, predominantly from foster care" and non-adopted children "raised continuously in their biological families in the United States" (pp. 427--428). Their group of post-institutionalized adoptees was made up of 42 Russian and Eastern European (R&EE) adoptees, 41 Asian adoptees, seven South American adoptees, and a single Ethiopian adoptee. As part of the project, the researchers extrapolated the children's IQs from their performance on the WISC-III's block design and vocabulary subtests, except for the children who "scored \>1 SD below the mean on either WISC-III subtest". The latter children then took the Leiter International Performance Scale-Revised, and their IQ was taken as their "Leiter Brief IQ score" instead (p. 429). [Table 3](#jintelligence-05-00001-t003){ref-type="table"} of Loman et al. \[[@B10-jintelligence-05-00001]\] broke down "Estimated IQ" by country of origin. Its mean and standard deviation were 99.6 ± 15.0 for the R&EE adoptees (*n* = 41), 95.6 ± 17.1 for the South American adoptees (*n* = 7), and 107.7 ± 18.3 for the Asian adoptees (*n* = 40). Loman et al.'s *F*-test of these gave a statistically insignificant result ("F(2,88) = 1.56, NS"), and their paper inferred "no region-of-origin differences in estimated IQ" (p. 431). However, a Welch's *t*-test of the difference between the Asian and the R&EE adoptees (8.1 points with a standard error of 3.7 points) rejects the null hypothesis of zero difference (*p* = 0.033, although one can argue that the explicit significance level should be less than 0.05 to account for potential multiple comparisons). This difference is meaningfully large and arguably statistically significant, but there is an obstacle to calling it a White-East Asian racial difference. The R&EE adoptees were presumably all White (coming from Russia, Romania, Bulgaria, Slovakia, Ukraine, Moldova, and Poland) but the Asian subsample comprised 22 Chinese children, three Filipino children, a Vietnamese child, and 15 Indian children, so it was roughly a 2:1 mixture of East Asian and South Asian adoptees. The observed difference is therefore one between White adoptees and an agglomeration of East Asian and South Asian adoptees. I might also compare the adoptees to the general population, although the necessary Flynn effect adjustment is compromised by the administration of only two WISC-III subtests. The study's IQ testing took place between 2005 and 2007 \[[@B40-jintelligence-05-00001]\]. The WISC-III was standardized in 1989 \[[@B7-jintelligence-05-00001]\] (p. 214) and the LIPS-R in 1995 and 1996 to match the 1993 US Census \[[@B41-jintelligence-05-00001],[@B42-jintelligence-05-00001]\]. Therefore the WISC-III takers were supported by a Flynn effect of 4.8--5.4 points, and the LIPS-R takers by one of 2.7--3.6 points. Taking the midpoints of those two intervals, and weighting the midpoints using the fact that 23.1% of the post-institutionalized adoptees took the LIPS-R \[[@B35-jintelligence-05-00001]\] (p. 429), the post-institutionalized adoptees would have had their average IQ estimates inflated by 4.6 points. Subtracting that figure from each subgroup mean gives estimated IQ means and standard errors of 95.0 ± 2.3 for the R&EE adoptees, 91.0 ± 6.5 for the South American adoptees, and 103.1 ± 2.9 for the Asian adoptees. Incorporating the expected adoptive boost, the Asian adoptees' mean extrapolated IQ would probably be similar to that of the general population, and the mean extrapolated IQs of the R&EE and South American adoptees appreciably less. 11. Dalen et al., 2008, and Lindblad et al., 2009 {#sec11-jintelligence-05-00001} ================================================= Dalen et al. \[[@B36-jintelligence-05-00001]\] and Lindblad et al. \[[@B37-jintelligence-05-00001]\] are the only studies I have found which took a systematic national sample instead of relying on a convenience sample. I discovered the Lindblad et al. paper first but focus on Dalen et al.'s as its sample greatly overlaps Lindblad et al.'s and is over twice as large. Dalen et al.'s "study population was drawn from all male residents in Sweden born between 1968 and 1976 who were conscripted before 20 years of age \[...\] and were still residents in Sweden at follow-up in December 2001", and had "complete information on all four intelligence test variables" (p. 1213). They divided the bounty of this expansive sampling frame into six groups: Korean adoptees, non-Korean foreign adoptees, domestic adoptees born in Sweden, non-adopted siblings of foreign adoptees, non-adopted siblings of domestic adoptees, and the general population, defined as "Swedish-born offspring of two Swedish-born parents with no record of ever having adopted a child". When registering for conscription, these men took "Enlistment battery 80", "an intelligence test" which gave "a global score derived from the four subtests"; the global scores "had a Gaussian distribution of scores between 1 and 9" (p. 1214). Dalen et al.'s [Table 2](#jintelligence-05-00001-t002){ref-type="table"} gives the mean and standard deviation of the global scores within all six groups, which I transform to an IQ scale, taking the general population's mean and standard deviation as 100 and 15 by definition. [Table 3](#jintelligence-05-00001-t003){ref-type="table"} summarizes the results. Lindblad et al. mention (p. 303) that the non-Korean foreign adoptees were a mixture of Black (Ethiopian) adoptees, East Asian (Thai) adoptees, and adoptees who were neither Black nor East Asian nor White (Indian, Chilean, Sri Lankan, Colombian, and Ecuadorian adoptees, among others), so the non-Korean foreign adoptees are a suboptimal comparison group for the Korean adoptees. The domestic adoptees, however, were presumably almost all White, and they scored 6.2 points below the Korean adoptees. Selective placement likely plays a part in the latter result, as the domestic adoptees' non-adopted siblings scored 7.5 points below the foreign adoptees' non-adopted siblings. I may also compare the Korean adoptees to the general population. (Unlike the older studies of East Asian adoptee IQ, the Flynn effect is irrelevant here because the adoptees were tested over the same period as the general population.) The Korean men scored only 1.5 IQ points above Swedish sons of non-adopting Swedish parents. This is rather less than the "10 or more points higher than \[Korean and Vietnamese adoptees'\] adoptive national norms" inferred by Rushton and Jensen \[[@B3-jintelligence-05-00001]\] (p. 276) from older adoption studies. Indeed, although statistically significant because of the samples' large size, the 1.5-point difference is scarcely practically significant, and would be reversed were one to allow for the usual adoptive IQ boost, especially a boost enhanced by selective placement. 12. Summarizing the Additional Studies {#sec12-jintelligence-05-00001} ====================================== Lien et al. \[[@B30-jintelligence-05-00001]\] and Wattier and Frydman \[[@B33-jintelligence-05-00001]\] suggest a mean East Asian transracial adoptee IQ below 100. Stams et al. \[[@B34-jintelligence-05-00001]\] is harder to interpret; although that study's East Asian adoptees were tested alongside non-East Asian adoptees, the non-East Asian adoptees do not fit cleanly into the usual hereditarian racial framework of Whites, Blacks, and East Asians, so it's unclear how to test the hereditarian hypothesis against these data. In any event, its 36 South Korean adoptees had a mean IQ 3 points higher than 11 Colombian adoptees, and 11 points higher than 100 Sri Lankan adoptees. These differences cannot be explained in terms of adoptive age, although the Sri Lankans may have suffered less healthy birth environments. Overall, the Colombian--Korean IQ difference remains marginal evidence in favour of the hereditarian hypothesis and against the nil hypothesis. Loman et al. \[[@B35-jintelligence-05-00001]\] likewise hints at a higher IQ for Asian adoptees than for White and South American adoptees, but that finding is blurred by the Asian subsample including South Asians, the study's use of incomplete test batteries, and the Flynn effect's ability to explain most of the apparent Asian adoptee advantage. Dalen et al. \[[@B36-jintelligence-05-00001]\] and Lindblad et al. \[[@B37-jintelligence-05-00001]\], the largest studies of East Asian transracial adoptees' IQs, found those IQs were only marginally higher than those of Swedish men in general, suggesting a mean IQ below 100 in the absence of the adoptive boost. Although Stams et al. \[[@B34-jintelligence-05-00001]\] and Loman et al. \[[@B35-jintelligence-05-00001]\] find that East Asian transracial adoptees outscored other adoptees, the overall drift of the additional studies is that East Asian transracial adoptees do not tend to have unusually good IQs in their adoptive nations. Appearances to the contrary are a result of ignoring their atypically salutary adoptive homes and of failing to accommodate the Flynn effect. 13. Conclusions {#sec13-jintelligence-05-00001} =============== Drawing together this paper's re-analyses, I conclude that East Asian adoptees raised by Western Whites score about on par with non-adopted Western Whites, and that there is no consistent IQ difference between Black adoptees raised by Whites and White adoptees raised by Whites. Meanwhile, some studies document East Asian adoptee samples with higher IQs than non-East Asian adoptee samples, but it is not clear that any offer a clean comparison of East Asian adoptees and White or Black adoptees in similar environments on complete IQ batteries. These inferences must be provisional because the studies give conflicting results, most of the studies are small, and all are methodologically flawed. Only two of the papers reviewed here drew on samples of foreign adoptees that were nationally representative, and then only of males, and *then* only of the adoptees in the destination country, not of inhabitants of their birth country. Indeed, given the obvious difficulties with ethically taking a random sample of newborns in one country and having them adopted into random foreign homes, it seems unlikely that any fully representative study will ever be done. It follows that transracial adoption studies are unlikely to conclusively settle the race and IQ debate, since commentators may always level the valid methodological objection of unrepresentative sampling. It is nonetheless worthwhile to correct misleading claims about transracial adoptee IQ data, which are still made: see Christainsen's remark that "East Asians growing up in white households in the US and Belgium have tended to score considerably above the white mean in terms of intelligence" \[[@B43-jintelligence-05-00001]\] (p. 168). (Christainsen wrote next that "East Asians, or at least the Chinese among them, also tend to be relatively quiescent *independently of their child-rearing environment* and also have less variable heart rates (Kagan, Resnick, \[sic\] and Snidman 1988)". However, Kagan, Reznick and Snidman \[[@B44-jintelligence-05-00001]\] (p. 167) studied "three cohorts of Caucasian children from working- and middle-class Boston homes", not Chinese children.) As well as correcting specific claims, my re-analyses enable a fresh comparison of the hereditarian model to the data. Provisionally, the hereditarian model fails to fit the data when one applies the level of standard applied by hereditarians such as Rushton, Jensen and Lynn. For instance, Rushton and Jensen \[[@B3-jintelligence-05-00001]\] (p. 276) wrote that "\[t\]he culture-only model cannot explain" the "finding" that "Korean and Vietnamese children adopted into White homes, even though as babies many had been hospitalized for malnutrition, nonetheless grew to have IQs 10 or more points higher than their adoptive national norms". That comparison was fallacious because it neglected the Flynn effect and the unrepresentative environments in which adoptees live. Allowing for both effects, I estimate that East Asian adoptees tend to have IQs about equal to the relevant norms, and possibly a little below. This is what a nil hypothesis, and presumably a "culture-only model", would predict, but it violates the hereditarian expectation of superior East Asian IQ. Contrary to Rushton and Jensen's \[[@B3-jintelligence-05-00001]\] (p. 276) allegation that "support for the hereditarian model again comes from adding the East Asian data to the mix", the hereditarian model has at least as much trouble with the East Asian data as with the Black data. The model is not definitively ruled out; the data are too weak for that. However, a hypothesis that fits these data, at least as well, is the nil hypothesis: adoptees of different races would have similar IQs if raised in the same environment. To the extent that the nil hypothesis is true, genes are not so likely to be the main cause of racial IQ differences. The author declares no conflict of interest. jintelligence-05-00001-t001_Table 1 ###### Studies of IQs of transracial adoptees adopted by Whites: unadjusted results. Study Group *N* Age at Testing Reported IQ ------------------------------------------ ----- ---------------- ------------- ---------- *Tizard, 1974* White 17 4.5 113.0 7.8 Black--White 7 4.5 119.9 16.6 *Winick, Meyer, and Harris, 1975* East Asian, "malnourished" 36 ≥6 102 9.7 ^b^ East Asian, "moderately nourished" 38 ≥6 106 12.5 ^b^ East Asian, "well-nourished" 37 ≥6 112 9.0 ^b^ *Scarr and Weinberg, 1976* White 25 5--21 111.5 16.1 Black--White 68 4--16 109.0 11.5 Black--Black 29 4--16 96.8 12.8 *Clark and Hanisee, 1982* East Asian 25 2.5--6 120 16 *Moore, 1986* Black--White 14 7--10 116.5 8.2 ^c^ Black--Black 9 7--10 118.0 10.5 ^c^ *Frydman and Lynn, 1989* East Asian 19 6--14 118.7 10.4 *Weinberg, Scarr, and Waldman, 1992* ^a^ White 16 15--31 105.6 14.9 Black--White 55 14--26 98.5 10.6 Black--Black 21 14--26 89.4 11.7 ^a^ Weinberg, Scarr, and Waldman's 1992 paper is a longitudinal follow-up to Scarr and Weinberg's 1976 paper, so the former's sample is a subset of that presented in the latter; ^b^ Standard deviations for Winick, Meyer, and Harris \[[@B10-jintelligence-05-00001]\] read off graphically from a digital copy of their paper; my estimates are probably accurate to 0.4 IQ points; ^c^ Pooled standard deviations computed from separate standard deviations for male and female subgroups. jintelligence-05-00001-t002_Table 2 ###### Studies of IQs of transracial adoptees adopted by Whites: partially adjusted results. Study Group *N* Mean IQ Adjusted for --------------------------------------- ---------------- ------------- -------------- -------------- *Tizard (1974)* White 17 113.0 -- nothing Black--White 7 119.9 -- nothing *Winick, Meyer, and Harris (1975)* East Asian, "malnourished" 36 102 91--102 Flynn effect East Asian, "moderately nourished" 38 106 95--106 Flynn effect East Asian, "well-nourished" 37 112 101--112 Flynn effect *Scarr and Weinberg (1976)* White 25 111.5 -- nothing Black--White 68 109.0 -- nothing Black--Black 29 96.8 -- nothing *Clark and Hanisee (1982)* East Asian 25 120 112--114 Flynn effect *Moore (1986)* Black--White 14 116.5 -- nothing Black--Black 9 118.0 -- nothing *Frydman and Lynn (1989)* East Asian 19 110.6 (VIQ) 99 (VIQ) Flynn effect 123.5 (PIQ) 100--103 (PIQ) *Weinberg, Scarr, and Waldman (1992)* White 16 105.6 101.8 attrition Black--White 55 98.5 98.3 attrition Black--Black 21 89.4 90.1 attrition jintelligence-05-00001-t003_Table 3 ###### Cognitive-test scores of Swedish men, adopted and non-adopted. Study Group Dalen et al. (IQ) Lindbland et al. (Stanine) ----------------------------------------- ------------------- ---------------------------- ------ ----------- ----------- General population 342,526 100 15 142,024 5.26 Korean adoptees 780 101.5 15.2 320 5.31 Non-Korean foreign adoptees 1558 88.6 13.7 1125 3.96 Foreign adoptees' non-adopted siblings 357 108.5 14.2 190 6.17 Domestic adoptees 1153 95.3 15.1 not given not given Domestic adoptees' non-adopted siblings 286 101.0 15.9 not given not given Results extracted from [Table 2](#jintelligence-05-00001-t002){ref-type="table"} of Lindblad et al.'s report \[[@B37-jintelligence-05-00001]\] (which gives mean scores on an apparent stanine scale, but not standard deviations, preventing conversion of the means to an IQ scale) and computed from [Table 2](#jintelligence-05-00001-t002){ref-type="table"} of Dalen et al.'s report \[[@B36-jintelligence-05-00001]\].

All relevant data can be found at the following URL: <https://doi.org/10.6084/m9.figshare.4640092.v1>. Introduction {#sec001} ============ Length-weight relationships (LWR) are used for estimating the weight corresponding to a given length \[[@pone.0171811.ref001]\]. As most observations in fisheries surveys are typically obtained as the number of specimens and length of each sampled specimen, LWR are widely used to transform the survey data into estimates of the biomass which is important for modelling aquatic ecosystems. Therefore, estimation of the LWR is common practice and essential in fisheries science \[[@pone.0171811.ref002]\]. There are two parameters in the LWR model (*W* = *a*×*L*^*b*^), the coefficient *a* and the exponent *b*. Parameter *a* is the condition factor, describing body shape which can be classified as four groups: eel-like, elongated, fusiform and short and deep \[[@pone.0171811.ref002]\]. Parameter *b* is the allometric growth parameter, which indicates isometric growth in body proportions if *b*≥3 where fish have more girth as it grows longer; the species tends to be more streamlined if the exponent *b*\<3 \[[@pone.0171811.ref002]\]. Numerous studies found that factors, such as geographic, seasonal, inter-annual, and environmental conditions, can affect the estimates of *a* and *b* in the LWR model \[[@pone.0171811.ref002]--[@pone.0171811.ref005]\]. Instead of constructing several models reflecting various situations (different regions and years), it is more reasonable to make use of generalized linear mixed model to cover all the spatial and temporal effects in a single model. Linear mixed-effects modelling, a mature model in the statistics community, has been used in multiple fields. Cnaan et al. (1997) provided two detailed case studies to sufficiently introduce the use of the general linear mixed effects model for the regression analysis of correlated data \[[@pone.0171811.ref006]\]. Xu et al. (2015, 2014) developed nonlinear mixed-effects model to study the individual-tree diameter growth and linear mixed effects model for individual-tree crown width of China-Fir trees in Southeast China \[[@pone.0171811.ref007], [@pone.0171811.ref008]\]. Baayen et al. (2008) provided an introduction of mixed-effects models and illustrated its advantages, which would allow the researcher to simultaneously consider all factors that potentially contribute to the understanding of the structure of the data \[[@pone.0171811.ref009]\]. The linear mixed-effects model provides a powerful and versatile approach to analyze a wide variety of data structures, in which the linear predictor contains random effects in addition to the usual fixed effects \[[@pone.0171811.ref007]\]. The random effects vary with respect to one or more grouping variables, e.g. regions and years, adding its contribution of variation to the residual error, which can account for the additional resources of random variation \[[@pone.0171811.ref007], [@pone.0171811.ref008], [@pone.0171811.ref010]\]. Yellow Croaker (*Larimichthys polyactis*) is a warm-temperate demersal fish species widely distributed throughout the northwest Pacific Ocean. As a commercially important species, Yellow Croaker experienced heavy fishing pressure in China. From catch data derived from *China Fishery Statistical Yearbook*, we found the variations of Yellow Croaker stock ([S1 Fig](#pone.0171811.s001){ref-type="supplementary-material"}). Yellow Croaker were abundant in 1950s and the beginning of 1960s; but the stock collapsed in the 1970s \[[@pone.0171811.ref011]\]. After a series of restoration efforts (i.e. seasonal fishery closure and protection of the spawning ground), the Yellow Croaker stock has been recovering since 1990 and the catch has increased continually in the subsequent two decades \[[@pone.0171811.ref012]--[@pone.0171811.ref014]\]. However, the characteristics of the stock have changed; as individuals are smaller, younger and reaching maturity earlier \[[@pone.0171811.ref013],[@pone.0171811.ref015],[@pone.0171811.ref016]\]. Quite many scientists have done much work about the growth of Yellow Croaker around the world. Li *et al*. (2013) applied simple linear regression to analyze LWRs of Yellow Croaker in Bohai Sea-Northern Yellow Sea from 1960 to 2004 and in the Southern Yellow Sea from 1960 to 2010 for male and female respectively \[[@pone.0171811.ref003]\]. Zhang *et al*. (2010) investigated the biological characteristics of Yellow Croaker in the central and southern Yellow Sea, including the LWRs in 1960, 1985, 1998 and 2008 \[[@pone.0171811.ref017]\]. Lin *et al*. (2004) studied the population biology of Yellow croaker in the East China Sea, which evaluated the LWR in 1963, 1983 and 2001 \[[@pone.0171811.ref018]\]. All these studies of LWR used simple linear regression to develop region specific or year specific models, rather than linear mixed-effects models, to detect the spatial and temporal variations of Yellow Croaker. In this study, our objective was to analyze the biological characteristics of Yellow Croaker in recent years. Specifically, we estimated the LWR of Yellow Croaker used linear mixed-effects model, condition factors relative to reference years, and explore the possible relationship between parameters in LWR model and environmental factors, based on the observations along north coast of China among 2008 and 2011 to 2015. Materials and methods {#sec002} ===================== Data collection {#sec003} --------------- Specimens were collected along the Shandong and Jiangsu province through 2008, and 2011--2015. These regions along northern Chinese coast include Yellow River Estuary (YE), Coastal Waters of Northern Shandong (NS), Jiaozhou Bay (JB), Coastal Waters of Qingdao (QD), Haizhou Bay (HB), and South Yellow Sea (SY) ([Table 1](#pone.0171811.t001){ref-type="table"}, [S1 Table](#pone.0171811.s006){ref-type="supplementary-material"}, [Fig 1](#pone.0171811.g001){ref-type="fig"}), which are important spawning and feeding grounds of Yellow Croaker \[[@pone.0171811.ref019]\]. The maps of surveying area were plotted using package maps and mapdata of the R (version: R i386 3.3.1) \[[@pone.0171811.ref020], [@pone.0171811.ref021]\]. 10.1371/journal.pone.0171811.t001 ###### Sample size and location of Yellow Croaker among regions and years. {#pone.0171811.t001g} Regions 2008 2011 2012 2013 2014 2015 Total ------------------------------------- ---- ------ ------ ------ ------ ------ ------ ------- Yellow River Estuary YE 40 40 Coastal Waters of Northern Shandong NS 35 35 Jiaozhou Bay JB 426 26 63 515 Coastal Waters of Qingdao QD 419 533 952 Haizhou Bay HB 921 81 100 579 1681 South Yellow Sea SY 148 11 308 **Total** 426 947 482 121 816 590 3382 The second column shows the abbreviations of regions. {ref-type="table"}.](pone.0171811.g001){#pone.0171811.g001} In Coastal waters of Northern Shandong, specimens were randomly collected from fishermen immediately after landing. For all other five regions, stratified random bottom trawling surveys were implemented to collect samples. In total, 3,275 individuals were collected, with sample size variations among years and regions (see [Table 1](#pone.0171811.t001){ref-type="table"} and [S1 Table](#pone.0171811.s006){ref-type="supplementary-material"}). For each individual, standard length was measured to the nearest mm and total weight was measured to the nearest gram. The surveys, which held in the *Seasonal Fishery Closure* (June, July and August) were approved by the Department of Marine and Fishery of Shandong and Jiangsu Province. No specific permissions were required for all the other surveys, because these surveys are traditional surveys, which did not cover any marine protected area or private area and this study did not involve endangered or protected species. LWR modelling {#sec004} ------------- We fitted an exponential function to the weight at length data that took the form \[[@pone.0171811.ref001]\]: $$W = aL^{b}$$ where *W* is the wet weight of an individual fish (g), *L* is the standard length (cm), *a* is the condition factor, and *b* is the allometric growth parameter. Because the variance of W increases when L increases, above equation was log-transformed and the equation becomes: $$\text{ln}\left( W \right) = \text{ln}\left( a \right) + b \times \text{ln}\left( L \right)$$ We fit a generalized linear model (GLM), simple linear models for individuals in different regions and years (SLMR and SLMY for regions and years respectively), as well as nine linear mixed effect models (LMEM) that treated region and/or year as random effects to coefficient *a* and/or the exponent *b* to explain the relationship between length and weight ([Table 2](#pone.0171811.t002){ref-type="table"}) \[[@pone.0171811.ref007]\]. Analysis of Variance performed by *F*-test between GLM and LMEM was computered to tests whether the temporal and spatial variations were significant. All these modelling processes were conducted, using package lme4 the R (version: R i386 3.3.1) \[[@pone.0171811.ref022]\]. Bootstrap the data with replacement for 1000 times was used to estimate the distributions and statistics of parameter estimates in these models. 10.1371/journal.pone.0171811.t002 ###### Models for length-weight relationships of Yellow Croaker. {#pone.0171811.t002g} Models Log-Transformed AIC ΔAIC MAE --------- ----------------------------------------------------------- -------------------------------------------------------- ------- ------ ------- GLM W = *a*\*L\^*b* ln(W) = ln(*a*)+*b*\*ln(L) -3777 581 0.106 SLMR W = *a*~*i*~\*L\^*b*~*i*~ ln(W) = ln(*a*~*i*~)+*b*~*i*~\*ln(L) -4002 356 0.097 SLMY W = *a*~*j*~\*L\^*b*~*j*~ ln(W) = ln(*a*~*j*~)+*b*~*j*~\*ln(L) -4254 104 0.102 R.I W = (*a*\*exp(ReR.I))\*L\^*b* ln(W) = (ln(*a*)+ReR.I)+*b*\*ln(L) -4005 353 0.102 Y.I W = (*a*\*exp(ReY.I))\*L\^*b* ln(W) = (ln(*a*)+ReY.I)+*b*\*ln(L) -4220 138 0.098 R&Y.I W = (*a*\* exp(ReR.I)\* exp(ReY.I))\*L\^*b* ln(W) = (ln(*a*)+ReR.I+ReY.I)+*b*\*ln(L) -4341 17 0.098 R.S W = *a*\*L\^(*b*+ ReR.S) ln(W) = ln(*a*)+(*b*+ReR.S)\*ln(L) -3999 359 0.102 Y.S W = *a*\*L\^(*b*+ ReY.S) ln(W) = ln(*a*)+(*b*+ReY.S)\*ln(L) -4212 146 0.098 R&Y.S W = *a*\*L\^(*b*+ReR.S+ReY.S) ln(W) = ln(*a*)+(*b*+ReR.S+ReY.S)\*ln(L) -4334 24 0.098 R.I&S W = (*a*\* exp(ReR.I))\*L\^(*b*+ReR.S) ln(W) = (ln(*a*)+ReR.I)+(*b*+ReR.S)\*ln(L) -4009 349 0.102 Y.I&S W = (*a*\* exp(ReY.I))\*L\^(*b*+ReY.S) ln(W) = (ln(*a*)+ReY.I)+(*b*+ReY.S)\*ln(L) -4267 91 0.097 R&Y.I&S W = (*a*\* exp(ReR.I)\* exp(ReY.I))\*L\^(*b*+ReR.S+ReY.S) ln(W) = (ln(*a*)+ReR.I+ReY.I)+(*b*+ReR.S+ReY.S)\*ln(L) -4358 0.095 The first column shows the abbreviations of models detailed in the second and third columns. L: standard length in centimeters; W: whole weight. R.I: random effects on intercept (ln(*a*)) of Regions (HB, JB, NS, QD, SY, and YE); R.S: random effects on slope (*b*) of Regions; Y.I: random effects on intercept (ln(*a*)) of Years (2008, 2011, 2012, 2013, 2014, and 2015); Y.S: random effects on slope (*b*) of Years. *i* is the *i*th region; *j* is the *j*th year. ΔAIC = AIC of models---AIC of best fitted model (R&Y.I&S). We explored the influence of marine environmental status on variability of the LWR. Environmental data were drawn from *Bulletin of Marine Environmental Status of China*, which were reported by State Oceanic Administration People's Republic of China \[[@pone.0171811.ref023]\]. According to *Sea Water Quality Standard* (National Standard GB 3097--1997) \[[@pone.0171811.ref024]\], five levels (the first, second, third, fourth and worse than fourth levels) were used to evaluate the water quality, in which higher level indicated poorer quality. The polluted water area percentage (PWAP), which included the third, fourth and worse than fourth level water, was used to exemplify water quality. We explored the influence of PWAP on the LWR along north coast of China during 2008, 2011 to 2015. Model comparison {#sec005} ---------------- Both the Akaike information criterion (AIC) and the mean absolute error (MAE) were calculated to compare the performance of the twelve candidate models. AIC was widely used to compare the quality of models \[[@pone.0171811.ref025],[@pone.0171811.ref026]\]. Lower AIC value indicates a better model. MAE is a quantity used as a measurement on how close forecasts or predictions are to the eventual outcomes \[[@pone.0171811.ref027]\]. MAE was the average of the absolute errors: $$MAE = \frac{1}{n}\sum\limits_{i = 1}^{n}\left( \left| {f_{i} - y_{i}} \right| \right)$$ where *f*~*i*~ was the estimation and *y*~*i*~ the observed value. Relative condition factor {#sec006} ------------------------- Relative condition factor (*K*) was calculated for each individual \[[@pone.0171811.ref028]\]: $$K = \frac{W}{aL^{b}}$$ where *W* and *L* were observed weight and standard length from our surveys, baseline model LWR with parameters *a* and *b* were drawn from historical references. Relative condition factor was recommended to measure the deviation of an individual from the baseline weight for length in the respective sample and to investigate changes in the stock over time \[[@pone.0171811.ref002],[@pone.0171811.ref028]\]. Here the LWR models of Yellow Croaker in 1960, 1986, 2005, 2007, 2008.04\~2009.03 (12 months) and 2010 were introduced to be the baselines ([S2 Table](#pone.0171811.s007){ref-type="supplementary-material"}) \[[@pone.0171811.ref003],[@pone.0171811.ref011]\]. In 1960, Yellow Croaker were abundant and the stock was not deeply disturbed by fishing \[[@pone.0171811.ref013]\]. However, Yellow Croaker experienced severe fishing pressure and reached its lowest biomass in 1986 \[[@pone.0171811.ref014]\]; after recovery, the catch of the stock was increasing in recent years \[[@pone.0171811.ref012],[@pone.0171811.ref014]\]. The current condition factors relative to reference years of 1960, 1986, 2005, 2007, 2008\~2009 and 2010 were represented by *K*~*cur/1960*~, *K*~*cur/1986*~, *K*~*cur/2005*~, *K*~*cur/2007*~, *K*~*cur/2008\ 9*~ and *K*~*cur/2010*~. Results {#sec007} ======= Length and weight of Yellow Croaker {#sec008} ----------------------------------- The length frequency of collected samples of Yellow Croaker is shown in [S2 Fig](#pone.0171811.s002){ref-type="supplementary-material"}. The maximum, minimum and mean (95% CI) lengths of the Yellow Croaker sample were 21.0 cm, 2.6 cm and 10.9 cm (5.1, 15.5 cm), respectively. The most commonly encountered length group was 11.5--12.4 cm, followed by length group 10.5--11.4 cm. The sample was dominated by individuals ranging from 8 cm to 14 cm in length, with variations among regions and years ([Fig 2](#pone.0171811.g002){ref-type="fig"}). {ref-type="table"}.](pone.0171811.g002){#pone.0171811.g002} Lengths of most individuals in JB, QD, NS, and YE were longer than 10 cm, ranging from 8--14 cm in HB and 6--12 cm in SY. Both the length and the weight had a positive relationship with latitude, where larger individuals were more common in northern than southern regions. Most lengths in 2008, 2011 and 2013 were longer than 10 cm and within 8--14 cm in 2012 and 2014, but 5--12 cm in 2015. Individuals in 2013 were larger than fish collected in other years, while a third of these specimen caught in Yellow River Estuary. All specimens were collected in the south area (Haizhou Bay and Southern Yellow Sea) in 2015 and were noticeably smaller in length and weight ([Fig 2](#pone.0171811.g002){ref-type="fig"}), might because of individual growth or the different age groups among specimens in different regions. Recommended LWR {#sec009} --------------- The LMEM (R&Y.I&S) with Regions and Years have random effects on both intercept and slope was recommended to be used as the best model after comparing with the other models based on AIC and MAE values (-4357 and 0.095, respectively) ([Table 2](#pone.0171811.t002){ref-type="table"}). According to this selected model, *a* and *b* were estimated as 0.0192 (95% CI = 0.0178, 0.0308) and 2.917 (95% CI = 2.731, 2.945), respectively ([Table 3](#pone.0171811.t003){ref-type="table"}). The estimates of *a* with the random effects from Region and Year ranged from 0.013--0.023, while the estimates of *b* with the random effects from Region and Year ranged from 2.835--3.017. 10.1371/journal.pone.0171811.t003 ###### Estimated values of parameter *a* and *b* in GLM and LEMM models. {#pone.0171811.t003g} Models *a* *b* --------- -------- -------- ------------------ --------- ------- ------- ---------------- ------- GLM 0.0170 0.0170 (0.0163, 0.0177) 0.00035 2.961 2.961 (2.944, 2.979) 0.009 R.I 0.0180 0.0180 (0.0171, 0.0189) 0.00047 2.934 2.934 (2.913, 2.956) 0.011 Y.I 0.0180 0.0181 (0.0172, 0.019) 0.00046 2.936 2.936 (2.915, 2.956) 0.011 R&Y.I 0.0189 0.0189 (0.0180, 0.0199) 0.00049 2.921 2.921 (2.900, 2.941) 0.010 R.S 0.0178 0.0178 (0.0170, 0.0187) 0.00043 2.938 2.938 (2.918, 2.957) 0.010 Y.S 0.0179 0.0179 (0.0170, 0.0188) 0.00045 2.940 2.940 (2.919, 2.960) 0.011 R&Y.S 0.0184 0.0184 (0.0175, 0.0193) 0.00047 2.930 2.930 (2.909, 2.953) 0.011 R.I&S 0.0195 0.0205 (0.0179, 0.0269) 0.00234 2.906 2.887 (2.783, 2.938) 0.039 Y.I&S 0.0176 0.0176 (0.0165, 0.0188) 0.00060 2.947 2.947 (2.920, 2.973) 0.014 R&Y.I&S 0.0192 0.0218 (0.0178, 0.0308) 0.00362 2.917 2.871 (2.731, 2.945) 0.058 Estimates in the best fitted model; mean, 95% confidence interval and standard deviation got from models after bootstrap the data. Spatial and temporal variations of LWR {#sec010} -------------------------------------- The selected model with random effects from both regions and years applied to both *a* and *b*, indicated that there were substantial spatial and temporal variations of LWR for Yellow Croaker ([Table 4](#pone.0171811.t004){ref-type="table"}, [Fig 3](#pone.0171811.g003){ref-type="fig"}, [S3](#pone.0171811.s003){ref-type="supplementary-material"} and [S4](#pone.0171811.s004){ref-type="supplementary-material"} Figs). The results of analysis of vriance between the selected LMEM and the global model GLM suggested that the spatial and temporal variations of LWR of Yellow Croaker are significant (F = 111434, Df = 1, *P*\<0.001). The *a* value in the length-weight model, decreased with latitude (32.75°-38.15° N), while the parameter *b* value increased with latitude. The estimates for Coastal Waters of Qingdao and Jiaozhou Bay were similar, while their latitudes covered 35.33°-36.49° N and 35.99°-36.13° N, respectively. Variations among regions were 0.0034 and 0.0055 for *a* and *b*, respectively; but were larger among years (0.0081 and 0.1791 for *a* and *b*, respectively). 10.1371/journal.pone.0171811.t004 ###### Estimates values for parameters *a* and *b* in the LMEM (R&Y.I&S). {#pone.0171811.t004g} Effects ln(*a*) *a* *b* ------------------------------------- ---------------------- --------- --------- --------- Fixed effects -3.954 0.019 2.917 Random Effects Yellow River Estuary 0.093 1.098 -0.0029 Coastal Waters of Northern Shandong 0.006 1.006 -0.0002 Coastal Waters of Qingdao 0.017 1.017 -0.0005 Jiaozhou Bay 0.012 1.012 -0.0004 Haizhou Bay -0.045 0.956 0.0014 South Yellow Sea -0.083 0.920 0.0026 2008 0.106 1.111 -0.0500 2011 0.065 1.068 -0.0109 2012 -0.041 0.960 0.0472 2013 0.165 1.179 -0.0791 2014 -0.284 0.752 0.1000 2015 -0.011 0.989 -0.0072 Fixed values of ln(*a*), *a* and *b*; random effects from six regions and six years respectively. {#pone.0171811.g003} Instead of a gradual trend of spatial variation, temporal variation was out of order, with complex influence. The estimate of *a* reached its maximum value (0.022) in 2013, declined by 0.001 through 2008, 2011, 2015 and 2012, and had its minimum value (0.014) in 2014. The largest estimate of parameter *b* appeared in 2012, followed by the estimate (2.96) in 2012, while the minimum estimates was 2.84 in 2013. In Yellow Sea, the PWAP occupied 3.2\~7.1 percentage of the whole area during the survey time, while the PWAP covered 17.7\~37.2% of the Bohai. In addition, the water quality of Chinese coast varied widely among years. When the water quality became poorer, condition factor *a* decreased, with the reverse condition for parameter *b* ([Fig 4](#pone.0171811.g004){ref-type="fig"}). Negative correlation (-0.47) and positive correlation (0.61) were found for PWAP with condition factor *a* and for PWAP parameter *b* respectively, while t-test indicated both the correlation coefficients were not significant (df = 4, *P*\>0.05). {#pone.0171811.g004} Relative condition factors {#sec011} -------------------------- Summary of condition factor relative to reference year of 1960, 1986, 2005, 2007, 2008\~2009 and 2010 were shown in [S2 Table](#pone.0171811.s007){ref-type="supplementary-material"}. The mean of *K*~*cur/2005*~ was the smallest value (0.750), followed by *K*~*cur/1960*~ (0.786) and the mean estimates of *K* ~*cur/2007*~ and *K* ~*cur/1986*~ (0.881 and 0.882, respectively), and those of K~*cur/2008\ 9*~ and K~*cur/2010*~ were the highest (0.906 for both). The distributions ([S5 Fig](#pone.0171811.s005){ref-type="supplementary-material"}) revealed that there was more variability in *K* ~*cur/1986*~, compared to others. Furthermore, variations of relative condition factors among years, months, regions and length shown in [Fig 5](#pone.0171811.g005){ref-type="fig"}. {ref-type="table"}. The dot lines show the relative condition factor equal to 1.0.](pone.0171811.g005){#pone.0171811.g005} The estimates of condition factors relative to each reference year were similar in 2008 and 2010, and reached their maximum in 2012. With the exception of *K* ~*cur/1960*~, relative condition factors began to decline in 2012. However, values of *K* ~*cur/1960*~ remained stable over time, again with the exception of 2012. Relative condition factors were much lower in summer than other seasons, and reached maximum during autumn and early winter, with diverse variations degree to different reference years. As for spatial variations, values of relative condition factors decreased with decreasing latitude, while there was little variation (0.71\~0.81) for *K* ~*cur/1960*~ among different regions. Generally, relative condition factors increased as Yellow Croaker grew, except for *K*~*cur/1960*~, while there were different variation ranges among different sizes of individuals for condition factors relative to different reference year([Fig 5](#pone.0171811.g005){ref-type="fig"}). When the body length of Yellow Croaker was less than 10 cm, values of condition factores relative to reference years 2005, 2007, 2008\~2009 and 2010 kept much constant. However, values of *K* ~*cur/1960*~, decreased as individuals grew larger, and remained stable at lengths 14--18 cm. When the water quality declined, values of *K* also decreased, while there was little variation in *K* ~*cur/1960*~ (0.75\~0.79), all with an exception of 2012 ([Fig 6](#pone.0171811.g006){ref-type="fig"}). Without data in 2012, correlation analysis indicated negative correlations for PWAP with condtion factors relative to reference years except 1960 (from -0.68 to -0.61), while no significance were found in t-test (df = 3, *P*\>0.05). {#pone.0171811.g006} Discussion {#sec012} ========== LWR of Yellow Croaker and its spatial-temporal variations {#sec013} --------------------------------------------------------- According to the results of LMEM (R&Y.I&S), the estimates of *a* (0.013 \~ 0.023), fell in the common range (0.001\~0.05) of most fish species \[[@pone.0171811.ref002]\]. Study of Froese \[[@pone.0171811.ref029]\] shows that values of parameter *a* of fish species, with fusiform body shape, was 0.0112 (geometric mean) with 95% range of 0.00514 to 0.0245. The results are consistent with the body shape of this species, which is described as fusiform in Fishbase \[[@pone.0171811.ref030]\]. The point estimate of exponent *b*, was a little lower than values 3.03 (2.91--3.15) in Fishbase \[[@pone.0171811.ref029]\]. Values of parameter *b* for fishes were suggested by Carlander \[[@pone.0171811.ref031]\] and Froese \[[@pone.0171811.ref002]\] to normally fall within the range of 2.5 \~ 3.5 and 2.7 \~ 3.4, respectively, which cover the range of parameter *b* of Yellow Croaker in this study. The fixed value of allometric growth parameter *b*, indicated a very slight decrease in plumpness or elongation in form with increase in length. However, when random effects were used, no significant difference with isomeric growth was found for Yellow Croaker \[[@pone.0171811.ref032]\]. Random region effects for *a* demonstrated that values of *a* reduced gradually as the latitude decreased, from north to south, except in Coastal Waters of Northern Shandong. Samples in the Coastal Waters of Northern Shandong were collected from fishermen, and the origin of the samples were not clear, which could be the reason that parameters with the random effect of NS did not follow this trend. Yellow Croaker distributed on China's coast were considered to be divided into various populations, while individuals in northern Yellow Sea and Bohai Sea (NYBS) were believed to be different from those in southern Yellow Sea (SYS) \[[@pone.0171811.ref019],[@pone.0171811.ref033],[@pone.0171811.ref034]\]. The results of this study suggested that Yellow Croaker sampled from Yellow River Estuary and Coastal Waters of Northern Shandong should be included in NYBS, while those in other regions belonged to SYS. Different subpopulation maybe the main reason that led to their significant spatial variation in length-weight relationships. Previous research has examined LWR variation for Yellow Croaker; including spatial, temporal, sexual, and length variations, which covered Bohai Sea, Yellow Sea and East China Sea from 1960 to 2010 \[[@pone.0171811.ref003],[@pone.0171811.ref011],[@pone.0171811.ref014],[@pone.0171811.ref017],[@pone.0171811.ref018], [@pone.0171811.ref035]--[@pone.0171811.ref041]\]. In these LWRs (51 models), values of parameter *a* ranged from 0.0061 to 0.1027, with mean of 0.028 ± 0.019; values of exponent *b* ranged from 2.32 to 3.35, with mean of 2.875 ± 0.217. This, accompanied with relative condition factors, could support the hypothesis, that individuals were becoming thinner and young-age in recent years. Several influencing factors of LWR were evaluated in previous studies \[[@pone.0171811.ref003],[@pone.0171811.ref042],[@pone.0171811.ref043]\]. Pollutions such as inorganic nitrogen, active phosphate and oil pollutant, frequent pollutants in the north coast of China, are believed to make food availability and oxygen level deteriorate, which would negatively influence the growth of Yellow Croaker \[[@pone.0171811.ref044]--[@pone.0171811.ref046]\]. In this study, the data revealed that values of condition factor *a* decreased, when the environmental pollution increased. Therefore, LWR could be a rough indicator for the environment quality. Limited sample years maybe the reason that resulted in the non-significant correlation from statistics test, so further long term monitoring is suggested in the future. Condition changes compared with historical records {#sec014} -------------------------------------------------- Relative condition factor provided an effective approach to compare the observed weight of an individual with the baseline weight of that length \[[@pone.0171811.ref028],[@pone.0171811.ref047]\]. Compared to individuals in previous decades \[[@pone.0171811.ref003],[@pone.0171811.ref011]\], Yellow Croaker condition declined in recent years. *K*~*cur/1960*~ remained comparatively stable, without obvious variation. *K*~*cur/2005*~, *K*~*cur/2007*~, *K*~*cur/2008\ 9*~ and *K*~*2010*~ had similar variation patterns, since the LWR of Yellow Croaker in these years were similar, while values of *K*~*cur/2005*~ were much lower. However, *K*~*cur/1986*~ have similar variation pattern but higher variation, since the stock of Yellow Croaker collapsed during the 1980s and had been in an unstable condition in 1986 \[[@pone.0171811.ref016],[@pone.0171811.ref048]\]. Condition factors relative to all reference years, reached the highest value in 2012, as specimens in 2012 were collected in May and September ([S1 Table](#pone.0171811.s006){ref-type="supplementary-material"}), which was just before spawning and after feeding. A weak El Nino in 2012 might be another possible reason that led to elevated relative condition factors. Monthly changes in condition factors are related to life history events of this fish, e.g., sexual maturation, spawning and feeding strategies \[[@pone.0171811.ref014],[@pone.0171811.ref049],[@pone.0171811.ref050]\]. Adult yellow croaker in these areas reach their full gonads maturity in spring, reproduce in June and then primarily feed in summer and autumn \[[@pone.0171811.ref049],[@pone.0171811.ref051]\]. The condition factor (*K*) behaved in accordance to this series of events. It reached a high value in May, dropped precipitately in June, gradually recovered in August and reached the highest value in December. Spatial variation of *K* discovered that Yellow Croaker was skinny in the south, with worse condition than in the north. It was widely reported the Yellow Croaker stock shifted in size and age structure to feature more small and young fish that mature earlier, in response to severe fishing pressure over the past half century \[[@pone.0171811.ref015],[@pone.0171811.ref016],[@pone.0171811.ref040],[@pone.0171811.ref048]\]. This circumstance was also reflected in the length variation of condition factor relative to reference year of 1960. Since there was only one sample of L = 21cm, the condition factor relative to all reference years became an exception in L = 21cm for length variation of Yellow Croaker. Moreover, just as values of parameter *a*, water pollution led to *K* worse. The exception value *K* in 2012 relative to all reference years, probably resulted from the more significant effects from monthly variation and nearly null El Nino in 2012 ([Fig 5](#pone.0171811.g005){ref-type="fig"}). Applications of mixed effects model in studying spatial-temporal variations of LWR {#sec015} ---------------------------------------------------------------------------------- According to AIC and MAE, linear mixed effect model with effects from both regions and years on both parameters *a* and *b* was supported as the best one. The recommended model matches previous studies that the exact relationship between length and weight differs among regions and years, according to the habitat condition \[[@pone.0171811.ref004],[@pone.0171811.ref005],[@pone.0171811.ref052],[@pone.0171811.ref053]\]. However, many studies on LWR for fishes used region specific or year specific models fitting to region or year specific dataset to detect the spatial and temporal variations of individuals \[[@pone.0171811.ref004],[@pone.0171811.ref054],[@pone.0171811.ref055]\], as seen from previous 51 LWR models for Yellow Croaker \[[@pone.0171811.ref003],[@pone.0171811.ref011],[@pone.0171811.ref014],[@pone.0171811.ref017],[@pone.0171811.ref018],[@pone.0171811.ref035]--[@pone.0171811.ref041]\]. The mixed effects model took account of regions and years as random effects in a single model, which was more convenient and reasonable to estimate the spatial and temporal variations. It also allow regions and years with limited samples to borrow strength from other regions and years \[[@pone.0171811.ref010]\]. Furthermore, mixed effects model provided an effective approach to estimate the effects of variables, observations of which could be unavailable \[[@pone.0171811.ref007],[@pone.0171811.ref008]\]. In this study, our survey data were quite poor: there were 23 blanks in the 6×6 (Regions-Years) data matrix ([Table 1](#pone.0171811.t001){ref-type="table"}), while mixed effects model effectively evaluated the effects of regions and years. In the future, more variations of LWR, such as sex, season, and growth stage should be considered in the mixed-effects model for Yellow Croaker. Conclusions {#sec016} =========== According to this study, the condition of Yellow Croaker declined in recent years and was skinny in the south, with worse condition than in the north. The values of condition factor *a* in LWR decreased, when the environmental pollution increased. Therefore, LWR could be a rough indicator for the environment quality. The mixed effects model provided a more convenient and reasonable approach to estimate the spatial and temporal variations of LWR for Yellow Croaker. Supporting information {#sec017} ====================== ###### Yield of Yellow Croaker in China. Data were derived from *China Fishery Statistical Yearbook* from 1956 to 2012. (TIF) ###### Click here for additional data file. ###### Length frequency of Yellow Croaker. (TIF) ###### Click here for additional data file. ###### LWR differences of Yellow Croaker among regions and years from the LMEM (R&Y.I&S). a, b, and c were plots among regions; d, e, and f were plots among years. a and d plots cover the whole range of length; b and e plots cover the median part (Length = 12--16 cm); c and f plots cover the large size (Length = 16--22 cm). (TIF) ###### Click here for additional data file. ###### Parameters *a* and *b* distributions among regions and years. A and B plots were parameters variation among regions, with C and D plots among years. B and D plots zoom the main parts with high frequency. (TIF) ###### Click here for additional data file. ###### Distribution of condition factors relative to reference years of 1960 (A), 1986 (B), 2005(C), 2007 (D), 2008\~2009(E) and 2010(F). (TIF) ###### Click here for additional data file. ###### Monthly sample size of Yellow Croaker among regions and years. The detailed information of regions presents in [Table 1](#pone.0171811.t001){ref-type="table"}; no individuals were caught during February, March and April. (DOCX) ###### Click here for additional data file. ###### Condition factors of Yellow Croaker relative to reference year of 1960, 1986 and 2007, and the references LWR models. See the text for the definition of Kcur/1960, Kcur/1960 and Kcur/2007. (DOCX) ###### Click here for additional data file. We gratefully acknowledge our colleagues for their help with collecting the survey data. The manuscript was improved by the comments from Can Zhou, Corbin Hilling, and Dongyan Han. We also thank Department of Fish and Wildlife Conservation at Virginia Polytechnic Institute and State University to opportunities for Qiuyun to work on this study under supervision of Professor Yan Jiao. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceptualization:** YJ.**Data curation:** QM.**Formal analysis:** QM YJ.**Funding acquisition:** YR.**Investigation:** YR.**Methodology:** QM.**Project administration:** YJ YR.**Resources:** YJ YR.**Software:** QM.**Supervision:** YJ YR.**Validation:** QM YJ.**Visualization:** QM.**Writing -- original draft:** QM.**Writing -- review & editing:** QM YJ YR.

INTRODUCTION {#SEC1} ============ Next-generation sequencing (NGS) and microarray technologies uncovered thousands of long non-coding RNAs (lncRNAs) encoded in the human genome ([@B1],[@B2]). The majority of those lncRNAs are transcribed and processed in a similar manner to mRNAs, however, lack protein-coding potential ([@B3],[@B4]). Although it is still unclear how many of those lncRNAs have a significant biological function, some of them have been found to be crucial players in the regulation of cellular processes such as proliferation, differentiation or development, as well as in a progression of a variety of human diseases including cancer ([@B5]). It has been shown that lncRNAs are key determinants of epigenetic regulation, modulation of chromatin structure, scaffolding or decoy function of mRNAs and post-transcriptional mRNA regulation ([@B11]).Gene regulation by lncRNAs can be a result of cis-action on nearby genes, or in trans by modulating mRNA stability, mRNA translation, or microRNA and RNA-binding-protein function ([@B16]). Cellular senescence was initially defined by Hayflick in 1965 as the limited lifespan of primary human fibroblasts in culture ([@B24]). It is a state of irreversible growth arrest which can be induced by different stimuli such as telomere shortening, DNA damage, oxidative stress or oncogene activation ([@B25]). Serrano *et al.*, were the first to observe that primary human and mouse fibroblasts enter senescence following the induction of oncogenic RAS, a process termed oncogene-induced senescence (OIS) ([@B26]). Cellular senescence has been studied most extensively as a strong tumor-suppressive mechanism against the emergence of oncogenes ([@B27]). Moreover, there is evidence indicating for a role of senescence in age-related conditions and diseases, including cancer, cardiovascular diseases, neurodegeneration, diabetes, sarcopenia and declining immune function in the elderly ([@B28]). In contrast, senescent cells can also contribute to tumorigenesis by secreting interleukins (e.g. IL-6, IL-8 and IL-1a), metalloproteases (e.g. MMP-1 and MMP-3) and other cytokines (e.g. granulocyte-macrophage colony-stimulating factor (GM-CSF)), as part of the senescence-associated secretory phenotype ([@B25],[@B30],[@B33]). Therefore, senescence may either suppress or promote tumor progression depending on the context where it occurs ([@B38],[@B39]). Given the impact of senescence on human physiology and pathology, it is of interest to understand the molecular mechanisms underlying senescence in order to utilize it for diagnosis and therapy. A number of factors have been implicated in regulating senescence, including transcription factors, RNA binding proteins and microRNAs, such as p53, Ets ([@B40]), HuR ([@B41]), AUF1 ([@B42]) and TTP ([@B43]), and miR-377 ([@B44]), miR-22 ([@B45]). In contrast, despite increasing interest in the expression and function of lncRNAs, their possible implication in senescence remains largely unexplored. Recent works indicated a role of MIR31HG and SALNR in senescence ([@B46],[@B47]), but a focused functional genetic screen was not described before. We therefore sought to identify senescence-associated lncRNAs using our established cellular system that induces senescence in primary human BJ fibroblasts ([@B48]). Using transcriptomic profiling we identified a number of differentially expressed lncRNAs following oncogene induction. Next, using functional screen, we discovered that one of the lncRNAs whose expression was induced upon oncogenic stress---lncRNA-OIS1---is required for OIS. We demonstrate that lncRNA-OIS1 is required for senescence by controlling a nearby DPP4 gene with a tumor suppressive activity. Collectively, our results provide a new lncRNA-mediated regulatory pathway for controlling DPP4 during OIS. Our findings support the role of lncRNAs as transcriptional regulators in critical processes such as cellular senescence and a potential role in cancer. MATERIALS AND METHODS {#SEC2} ===================== Cell culture, transfection, retroviral and lentiviral transduction {#SEC2-1} ------------------------------------------------------------------ BJ/ET/Ras^V12^, TIG3/ET/RAS^V12^, Ecopack 2 and HEK293-T cells were cultured in Dulbecco's modified Eagle's medium (Gibco), supplemented with 10% fetal calf serum (FCS) (Hyclone) and 1% penicillin/streptomycin (Gibco). Senescence was induced by treatment with 100 nM 4-OHT (Sigma) for 14 days. Retroviruses were made by calcium phosphate transfection of Ecopack 2 cells and harvest at 40 and 64 h later. Lentiviruses were made by polyethylenimine (PEI) transfection of HEK293T. Medium was refreshed after 16 h and collect the lentivirus by filtering through a 0.45 μm membrane (Milipore Steriflip HV/PVDF) 40 h post-transfection and stored at −80°C. Cells were selected with the proper selection medium 48 h after transduction for at least 4 days until no surviving cells remained in the no-transduction control plate. RNA-seq and analysis {#SEC2-2} -------------------- RNA-seq samples were processed with TruSeq RNA library prep kit v2 (Illumina) and sequenced in a HiSeq 2500 (Illumina). Sequenced reads were aligned to the human genome (hg19) using TopHat2 ([@B49]) and gene expression levels were counted using HTseq ([@B50]) and normalized using quantile normalization. To avoid inflation of lowly expressed genes among the genes called as differentially expressed, we applied a dynamic cut-off which takes into account that technical variation varies with expression level. Specifically, in the comparison between two conditions, we divided the genes into 20 bins according to their average expression level, and calculated the standard deviation (SD) of fold-change within each bin. Genes whose expression was changed by at least 1.75-fold and this fold-change was above the bin's 1.75 SD (dashed curve in Figures [1B](#F1){ref-type="fig"} and [3B](#F3){ref-type="fig"}) were called as differentially expressed. To further avoid false positive calls among lowly expressed genes we set a floor level of five counts (i.e. any level below five was set to five). Functional enrichment analysis was done using DAVID ([@B51]). Global characterization of pathways that were deregulated upon knockdown of lncRNA-OIS1 was done using gene set enrichment analysis (GSEA) ([@B52]). {#F1} *In situ* hybridization {#SEC2-3} ----------------------- *In situ* hybridization (ISH) was performed using double-FAM labeled locked nucleic acid (LNA) probes (Exiqon) as described previously ([@B53]). Briefly, cells were fixed, permeabilized and pre-hybridized in hybridization buffer and then hybridized at 55°C for 1 h with LNA probes for lncRNA-OIS1: 5-TTGAAAACCCATCACTCCT-3, or with a scramble probe 5-TGTAACACGTCTATACGCCCA-3 as negative control, all at 25 nM. Cells were subsequently incubated with 3% hydrogen peroxide to block potential endogenous peroxidase, and then probes were detected with peroxidase-conjugated anti-fluorescein-Ab (Roche applied Sciences) diluted 1:400 followed by addition of Cy3-labeled TSA substrate for 10 min (Perkin Elmer). All cells were mounted with ProLong^®^GoldAntifade Mountant containing DAPI nuclear stain (ThermoFisher Scientific). Images were acquired using a Zeiss Axio Imager Z1 epi-fluorescence microscope equipped with an AxioCamMRm CCD camera and a Plan-APOCHROMAT 63×/1.4 objective (Zeiss). Within the same experiment, images were acquired at the same exposure conditions. BrdU proliferation assay {#SEC2-4} ------------------------ BJ and TIG3 Cells were pulsed for 3 h with 30 μM bromodeoxyuridine (BrdU, Sigma), washed two times with phosphate-buffered saline (PBS) and then fixed with 4% formaldehyde, wash two times with PBS and treated with 5M HCl/0.5% Triton to denature DNA and neutralized with 0.1M Na~2~B~4~O~7~, incubated with anti-BrdU (Dako) for 2 h in RT after 30 min blocking with 3% bovine serum albumin (BSA) in 0.5% Tween PBS, washed in blocking buffer (PBS, Tween 0.5%, 3% BSA) three times, and finally incubated with FITC-conjugated anti-mouse Alexa FLOUR 488 secondary antibody (Dako) for 1 h, washed three times, stained with propidium iodide for 30 min. BrdU incorporation was measured by immunofluorescence (at least 300 cells were scored for each condition). Senescence-associated β-galactosidase (SA-β-Gal) assay {#SEC2-5} ------------------------------------------------------ BJ and TIG3 cells were transduced with different shRNAs constructs, plated in triplicate and treated with 100 nM 4-OHT for 14 days. β-galactosidase activity was determined by using the kit (Cell Signaling), and at least 300 cells were analyzed for each condition. Ribosome profiling (Ribo-seq) {#SEC2-6} ----------------------------- BJ Cells were treated with cycloheximide (100 μg/ml) for 5 min, and lysed 20 mM Tris--HCl, pH 7.8, 100 mM KCl, 10 mM MgCl~2~, 1% Triton X-100, 2 mM dithiothreitol (DTT), 100 μg/ml cycloheximide, 1× complete protease inhibitor. Lysates were centrifuged at 1300 *g* and the supernatant was treated with 2 U/μl of RNase I (Invitrogen) for 45 min at room temperature. Lysates were fractionated on a linear sucrose gradient (7--47%) using the SW-41Ti rotor at 36 000 rpm for 2 h. Fractions enriched in monosomes were pooled and treated with proteinase K (Roche, Mannheim, Germany) in a 1% sodium dodecyl sulphate (SDS) solution. Released RNA fragments were purified using Trizol reagent and precipitated in the presence of glycogen. For libraries preparation, RNA was gel-purified on a denaturing 10% polyacrylamide urea (7 M) gel. A section corresponding to 30--33 nt, the region where most of the ribosome-protected fragments are comprised, was excised, eluted and ethanol precipitated. The resulting fragments were 3′-dephosphorylated using T4 polynucleotide kinase (New England Biolabs Inc. Beverly, MA, USA) for 6 h at 37°C in 2-(N-morpholino) ethanesulfonic acid (MES) buffer (100 mM MES-NaOH, pH 5.5, 10 mM MgCl~2~, 10 mM β-mercaptoethanol, 300 mM NaCl). 3′ adaptor was added with T4 RNA ligase 1 (New England Biolabs Inc. Beverly, MA, USA) for 2.5 h at 37°C. Ligation products were 5′-phosphorylated with T4 polynucleotide kinase for 30 min at 37°C. 5′ adaptor was added with T4 RNA ligase 1 for 18 h at 22°C. The library was sequenced in illumina HiSeq2000 machine. The data were analyzed as described ([@B54]). GRO-seq {#SEC2-7} ------- Briefly, 5 × 10^6^ nuclei were isolated and incubated 5 min at 30°C with equal volume of reaction buffer (10 mM Tris--Cl pH 8.0, 5 mM MgCl~2~, 1 mM DTT, 300 mM KCL, 20 units of SUPERase In, 1% sarkosyl, 500 μM adenosine triphosphate (ATP), Guanosine triphosphate (GTP) and Br-Uridine triphosphate (UTP), 0.2 μM CTP+32P Cytidine triphosphate (CTP)) for the nuclear run-on. The reaction was stopped and total RNA was extracted with Trizol LS (Invitrogen) according to the manufacturer's instructions. RNA was fragmented using fragmentation reagents (Ambion) and the reaction was purified through p-30 RNase free spin column (BioRad). BrU-labeled RNA was immunoprecipitated with anti-BrdU agarose beads (Santa Cruz), washed one time in binding buffer, one time in low salt buffer (0.2 × SSPE, 1 mM ethylenediaminetetraacetic acid (EDTA), 0.05% Tween-20), one time high-salt buffer (0.25 × SSPE, 1 mM EDTA, 0.05% Tween-20, 137.5 mM NaCl) and two times in TET buffer (TE with 0.05% Tween-20). RNA was eluted with elution buffer (20 mM DTT, 300 mM NaCl, 5 mM Tris--Cl pH 7.5, 1 mM EDTA and 0.1% SDS) and isolated with Trizol LS. After the binding step, BrU-labeled RNA was treated with tobacco acid pyrophosphatase (Epicenter) to remove 5′-methyl guanosine cap, followed by T4 polynucleotide kinase (PNK; NEB) to remove 3′-phosphate group. BrU-containing RNA was treated with T4 PNK again at high pH in the presence of ATP to add 5′-phosphate group. The reaction was stopped and RNA was extracted with Trizol LS. Sequencing libraries were prepared using TruSeq Small RNA kit (Illumina) following manufacturer's instructions. Briefly, end-repaired RNA was ligated to RNA 3′ and 5′ adapters, followed by RT-PCR amplification. cDNA was purified using Agencourt AMPure XP (Beckman Coulter) and amplified by polymerase chain reaction (PCR) for 12 cycles. Finally, amplicons were cleaned and size-selected using Agencourt AMPure XP (Beckman Coulter), quantified in a Bioanalyzer 2100 (Agilent), and sequenced in a HiSeq 2500 (Illumina). Sequenced reads were aligned to the human genome (hg19) using bowtie2 ([@B55]). RNA isolation, reverse-transcription and quantitative real-time PCR (qPCR) {#SEC2-8} -------------------------------------------------------------------------- Total RNA was extracted by using TRIsure (Bioline) reagent and following the manufacturer's protocol. cDNA was produced with SuperScript III (Invitrogen) using 4 μg of total RNA per reaction. qPCR reaction was performed with SYBR green I Master mix in a LightCycler 480 (Roche). Primers used in qPCR are listed in [Supplementary Table S5](#sup1){ref-type="supplementary-material"}. Western blot analysis {#SEC2-9} --------------------- Whole-cell lysates were prepared as previously described ([@B56]). Membranes were immunoblotted with the following antibodies: CDKN1A (Sc-397, Santa Cruz; 1: 1000), HRAS (C-20, Santa Cruz; 1: 1000), DPP4 (ab28340, abcam; 1: 2000), GAPDH (Sc-47724, Santa Cruz; 1: 5000). Protein bands were visualized using corresponding secondary antibodies (Dako) and ECL reagent (GE Healthcare). Chromosome conformation capture combined with sequencing (4C-seq) {#SEC2-10} ----------------------------------------------------------------- Briefly, BJ cells were treated with or without 4-OHT for 14 days and 10^7^ of cells for each condition were harvested and we performed 4C as previously described ([@B57]). An adapted two-step 4C-PCR was performed as previously described ([@B58]) to introduce template specific indexes. We had two viewpoints and used the following primers in the first PCR: vp1_forward AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTCTTTGCTACTCTGTGAGATC vp1_reverse ACTGGAGTTCAGACGTGTGCTCTTCCGATCTATAGGGCTCTGGAGTCAG vp2_forward AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTGTATTTCTCTAGCTGGGATC vp2_reverse ACTGGAGTTCAGACGTGTGCTCTTCCGATCAACCGTAAAGTCTTCGCTC We used the forward primers from the first PCR and combined the following reverse primers for the second PCR: BJ -- 4-OHT rep1 CAAGCAGAAGACGGCATACGAGAT CGTGAT GTGACTGGAGTTCAGACGTGTGCT BJ-- 4-OHT rep2 CAAGCAGAAGACGGCATACGAGAT GCCTAA GTGACTGGAGTTCAGACGTGTGCT BJ + 4-OHT rep1 CAAGCAGAAGACGGCATACGAGAT GGAACT GTGACTGGAGTTCAGACGTGTGCT BJ + 4-OHT rep2 CAAGCAGAAGACGGCATACGAGAT GCGGAC GTGACTGGAGTTCAGACGTGTGCT LncRNA-OIS1 expression analysis in tumors {#SEC2-11} ----------------------------------------- Gene expression data was obtained from the TCGA Data Portal (<https://tcga-data.nci.nih.gov>). We selected those cancer types with transcriptome data available for at least five normal and five tumor samples, belonging to phenotypes 'solid tissue normal' and 'primary solid tumor', respectively. Lowly expressed genes (genes with raw read counts in less than half the normal samples and half the tumor samples) were removed within each cancer type data. Differential expression analysis was carried out with R/Bioconductor package limma ([@B59]) using voom normalization ([@B60]). Pearson correlation calculation was carried out using normalized gene expression values, also in R/Bioconductor. RESULTS {#SEC3} ======= Genome-wide identification of lncRNAs responsive to OIS {#SEC3-1} ------------------------------------------------------- To identify lncRNAs with a role in OIS, we used the model of primary human BJ fibroblasts expressing hTERT and 4-OH-tamoxifen (4-OHT)-inducible oncogenic H-Ras^V12^ (BJ/ET/Ras^V12^ER cells) ([@B48]). RNA sequencing (RNA-seq) in senescent cells and non-senescent control cells revealed senescence-associated differentially expressed transcripts (Figure [1B](#F1){ref-type="fig"}). Of those transcripts, we found 34 and 6 lncRNAs upregulated and downregulated respectively during OIS ([Supplementary Table S1](#sup1){ref-type="supplementary-material"}). Ribosome profiling confirmed the non-coding nature of these RNAs (Figure [1C](#F1){ref-type="fig"}). We also confirmed by qRT-PCR the induction of some lncRNAs following H-Ras^V12^ induction ([Supplementary Figure S1A](#sup1){ref-type="supplementary-material"}). A focused loss-of-function screen for lncRNAs required for OIS identifies lncRNA-OIS1. {#SEC3-2} -------------------------------------------------------------------------------------- To examine possible causal roles for lncRNAs in OIS, we developed RNAi tools to target the 40 lncRNAs that were differentially expressed in OIS. We generated a pooled library consisting of five different shRNAs against each lncRNA, and included four non-targeting shRNAs as negative controls, as well as two positive control shRNAs targeting BRD7---a gene identified as a tumor suppressor in OIS ([@B48]) ([Supplementary Table S2](#sup1){ref-type="supplementary-material"}). We transduced cells with three independent retroviral pools of the shRNAs library, and following puro selection harvested half of each cell population as control (T0, Time 0). We cultured the rest of the cells with 4--OHT treatment for 4 weeks, then harvested the cells (T4, Time 4 weeks) and performed NGS to identify shRNAs enriched in the final populations (T4) compared to the initial (T0) pool (Figure [1D](#F1){ref-type="fig"}). Our screen detected the positive control shRNAs against BRD7, as well as few shRNAs targeting different lncRNAs enriched in the RAS-induced cell populations, suggesting that the knockdown of these lncRNAs conferred a growth advantage in BJ/ ET cells expressing Ras^V12^ (Figure [1E](#F1){ref-type="fig"}). For further validation, we selected two hits: lncRNA\#32, which has one shRNA (shRNA3) at the top of the enrichment list in all three replicates, and another shRNA (shRNA2) giving minor enrichment; and lncRNA\#30 with two shRNAs (shRNA5 and shRNA3) showing consistent enrichment in all three replicates (Figure [1E](#F1){ref-type="fig"} and [Supplementary Table S3](#sup1){ref-type="supplementary-material"}). We validated the hits by repeating the OIS experiment using individual shRNAs. We used an shRNA targeting BRD7 (BRD7_shRNA4) as a positive control, and two non-targeting shRNA as negative controls. A proliferation assay (using BrdU labeling) indicated bypass of oncogene-induced cellular arrest by one shRNA (\#30--5) targeting lncRNA\#30 and two shRNAs (\#32--2 and \#32--3) targeting lncRNA\#32 (Figure [2A](#F2){ref-type="fig"} and [Supplementary Figure S2A](#sup1){ref-type="supplementary-material"}). To further examine the effect of loss-of lncRNA\#30 and \#32 in OIS, we measured the induction of senescence-associated β-galactosidase (SA-β-Gal). In comparison with negative control cells, a marked decrease in SA-β-Gal was observed in Ras^V12^-expressing BRD7-knockdown (BRD7 kd), lncRNA\#30--5 and lncRNA\#32--2 and \#32--3 cells (Figure [2B](#F2){ref-type="fig"} and [Supplementary Figure S2B](#sup1){ref-type="supplementary-material"}). In contrast, shRNA (\#30--3) was not validated as expected from the screen outcome. Interestingly, RNA expression analysis indicated that only shRNAs \#30--5, \#32--2 and \#32--3 were effective toward their lncRNA targets, suggesting on-target activity (Figure [2C](#F2){ref-type="fig"} and [D](#F2){ref-type="fig"}). To exclude off-target effects of the shRNAs, we designed additional vectors targeting lncRNA\#30 and \#32 ([Supplementary Table S6](#sup1){ref-type="supplementary-material"}), and repeated the proliferation and SA-β-Gal assays. This experiment identified more functional shRNAs (\#32--8, \#32--9) targeting lncRNA\#32, but no additional shRNAs targeting lncRNA\#30 (Figure [2E](#F2){ref-type="fig"} and [F](#F2){ref-type="fig"}; [Supplementary Figure S3A](#sup1){ref-type="supplementary-material"} and B). qRT-PCR confirmed loss-of expression of lncRNA\#32 by all four active shRNA vectors (\#32--2, 3, 8 and 9) (Figure [2H](#F2){ref-type="fig"}). In contrast, two new shRNAs (\#30--8, \#30--9) showed efficient loss-of lncRNA\#30 (Figure [2G](#F2){ref-type="fig"}) but did not induce bypass of OIS (Figure [2E](#F2){ref-type="fig"} and [F](#F2){ref-type="fig"}; [Supplementary Figure S3A and B](#sup1){ref-type="supplementary-material"}), indicating that the bypass of OIS by shRNA\#30--5 was not mediated by its targeted lncRNA. Altogether, these results demonstrate that lncRNA\#32 is both induced by oncogenic RAS and is required for the establishment of the OIS phenotype. {#F2} To further solidify the role of lncRNA\#32 in OIS we made use of a dual CRISPR-Cas9 system ([@B61]), and induced deletions of the lncRNA\#32 locus. As BJ cells do not form single clones, generation of monoclonal population of deleted cells was not possible. Instead, we performed a functional genetic experiment to test whether the cells containing the lncRNA\#32 deletion are enriched in cells undergoing OIS. Notably, [Supplementary Figure S4A](#sup1){ref-type="supplementary-material"} shows that control-transduced BJ cells completely senesced, p53 knockout BJ cells strongly bypassed OIS, and targeting lncRNA\#32 attenuated senescence, albeit to a lesser extent than p53KO. To confirm that the dual CRISPR-Cas9 system triggered deletion of lncRNA\#32, we isolated genomic DNA and performed semi-quantitative PCR to detect lncRNA\#32 with oligos (FW: TGGAGGGCTGAATCATCAAGTT, REV: ACTTCAAAGGGCAATTGCTGAAC) surrounding the CRISPR-target region. While wild-type and control-transduced cells produced only one band of about 1.8 Kb, cells transduced with the lncRNA\#32-targeting vector showed lncRNA\#32 deleted bands (∼350 bp), indicating the functionality of the CRISPR vector ([Supplementary Figure S4B](#sup1){ref-type="supplementary-material"}). Intriguingly, the PCR signal of the deletion band increased after 2 and 3 weeks following OIS induction, in line with a bypass of the OIS phenotype. In comparison, no enrichment of the deleted allele was noted following 3 weeks of culturing without induction of OIS ([Supplementary Figure S4B](#sup1){ref-type="supplementary-material"}). This indicates that lncRNA\#32 deletion gives growth advantage only under OIS conditions. As expected, we found by qRT-PCR that cells expressing sgRNAs targeting lncRNA\#32 have reduced level of lncRNA\#32 ([Supplementary Figure S4C](#sup1){ref-type="supplementary-material"}). Sanger sequencing confirmed the correct deletion of lncRNA\#32 ([Supplementary Figure S4D](#sup1){ref-type="supplementary-material"}). For simplicity and in conjunction with its function, we hereafter refer to lncRNA\#32 as lncRNA-OIS1. To extend our finding on the role of lncRNA-OIS1 in OIS we employed a different cell system. We transduced all four functional shRNAs targeting lncRNA-OIS1 (\#32--2, 3, 8 and 9, which we renamed KD1, 2, 3 and 4, respectively) into TIG3 cells expressing hTERT and 4-OH-tamoxifen (4-OHT)-inducible oncogenic H-Ras^V12^, and repeated the BrdU labeling and SA-β-Gal experiments. First, q-RT-PCR and GRO-seq analysis indicated upregulation of lncRNA-OIS1 following oncogenic RAS induction ([Supplementary Figures S5E and S9A](#sup1){ref-type="supplementary-material"}). Second, as expected, the introduction of all four lncRNA-OIS1 shRNAs reduced lncRNA-OIS1 expression ([Supplementary Figure S5E](#sup1){ref-type="supplementary-material"}). Last, and most profoundly, all four lncRNA-OIS1 shRNAs very effectively bypassed OIS as measured by the proliferation and senescent assays BrdU and SA-β-Gal, respectively ([Supplementary Figure S5A--D](#sup1){ref-type="supplementary-material"}). Altogether, our results demonstrate that intact lncRNA-OIS1 is required for senescence induction following RAS^V12^ activation in primary human cells. Knockdown of lncRNA-OIS1 abolishes OIS gene expression signature {#SEC3-3} ---------------------------------------------------------------- Next, we sought to explore the mode of action of lncRNA-OIS1 in senescence. To this goal, we first performed RNA-seq of cells transduced with shRNAs against lncRNA-OIS1, the positive controls p53 and BRD7, and negative controls (Figure [3A](#F3){ref-type="fig"}). Comparison of gene expression profiles in negative controls and p53kd cells upon activation of oncogenic RAS (the former enters senescence while the latter bypasses it) identified 885 differentially expressed genes (386 up- and 499 downregulated in senescent cells) (Figure [3B](#F3){ref-type="fig"}). Functional enrichment analysis showed that the set of genes whose expression was significantly repressed in senescent cells (compared to the p53kd cells) was markedly enriched for cell-cycle-related genes (Figure [3C](#F3){ref-type="fig"}), reflecting the strong proliferation arrest that is imposed in negative control cells upon oncogenic stress. This sharp downregulation of cell-cycle genes defines a molecular signature that characterizes the induction of the senescent physiological state. Remarkably, knocking--down lncRNA-OIS1 significantly abolished the repression of these genes (Figure [3D](#F3){ref-type="fig"}). The effect observed for lncRNA-OIS1-kd was comparable to the effect obtained by BRD7-kd but weaker than the effect elicited by p53-kd (Figure [3D](#F3){ref-type="fig"}). In accordance with the phenotypic effect of OIS-bypass, we observed that lncRNA-OIS1-kd resulted in attenuation of the induction of CDKN1A (p21), a prime target of p53 that is required for OIS in BJ cells ([Supplementary Figure S6A](#sup1){ref-type="supplementary-material"}). We confirmed this result at the protein level using western blot analysis (Figure [3E](#F3){ref-type="fig"} and [Supplementary Figure S6B](#sup1){ref-type="supplementary-material"}). We included one shRNA (\#32--6) which did not give knockdown of lncRNA-OIS1 and showed no bypass of the senescence phenotype (Figure [3E](#F3){ref-type="fig"} and [Supplementary Figure S6B](#sup1){ref-type="supplementary-material"}) to demonstrate specificity of the decreased expression of CDKN1A due to lncRNA-OIS1-kd. {#F3} To further characterize the effect of knocking-down lncRNA-OIS1 on the cellular transcriptome, we systematically compared, using GSEA analysis ([@B52]), gene expression profiles in cells induced for oncogenic RAS and transduced either with shRNAs against lncRNA-OIS1 or with non-targeting shRNAs. As expected from the phenotypic effect and inline with the above analysis, the strongest gene sets that were upregulated upon knocking-down lncRNA-OIS1 were related to proliferation and cell-cycle ([Supplementary Figure S6C](#sup1){ref-type="supplementary-material"}). A set of genes that are induced in response to ionizing irradiation was the most significantly downregulated gene set in the lncRNA-OIS1 kd cells. This set contains numerous p53 direct target genes, indicating that attenuated expression of lncRNA-OIS1 compromises the activation of the p53 network ([Supplementary Figure S6C](#sup1){ref-type="supplementary-material"}). In addition, genes of the oxidative phosphorylation pathway are downregulated too in the lncRNA-OIS1 kd cells. Notably, all these gene sets show the opposite response in cells that enter senescence in response to oncogenic stress ([Supplementary Figure S6C](#sup1){ref-type="supplementary-material"}), demonstrating that loss of lncRNA-OIS1 abolishes OIS gene expression signature. Loss-of lncRNA-OIS1 compromises the induction of DPP4 by OIS {#SEC3-4} ------------------------------------------------------------ To investigate the mechanism(s) by which lncRNA-OIS1 affects OIS induction, we first examined its subcellular localization. In control BJ/ET/Ras^V12^ER cells, lncRNA-OIS1 was located both in the nucleus and the cytosol. Following RAS^V12^ activation, lncRNA-OIS1 maintained a similar pattern in these two compartments (Figure [4A](#F4){ref-type="fig"}). ISH analysis confirmed lncRNA-OIS1 increased expression following RAS^V12^ induction, and its localization in the nucleus and cytosol. Loss-of lncRNA-OIS1 confirmed the specificity of the signal to lncRNA-OIS1 (Figure [4B](#F4){ref-type="fig"}). {#F4} LncRNAs can impact the expression of nearby genes on the chromatin (*cis* function), or affect gene expression in trans (for example by controlling mRNA transcription, splicing and translation). We therefore first interrogated whether lncRNA-OIS1 functions *in trans*, and whether ectopic expression of lncRNA-OIS1 can drive cells into senescence without RAS induction. We over-expressed lncRNA-OIS1 in primary BJ cells (full length or exons; [Supplementary Figure S7A](#sup1){ref-type="supplementary-material"}), but observed no induction of senescence as measured by BrdU labeling and SA-β-Gal assays ([Supplementary Figure S7B--D](#sup1){ref-type="supplementary-material"}). Second, we overexpressed lncRNA-OIS1 (both full length and exons) in lncRNA-OIS1-kd cells to test whether ectopic expression of lncRNA-OIS1 can restore the senescence phenotype. However, despite the high expression of lncRNA-OIS1 ([Supplementary Figure S8A](#sup1){ref-type="supplementary-material"}), OIS-bypass by lncRNA-OIS1-kd was maintained ([Supplementary Figure S8B--D](#sup1){ref-type="supplementary-material"}). These data indicated that lncRNA-OIS1 does not function in trans, rather, a localized expression and effect on neighboring genes is required (*cis* effect). In general, lncRNAs can be physically linked to the locus from which they are encoded, and exert its function during transcription without the need for processing or shuttling. Well-studied examples of *cis*-acting lncRNAs are those that cause X-inactivation ([@B62],[@B63]). Examples of other *cis*-regulatory lncRNAs include ncRNA-a1--7, Hottip and Mistral, the perturbation of which lead to decreased expression of nearby genes ([@B64]), suggesting that gene regulation in *cis* is a very important mode of lncRNA action. To investigate whether lncRNA-OIS1 expression influences nearby genes, we analyzed Global Run-On Sequencing data (GRO-Seq) of senescent and proliferation BJ cells ([@B55]). We observed that both lncRNA-OIS1 and its nearby gene DPP4 were increased in the BJ cells upon RAS induction (Figure [4C](#F4){ref-type="fig"}). We also observed the same effect in TIG3 cells ([Supplementary Figure S9A](#sup1){ref-type="supplementary-material"}). Additionally, loss-of lncRNA-OIS1 abolished the activation of DPP4 following oncogene induction based on BJ cells RNA-seq data ([Supplementary Figure S9B](#sup1){ref-type="supplementary-material"}). We solidified these results by qRT-PCR (Figure [4D](#F4){ref-type="fig"} and [Supplementary Figure S9C](#sup1){ref-type="supplementary-material"}) and chose the best two lncRNA-OIS1 knockdowns (KD2 and 4) for western blot analyses of DPP4 expression four days following RAS^V12^ induction, before the cell-cycle is arrested and senescence is established (Figure [4D](#F4){ref-type="fig"} and [E](#F4){ref-type="fig"}). Indeed, attenuated activation of DPP4 protein expression was obtained in cells with lncRNA-OIS knockdown. A similar effect was also observed in TIG3 lncRNA-OIS1-kd cells 4 days following RAS^V12^ induction ([Supplementary Figure S9D](#sup1){ref-type="supplementary-material"}). Altogether, these data link lncRNA-OIS1 to regulation of DPP4 expression and to the senescent phenotype induced by oncogenic RAS. Loss-of DPP4 bypasses OIS {#SEC3-5} ------------------------- Interestingly, it has been reported that DPP4 is a tumor suppressor in melanoma ([@B68],[@B69]), non-small cell lung cancer ([@B70]), ovarian cancer ([@B71]), endometrial carcinoma ([@B74]), prostate cancer ([@B75]), neuroblastoma ([@B76]) and glioma ([@B77]). We therefore hypothesized that the tumor suppressive role of DPP4 is linked to OIS. To examine this issue, we generated shRNAs ([Supplementary Table S7](#sup1){ref-type="supplementary-material"}) transduced DPP4 knockdown BJ cells. qRT-PCR and western blot analyses confirmed significant reduction of DPP4 mRNA and protein levels upon knockdown (Figure [5A](#F5){ref-type="fig"} and [B](#F5){ref-type="fig"}). As predicted, loss-of DPP4 bypassed OIS, as determined by proliferation and SA-β-Gal assays (Figure [5C](#F5){ref-type="fig"} and [D](#F5){ref-type="fig"}; [Supplementary Figure S9E](#sup1){ref-type="supplementary-material"}). Next, we examined whether lncRNA-OIS1 regulates senescence through DPP4. We cloned DPP4 in a lentiviral vector, ectopically expressed it in lncRNA-OIS1-kd cells and induced OIS. Intriguingly, proliferation (BrdU labeling) and SA-β-Gal assays demonstrated that ectopic expression of DPP4 abolished the senescence bypass phenotype of lncRNA-OIS1-kd cells, while a control vector did not (Figure [6A](#F6){ref-type="fig"} and [B](#F6){ref-type="fig"}; [Supplementary Figure S10A and B](#sup1){ref-type="supplementary-material"}). We confirmed the overexpression of DPP4 by western blot (Figure [6C](#F6){ref-type="fig"} and [Supplementary Figure S10C](#sup1){ref-type="supplementary-material"}). These experiments indicate that DPP4 is the relevant target gene of lncRNA-OIS1 during OIS, and that lncRNA-OIS1 is a major determinant of DPP4 function in OIS. {#F5} {#F6} Association of LncRNA-OIS1and DPP4 in the tumors {#SEC3-6} ------------------------------------------------ Last, we interrogated lncRNA-OIS1 expression in tumors and its correlation with that of DPP4 by analyzing TCGA data. LncRNA-OIS1 is very lowly expressed in most tumor types ([Supplementary Figure S11A](#sup1){ref-type="supplementary-material"}). We plotted the read counts of the lncRNA-OIS1 among normal and tumor samples, indicating in each type the number of samples with at least one read count for lncRNA-OIS1. Interestingly, prostate adenocarcinoma (PRAD) samples showed clear lncRNA-OIS1 expression. Using this dataset for differential expression analysis, we observed no change between tumor and normal samples (empirical Bayes test, B = −5.79, *P*-value = 0.28) ([Supplementary Table S4](#sup1){ref-type="supplementary-material"}), but a significant positive correlation between lncRNA-OIS1 and DPP4 expression in the tumor samples (*r* = 0.469, *P*-value = 2.5e^−31^) (Figure [6D](#F6){ref-type="fig"}), suggesting that at least in PRAD DPP4 expression is controlled by lncRNA-OIS1. DISCUSSION {#SEC4} ========== Over the past few years, numerous lncRNAs have been discovered and characterized as critical factors in physiological and pathological processes. However, the role of lncRNAs in OIS remained unexplored. Here, we contribute to the understanding of the function of lncRNAs by describing a role of lncRNA-OIS1 in cellular senescence provoked by the expression of oncogenic RAS (OIS). Upregulation of lncRNA-OIS1 following OIS was required for the induction of DPP4, a well-described gene with tumor suppressive activity. Differential gene expression analyses of lncRNA-OIS1 knockdown cells indicated attenuated activation of CDKN1A following OIS induction, and confirmed changes in cell-cycle regulatory genes favoring cellular proliferation. Gene complementation experiments indicated that DPP4, a lncRNA-OIS1 neighboring gene, is the downstream target of lncRNA-OIS1 in senescence. Exactly how DPP4 affects CDKN1A and cell-cycle genes, and how lncRNA-OIS1 controls DPP4 expression, remains to be uncovered. Nevertheless, we describe here an important function of lncRNAs with potentially influential implications in cancer biology. OIS is a major senescence type and it poses a critical barrier to cancer. A recent study has shown that the lncRNA-MIR31HG was a senescence modulator during BRAF-V600 induced senescence in TIG3 cells ([@B46]). It has also been shown that loss-of MIR31HG reduces cell growth and promotes a strong senescence phenotype through the regulation of the tumor suppressor P16^INK4A^. Here, we add to this knowledge by identifying and characterizing the role of lncRNA-OIS1 in regulating senescence through the control of a nearby gene DPP4. Interestingly, we also overexpressed lncRNA-OIS1 in BJ cells to examine whether high levels of the lncRNA-OIS1 can drive cells into senescence. However, we neither observed senescence induction nor DPP4 was activated ([Supplementary Figure S7](#sup1){ref-type="supplementary-material"}). Additionally, also the ectopic expression of lncRNA-OIS1 was not able to revert the bypass of senescence and the reduced DPP4 activation induced by lncRNA-OIS1 knockdown under OIS ([Supplementary Figure S8](#sup1){ref-type="supplementary-material"}). This is indicative of a cis function of lncRNA-OIS1. Indeed, we identified DPP4, a nearby gene to lncRNA-OIS1, as a key component of OIS. First, loss-of DPP4, similar to lncRNA-OIS1 loss, resulted in bypass of senescence (Figure [5](#F5){ref-type="fig"}). Second, ectopic expression of DPP4 reverted the bypass of senescence induced by the loss-of lncRNA-OIS1 (Figure [6A](#F6){ref-type="fig"}--[C](#F6){ref-type="fig"} and [Supplementary Figure S10](#sup1){ref-type="supplementary-material"}). Additionally, a recent research found that DPP4 can regulate senescence in WI-38 cells, strongly supporting our observations ([@B78]). However, although both lncRNA-OIS1 and DPP4 genes reside in the same topologically associating chromatin domain through a CCCTC-binding factor (CTCF) loop ([Supplementary Figure S12A](#sup1){ref-type="supplementary-material"}), and chromatin loops can be identified in various ChIA-PET and Hi-C chromatin conformation capture datasets ([Supplementary Figure S12B](#sup1){ref-type="supplementary-material"}), we did not observe a clear direct interaction of lncRNA-OIS1 locus with the promoter of DPP4 using 4C, a chromatin capture analysis technique, through two distinct view point sites ([Supplementary Figure S13A](#sup1){ref-type="supplementary-material"}). Thus, how exactly the expression of DPP4 depends on lncRNA-OIS1 remains unclear. We speculate that lncRNA-OIS1 expression may be required to allow high chromatin accessibility to senescence-associated DPP4-activating transcription factors by directly recruiting essential transcription factors, or alternatively by counteracting chromatin-repressive components of the chromatin (Figure [7](#F7){ref-type="fig"}). Nevertheless, our findings here elucidate the importance of lncRNA-OIS1 for eliciting a proper cellular response to the emergence of oncogenic stress. {#F7} DATA AVAILABILITY {#SEC5} ================= RNA-seq, Ribosome profiling (Ribo-seq), GRO-seq data are deposited in GEO DB (accession number GSE42509, GSE106414, GSE109290). Supplementary Material ====================== ###### Click here for additional data file. We thank the China Scholarship Council for support. RNA Train supports our research and provides a great platform for training, learning, calibration. We thank all the members of the Agami group for helpful discussions. We thank Andrea Ventura for kindly providing the plasmid PX333 and all the reagents for lncRNA-OIS1 deletion. We also want to thank Jing Li (Cnkingbio Company) for helpful suggestions. SUPPLEMENTARY DATA {#SEC6} ================== [Supplementary Data](https://academic.oup.com/nar/article-lookup/doi/10.1093/nar/gky087#supplementary-data) are available at NAR Online. FUNDING {#SEC7} ======= This work was supported by the China Scholarship Council (CSC) (to L.L.); ERC-AdG enhReg \[322493 to R.A.\]; ERC-ITN RNA TRAIN \[607720 to R.A.\]; Edmond J. Safra Center for Bioinformatics Fellowship (to R.E.). The Human Frontier Science Program \[LT000640/2013 to Alejandro Pineiro Ugalde\]. Funding for open access charge: CSC; ERC-AdG enhReg \[322493\]; RNATrain \[607720\]. *Conflict of interest statement*. None declared.

Introduction {#s1} ============ Osteoarthritis is a multifactorial disease characterized by destruction of the articular cartilage due to genetic, mechanical and environmental components [@pone.0003740-Spector1], affecting more than 20 million people in the US [@pone.0003740-Lawrence1]. Despite its high prevalence there are few studies concerning the molecular pathobiology and the involvement of genetic factors in the pathogenesis of osteoarthritis [@pone.0003740-Miyamoto1], [@pone.0003740-Kizawa1]. Several clinical studies have implicated the causative role of obesity in osteoarthritis development [@pone.0003740-Lohmander1], [@pone.0003740-Lementowski1], however there are few molecular studies correlating metabolism with osteoarthritis [@pone.0003740-Aspden1], [@pone.0003740-Ostalowska1]. Achieving a deeper understanding of osteoarthritis molecular mechanisms requires global strategies aimed at modelling the functional interrelationships between genes as complex interdependent networks. Lately, it has become evident that genetic alterations in non-coding genes can also contribute to the pathogenesis of human disease [@pone.0003740-McManus1]. A new class of small non-coding RNAs, named microRNAs, regulate gene expression by inhibition of translation or mRNA cleavage [@pone.0003740-EsquelaKerscher1]. MicroRNAs have been implicated in important cellular processes such as lipid metabolism [@pone.0003740-Esau1], apoptosis [@pone.0003740-He1], differentiation [@pone.0003740-Chen1] and organ development [@pone.0003740-Callis1]. Furthermore microRNAs expression signatures have been associated with well-defined clinicopathological features and disease outcome [@pone.0003740-Calin1]. It is known that microRNAs exert their biological functions through suppression of their target genes. Several bioinformatic algorithms have been constructed in order to predict microRNA gene targets. Most of these algorithms search for sequence complementarity between the microRNA and the 3′ UTR of the gene target. These algorithms predict hundreds of potential gene targets, which can not all be experimentally validated. Previous studies have tried to identify microRNA gene targets using cDNA microarray data [@pone.0003740-Huang1]. However, it has been shown that a microRNA (miR-10b) regulates gene expression only at the protein level, while mRNA levels were not affected [@pone.0003740-Ma1]. In addition, very recently Selbach et al, showed that a microRNA can repress the production of hundred of proteins [@pone.0003740-Selbach1]. Therefore, it becomes evident that proteomic data are needed in order to accurately detect microRNA gene targets. Up to now there are few studies trying to characterize the cartilage proteome. More specifically, recently Vincourt et al, performed a detailed two dimensional electrophoresis-based proteomic analysis of articular cartilage [@pone.0003740-Vincourt1]. In addition Wu et al. performed a comparative proteomic analysis of cartilage from healthy donors and osteoarthritis patients, however the number of samples used was very small [@pone.0003740-Wu1]. For all above reasons, we undertook to associate specific microRNAs and proteins with the development of osteoarthritis and clinicopathological parameters, in order to identify new signalling pathways involved in its pathogenesis. Here, we report a novel approach of studying multi-aetiological diseases and identifying new genes involved in the pathogenesis of a complex disease. Integration of microRNA microarray and proteomic analysis data together with computational approaches, such as microRNA gene target prediction algorithms and gene network construction, revealed the role of microRNAs in cartilage destruction and linked inflammatory and metabolic gene networks with cartilage homeostasis. Materials and Methods {#s2} ===================== Cartilage tissue samples {#s2a} ------------------------ Articular cartilage samples were obtained from femoral heads, femoral condyles and tibial plateaus of patients with primary osteoarthritis undergoing hip or knee replacement surgery at the Orthopaedics Department of University Hospital of Larissa. A total of 33 patients were included in this study (twenty eight females and five males; mean age 68.91±6.97 years, range 57--83; mean Body Mass Index (BMI) 30.51±5.23, range 22.67--43.96) who had undergone total knee replacement surgery. Each sample was categorized according to its gross morphology, as severely damaged and was taken from the main defective area of maximal load. Macroscopic findings were validated by histological studies performed on 5 mm serial sections of cartilage samples and graded using the Mankin score. Specimens with osteoarthritis had Mankin score 10--14. Normal cartilage was obtained from small free cartilaginous fragments from ten individuals (six females and four males; mean age 61.70±18.17, range 27--78; mean BMI 23.80±4.34, range 19.03--35.06) with 0 Mankin score, undergoing fracture repair surgery, with no history of joint disease. Both patients and healthy individual\'s cartilage samples were obtained upon individuals\' verbal informed consent. The method of obtaining verbal approval by all individuals was approved by Institutional Review Board of the University Hospital of Larissa. Also the protocol was approved by the local ethics committee of University Hospital of Larissa. Detection of microRNA expression {#s2b} -------------------------------- Expression levels of 365 microRNAs were evaluated with TaqMan microRNA microarray assays as previously described [@pone.0003740-Thum1]. Validation of these results was performed using the mirVana qRT--PCR miRNA Detection Kit and qRT--PCR Primer Sets, according to the manufacturer\'s instructions (Ambion Inc, TX, USA). The U6 small nuclear RNA was used as an internal control. MicroRNA Northern Blot Analysis {#s2c} ------------------------------- For microRNA Northern Blot Analysis, 10ug of RNA were separated on 12% denaturating polyacrylamide gels and transferred to GeneScreen Plus membrane (PerkinElmer, Waltham). MiRCURY LNA Probes for miR-483 and miR-22 (Exiqon, Denmark) were end-labeled with T4 polynucleotide kinase. Prehybridization of the filters was carried out in 50% formamide, 0.5% SDS, 5· SSPE, 5·Denhardt\'s solution and 20 mg/ml sheared, denatured salmon sperm DNA. Hybridizations were performed in the same solution at 42°C. The labeled probes were heated for 1 min at 95°C before addition to the filters in the prehybridization solution. After hybridization, the membranes were washed in 0.1 SSC, 0.1% SDS at 42°C twice for 10 min. Reverse-Phase protein microarray analysis {#s2d} ----------------------------------------- Chondrocyte cell lysates were boiled for 5 min and were loaded into 384-well plates in serial dilutions (neat, 1∶2, 1∶4, 1∶8, and 1∶16) with negative control wells containing only lysis buffer. These samples were printed in duplicate onto nitrocellulose-coated glass slides (Schleicher & Schuell Bioscience, Keene, NH) using a ring-and-pin robotic arrayer (GMS 417, Affymetrix, Santa Clara, CA). The arrays were stained as previously described [@pone.0003740-Sheehan1] on an autostainer (DAKO, Carpinteria, CA) using a biotinyl-linked catalyzed signal amplification system (DAKO). Specificity of each antibody was tested by western blot analysis. Statistical analysis {#s2e} -------------------- All calculations were performed on a Microsoft computer, using the SPSS software (version 12.0). Correlation of between microRNA and protein expression levels with BMI was identified by correlation coefficients, calculated by Pearson rank correlation (*r*) and Spearman rank correlation. Statistical methods regarding the proteomic analysis are described analytically in the suppl. [Methods](#s2){ref-type="sec"} section. Construction and statistical significance of gene networks was performed by Ingenuity pathway analysis. Statistical significant networks were considered those with p value higher than 10^−5^. In addition clustering of the protein data in functional groups was performed using DAVID NIH Bioinformatics Database with a p value cut-off of 10^−5^. Quantification of western blots was performed by standard densitometric analysis. All transfection experiments were performed in triplicate and the results were compared by student\'s t-test analysis. Additional methods {#s2f} ------------------ Detailed experimental methods are described in the supplemental methods section ([**Methods S1**](#pone.0003740.s001){ref-type="supplementary-material"}). Results {#s3} ======= MicroRNA gene signature of osteoarthritis {#s3a} ----------------------------------------- To identify microRNAs involved in osteoarthritis, we tested the expression of 365 microRNAs in articular cartilage obtained from patients with osteoarthritis undergoing knee replacement surgery and from normal individuals with no history of joint disease. We identified 16 microRNAs differentially expressed in osteoarthritic compared to normal cartilage ([**Figure 1A**](#pone-0003740-g001){ref-type="fig"}). Specifically we detected nine up-regulated and seven down-regulated microRNAs in osteoarthritic cartilage compared to normal ([**Table S1**](#pone.0003740.s002){ref-type="supplementary-material"}). Real-time PCR and Northern blot analysis ([**Figure 1B, C**](#pone-0003740-g001){ref-type="fig"} **;** [**Table S2**](#pone.0003740.s003){ref-type="supplementary-material"}) validated that this 16 microRNA gene signature was able to distinguish osteoarthritic from normal chondrocytes. {#pone-0003740-g001} microRNA expression correlates with BMI {#s3b} --------------------------------------- Clinical characteristics of the patients and normal individuals ([**Table 1**](#pone-0003740-t001){ref-type="table"}) allowed us to study potential correlations between microRNAs expression and clinicopathological parameters. Very interestingly, we found five microRNAs to be statistically correlated with body mass index (BMI) ([**Figure 1C**](#pone-0003740-g001){ref-type="fig"}). miR-22 and miR-103 expression was positively correlated with BMI, while miR-25, miR-337 and miR-29a expression was inversely correlated, pointing towards the potential role of microRNAs in lipid metabolism and osteoarthritis pathogenesis. 10.1371/journal.pone.0003740.t001 ###### Clinicopathological characteristics of osteoarthritis patients and normal individuals {#pone-0003740-t001-1} OA Normal --------------------------- -------------- -------------- **Characteristic** **Female Sex -- no (%)** 28 (84.8%) 6 (60%) **Age at diagnosis -yr** Median 68.91±6.97 61.70±18.17 Range 57--83 27--78 **Body Mass Index (BMI)** Median 30.51±5.23 23.80±4.34 Range 22.67--43.96 19.03--35.06 Normal 4 6 Overweight 11 2 Obese 18 2 **Kellgrene-Lawrence** Median 3.72±0.51 0 Range 2--4 0 Proteomic analysis of articular cartilage {#s3c} ----------------------------------------- In order to identify deregulated proteins in osteoarthritic chondrocytes and study in detail whether obesity and osteoarthritis are correlated at a molecular level, we performed proteomic analysis in the same articular cartilage samples that we performed microRNA expression analysis. Specifically reverse-phase protein arrays were constructed [@pone.0003740-Gulmann1] ([**Figure S1A**](#pone.0003740.s006){ref-type="supplementary-material"}) and probed with antibodies to 214 proteins expressed in articular cartilage ([**Table S3**](#pone.0003740.s004){ref-type="supplementary-material"}). All antibodies were tested for their quality and specificity by Western blot analysis ([**Figure S1B**](#pone.0003740.s006){ref-type="supplementary-material"}) Arrays were scanned and dilution curves were used to quantify relative protein expression ([**Figure S1C**](#pone.0003740.s006){ref-type="supplementary-material"}). We detected that 76 proteins, (48 up-regulated and 28 down-regulated) were differentially expressed between osteoarthritic and normal chondrocytes ([**Figure 2A**](#pone-0003740-g002){ref-type="fig"}). These results were validated by western blot analysis ([**Figure 2B**](#pone-0003740-g002){ref-type="fig"}). We were able to identify for the first time deregulated proteins in osteoarthritic chondrocytes that were implicated in inflammatory and lipid metabolism pathways. More specifically, we detected up-regulation of proteins involved in inflammatory pathways such as IL1B, IL6 and CCR3, while proteins (PPARA, PPARG, ACOX1) involved in lipid metabolism mechanisms, were found highly down-regulated in osteoarthritic chondrocytes ([**Figure 2A, B**](#pone-0003740-g002){ref-type="fig"}). In addition we detected novel proteins that were deregulated in osteoarthritis, such as SOX11, FGF23, KLF6, WWOX and GDF15. {#pone-0003740-g002} Around half of the deregulated proteins in osteoarthritic chondrocytes were located in the extracellular space (50.67%), while the rest had cytoplasmic (20%), plasma membrane (16%) and nuclear (13.33%) localization ([**Figure 2C**](#pone-0003740-g002){ref-type="fig"}). Functional clustering analysis categorized the differentially expressed proteins in nine statistically significant pathways. Specifically, 45.33% of the deregulated proteins were involved in cartilage homeostasis pathways ([**Figure 2D**](#pone-0003740-g002){ref-type="fig"}), while 18.67% and 9.33% were involved in lipid metabolism and inflammation pathways, respectively. Metabolism-related proteins correlate with BMI {#s3d} ---------------------------------------------- Furthermore, we tried to correlate the expression levels of differentially expressed proteins in osteoarthritic cartilage with clinicopathological parameters, such as Body Mass Index. A recent study suggested that BMI is a significant risk factor for knee osteoarthritis leading to arthroplasty, speculating that biomechanics and metabolic factors associated with adipose tissue contribute to this phenotype [@pone.0003740-Lohmander1]. We identified that PPARA, BMP7, IL1B, LEP, ITGA5 and SREBP1 protein levels in osteoarthritic chondrocytes were highly correlated with BMI ([**Figure 2E**](#pone-0003740-g002){ref-type="fig"}), suggesting the potential role of metabolic-related proteins in the development of osteoarthritis. Detection of microRNA gene targets {#s3e} ---------------------------------- Since microRNAs exert their biological functions through suppression of target genes, it is important to identify microRNA-target pairs. As the available bioinformatic algorithms predict hundreds of microRNA-gene target pairs, it is evident that experimental data are needed for verification of these pairs. However, as it is known that several microRNAs target gene expression only at the protein and not at the mRNA level, the integration of microRNA along with protein data sets could be considered more effective for microRNA-gene target verification. Recently Huang *et al.,* used microRNA with cDNA expression profiling data to identify human microRNA targets using Bayesian data analysis algorithm [@pone.0003740-Huang1]. In our study, we identified microRNA-gene target pairs by matching microRNA and protein data. Subsequently, we filtered these data through three different selection criteria ([**Figure S2**](#pone.0003740.s007){ref-type="supplementary-material"}) and revealed 17 microRNA-gene target pairs implicated in osteoarthritis pathogenesis ([**Table S4**](#pone.0003740.s005){ref-type="supplementary-material"}). More specifically, we found microRNA-gene target pairs potentially involved in cartilage homeostasis and structure (miR-377-CART1, miR-140-ADAMTS5, miR-483-ACAN, miR-23b-CRTAP, miR-16-TPM2, miR-223-GDF5, miR-509-SOX9, miR-26a-ASPN), in biomechanic pathways (miR-25-ITGA5), in apoptotic mechanisms (miR-373-CASP6, miR-210-CASP10) and in lipid metabolism pathways (miR-22-PPARA, miR-22-BMP7, miR-103-ACOX1, miR-337-RETN, miR-29a-LEP). Several of these target genes such as ADAMTS5, GDF5 and LEP have been previously correlated with osteoarthritis [@pone.0003740-Miyamoto1], [@pone.0003740-Glasson1], [@pone.0003740-Iliopoulos1]. Gene networks in osteoarthritis {#s3f} ------------------------------- It is becoming increasingly clear that most proteins interact in complex cellular networks, the properties of which might be altered in osteoarthritic compared to normal chondrocytes. Therefore, global strategies aimed at modeling the functional interrelationships between microRNA and proteins, as complex interdependent networks, are required. We integrated microRNA and protein data sets in order to generate a model of macromolecular network that is perturbed in osteoarthritis using Ingenuity program analysis [@pone.0003740-Tongbai1]. The resulting network contained 11 microRNAs, 58 proteins and 414 potential functional associations ([**Figure 3A**](#pone-0003740-g003){ref-type="fig"}). We were able to detect three sub-networks representing key functional units that make up the co-expression network. A metabolism-related, an inflammation and a cartilage homeostasis sub-network were found to be interrelated contributing all together to cartilage destruction and osteoarthritis development ([**Figure S3**](#pone.0003740.s008){ref-type="supplementary-material"}). {#pone-0003740-g003} miR-22 correlates with PPARA and BMP7 protein expression {#s3g} -------------------------------------------------------- The functional significance of our predicted "interactome" network was tested by experimental validation. Our finding that specific microRNAs and proteins were related to BMI, focused our interest in identifying functional microRNA-gene target pairs relating obesity with osteoarthritis pathogenesis mechanisms. Epidemiological studies have shown that the risk for knee osteoarthritis is increased by 36% for every 2 units of BMI (5 kg) of weight gain [@pone.0003740-March1]. At first we tried to identify which microRNA-gene target interactions have biological significance (inverse correlation in expression levels) in our network. According to our previous combined *in silico* and expression data analysis we identified 5 microRNAs (miR-22, miR-103, miR-337, miR-25, miR-29a) and their 6 targets (PPARA, BMP7, ACOX1, RETN, ITGA5, LEP) that were highly correlated with BMI ([**Figure 1C**](#pone-0003740-g001){ref-type="fig"} **,** [**Figure 2E**](#pone-0003740-g002){ref-type="fig"}). Correlation of microRNA-gene target expression levels revealed that miR-22 was highly inversely correlated with PPARA (r^2^ = 0.919) and BMP7 (r^2^ = 0.816). ([**Figure 3B**](#pone-0003740-g003){ref-type="fig"}), while we detected low correlation between miR-29a with LEP (r^2^ = 0.491), miR-25 and ITGA5 (r^2^ = 0.456), miR-337 and RETN (r^2^ = 0.385) and miR-103 and ACOX1 (r^2^ = 0.252). The above results suggested the potential functional relationship between miR-22, PPARA and BMP7. miR-22 regulates PPARA and BMP7 in normal and osteoarthritic chondrocytes {#s3h} ------------------------------------------------------------------------- In order to detect whether PPARA and BMP7 were direct targets of miR-22, we performed luciferase assay. We found that BMP7-encoded mRNA contains a 3′UTR element that is partially complementary to miR-22 ([**Figure 3C**](#pone-0003740-g003){ref-type="fig"}) and luciferase assay showed that BMP-7 is a direct target of miR-22 (69% reduction, p\<0.001, [**Figure S4A**](#pone.0003740.s009){ref-type="supplementary-material"}). Furthermore, miR-22 overexpression in chondrocytes reduced the activity of a luciferase reporter gene fused to the PPARA 3′UTR (52% reduction, p\<0.001, [**Figure S4B**](#pone.0003740.s009){ref-type="supplementary-material"}). Evaluation of BMP7 and PPARA mRNA expression levels after miR-22 expression revealed that only BMP7 mRNA levels were significantly down-regulated (p\<0.001), while PPARA were not (p = 0.492) ([**Figure S5**](#pone.0003740.s010){ref-type="supplementary-material"}). Western blot analysis revealed that miR-22 regulates both PPARA and BMP7 protein expression levels in normal and osteoarthritic chondrocytes ([**Figure 3D**](#pone-0003740-g003){ref-type="fig"}). Overexpression of miR-22 inhibited BMP-7 (76%) and PPARA (93%) protein expression in normal chondrocytes. Subsequently, inhibition of miR-22 in osteoarthritic chondrocytes by antisense miR-22 treatment, highly up-regulated BMP-7 (8.35 fold) and PPARA (12.55 fold) expression, suggesting that miR-22 is a strong regulator of BMP-7 and PPARA proteins. All above data suggest that miR-22 regulates BMP-7 at the mRNA level and PPARA at the protein level, revealing thus the advantage of using proteomic instead of cDNA microarray data for detecting microRNA gene targets. Metabolic, inflammatory and cartilage homeostasis networks are inter-related {#s3i} ---------------------------------------------------------------------------- Subsequently, we proceeded by investigating how these miR-22 target gene pairs are correlated with the rest of the proteins present in the "interactome" network. Specifically, PPARA belongs to the metabolism sub-network, which is connected with the inflammation sub-network through IL1B. A recent report suggested that PPARA is a receptor involved in inflammatory processes [@pone.0003740-Cuzzocrea1] and was recently found down-regulated in osteoarthritic cartilage [@pone.0003740-Watters1]. In our study, the inflammatory sub-network having IL1B and IL6 as central nodes is connected with MMP13, which is central node of the cartilage structure sub-network. To test this hypothesis predicted by the gene network of IL1B-MMP13 interaction, we treated normal and osteoarthritic chondrocytes with IL1B and examined MMP13 expression. We found that IL1B up-regulated MMP13 expression both at mRNA and protein levels ([**Figure 4A, B**](#pone-0003740-g004){ref-type="fig"}), verifying the correlation that we had previously described between IL1B and MMP13 expression levels in clinical samples [@pone.0003740-Simopoulou1]. In order to understand how IL1B affects not just MMP13 but cartilage homeostasis pathways, we over-expressed IL1B in normal chondrocytes and monitored the expression of the proteins involved in the cartilage network. IL1B over-expression pertubated the cartilage homeostasis sub-network by activation of metalloproteinases and aggrecanases and down-regulation of cartilage structural proteins. ([**Figure 4C**](#pone-0003740-g004){ref-type="fig"}), connecting thus inflammation and cartilage homeostasis sub-networks with osteoarthritis development. {#pone-0003740-g004} PPARA and BMP7 regulate IL1B and MMP13 expression in chondrocytes {#s3j} ----------------------------------------------------------------- In order to delineate the PPARA-IL1B-MMP13 potential pathway we followed an RNA interference strategy. More specifically, siRNA inhibition of PPARA increased IL1B (2.5 fold) and MMP13 (3.5 fold) expression levels ([**Figure 5A**](#pone-0003740-g005){ref-type="fig"}). IL1B was found to be highly induced 24 h after siRNA PPARA treatment, while MMP13 was highly induced 48 h after siRNA treatment, suggesting a sequential activation of MMP13 by IL1B. {#pone-0003740-g005} The second target of miR-22, BMP7, is frequently down-regulated in osteoarthritic cartilage, while BMP7 overexpression induces cartilage formation *in vitro* and *in vivo* [@pone.0003740-Im1]. It has been shown that IL1B and ACAN (aggrecan) [@pone.0003740-Yeh1] levels are regulated by BMP7 in osteoarthritic cartilage. These correlations were present in the cartilage homeostasis sub-network and in addition we found that siRNA against BMP7 resulted in increased IL1B and MMP13 (2 fold) expression in normal chondrocytes ([**Figure 5B**](#pone-0003740-g005){ref-type="fig"}). Furthermore, we detected high correlation between BMP7 and ACAN mRNA levels ([**Figure 5C**](#pone-0003740-g005){ref-type="fig"}) and found that BMP7 siRNA down-regulation blocked effectively ACAN expression ([**Figure 5D**](#pone-0003740-g005){ref-type="fig"}). miR-22 blocks MMP13 activity and inhibits cartilage destruction {#s3k} --------------------------------------------------------------- Overexpression of miR-22 in normal chondrocytes resulted in increased IL1B (5.8 fold) and MMP13 (8.1 fold) expression and decreased aggrecan (4.9 fold) expression ([**Figure 6A**](#pone-0003740-g006){ref-type="fig"}). These results point towards the implication of the combinatory effect in MMP13 up-regulation (8.1 fold instead of 3.5 and 2 fold) by miR-22 overexpression through PPARA and BMP7. Additionally, inhibition of miR-22 in osteoarthritic chondrocytes up-regulated PPARA (4.9 fold) and BMP7 (5.8 fold) expression, blocked the inflammatory process, through inhibition of IL1B (7.6 fold), inhibited catabolic changes such as MMP13 expression (7.9 fold) and activated the cartilage repair protein aggrecan (3.1 fold) ([**Figure 6B**](#pone-0003740-g006){ref-type="fig"}), suggesting the therapeutic potential of microRNA inhibition in osteoarthritis. Furthermore, Western blot, ELISA and immunofluorescence experiments revealed a high decrease of MMP13 expression in osteoarthritic chondrocytes after miR-22 inhibition ([**Figure 6 C--E**](#pone-0003740-g006){ref-type="fig"}). MMP13 is one of the major pathophysiological mediators of cartilage destruction, through degradation of type II collagen in osteoarthritis [@pone.0003740-Billinghurst1] and therefore its down-regulation is of great clinical importance. {#pone-0003740-g006} Discussion {#s4} ========== We detected, for the first time to our knowledge, a 16 microRNA gene signature differentially expressed in osteoarthritis. The biological significance of microRNAs is determined by their gene targets which currently can be predicted computationally by prediction algorithms. In order to have experimental data for identification of microRNA gene targets, we performed proteomic analysis, using reverse-phase protein arrays, which revealed a 76 protein signature differentially expressed in osteoarthritis. This technology platform has been designed for quantitative multiplexed analysis of cellular proteins from a limited amount of sample [@pone.0003740-VanMeter1], offering thus a major advantage for protein quantification levels from tissues such as cartilage, where the material is frequently limited. Proteomic analysis was validated by detecting differentially expressed genes such as GDF5, that have been identified as osteoarthritis susceptibility genes by gene association studies [@pone.0003740-Miyamoto1]. Up to date most osteoarthritis molecular studies focus their interest in cartilage homeostasis mechanisms. However our global approach identified proteins related with osteoarthritis pathobiology, that are involved in lipid metabolism and inflammatory pathways, providing a new source of protein expression data that need to be explored in greater detail. For example, we identified novel proteins such as SOX11, CCR3, WWOX to be differentially expressed in osteoarthritic cartilage. Sock et al, have shown that SOX11 is an important transcription factor related with skeletal malformations [@pone.0003740-Sock1]. Similarly, recently Aqeilan et al showed that WWOX -/- mice develop metabolic bone disease [@pone.0003740-Aqeilan1]. In addition CCR3 has been found up-regulated in adipose tissue from obese individuals [@pone.0003740-Huber1]. All these studies suggest that these proteins are directly or indirectly related with osteoarthritis and their differential expression observed in our study in osteoarthritic chondrocytes supports the idea that these proteins are involved in cartilage destruction pathways. However, additional molecular studies are needed to clarify the role and importance of these proteins in osteoarthritis development. Comparison of proteomic and clinical data revealed metabolism-related proteins differentially expressed in osteoarthritic chondrocytes to be correlated with BMI. It has been described that mechanoreceptors are activated in knee chondrocytes due to excess of weight contributing to cartilage destruction. Very interestingly, our proteomic analysis revealed that ITGA5 mechanoreceptor protein was highly up-regulated in osteoarthritic chondrocytes and was correlated with BMI. Previous studies have described integrin-dependent signalling cascades in chondrocyte mechanotransduction. Furthermore Chowdhury et al., showed an integrin-mediated mechanotransduction in IL1B-stimulated chondrocytes, suggesting the relationship between cartilage structure and inflammatory pathways [@pone.0003740-Chowdhury1]. The present work has led to the identification of an "interactome" network involved in the pathogenesis of osteoarthritis. We have shown that integration of microRNA, protein expression and clinical data can be used to generate a network of potential functional associations with osteoarthritis. We were able to validate experimentally the interplay between metabolism genes, inflammatory molecules and cartilage homeostasis enzymes through microRNA mechanism of action. Specifically we found regulation of IL1B and MMP13 by PPARA and BMP7 through miR-22. Consistent with our studies, Watters *et al*, showed that PPARA expression is reduced in a STR/Ort osteoarthritis mouse model [@pone.0003740-Watters1]. IL1B is a central node in our network and perturbation of its expression (over-expression) contributes to cartilage destruction. Our molecular data are consistent with several clinical studies implicating the role of obesity and inflammation in cartilage destruction. Several clinical studies have suggested the effect of obesity in the development of osteoarthritis [@pone.0003740-Lementowski1], [@pone.0003740-Aspden1], however there are no extensive studies correlating lipid metabolism with osteoarthritis at the molecular level. Specifically, Lohmander et al., in a large population study revealed that body mass is a significant risk factor for osteoarthritis leading to arthroplasty [@pone.0003740-Lohmander1]. In addition Marks proposed that a high body mass is present in most adults with osteoarthritis [@pone.0003740-Marks1]. Gene network approaches provide new insights for elucidating the complexity of diseases such as osteoarthritis. Several genes consist the sub-networks that are interconnected creating large gene networks. Alterations in gene expression that are able to perturb a network have a causal relationship with disease [@pone.0003740-Sieberts1]. The integration of gene expression profiling and clinical data provides a detailed picture of how a network state is correlated with disease and furthermore leads to the development of new treatments that target the gene network as opposed to current therapeutic approaches focused on targeting one specific gene only. This strategy will help to improve the understanding of the pathogenesis of multifactorial diseases such as osteoarthritis and provide possible novel therapeutic targets. Supporting Information {#s5} ====================== ###### (0.04 MB DOC) ###### Click here for additional data file. ###### (0.07 MB PPT) ###### Click here for additional data file. ###### (0.07 MB PPT) ###### Click here for additional data file. ###### (0.13 MB PPT) ###### Click here for additional data file. ###### (0.07 MB PPT) ###### Click here for additional data file. ###### (0.16 MB PPT) ###### Click here for additional data file. ###### (0.12 MB PPT) ###### Click here for additional data file. ###### (0.68 MB PPT) ###### Click here for additional data file. ###### (0.14 MB PPT) ###### Click here for additional data file. ###### (0.14 MB PPT) ###### Click here for additional data file. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**Funding support was provided from Hellenic Association of Orthopaedic Surgery and Traumatology. Funding organizations did not have any role in study design, data collection, interpretation of the results, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: DI AT. Performed the experiments: DI PO. Analyzed the data: DI PO AT. Contributed reagents/materials/analysis tools: DI KNM AT. Wrote the paper: DI AT. [^2]: Current address: Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School, Boston, Massachusetts, United States of America

Introduction {#sec1-1} ============ Obstructive sleep apnea (OSA) is an increasingly prevalent condition that is characterised by repetitive upper airway obstructions resulting in intermittent hypoxia and sleep fragmentation caused by arousals \[[@ref1]\]. Among adults, 30--70 years of age, approximately 13% of men and 6% of women, have moderate to severe forms of OSA \[[@ref2]\]. OSA is often closely associated with other conditions which are recognised causes of morbidity and mortality such as obesity, metabolic syndrome, atherosclerosis, systemic inflammation, insulin resistance and type 2 diabetes mellitus \[[@ref3], [@ref4]\]. Recently, there has been a great interest in the interaction between OSA and metabolic dysfunction. There is no consistent data suggesting that OSA is a risk factor for dyslipidemia. Indeed, conflicting results have been observed in cross-sectional and interventional studies \[[@ref5]\]. Taking into account components of the metabolic syndrome, some reports found increased levels of triglycerides \[[@ref6]-[@ref9]\] and reduced levels of high-density lipoproteins (HDL) in patients with OSA \[[@ref8]-[@ref10]\], while others studies did not find the correlation between OSA and dyslipidemia \[[@ref11],[@ref12]\]. Of note, the majority of the studies were not specifically designed to evaluate the lipid profile. Therefore, more evidence is still needed. Increased understanding of the independent associations between OSA, metabolic syndrome and insulin resistance is important to develop appropriate therapeutic strategies to reduce the high cardiometabolic risks in patients with OSA. The aim of this cross-sectional study was to evaluate the prevalence of lipid abnormalities in patients suspected for OSA referred to our sleep laboratory for polysomnography. Material and Methods {#sec1-2} ==================== The study included 200 patients. It was conducted at University Clinic of Pulmonology and Allergy in Skopje. Inclusion criteria for patients were age from 35 to 60 years and persistence of minimum 2 of 3 clinical symptoms of OSA. The symptoms were snoring, witnessed apnea and daytime sleepiness. Exclusion criteria were previous history and treatment of diabetes and lipid abnormalities. The study was approved by Ethical Committee of the Faculty of Medicine with No. 03-941/2, and before the study procedures, informed consent was obtained from all patients. Body mass index (BMI) was calculated, and patients were divided into two groups according to the BMI. All patients underwent polysomnography (Respironix, model Alice 5). All results from polysomnography were scored manually according to standard criteria \[[@ref13]\]. Apnea, hypopnea and arousals were also identified according to the standard criteria and summarised in the form of a respiratory disturbance index (RDI). All patients with RDI above 15 were diagnosed with OSA. In the morning after 12 hours fasting, a blood sample was collected from all patients. Blood levels of triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL) and low-density lipoprotein cholesterol (LDL), were assessed. Biochemical measurements were conducted using an Architect Abbott C8000 auto analyser. Statistical analyses were performed using Statistical software (Stat Soft). Data were expressed as *mean (X)* and *standard deviation (SD)*. Comparisons between variables were made using the unpaired t test for parametric data. The multiple linear regressions were used to determine the association between OSA and metabolic parameters. Statistical significance was considered at p less than 0.05. Results {#sec1-3} ======= From all study patients, 51 were female with an average age of 49 ± 9 years, and 149 were men average age of 47 ± 9 years. There was no significant difference in age, BMI and RDI between male and female. There was the significant difference in the occurrence of OSA in men versus women, 109 (73.2%) of males and 31 (62.8%) of females were OSA positive (p \< 0.03). According to BMI, patients in the study were divided into two groups. There were 120 non-obese patients with BMI ≤ 30 kg/m\^2, and 80 obese patients with BMI \> 30 kg/m\^2. In a non obese group with BMI ≤ 30, 62 patients were OSA negative and 58 patients were OSA positive. In an obese group with BMI \> 30, 14 patients were OSA negative, and 66 patients were OSA positive ([Figure 1](#F1){ref-type="fig"}). {#F1} OSA patients had statistically significant higher BMI, triglycerides, total cholesterol and lower HDL when compared to OSA negative patients ([Table 1](#T1){ref-type="table"}). There was no statistical difference in age and LDL levels between these two groups of patients. ###### Comparison between OSA positive and OSA negative patients OSA negative RDI \< 15 (76 patients) OSA positive RDI \> 15 (124 patients) --------------- -------------------------------------- --------------------------------------- ------- ------- ------- RDI 5.04 3.81 43.78 18.71 0.000 Age (years) 47.29 9.73 48.22 8.49 NS BMI (kg/m\^2) 27.58 3.14 31.11 4.35 0.000 TG (mmol/l) 1.60 0.36 1.76 0.26 0.000 TC (mmol/l) 4.94 0.50 5.22 0.34 0.000 HDL (mmol/l) 1.45 0.23 1.34 0.23 0.001 LDL (mmol/l) 2.85 0.53 2.94 0.49 NS OSA = Obstructive sleep apnea; RDI = Respiratory disturbance index; BMI = Body mass index; TG = Triglycerides; TC = Total cholesterol; HDL= High-density lipoprotein cholesterol; LDL = Low-density lipoprotein cholesterol. In the study, both OSA positive and OSA negative patients were divided according to BMI in two groups, first group with BMI ≤ 30 and the second group with BMI \> 30. OSA positive patients with BMI ≤ 30 had statistically significant higher levels of triglycerides and total cholesterol, and statistically significant lower level of HDL compared to OSA negative patients with BMI ≤ 30. There were no statistically significant differences in age and LDL levels between these groups ([Table 2](#T2){ref-type="table"}). ###### Comparison between OSA positive and OSA negative patients with BMI ≤ 30 BMI ≤ 30 --------------- ---------- ------ ------- ------- ------- RDI 4.65 3.41 38.68 16.92 0.000 Age (years) 47.08 9.56 47.62 8.38 NS BMI (kg/m\^2) 26.55 2.40 27.38 1.80 0.035 TG (mmol/l) 1.53 0.28 1.66 0.18 0.003 TC (mmol/l) 4.90 0.49 5.10 0.28 0.010 HDL (mmol/l) 1.55 0.22 1.35 0.19 0.000 LDL (mmol/l) 2.84 0.52 2.79 0.36 NS RDI = Respiratory disturbance index; BMI = Body mass index; TG = Triglycerides; TC = Total cholesterol; HDL= High-density lipoprotein cholesterol; LDL = Low-density lipoprotein cholesterol. OSA positive patients with BMI\>30 had higher triglycerides, total cholesterol and LDL and lower HDL versus OSA negative patients with BMI\>30, but without statistically significant differences ([Table 3](#T3){ref-type="table"}.) ###### Comparison between OSA positive and negative patients with BMI \> 30 BMI \>30 --------------- ---------- ------- ------- ------- ------- RDI 6.81 5.01 48.26 19.17 0.000 Age (years) 48.21 10.76 48.74 8.62 NS BMI (kg/m\^2) 32.14 1.59 34.38 3.11 0.011 TG (mmol/l) 1.85 0.49 1.90 0.28 NS TC (mmol/l) 5.12 0.53 5.33 0.35 NS HDL (mmol/l) 1.37 0.25 1.30 0.23 NS LDL (mmol/l) 2.94 0.60 3.08 0.55 NS RDI = Respiratory disturbance index; BMI = Body mass index; TG = Triglycerides; TC = Total cholesterol; HDL= High-density lipoprotein cholesterol; LDL = Low-density lipoprotein cholesterol. When all parameters were analysed with multiple linear regressions, only BMI, total cholesterol levels and LDL levels were found to be independent predictors of OSA ([Table 4](#T4){ref-type="table"}). ###### Independent predictors of OSA OR 95%CI p ----- ------ ----------- ------- BMI 1.6 1.39-1.83 0.000 TC 1.73 1.38-2.16 0.000 LDL 1.47 1.19-1.81 0.000 BMI = Body mass index; TC = Total cholesterol; LDL = Low-density lipoprotein cholesterol. Discussion {#sec1-4} ========== OSA is the potent risk factor for metabolic disorders. The mechanisms through which OSA may worsen metabolism are complex. It may trigger several pathological mediating pathways (sympathetic activation, neurohumoral changes, glucose homoeostasis disruption, inflammation and oxidative stress) through chronic intermittent hypoxia (CIH), and these may ultimately cause deterioration in the metabolic function \[[@ref14], [@ref15]\]. According to previous studies, the prevalence of OSA is increased fourfold in patients with obesity. Obesity plays a major part in the development of the metabolic syndrome, which consists of insulin resistance, diabetes or impaired glucose tolerance, hypertension, and lipoproteinemia \[[@ref16]\]. In this study, we have demonstrated that OSA positive patients had significantly higher level of triglycerides, total cholesterol and decreased HDL cholesterol levels versus OSA negative patients. LDL was also higher in OSA patients but with no significant value. There were statistically significant differences in BMI between OSA positive and negative patients. So, the question is, does OSA affect lipid metabolism by itself or obesity is playing the major role in metabolic changes in these groups of patients. The relationships between OSA and various lipid parameters have not been extensively investigated like other components in the metabolic syndrome and the results have been more diverse. Studies of sleep clinic cohorts have consistently reported a higher prevalence of dyslipidemia in OSA positive subjects compared to those without OSA \[[@ref17], [@ref18]\]. The American Heart Health Sleep Study reported that apnea-hypopnea index (AHI) was inversely related to HDL-cholesterol levels in younger men and women, but not in older men, and triglyceride levels in younger men and women only \[[@ref19]\]. In contrast, Lam et al. evaluated 255 patients between 30 and 60 years, and they did not find the association between OSA and HDL or TG levels, after controlling for confounding variables \[[@ref20]\]. In our study, after dividing patients according to BMI, OSA positive patients with BMI \< 30 had statistically significant higher levels of triglycerides and total cholesterol, and statistically significantly lower levels of HDL versus OSA negative patients with BMI ≤ 30. This result is corresponding with previously cited studies \[[@ref6]-[@ref8], [@ref9], [@ref17]-[@ref19]\]. However, several studies that were searching for an association between metabolic syndrome and normal weight, over weight and obese patients, reported that prevalence of metabolic syndrome in non-obese patients (BMI 25.0-26.9 kg/m\^2) is between 9.6-22.5%, depending on ethnicity and sex \[[@ref21], [@ref22]\]. So the possibility that OSA mechanisms are worsening lipid metabolism in a non obese group of patients is very high rather that weight gain. In obese patients, both OSA positive and OSA negative, there were no statistically significant differences in lipid blood levels. This result is corresponding with others studies. Sahin et al. in their study found out that OSA positive obese patients had statistically significant higher levels of lipids compared to OSA positive patients with normal weight. But in obese patients, both OSA positive and OSA negative, there were no statistically significant differences in blood levels of lipids \[[@ref23]\]. Sharma et al. compared three groups of patients, 40 obese OSA positive patients, 40 obese OSA negative patients and 40 normal weight OSA negative patients and found that there was no difference in metabolic status between obese OSA positive and obese OSA negative patients \[[@ref24]\]. Schäfer et al. found no relationship between OSA and concentration of lipoproteins in 81 male subjects \[[@ref25]\]. Results from national surveys suggest that dyslipidemias are the most common comorbidities associated with a range of body mass indices (BMI), with substantial increases found with increased body weight. It is estimated that about 68% of obese adults in the National Health and Nutrition Examination Survey population had metabolic abnormalities \[26\]. However, the limiting factor of this study may be not so large number of patients, particularly the small number of OSA negative patients with a BMI \> 30. It should be noted that patients with previous history and treatment of diabetes and lipid disorders were excluded from the study. In conclusion, OSA and obesity are the potent risk factor for dyslipidemias. OSA could play the significant part in worsening of lipid metabolism in non-obese patients. But in obese patients, the extra weight makes the metabolic changes of lipid metabolism, and the roll of OSA is not that very important like in non-obese patients. **Funding:** This research did not receive any financial support. **Competing Interests:** The authors have declared that no competing interests exist.

Introduction {#Sec1} ============ Recent times have seen higher demand for sustainable and green products reproduced from waste materials^[@CR1]^. In particular, this has called for more research and development endeavors to recycle waste materials into biodegradable products with low environmental impacts^[@CR2]^. The potential of several types of cellulose-based materials from agro-wastes such as palm oil, pineapple, kenaf, sisal, etc. as reinforcing materials in cement composites has been discovered decades ago, but yet to be thoroughly understood and applied in real-world construction^[@CR3],[@CR4]^. Cellulose can be extracted from various plants while oil palm empty fruit bunch is one of its sources. Malaysia is the second-largest producer of palm oil and the country's palm oil industry produces about 90 million tonnes of lignocellulosic biomass, including empty fruit bunches, oil palm trunks, and oil palm fronds, as well as palm oil mill effluent^[@CR5]^. Reinforcement of natural fibers and cementitious matrices from various sources of plant fiber have found to improve the mechanical strength of the composites and currently is being applied in various industrial sectors including construction, automobiles, aerospace's, etc.^[@CR6],[@CR7]^. In particular, renewable waste materials such as agro-waste have enhanced the mechanical properties of cement mortar^[@CR8]^. However, direct incorporation of natural fiber material into mortar leads to low concrete workability, decay problems, low resistance to chemical attack, and other structural problems^[@CR9]^. The most common approach to overcome these problems is through surface modification of plant fibers using various chemical treatments (alkali, silane, ozone, etc.) and physical treatment (plasma, UV radiation, corona, etc.)^[@CR10]--[@CR12]^. Another approach to improve the properties of natural fibers is through a top-down approach by deconstructing the structure of the plant itself into nano or microfibrous materials such as nanocrystalline cellulose (CNCs) and microcrystalline cellulose (MCC)^[@CR13]^. In recent days, construction industries conduct various researches to discover new eco-friendly reinforcement materials to reduce cement usage. Previous few examples of the reinforcement material that contributes to minimizing the cement usage include steel, glass, carbon, etc.^[@CR14],[@CR15]^. In order to replace this material, various high-performance nanostructures such as carbon nanotubes have been utilized to improve the performance of cement composites^[@CR16],[@CR17]^. However, troubles in dispersion and cost-effectiveness are the major crisis with these materials, which are needed to be addressed before commercialization^[@CR7]^. At present, the performance of nano and micro cellulose materials as reinforcement of cementitious composites is getting research attention worldwide. Positive result discovered by Cao *et al*.^[@CR12]^ through the use of CNCs extracted by using the acid hydrolysis process. An enhancement of cement mortar up to 30% for the flexural strength with the addition of 0.2% CNCs was found^[@CR12]^. Another achievement found with the addition of CNCs in oil well cement studied by Reza *et al*.^[@CR18]^, revealed that CNCs reduced the porosity up to 33%, surface area to 66% and 0.7% of the total design water by mass. Also, CNCs have raised the compressive and tensile strength by 60% in the first 24 hours of composite's age. However, research on CNCs based cementitious composites extracted from palm oil fruit bunches are very rare in the existing literature. Therefore, this present study intends to report the effect of CNCs suspension on the mechanical properties of mortar via different: i) curing environment, ii) percentage of CNCs added and iii) morphological observation. A series of experiments were conducted to examine the effect of different curing methods (water, lime and wrapping curing) to find the best method of curing, as well as the microstructural and mechanical properties of the cement composites after adding CNCs. Materials and Methods {#Sec2} ===================== The cement mortar mix used in this study were prepared by mixing the CNCs aqueous suspension together with fine aggregates and cement at 0.5 water/cement ratio^[@CR19]^. The amount of CNCs liquid suspension added to cement composites was from 0% to 0.8% by volume of cement content. The dispersion behavior of the CNCs in aqueous suspension was found to be more stable compared to the powder form. Preparation of the CNCs {#Sec3} ----------------------- The cellulose was supplied by a local company from Waris Nove Sdn. Bhd. at Kuantan, Pahang Malaysia that commercially produces α-cellulose from palm oil wastes for industrial applications. The extraction process began by referring to the extraction methods from Dong *et al*.^[@CR20]^ and Lu and Hsieh^[@CR21]^ with the production of CNCs from α-cellulose. The first step is to extract the microcrystalline cellulose (MCCs) from α-cellulose^[@CR20]^. In order to produce MCC, 2.5 N hydrochloric acid (HCl) was mixed with α-cellulose and incubated for 15 minutes at the controlled temperature of 105 °C. This is followed by the addition of cold water to the hot mixture, stirred and the mixture was left overnight. The mixture was filtered and then washed with water until the pH reached 6\~7. The filtered sample was dried in a hot air oven for 60 minutes at 60 °C. The finally the sample was ground and sieved using 60 µm sieve aperture before it was extracted for CNCs. Based on Kumar *et al*.^[@CR22]^, during the CNCs extraction, 64% w/v sulphuric acid (H~2~SO~4~) was used for the acid hydrolysis process. The acid solution was initially preheated to 45 °C and MCCs were added at a ratio of 10:1 (diluted H~2~SO~4~: MCCs). Subsequently, the solution was stirred for 60 minutes, mixed with 1/10 fold of chilled distilled water, and centrifuged at 6000 rpm for 15 minutes to remove any excessive acid. Then, the remaining precipitate underwent dialysis for 5-7 days for neutralization. Finally, the solution was centrifuged and subjected to sonification for 15 minutes to form CNCs aggregates. The final product was refrigerated at 4 °C until the further application. In general, when MCCs was mixed with sulphuric acid (through acid hydrolysis process) it changes the MCCs to CNCs^[@CR23]--[@CR25]^. Sulfuric acid hydrolysis of cellulose is a heterogeneous process where the acid diffuses into the pulp fiber and cleaves the glycosidic bonds in the cellulose polymer. Depending on reaction times, temperature, and how the heating rate is controlled, the hydrolysis could also occur on the crystalline regions and some of the hydroxyl groups on the crystalline surface and convert into sulfate groups (e.g., conversion of cellulose-OH to cellulose-OSO~3~−H^+^). CNCs can be generated which is in a milky white color but not as brownish or blackish color. The brownish and blackish color solution shows that the CNC is burned due to improper selection of acid concentration and temperature. Therefore, in this study concentration of sulphuric acid used was 64% w/v with a temperature of 45 °C for 60 minutes. With this condition of acid hydrolysis, the end product of CNC comes out to be milky white in color. Figure [1(a)](#Fig1){ref-type="fig"} shows the MCCs powder after the α-cellulose pre-treatment process and Fig. [1(b)](#Fig1){ref-type="fig"} displays the CNCs aqueous suspension after the acid hydrolysis process (milky white in color). The schematic illustration of the CNCs extraction process was simplified in Fig. [2](#Fig2){ref-type="fig"}. The chemical composition of the extracted CNCs from palm oil wastes used in this study was evaluated via X-ray Fluorescence (XRF) assessment and tabulated in Table [1](#Tab1){ref-type="table"}. The main constituents of the CNCs were detected as carbon and oxygen. A low amount of sulfur was also detected, most probably from the leftover acid during the acid hydrolysis process with the H~2~SO~4~ solution.Figure 1Two different types of cellulose production (**a**) MCC powder (**b**) CNCs aqueous suspension.Figure 2Schematic illustration of CNCs extraction process.Table 1Chemical composition of cellulose nanocrystals used as an admixture in cement composites.Chemical CompositionMass Percentage (%)Carbon, C43.76Oxygen, O55.12Sulfur, S1.01Others0.11 Cement mortar samples preparation and strength tests {#Sec4} ---------------------------------------------------- ### Cement and sand preparation {#Sec5} Portland Cement Type I was used throughout the study. This was obtained from a Tasek Cement Company located at Ipoh, Malaysia. Based on ASTM C778-113^[@CR26]^, two types of sand with different sizes were blended in the equal portion into the mortar mixes, i.e passing 850-µm sieve and retained at 600-µm sieve and passing 600-µm sieve and retained at 150-µm sieve^[@CR26]^. ### Mortar samples casting {#Sec6} The specimens consisted of 50 mm cubes and 40 mm by 40 mm by 160 mm prism for compressive and flexural strength tests, respectively. By following the ASTM C109/C109M^[@CR19]^, the specimens were prepared based on the optimum mix design of 1:2.75 (cement:sand) with a water/cement (w/c) ratio of 0.5 and five different content of CNCs by volume of cement (0% for control specimen, increased every 0.2% until reaching 0.8%) as additive. Based on previous literature, the addition of CNCs in cement composites only up to 0.2% and 0.4%. Therefore, a range of 0% (as a control specimen) to 0.8% is chosen in this study to understand the effect of CNCs addition in mortar can improve or reduce its performance in strength^[@CR27]--[@CR29]^. The workability of the mortar was obtained via the flow table method as stated in ASTM C230/C230M-14^[@CR30]^. After mixing, the mortar was poured into the specimen moulds and compacted using a vibrating table for two minutes with a vibrating rate of 12000 ± 400/minute^[@CR30]^. ### Mortar samples curing procedures {#Sec7} The hardened mortar specimen was cured in three different conditions -- water, lime, and wrap curing for 28 days. Water curing was done by immersing the mortar samples into tap water until it reaches the testing day. As for lime curing, the specimens were prepared using a saturated lime solution, which was prepared by mixing 2 grams of hydrated lime in every liter of water. Wrap curing was conducted with at least three layers of polyethylene film wrapping onto mortar samples to prevent water from evaporating during the hydration process. The ambient temperature was maintained at 23 ± 2 °C with more than 50% humidity. Finally, all samples were tested for compressive strength. All the experiments were performed in triplicates. ### Mechanical properties {#Sec8} The mechanical properties of the cement composite with CNCs were studied for its compressive and flexural strength for 7, 14 and 28 days. Compressive and flexural strength test was evaluated using a Universal Testing Machine (MATEST3000) with a maximum loading capacity of 3000 kN and a pacing rate of 0.9 kN/s for compressive strength and 0.04 kN/s for testing flexural strength test^[@CR31]^. CNCs-mortar characterization {#Sec9} ---------------------------- The microstructure of the CNCs mortar was studied under Field Emission Scanning Electron Microscope (FESEM, JSM 670-1 F, JEOL, Japan) and Scanning Electron Microscope/Energy Dispersive X-ray Spectroscopy (SEM/EDS, Quanta 650 MLA-FEG, FEI, Australia, acceleration voltage: 20 kV, coating:30 nm Au-Pd) to understand the changes induced by CNCs on cement composites. Since the resolution of SEM/EDS observation has its limitations, the FESEM test was conducted to further observe morphological changes previously identified by the SEM/EDS under the higher resolution. During this morphological study, various magnifications were applied to the control sample and CNCs-mortar sample. For instance, using FESEM, the magnification was 25,000 times for the control sample, but 10,000 times for the CNCs-mortar sample. The same applies to the SEM/EDS monitoring, with magnification was set at 766 times for the control sample and 854 times for CNCs-mortar sample. In order to study the crystalline pattern of calcium hydroxide formation (portlandite) in cement composites after incorporating CNCs, the X-ray Diffraction (XRD, Rint 2000 Ultima- III, RIGAKU Corporation, Japan) test was conducted using an X-ray Diffractometer with graphite filtered CuK (λ = 1.5433 Å) radiation at 40 kV and 30 mA. The data were collected on a 2θ scale from 0 to 80°. The zeta potential evaluation was done by Zetasizer Nano ZS90: DLS Zeta Potential Analyzer from Malvern Instrument for cement particle and CNCs aqueous suspension to study the dispersion behavior of CNCs before and after mixing with the cement materials. This is important to make sure the CNCs can be uniformly distributed instead of becoming agglomerated during the mixing process. For this reason, both samples were diluted with distilled water until achieving a concentration of 0.01 mol/l with pH 7. The refractive index of the CNCs and cement particles were set to 1.46 and 1.35, respectively. Results and Discussion {#Sec10} ====================== The effects of different curing methods on CNCs mortar {#Sec11} ------------------------------------------------------ As depicted in Fig. [3](#Fig3){ref-type="fig"}, all operational parameters of different curing methods have resulted in a myriad of strengths. Among all curing methods, the wrapping method showed the highest improvement in strength after 0.2% of CNCs were added into the mortar matrix, achieving a compressive strength of 49 MPa at the age of 28 days with a 48% increment compared to the control mortar. On the other hand, for lime and water curing methods, the compressive strength of 47 MPa and 43 MPa, respectively, was recorded with the same amount of CNCs added; these are translated into 43% and 42% increment, respectively, compared to the control sample. Therefore, the wrap method was chosen to be used as the curing method throughout the study due to its effect on compressive strength.Figure 3Compressive strength of cement mortar at age of 28 days with different curing methods and CNCs concentration. The hydration of cement mortar must have adequate control of moisture and temperature movement from and into the composites' matrices. Furthermore, a proper curing process -ensures good protection of cement mortar against damage due to loading and mechanical interference^[@CR2],[@CR32]^. Fundamentally, the resistance towards loading and mechanical impacts comes from a good production of calcium crystals (calcium-silicate-hydrate, C-S-H, and calcium hydroxide, Ca(OH)~2~) compounds in cement mortar^[@CR33]^. These compounds form after the water combines with cement particle compounds (tricalcium silicate, C~3~S), as stated in Eq. ([1](#Equ1){ref-type=""}) below:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{C}}}_{3}{\rm{S}}+(1.3+{\rm{x}}){\rm{H}}\to {{\rm{C}}}_{17}{{\rm{SH}}}_{{\rm{x}}}+1.3{\rm{CH}}.$$\end{document}$$Where: C~3~S = tricalcium silicate CSH = calcium silicate hydrate CH = calcium hydroxide Equation ([1](#Equ1){ref-type=""}) was developed by Allen *et al*.^[@CR34]^ where "x" denotes the variables of water. This shows that the amount of water available during the curing process greatly affects the production of calcium crystals during hydration. On the contrary, insufficient moisture promotes the formation of pores and microcapillary pores, which directly decreases the strength and long-term durability of the concrete. The lime and water curing methods stated in this study have undoubtedly provided sufficient moisture for hydration compared to the wrapping method. However, the wrapping method had unexpectedly given the best result in terms of strength. This is attributed to the ability of moisture preservation (vapor pressure) for optimum hydration. Hence, by controlling the movement of water outside and inside the cement matrices during the curing process, a cement composite with higher resistance to loading impact can be produced^[@CR35]--[@CR37]^. The different compressive strength recorded for different curing methods is also attributed to the addition of CNCs into the matrices. CNCs are known as cellulose-based nanofibers which are built up of hydrophilic nanoporous structure. The expansion and shrinkage of this structure during and after the curing period might have led to microcracking, and consequently, a lower strength than the designed strength. With an obvious lower strength recorded for water and lime curing methods, it becomes clear that cement mortar containing cellulose-based materials such as CNCs ought to be cured using unconventional methods. Zheng *et al*.^[@CR32]^ and Mariano *et al*.^[@CR33]^. have also stated this, where the authors affirmed that the hydrophilic nature of CNCs which causes the structure to absorb and store water is profound and should not be ignored. As such, its curing process should be investigated and studied in-depth. Mechanical properties {#Sec12} --------------------- ### Effect of CNCs on cement mortar strength {#Sec13} The compressive strength performance of cement mortar with CNCs cured under wrapping technique is graphically represented in Fig. [4](#Fig4){ref-type="fig"}. With the addition of 0.4% CNCs, the compressive strength increased between 43% and 46% compared to the control sample after 28 days of curing period. Further addition of CNCs did not show a significant increase in strength. For instance, the addition of 0.6% and 0.8%, CNCs showed the compressive strength up to 16% and 38%, respectively. Comparison of previous reports in Table [2](#Tab2){ref-type="table"} shows that CNCs have a high potential in improving the strength of cement composites after its addition. In this study, the optimum CNCs content was found at 0.4%. This was due to the reason that the CNCs had changed the structural properties of the cement mortar. An increase in strength could mean that the CNCs had triggered a good production of C-S-H gel and Ca(OH)~2~ during the hydration process, which is an important component for strength development other than Ca(OH)~2~^[@CR12]^. The function of these hydration products is to bind the cement particles together^[@CR38]^. Though this may be interpreted as the more the binder, the stronger the structure could be, it is apparent that a CNCs content of more than 0.4% does not cause any changes in strength compared to the conventional mortar. This is because the high amount of hydration products formed could have made the structure more brittle and thus, more prone to cracking^[@CR39]^.Figure 4Compressive strength of cement mortar incorporating different percentage of CNCs concentrations under wrap curing condition.Table 2Comparison of previous reports on the effect of CNCs addition in to mortar.ParametersFindingsLiteratureReflectionExtraction method-Cellulose source: Palm oil empty fruit bunch\[1\] Fortunati *et al*., (2013)^[@CR52]^-Cellulose source: Okra bahmia-Process: Acid hydrolysis-Condition: 64 wt/wt % H~2~SO~4~ at 45 °C for 30 minDifferent raw sources of cellulose required different acid hydrolysis condition to extract the CNCs-Process: Acid hydrolysis\[2\] Nuruddin *et al*., (2014)^[@CR53]^-Cellulose source: Kenaf fiber-Process: Acid hydrolysis-Condition :60% H~2~SO~4~ (v/v) at 50 °C for 60 min-Condition: 64% w/v H~2~SO~4~ at 45 °C for 60 min\[3\] Darpentigny *et al*., (2019)^[@CR54]^-Cellulose source: Tunicate-Process: Acid hydrolysis-Condition: 50 wt% H~2~SO~4~ at50 °C for 20 hoursCompressive strengthImproved by 43% to 46% compared to conventional mortar\[1\] Barnat-Hunek (2019)^[@CR55]^ found that with 1.5% addition of CNCs improved 27.6% of compressive strengthThe addition of CNC in cement composites improved the compressive strength performance of more than 20% of the cement composite's original strength. Compressive strength of cement composites\[2\] Aloulou *et al*., (2019)^[@CR56]^ proved that compressive strength improved by 50% with the addition of 1% nano wood fiber.\[3\] As mentioned by Jiao *et al*., (2016)^[@CR57]^, 0.15% addition of CNCs will improve the compressive strength of cement by 20%Flexural strengthImproved by 20% compared to conventional mortar\[1\] Barnat-Hunek (2019)^[@CR55]^ mentioned that with the addition of 1.5%CNCs improved flexural strength of mortar by 10.9%CNCs have a high potential in improving cement composite flexural strength more than\[2\] Fu *et al*., (2017)^[@CR58]^ reported that flexural strength improved by 20% with the addition of 0.2% of CNCs10% of normal cement composites strength.\[3\] Jiao *et al*., (2016)^[@CR57]^ presented the data of flexural strength improvement by 15% after the addition of 0.15% CNCs in cement paste.\[4\] Cao *et al*., (2013)^[@CR29]^ found that with 0.2% CNCs improve flexural strength by 30% With this, morphological studies were conducted to further understand the effect of adding CNCs into cement mortar using FESEM and SEM/EDX plus XRD characterization. Figure [5](#Fig5){ref-type="fig"} shows the FESEM results between conventional mortar (Fig. [5a](#Fig5){ref-type="fig"}) and mortar added with CNCs (Fig. [5b](#Fig5){ref-type="fig"}). Apparently, the formation of hydration products was more obvious in CNCs-mortar. The C-S-H gel and Ca(OH)~2~ compounds had also filled in the pores, making the mortar structure denser and more compacted; the CNCs had not only functioned as nuclei to induce nucleation of hydration products but also a bio base nano-filler^[@CR40]^. The second function has been extensively discussed by Maraino *et al*.^[@CR41]^, where it is stated that its unique and favorable nano-characteristics include large surface area and aspect ratio.Figure 5FESEM image of calcium crystals (C-S-H and Ca(OH)~2~) filled in mortar specimen (**a**) control (**b**) 0.2% CNCs. The XRD data were used to further confirm the existence of hydration products in addition to FESEM images (see Fig. [6](#Fig6){ref-type="fig"}). The highest peaks at 18, 34, 47 and 51 denoted the existence of calcium hydroxide, Ca(OH)~2~, also known by its mineral name portlandite, was found higher in the mortar-CNCs sample than the control^[@CR42]^. Calcium hydroxide forms as crystals with a wide range of shapes and sizes, which can completely engulf a small cement particle next to it^[@CR43]^. A significant proportion of the calcium hydroxide formed as an intimate mixture with the C-S-H gel. C-S-H gel tends to be much smaller in which their growth could have impeded by the surrounding solid^[@CR44]^. Therefore, it's hard to distinguish since the latter contains a significant proportion of Ca-OH bonds. Since, the strength of cement mortar depends on the production of C-S-H gel, more formation of C-S-H gel, results in better strength of cement mortar. However, from the XRD test the C-S-H formation is hard to measure correctly due to its amorphous behavior and poorly crystalline. Therefore, the crystallinity of portlandite (Ca(OH)~2~) was measure instead. This is because. the portlandite is a secondary product of hydration (Eq. [1](#Equ1){ref-type=""}) and its highly crystalline^[@CR45]^. C-S-H size has nanometer level morphology and the observation of peak in Fig. [6](#Fig6){ref-type="fig"} confirms the presence of portlandite formation as a binding agent which in turn contributes to the higher mechanical strength of cement mortar.Figure 6XRD pattern of control sample and 0.2% CNCs-mortar at 28 days. P: Portlandite (Ca(OH)~2~), CS: Calcium-Silicate-Hydrate (C-S-H), Et: Ettringite. Since the CNCs added in the cement mortar was in aqueous suspension form, the zeta potential of the solution was also determined to affirm its dispersion characteristic during the mixing process as well as its effects, such as whether the CNCs particles had bonded with itself or cement particles. Table [3](#Tab3){ref-type="table"} shows the zeta potential value of cement and CNCs. The zeta potential of the CNCs aqueous suspension (−50.4 mV), CNCs powder form (−36.1 mV) was much higher than cement particles, which means that it does not agglomerate easily by itself due to the high degree of electrostatic repulsion. This characteristic has arisen due to the earlier acid hydrolysis process which has promoted the formation of sulfate group^[@CR46]^. The lower zeta potential value of −16.6 mV for cement particles indicated easy agglomeration and the significant difference of zeta potential value between CNCs and cement particles, which attract it to stick together due to make the attractive force exceed the repulsion force. The CNCs particles are more likely to stick to the cement particles since it has higher zeta potential^[@CR47]^.Table 3Different zeta potential value between cement particle and cellulose nanocrystals.Types of particleZeta Potential (mV)Cement−16.6CNCs−50.4 Figure [6](#Fig6){ref-type="fig"} reported the SEM/EDS data of the unhydrated cement particles for control mortar (Fig. [7a](#Fig7){ref-type="fig"}) and the adhering CNCs around the cement particles for CNCs-mortar (Fig. [7b](#Fig7){ref-type="fig"}). Different magnification was used to observe the dark spot around the unhydrated compound. The dark spot occurred only in the sample containing CNCs. The EDS result of the dark spot confirmed the presence of Sulphur (S) as the CNCs contained some sulfate groups (Table [1](#Tab1){ref-type="table"}) that is formed during the acid hydrolysis process. It is expected that CNCs agglomerate around the unhydrated cement particle. However, a dark spot from EDS is difficult to be determined accurately at this point because the CNCs are known to contain light chemical compounds^[@CR29]^. As preventive measures, 30 spots of EDS pointed at the dark spot and the finding is almost similar to the presence of Sulphur. In order to support the idea of the dark spot identity, the result of X-ray Fluoresces (XRF) of cement particle is given in Table [4](#Tab4){ref-type="table"}. The EDS data of unhydrated cement particles illustrated in Fig. [6a](#Fig6){ref-type="fig"} shows the absence of Sulphur content.Figure 7SEM/EDS observation to show the (**a**) unhydrated cement particle of conventional mortar (**b**) CNCs-mortar with dark spot formation around unhydrated cement particles.Table 4Chemical composition of cement by XRF analysis.ComponentPercentage by weight (%)Silica, SiO~2~19.8Alumina, Al~2~O~3~5.6Iron oxide, Fe~2~O~3~3.4Calcium oxide, CaO62.7Magnesium oxide, MgO1.2Sodium oxide, Na~2~O0.02Phosphorus pentoxide, P~2~O~5~0.1Loss of Ignition, LOI2.1Lime saturated factor1.0 The black or darker spot around the cement particle was found to be the CNCs nanofiber sticking to the cement particles. The dark spot was another reason behind the high formation of calcium crystals when CNCs were added where its hydrophilic behavior had caused absorption of water into the porous structure. The inference made here is that the absorbed water had been the major water supply for the unhydrated cement particles to go through the hydration process and form extra calcium crystals even after the curing period. A similar finding has been described by Cao *et al*.^[@CR12]^, where the identical dark ring was found around the cement particle in the cement matrix, which was later identified as concentrated CNCs lingering around the cement particles. The finding shows that CNCs can potentially provide continuous strength development in aging cement mortar^[@CR48]^. ### Flexural strength performance effected by the addition of CNCs {#Sec14} The performance of flexural strength shows a substantial improvement after the addition of CNCs into the mortar matrix (see Fig. [8](#Fig8){ref-type="fig"}). Based on that, the optimum limit of CNCs was 0.4% where an increase in flexural strength of up to 20% stronger than conventional mortar was recorded. However, further addition of CNCs in mortar had resulted in a reduction of 7% to 10% in flexural strength.Figure 8The effect of cement composites containing CNCs as nano reinforced agent. The optimum CNCs content for flexural strength was different from the results of compressive strength because of the CNCs could function better in resisting flexural failure when subjected to axial loading^[@CR42]^. This attribute has yet to be discussed extensively in most of the studies - though the same behavior has been found in many works using carbon nanotubes as the flexural strength booster. These two materials share almost the same nanofiber shape, and thus, theoretically, both can act as the bridging agent or reinforcement material^[@CR8]^. Parveen *et al*.,^[@CR49]^ reported that nanomaterials can be effective in preventing the initiation of crack growth in nanoscale level. Without CNCs, cement composites are vulnerable to freeze and thaw damage, chloride penetration, corrosion and alkali-silica reaction, which degrade the overall performance of composites^[@CR50]^. This occurs when the interfacial interaction between the nanofiber and the bonding agents of the hydration products (C-S-H and Ca(OH)~2~) has enhanced the load transfer between the matrix and reinforcement. Similar findings reported by Nochaiya and Chaipanich^[@CR51]^, shows that the addition of 0.5% carbon nanotube cause the pore reduction left between C-S-H and Ca(OH)~2~ due to the presence of carbon nanotubes as a filler and bridging agents of hydrated cement. Conclusion {#Sec15} ========== The present work concludes the significant impact of CNCs on cement mortar which the CNCs have affected the curing performance of the cement mortar. Wrap curing with polyethylene film was the most effective method, evident through the highest compressive strength recorded. Moreover, the compressive strength of the cement mortar increased 43% to 46% from its original strength when 0.4% of CNCs were added. In addition, the effect of CNCs addition also positively affected the flexural strength of cement mortar that increased about 20% when up to 0.4% CNCs were added. Finally, the addition of CNCs found to change the inner structure of the composites where the formation of calcium crystals (C-S-H or Ca(OH)~2~) that would strengthen the structure continued even after the curing process ended. Further studies on the other potentials of CNCs in improving cement mortar should be examined thoroughly due to its promising characteristics and the possibility of being applied in real-world construction projects. Novelty statement {#Sec16} ----------------- This study is reported to be the novel study on examining the effect of cellulose nanocrystals (CNCs) derived from palm oil empty fruit bunch (EFB) fiber into cement mortar as a green admixture to increase its structural performance. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors wish to acknowledge the support of the School of Civil Engineering, Universiti Teknologi Malaysia and School of Creative Science and Engineering, Waseda University. This work was done under the financial support of AUN/SEED-Net and JICA Collaborative Research (CR) grant under vot R.J130000.7317.4B194, Collaborative Research Program for Common Regional Issues (CRC) grant under vot R.J130000.7317.4B189, PDRU Grant- Vot No. Q.J130000.21A2.04E53, Hitachi Scholarship Program 2019, MRUN R.J130000.7805.4L886, and UTM Matching Grant (Q.J130000.3017.00M90). Diana Mazlan-Experimentation, Data curation, and Writing; Santhana Krishnan- Revised writing, Investigation, and Conceptualization; Mohd Fadhil Md Din --Supervision; Chiharu Tokoro- Supervsion; Nur Hafizah Abd Khalid- Sofware; Izni Syahrizal Ibrahim- Software; Hideki Takahashi- Proof read; Daisuke Komori- Resources. The authors declare no competing interests.

Introduction ============ Caridean shrimps (infraorder Caridea) belonging to the genus *Lysmata*, Risso 1816, are heavily targeted in the marine ornamental trade ([@ref-20]; [@ref-8]; [@ref-36]) and have experienced increases in market demand ([@ref-35]). Many marine ornamentals provide needed ecosystem services in home aquariums, such as algal--grazing and scavenging and are a biological alternative to other mechanical and chemical methods for aquarium maintenance ([@ref-21]; [@ref-35]). In the United States, Florida is the center of the ornamental fishery ([@ref-30]), where *Lysmata boggessi* and *Lysmata wurdemanni* are collected as bycatch between October and May from commercial stone crab traps and are landed year-round in the commercial bait shrimp trawl fishery ([@ref-8]; [@ref-33]). These shrimp are desired among aquarists specifically for their ability to regulate the pest anemone *Aiptasia* spp., Gosse 1858 ([@ref-34]), which is often introduced to aquaria via live rock. Previous studies suggest that their recent increase in popularity is in part driven by this biological control ([@ref-35]). The overall increase in landings of marine invertebrates that provide ecosystem services has raised concerns regarding the impact of harvest on wild stocks and their surrounding environment ([@ref-1]; [@ref-19]; [@ref-35]; [@ref-8]). For instance, [@ref-19] found that many ornamental fisheries operated on small spatial scales and that harvest could result in a nearly 50% localized reduction in population size. Unfortunately, there are severe data gaps in the life history, reproductive biology, population structure and growth characteristics for many of these organisms ([@ref-21]; [@ref-35]), which is especially true for ornamental crustaceans and *Lysmata* spp. The alarming result is that ornamental fisheries in Florida have not been managed using fisheries data and no management strategies currently inform the catch targets (i.e., maximum sustainable yield, total allowable catch) necessary to ensure stock sustainability ([@ref-19]). If these data gaps were to lead to fisheries mismanagement and the overexploitation of ecosystem service providers such as *Lysmata*, then an unintended ecological consequence would be the loss of those same services from the wild ([@ref-35]; [@ref-8]). In addition to their fishery and ecosystem value, shrimp from the genus *Lysmata* are notable for their rare sexual system, protandric simultaneous hermaphroditism (PSH). Individuals displaying PSH settle and mature initially as males and over a series of transitional molts develop functional female gonads to become simultaneous hermaphrodites ([@ref-25]; [@ref-39]). The latter sex phase, characterized by the presence of ovotestes and gonopores ([@ref-16]) is capable of functioning as either sex but cannot self-fertilize ([@ref-13]). Phylogenetic studies have determined that PSH is a fixed and conserved trait within the genus ([@ref-4], [@ref-5]; [@ref-11]), which is thought to be advantageous for increasing mating opportunities in low density populations and fitness specifically for large male phase shrimp ([@ref-3], [@ref-6]; [@ref-31]). Interestingly, the size at which *L. wurdemanni* males transition to hermaphrodite varies with season and appears flexible, which is suspected to be a strategy to decrease reproductive output when conditions are unfavorable ([@ref-7]; [@ref-12]). If true across the genus, this adaptive capability may be an important consideration in the management of other harvested populations. [@ref-9] produced a snapshot of the life history parameters of a Florida *L. boggessi* population, but they did not capture the important detailed seasonality to these parameters. We implemented a similar methodology as used by [@ref-9] and [@ref-12] to assess life history and reproductive characteristics, but focused on a fished population of *L. boggessi* on the west Florida shelf. These shrimp are found in nearshore waters and are landed year round as bycatch in the commercial bait shrimp (*Farfantepenaeus duorarum*) industry. The primary objective was to describe the seasonality of sex phase ratio and size at sex change at the population level and fecundity, embryo volume and reproductive investment at the individual level. Materials and Methods ===================== Study site, sampling protocol and shrimp measurements ----------------------------------------------------- We sampled a segment of the *L. boggessi* population on the Florida west coast using fisheries-dependent techniques for a full year. This genetically homogeneous population ranges from Key West to approximately Cedar Key, Florida ([@ref-8]). Shrimp were collected by a trained observer at night and as bycatch once per month from December 2012 to November 2013 via roller-frame trawlers ([@ref-18]) in a shallow subtidal region off the west coast of Florida. This area, locally referred to as The Reef, is located adjacent to the St. Martin's Aquatic Preserve, 6--8 km offshore of Citrus County, Florida ([Fig. 1](#fig-1){ref-type="fig"}). The depth ranged from 2 to 5 m and the benthic substrate was composed of heterogeneous coarse sand, seagrass and low relief hard bottom areas. Seagrass areas were dominated by *Thalassia testudinum* and *Syringodium filiforme* and the hard bottom was characterized by exposed limestone covered with a thin (\<2 cm) layer of sediment. *L. boggessi* shelter diurnally within crevices found in hard bottom ([@ref-9]), but their cryptic and nocturnal nature poses a logistical challenge for collecting specimens. We were able to circumvent this limitation by using commercial trawlers, which provided sufficient catch efficiency to achieve the sample sizes required for a statistically robust population assessment. Nocturnal segregation of shrimps by size or ontogenetic phase was not suspected in *L. boggessi* based on previous population studies of *L. wurdemanni* ([@ref-12]). Therefore, we are confident that the haphazard trawler coverages collected representative samples of *L. boggessi* on The Reef. {#fig-1} During each sampling event the trawl vessel simultaneously deployed two roller-frame trawls (4.27 m height × 0.61 m width), one on the port and one on the starboard side, 8--10 times between sunset and 02:30 h. Trawl durations were 30--45 min. A mesh size of 25.4 mm was used near the mouth of the trawl net and throughout the tapered body, with a finer mesh catch bag (19.1 mm) woven into the tailing end. The vessel tracks, average speeds and tow times were recorded with a hand-held Garmin^™^ GPS Map 60CSX. *L. boggessi* landed during each deployment were counted and approximately 30 individuals on alternating deployments were haphazardly sub-sampled for further measurements back at the laboratory. The sub-sampled shrimp were fixed onboard the fishing vessel in 10% neutral buffered formalin and were transferred after 48 h to 70% ethanol for preservation. Preserved specimens were viewed under a Leica^®^ Model S8AP0 dissecting stereomicroscope to measure body size, determine sex phase and assess embryonic development in gravid hermaphrodites. Carapace length (CL) was measured using Leica^®^ Application Suite V4 image analysis software to 0.1 mm from the mid-dorsal posterior margin of the carapace to the posterior edge of the eye orbital. Sex phase was determined by observing the second pair of pleopods for either the presence or absence of the appendices masculinae, where presence indicated male phase and absence indicated hermaphroditic phase ([@ref-16]). Hermaphrodites were considered gravid if they retained an embryonic mass on the ventral side of their abdomen. Masses were delicately removed with forceps and the embryos were then counted at 10× power under the stereomicroscope to determine batch fecundity. We were not able to confidently categorize embryo maturation based on eyespot and yolk sac development alone, so we used only early embryo masses void of either feature for this analysis. Therefore, fecundity estimates were not corrected for embryo loss. To calculate embryo volume, 10 embryos were haphazardly sub-sampled from each removed mass and measured along their short and long axes to a precision of 0.001 mm. Lastly, hermaphrodites and their corresponding embryo mass were dried at 60 °C for 24 h and weighed separately to approximate reproductive output. Population-level life history parameters in *Lysmata boggessi* -------------------------------------------------------------- Temporal differences in sex phase ratio, in terms of the proportion of male phase shrimp and size at sex change (CL~50~) were assessed from the monthly sub-samples. Sex phase ratio was estimated as the number of male phase individuals divided by the total number of males plus hermaphrodites in the population during each month ([@ref-10]). CL~50~ was calculated via binomial logistic regression with 95% confidence interval bounds. The CL~50~ estimates represented the CL at which *L*. *boggessi* would have a 50% probability of being either male phase or hermaphroditic, with the likelihood of hermaphroditism increasing as CL exceeded the CL~50~ point estimate ([@ref-9]). Individual-level reproductive characteristics in *Lysmata boggessi* ------------------------------------------------------------------- For hermaphrodites with early stage embryo masses, fecundity was determined as the number of embryos per individual. Reproductive investment was calculated as the quotient dry embryo mass divided by its respective dry hermaphrodite mass and represents the resource allocation of an individual towards reproduction ([@ref-9]). For embryos that were removed for measurement, their volume was calculated with the formula ([@ref-41]): $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$${\rm EV }= \frac{1}{6} ({LS}^{2}{\rm \pi})$$\end{document}$$where *L* denotes the long axis measurement and *S* the short axis. The mean volume estimate for an individual hermaphrodite was the average of the 10 sub-sampled embryos. For the following seasonal analyses, the categorization of months into seasons was based on local temperature readings and photoperiod oscillations representative of the study area ([Fig. 2](#fig-2){ref-type="fig"}). Hereafter, December--February is referred to as winter, March--May as spring, June--August as summer and September--November as fall. {#fig-2} For fecundity and embryo volume, their relationships with CL and hermaphrodite body mass, respectively, were fit with general linear models and their slopes tested (*t*-test) to determine if they differed significantly from zero. Analyses of covariance (ANCOVA) were used to determine the effect of the primary factor (i.e., season) on reproductive output, fecundity and embryo volume. Monthly samples were pooled by season to achieve adequate sample sizes and only individuals with early stage embryos were used for the reproduction analyses. Only one covariate, either hermaphrodite body mass or CL, was controlled for in each independent analysis. The linear relationship between reproductive investment and hermaphrodite dry body mass was also tested and assessed using the allometric model *y = ax^b^*, where *b* indicates the exponential increase or decrease of reproductive allocation per unit of shrimp body mass. A log--log transformed least-squares regression of the variables reproductive output and hermaphrodite dry body mass was analyzed using a *t*-test to determine if the slope deviated from expected unity (*b* = 1) ([@ref-26]). Significance for all tests was achieved at α = 0.05. Results ======= Landings of *Lysmata boggessi* ------------------------------ Sampling commenced on 6th December 2012 and concluded on 7th November 2013. Although intra-month variability in landings could not be accounted for due to the nature of scheduling fisheries dependent sampling, monthly sampling was achieved with a median number of 31 days between each sampling event. A total of 5,131 shrimp were landed over this study period ([Fig. 3](#fig-3){ref-type="fig"}). Each month, 118 ± 10 (mean ± 1 SD) individuals were sub-sampled for additional detailed measurements. CLs ranged from 3.9 to 11.3 mm and 776 individuals were determined to be male phase. Of the remaining 638 hermaphrodites, 498 were gravid at the time of capture. {#fig-3} Population-level life history parameters in *Lysmata boggessi* -------------------------------------------------------------- Sex phase ratio varied between 0.18 in March and 0.90 in October with a mean of 0.54 ± 0.24, *n* = 12. Variation was plotted against a dashed line representing a 1:1 male: hermaphrodite ratio ([Fig. 4](#fig-4){ref-type="fig"}). The population was skewed toward males from September to January (0.65--0.90) and toward hermaphrodites from February to May (male proportions 0.18--0.33). {#fig-4} The monthly CL~50~ was highly variable and ranged from 6.20 ± 0.09 mm in March to 9.17 ± 0.12 mm in January ([Fig. 5](#fig-5){ref-type="fig"}). The proportion of ovigerous hermaphrodites ranged from 0.05 in December 2012 to 0.94 in February 2013, with the highest proportions occurring from February to May ([Fig. 6](#fig-6){ref-type="fig"}). These proportions remained consistently high (\>0.50) throughout the year, except during October, November and December, when they were less than 0.40. Generally, high proportions of ovigerous hermaphrodites coincided with low CL~50~ estimates. {#fig-5} {#fig-6} Individual-level reproductive characteristics in *Lysmata boggessi* ------------------------------------------------------------------- Individual fecundity ranged from 45 to 1,614 embryos per shrimp, 642 ± 293 (mean ± 1 SD) and was highly variable among seasons and hermaphrodite sizes. All seasonal fecundities were positively correlated with CL and hermaphrodite body mass and mean embryos reached a high in spring and a low in fall ([Table 1](#table-1){ref-type="table"}; [Figs. 7](#fig-7){ref-type="fig"} and [8](#fig-8){ref-type="fig"}). The two ANCOVA analyses used to test the effect of season on fecundity revealed strong seasonality. The covariates of CL and hermaphrodite body mass both resulted in significant interaction effects (*p*-values \< 0.001), indicating that the relationship between each covariate and fecundity differed between seasons. Posthoc Tukey's multiple comparison tests showed that the spring and summer fecundities were greater than fall, spring was greater than summer and summer was greater than winter (*p*-values \< 0.001). 10.7717/peerj.8231/table-1 ###### Seasonal fecundity estimates and regression parameters for the relationships between the dependent variable fecundity and both independent variables carapace length and hermaphrodite dry body mass.  Season *n* Mean (SD) CL parameters Body mass parameters -------- ----- ----------- ---------------------------------------------- ---------------------- ------ --------------- ----------- ------ Fall 25 472 (137) 141.0 (24.3)[\*](#table-1fn1){ref-type="fn"} −662 (196) 0.59 2,871 (946) 181 (98) 0.25 Winter 69 474 (160) 140.1 (14.8)[\*](#table-1fn1){ref-type="fn"} −779 (133) 0.57 2,613 (380) 106 (55) 0.4 Spring 158 748 (293) 207.2 (17.6)[\*](#table-1fn1){ref-type="fn"} −1,094 (157) 0.47 4,513 (4,340) 84 (66) 0.41 Summer 80 700 (337) 259.0 (16.3)[\*](#table-1fn1){ref-type="fn"} −1,490 (139) 0.76 6,713 (421) −125 (54) 0.76 **Note:** Significant at α = 0.05. {#fig-7} {#fig-8} Embryo volume ranged from 0.056 to 0.230 mm^3^, with winter having the highest mean of 0.15 ± 0.02 mm^3^, *n* = 69 and summer having the lowest mean of 0.10 ± 0.01 mm^3^, *n* = 80. Other than a slightly positive, but significant, correlation between volume and hermaphrodite body mass during winter, there were no statistically significant relationships between embryo volume and either CL or body mass ([Table 2](#table-2){ref-type="table"}). The results of the ANCOVA showed an effect of season on embryo volume (*p*-value \< 0.001), though the interaction term was not significant (*p*-value = 0.10) and therefore the interaction term was removed and the ANCOVA analyses re-run. This second test produced similar main effect outcomes. Posthoc Tukey multiple comparison tests showed that during winter hermaphrodites produced larger embryos than hermaphrodites in any other season ([Table 2](#table-2){ref-type="table"}). Spring hermaphrodites produced the second largest embryos and no significant difference in volume was found between summer and fall. 10.7717/peerj.8231/table-2 ###### Seasonal embryo volume estimates and regression parameters for the relationships between the dependent variable embryo volume and independent variables carapace length and hermaphrodite dry body mass.  Season *n* Mean (SD) CL parameters Body mass parameters -------- ----- ------------- --------------------------------------------- ---------------------- -------- ------------- ------------- ------ Fall 25 0.11 (0.01) 0.0007 (--) 0.06 (0.04) 0.03 0.23 (0.14) 0.09 (0.01) 0.58 Winter 69 0.15 (0.02) 0.0007 (--)[\*](#table-2fn1){ref-type="fn"} 0.09 (0.03) 0.04 0.22 (0.07) 0.12 (0.01) 0.57 Spring 158 0.13 (0.02) 0.0001 (--) 0.14 (0.01) 0.0003 0.02 (0.04) 0.13 (0.01) 0.47 Summer 80 0.10 (0.01) 0.0002 (--) 0.09 (0.01) 0.01 0.09 (0.04) 0.09 (0.01) 0.76 **Note:** Significant at α = 0.05. The mean reproductive investment ranged from 0.14 ± 0.04, *n* = 69, in winter to 0.20 ± 0.05, *n* = 158, in spring ([Table 3](#table-3){ref-type="table"}). In all seasons other than summer, reproductive investment increased linearly with hermaphrodite body mass, as the slope describing the relationship between these two variables did not differ from unity ([Fig. 9](#fig-9){ref-type="fig"}). For summer the slope *b* did significantly differ from unity, as reproductive investment decreased with hermaphrodite body mass (Paired *t*-test, *t*(79) = −3.30, *p*-value \< 0.001). The primary factor (season) assessed in an ANCOVA, with body mass as the covariate, was significant (*p*-value \< 0.001) and showed that there were seasonal differences in the dependent variable (reproductive investment). The interaction term was not significant and therefore was removed and the model re-run (Season: ANOVA, *F*(3,331) = 27.64, *p* \< 0.001). After it was re-run, the primary factor season was still significant. Posthoc Tukey multiple comparison tests concluded that reproductive investment in spring was significantly greater than all other seasons and that summer was greater than winter (*p* \< 0.05). 10.7717/peerj.8231/table-3 ###### Reproductive investment for hermaphrodites in each season, and regression parameters showing the relationship between hermaphrodite dry body mass and embryo mass after a log-log data transformation.  Season *n* R.O. (SD) Regression parameters Hypothesis testing -------- ----- ------------- ----------------------- -------------------- -------- ------- ---------------------------------------- Fall 25 0.15 (0.03) −0.9049 0.90 (0.18) −0.22 1.24 0.82 Winter 69 0.14 (0.04) −0.7466 1.14 (0.12) −0.64 1.68 0.52 Spring 158 0.20 (0.05) −0.6895 1.03 (0.10) −0.181 1.154 0.86 Summer 80 0.17 (0.04) −0.584 1.21 (0.08) −3.3 1.79 \<0.01[\*](#table-3fn1){ref-type="fn"} **Note:** Significant at α = 0.05. {#fig-9} Discussion ========== The trawl gear overall yielded catches representative of the *L. boggessi* population, despite their small individual sizes, due to debris (i.e., macroalgae, seagrass) reducing pore openings by lining the mesh during tows. Under stress *L. boggessi* were also presumed to seek refuge within this bycatch debris rather than escaping through the net. However, one exception was that this gear type did appear to underrepresent extremely small shrimp with a CL less than three mm. It is possible that these individuals were capable of passing through small openings, or were degraded beyond recognition by debris and biomass in the catch bag. These small shrimp would have been new recruits and likely male phase. *L. boggessi* are also gregarious and are often found cohabiting small crevices in limestone outcroppings ([@ref-9]). Therefore, there was the possibility that the population on The Reef was aggregated at scales smaller than what we were able to measure using the trawl. However, the nocturnal roaming activity of *L. boggessi* (Dickson et al., 2012, unpublished data) was presumed to shift the population more towards a normal distribution during sampling and reduce the potential clustering bias. Although sampling limitations still exist, fisheries dependent sampling via roller-frame trawls is the most effective method for sampling cryptic *L. boggessi*. Thus, this methodology was deemed adequate for our study objectives. The abundance of *L. boggessi* on The Reef was relatively low throughout the year, but peaked in the winter. Sex phase ratios were also male-dominated from fall to early winter and size at sex change decreased dramatically during spring months. Delayed sex change in male shrimp was not directly measured in this study but the similar trends in life history parameters reported here and by [@ref-12] for *L. wurdemanni* suggest that this phenomenon may also exist for *L. boggessi* on The Reef. Seasonality was also observed in fecundity, reproductive investment and embryo volume. Even though delayed sex change contradicts the general hypothesis that males should transition into hermaphrodites as soon as possible to increase fitness ([@ref-4]) this phenomenon may in fact be beneficial if it occurs for *L. boggessi* on The Reef. For instance, it might allow males to capitalize on a sex specific mating advantage, as was observed in *L. wurdemanni* ([@ref-14]), where large males are more competitive in spawning with post-paturial molt hermaphrodites than small males. However, [@ref-12] suggested that delayed sex change in temperate *L. wurdemanni* was largely driven by abiotic factors and was a response to suboptimal periods for reproduction. If delayed sex change does occur in *L. boggessi*, then this latter notion is suspected for this population on the west coast of Florida. Similar to the study performed by [@ref-12] the months where life history characteristics insinuate delayed sex change in *L. boggessi* (high proportions of male phase shrimp and CL~50~ values) coincided with cool temperatures and short photoperiods. These abiotic conditions are presumed to be suboptimal for embryo development and therefore, these males may also have transitioned later to temporarily forego the associated cost of functioning as hermaphrodites ([@ref-14], [@ref-15]; [@ref-2]; [@ref-40]). If true, then the advantage of delayed sex change is the ability for males to allocate resources toward growth and survival during suboptimal reproductive periods and then transition into large and more fecund hermaphrodites when conditions improve. It is also possible that delayed sex change for *L. boggessi* is not adaptive. For instance, assuming that *L*. *boggessi* have short life expectancies (approximately 1 year) similar to *L. wurdemanni* ([@ref-12]) the high proportion of males during fall may simply be a result of the population consisting of only a few large old individuals and a strong cohort of small young recruits. If true, then the high CL~50~ values during this cool season may be interpreted more as a threshold size at sex change for the species. Additional information on population structure, growth and lifespan is needed to determine the relationships among these factors and life history. Regardless of whether or not delayed sex change is adaptive, there is undoubtedly a strong demographic shift in *L. boggessi* on The Reef from February through May, where the majority of male individuals across all size classes transition into hermaphrodites. This observation supports the notion that PSH *L. boggessi* utilize their sex change plasticity to optimize reproduction and fitness. The peak reproductive seasons for *L. boggessi* were spring and summer, which were first evident from the high mean fecundity and reproductive investment estimates for individuals during these months ([Tables 1](#table-1){ref-type="table"} and [3](#table-3){ref-type="table"}). Although these estimates were greater than for fall and winter, they do not account for the positive correlation between individual size and fecundity and therefore may be misleading if the size structure of the population was different among seasons. However, the ANCOVA regressions ([Figs. 7](#fig-7){ref-type="fig"} and [8](#fig-8){ref-type="fig"}) did suggest that fecundity for small individuals was equal across all seasons and greater for larger individuals during spring and summer. In addition, spring and summer exhibited lower proportions of males, higher proportions of gravid hermaphrodites and lower CL~50~ estimates than compared to fall and winter. These reproductive characteristics alone do not identify these seasons as being the optimal reproductive period, but in conjunction with the previous life history parameters support this claim. The results for embryo volume (i.e., greater in winter than summer) were a surprise considering that the high winter estimate (0.15 ± 0.02) more closely resembled the volume for spring than for fall or summer. If winter is truly suboptimal for reproduction, then this finding may be explained by a common tradeoff between fecundity and embryo volume, as reported for other decapod crustaceans ([@ref-24]; [@ref-23]; [@ref-28]; [@ref-29]). For *L. boggessi* this may involve hermaphrodites investing limited resources into a few large embryos vs. numerous small ones, with the rationale that larger embryos contain more nutrients and therefore have a greater probability for individual survival. Alternatively, the winter estimate may be biased due to the categorization of seasons used in this study, considering that the life history parameters (sex ratio and CL~50~) for February appeared to be precursors for a high reproductive period. In addition, there was also a positive correlation between body size and embryo volume during winter ([Table 2](#table-2){ref-type="table"}), which could suggest that the winter estimate is an artifact of the population being skewed towards large individuals. Conclusion ========== Our findings show that the primary breeding seasons for *L. boggessi* on The Reef are spring and summer and high reproductive activity occurs from February to September. We suspect that the seasonality of life history parameters and reproductive characteristics observed here may be, in part, a function of the abiotic conditions (i.e., temperature, photoperiod) at this location. Due to the plasticity of *L. boggessi* size at sex change, seasonality may vary among *L. boggessi* across a latitudinal range, which could have implications for management. To test this hypothesis and to further fill data gaps for fisheries management, similar studies are needed for *L. boggessi* populations in other geographic areas. Future studies should also incorporate methods to assess population size distributions and cohort growth. Lastly, research addressing the influence of biotic factors on *L. boggessi* populations is limited and warrants more attention. For instance, peaks in abundance for *L. boggessi* on The Reef coincided with observed increases in the abundance of *Laurencia* spp. macroalgae. *Laurencia* is ubiquitous throughout Florida waters ([@ref-22]), is characterized by high productivity during winter months ([@ref-38]) and can serve as refugia and transport vehicle for small invertebrates across the benthos ([@ref-27]; [@ref-37]; [@ref-17]; [@ref-32]). It is unclear whether *Laurencia* increases the susceptibility to capture for *L. boggessi*, or instead facilitates their immigration from elsewhere. Understanding of the ecology of ornamental species such as *L. boggessi* is equally important as determining the life history and reproductive characteristics for ensuring sustainable harvest. Supplemental Information ======================== 10.7717/peerj.8231/supp-1 ###### ANCOVA analyses summarizing the effect of the primary factor season on reproductive output, fecundity and embryo volume. Hermaphrodite body mass and carapace length were controlled for as covariates. ###### Click here for additional data file. 10.7717/peerj.8231/supp-2 ###### Raw Lysmata Data containing trawl tow information and shrimp sample information. ###### Click here for additional data file. Special thanks to H. Patrick, P. Gardner and G. Smith for help with field and laboratory work and M. Helmholtz, R. Burnen, Frog, Eric, Aaron and J.T. Christmas from the Hernando Beach, Florida shrimp trawling community for generously allowing us to sample from onboard their trawlers. Additional Information and Declarations ======================================= The authors declare that they have no competing interests. [Michael D. Dickson](#author-1){ref-type="contrib"} conceived and designed the experiments, performed the experiments, analyzed the data, prepared figures and/or tables, authored or reviewed drafts of the paper, and approved the final draft. [Donald C. Behringer](#author-2){ref-type="contrib"} conceived and designed the experiments, authored or reviewed drafts of the paper, and approved the final draft. [J. Antonio Baeza](#author-3){ref-type="contrib"} conceived and designed the experiments, authored or reviewed drafts of the paper, and approved the final draft. The following information was supplied regarding data availability: Raw data is available in the [Supplemental Files](#supplemental-information){ref-type="supplementary-material"}.

"I soon noticed that many cases of congenital hypoplasia of the mandible occurred and that the organovegetative and psychic life of the infants was more disturbed when the hypoplasia was more pronounced. I have never seen babies live for more than sixteen or eighteen months who presented hypoplasia such that the lower maxilla was pushed more than 1 cm behind the upper.(Pierre Robin 1934)" Introduction {#Sec1} ============ In 1923 Pierre Robin described a constellation of findings that bears his name today \[[@CR1]\]. The triad of findings included micrognathia, glossoptosis and respiratory obstruction; however, considerable confusion in the medical literature delineating Robin sequence has been demonstrated \[[@CR2], [@CR3]\]. Pediatricians often encounter the entity "Robin sequence"; however, there are still many unanswered questions surrounding this disorder. Robin sequence can still be associated with significant morbidity and even mortality \[[@CR4]\]. Glossoptosis associated with airway compromise is most often the culprit instigating respiratory insufficiency (Fig. [1](#Fig1){ref-type="fig"}). However, other causes can cause breathing problems, and these patients should be carefully investigated preferably by a multidisciplinary team \[[@CR5]\]. Traditionally tracheotomy has been considered the definitive treatment in securing a stable airway when the airway was compromised. However, tracheotomy can be associated with significant morbidity and even mortality \[[@CR6], [@CR7]\].Fig. 1Typical cases of glossoptosis. Patient has a cleft of the small palate (not visible on photo). Note retrusion of mandibula with regard to maxilla Distraction of the mandible has become an accepted method to treat the micrognathia and subsequently the airway compromise \[[@CR8]--[@CR12]\]. Distraction osteogenesis (DO) is a technique in which bone is gradually lengthened after performing an osteotomy. After a short latency period, the bone segments are distracted. The bone segments are separated from each other at a slow, steady rate. Similar to fracture healing new bone will subsequently be formed between these segments. After the acquired bone length is achieved the consolidation period ensues in which the bone segments are held in their advanced positions. This is needed because the newly formed bone has to mature and consolidate. During DO the distraction proceeds at a slow, steady state ensuing not only bone lengthening but also concomitant soft tissue expansion. Subsequently will not only new bone be formed, but the muscles, blood vessels, nerves and mucosa will also be elongated. Ilizarov popularized distraction on the lower extremity in the 1940s \[[@CR13]\], although Codvilla introduced distraction nearly 100 years ago \[[@CR14]\]. Following in the footsteps of Ilizarov, mandibular distraction was first performed experimentally by Snyder \[[@CR15]\]. The first clinical report of mandibular distraction in the English literature was reported by McCarthy et al. in 1992 \[[@CR16]\]. Like Ilizarov did, mandibular distraction was performed with an external device. Since then, numerous reports have been published demonstrating the feasibility in relieving airway obstruction \[[@CR8]--[@CR12]\]. However, an external distraction system is cumbersome to take care of; it leaves external scars and always needs a second operation to remove the distraction device. In an attempt to alleviate these disadvantages, an internal and resorbable distraction device (located under the skin) was developed \[[@CR17]\]. The goal of this manuscript is to review our results of performing mandibular distraction with a resorbable system in patients with Robin sequence and life-threatening airway compromise. Methods {#Sec2} ======= For this study we looked at the patients we treated early, i.e. in the first 3 months after births. Patients were considered for distraction only after a diagnosis of Robin sequence was made (glossoptosis, micrognathia and airway compromise). The medical ethical board approved this study. Patients were seen by a multidisciplinary team consisting of a pediatrician, ENT surgeon, geneticist, dietician and plastic surgeon. Non-invasive treatment options such as prone positioning and nasal continuous positive pressure are sufficient measures for most newborn babies with Robin sequence. Only patients that could not be treated conservatively and would traditionally be considered candidates for a tracheotomy were candidates for distraction osteogenesis. Before intervention patients were observed with continuous pulse oximetry and blood gas evaluation (pCO2, HCO 3 etc). Saturation measured over 12 h in all patients was \< 90% for \> 5% of the 12 h \[[@CR10]\]. Polysomnography was only used if the aforementioned results were not comparable to the clinical picture. Patients received an endoscopy by the ENT surgeon prior to DO to exclude any other cause of airway obstruction (e.g. tracheomalacia, stenosis etc) besides the glossoptosis. The first patient treated (Table [1](#Tab1){ref-type="table"}) had already a tracheotomy, while the others were treated primarily for airway compromise. The aim in the first patient was to relieve him of his tracheostoma.Table 1Patient characteristicsNameDate of birth (day.month.year)Age at surgery (days)Amount of distraction (mm)Associated malformations (syndrome)OutcomeDuration of hospital stay (days)1. JH09.19.20068320COL 11a2 gene mutation (anocular Stickler syndrome)Admission with tracheacanule. Minor local symptoms of infection at pin site. Removal 10 months post op.182. NS10.04.20071518COL 11a2 gene mutation (anocular Stickler syndrome)Successful detubation on day 9 post op. Minor local symptoms of infection at one pin site163. SS10.03.20071916NoneSuccessful detubation on day 8 post op.114. LB11.16.20071720No mutation on Col2A1 and Col11A1 genes. No definite exclusion of Stickler because of severe myopiaSuccessful detubation on day 11 post op. Technical failure of one distraction screw 5 weeks after surgery185. LN01.17.20081318NoneSuccessful detubation on day 8 post op.236. LK03.30.20089418NoneSuccessful detubation on day 5 post op.147. RS06.26.20082720Megaencephaly and retardation, no genetic mutation foundSuccessful detubation on day 8 post op.278. LW02.08.20104522NoneSuccessful detubation on day 5 post op.169. RS06.19.201016202.19 Mb deletion in 3q22.2q22.3. Further research is ongoingSuccessful detubation on day 7 post op.1510. GH11.03.20092218NoneSuccessful detubation on day 6 post op.2011. JH07.31.20081118NoneSuccessful detubation on day 8 post op.1712. AE04.23.20102416Suspicion of Stickler due to familiar myopiaSuccessful detubation on day 8 post op.14*op.* operation All patients were treated with the Lactosorb internal distractor distributed by W. Lorenz Surgical, a Biomet company. The precise placement has been described previously by Burstein \[[@CR17]\]. Briefly, the surgical approach was a submandibular incision (2--2.5 cm) with dissection to the mandibular body and angle while preserving the mandibular branch of the facial nerve. The two dissolvable plates were placed after the vector of distraction was determined from a mandibular X-ray or a CT scan. An osteotomy was performed after the plates were fixated with soluble screws (Fig. [2](#Fig2){ref-type="fig"}). The distractor wire was subsequently placed subperiostealy and protruded the skin through an incision placed above the ear (Fig. [3](#Fig3){ref-type="fig"}). After the placement of the distractor, we waited for 36--48 h before the distraction was started. A postoperative X-ray was made. Distraction was performed at a rate of 1 mm twice daily (Figs. [4](#Fig4){ref-type="fig"} and [5](#Fig5){ref-type="fig"}). After surgery all patients were treated in the pediatric intensive care unit, until the intubation tube could be removed. On average this was performed 5--7 days after the actual distraction was initiated, i.e. when 10--14 mm of bone lengthening was achieved. Distraction was continued until the mandibular alveolus was in a normal position with regard to the maxillary alveolus or until the maximum technical length of distraction with this device (20--25 mm) was achieved (Fig. [6](#Fig6){ref-type="fig"}). After a consolidation phase of 4 weeks the distraction screw was removed in the outpatient clinic with patients receiving only paracetamol 30 min before removal of the screw. An X-ray was performed before the distraction screw was removed to demonstrate bone consolidation.Fig. 2Location of osteotomyFig. 3Placement of internal device with distractor wire visible above ear. This could easily be concealed with a baby hatFig. 4After osteotomy the mandibular is gradually lengthened with the distraction. **a** Prior to distraction. This brings the tongue forward (**b**) and alleviates the respiratory obstructionFig. 5Comparison of resorbable plate size with 2-euro coinFig. 6Example of patient before (**a**) and after (**b**) surgery. Notice the extra space in the oropharynx after the distraction and that the nasogastric tube has been removed Results {#Sec3} ======= Twelve patients with Robin sequence were included (Table [1](#Tab1){ref-type="table"}). All our patients had an associated cleft palate. Beside our first patient who already had a tracheostoma prior to distraction, a tracheotomy was prevented in all other patients. The mean age at surgery was 32 days (range 11--94 days). The average amount of distraction performed wa**s** 18 mm. All patients were extubated after an average of 7.5 days. The average length of stay in the hospital was 17 days after surgery (range 11--27 days). All patients were discharged without any nasal continuous positive pressure. Although feeding issues are not the aim of this manuscript, it should be noted that six of the patients went home without nasogastric feeding and another four patients had the nasogastric feeding discontinued before the distraction screw was removed. Our first patient treated with internal distraction could not be decannulated after the distraction process that started at the age of 3 months. The X-ray showed only about 8--10-mm distraction, despite the expected 20-mm distraction. No surgical re-exploration was performed, but we expect that an incomplete osteotomy or possible mechanical default of the apparatus was the cause. The patient was eventually decannulated at 7 months of age, and it is unknown whether the distraction influenced this in a positive way. In another patient the distraction screw fell out after 95% of the consolidation phase was completed. The patient showed no symptoms, and the technical failure did not lead to any delay or problems. Patient No. 7 in Table [1](#Tab1){ref-type="table"} developed some redness in the skin around the distraction screw but with antibiotic ointment and oral antibiotics; this resolved without complications. Discussion {#Sec4} ========== This study demonstrates that the use of an internal bioresorbable distraction system for the treatment of airway compromise in Robin sequence seems a safe procedure with no serious short-term sequelae. The treatment of patients with airway compromise and associated micrognathia and glossoptosis has been an ongoing research field for many physicians involved in pediatric care. There are numerous ways to address the airway obstruction in newborns ranging form prone positioning to nasopharyngeal airway placement and surgical intervention. Recently the " pre-epiglottic baton plate" (PEBP) has been described as another method to treat sleep apnea in infants with isolated Robin sequence \[[@CR18]\]. The aim of our manuscript was not to compare the different treatment methods but to investigate an innovative method. We have previously demonstrated that there is widespread confusion regarding the description of this disorder \[[@CR2], [@CR3]\]. Moreover, by having different descriptions of Robin sequence, it is not possible to compare various treatment options. Robin sequence affects approximately 1:8000--8500 live births. Additionally it has been demonstrated that many different syndromes could be associated with Robin sequence \[[@CR2], [@CR3]\]. Some patients have multiple congenital malformations that do not fall within diagnostic criteria for a specific syndrome. It has been demonstrated that syndromic Robin sequence patients are associated with worse outcomes regarding the severity of feeding problems and airway occlusion \[[@CR7]\]. For this study we used the definition described originally by Pierre Robin, consisting of micrognathia, glossoptosis and airway compromise. All our patients had an associated cleft palate. It is well known that most patients with Robin sequence can be treated with positional changes and nasal continuous positive pressure without surgical intervention \[[@CR4]\]. However, it is also recognized that a small subgroup needs some form of intervention to maintain an adequate airway \[[@CR4], [@CR10]--[@CR12]\]. Tracheotomies for example can be associated with significant morbidity for the patient and places a huge social burden and responsibility on the family of the patient \[[@CR6]\]. Average age at decannulation is 3.1 years, and the long-term sequelae of tracheal stenosis or tracheomalacia may be present in up to 50% to 75% of cases \[[@CR6], [@CR7], [@CR12]\]. Other complications that could be associated with tracheotomy include sudden airway obstruction from mucous plugging or accidental decannulation. Additional concerns include airway infection, airway bleeding and possible inhibition of proper speech and swallowing development. Tongue--lip adhesion was introduced in 1946 and has long been an alternative to tracheotomies. Success rates have been determined between 50 and 80% although patient characteristics were not always clearly defined in the manuscripts \[[@CR19], [@CR20]\]. Complications associated with glossopexy include a dehiscence of the adhesion and scarring of the salivary glands. Patients also need a second operation to undo the tongue--lip adhesion. The feasibility of distraction osteogenesis in the treatment of airway problems was recently assessed by a comprehensive meta-analysis performed by Ow and Cheung \[[@CR21]\]. This review retrieved 646 patients in which a bilateral distraction was performed to treat upper airway obstruction. Tracheotomy was prevented in 91.3% of neonates. However, distraction osteogenesis is still a relatively new technique and is performed with an external device in most cases \[[@CR9]--[@CR12], [@CR16], [@CR22], [@CR23]\]. External distraction leaves scars on the side of the face and always needs a second operation to remove the pins \[[@CR21]\]. The internal device is small (Fig. [6](#Fig6){ref-type="fig"}). Patients need only one operation as the material is dissolvable. The inconspicuous scar is located under the border of the mandible and above the ear. The external distraction has the added benefit that multiple vectors of distraction are possible, making it a more suitable distractor in patients with, for example, hemifacial microsomia and an absent condyle of the mandible (class II and III mandibular hypoplasia). However, the external distraction device is cumbersome and could definitely be inconvenient for parents and caretakers. For this reason patients are often admitted to the hospital for extended periods of time \[[@CR12]\]. The distractor wire of the internal device above the ear is small, and this could easily be concealed under a baby hat. In this study patients had a distraction at an early age. This obviously was done to prevent a tracheostomy. However, at this age the mandible is also small and soft, and if screws are not adequately fixed they will break out. As the child gets older, the bone will become harder and more stable with subsequent easier fixation of the distractor. A recent study has demonstrated that the long-term results of distraction osteogenesis are sustained \[[@CR11], [@CR12]\]. However, the entire process of distraction osteogenesis has multiple steps that each have potential complications and subsequently presents a unique challenge to the surgeon. Potential complications such as open bite deformities, tooth malformations or losses and possible nerve damage should be discussed before every intervention. A recent review has demonstrated that the external distraction device is often associated with the following complications: tooth injury (22.5%), hypertrophic scarring (15.6%), nerve injury (facial and inferior alveolar) (11.4%), infection (9.5%), inappropriate vector (8.8%), device failure (7.9%), fusion error (2.4%) and temporo-mandibular joint injury (0.7%) \[[@CR22]--[@CR24]\]. However, when we compare our study with the only other study population where the same internal resorbable device was used \[[@CR8], [@CR17]\], it seems that the internal device is associated with less morbidity than the external device. Although it should be mentioned that our study population is small, and long-term follow-up is needed to determine which device is superior. In our study we had one patient where an "unsuccessful" distraction was achieved. Prior to distraction she had a tracheotomy, as was custom in our hospital at that stage. Objectively we achieved only 10 mm of distraction despite the expected 20 mm. However, we were able to decannulate her at 7 months of age. Since literature demonstrates that the average age of decannulation for children with Robin sequence is 3.1 years, it is possible that the distraction did shorten her tracheotomy time \[[@CR25]\]. It is often stated that the mandible in Robin sequence always has a "catch-up" phase and that patients have a normal mandible in the long-term. However, it is demonstrated in the literature that micrognathia seldom recovers fully and that the previously reported "catch-up growth" often does not occur \[[@CR26], [@CR27]\]. Neonates with Robin sequence suffer from two main problems: airway obstruction and feeding difficulties. The main aim of this study was to determine the feasibility of this internal resorbable device to prevent tracheotomies; however, the impact distraction has on feeding was not studied and will be investigated in the future. Still we can address that the majority of our patients were dismissed without the need for a nasogastric tube and were able to be fed with a bottle and a Haberman teat feeder. Many other factors must be taken into consideration before deciding which intervention is best for the patient. In some patients with Robin sequence, mandibular distraction can permanently correct the obstructed airway, and subsequent inconvenience and costs associated with the maintenance of the tracheotomy can be avoided \[[@CR8], [@CR9], [@CR11], [@CR12]\]. It has also been demonstrated that some patients need multiple distractions and some patients will only benefit from a tracheotomy because of neurological impairment \[[@CR28]\]. Conclusion {#Sec5} ========== The internal distraction system seems safe for infants with micrognathia and has certain benefits when compared to the external distractor. A tracheotomy was prevented in all our patients, and complications were limited. Long-term studies are needed to evaluate the influence that internal distraction has on the growth of the mandible and teeth. DO : Distraction osteogenesis ENT : Ear nose and throat CT : Computed tomography Each author listed above has participated in the concept and design, analysis and interpretation of data, and drafting or revising of this manuscript; each author has approved the manuscript as submitted. An erratum to this article is available at <http://dx.doi.org/10.1007/s00784-015-1519-z>. Conflict of interest {#d30e884} ==================== The authors have no conflicts of interest to disclose. Open Access {#d30e889} =========== This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

1. Introduction {#sec1-nanomaterials-10-00919} =============== Biological apatite (non-stoichiometric hydroxyapatite) (BAp) is the main inorganic constituent in human bones and teeth \[[@B1-nanomaterials-10-00919]\]. Pure hydroxyapatite (HAp, Ca~10~(PO~4~)~6~(OH)~2~) has inorganic components of Ca and P ions; however, biological apatite is amorphous and contains several other ions, such as carbonate, Mg^2+^, SO~4~^3−^, Na^+^, CO~3~^2−^, K^+^ and Cl^−^, among others \[[@B2-nanomaterials-10-00919]\]. In addition, the 2D plate-like structure is the predominant phase in biological apatite, due to its intrinsic tendency to grow in a plate-like structure under physiological conditions \[[@B2-nanomaterials-10-00919],[@B3-nanomaterials-10-00919]\]. The fabrication of BAp with a similar chemical composition to natural bone with a controlled 2D plate-like structure is extremely difficult, especially when the mineralization occurs in a complex bioinspired condition. Designing BAp 2D nanoplates is a crucial factor for the improvement of its biological properties \[[@B2-nanomaterials-10-00919],[@B4-nanomaterials-10-00919],[@B5-nanomaterials-10-00919]\]. It has been reported that the composition, size, shape and controlled structure of bioactive bone minerals play a crucial role in determining the physical and chemical properties enabling its biomedical applications \[[@B5-nanomaterials-10-00919]\]. Although there have been many attempts to synthesize BAp minerals with nanoplate morphology, the proposed methods are replete with several problems, including the use of a toxic template and an organic solvent, and creating hazardous by-products \[[@B5-nanomaterials-10-00919],[@B6-nanomaterials-10-00919],[@B7-nanomaterials-10-00919]\], and it is difficult to obtain a 2D plate-like structure simulating the morphology of natural bone \[[@B3-nanomaterials-10-00919],[@B8-nanomaterials-10-00919],[@B9-nanomaterials-10-00919]\]. To drive homogeneous nucleation, a very high degree of supersaturation is required. This is dependent on the crystal-like arrangement and formation of ions meeting the thermodynamic criteria of the critical dimensions. Therefore, the synthesis of BAp nanocrystals using efficient green chemistry routes can be an alternative method of incorporating other biological minerals into precipitated bone minerals. Green synthesis has been intensively employed in recent studies because it is simple, biocompatible, eco-friendly and cost-effective \[[@B5-nanomaterials-10-00919],[@B10-nanomaterials-10-00919]\]. Biomimetic synthesis of its outstanding properties, including the absence of toxic chemicals, high pressure, temperature and/or energy, and its ability to easily scale-up for the large-scale synthesis of nanomaterials \[[@B9-nanomaterials-10-00919]\]. Therefore, the ecofriendly synthesis method is employed in the present study to incorporate bone minerals and control the growth rate of crystal size. Polyphenolic provenance from plants is comprehensively employed in the synthesis of inorganic nanomaterials as stabilizing, reducing and chelating factors \[[@B11-nanomaterials-10-00919],[@B12-nanomaterials-10-00919],[@B13-nanomaterials-10-00919],[@B14-nanomaterials-10-00919],[@B15-nanomaterials-10-00919],[@B16-nanomaterials-10-00919]\], and as eco-friendly alternatives to chemical and physical routes. Fenugreek (FG) is an annual plant belonging to the legume family, which has been used as one of the most promising traditional medicinal herbs. It has been widely used for medicinal purposes such as remedying pain, anti-cancer effects, anti-diabetic medication, anti-microbial effects and as an anti-oxidant reagent \[[@B17-nanomaterials-10-00919],[@B18-nanomaterials-10-00919],[@B19-nanomaterials-10-00919]\]. This is due to the pharmacological activity of the major compounds in FG seeds, such as linoleic acid, palmitic acid, pinene, 4-Pentyl-1-(4-propylcyclohexyl)-1-cyclohexene and linoleic acid methyl ester \[[@B10-nanomaterials-10-00919],[@B17-nanomaterials-10-00919],[@B20-nanomaterials-10-00919]\]. FG is also being studied for its cardiovascular benefits \[[@B18-nanomaterials-10-00919]\]. In particular interest, previous studies including our own ([Table 1](#nanomaterials-10-00919-t001){ref-type="table"}) showed that FG plants can be a good source of essential minerals, such as Mg^+2^, Zn^+2^, Ca^+2^, PO~4~^3−^, K^+^, Na^+^ and Fe^2+^ \[[@B17-nanomaterials-10-00919],[@B18-nanomaterials-10-00919]\]. Therefore, the incorporation of such ions into the synthesized bone minerals can have a favorable impact on fast bone formation. On the other hand, it is speculated that, by increasing the concentration of the hydroxyl groups in the precursor solution, the interaction between the organic molecules existing in the FG extract and the metallic ions of the calcium phosphate is increased, which means higher interference by apatite crystals during the biosynthesis process. This phenomenon might lead to denser structures being formed by FG organic molecules of the BAp minerals. A denser structure leads to smaller interplanar spacings in the obtained bone minerals. Accordingly, the crystallite size can be reduced to its smallest size to obtain apatite with 2D fine plate-like structures. Controlling the structure of apatite crystals at the nano-level is vital for acquiring a favorable commercial product. In this work, BAp minerals with a 2D plate-like morphology mimicking the natural bones' composition and structural features, fabricated by biosynthesis of the plant extraction method, have been prepared successfully. The plant polyphenolics possess high cognation for metal ions because of the existence of the hydroxyl groups of phenolic compounds and their molecular structure. As an effort to prepare BAp 2D plate-like structures mimicking the morphology of biological apatite, a biosynthesis process was developed based on the use of natural macromolecules from FG extracts. The bioactive ceramic formed in FG extraction has been called biosynthesized BAp nanoplate, because its composition and structure are similar to those of natural bone rather than of sintered stoichiometric HAp, and it has important characteristics such as low crystallinity and nanoscale sizes that are important for the reabsorption and remodeling found in bone \[[@B21-nanomaterials-10-00919],[@B22-nanomaterials-10-00919]\]. The BAp with the 2D nanostructures formed in this FG extract is believed to exhibit even higher bioactivity and biocompatibility than stoichiometric HAp \[[@B22-nanomaterials-10-00919],[@B23-nanomaterials-10-00919]\]. Furthermore, BAp ultrafine 2D plate-like structures were prepared by employing FG seed extract using biosynthesis wet-chemical precipitation as a simple, efficient, economic and non-toxic route. 2. Materials and Methods {#sec2-nanomaterials-10-00919} ======================== 2.1. Biosynthesis Process {#sec2dot1-nanomaterials-10-00919} ------------------------- In total, 20 g of commonly used FG seeds were washed several times before being boiled for 15 min \[[@B24-nanomaterials-10-00919]\] in 100 mL Milli-Q water (18.2 MΩ·cm), as typically prepared in traditional medicine. The extracted solution was filtered twice through Whatman no. 1 filter paper. The extract solution was then used for the synthesis of BAp powders by the wet-chemical precipitation route, as described in our previous report \[[@B4-nanomaterials-10-00919]\]. Briefly, calcium cations (1 M Ca(NO~3~)~2~·4H~2~O) (Sigma Aldrich, South Korea) and phosphate anions (0.6 M (NH~4~)~2~HPO~4~) (Sigma Aldrich, South Korea) were separately dissolved in 100% concentration FG extract solution. The phosphate solution was added dropwise at a rate of 0.4 mL min^−1^, under vigorous mixing, into the calcium solution. Ca/P ratio was controlled to be 1.67, the stoichiometric value of HAp. The resultant precipitate slurries were dispersed in a mixing solution of pure Milli-Q water and ethanol (volume ratio = 1:1) and then left to dry in a vacuum for 24 h. The dried powder was further heat-treated at 650 °C for 4 h, with a heating rate of 20 °C/min \[[@B4-nanomaterials-10-00919]\]. The chemical element concentration of the FG seed extract and the biosynthesized powder (0.1 mg dispersed in 5 mL Milli-Q water) were analyzed at 670.783 nm wavelength using a Varian (Inc., Melbourne, Australia) Vista Pro (MPX) radial inductively coupled plasma atomic emission spectroscopy (ICP-OES) instrument. Standards from 0 to 5 mg/L metallic ions were prepared from Fluka, TraceCERT 1000 mg/L stock standard. In addition, due to the limitation of ICP-OES in detecting SO~4~, Cl and CO~3~, the colorimetric analysis method was used to measure the concentration of these elements in both the FG extracts and obtained BAp powders. Electrical conductivity was measured by EC meter CM 40G Ver 1.09 (DKK TOA Co., Tokyo, Japan). The molecular structure of the FG seeds was analyzed using Fourier transform infrared (FTIR) spectroscopy (Prestige-21 FTIR spectrophotometer) spectra in transmission mode. The morphology of the biosynthesized powders was examined using a Scanning Electron Microscope equipped with Energy Dispersive Spectroscopy (EDS) (SEM, JEOL JSM-6400, Tokyo, Japan) and Transmission Electron Microscopy (TEM, CM 200, Philips, PA, USA). The phase composition, molecular structure and interactions of the obtained powders were analyzed using an X-ray diffractometer (XRD; Rigaku, Tokyo, Japan) and FTIR spectroscopy. 2.2. In Vitro Cell Culture {#sec2dot2-nanomaterials-10-00919} -------------------------- The cell culture was carried out by the indirect extraction method \[[@B4-nanomaterials-10-00919]\]. Briefly, the biosynthesized BAp powders were immersed in 70% (w/v) ethanol for 1h and air-dried for 48 h in a biosafety cabinet class II. Then, the apatite particles were dispersed in Dulbecco's modified Eagle's medium (αMEM) supplemented with 10% fetal bovine serum (FBS) in a humidified incubator with 95% relative humidity and 5% CO~2~ at 37 °C, under shaking, for 4 days. Subsequently, the extraction of the previous step was centrifuged for 4 min at 1000 rpm. After incubation, the suspension extracts were used for culturing cells. 10 × 10^3^ MC3T3 cells in 20 µL culture medium were carefully pipetted into the 48 well-plate containing 1 mL extraction. After incubation for 1 and 3 days, live/dead and MTT assays were carried out. The stained scaffolds were further washed two times using Phosphate Buffered Saline (PBS) and observed under a confocal microscope (Nikon ECLIPSE Ti) at 10× magnification. PI stained red-colored dead cells (Exc/Em: 561 nm/570--1000 nm) and FDA stained green colored live cells (Exc/Em: 488 nm/500--550 nm) were observed at 200 µm depth. For SEM observation, at 1 and 3 days, the samples were fixed with 3% Glutaraldehyde in 0.1M Cacodylate buffer and sequentially dehydrated in ethanol (at 25%, 40%, 50%, 80%, 90% and 100%) and treated with hexamethyldisilazane for 1 h. Before SEM observation, the samples were coated to 10 nm thickness using gold sputtering. 3. Results and Discussion {#sec3-nanomaterials-10-00919} ========================= To investigate the elements' concentration, conductivity and chemical constituents at the molecule level of the FG seed extract, ICP-OES, FTIR and solution conductivity were performed. It seems that FG seed extract has an element concentration in this order: K \> Cl \> CO~3~ \> P \> Na \> SO~4~ \> Mg \> Ca, and a trace element of Fe and Zn ([Table 1](#nanomaterials-10-00919-t001){ref-type="table"}). This is in a good agreement with the higher solution electrical conductivity of FG seed extract (1145 µS/m) than that of pure Milli-Q water as a control (93 µS/cm). This indicates that FG seed extract contains soluble metallic ions, resulting in an increased solution conductivity. Wave numbers of 569, 1050, 1250, 1330, 1397, 1545, 1651, 1746, 2956 and 3354 cm^−1^ of FG seeds were detected by the FTIR spectrum analysis ([Figure 1](#nanomaterials-10-00919-f001){ref-type="fig"}). This can be attributed to the existence of organic constituents (such as hydroxyl and carboxyl groups) in FG seeds. The FTIR's characteristic peaks in the FG seeds were similar to those in the in situ plant cell wall. The distinctive band at 3354 cm^−1^ was due to the vibration of O--H. The presence of water resulted in a weak broad intensity peak in the FG seeds. The peaks located at 2956 and 1397 cm^−1^ are a signature of C--H vibration. The three peaks at 1746, 1397 and 1330 cm^−1^ indicated the presence of the carboxylic group. The existence of -O- was resolved by combining scans showing the peak located at 1050 cm^−1^ with the antisymmetric peak at 1250 cm^−1^ and the symmetric peak at 1050 cm^−1^. The band at 1651 cm^−1^ indicated the presence of the C=C in the material. The SEM images, shown in [Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}A,B, show no significant difference between the biosynthesized bone mineral powder's distribution before and after heat sintering. Furthermore, it can be seen that a uniform distribution of nanoparticles was obtained ([Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}A,B). To have a close look, the shape and the morphology of biosynthesized BAp powder was further investigated by TEM analysis. TEM results showed that the non-sintered (dried) and sintered (treated at 650 °C) powders had a 2D plate-like nanostructure morphology with an average width (or length) of 12.5 ± 2 and thickness of 3.8 ± 1.2 nm ([Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}). However, the dried samples had a higher tendency to agglomerate ([Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}A) than the sintered samples ([Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}B). Although we have employed the exact chemical precipitation route which has been previously used in several reports, including ours \[[@B4-nanomaterials-10-00919],[@B25-nanomaterials-10-00919],[@B26-nanomaterials-10-00919]\], in the present study, we obtained 2D plate-like nanostructures with a smaller size. This is likely due to the contribution of FG extracts and their organic molecules, which controlled the bone mineral crystals' growth rate during the biosynthesis process. Formation of 2D plate-like nanostructures may be understood by the concept of driving force playing an important role in the structural changes of a material \[[@B4-nanomaterials-10-00919]\]. Driving force acting during the biosynthesis process of a material generally minimizes the surface area to maintain a smaller surface to volume ratio of the particle, resulting in the lower energy state of the obtained material, which means a more stable structure. Further discussion of the contribution of driving force and controlling the energy state of crystal growth of mineralized nanoparticles was detailed previously \[[@B4-nanomaterials-10-00919]\]. Notably, the morphology of apatite 2D plate-like nanostructures is similar to that of the apatite-like crystals in natural bones, though the plate size of the former is thinner than that of the latter \[[@B2-nanomaterials-10-00919],[@B27-nanomaterials-10-00919]\]. The elemental analysis in mg/L of the obtained bonelike apatite nanoplates is summarized in [Table 1](#nanomaterials-10-00919-t001){ref-type="table"}. From the obtained results, the obtained 2D nanoplates have most of the elements that are in natural bone. In addition, it is shown clearly that the Ca/P ratio is about 2.93, which is higher than that of stoichiometric HAp fabricated by a similar method. It is likely that the loading of other elements distracts the ratio of Ca/P in the fabricated BAp minerals. The EDS elemental analysis results of the obtained powders are shown in [Figure 2](#nanomaterials-10-00919-f002){ref-type="fig"}C,D. From EDS analysis, the Ca, P and O elements are mainly derived from the Ca--P reagents; the C element is derived from FG extract, as verified by colorimetric analysis ([Table 1](#nanomaterials-10-00919-t001){ref-type="table"}). In addition, it is obvious that the biosynthesized powders contained ion substitutions, such as Na, Mg and K. The fabricated bone mineral nanoplates showed a high content of potassium ions (822 mg/L), which were derived from FG seed extract ([Table 1](#nanomaterials-10-00919-t001){ref-type="table"}). Previous epidemiological and clinical studies have demonstrated that potassium ions have a crucial role in regulating the blood pressure of high blood pressure patients and reducing the stroke risk \[[@B28-nanomaterials-10-00919]\], reducing the demineralization of bone (osteoporosis), and reducing the formation of kidney stones \[[@B29-nanomaterials-10-00919]\]. The bone minerals prepared by the biosynthesis method produced a powder of a greenish-beige color \[[@B30-nanomaterials-10-00919]\] (inset of [Figure 3](#nanomaterials-10-00919-f003){ref-type="fig"}A,B). It is likely that the presence of Mg, K and Na compound content in the obtained BAp powder is responsible for producing the color, whereas the appearance of HAp powder obtained by free-FG extract is always white in color \[[@B2-nanomaterials-10-00919],[@B27-nanomaterials-10-00919]\]. These ions are derived from the FG extract solution used as the ion source during the reaction process ([Table 1](#nanomaterials-10-00919-t001){ref-type="table"}). FG seed extract has been reported to be generally a good source of inorganic minerals \[[@B17-nanomaterials-10-00919],[@B18-nanomaterials-10-00919]\]. The formation of 2D plate-like morphology is therefore likely attributed to the effect of Mg^+^ ions \[[@B3-nanomaterials-10-00919],[@B31-nanomaterials-10-00919]\], which are leached out of FG seeds and incorporated into the biosynthesized bone minerals. Furthermore, previous reports, including ours, have proved that these kinds of substitutions play crucial roles in the biological performance of the bone materials in comparison to pure stoichiometric Hap, and become similar to that of the natural apatite \[[@B2-nanomaterials-10-00919]\]. In particular, Mg plays an important role in the early stages of osteogenesis, stimulating osteoblast growth and differentiation \[[@B4-nanomaterials-10-00919],[@B32-nanomaterials-10-00919]\]. Sodium, available in abundance next to calcium and phosphorus, plays a significant role in bone metabolism and osteoporosis. Fe can be an interesting growth factor molecule related to selective activation of mechanosensitive ion channels using magnetic particles, as reported by Hughes et al. \[[@B33-nanomaterials-10-00919]\]. The final BAp biphasic powder was obtained as illustrated by XRD (JCPDS card: No. 09-0432) analysis ([Figure 4](#nanomaterials-10-00919-f004){ref-type="fig"}A). From X-ray diffractometry profiles, it is revealed that biosynthesis of BAp minerals using FG seed extract is a suitable method to produce bone mineral nanoplates before and after heat sintering. No noticeable diffraction peak appeared, other than that of the calcium phosphate phase structure; however, EDS showed doped trace elements in the fabricated bone minerals. This is likely due to the small amount of these elements that could not be detected by XRD. Thus, the calcium phosphate phase still has its crystal structure and there was no second phase to be detected after heat sintering. XRD was also used to characterize the crystal size. The Scherrer equation (Equation (1)) was applied to calculate the crystal size as follows:$$d = \frac{K\lambda}{B\ cos\theta}$$ where *d* is the average diameter, *K* is the shape factor, *B* is the FWHM of the diffraction peak measured in radios, *λ* is the wavelength of the X-rays and $\theta$ is the Bragg's diffraction angle. The diffraction peak at (0 0 2) plane was chosen for the calculation of crystallite size, since it is sharper and isolated from the others. It was found that the crystal size of biosynthesized 2D nanoplates was 5.75 and 5.53 nm before and after heat sintering, respectively. Such behavior is in good agreement with the FTIR analysis which was used to follow the evolution of the chemical compounds contained in the initial, and up to the final, BAp plate-like nanostructures. In particular, the FTIR spectrum of the non-sintered and sintered bone mineral nanoplates ([Figure 4](#nanomaterials-10-00919-f004){ref-type="fig"}B) shows peaks typical of carbonated apatite products, and the characteristic bands of carboxylic and carbonate (1400--1600 cm^−1^), phosphate (564, 603, and 900--1100 cm^−1^), amino (≈1405, 1600 and 3200 cm^−1^), hydroxide (630 and 3560 cm^−1^) and acetate (≈2810, 2307 cm^−1^) groups are distinguishable \[[@B27-nanomaterials-10-00919]\]. Together with the strong bands of the CO~3~^2−^ group at ≈1540 cm^−1^, this suggests that it is carbonated apatite \[[@B27-nanomaterials-10-00919]\] as expected, since the precipitation process involves organic reagents from FG extracts \[[@B25-nanomaterials-10-00919],[@B26-nanomaterials-10-00919]\]. Therefore, the phase purity of the final biosynthesized powder in terms of chemical substitutions (HPO~4~^2−^ and CO~3~^2−^ can be present as partially substituting groups of PO~4~^3−^ and/or OH^−^ in the carbonated apatite structure) have been confirmed by the FTIR analysis. The existence of carbonate ions has remarkable advantages for bone mineral crystals in terms of excellent biocompatibility and high resorbability, which make it one of the best candidate materials as a bioresorbable bone substitute \[[@B27-nanomaterials-10-00919]\]. Furthermore, the addition of carbonate in the fabricated bioactive minerals affects the charge balance and crystal structure of the BAp \[[@B1-nanomaterials-10-00919],[@B4-nanomaterials-10-00919]\]. The following mechanism of the formation of BAp 2D plate-like nanostructures can be proposed: the polyphenolic OH^−^ groups of FG extract produced a p-track conjugation effect and bound with Ca^2+^ ions to form Ca^−^ complex by the conjugation effect \[[@B17-nanomaterials-10-00919]\]. During the addition of (NH~4~)~2~HPO~4~ to the Ca(NO~3~)~2~·4H~2~O) while both of them were dissolved in the FG extract solution dropwise, the negative PO~4~^3−^ of the group had an ionic interaction with positive Ca^2+^ ions for the formation of calcium phosphate--complex. Once the concentration of the Ca^2+^ and PO~4~^3−^ reached super saturation after the increase in the ion-release amount, bone mineral nucleates assembled into the nanoplates' structure with the growth in anisotropic characteristics of a, b and c- planes of apatite crystals. The greater concentration of reactants increased the saturation, resulting in the acceleration of the nucleation of nuclei formation than that of their growth. It is widely accepted that organic molecules adjust many aspects of BAp formation, including control of the phase, shape, orientation and organization of mineral deposits in mineralized tissues. The FG seed extract possesses high flavonoids and other natural bioactive products such as lignin, saponin and vitamins \[[@B11-nanomaterials-10-00919]\]. The reduction of chloroauric acid by using the powerful reducing agents in FG seed extract acts as a better eco-friendly surfactant \[[@B18-nanomaterials-10-00919],[@B34-nanomaterials-10-00919],[@B35-nanomaterials-10-00919]\]. The carboxylic (COO^−^) group, CN and CC metabolites functional groups exist in the seed extract. The metabolites might act as a surfactant of the crystals and the flavonoids might stabilize the electrostatic stabilization of plate-like nanostructures \[[@B11-nanomaterials-10-00919],[@B19-nanomaterials-10-00919]\]. Thus, the seed extract acts as a soft template and controls the growth formation of the BAp plate-like fine structure. The metabolites' interaction can form relatively tight coverage on the apatite plate-like surface to improve the surface's ordered structure and crystallinity. Finally, the resulting product was heat-treated at a higher temperature and the impure organics should be eliminated to get higher crystalline apatite phase structure. [Figure 5](#nanomaterials-10-00919-f005){ref-type="fig"}A--C presents the live/dead cytotoxicity assay results of the viability of the cells on the in vitro cell culture. The results indicate that the sintered BAp mineral 2D plate-like samples showed good viability over 1 and 3 days of culture, consistent with previous reports \[[@B36-nanomaterials-10-00919],[@B37-nanomaterials-10-00919]\], but with no significant difference compared to the control group (one way ANOVA, *p* \> 0.05). The quantification results of the MTT proliferation assay to investigate the metabolic activity of the osteoblast cells after 1 and 3 days of culture are shown in [Figure 6](#nanomaterials-10-00919-f006){ref-type="fig"}A. There was a statistical difference in the cell growth between the control (tissue culture plate) and sintered samples. An increase in cell growth was experienced between days 1 and 3 for sintered samples. This growth may indicate the active proliferation of the cells on the mineralized samples and the areas of the samples where calcification is denser. SEM images ([Figure 6](#nanomaterials-10-00919-f006){ref-type="fig"}B,C) of the osteoblast cells showed heather morphology and higher growth on BAp mineral 2D plate-like control samples with no significant difference, *p* \> 0.05. Based on the obtained results, BAp displayed the desirable properties for developing bioceramic bone implants. 4. Conclusions {#sec4-nanomaterials-10-00919} ============== Our goal in the present study was to synthesize carbonated bonelike apatite 2D plate-like nanostructures in the presence of fenugreek seed extract aqueous solution as a simple, eco-friendly and economic method. The method was successful in producing bone mineral 2D nanoplates very similar to the chemical composition and structure of natural biological apatite. The TEM analysis showed the presence of well-dispersed 2D plate-like structures in the size range of 12.67 nm, which aligned with the existence of FG organic molecules. Therefore, as anticipated, increases in the organic molecules' concentration can manipulate the role of hydroxyl pendant groups of FG extracts as a nucleation site for apatite crystals. Hence, more of the OH^−^ group enhances the formation of bone mineral 2D nanoplates in the reaction process, and it will hinder the absorption of Ca^2+^ on hydroxyl pendant groups located in the FG seeds extract. The formation of BAp nanoplates would be more sluggish while there is more intimate interaction between the organic molecules and ionic phases during the reaction. Thus, for the best interaction between both compartments while at the same time allowing the formation of BAp 2D nanoplates within the structure at a higher concentration of hydroxyl groups. The bone mineral 2D nanoplates synthesized by this green precipitation mediated method, with high potassium content besides other useful minerals, can be used as good biomaterials for various biomedical applications. The as-synthesized BAp nanoplates can also be used as a coating material or a nanofiller for orthopedic applications and further studies will be published later. Methodology and writing---original draft preparation, A.A.-h.; writing---review and editing, K.A.K.; project administration, H.F.; funding acquisition, B.A.A. All authors have read and agreed to the published version of the manuscript. This project was funded by the National Plan for Science, Technology and Innovation (MAARIFAH), King Abdulaziz City for Science and Technology, Kingdom of Saudi Arabia, Award Number (15-BIO4042-02). The authors declare no conflict of interest. {#nanomaterials-10-00919-f001} {#nanomaterials-10-00919-f002} {#nanomaterials-10-00919-f003} {#nanomaterials-10-00919-f004} {#nanomaterials-10-00919-f005} {#nanomaterials-10-00919-f006} nanomaterials-10-00919-t001_Table 1 ###### Chemical analysis of fenugreek seed extract and biosynthesized BAp powders in mg/L. Samples Ca Fe K Mg Na P Zn SO~4~ Cl CO~3~ ------------------- ------ ------ ----- ---- ------ ------ ------ ------- ------ ------- FG seeds extracts 17 0.39 751 20 32.2 35 0.14 32 83.7 56 BAp minerals 2978 0.45 822 32 39.7 1014 0.24 38 78.5 169 Samples were analysed by ICP-OES, whereas, other results were analysed calorimetrically for SO~4~, Cl and CO~3.~

Introduction {#section1-0963689719854363} ============ Urethral strictures represent a pathologic narrowing of the urethral lumen. This pathology affects mainly men and is manifested by lower urinary tract symptoms: poor stream, hesitancy, terminal dribbling, incomplete voiding, etc^[@bibr1-0963689719854363]^. Female urethral strictures are rare. Urinary tract infections, overactive bladder, and stress incontinence are common conditions affecting women´s urinary tracts^[@bibr2-0963689719854363]^. Management of the female urethral strictures involves urethral dilatation and urethral reconstruction using vaginal flaps, vaginal grafts, and oral mucosal grafts^[@bibr3-0963689719854363]^. In men, iatrogenic injury is the most common cause of urethral stricture followed by idiopathy, trauma, and inflammation. Systemic diseases (e.g., lichen sclerosus) can also lead to urethral strictures. Scarring of the urethral tissue is the causative process leading to the replacement of the vascular tissue of the corpus spongiosum, which leads to ischemic spongiofibrosis of the urethra. Urethral stricture is often manifested by other complications that can be presented as recurrent or chronic infection, the formation of bladder calculi, fistulas, development of sepsis, or renal failure. Therefore, urethral strictures present a serious health condition that significantly impairs quality of life and may lead to the failure of vital organs if left untreated^[@bibr4-0963689719854363]^. It is estimated that the incidence of the urethral strictures is approximately 1% in males over the age of 55. However, the real incidence of male urethral stricture disease is unknown and strongly depends on certain populations, geography, and income^[@bibr5-0963689719854363]^. The choice of surgical technique depends on the location and length of the stricture. Approximately 50% of urethral strictures are located in the bulbar urethra, 30% in the penile urethra, with the rest in a combination of both. There is a significant interest in optimizing surgical techniques to avoid further interventions and obtain a satisfactory high long-term success rate. Urethral dilatation and direct visualization internal urethrotomy are the preferred techniques for urethral stricture management. However, the management of long urethral strictures remains challenging, as outcomes seem to be poor^[@bibr6-0963689719854363],[@bibr7-0963689719854363]^. The success rate of urethroplasty techniques is between 80--90%. The outcomes of anastomotic and graft substitution urethroplasties showed a satisfactory long-term success rate. In pioneering works, different tissues such as the buccal mucosa and free-skin grafts have been applied to treat these pathological conditions. However, both are accessible in limited quantities and their utilization is accompanied by donor site morbidity and various post-surgery complications, so alternative treatment procedures are needed to improve long-term outcomes^[@bibr8-0963689719854363],[@bibr9-0963689719854363]^. Tissue engineering (TE) offers several promising approaches that may help to overcome these problems. TE, in the context of the urethral repair, uses different cell types alone or in combination with different functional biomaterials and specific growth factors to engineer functional urethral tissue suitable for transplantation purposes. Choosing the right cell type together with a supporting matrix (scaffold) is a crucial step in urethral TE^[@bibr10-0963689719854363][@bibr11-0963689719854363]--[@bibr12-0963689719854363]^. So far, there have been only a few clinical studies describing the use of cell-seeded templates in urethral reconstruction. Cells isolated from the buccal mucosa and bladder tissue were applied^[@bibr13-0963689719854363]^. According to our search criteria, only one article describing clinical application on in vitro cultivated cells was suitable for review. The main goal of this article is to focus on the various types of cells involved in urethral reconstruction using approach of TE. Materials and Methods {#section2-0963689719854363} ===================== Literature search methodology {#section3-0963689719854363} ----------------------------- A search was performed (3 January 2019) of the PubMed/Medline databases. Keywords related to TE were combined with synonyms for the urethra, urethral tissue, urothelium, smooth muscle cells (SMCs), stem cells, and urethral TE. The search was restricted to the last 10 years, the English language, and studies performed on humans or animals. A Prisma Flow Diagram represents the outline of the literature search ([Figure 1](#fig1-0963689719854363){ref-type="fig"}). {#fig1-0963689719854363} Results {#section4-0963689719854363} ======= Urine-derived stem sells {#section5-0963689719854363} ------------------------ Urine-derived stem cells (UDSCs) were the point of interest in the following studies. Either human or animal urine samples were the source of these cells. Urethral catheterization, spontaneously voided urine, bladder irrigation, and bladder washing were the methods used for urine harvesting. To collect the cells, urine samples were centrifuged. Cells were cultured in initiation media, which mainly consisted of a mixture of embryonic fibroblast (EFM) and keratinocyte serum free medium ([Figure 2](#fig2-0963689719854363){ref-type="fig"}). {#fig2-0963689719854363} Tayhan et al. used six fresh urine samples harvested from healthy patients via urethral catheterization. Human UDSCs and urine-derived urothelial cells were studied. Immunocytochemical analysis was performed to characterize isolated cells. Antibodies against cytokeratin 7 were used as urothelial cell markers. Antibodies against CD45 and CD90 were used to determine the presence of the mesenchymal stem cells (MSCs). Results showed that epithelial cell colonies were observed up to 2 days after initial seeding. Overall, 80--90% confluency of human UDSCs was reached within 12 days. Some of these cells were also positive for cytokeratin 7. Those that were positive for CD90 were negative for CD45. This study demonstrated the presence of both cell types in fresh urine samples^[@bibr14-0963689719854363]^. Yang et al. also focused on the characterization of UDSCs, but in this study cells were of the animal origin (rabbit). A total of 12 urine and 13 bladder wash samples were used for cell isolation. For the characterization, cell proliferation assay, flow cytometry, Western blot, and immunocytochemistry were used. A differentiation experiment was also performed and stem cells were successfully differentiated into smooth muscle, urothelial, and osteogenic cell lines^[@bibr15-0963689719854363]^. UDSCs seeded on small intestinal submucosa (SIS) were examined in two studies^[@bibr16-0963689719854363],[@bibr17-0963689719854363]^. The aim was to engineer a cell-seeded construct that could be applicable for urethral repair. In one study, modified three-dimensional (3D) porous SIS was colonized with human UDSCs, which were differentiated into urothelial and smooth muscle cell lines. The cell source was 12 voided urine samples. The culture medium for smooth muscle differentiation consisted of Dulbecco's modified Eagle's medium, EFM, platelet-derived growth factor-BB, and transforming growth factor β1 (TGF-β1). Cells were analyzed after 7 and 14 days. Cell-seeded constructs were cultured under static and dynamic conditions and also applied in vivo. To confirm urothelial and myogenic differentiation, immunohistochemical tests were performed. The results showed the multilayered mucosal structure was formed under dynamic conditions with similar features to the native urothelial tissue^[@bibr16-0963689719854363]^. In another study, autologous rabbit UDSCs were obtained from bladder irrigation solution samples. The media for urothelial and smooth muscle differentiation were the same as in the study above. Seeded UDSCs were labeled with PKH67 to establish cell differentiation. Labeled cell-seeded constructs were transplanted into rabbits to repair the ventral urethral defect. Histological analyses and retrograde urethrograms were performed at various time points. The results revealed that transplanted UDSCs could differentiate into required cell lineages and, when seeded on SIS, the urethral defect could be regenerated^[@bibr17-0963689719854363]^. Using human UDSCs to optimize their differentiation into a functional urothelium together with the emphasis on proper urothelial barrier function was the main aim in the study Wan et al. carried out^[@bibr18-0963689719854363]^. The harvested stem cells were cultured both under static and dynamic conditions for 1, 2, or 3 weeks. The following media were used for induction of the UDSCs: conditioned medium obtained from urothelial cells, induced medium supplemented with epidermal growth factor (EGF), and conditioned medium from SMCs supplemented with EGF. An orbital shaker was used for culturing under dynamic conditions. Stem cells were also seeded on a scaffold generated from SIS as well as on the Transwell system to form a multilayered uroepithelium in vitro. Successful differentiation was confirmed via reverse transcription polymerase chain reaction (RT-PCR), Western blotting, and immunofluorescent staining. Permeability assays and transmission electron microscopy were used to assess the properties of the urothelial barrier function. Required phenotypical and functional features were observed in induced UDSCs. When seeded on SIS, induced UDSCs were able to form a multilayered urothelium within 14 days. The role of the bioreactors and mechanical stimuli in TE {#section6-0963689719854363} -------------------------------------------------------- Use of the bioreactor and the influence of in vitro mechanical stimuli was tested in the following studies. Cattan et al. observed terminal urothelium differentiation in engineered tubular grafts that were subjected to dynamic culture conditions. The graft was composed of human dermal fibroblasts and human urothelial cells. No exogenous scaffolding was used. Fibroblasts were cultured for 1 month until they formed tissue sheets. These sheets were subsequently placed on a tubular support and left to mature for another 3 weeks. After this, intraluminal seeding with human urothelial cells was performed. This construct was placed in the bioreactor in which dynamic flow and hydrostatic pressure acted on the graft for 14 days. Histology, immunofluorescence, and RT-PCR were performed as well as electron microscopy and permeation studies. Terminal urothelium differentiation was determined in the results of this study^[@bibr19-0963689719854363]^. In another study, adipose-derived stem cells (ADSCs) and sorted primary epithelial cells were seeded onto a biodegradable poly-glycolic acid (PGA) scaffold and constructs were placed in the bioreactor. The PGA scaffold had a fibrous tubular structure and layered seeding technology was used to load both types of the cell onto the scaffold. Constructs in a bioreactor were subjected to a mechanical extension. Engineered grafts were applied in vivo (beagle dogs) to repair a 1 cm-long urethral defect. Results showed a positive effect of the mechanical extension on viability and differentiation capability of ADSCs. The use of mechanical stimuli together with cell sorting of primary epithelial cells resulted in the successful engineering of the two-layered epithelial-muscular urethra^[@bibr20-0963689719854363]^. The bioreactor was also used in another study in which ADSCs were seeded on a PGA mesh. The goal was to engineer a muscular tube suitable for potential urethral repair. Adipose tissue was harvested from the inguinal regions of adult dogs and the first passage of the isolated stem cells was used for seeding. When induced by 5-azacytidine, ADSCs acquired a myoblast phenotype. The cell-scaffold construct was statically cultured for 7 days and subsequently placed into the bioreactor to be subjected to the mechanical extension for another 5 weeks. The muscular tubes of the urethra were successfully engineered and the importance of the mechanical stimuli on the tissue maturation was highlighted^[@bibr21-0963689719854363]^. Seifarth et al. applied a bioreactor-based technique of mechanotransduction to achieve the bi-directional orientation of porcine bladder SMCs within a structure of the tubular biohybrids. SMCs were isolated from porcine bladders and seeded onto tubular fibrin-poly(vinylidene fluoride) scaffolds. The engineered biohybrid was mechanically stimulated using a balloon kyphoplasty catheter. The bursting pressure together with the permeability of urea and creatinine was also evaluated. The results showed the bidirectional orientation of the primary porcine SMCs in both a circumferential and axial direction^[@bibr22-0963689719854363]^. The ventral penile subcutaneous cavity was used as a bioreactor in a study carried out by Sun et al. Hypoxia-activated human umbilical cord MSCs (hUCMSCs) and pedicled rabbit muscles served as cell sources. hUCMSCs were isolated from fresh human placentas. These cells were cultured under normoxic or hypoxic conditions. Pieces of skeletal muscles were harvested from the rabbits. To prepare a construct suitable for urethral repair, hUCMSCs were mixed with muscle fragments and injected into the ventral penile subcutaneous cavity. Pre-incubation in a bioreactor lasted for 3 weeks. Afterwards, the generated urethral patch was used around the ventral urethral defect. Cells and tissues were tested using histological, immunohistochemical, microscopic, and molecular-genetic analyses. Results showed that hypoxia-activated hUCMSCs continually secreted angiogenic cytokines. The muscle-derived construct was successfully engineered as a living graft and used for urethral reconstruction^[@bibr23-0963689719854363]^. Modulation of the inflammatory response and stricture formation {#section7-0963689719854363} --------------------------------------------------------------- MSCs combined with CD34^+^ hematopoietic stem/progenitor cells were used in a study carried out by Liu et al. The aim was to determine whether a graft composed of these cells, which were seeded on poly(1,8-octanediol-co-citrate) (POC), could influence the inflammatory response and enhance wound regeneration when applied in vivo. In this rat-model study, polymerized scaffolds were cut into the desired parameters (2 x 8 x 0.15 mm) and subsequently seeded with bone marrow-derived MSCs and CD14^+^ hematopoietic stem/ progenitor cells. These grafts were implanted into nude athymic male rats and an unseeded POC scaffold was used in a control group. Histological, immunohistochemical, and microscopic analyses were performed. Postoperatively, the proliferation of MSCs and CD34^+^ hematopoietic stem/progenitor cells in the area of the substitution was demonstrated by using human-specific γ-tubulin antibodies. A statistically significant decrease in pro-inflammatory markers such as tumor necrosis factor α, interleukin-1β (IL-1β), neutrophil, and macrophage markers myeloperoxidase and CD68 was detected. Results showed that implanted grafts could modulate the inflammatory response, scar formation and promote angiogenesis^[@bibr24-0963689719854363]^. The formation of a stricture after urethral surgery is a common complication that often requires re-operation. Overcoming the scarring and contraction of the urethral tissue was the main interest in the following studies. Li et al. studied whether TGF-β1 small interfering RNA (siRNA)-transfected fibroblasts could inhibit the secretion of collagen type I and thus reduce stricture formation. Both rabbit oral keratinocytes (isolated from the buccal mucosa) and fibroblasts (harvested from dermal tissue) were used in this experiment. Bladders from male rabbits were decellularized and prepared as bladder acellular matrix grafts (BAMGs). Rabbit fibroblasts were transfected with TGF-β1 siRNA and cell culture medium was collected for 1 week. The concentration of collagen type I was assessed using enzyme-linked immunosorbent assay (ELISA). The second passage of oral keratinocytes was seeded on to BAMG and another side was covered with transfected fibroblasts. Compound grafts were left to mature for 1 week and subsequently implanted into rabbits. The control groups used the following grafts: BAMGs seeded with autologous oral keratinocytes and cell-free BAMGs. Urethral calibers were assessed using retrograde urethrography. Histological, immunohistochemical testing, and scanning electron microscopy evaluated the neo-urethra features. The results showed that transfected fibroblasts could inhibit the secretion of collagen type I. In the experimental group, defected mucosa was successfully repaired without the presence of the strictures^[@bibr25-0963689719854363]^. Epithelial cell-conditioned medium (ECCM) was used to demonstrate the proliferation and migration of the stricture fibroblasts in a study carried out by Nath et al. Patients undergoing urethroplasty were the donors of the urethral mucosa, urethral stricture samples, buccal mucosa, and penile skin. The control group was represented by normal urethras. Hematoxylin and eosin (H&E) and Masson trichrome staining were used for the histological analysis of normal urethral and stricture tissue histology. Fibroblasts were obtained from human urethral stricture samples. Human penile skin, buccal, and urethral mucosa served as a source for the epithelial cells. ECCM was collected from confluent cell cultures of the penile skin, buccal, and urethral mucosa and this medium was added to the fibroblast culture. In vitro scratch assay was chosen as the preferred test and the behavior of the fibroblasts was observed for 3 days. Images were taken using a phase contrast inverted microscope. The results revealed that all three different types of ECCM had an inhibitory potential on stricture fibroblast proliferation and migration in the scratch area. An explanation of this effect could be that specific soluble molecules might have been present in the ECCM^[@bibr26-0963689719854363]^. Vascularization of the engineered tissues {#section8-0963689719854363} ----------------------------------------- Sufficient vascularization of the engineered tissue is another challenge in regenerative medicine. A cell-based genetic strategy was developed to improve the vascularization of the urethral grafts in a study carried out by Guan et al. The goal was to modify urothelial cells with human vascular EGF (VEGF) via retrovirus transduction. In a control group, urothelial cells were modified with green fluorescent protein (GFP). Eight whole bladders from New Zealand rabbits were used to isolate epithelial cells. Cells from the third passage were used for viral transduction. Transduction efficiency was evaluated 4 days after the procedure. Transgene expression was examined by immunofluorescence and RT-PCR. Western blot and ELISA analyses were also performed. Cells were seeded onto rabbit carotid arteries. Grafts were analyzed by H&E staining or Weigert's elastic staining. The porosity of the material was assessed by scanning electron microscopy. Modified cells were statically seeded onto scaffolds and left to culture for 1 week. Then 1 cm-long grafts were subsequently implanted subcutaneously into nude mice. Implants were left in vivo for 1 month and analyzed. Results showed that VEGF-modified cells enhanced the neovascularization of the urethral wall and the formation of the urothelium. Moreover, secretion of VEGF appeared to be in a time-dependent manner^[@bibr27-0963689719854363]^. Native collagen and cross-linked collagen membranes were seeded with a tri-culture of primary buccal epithelial cells, fibroblasts, and microvascular endothelial cells to engineer a pre-vascularized buccal mucosa equivalent. Human gingiva served as a source for epithelial cells and fibroblasts. Microvascular endothelial cells were harvested from the human juvenile foreskin. At first, endothelial cells were seeded on to the rough site of the scaffold and left to incubate for 24 hours. Subsequently, fibroblasts were added and cells cultured for 3 days. After this, the scaffold was carefully turned upside-down and epithelial cells were seeded. Cell-seeded scaffolds were cultured for 21 days, implanted in vivo (a murine model) and then analyzed. Results demonstrated that tri-culture of human epithelial cells, fibroblasts, and endothelial cells together with a collagen membrane could form a pre-vascularized buccal mucosa equivalent. Capillary-like structures were better formed on native collagen membrane^[@bibr28-0963689719854363]^. Other approaches used in the urethral TE {#section9-0963689719854363} ---------------------------------------- In a study carried out by Jiang et al., a natural autologous tubular graft was used for long-segmental urethral reconstruction. At first, silastic tubes were implanted subcutaneously into male New Zealand white rabbits. Tubes were totally covered with granulation tissue 14 days post-implantation. Mesenchymal cells were obtained from omentum biopsies. Pancytokeratin AE1/AE3 and vimentin were used as staining antibodies to prove cells' phenotypic features. At the third passage, mesenchymal cells were used for seeding onto the outer surface of the tubular tissue (grafts were everted prior to the urethral surgery). Prepared compound grafts, with a length of 2 cm, were incubated for 1 week prior to implantation in animals. Grafts were also analyzed using transmission electron microscopy. Unseeded tubular grafts were used in the control group. Urethral calibers were assessed by retrograde urethrography. Results showed compound grafts consisted of the inner lining of mesothelium, myofibroblasts, and collagen. The urethral lumen in the control group was severely contracted, whereas in the experimental group wide urethral calibers were observed. Moreover, the newly formed urothelium replaced the mesothelium 4 weeks post-implantation^[@bibr29-0963689719854363]^. The transdifferentiation potential of epidermal keratinocytes was investigated in a study in which living skin equivalent (LSE) was considered an alternative graft source for urethral replacement. In total, 20 male chinchilla rabbits were involved. Briefly, the skin biopsy was performed and epidermal rabbit keratinocytes subsequently isolated from the samples. After the destratification and trypsinization of the cell cultures, cells were plated on LSE. A Spongostan sponge and prepared collagen gel containing postnatal human or animal (rabbit) fibroblasts were put into a Petri dish. After 24 hours of incubation, this construct formed a connective tissue equivalent. Autologous keratinocytes used for in vivo testing were labeled with a DiI membrane tracer and transfected with enhanced GFP as well. Cells were subsequently seeded onto the LSE surface. The LSE graft was implanted into the de-epithelialized urethra. Histological and immunohistochemical analyses were performed. The results revealed that keratinocytes acquired phenotypic features characteristic for the urothelium. LSE together with autologous epidermal keratinocytes could fully restore damaged urothelium^[@bibr30-0963689719854363]^. A study carried out by Li et al. investigated the feasibility of epithelial-differentiated rabbit ADSCs (Epith-rASCs) to be the proper cell source for the substitution urethroplasty. Adipose tissue donors were New Zealand rabbits. Passage three of the ADSCs was used for the experiments. Epithelial differentiation was performed in a specific microenvironment established using a 3D culture system. Epith-rASCs were used for in vivo implantation 12 days after initial induction. Epithelial and smooth muscle phenotypes were detected by proteinic and genetic analyses. To establish Epith-rASC differentiation and proliferation capacity, cells were labeled with BrdU. Rabbit BAMGs were used as scaffolds. Matrices were characterized using histological staining and scanning electron microscopy. In the experimental group, BAMGs seeded with Epith-rASCs were incubated for 1 week and subsequently implanted into rabbits. Cell-free BAMGs and BAMGs seeded with undifferentiated rADSCs formed the control groups. A ventral urethral defect with a mean length of 2 cm was repaired using mentioned grafts. Cell proliferation assay, transmission electron microscopy, retrograde urethrography, Western blot and multicolor immunofluorescence analyses were performed. Results showed that Epith-rASCs differentiated into the urothelium and prevented the contraction of the urethral lumen^[@bibr31-0963689719854363]^. Cell sheet technology was used to fabricate the bionic urethra in a study carried out by Zhou et al. The following cell types were used: ADSCs, oral mucosal epithelial cells, and oral mucosal fibroblasts. Cell sheets were prepared from mentioned cell types. For myoblast induction of adipose-derived cell sheets, 5-azacytidine was used. Cell sheets were subsequently tabularized to create a hierarchical structure. Ultra-small super-magnetic iron oxide (USPIO) nanoparticles were synthesized and used for labeling the tissue-engineered urethras. Constructed neo-urethras were subcutaneously implanted in vivo (canine model) for 3 weeks to support the neo-vascularization and mechanical strength of the construct. Afterward, a 2 cm-long urethral defect was created and prepared tissue-engineered urethras were used for the surgical repair. USPIO nanoparticles could be detected by magnetic resonance imaging at the transplant site. Engineered urethras formed a functional three-layer structure and promoted urethral regeneration^[@bibr32-0963689719854363]^. In vitro cultured cells applied in a clinical study {#section10-0963689719854363} --------------------------------------------------- Autologous in vitro cultured urothelial cells were used for hypospadias repair in a clinical study carried out by Fossum et al. Cells were harvested from bladder washings of patients with scrotal or perineal hypospadias and pronounced chordee. The second passage of the urothelial cells was used and seeded on the allogenic acellular dermis. The prepared transplant was subsequently applied in vivo (six patients). Patients were evaluated 6--8 years post-operatively: results revealed the neourethras had a satisfactory function and the cosmetic appearance was good^[@bibr33-0963689719854363]^. Challenges and future directions in urethral TE {#section11-0963689719854363} ----------------------------------------------- As MSCs represent a promising cell source for TE applications, proper culture conditions for sufficient proliferation are crucial. To maintain a typical MSC phenotype not only during the early stages of culture but also after numerous passages, new synthetic culture media together with the optimized combination of the growth and differentiation factors might be a point of interest. Focusing on stem cell reprogramming might also broaden their application. An engineered functional urothelium should fulfill several requirements that also seem to be major challenges -- the cells are interconnected by tight junctions, providing sufficient vascularization and innervation of the new tissue^[@bibr34-0963689719854363]^. Discussion {#section12-0963689719854363} ========== Urethral strictures are a common pathology of the urethra and management of long-segment strictures presents a challenging surgical problem primarily because of stricture recurrence^[@bibr3-0963689719854363]^. So far, the treatment of urethral strictures depends on the length, location, and severity of the urethral strictures. The aim of TE is to reduce recurrence of strictures, morbidity, and the use of invasive methods. TE uses several approaches, scaffolds, cells, cell sources, and techniques mainly in an experimental setting^[@bibr5-0963689719854363]^. In this article, we performed a search within PubMed/Medline to select papers focused on cells used in urethral TE. Based on results, various cell types such as hUDSCs, UDSCs, MSCs, and ADSCs and several techniques for urine cell harvesting (urethral catheterization, spontaneously voided urine, bladder irrigation, and bladder washing) have been utilized in this context ([Table 1](#table1-0963689719854363){ref-type="table"}). ###### Overview of Cells Used in Urethral Tissue Engineering.  ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Cell type Use of scaffold Aim of the study Reference ---------------------------------------------------------------------------- ------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------- --------------------------------------------- hUDSCs and urothelial cells No To harvest cells using a non-invasive method and culture them Tayhan et al.^[@bibr14-0963689719854363]^ Rabbit UDSCs No Cell characterization, differentiation into urothelial, myogenic, and osteogenic cell lines Yang H et al.^[@bibr15-0963689719854363]^ hUDSCs 3D porous small intestinal submucosa To engineer urethras using scaffold and differentiated hUDSCs Wu et al.^[@bibr16-0963689719854363]^ Autologous rabbit UDSCs Porcine small intestinal submucosa To evaluate the feasibility of the urethral repair using UDSCs Liu Y et al.^[@bibr17-0963689719854363]^ hUDSCs Porcine small intestinal submucosa To optimize differentiation methods to engineer functional urothelium with proper barrier function Wan Q et al.^[@bibr18-0963689719854363]^ Human dermal fibroblasts and urothelial cells No To evaluate the influence of mechanical stimuli on engineered tissue Cattan et al.^[@bibr19-0963689719854363]^ ADSCs, sorted primary epithelial cells Polyglycolic acid To evaluate the effect of mechanical extension stimulation and primary cell sorting on the engineered muscular urethras Fu et al.^[@bibr20-0963689719854363]^ ADSCs Polyglycolic acid mesh To engineer a muscular tube for urethroplasty Wang et al.^[@bibr21-0963689719854363]^ Primary porcine SMCs Fibrin gel with stabilizing poly(vinylidene fluoride) mesh To apply mechanical stimulation to engineer tubular structure with the bidirectional orientation of cells Seifarth et al.^[@bibr22-0963689719854363]^ Hypoxia-activated human umbilical cord MSCs, rabbit muscle cells No To engineer a pre-vascularized urethral patch Sun et al.^[@bibr23-0963689719854363]^ MSCs, CD34^+^ hematopoietic stem/progenitor stem cells POC To assess the effect on inflammatory response and wound regeneration Liu JS et al.^[@bibr24-0963689719854363]^ Autologous rabbit oral keratinocytes, TGF-β1 siRNA transfected fibroblasts Rabbit bladder acellular matrix To repair the ventral urethral defect and minimize scar formation Li C et al.^[@bibr25-0963689719854363]^ Human fibroblasts, epithelial cells No To reduce stricture formation by modulation of scar fibroblasts with cell-conditioned media Nath et al.^[@bibr26-0963689719854363]^ Modified urothelial cells with human VEGF (virus transduction) Decellularized artery matrix To improve urethral vascularization via a cell-based genetic strategy Guan et al.^[@bibr27-0963689719854363]^ Human epithelial cells, fibroblasts and microvascular endothelial cells Native and cross-linked collagen membrane To engineer pre-vascularized human buccal mucosa equivalent Heller et al.^[@bibr28-0963689719854363]^ MSCs Autologous granulation tissue tube To reconstruct urethra using MSCs seeded on autologous granulation tissue tube Jiang et al.^[@bibr29-0963689719854363]^ Rabbit epidermal keratinocytes Living skin equivalent To reconstruct urothelium using skin keratinocytes in a rabbit model Rogovaya et al.^[@bibr30-0963689719854363]^ Epithelial differentiated rabbit ADSCs Rabbit acellular bladder matrix To reconstruct damaged urethra using cell-seeded grafts Li H et al.^[@bibr31-0963689719854363]^ ADSCs, oral mucosal epithelial cells, oral mucosal fibroblasts No To engineer three-layered cell sheet labeled with USPIO for full thickness urethral repair Zhou et al.^[@bibr32-0963689719854363]^ Human autologous urothelial cells Allogenic acellular dermis To engineer neourethra for hypospadias repair -- clinical model\ Fossum et al.^[@bibr33-0963689719854363]^ ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- hUDSCs: human urine-derived stem cells; UDSCs: urine-derived stem cells; POC: poly(1,8-octanediol-co-citrate); MSCs: mesenchymal stem cells; VEGF: vascular endothelial growth factor; TGF-β1: transforming growth factor beta 1; ADSCs: adipose-derived stem cells; USPIO: ultra-small super-magnetic iron oxide; SMCs: smooth muscle cells. Based on a comparison of different tissue sources and because spontaneously voided urine, which is considered biological waste, is a non-invasive and low-cost technique, urine should be considered a reliable source of cells. These cells display characteristics comparable to MS (stromal) cells^[@bibr35-0963689719854363]^. However, articles describing the use of this source are still scarce^[@bibr7-0963689719854363],[@bibr33-0963689719854363]^. Previously mentioned studies in this review prove that urine is a suitable cell source for urethral TE and that using proper scaffolds and bioreactors can create biologically and biomechanically functional tissue that can be used for urethral TE. Because of telomerase activity, the proliferation capacity of these cells is high. Moreover, they attach well on various surfaces, which highlights their MSC origin. When differentiated into smooth muscle or urothelial cell lines, typical markers can be detected (e.g., α- smooth muscle actin, desmin, myosin, cytokeratins, and uroplakins). Within this context, one study compared the differentiation capability of UDSCs with bone marrow-derived stem cells. Results showed that for both cell types, differentiation into smooth muscle cell line was successful. However, generating urothelial cells was much more efficient in UDSCs (70:5% success rate). This phenomenon might be due to their uroepithelial origin^[@bibr36-0963689719854363],[@bibr37-0963689719854363]^. Until now, the kind of cultivation method and type of cell that should be used for urethral TE has been a matter of debate. The results from experimental studies demonstrate the importance of dynamic conditions during in vitro cultivation, which are provided by bioreactors. They positively influenced the proliferation of cells, their growth into the scaffolds, and the maturation of engineered tissues^[@bibr17-0963689719854363],[@bibr19-0963689719854363],[@bibr20-0963689719854363][@bibr21-0963689719854363][@bibr22-0963689719854363]--[@bibr23-0963689719854363]^. Moreover, several interesting techniques were introduced to reduce scarring and scar formation (TGF-β1 siRNA-transfected fibroblasts, modulation of scar fibroblasts with cell-conditioned media), which could be used alone or in combination with other techniques (e.g., using modified urothelial cells with human VEGF by virus transduction) ^[@bibr25-0963689719854363][@bibr26-0963689719854363][@bibr27-0963689719854363]--[@bibr28-0963689719854363]^. For clinical use and implantation in vivo, tubular multilayered tissue engineered grafts show much promise, but the lack of clinical trials limits their broader use in a clinical setting. Conclusions {#section13-0963689719854363} =========== This systematic review summarizes recent articles on cells used in urethral TE. Cell culture techniques and ADSC (human and rabbit), UDSC, human urothelial, fibroblasts, and keratinocytes cell types were studied using different types of scaffold or a scaffold-free approach. Studies show that proliferation and differentiation of cultured cells are affected both by the culture conditions and type of scaffold. After choosing the cell source and cell culture technique, the choice of a scaffold (biological, synthetic, composite) is another determinant of the clinical outcome. Currently used scaffolds show varied results. However, there is no standardized or optimal scaffold that could be recommended for use in a clinical setting. In the near future, development can be expected in the construction of new bioreactors providing dynamic cultivation conditions that may improve the process of cell proliferation and differentiation. To reduce invasivity, autologous urinary stem cells and autologous ADSCs show much promise, as they can be expanded in vitro and used for TE and 3D bioprinting. Further development of urethral TE may be associated mainly with introducing new composite biomaterial and 3D bioprinting technology in combination with various cells including urinary stem cells to mimic the native urethral architecture. Current trends of experimental studies in the context of urethral TE might be summarized as follows. The focus is on the use of UDSCs, as their harvesting is non-invasive and economic, and efficient stem cell differentiation with proper culture media supplemented with regulatory factors (growth, differentiation) is highlighted, as the application of an inadequately engineered construct in vivo might lead to urine leakage or insufficient vascular supply with resultant stricture formation. A detailed understanding of immunomodulatory processes under in vitro conditions provides the necessary insight on how urethral strictures are formed. As contraction and expansion are the physiological features of the urethra that might be mimicked in vitro under dynamic conditions, bioreactors play a key role in the process of tissue maturation. Another crucial point is to develop a technique to provide sufficient nutrient supply for the engineered construct, which could also enhance long-urethral segment reconstruction using TE approaches. This type of reconstruction requires an adequate tubular cell-seeded graft with the inner lining of the urothelium and outer lining composed of SMCs. Therefore tubularization of TE constructs is also a point of interest. Cell labeling with specific particles allows better visualization of the reconstructed tissue site. This review shows that many studies have been done on animal experiments. However, the transformation of these results is still lacking in the field of clinical medicine. **Ethical Approval:** This study was approved by our institutional review board. **Statement of Human and Animal Rights:** This article does not contain any studies with human or animal subjects. **Statement of Informed Consent:** There are no human subjects in this article and informed consent is not applicable. **Declaration of Conflicting Interests:** The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. **Funding:** The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Presented work was supported by the grants APVV-15-0111 and UK/240/2018. **ORCID iD:** Lubos Danisovic  <https://orcid.org/0000-0002-5074-9621>

1. Introduction {#sec1} =============== Deep resources such as oil, gas, and solid mineral have drawn more interest. Generally, the deeper drill is characterized by higher pressure and temperature, which make the drilling and borehole stability harder \[[@B1]--[@B7]\]. However, in Mexico Bay, North Sea Basin, Sichuan Basin, and the South Sea of China \[[@B8]\], for example, the gas and oil reservoirs in layers over 200°C have been successfully exploited. When the fluid circles, the upper surrounding rock will be heated; when the fluid ceases to work, however, the lower one will be heated. Balanced by the fluid column pressure and the rock confining pressure \[[@B9], [@B10]\], the heated rock will fail to expand, generating thermal stress as a result \[[@B11]\]. Maury and Guenot claim that the thermal stress contributes most to the instability of the borehole \[[@B12]\]. The outcome they obtained shows that when the temperature of the midhard rock rises up by 1 centigrade, the stress can increase by 0.4 MPa, up to 1 MPa for the harder rock as a result. The thermal stress in 25 MPa to 50 MPa is practically common in 4000 meters boreholes. Consequently, the initial borehole stress and the common thermal stress can work together leading to collapse and fracture. Wang et al.\'s research \[[@B13]\] shows that the Westerly granite can generate thermal cracking when heated up to 75°C. And the threshold value of 60\~70°C is suggested by Chen et al. \[[@B14]\]. Impacted by hydrostatic stress and thermal cracking, the granite\'s peak of the permeability, up to 3.5 × 10\~4 mD/°C, to the initial one reaches up to 93 \[[@B15]\]. This indicates that a field characterized by high permeability is developed around the borehole, triggering another stress field. In the borehole, the initial stress, temperature, and the stress field were triggered by overlapping the fluid together, which led to the deformation instability and leakage \[[@B9], [@B16], [@B17]\]. Consequently, the instability may make the drill stick or damage the casing. Since the 1980s, in order to dispose the permanent nuclear waste, people began researching the coupling of THM (thermo-hydro-mechanical) \[[@B18], [@B19]\]. A global International cooperation project named DECOVLEX was established in 1992. Since then, a series of experiments, including modeling, have been conducted and some invaluable outcomes have been obtained as a result \[[@B20]--[@B24]\]. At the fourth stage of this project, the aim mainly was to study the mechanics of crystalline rock and the process in which the mechanical and hydraulic properties of the EDZ (excavation damage zone) are transformed. This process can harden or soften the rock \[[@B25]\]. In this paper, the thermal physical and mechanical properties of the granite are developed and researched under high temperature and three-dimensional stress. By utilizing*ANASYS-APDL*(ANSYS Parameter Design Language-APDL) \[[@B26], [@B27]\], the dynamic evolution equations of elastic modulus, Poisson ratio, uniaxial compressive strength, and permeability of granite with temperature are built and run. The temperature-fluid-stress coupling model to analyze the granite\'s stability is established and simulated to figure out the temperature\'s influence on collapse pressure, fracture pressure, and stress near the borehole, which can provide theoretical guidance for borehole stability and safety drilling in granite formations. 2. Thermophysical and Mechanical Properties {#sec2} =========================================== 2.1. Overview of Experiment {#sec2.1} --------------------------- The sample, obtained from a 1000 meter deep borehole in Mount Yan, North China, is about 100 mm with a diameter of 50 mm. The density is about 2.54 g/cm^3^. TAW-1000 deep pore pressure servo experimental system was employed to test the sample. It consists of quartz, feldspar, and hornblende. All the samples were processed on the basis of Chinese national standard of GB50128-94 (shown in [Figure 1](#fig1){ref-type="fig"}). In order to avoid being contaminated by the hydraulic oil, we encapsulated the sample with a 3 mm thickness hot pyrocondensation pipe. The experiments were conducted in a 1000°C electrothermal furnace whose space is 300 × 200 × 120 mm. The samples were placed at the center of the furnace, to whose front and rear it is about 3 mm far from the sample. All the samples were divided into 5 groups, with each was heated to room temperature, 100°C, 200°C, 300°C, 400°C and insulated for 2 hours, respectively. Compared with the original sample in [Figure 2](#fig2){ref-type="fig"}, these heated to 300°C and 400°C is dark red, owing to the Fe^3+^ transformed from Fe. 2.2. Longitudinal Wave Velocity Characteristics {#sec2.2} ----------------------------------------------- [Figure 3](#fig3){ref-type="fig"} plots the link between longitudinal wave velocity and the temperature. The curve shows that the speed varies inversely with the temperature. This can be accounted for as follows: (I) as free water inside the rock evaporates, the pore becomes bigger; (II) when the temperature increases, the thermal stress will be triggered between minerals, due to their different coefficients of thermal expansion and anisotropy, generating new fractures or expanding the old. 2.3. Uniaxial Compression Tests {#sec2.3} ------------------------------- ### 2.3.1. Uniaxial Strength and Strain {#sec2.3.1} [Figure 4](#fig4){ref-type="fig"} plots the link between temperature and uniaxial strength. It shows that the threshold temperature is 200°C, in accordance with the result obtained from [Figure 5](#fig5){ref-type="fig"}. Below 200°C, the sample mainly undergoes brittle fracture, specially divided into compacting and linear elastic phases. On the other hand, over 400°C, the sample mainly undergoes the shear and tensile fractures. Below 200 centigrade, the peak stress increases slowly but rapidly when it is over 200 centigrade. It shows that the threshold temperature is 200 centigrade, which accords with the outcome obtained from the link between the temperature and the uniaxial strength. ### 2.3.2. Elasticity Parameters of the Sample {#sec2.3.2} The thermal damage is introduced to reflect the fluctuation of the elastic modulus of the samples before and after the heating the sample. The thermal stress will be produced between different mineral compositions due to the temperature change \[[@B28]\]. The thermal damage is calculated as follows: $$\begin{matrix} {D\left( T \right) = 1 - \frac{E_{(T)}}{E_{(0)}}.} \\ \end{matrix}$$ The elastic modulus decreases with the increase of the temperature. Additionally, *e* the relationship between the elastic modulus and temperature is fitted by the data, and its fitting formula is *E* = −0.0145*T* + 29.997, with a goodness of 0.955. [Figure 6](#fig6){ref-type="fig"} displays the increase of thermal damage after the sample was heated. As same as aforementioned, the threshold temperature obtained from the*D*-*T* curve is also 200 centigrade. When the temperature is from 0 centigrade to 100 centigrade and over 200 centigrade, the thermal damage of the sample is increasing while the thermal damage is unchanged from 100 centigrade to 200 centigrade. The Poisson ratio is characterized by polymeric. As shown in [Figure 7](#fig7){ref-type="fig"}, with the increasing of the temperature, the Poisson ratio of the granite samples is more and more mounting. The proportion between Poisson ratio and temperature is mainly accounted for two reasons: (I) the increase of the temperature leads to the changes of the sample\'s interior structure, the water content, and the porosity; (II) and the temperature and stress are beyond the sample\'s elasticity. ### 2.3.3. Damage States of Samples {#sec2.3.3} The sample was experimentally damaged under uniaxial pressure in three ways as shown in [Figure 8](#fig8){ref-type="fig"}: (I) under room temperature, the sample undergoes the brittle fractures developing along the axial direction. (II) Under 100--200°C, the sample undergoes the shear fracture. If loaded, the softer part would be damaged without losing its bearing capacity. (III) Under 300--400°C, being sheared and tensioned, the sample undergoes the column fractures. 2.4. Triaxial Compression Tests {#sec2.4} ------------------------------- ### 2.4.1. Mechanical Properties with Different Confining Pressure {#sec2.4.1} [Figure 9](#fig9){ref-type="fig"} displays the link between triaxial compressive strength and confining pressure. It shows that as the confining pressure rises, the triaxial compressive strength virtually and nonlinearly increases. With *R* = 0.996, the nonlinear link can be expressed as $$\begin{matrix} {\sigma_{s} = 0.834\sigma_{w}^{2} - 14.05\sigma_{w} + 269.65.} \\ \end{matrix}$$ The link between elastic modulus and confining pressure was displayed in [Figure 10](#fig10){ref-type="fig"}. The elastic modulus changed the same as the confining pressure except 20 MPa. The threshold pressure is 15 MPa. The elastic modulus can be expressed as $$\begin{matrix} {E = - 0.095\sigma_{w}^{2} + 3.085\sigma_{w} + 8.775.} \\ \end{matrix}$$ ### 2.4.2. Mechanical Properties with Different Temperatures {#sec2.4.2} Figures [11](#fig11){ref-type="fig"}, [12](#fig12){ref-type="fig"}, and [13](#fig13){ref-type="fig"}, respectively, present the influence exerted by a given temperature on triaxial compressive strength, axial strain at failure, and elastic modulus, which are characterized by discreteness. The three figures confirm that 200°C is the threshold temperature and pressure. ### 2.4.3. Damage States of Samples {#sec2.4.3} Tested by deep pore pressure servo experimental system, the samples were broken by two ways: (I) when heated to 200°C or lower, the sample undergoes the brittle fracture. However, when the confining pressure increased to 20 MPa, the shear and tension fracture dominated. (II) When heated over 200°C, the sample undergoes the compression shear and fracture ([Figure 14](#fig14){ref-type="fig"}). 2.5. Permeability Effected by Temperature {#sec2.5} ----------------------------------------- The permeability was measured by TAW-1000 deep pore pressure servo experimental system. The sample was enwrapped by a 3 mm thickness hot pyrocondensation pipe. Pressed around by 20 MPa, the sample\'s one end was ventilated by N~2~ and a highly precise gas flowmeter was installed at its other end. [Figure 15](#fig15){ref-type="fig"} indicates that the threshold temperature is 200°C. The thermal fracture improves the permeability. 3. Finite Element Simulation and Experiment {#sec3} =========================================== 3.1. Basic Equations {#sec3.1} -------------------- Adopting the definition of Biot\'s effective stress, the relationship between effective stress and total stress is $$\begin{matrix} {\sigma^{\prime} = \sigma + p_{w}I.} \\ \end{matrix}$$ The mass conservation equation of fluid is $$\begin{matrix} {\frac{\partial}{\partial x_{i}}\left\lbrack {\frac{\rho_{1}k_{1ij}}{\mu_{1}}\left( {\frac{\partial p_{1}}{\partial x_{j}} + \rho_{1}g_{j}} \right) + \rho_{1}k_{1Tij}\frac{\partial T}{\partial x_{j}}} \right\rbrack} \\ {\quad\quad = n\frac{\partial p_{1}}{\partial t} + \rho_{1}\frac{dn}{dt}.} \\ \end{matrix}$$ The energy conservation equation of solid is $$\begin{matrix} {\frac{\partial}{\partial t}\left\lbrack {\left( 1 - n \right)\rho_{s} \cdot C_{s} \cdot \Delta T} \right\rbrack = - \frac{\partial q_{si}}{\partial x_{j}} + Q_{s}.} \\ \end{matrix}$$ The energy conservation equation of fluid is $$\begin{matrix} {\frac{\partial}{\partial t} = n \cdot \rho_{1} \cdot C_{1} \cdot \Delta T = - \frac{\partial}{\partial x_{j}}\left( {q_{1i} + q_{1i}^{c}} \right),} \\ {q_{1i} = - \lambda_{1ij}n\frac{\partial T}{\partial x_{j}}q_{1i},} \\ {q_{1i}^{c} = \rho_{1} \cdot C_{1} \cdot \Delta T \cdot \nu_{1i}^{r}} \\ {= - \rho_{1} \cdot C_{1} \cdot \Delta T \cdot \left\lbrack {\frac{k_{1ij}}{\mu_{1}}\left( {\frac{\partial p_{1}}{\partial x_{j}} + \rho_{1} \cdot g_{j}} \right) + k_{1Tij}\frac{\partial T}{\partial x_{j}}} \right\rbrack.} \\ \end{matrix}$$ Assume that at any point inside the solid phase and liquid phase has the same temperature, the total energy conservation equation \[[@B29]\] can be expressed as $$\begin{matrix} {\frac{\partial}{\partial t}\left\lbrack {\left( {1 - n} \right)\rho_{s} \cdot C_{s} \cdot \Delta T + n \cdot \rho_{1} \cdot C_{1} \cdot \Delta T} \right\rbrack} \\ {\quad\quad = - \frac{\partial q_{si}}{\partial x_{j}}\left( q_{mi} + q_{1i}^{c} \right) + Q_{s}.} \\ \end{matrix}$$ The total heat flux density of rock and fluid can be expressed as $$\begin{matrix} {q_{mi} = q_{si} + q_{1i} = - \left\lbrack {\lambda_{sij} \cdot \left( {1 - n} \right) + \lambda_{1ij} \cdot n} \right\rbrack \cdot \frac{\partial T}{\partial x_{j}}.} \\ \end{matrix}$$ Based on mixture theory, the equivalent thermal conductivity can be defined; namely, $$\begin{matrix} {\lambda_{mij} = \lambda_{sij} \cdot \left( {1 - n} \right) + \lambda_{1ij} \cdot n.} \\ \end{matrix}$$ According to the principle of virtual displacement, the whole equilibrium differential equations in solution domain can be represented as $$\begin{matrix} {{\int_{\Omega}{\delta \cdot \varepsilon^{T} \cdot \sigma^{\prime}}} \cdot d\Omega - {\int_{\Omega}{\delta \cdot u^{T} \cdot b}} \cdot d\Omega - {\int_{\Omega}{\delta \cdot u^{T} \cdot t}} \cdot ds = 0.} \\ \end{matrix}$$ We take the effective stress of rock skeleton equation into ([11](#EEq9){ref-type="disp-formula"}), and according to the mass conservation equation of the fluid and fluid-solid overall energy conservation equation to form the control equations under the heat-flow-solid coupling. The finite element discretization method can be used to solve the equations\' system after it has been transferred to the equivalent credits\' weak formulation. 3.2. Dynamic Evolution Equations {#sec3.2} -------------------------------- Based on indoor experiment of this study paper, it was found that the dynamic evolution equations of elastic modulus, Poisson ratio, uniaxial compressive strength, and permeability of Granite with temperature can be represented as $$\begin{matrix} {E = - 0.0145T + 29.977,} \\ {\upsilon = 0.0004T + 0.1185,} \\ {\text{UCS} = - 0.0001T^{2} - 0.0284T + 64.05,} \\ {K = 1E - 08T^{2} - 2E - 06T + 0.0002.} \\ \end{matrix}$$ 3.3. Engineering Application Example {#sec3.3} ------------------------------------ The paper uses the ANSYS secondary development function of the fluid-solid interaction module and temperature-structure coupling calculation module for the solver, according to the decoupling method; firstly we do numerical calculation of the granite-borehole temperature field and then put the results into ANSYS fluid-solid interaction of calculation module. The units\' segmentation of temperature field and the units\' segmentation of flow-solid coupling calculation is the same, such that the plane-strain problems use four-node units. The dynamic evolution of elastic modulus, Poisson ratio, uniaxial compressive strength, and permeability of granite using secondary development of the ANSYS parametric design language (ANSYS Parameter Design Language-APDL) to achieve. Firstly, to extract the temperature of the unit in the process of thermal analysis calculation, and modify the unit parameters of the thermal and mechanical properties, to form Loop iteration control process, and realize the Granite-borehole temperature coupling. The stimulation was performed on a one-fourth sample of symmetry. The sample was divided into 612 four-point units in [Figure 16](#fig16){ref-type="fig"}. We compared the finite element calculation results with the analytical solutions of Marshall and Bentsen \[[@B30]\] to verify the reliability of the model adopted in this paper. According to the relationship of the rock mechanics of granite that tested indoor and confining pressure, then converted it to the parameters under the condition of confining pressure in this area and applied it to this model, finally the temperature distribution of the borehole surrounding rock was acquired. [Figure 17](#fig17){ref-type="fig"} illustrated the temperature distribution of the borehole surrounding rock after 8-hour drilling, where the finite element calculation results matched borehole with the analytical solutions. The of the wall and surrounding rock decreased gradually along with the decreasing of drilling fluid temperature, besides the thermal stress of the wall down to the minimum. With the increasing of distance from the borehole, the formation temperature increased gradually until it reached the original formation temperature. The formation that far away from the wall approximately equal five times the boreholebore radius, its internal temperature almost no change, and stayed at the original formation temperature 182°C. The influence that impacted the granite strata borehole wall stability in the temperature field, the stress field, and the seepage field mainly was exerted by changing the stress state of the borehole \[[@B31]\]. As a result, the original formation of equilibrium was destroyed so that the stress concentration produced around the borehole easily brought up the sidewall instability. The below three kinds of conditions were accounted for to explain the influence on the sidewall stress brought by interconnection. Firstly, do not consider the interconnection of the temperature field and the stress field but consider the interconnection of the seepage field and the stress field. Secondly, do not consider the interconnection of the seepage field and the stress field, but consider the interconnection of the temperature field and the stress field. Thirdly, consider the interconnection of the temperature field and the stress field and the seepage field simultaneously. ### 3.3.1. Stress in Borehole {#sec3.3.1} Figures [18](#fig18){ref-type="fig"} and [19](#fig19){ref-type="fig"} display the distribution of radial and tangential stresses peripheral to the borehole. It indicates that temperature and percolation accordantly influence the stress. The minimum stress occurs near the borehole; on the other hand, the samples virtually undergo the same stress under the above three conditions in the further field. ### 3.3.2. Borehole Stability Effected by Temperature {#sec3.3.2} The shear fracture of the rock, subject to Mohr-Coulomb, expressed by the main stress is described as $$\begin{matrix} {\sigma_{1} = \sigma_{3}\text{ta}\text{n}^{2}\left( {\frac{\pi}{4} + \frac{\varphi}{2}} \right) + 2C\tan\left( {\frac{\pi}{4} + \frac{\varphi}{2}} \right).} \\ \end{matrix}$$ The shear fracture will occur when the maximum and minimum effective principal stresses are beyond the breaking strength of the rock. The layer will collapse when the tangential effective stress is over the tensile strength of the rock: $$\begin{matrix} {\sigma_{\theta} - \alpha P_{P} = - S_{t}.} \\ \end{matrix}$$ The stress distribution is calculated on the basis of finite element. Considering the shear failure and tensile failure, the collapse pressure and tensile pressure are calculated. Suppose the uniaxial compressive strength is subject to temperature. Based on Griffith, $$\begin{matrix} {\sigma_{c} = \left( {\left. 8 \right.\sim 12} \right)S_{t}.} \\ \end{matrix}$$ The variations of collapse pressure and fracture pressure with temperature increase and decrease are shown in Figures [20](#fig20){ref-type="fig"} and [21](#fig21){ref-type="fig"}, respectively. ### 3.3.3. Borehole Stability Affected by Permeability {#sec3.3.3} A filter cake can be developed as the fluid seeps through the permeable reservoir. In this case where the fluid will be constrained, the pore pressure is not equal to the drilling fluid column pressure. [Figure 22](#fig22){ref-type="fig"} plots the link between the permeability coefficient and the collapse and fracture pressure. The fact that the value by which the fracture decrease is bigger than the collapse pressure increase indicates that the permeability coefficient influences the fracture pressure more. Consider $$\begin{matrix} {\delta = \frac{\left( {p_{w} - p_{o}} \right)}{\left( {p - p_{o}} \right)},\quad 0 \leq \delta \leq 1.} \\ \end{matrix}$$ ### 3.3.4. Stability in Deviated Borehole {#sec3.3.4} * * *(I) Collapse Pressure in Deviated Borehole*. Under different conditions, in Figures [23](#fig23){ref-type="fig"}, [24](#fig24){ref-type="fig"}, and [25](#fig25){ref-type="fig"} the distributions of the collapse pressure were performed. Suppose north-south and east-west as the directions along which the horizontal maximum and minimum stresses developed, respectively. It can be concluded that the seepage can cut the maximum and add the minimum collapse pressures; the decrease of the temperature, however, leads to the increase of the maximum and minimum collapse pressure. What is more, [Figure 25](#fig25){ref-type="fig"} shows that the minimum collapse pressure can reach the smallest, with the maximum ranking the middle. Without considering the fluid, the result will be below the prediction; on the other hand, without considering the temperature, the result will be beyond the prediction. *(II) Fracture Pressure in Deviated Borehole.* Additionally, the distribution of the fracture pressure was performed in Figures [26](#fig26){ref-type="fig"}, [27](#fig27){ref-type="fig"}, and [28](#fig28){ref-type="fig"}. When drilling along the direction of the minimum principal stress, the fracture pressure reached the biggest, increasing the upper boundary of the fluid\'s density. It shows that the wider the window of the fluid is, the safer the drilling is. When drilling along the direction of the maximum stress, the fracture pressure reaches the minimum. As a result, it is suggested that in order to ensure the borehole stability, we should drill along the direction of the maximum stress. If the fracture pressure is beyond the expectation, the sloughing formation will be developed. 4. Conclusion {#sec4} ============= It is shown that the threshold temperature of strength and elastic modulus of granite are both 200 centigrade. Below this, the sample mainly undergoes the brittle fracture and the rupture surface is along the axial direction under small confining pressure, while shear compression failure is the main state when the confining pressure is over 20 MPa. Above 200 centigrade, the damage modes are mixing shear compression and brittle fracture failure, and shear compression failure is positively correlated with the increasing of confining pressure and temperature.The compressional wave velocity, elastic modulus, and uniaxial compression strength will decrease as the temperature rises. Additionally, when the temperature is given, the elastic modulus and strength will increase as the surrounding pressure rises. The threshold pressure and temperature are 15 MPa and 200°C, respectively. The threshold thermal fracture temperature is 200°C. The permeability will dramatically increase with the rise of temperature up to 10^−3^\~10^−4^ mD.The coupling borehole stability model of thermo-fluid-solid is developed by the ANASYS-APDL. The dynamic evolution equations of elastic modulus, Poisson ratio, uniaxial compressive strength, and permeability of granite with temperature are built and run. The results show that the radical stress and tangential stress are greatly different in full coupling model and in other physical field models. The results simulated by full coupling model are more precise and reliable than other models.The temperature affects the fracture pressure more than the collapse pressure. In order to avoid losing fluid, we suggest lowering the fluid\'s density when the temperature of the borehole wall decreases. As for the permeability, its rise leads to the decrease of the fracture pressure but increase of the collapse pressure, which indicates that the low-density fluid is better.The seepage degrades the upper limit of collapse pressure and heightens the lower limit. The fall of temperature heightens both upper and lower limits of collapse pressure in borehole. As a result, in order to accurately predict the collapse pressure, the seepage and temperature are supposed to be taken into account. The authors gratefully acknowledge the support by the Fundamental Research Funds for the Central Universities (Grant no. 2652011273), the International Scientific and Technological Cooperation projects (Grants nos. 2010DFR70920 and 2011DFR71170), the National Natural Science Foundation of China (Grant no. 51004086), and the open Funds of Key Laboratory on Deep Geo-Drilling Technology, Ministry of Land and Resources (Grant no. NLSD201210). Meanwhile, great thanks also go to former researchers for their excellent works, which was of great help to our academic study. *D*(*T*): : Thermal damage coefficient *T*: : Temperature, °C *E*~(*T*)~: : Elastic modulus at *T*°C, GPa *E*~(0)~: : Elastic modulus at 20°C, GPa *E*: : Elastic modulus, GPa *R*: : Goodness of fit *σ*~*s*~: : Triaxial compressive strength, MPa *σ*~*w*~: : Confining pressure, MPa *σ*′: : Matrix of effective stress, MPa *σ*: : Matrix of total stress, MPa *I*: : Second-order unit tensor *p*~*w*~: : Absolute value of pressure, MPa *n*: : Porosity *k*~1*ij*~: : Permeability coefficient of fluid *k*~1*Tij*~: : Velocity of flow coefficient affected by temperature *μ*~1~: : Viscosity coefficient of fluid *ρ*~1~: : Density of fluid, kg/m^3^ *p*~1~: : Hydraulic pressure, Pa *g*~*j*~: : Acceleration of gravity of fluid, m/s^2^ *C*~*s*~: : Specific heat capacity, J/kg · K *ρ*~*s*~: : Density of rock, kg/m^3^ *q*~*si*~: : Heat flux density of rock, J/m^2^ · s *Q*~*s*~: : Energy conversion coefficient, J/m^3^ · s *C*~1~: : Specific heat capacity of fluid, J/kg · K *v*~*li*~^*r*^: : Relative density of fluid *q*~*li*~^*c*^: : Heat flow, W/m^2^ *q*~*mi*~: : Total heat flux density, J/m^2^ · s *q*~1*i*~: : Heat flux density of fluid, J/m^2^ · s *λ*~*sij*~: : Heat transfer coefficient of rock, W/m^2^ · K *λ*~1*ij*~: : Heat transfer coefficient of fluid, W/m^2^ · K *λ*~*mij*~: : Equivalent thermal conductivity coefficient, W/m^2^ · K *ε*: : Strain *b*: : Three-dimensional force, N *t*: : Plane vector force, N *δ*: : Cake permeability *υ*: : Poisson ratio UCS: : Uniaxial compressive strength, MPa *K*: : Permeability, mD *σ*~1~: : Maximum main stress, MPa *σ*~3~: : Minimum main stress, MPa *φ*: : Internal friction angle, rad *C*: : Cohesive force, N *σ*~*θ*~: : Tangential effective stress in borehole, MPa *α*: : Effective stress coefficient *P*~*P*~: : Pore pressure, MPa *S*~*t*~: : Tensile strength, MPa *σ*~*c*~: : Uniaxial compressive strength, MPa *p*: : Drilling fluid column pressure, MPa *p*~*w*~: : Borehole pore pressure, MPa *p*~0~: : Formation pore pressure. Conflict of Interests ===================== The authors declare that there is no conflict of interests regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} {#fig9} {#fig10} {#fig11} {#fig12} {#fig13} {#fig14} {#fig15} {#fig16} {#fig17} {#fig18} {#fig19} {#fig20} {#fig21} {#fig22} {#fig23} {#fig24} {#fig25} {#fig26} {#fig27} {#fig28} [^1]: Academic Editors: C. Nah and A. Tonkikh

Introduction {#s1} ============ Based on the current trends in fossil energy production and use, deforestation, and population growth, it is expected that the increase of global mean surface temperatures for 2081--2100 relative to 1986--2005 is projected to be in the ranges of 0.3 to 1.7°C (RCP2.6), 1.1 to 2.6°C (RCP4.5), 1.4 to 3.1°C (RCP6.0), and 2.6 to 4.8°C (RCP8.5), which will have dramatic effects on economics, agriculture, and environment (AR5, IPCC, [@B26]). Plant traits are sensitive to climate warming and ecologists use plant trait-climate relationships to simulate plant physiology and growth in current and future climate situations (Farquhar and Sharkey, [@B18]; Wang et al., [@B66]; Jing et al., [@B28]). Therefore, understanding the patterns of plant physiological and morphological responses to global warming is of great importance in simulating and predicting the impact of global change on natural systems and agriculture. Predictions of response to global warming may be derived from experimental and observational studies (Tilman, [@B58]; Wang et al., [@B65], [@B67]; Knapp et al., [@B33]). While both types of study are common, relatively few authors have investigated whether they produce similar predictions or reflect reality (Dunne et al., [@B16]; Knapp et al., [@B32]). Experimental global change studies are typically limited in scope both spatially and temporally (Rustad et al., [@B49]). Observational studies often have broader spatial and temporal scales but suffer from a lack of control over covariates in biophysical and biochemical parameters of weather and soil. To minimize the weaknesses of each approach, it has been suggested that more research should explicitly unite observational and experimental work, perhaps by nesting experiments at multiple sites within a larger observational context or through summarized meta-analysis (Dunne et al., [@B16]; Jing et al., [@B28]). Many manipulative experiments controlling physical and environmental factors have been conducted around the world to investigate the potential effects of global change on plants and terrestrial ecosystems (Sage and Kubien, [@B51]; Rustad, [@B50]; Wang et al., [@B67]). However, the methodology used in these experiments was different in their research settings, treatment intensities and durations and targeted species. The impact of short-term vs. long-term warming on plants traits would probably be different due to plants\' acclimation capacity in photosynthesis, respiration and other physiological processes and these impacts would vary among different plant functional types (PFTs) under natural or controlled settings (Smith and Dukes, [@B54]). Plants\' physiological and morphological responses to short-term warming treatment, however, are often used to parameterize the sub-models of photosynthesis, stomatal conductance, and respiration in plant growth and terrestrial ecosystem models, which would likely unrealistically simulate plant energy, carbon, and water fluxes in the long term. Indoor or outdoor settings and pot sizes could also affect the magnitude of ecophysiologial responses to temperature increase by implicating root growth and plant above-ground and below-ground tissue interactions (Arp, [@B3]). To accurately predict the impacts of climatic change and develop proper adaptive agricultural management practices, it is imperative to understand how temperature changes of different intensities and duration and changes manipulated under different experimental settings affect photosynthetic carbon gain, loss and allocation through a comprehensive analysis of relevant studies. Previous research and meta-analyses have indicated that global warming will promote plant photosynthesis, dark respiration, leaf nitrogen content, specific leaf area, and other metabolisms (Poorter et al., [@B41]). It has been reported that the modulation of leaf traits and trait relationships by site climatic properties was modest (Wright et al., [@B71]). However, the modulation of leaf traits by warming treatment of different intensities and duration has not been extensively analyzed. Understanding how these processes vary among different species and plant functional types is a major goal for plant ecology and crucial for modeling how nutrient fluxes and vegetation boundaries will shift under global warming. The effect of the intensities and the treatment duration of global warming manipulative experiments on the plant physiology and growth among different plant functional groups, however, remain unclear. Therefore, the main objective of this study was to investigate the effects of global warming treatment with different magnitudes and durations on plant response in ecophysiological traits. Specifically, we aim to: (1) assess the impact of global warming of different magnitudes and durations on plant ecophysiological traits at leaf level; (2) detect the variations of ecophysiological traits response of different plant functional types to warming treatment of different durations; (3) explore the effect of different experimental settings on the response of a plant\'s traits to global warming. Accordingly, we propose: (1) due to plant acclimation capacity, short-term vs. long-term warming has different impacts on plant traits, with short-term warming having a more stimulating effect on the physiological functions of plants; (2) different experimental facilities may change the response of plants traits to warming treatment. To test these hypotheses, we conducted a comprehensive meta-analysis of the warming manipulating studies published from 1980 to 2018. Materials and Methods {#s2} ===================== Data Collection --------------- Journal articles were searched on the Web of Science database with the keyword "leaf traits & warming," "leaf traits & temperature increase" and etc. The articles were later cross-checked with review articles and book chapters. The articles were imported into EndNote software and formed a database. All articles about warming effects on leaf traits were screened to ensure that all the articles available were included for the analysis. The articles published from 1980 to 2018 and meeting the following two conditions were included in the analysis: (1) the control group in the experiment was treated at ambient temperature situation; (2) physiological and morphological measurements were performed on both ambient and manipulated groups. Articles were rejected if: (1) plant physiological changes under warming treatments led to death of or severe damage to the plant; (2) there were other stressing factors impacting the warming treatments. Finally, 80 papers meeting the requirements were included in the database ([Supplementary Material S1](#SM1){ref-type="supplementary-material"}). Data was obtained directly from the table or was extracted using the GetData Graph Digitizer software from the selected articles. In these studies, the magnitude of warming treatment ranged between 0.3 and 25°C, with only two studies showing a warming treatment above 20°C above AT ([Supplementary Material S1](#SM1){ref-type="supplementary-material"}). Response variables collected from these articles included net photosynthetic rate (A~net~), stomatal conductance (G~s~), leaf nitrogen (LN), dark respiration (R~d~), and specific leaf area (SLA). When A~net~, R~d~, and G~s~ of one species with the same unit were all provided in the study (including measurements conducted on the same leaves/individuals and those across individuals), the R~d~/A~net~, and A~net~/G~s~ in the control and warming treatments were calculated. In addition to the above responsive variables under different treatments, plant species, sample size, growth facilities, and duration of warming treatment were also collected. To ensure the independent nature of the data, we excluded duplicate results collected from the same studies. However, our analyses were not completely independent because individual study often provided data with more than one treatment (e.g., different warming treatment intensities) and/or different response variables. To examine the influence of non-independence of data, we first averaged those data from the same published study by PFTs so that only one comparison was used from a published study for each PFT. Nonetheless, we found that most of the response patterns were unchanged; therefore, all data were used in our study. Categorization of the Studies ----------------------------- Temperature treatment was divided into two categories: AT (ambient temperature) and ET (elevated temperature). Plant species were classified into different photosynthetic pathways (C~3~, C~4~, or CAM), growth forms (herb or wood) and economic values (crop or non-crop). Experimental facilities were categorized into indoor (growth chambers or greenhouses) and outdoor (open top chambers or fully-open) settings and \<10 L and \>10 L growing pots. In our dataset, exposure time (i.e., how long plants were exposed to warming) ranged from \<10 days to \>10 years. To analyze the possible different responses under various warming durations, we banded the temperature treatment into two categories: short-term (\<1 year) and long-term (\>1 year). Warming treatments that were applied through air warming were included in the analyses. We listed the species, PFTs information and relevant experimental methodology used in this study ([Supplementary Material S1](#SM1){ref-type="supplementary-material"}). Meta-Analysis Methods --------------------- To avoid the adverse effects of different units, we used the response ratio r = X~t~/X~c~ to estimate the magnitude of the effect of warming treatment, where X~t~ is the treatment mean and X~c~ is the control mean. For ease of comparison, we calculated the natural logarithm of the response ratio (lnr). The standard deviation (SD) and the sample size (*n*) for each observation were collected to calculate the variance of the effect size. The lnr was calculated without and with being standardized by warming magnitude (Equations 1, 2). log e r = log e ( X t X c ) = log e ( X t ) − log e ( X c ) log e r = log e ( X t X c ) T t − T c = log e ( X t ) T t − T c − log e ( X c ) T t − T c where T~t~ and T~c~ are the temperature in the warming and control treatments, respectively. Using METAWIN software 2.1 (Sinauer Associates, Inc. Sunderland, MA, USA), we calculated the effect size of the target variables and used a weighted fixed-effect model to assess the effect of plant functional types, experimental settings, and treatment duration. If the 95% confidence interval (CI) of the effect size produced by the fixed-effect model overlaps with 0, no significant effect was detected on the response variables. If the upper limit of 95% CI is less than 0, the effect is considered significantly negative. In contrast, if the lower limit of 95% CI is greater than 0, the effect is considered significantly positive. If the 95% CI of the effect size among different species, pot size, and treatment duration does not overlap, their response is considered significantly different. Unless otherwise indicated, significance level was set at *p* \< 0.05. The publication bias for effect size (lnr) in this meta-analysis was also calculated. We calculated Spearman\'s rank order correlation (rs) which indicates the relationship between the effect size (lnr) and the sample size (Begg and Mazumdar, [@B6]), and Rosenthal\'s fail-safe number which represents the number of additional studies with a mean effect size of zero needed to eliminate the significance of a significant effect (Rosenthal, [@B47]). Publication bias was significant if *p*-value of rs was smaller than 0.05. However, the publication bias may be safely ignored if the fail-safe number is larger than a critical value of 5n+10 where n is the number of studies (Rosenberg, [@B46]). Statistical Analysis -------------------- Original data collected from these studies were arranged into a database in which the value of response variables was lnr. The effect of warming duration on lnr was considered significant if the 95% confidence interval (CI) of lnr does not overlap with 0. And when the 95% confidence intervals (CI) of lnr of different PFTs, facilities or pot size did not overlap with each other, the response was considered significantly different among different categories, the means of the ratio of the R~d~/A~net~ and A~net~/G~s~ in the control and warming treatments were compared using paired *t*-test. The relationship between lnr of all the variables and the magnitude of warming treatments were evaluated by a second-degree polynomial or linear regression analysis with the R statistical programming language (R 3.2.2 for Windows GUI front-end). Results {#s3} ======= Effects of the Duration of Warming Treatment on Plant Ecophysiological Traits Across Plant Functional Types (PFTs) and Growth Forms ----------------------------------------------------------------------------------------------------------------------------------- Warming treatment increased dark respiration (R~d~) and specific leaf area (SLA) and decreased net photosynthetic rate (A~net~) and leaf N concentration (LN) across all the experiments ([Figure 1](#F1){ref-type="fig"}). The response of standardized (triangle symbols) or unstandardized (circle symbols) rate of A~net~, G~s~, R~d~, LN, and SLA to warming treatment differed with different warming durations ([Figure 2](#F2){ref-type="fig"}). Long-term warming treatment (\>1 year) had a greater positive effect on R~d~ than short-term (\<1 year), regardless of whether the effect was standardized or unstandardized. LN was decreased by long-term warming but was increased or not changed by short-term warming treatment for unstandardized and standardized effect, respectively. Long-term warming treatment increased SLA, while short-term treatment had no effect on SLA. For standardized response of SLA, there was no difference between long and short-term treatments. For G~s~, long term treatment had a positive but short-term treatment had a negative effect on the standardized effect size. However, for the unstandardized effect size, short-term treatment did not have a significant but long-term had a negative effect on G~s~. Short-term had a positive and long-term treatment a negative effect on A~net~ for the unstandardized form of the effect. And for standardized effect of A~net~, long-term treatment had a more negative effect than short-term treatment ([Figure 2](#F2){ref-type="fig"}). {#F1} {#F2} The response of A~net~, G~s~, R~d~, LN, and SLA to warming treatment differed among PFTs with different photosynthetic pathways ([Figure 3](#F3){ref-type="fig"}). Warming had a more positive effect on R~d~ for C~4~ species than for C~3~ species, regardless of whether the effect size was standardized. Warming had a negative effect for C~3~ but a positive effect for C~4~ species on LN, SLA, and G~s~. In contrast, warming had a negative effect for C~4~ but near-zero effect for C~3~ species on A~net~ ([Figure 3](#F3){ref-type="fig"}). {#F3} Warming duration had a significant effect on the response of A~net~, G~s~, R~d~, LN, and SLA for PFTs with different photosynthetic pathways ([Figure 4](#F4){ref-type="fig"}). Long term warming treatment had a more positive effect than short-term on R~d~ for both C~3~ and C~4~ species, regardless of whether the effect was standardized. For LN, long term treatment had a negative effect but short-term treatment had a positive effect for C~3~ and C~4~ species. For C~3~ species, short term warming treatment had a positive and long-term had a negative effect on A~net~; for C~4~ species, long term warming treatment had a positive but short term a negative effect on A~net~. Similar trend was found for standardized A~net~, even though the magnitude of the effect differed. {#F4} Effects of Warming Duration on Plant Traits Across Different Experimental Settings ---------------------------------------------------------------------------------- The responses of A~net~, G~s~, R~d~, LN, and SLA to warming treatment differed among in-door and outdoor experimental settings ([Figure 5](#F5){ref-type="fig"}). Warming had a more positive impact on R~d~ in the in-door than the out-door settings for unstandardized effect size. Warming had a positive effect on LN for in-door, but a negative effect for outdoor settings. Being standardized with temperature treatment, warming had no impact on LN for the in-door but negative impact on outdoor experimental settings. Warming had a positive effect on SLA for in-door settings but tended to have a negative effect for outdoor settings. G~s~ responded positively to warming under in-door but negatively under outdoor settings. Warming treatment had a positive effect on unstandardized A~net~ under in-door settings but a negative effect under outdoor settings. For standardized A~net~, warming had a negative effect for both in-door and outdoor settings ([Figure 5](#F5){ref-type="fig"}). {#F5} The response of A~net~, G~s~, R~d~, LN, and SLA to warming treatment under indoor and outdoor experiment settings also differed with different treatment durations ([Figure 6](#F6){ref-type="fig"}). Short-term warming had a positive effect but long-term had a negative effect on R~d~ for indoor experimental settings. Long-term warming had a more positive impact on R~d~ than short-term for outdoor experimental settings for both standardized and unstandardized effect size. Short-term warming treatment had a more positive impact than long-term treatment on A~net~ for unstandardized effect size but had no difference on standardized effect size. Short-term had a positive impact on A~net~ for outdoor settings, but long-term treatment had a negative impact on A~net~ for unstandardized effect. Long-term warming treatment had a more negative effect on standardized A~net~ than short term for outdoor settings ([Figure 6](#F6){ref-type="fig"}). {#F6} Pot size had a significant impact on the responses of A~net~, G~s~, R~d~, and LN to warming treatment ([Figure 7](#F7){ref-type="fig"}). Warming had a positive impact on R~d~ for plants grown in pots larger than 10 L, while a negative effect for plants grown in pots smaller than 10 L. G~s~ responded positively to warming when grown at \<10 L plots but negatively at \>10 L plots. A~net~ of plants grown at \>10 L pots responded negatively to warming. Warming had no impacts on unstandardized A~net~ but a negative effect on standardized A~net~ of plants grown at \<10 L pots. {#F7} The response of A~net~, G~s~, LN, and SLA to warming treatment differed among different treatment durations when plants were grown in pots of different volumes ([Figure 8](#F8){ref-type="fig"}). Short-term warming had a positive effect but long-term, a positive effect on LN for plants grown at both \<10 L and \>10 L pots. Short-term warming had a negative effect on SLA, but long-term a positive effect for both \<10 L and \>10 L pots. G~s~ responded positively with both short and long-term warming treatments at \<10 L pots but negatively at \>10 L pots. A~net~ responded positively to long-term warming treatment at \<10 L pots but negatively at \>10 L pots ([Figure 8](#F8){ref-type="fig"}). {#F8} Effects of Warming Magnitude on Plant Traits Across Different Experimental Settings ----------------------------------------------------------------------------------- A~net~, R~d~, LN, and SLA formed a quadratic relationship to warming treatment ([Figure 9](#F9){ref-type="fig"}). The effect size of A~net~, R~d~, LN, and SLA to warming was highest or lowest when temperature change was 6.6, 2.5, 6.6, and 5.2°C above ambient temperature, respectively ([Figure 9](#F9){ref-type="fig"}). {#F9} Discussion {#s4} ========== Several meta-analyses have investigated the general tendency of warming impacts on plant physiology and production (Rustad et al., [@B49]; Jing et al., [@B28]). However, it remains unclear how the experimental methodology of warming treatment affects the responses of plant ecophysiological traits to warming at leaf level. In this study, we collected data from warming manipulative studies and analyzed changes in the ecophysiological responses in the leaf traits. Overall, we found that (1) the direction and degree of the effect of warming treatment of different durations and settings on plant ecophysiological traits varied significantly; (2) there were significant variations among plant functional types in response to warming treatment of different methodology. Consistent with previous findings from other studies, this meta-analysis confirmed that R~d~ and SLA were stimulated by warming treatment (Rustad et al., [@B49]; Jing et al., [@B28]). Increasing, decreasing or neutral impacts of experimental warming have been observed for net photosynthetic rates (Bruhn et al., [@B8]; Bronson and Gower, [@B7]; Li et al., [@B36]). The net photosynthetic rate in this analysis was significantly decreased by warming treatment. The decrease in plant photosynthetic capacity may be attributed to the decreased LN under warmed conditions. Many studies showed that plant photosynthetic capacity was positively related to leaf N concentrations (Kattge et al., [@B30]; Reich et al., [@B44]). Compared with the negative effect of warming for non-legumes, there was a positive or neutral effect on LN and A~net~ for legume species ([Supplementary Material S3](#SM1){ref-type="supplementary-material"}). Contrary to the expectations, stomatal conductance remained unchanged under warming, thus highlighting the key roles of biochemical and nutritional limitations on the negative responses of net photosynthesis to warming treatment. The response of G~s~ to global warming is critical for modeling ecosystem and landscape-scale water fluxes and CO~2~ exchange. The ratio of respiration to photosynthesis (*R/P*) has been used to express the proportion of consumed to fixed C of plants (Atkin et al., [@B5]; Campbell et al., [@B9]) and shown to be enhanced (Danby and Hik, [@B13]; Wan et al., [@B64]), suppressed (Jochum et al., [@B29]), or maintained (He et al., [@B24]) by experimental warming. The ratio of R~d~/A~net~ was increased at warming conditions (effect size is 0.3623, *n* = 275) in this study, suggesting that the respiration was more affected and a greater proportion of fixed C was consumed, implying a decline of the net amount of C fixed by leaves by warming, at least in the controlled experiments. Ecophysiological traits responses of terrestrial plants to increased temperature varied among plant functional types with different photosynthetic pathways (PFTs; Wang et al., [@B66]; Jing et al., [@B28]). Previous studies indicated that global warming had stronger effects on A~net~ of C~3~ species than C~4~ species (Wahid et al., [@B62]). In this study, the positive and negative effects of warming on R~d~ and A~net~ were greater for C~4~ species than C~3~ species, in spite of positive or neutral effects of warming on LN, SLA, and G~S~ for C~4~ and C~3~ species, respectively. The contradictory findings posed great challenges for projecting the responses and feedbacks of terrestrial ecosystems to global warming. The more disadvantaged situation for C~4~ species under warming might be associated with higher growth and treatment temperature applied in the experiment ([Supplementary Material S1](#SM1){ref-type="supplementary-material"}). The metabolic balance of the photosynthetic and respiratory processes under climate warming plays a critical role in regulating ecosystem carbon storage and cycling (Schimel, [@B52]; King et al., [@B31]). Warming stimulated A~net~ in woody but suppressed it in herbaceous plants ([Supplementary Material S4](#SM1){ref-type="supplementary-material"}). The positive effect of warming on A~net~ for woody species was unrelated to either G~s~ or LN, as G~s~ and LN both were decreased under warming treatments ([Supplementary Material S4](#SM1){ref-type="supplementary-material"}). The results from this study were similar to the trend reported for trees showing a lower percentage decrease in G~s~ compared to herbaceous species (Wang et al., [@B66]). Warming had a positive effect on G~s~ and LN for crops, while a negative effect for non-crops ([Supplementary Material S5](#SM1){ref-type="supplementary-material"}). The changes in G~s~ at warming treatment may alter leaf temperature and result in a change in latent heat loss through evaporation, which may further affect net carbon balance (Warren et al., [@B69]). Warming could influence vegetation dynamics and ecosystem structure through shifting competitive interactions among different functional groups in natural or agricultural systems. Therefore, knowledge of photosynthetic and stomatal responses to increased temperature of different PFTs instead of species will facilitate the prediction of terrestrial C- and water- cycle feedback to climate warming. Ecophysiological trait responses of terrestrial plants to increased temperature varied among warming treatments of differing durations. The physiological acclimation can lead to smaller enhancements of plant photosynthesis and respiration under long term warmer conditions than predicted with photosynthesis/respiration-temperature relationships (Medlyn et al., [@B39]; Dwyer et al., [@B17]; Tjoelker and Zhou, [@B59]; Gunderson et al., [@B21]). The thermal acclimation of R~d~ could minimize the effects of climate warming on C loss via plant respiration (Gifford, [@B20]; Ziska and Bunce, [@B76]; Loveys et al., [@B37]) and mitigate the positive feedback between climate change and atmospheric CO~2~ (King et al., [@B31]; Atkin et al., [@B4]). The findings in this meta-analysis indicated that the negative effect of warming treatment on A~net~ and LN and the positive effect on R~d~ were more evident under \>1 year warming treatment and the trend was confirmed for both C~3~ and C~4~ species ([Figure 4](#F4){ref-type="fig"}), which contrasted to other studies showing significant declines in the photosynthetic and/or respiratory response with increasing exposure time, a thermal acclimation to warming (Hikosaka et al., [@B25]; Gunderson et al., [@B21]). Potential confounding factors must be accounted in the meta-analysis because many studies were conducted under variable conditions and targeted on different species. In this analysis, studies in which plants were grown under other environmental stresses such as drought, low nutrients, light deficiency or elevated ozone were excluded. In addition to the variation caused by plant functional types and treatment duration, different experimental facilities could be responsible for the responses of different PFTs (Cheesman and Klaus, [@B10]; Rehmani et al., [@B43]). This study mainly focused on the effects of pot size (\<10 L vs. \>10 L) and experimental settings (in-door vs. out-door) on plant ecophysiological responses. Warming had a negative and positive effect on LN and G~s~ when plants were grown at outdoor and in-door settings, respectively. Pot size significantly altered the responses of R~d~, LN and SLA to warming treatments. Warming had a negative effect on R~d~ for plants grown at \<10 L pots, while a positive effect at \>10 L pot. For both LN and G~s~, warming had a negative effect for plants grown at \>10 L pots, while a neutral effect at \<10 L pot. We were expecting that warming would have a more negative effect on LN and G~s~ in smaller pots or in-door settings considering that below-ground growth would be more constrained and thus limited the nutrients and water supply to the aboveground growth (Walters and Reich, [@B63]; Climent et al., [@B11]), the analysis indicated that this was true only when experiment duration was longer than 1 year when negative effects of warming was more evident for plants grown at \<10 L pots. Warming treatment duration had a significant interactive effect with experimental settings (in-door vs. outdoor) on R~d~ and A~net~. Long-term warming had a negative effect on R~d~ for in-door and on A~net~ for outdoor experimental settings. The findings in this meta-analysis indicated that the negative effect of warming treatment on A~net~ and LN and the positive effect on R~d~ were more evident under \>1 year warming treatment and the trend was confirmed for both C~3~ and C~4~ species ([Figure 4](#F4){ref-type="fig"}). The negative effect of warming on R~d~ could be related to the higher treatment temperature applied at the in-door settings ([Supplementary Material S2](#SM1){ref-type="supplementary-material"}). Temperature conditions in which plants live may be another possible reason for the contradictory findings (Rustad et al., [@B49]). The discrepancy of the response of A~net~ and R~d~ to warming treatment under different experimental settings provided difficulty in parameterizing ecosystem models and raised concerns in proper experimental designs when dealing with climate change questions. The intensities of temperature treatment also had a significant impact on most of the parameters that were investigated in the study. The effect size of A~net~, R~d~, LN, and SLA responded to temperature increase in a quadratic relationship. Consistent with the results discussed before, the peak value of the ecophysiological traits of A~net~, R~d~, and LN occurred at temperatures higher than the ambient. Plant physiological responses to warming may also depend on the temperature regime they are grown at. Studies often report a positive response to warming in Rubisco carboxylation, photosynthesis, and growth in cool-climate species but reduced growth and carbon gain in species that exist in warm low-latitude climates (Way and Oren, [@B70]; Crous et al., [@B12]). Conclusion {#s5} ========== Overall, we found that warming treatment of different durations and settings had different impacts on plant ecophysiological traits and the responses varied significantly among different plant functional types. Warming stimulated R~d~ and SLA but suppressed A~net~ and LN and the effect varied among different PFTs and experimental designs. The positive and negative effects of warming on R~d~ and A~net~, were greater for C~4~ than C~3~ species, in spite of the positive or neutral effects of warming on LN, SLA, and G~S~ for C~4~ and C~3~ species, respectively. The findings in this meta-analysis also indicated that the negative effect of warming treatment on A~net~ and LN and the positive effect on R~d~ were more evident under \>1 year warming treatment and the trend was confirmed for both C~3~ and C~4~ species. Negative effect of warming was more evident for plants grown at \<10 L pots only when experiment duration was longer than 1 year. The magnitude of temperature treatment did have an impact on most of the parameters that were investigated in the study. The functional type specific response patterns of plant traits to warming are critical for obtaining credible predictions of the changes in food production, carbon sequestration and climate regulation. This result also highlights the need for cautiously selecting parameter values in forecasting ecosystem function changes in future climate regimes, evaluating much more broadly what can and cannot be learned from experimental studies and designing controlled experiments to realistically reflecting ecosystems responses to future global warming. Data Availability {#s6} ================= All datasets for this study are included in the manuscript and the [supplementary files](#s8){ref-type="supplementary-material"}. Author Contributions {#s7} ==================== DW and ZY conceived and wrote the paper. The rest of the authors helped collection data and ran data analysis. Conflict of Interest Statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Funding for this research was provided by The National Natural Science Foundation of China (31500503 & 31770485), Nanjing University of Information Science and Technology (2013r115), Jiangsu Distinguished Professor Scholarship, Jiangsu six talent peaks (R2016L15), Jiangsu Natural Science Foundation (BK20150894), and the Jiangsu Overseas Research & Training Program for University Prominent Young & Middle-aged Teachers and Presidents through DW. Supplementary Material {#s8} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpls.2019.00957/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Iker Aranjuelo, Institute of Agrobiotechnology, Superior Council of Scientific Investigations, Spain [^2]: Reviewed by: Lina Fusaro, Sapienza University of Rome, Italy; Elisa Pellegrini, University of Pisa, Italy [^3]: This article was submitted to Plant Abiotic Stress, a section of the journal Frontiers in Plant Science

Kuntz‐Melcavage KL, Gahagan KDC, Fracasso MR, Marsteller JA. Enhancing knowledge of authorization requests through registry development. Learn Health Sys. 2018;2:e10050 10.1002/lrh2.10050 All authors are employed by Johns Hopkins HealthCare LLC, which is the company that funded this study. 1. INTRODUCTION {#lrh210050-sec-0001} =============== Health insurers are a crucial part of health care systems and have a responsibility to provide wise medical coverage decisions for their members. Coverage decisions are determined by regulations, benefits, and medical policies, with each insurer developing its own company‐specific policies. It is essential that medical coverage policies be written in a manner that promotes the best health for insured members while aligning with the mission of a company. Sometimes, there is insufficient evidence on a new procedure or drug to know for sure which patients will benefit from it most and which may not benefit at all. As part of an academic health system, our organization strives to maximize the knowledge obtained from data we collect. Current research in the industry relies heavily on administrative claims[1](#lrh210050-bib-0001){ref-type="ref"}, [2](#lrh210050-bib-0002){ref-type="ref"} that are useful for providing data about use and costs. However, there are other sources of information that can be informative to medical policy decisions that remain largely untapped. Claims data do not contain information that is submitted as part of pre‐authorization requests (such as clinical measurements and laboratory values). Additionally, data from claims do not contain information about members who were denied authorization for a procedure. A source of data beyond claims can be helpful for providing a more complete profile of a member population seeking services. We are currently working to expand knowledge at the juncture between the medical‐policy setting and the subset of members who are affected by coverage decisions. The mechanism we have chosen to increase our knowledge in this area is the development of registries. The need for a registry to record member‐level information surrounding certain procedures first became apparent, as the medical policy unit at our company planned a revision to the coverage policy for bariatric surgery. Anecdotal stories and opinions about bariatric surgery were prevalent, but no one was able to provide information derived from an aggregate collection of authorization requests. To address the uncertainty about details contained in requests for bariatric surgery, we developed a registry of bariatric surgery requests containing data to inform future policy decisions. The registry records information from medical records, submitted as part of the pre‐authorization process for bariatric surgery. This paper describes the development of the registry and workflow to incorporate the registry into the process of medical pre‐authorization review. Results from a descriptive analysis of the population seeking surgery are presented, along with an examination of characteristics of members whose pre‐authorization requests for bariatric surgery are initially denied. Two examples of how the registry has broadened our knowledge about the pre‐authorization requests received by our company are presented. This report is relevant for health systems because it demonstrates an approach for gathering meaningful data from an operational process. Data are the foundation for information, which ultimately leads to knowledge. Analyzing data from authorization requests has allowed our company to be more aware of approval patterns for requests and has provided the basis for us to learn more about factors that influence medical decisions. 2. METHOD {#lrh210050-sec-0002} ========= This study was reviewed and approved by the Institutional Review Board of our institution (IRB number IRB00084964). 2.1. Registry design {#lrh210050-sec-0003} -------------------- The initial step in developing a registry was to determine the proper electronic platform to be used. We chose to use Microsoft Access (Microsoft Corp, Redmond, Washington) because of 2 features considered to be of great importance: It is compatible with a variety of data analysis programs, and it provides the ability to develop a user‐friendly electronic data entry form. Both of these considerations were important for ensuring the functionality and usefulness of the registry. After deciding upon the software to be used for registry development, we identified the data fields to be included in the registry. Variables that are important when making authorization decisions were identified through literature reviews and interviews with medical directors. Health conditions that are associated with candidates for bariatric surgery were identified from literature, while conversations with medical directors revealed laboratory measures that were worthy of recording. A data dictionary was developed to provide a framework for the fields to be developed in the database (Table [1](#lrh210050-tbl-0001){ref-type="table"}). Each of the fields is able to be populated from information included in authorization requests. Because of the limited demographic information included in pre‐authorization requests, the registry similarly contains limited demographic information. ###### Kuntz‐Melcavage et al Name Description ------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------- Database identifier Unique number assigned to database entry Study identifier Number assigned to member seeking pre‐authorization for bariatric surgery Member gender Gender of member seeking pre‐authorization Age Age of member seeking pre‐authorization for bariatric surgery at time of application Line of business Line of business of which member seeking pre‐authorization is a member Date of letter seeking authorization Date that appears on the cover letter that was sent to seek pre‐authorization for bariatric surgery Date of disenrollment If member disenrolls during study period; date on which member disenrolled Date of baseline measurements Date on which baseline measurements of weight and BMI were made Baseline weight Baseline weight (rounded to nearest whole number) Baseline BMI Baseline BMI (body mass index) Date of 1‐month postbaseline measurements Date on which measurements of weight and BMI 1 month after baseline were made One‐month postbaseline weight Weight at timepoint approximately 1 month after baseline measurement (rounded to nearest whole number) One‐month postbaseline BMI BMI at timepoint approximately 1 month after baseline Date of 2‐month postbaseline measurements Date on which measurements of weight and BMI 2 months after baseline were made Two‐month postbaseline weight Weight at timepoint approximately 2 months after baseline (rounded to nearest whole number) Two‐month postbaseline BMI BMI at timepoint approximately 2 months after baseline Date of 3‐month postbaseline measurements Date on which measurements of weight and BMI 3 months after baseline were made Three‐month postbaseline weight Weight at timepoint approximately 3 months after baseline (rounded to nearest whole number) Three‐month postbaseline BMI BMI at timepoint approximately 3 months after baseline Date of 4‐month postbaseline measurements Date on which measurements of weight and BMI 4 months after baseline were made Four‐month postbaseline weight Weight at timepoint approximately 4 months after baseline (rounded to nearest whole number) Four‐month postbaseline BMI BMI at timepoint approximately 4 months after baseline Date of 5‐month postbaseline measurements Date on which measurements of weight and BMI 5 months after baseline were made Five‐month postbaseline weight Weight at timepoint approximately 5 months after baseline (rounded to nearest whole number) Five‐month postbaseline BMI BMI at timepoint approximately 5 months after baseline Date of 6‐month postbaseline measurements Date on which measurements of weight and BMI 6 months after baseline were made Six‐month postbaseline weight Weight at timepoint approximately 6 months after baseline (rounded to nearest whole number) Six‐month postbaseline BMI BMI at timepoint approximately 6 months after baseline Date of 7‐month postbaseline measurements Date on which measurements of weight and BMI 7 months after baseline were made Seven‐month postbaseline weight Weight at timepoint approximately 7 months after baseline (rounded to nearest whole number) Seven‐month postbaseline BMI BMI at timepoint approximately 7 months after baseline Comorbidities Comorbidity noted in letter requesting authorization for bariatric surgery Comorbidities 2 Comorbidity noted in letter requesting authorization for bariatric surgery Comorbidites 3 Comorbidity noted in letter requesting authorization for bariatric surgery Comorbidities 4 Comorbidity noted in letter requesting authorization for bariatric surgery Comorbidities 5 Comorbidity noted in letter requesting authorization for bariatric surgery TSH laboratory results Laboratory results for thyroid‐stimulating hormone Total cholesterol Laboratory results for total cholesterol level Triglycerides Laboratory results for triglycerides level LDL cholesterol Laboratory results for LDL cholesterol level HbA1c laboratory results Laboratory results for hemoglobin A1c level Complete? Has all available information been entered to the comorbidities/lab form? *H*. *pylori* laboratory results Laboratory results for *Helicobacter pylori* test Type of surgery requested The type of surgery that is requested (open or laparoscopic) Initial determination Initial authorization determination (yes or no) If no, action taken after initial denial If initial authorization determination was *no*, what action was taken after denial (peer‐to‐peer discussion, appeal 1, appeal 2, none) Did reapplication occur? Did reapplication occur? Outcome of reapplication? Approval or denial of reapplication submission? Did a peer‐to‐peer consult occur? Whether a peer‐to‐peer consult occurred Was initial determination overturned or upheld? Following action taken after the initial denial, was the initial determination overturned or upheld? (overturned, upheld, NA) Outcome of peer‐to‐peer consult If a peer‐to‐peer consult occurred, was the outcome of the consult an approval of bariatric procedure or a denial of bariatric procedure? Did a first‐level review occur? Whether a first‐level review occurred Outcome of first‐level review If a first‐level review occurred, was the outcome of the review an approval of bariatric procedure or a denial of bariatric procedure? Did a second‐level review occur? Whether a second‐level review occurred Outcome of second‐level review If a second‐level review occurred, was the outcome of the review an approval of bariatric procedure or a denial of bariatric procedure? Did an outside consult occur? Whether an outside consult occurred Outcome of outside consult If an outside consult occurred, was the outcome of the consult an approval of bariatric procedure or a denial of bariatric procedure? Date of bariatric surgery Date on which bariatric surgery was performed Type of bariatric surgery Type of bariatric surgery that was performed Facility Facility at which bariatric surgery was performed Name of surgeon Name of surgeon who performed bariatric surgery Abbreviation: LDL, low‐density lipoprotein; NA, not applicable. Development of the database within access occurred through a 2‐step process. The first step involved creating a data field for each metric that was planned to be recorded, and the second step focused on generating electronic data entry forms. Because of the large quantity of data fields included in the registry, multiple data entry forms were created with each form containing a limited set of related variables. For example, a form containing only comorbidity and laboratory data was developed. A total of 4 forms were developed to facilitate entry of data from pre‐authorization forms: member information, comorbidities and labs, pre‐op weight and body mass index, and authorization and surgery. The data fields on each of these forms were a mixture of drop‐down boxes and numerical fields. Data fields allowing free‐text entry were limited because of the challenges free text presents when performing data analysis. Confidentiality was an important consideration when designing the registry. The purpose of the registry is to provide population‐level knowledge of presurgical measures of individuals seeking bariatric surgery, and therefore, we minimized the entry of protected health information from medical records to the registry. Each medical record is assigned a randomized identification number (termed study ID), and a separate key has been created containing a link between study ID numbers and member identification numbers. The linking file is necessary to enable future studies that will combine information from authorization requests and claims records. 2.2. Incorporation into workflow {#lrh210050-sec-0004} -------------------------------- With the registry developed, we established an operational process to capture the information. Given the business needs of the company and our responsibilities to members, it was necessary to develop a process that was minimally intrusive on the day‐to‐day operations for reviewing authorization requests so as not to delay determinations. An effective process for collecting data for the registry was accomplished via the incorporation of a single additional step into the typical workflow surrounding pre‐authorization requests for bariatric surgery (Figure [1](#lrh210050-fig-0001){ref-type="fig"}). The standard business protocol for reviewing requests that are made for bariatric surgery is as follows: A member\'s pre‐authorization forms are sent to our company. The information is transmitted to a medical review nurse. Once a medical review nurse determines that all required information has been submitted, the request is reviewed by a medical director. The medical director reviews the pre‐authorization request and makes a decision about whether the surgery is authorized for payment by the insurance plan to which the member belongs. The medical review nurse is informed of the decision and takes the appropriate steps to inform the member\'s physician of the decision regarding authorization for bariatric surgery. {#lrh210050-fig-0001} The bariatric registry has been incorporated into the standard business workflow via the addition of a step in which the medical review nurse sends copies of the pre‐authorization forms, including the medical director\'s decision to approve or deny the request, to a medical policy research associate (MPRA). The MPRA assigns a randomized study ID to each pre‐authorization packet and enters relevant information to the registry. These additional steps occur without any disruption to the normal time frame for pre‐authorization request reviews. The MPRA is employed by our company to perform a variety of research tasks, and the time required for entry of data to the registry has not had an adverse impact on that person\'s overall productivity. 2.3. Analysis {#lrh210050-sec-0005} ------------- Analyses of data collected from pre‐authorization forms have been performed using SAS version 9.4 (SAS Institute Inc, Cary, North Carolina). Figure [2](#lrh210050-fig-0002){ref-type="fig"} provides an overview of questions that have been answered using data stored in the registry. The ovals in the figure depict the statistical approach for addressing each question. {#lrh210050-fig-0002} A descriptive analysis was performed on the entire population of cases in the registry. Variables examined in this analysis were gender, age, body mass index, approval decision, and selected comorbidities known to be prevalent among individuals who seek bariatric surgery. Means or percentages were calculated, as appropriate, for each variable. Logistic regression was performed to examine the relationship between review decision and 7 variables reported on authorization applications. The initial review decision (approval vs denial) served as the outcome variable in our statistical model, and predictor variables were age, gender, hypertension, diabetes, obstructive sleep apnea, back pain, arthritis, and gastroesophageal reflux disease (GERD). The amount of applications that originated from each insurance product managed by our company was calculated and compared with the proportion of total enrollees in each insurance product. A *z* test was used to examine the significance of the difference between proportions. The presence of a *Helicobacter pylori* test in applications and the resulting approval decision was examined for requests originating from one specific facility. Proportions were calculated to examine approval decisions (approved or denied) and test result (positive or negative). A chi‐square test of independence examined whether a difference in approval rates exists between applicants with positive test results vs applicants with negative test results. 3. RESULTS {#lrh210050-sec-0006} ========== Using the new workflow to enter data into the bariatric registry, all authorization requests for bariatric surgery that were submitted to our company beginning July 1, 2013, have been entered, and the registry is continuously updated. Descriptive statistics about the population seeking bariatric surgery are provided below. 3.1. Population overview {#lrh210050-sec-0007} ------------------------ The data analysis described in this publication is based on data collected from pre‐authorization requests submitted during the first 40 months of the registry\'s existence. During this time, 504 members applied for bariatric surgery pre‐authorization, and of those, 87% were female. The age distribution of members seeking bariatric surgery was normally distributed (Figure [3](#lrh210050-fig-0003){ref-type="fig"}), with an average age of 41 years. The minimum age was 13 years, and the maximum age was 65 years. Only 2 applicants were under the age of 18, both of whom were female. {#lrh210050-fig-0003} We examined 3 of our company\'s insurance products: a commercial plan, a Medicaid plan, and a plan for military dependents and retirees. Of the members in the registry, 19% (n = 97) are members of the commercial plan, 71% (n = 359) are members of the Medicaid plan, and 10% (n = 48) are members of the military plan (Figure [4](#lrh210050-fig-0004){ref-type="fig"}). {#lrh210050-fig-0004} The prevalence of selected comorbidities that were reported on letters requesting prior authorizations for surgery is depicted in Table [2](#lrh210050-tbl-0002){ref-type="table"}. Hypertension was reported in nearly half of applicants, and obstructive sleep apnea was reported in 37% of applicants. Arthritis was the least reported comorbidity among those recorded in the registry. ###### Kuntz‐Melcavage et al Comorbidity Percent --------------------------------- --------- Hypertension 49 Diabetes 27 Obstructive sleep apnea 37 Back pain 16 Arthritis 6 Gastroesophageal reflux disease 23 3.2. Logistic regression {#lrh210050-sec-0008} ------------------------ Reported presence of GERD decreased the odds of denial of authorization requests for bariatric surgery (odds ratio, 0.573; 95% CI, 0.331‐0.993; *P* = .0470). None of the other comorbidities examined were deemed to be significant because the confidence interval crossed 1. 3.3. Origin of surgery requests {#lrh210050-sec-0009} ------------------------------- Most requests for bariatric surgery originate from Medicaid members. Total member enrollment is also greatest for Medicaid. The percent of total requests for bariatric surgery that can be attributed to each insurance product is displayed in the *requests* column of Table [3](#lrh210050-tbl-0003){ref-type="table"}. The percent of total enrollment that can be attributed to each insurance product is displayed in the *enrollment* column of Table [3](#lrh210050-tbl-0003){ref-type="table"}. For both the commercial product and the Medicaid product, no significant difference existed between the proportion of members who apply for bariatric surgery and the proportion of total enrolled members who belong to each product (*P* = .514 and *P* = .493). A *z* test was not performed for the military health plan because less than 30 applications exist in the registry for this insurance product. ###### Kuntz‐Melcavage et al Line of Business Requests Enrollment ------------------ ---------- ------------ Commercial .17 .15 Medicaid .76 .73 Military .07 .12 3.4. Relationship between approval decisions and laboratory results {#lrh210050-sec-0010} ------------------------------------------------------------------- Total requests from one facility in the registry were examined to determine the relationship between approval decisions and results of *H. pylori* tests. A total of 209 requests were identified, and of those, 19% (n = 39) did not contain *H. pylori* test results. Among requests containing *H. pylori* test results, no significant difference in approval decisions was detected (Figure [5](#lrh210050-fig-0005){ref-type="fig"}). {#lrh210050-fig-0005} 4. DISCUSSION {#lrh210050-sec-0011} ============= Identifying a need for data regarding the population of members who seek bariatric surgery led us to develop a data collection workflow and a registry that has enhanced our knowledge about this population. Because operational efficiency is important in the health insurance industry, research may not always be prioritized and may be impeded by limited data collection or insufficient analytic tools. The present report demonstrates a simple addition to a standard medical review workflow that has greatly increased knowledge about a specific population of insured members. The registry has provided us with a tool to answer questions about members seeking bariatric surgery. We are now able to provide an accurate description of the members who seek bariatric surgery, and we can examine data to discover trends that may affect authorization decisions. For example, an association was detected for GERD in the results from the logistic regression analysis. It is possible that when reviewers notice GERD on an authorization application, they are inclined to approve the request. Further study of this observation could include interviews with medical directors to determine whether they feel GERD places members at a higher need for bariatric surgery. Discovering approval patterns that have previously gone unnoticed is a benefit of capturing pre‐authorization data in a registry. One could envision an automated decision tool resulting from the collection and analysis of authorization data. Some assumptions have been supported by data that have been captured, while others have been called into question. We can now verify that most applications for bariatric surgery originate from female members. In a previous study of bariatric surgery, Fuchs et al reported that in a population of nearly 190 000 patients who underwent bariatric surgery, 80% of the patients were female.[3](#lrh210050-bib-0003){ref-type="ref"} We can now confidently state that the prevalence of females who seek bariatric surgery in our insured population agrees with other reports of groups seeking bariatric surgery. The comparison of proportions of members in each insurance product who seek bariatric surgery has been enlightening and resulted in questions that require further study. The predominance of applications originating from Medicaid members has historically been attributed to the socioeconomic barriers faced by many who are eligible for government assistance. Challenges that include limited access to healthy food, a negative environment, and detrimental behavior are more prevalent among population that is eligible for Medicaid.[4](#lrh210050-bib-0004){ref-type="ref"}, [5](#lrh210050-bib-0005){ref-type="ref"} Surprisingly, we observed that the proportions of applications for bariatric surgery are similar between Medicaid members and members enrolled in commercial insurance. We plan to delve further into this finding by examining approval rates among the populations and overall health status of applicants at the time of application. Our report of laboratory results and approval decisions is an example of the use of a registry for enhancing knowledge among the medical policy unit. This report was prompted by a provider inquiry regarding the necessity of *H*. *pylori* test results for making approval decisions. Our analysis determined that there is not a significant difference in approval decisions depending on whether *H*. *pylori* test results are positive or negative. Because medical review involves a more complete assessment than merely considering 1 test result, one cannot conclude that the test is unnecessary for the approval decision process. However, based on this knowledge, a policy revision has been made that eliminates the need for a specific test but includes a more general statement regarding the necessity of gastrointestinal health. The importance of registries for monitoring health care quality has been documented, along with the observation that there is currently ample opportunity to develop medical registries in the United States.[6](#lrh210050-bib-0006){ref-type="ref"} Our work demonstrates the feasibility of developing condition‐specific registries using a modest amount of resources and having a minimal impact on established business workflows. A prominent health insurer within United States has previously reported on the positive impact of its disease registries on patient safety, quality improvement, cost‐effectiveness, and research.[7](#lrh210050-bib-0007){ref-type="ref"} Because these areas are places where most health care companies seek to improve, it is reasonable to conclude that registries can be instrumental for helping health insurers achieve their goals. While large insurers may have the analytics in place to examine data from subpopulations, there is room for smaller companies to improve their abilities to remain informed about specifics groups of members. The process of extracting and recording data from all pre‐authorization requests generates data that are not available via claims records because requests that are denied will never appear as claims. Just as this tool has helped us ensure that our medical policy is relevant and fair for the subpopulation to which it applies, future registries will be developed as a means to improve the quality of our medical policies.

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi:10.1111/hae.14070

INTRODUCTION ============ Globally, the incidence rate of malaria has decreased from 72 to 59 cases *per* 1,000 at risk inhabitants between 2010 and 2017, a 18% reduction^[@B1]^ . An estimated 219 million cases of malaria occurred worldwide in 2017 compared to 239 million cases in 2010. However, the estimates for 2015-2017 were almost similar, suggesting no progress in reducing the malaria burden during these last three years. In 2017, malaria resulted in an estimated 435,000 deaths globally compared to 607,000 deaths in 2010, with a 28% decrease in mortality^[@B1]^ . This reduction was attributed to the availability of highly effective antimalarial drugs and long-lasting insecticide nets (LLINs), as well as a mutual investment to provide treatment and preventive measures to the people in need^[@B1]\ -\ [@B3]^ . Most malaria cases were reported from the African World Health Organization (WHO) (92%), followed by the South-East Asia Region WHO (5%), and the Eastern Mediterranean Region WHO (2%). Notably, the highest numbers of malaria cases and deaths (93%) were reported from the African WHO, mostly in children under five years of age^[@B1]^ . To accelerate progress in reducing the burden of malaria, WHO endorsed the Global Technical Strategy for Malaria 2016--2030 (GTS) which set a vision to eliminate malaria in 35 countries by 2030 and in at least ten countries by 2020^[@B1]^ . Bhutan has achieved a significant reduction of malaria incidence from 1,868 indigenous cases in 2006 to only six indigenous cases in 2018^[@B4]\ -\ [@B8]^ . The geographical distribution of malaria infection has also decreased from 15 of the 20 districts in 2006 to only two districts in 2018^[@B4]\ ,\ [@B6]^ . The dramatic decline in malaria cases is believed to be due to the high coverage of LLINs, intensified surveillance and early diagnosis and treatment^[@B5]\ ,\ [@B6]\ ,\ [@B9]\ ,\ [@B10]^ . Bhutan plans to eliminate malaria by 2025^[@B9]^ . However, imported and reintroduced cases along the international border with some Indian States remain a significant concern as seven of the seven Bhutanese districts (Chukha, Dagana, Pemagatshel, Samtse, Samdrup Jongkhar, Sarpang, and Zhemgang) share porous international borders with Assam, West Bengal, Arunachal Pradesh and Sikkim, in India. Among these States, Assam and West Bengal borders have intense cross-border activity, and most cases are reported in areas bordering Assam^[@B5]\ ,\ [@B11]\ ,\ [@B12]^ . Percentages of imported malaria cases have increased from 79.7% of the total malaria cases detected in Bhutan in 2016 to 82.33% in 2017, and 88.9% in 2018^[@B4]\ ,\ [@B10]\ ,\ [@B13]^ . In Bhutan, *Plasmodium* species that cause malaria are *P. falciparum* and *P. vivax* . *Anopheles* species recorded and considered as malaria vectors in Bhutan are *An. minimus* , *An. fluviatilis, An. dirus, An. pseudowillmori* and *An. culicifacies* ^[@B13]^ . However, no studies on vectors including their ecology and behaviors have been conducted in Bhutan. The primary malaria control intervention adopted in the country includes mandatory screening for plasmodial infections in any fever case, early detection and treatment, active case finding to detect foci of transmission, community awareness and education. In addition, vector control by the universal coverage of LLINs, indoor residual spraying (IRS), clearing bushes and avoiding stagnation of water in the surrounding have also been implemented. While malaria incidence has dramatically declined, there is not much information on asymptomatic reservoirs in the country. Some evidences suggest that a significant proportion of asymptomatic reservoirs are present in both, high and low transmission settings^[@B14]\ ,\ [@B15]^ . The diagnosis of asymptomatic plasmodial infections in people living in low transmission settings cannot be made by commonly used diagnostic methods, *i.e.,* microscopic examination and rapid diagnostic tests (RDT)^[@B16]^ . To achieve malaria elimination, it is essential to ascertain the burden of asymptomatic reservoirs in the population at risk, as well as in migrant workers from malaria-endemic countries, particularly India^[@B14]\ ,\ [@B17]^ , and proactively detect and treat asymptomatic plasmodial infections with effective antimalarial drugs^[@B18]\ -\ [@B21]^ . This study aimed to estimate the prevalence of asymptomatic plasmodial infections in the population living in at risk malaria areas in Bhutan, as well as in migrant workers from India. This information is essential to support the implementation of malaria elimination strategies in pursuit of elimination by 2025^[@B9]^ . MATERIALS AND METHODS ===================== Study area and sample size -------------------------- A cross-sectional survey was conducted to determine the prevalence of asymptomatic *P. vivax* and *P. falciparum* infections targeting populations living in risk areas for malaria in seven districts of Bhutan, *i.e.,* Chukha, Dagana, Pemagatshel, Samtse, Sarpang Samdrup Jongkhar, and Zhemgang ( [Figure 1](#f01){ref-type="fig"} ). The study period coincided with the peak of the malaria season in April to May 2016. Based on the records maintained by 16 health centers in the risk areas they consisted of 6,319 households and 28,583 people, accounting for approximately 4% of the country's population. For the estimation of the prevalence of asymptomatic plasmodial infections ( *P. vivax and P. falciparum* ), catchment areas of two health centers in each of these districts located in malaria risk areas (based on the number of cases detected from 2011-2015) were intentionally selected^[@B22]^ . From each selected health center, approximately four villages (primary sampling units) were randomly selected and from each selected village, a maximum of ten households (secondary sampling units) were selected using a systematic random sampling technique (based on data maintained by VDCP and health centers). When the selected village had less than ten households, additional village(s) were randomly sampled, from the eligible list of villages in the catchment area of the selected health centers. From each household, a single member in the eligible list of household members was randomly sampled based on the inclusion criteria: (i) individual residing in a household that received LLINs distributed by the Vector-Borne Disease Control program in 2014; (ii) had not been diagnosed or treated for malaria during the last 21 days (based on the maximum incubation period^[@B23]^ ), (iii) aged ≥18 years on the date of the survey, (iv) resident in the locality for ≥ 1 years and (v) agreed to participate in the study. Figure 1- Bhutan's chiwog map (small administrative units are in dotted lines) and the district map (bold lines) shows the sampling sites (shaded in black color) to estimate the prevalence of asymptomatic malaria in the community and in migrant workers at the three border entry points (triangles). Using a 95% confidence interval (95% CI), an error of 2% and an anticipated asymptomatic prevalence of 5% (the expected prevalence of 5% was assumed although previous small-scale studies reported a prevalence of asymptomatic plasmodial infections \< 1%^[@B5]^ ). The required sample size was 457 participants. To account for the loss and missing information (approximately 10%), the sample size was rounded to 500. Taking the maximum design effect of 1.5^[@B18]^ , 750 individuals were included in the analysis using the formula N= (Zα/2)^[@B2]^ \*P (1-P)\*DEFF/ ME^[@B2]^ , where N is the sample size, Zα/2 is the critical α level, P is the anticipated asymptomatic malaria prevalence, DEFF is the design effect, and ME is the marginal error. Migrant workers were enrolled in three main entry points, *i.e.,* Phuntsholing, Gelephu and Samdrup Jongkhar. All migrant workers underwent compulsory medical examination for entry into Bhutan. To this end, there were five registered private diagnostic centers in Phuntsholing, two in Gelephu and one in Samdrup Jongkhar. The sample size required for the estimation of the prevalence of asymptomatic malaria in this group of migrant workers was estimated using the same formula (without adjusting for the design effect DEFF assuming a within-private-diagnostic-centers variance of zero). Therefore, the estimated sample size for migrant workers entering Bhutan through these entry points was 500 individuals. Since over 50% of the workers entered through Phuntsholing, and around 50% through Gelephu and Samdrup Jongkhar, 250 individuals were randomly sampled in Phuntsholing and 125 individuals in Gelephu and the same number in Samdrup Jongkhar. Approval of the study protocol was obtained from the Research Ethics Board of Health (REBH), Ministry of Health, Royal Government of Bhutan (approval No. REBH/Approval/2016/016). Written informed consent was obtained from the head of each household (HH) family. The interviewers explained the purpose, the risks and benefits of the study in the participant language and participation in the survey were voluntary. Written information on the survey was translated into Dzongkha for Bhutanese and Hindi for migrant workers and provided to all participants. Participants' demographics were collected using a structured questionnaire. Sample collection and testing ----------------------------- Venous blood samples were collected from all the subjects for the RDT and three aliquots of 50 μL were directly spotted onto filter papers (Whatman No. 3 MM, GE Healthcare UK Limited, Buckinghamshire, United Kingdom). Venous blood sample collection was chosen due to the difficulty to obtain adequate blood samples volumes by the finger-prick procedure as most farmers in the community had hard finger skins due to the type of work they do. Each filter paper (Dried Blood Spot, DBS) was air-dried at 25 ^o^C overnight, kept in a zip-lock plastic bag containing desiccants and stored at 25 ^o^C until it was sent to the laboratory and processed^[@B24]\ ,\ [@B25]^ . The RDT used in the study was the FirstSign^TM^-- ParaView-2 (Unimed, Volmolenheide, Belgium) and the test was performed according to the manufacturer's instructions. This immuno-chromatography test detects *P. falciparum* specific histidine-rich protein-2 (HRP-2) and *P. vivax* specific pLDH *.* Genomic DNA was extracted from DBS samples using the QIAamp DNA Mini Kit (QIAGEN^®^, Hilden, Germany) according to the manufacturer's instructions. The extracted DNA was stored at -20 °C until use. Nested PCR assay was carried out as previously described^[@B26]^ . DNA samples were amplified using species-specific primer pairs designed to amplify small subunit ribosomal ribonucleic acid (ssrRNA) genes of *P. falciparum* and *P. vivax.* The outer primers were rPLU 1: 5'TCAAAGATTAAGCCATGCAAGTGA3' and rPLU5: 5'CCTGTTGTTGCCTTAAACTCC3'. The primers used for nested PCR for *P. falciparum* were FAL1:5'TTAAACTGGTTTGGGAAAACCAAATATATT3' and FAL2: 5'ACACAATGAACTCAATCATGACTACCCGTC3' and for *P. vivax* VIV1: 5'CGCTTCTAGCTTAATCCACATAACTGATAC3' and VIV2:5'ACTTCCAAGCCG AAGCAAAGAAAGTCCTTA3. In brief, the primary amplification was carried out in a total volume of 25 μL containing 1X buffer, 3.5 mM MgCl~2~, 0.2 mM dNTPs, 0.2 µM primers and 1 U *Taq* DNA-polymerase, 2.0 μL of extracted DNA (25 ng), and nuclease-free water in a T100^TM^ Thermal cycler (BioRaid, Singapore). The nested amplification was carried out in a 20 μL reaction volume containing the same reaction mixture, *P. falciparum* and *P. vivax* species-specific primer, and 1 μL of the PCR product from the primary amplification. Thermal cycling conditions for the first round of amplification were as follows: initial denaturation at 94 °C for 1 min, followed by 35 cycles of 30 s at 94 °C, 60 s for 55 °C and 60 s at 72 °C, followed by a final extension of 10 min at 72 °C. The cycling conditions for the second PCR were similar, excepting for the annealing temperature that was set at 58 °C. Negative and positive controls were included in each experiment. Amplicons were separated by 2% agarose gel electrophoresis at 100 V with the molecular weight marker 100 *bp* DNA ladder (Bioline, Memphis, USA). The presence or absence of different *Plasmodium* species was confirmed by the presence of a 205 bp amplicon for *P. falciparum* and a 117 bp amplicon for *P. vivax* . Statistical analysis -------------------- All data were entered into the Epi Info™ version 7.2.0.1, and statistical analysis was performed using the Stata version 14.0 (Stata Statistical Software, Release 14: Stata Corp, College Station, TX, USA). The frequencies of asymptomatic malaria caused by the two species are summarized as proportions, with 95% confidence intervals (95% CI). RESULTS ======= Demographic characteristics of the participants ----------------------------------------------- ### Participants from malaria risk areas A total of 750 participants were enrolled from 75 villages in malaria risk areas of the seven districts of Bhutan. The participants' age ranged from 18 to 84 years, with a mean age of 45 years old. The male to female participants' ratio was 1:1.2. The participants were predominantly farmers (91.87%), and the remaining participants were students, business people, government/private employees and religious practitioners. ### Migrant workers From a total of 500 migrant workers, 473 were enrolled in five private diagnostic centers (Druk Diagnostic, Garuda Diagnostic Center, New Life Diagnostic Service, Phuntsholing Diagnostic Center, and Samdrup Jongkhar Diagnostic Center) at the three border entry points (Phuntsholing, Gelephu, and Samdrup Jongkhar). All of them were males with age ranging from 18 to 66 years (mean of 30 years). Most were from West Bengal (57.51%), followed by Assam (12.26%) and the remaining were from Bihar, Uttar Pradesh, Himachal Pradesh, Manipur and Orissa. Information on the States in India where they used to live was not obtained from a significant part of the migrant workers (27.27%). Detection of P. falciparum and P. vivax parasitemia by RDT and PCR ================================================================== Participants from malaria risk areas ------------------------------------ From the total of 750 samples collected from villagers, 11 (1.46%) samples were not tested by RDT as one of the clusters had inadequate RDT. All of the 739 samples showed negative results by RDT. Based on the PCR analysis, 2 of the 739 samples were positive (0.27%) for *P. vivax* and none for *P. falciparum.* The estimated prevalence of asymptomatic plasmodial infections in the studied areas was 0.27% (95% CI: 0.05-1.07%) ( [Table 1](#t1){ref-type="table"} ). Positive samples were obtained from Chukha district, one sample from Lingden village (32 years old male) and another sample from Kothilina villager (a 38 years old male). All positive cases were treated according to the national protocol. Table 1- Detection of *P. falciparum* and *P. vivax* infections by RDT and PCR in participants from malaria risk areas (750 cases) and migrant workers from the three border entry points (473 cases). Data are presented as numbers (percentages) and 95% confidence intervals (CI) when appropriate. Malaria risk areas ( N = 750)Migrant workers (N = 473)RDT (n = 739)PCR (n = 735)RDT (n =473)PCR (n =473)*P. falciparum*0 (0)0 (0)0 (0)0 (0)*P. vivax*0 (0)2 (0.27%) (95% CI: 0.05-1.07%)0 (0)2 (0.42%) (95% CI: 0.07 -- 1.69%) Migrant workers --------------- From the total of 473 migrant worker samples, all of them showed negative results by RDT, and only 2 samples showed positive results for *P. vivax* PCR *.* One *P. vivax* positive infection was from Assam (a 21 years old male), and another one was from West Bengal (a 27 years old male). The estimated prevalence of asymptomatic plasmodial infections in this group of migrant workers was 0.42% (95% CI: 0.07-1.69%) ( [Table 1](#t1){ref-type="table"} ). DISCUSSION ========== Asymptomatic and submicroscopic parasitemia reservoirs can sustain the continuous transmission of malaria even in low transmission settings if they are not accurately detected by a highly sensitive detection method and immediately treated by an effective antimalarial drug^[@B14]\ -\ [@B16]^ . Therefore, it is essential to address the asymptomatic and submicroscopic parasitemia in the population to achieve malaria elimination^[@B16]^ . Most countries with low transmission settings and striving to eliminate malaria, including Bhutan, have observed a large proportion of asymptomatic plasmodial infections, particularly of cases with submicroscopic parasitemia^[@B3]\ ,\ [@B20]\ ,\ [@B21]\ ,\ [@B27]^ . This study is the first report on asymptomatic plasmodial infections that represented all malaria risk areas in Bhutan^[@B5]^ . PCR was used in addition to RDT for malaria detection. This study was conducted during the pre-monsoon season (April-May 2016) coinciding with the peak of the malaria season in the country, to maximize the detection of asymptomatic reservoirs. Results suggest that the prevalence of asymptomatic and submicroscopic parasitemia cases could be very low (0.27%) in the malaria risk areas in Bhutan. The two positive cases detected were *P. vivax* infections. Although this study was conducted in 2016, we believe that the prevalence of asymptomatic plasmodial infections remains unchanged. The prevention and control program in these areas have been maintained at the same level. There was no significant political and socio-economic changes in the country. The malaria cases decreased in 2018 compared to 2016. Nevertheless, Bhutan needs to strengthen and sustain an optimal surveillance, even though the prevalence of asymptomatic plasmodial infections was very low. The following reasons support this proposal: (i) Bhutan shares open borders with endemic areas of Indian States (Assam and West Bengal), (ii) The climate conditions in these border areas include other high-risk areas with tropical and subtropical climate and (iii) there is abundance of vector mosquito populations in these areas. Results indicate that the current elimination strategies and interventions have been effective in interrupting malaria transmission in populations at risk, and asymptomatic and submicroscopic parasitemia cases are an insignificant source of malaria transmission in Bhutan. The observation is in agreement with previous reports from Sri Lanka^[@B28]\ ,\ [@B29]^ , Iran^[@B30]^ , Soloman Islands^[@B31]^ and Paraguay^[@B32]^ . All these reports suggest the reduction of cases of asymptomatic and submicroscopic parasitemia malaria with decreased symptomatic malaria cases in the population at risk. The absence of asymptomatic plasmodial infections in Bhutan Has been previously reported^[@B5]^ , but the study was limited to only four risk areas of the two districts bordering the Assam State of India (Sarpang and Samdrup Jongkhar). Besides, the method used for the detection of asymptomatic plasmodial infections was the RDT, which has poor sensitivity compared to PCR^[@B3]\ ,\ [@B5]\ ,\ [@B19]\ -\ [@B21]^ . A low prevalence of asymptomatic plasmodial infections was expected in Bhutan due to its low malaria transmission sustained over a long period. This could be attributed to the country's relentless effort in ensuring preventive measures such as the provision of LLINs and IRS, as well as the policy to perform the compulsory screening of all fever cases in malaria risk areas for early detection and treatment of the disease. Bhutan launched active case finding strategies targeting asymptomatic plasmodial infections as early as 2012. For the preparation towards the malaria elimination, the program included reactive case detection methods for the identification of parasite transmission foci, and proactive case detection methods in the screening of all migrant workers at border districts^[@B10]\ ,\ [@B33]\ -\ [@B36]^ . The LLIN's coverage exceeded 90% since 2006, and successive surveys on indicators have also reported very high coverages^[@B13]\ ,\ [@B34]\ ,\ [@B35]^ . Malaria cases were also dramatically reduced from 465 cases in 2010 to 104 cases in 2015, with 65% of imported cases. Notably, 54% of the imported cases were non-national^[@B13]^ . Among the indigenous cases, 74% were residing close to the Bhutan-India border, suggesting an active transmission in pockets of border areas due to the continuous movement of people across the border^[@B12]^ . The genetic analyses have also suggested an active transmission in border areas, as the same genotype was found in both, Bhutanese and non-national (Indian) people^[@B37]^ . Most of the migrant workers were from Indian States of West Bengal (57.51%), followed by Assam (12.26%). The remaining cases were from Bihar, Uttar Pradesh, Himachal Pradesh, Manipur and Orissa. These figures were in agreement with previous reports on the high number of imported cases from West Bengal, Assam and Bihar in the past years due to its proximity^[@B38]\ -\ [@B40]^ and the majority of migrant workers coming from these two regions. All of these States are endemic to malaria and responsible for the highest number of malaria cases and deaths in India^[@B11]\ ,\ [@B41]^ . High prevalences of asymptomatic malaria cases have also been reported in a tribal population in Eastern and North East India^[@B14]\ ,\ [@B41]^ . Contrary to our expectations, despite the high incidence of malaria cases in these States, a low prevalence of asymptomatic malaria was found in these imported cases (0.48%). Mass malaria screening of migrant workers has been routinely implemented, annually, by the Vector-Borne Disease Control Program (VDCP) in different hydropower project sites in which very low prevalences of asymptomatic malaria cases were reported in Bhutan. Although a regular screening process at the point of entry is mandatory, the current RDT may not be sufficient to prevent asymptomatic carriers from entering Bhutan indicating the need for more regular and strengthened surveillance systems to monitor this population. In 2014, VDCP screened 4,689 migrant workers in various hydropower projects sites also including Bhutanese people and only three *P. vivax* cases (0.06%) were detected. In 2015, only one *P. vivax* and one *P. falciparum* (0.05%) cases were detected from 3,754 migrant workers screened with no asymptomatic case detected from 1,534 people screened in 2016^[@B39]\ ,\ [@B40]^ . However, the method used for the screening at these sites was microscopy and RDTs. Based on PCR analysis, absence of asymptomatic malaria cases has also been reported in the malaria-endemic area of Assam State^[@B42]^ . The low prevalence among migrant workers may also be attributed to the intensified malaria prevention and control activities in India^[@B43]^ . The forest and illegal population dwelling inhabiting the international border may be greatly affected by malaria, but they may not be among the screened migrant workers. Only migrant workers moving for longer periods in Bhutan, but not daily workers who move on a daily basis were screened for malaria. According to the annual report of National Vector Borne Disease Control Program of India, the number of malaria cases decreased from 55,793 in 2012 to 26,240 in 2016 in West Bengal (a 47% reduction) and from 29,999 in 2012 to 6,948 in 2016 in Assam (a 23% reduction)^[@B44]^ . The endemicity greatly varies within areas in these States^[@B44]^ . Based on the results of this study, the asymptomatic reservoirs may not be a significant concern or barrier to the Bhutan's malaria elimination effort. However, a continuous high LLIN's coverage, special attention to the screening of migrant workers at border points and periodic mass screening at hydropower projects using more sensitive malaria detection methods such as RDT and PCR is required to eliminate any potential source of infection represented by both symptomatic and asymptomatic cases. To achieve Bhutan effort of malaria elimination in 2020, the current strategy of screening migrant workers at entry points and hydropower projects should be continued along with supportive control measures. Low-density infections are common in asymptomatic individuals. Both, microscopy and RDT usually miss detection of infections when parasite densities are low (\<10 parasites/µL). One of the most used RDT, is based on the capture of the parasite antigen by monoclonal antibodies ( *P. falciparum* specific HRP-2 and *P. vivax* specific pLDH). The test is very convenient, but is expensive. In addition, it can result in false-positives with variations of sensitivity and specificity. Nested PCR targeting the ssrRNA genes is the most sensitive and specific method, but it is also expensive, requiring an amplification infrastructure, equipment and reagents. Besides, it is a relatively sophisticated and time-consuming procedure, which may not be applicable to malaria diagnosis in remote areas. On the other hand, the detection limit of nested-PCR and nRT-PCR are as low as 5-10 and 22 parasites/µL of blood, respectively^[@B45]\ ,\ [@B46]^ . Therefore, it is an important technique to investigate low parasite density in asymptomatic malaria patients. The study was supported by the Vector Borne Disease Control Program under the Department of Public Health, Ministry of Health through Global Fund project. KN and WC were supported by Thammasat University, Thailand (Center of Excellence in Pharmacology and Molecular Biology of Malaria and Cholangiocarcinoma), National Research Council of Thailand. [^1]: CONFLICT OF INTERESTS The authors declare that there are no conflict of interests

1. Introduction {#sec1} =============== *Histoplasma capsulatum* (*H. capsulatum*) is a dimorphic fungus, which is endemic to North America (Central United States; Ohio-Mississippi valley) and Latin America \[[@B1]--[@B3]\]. It is associated with exposure to bat caves and avian droppings \[[@B4], [@B5]\].*H. Capsulatum* peritonitis should be suspected in patients on peritoneal dialysis (PD) from endemic areas who have the potential for exposure. Review of data from United States Renal Data Systems from 1992 to 1997 demonstrated age-adjusted incidence ratio of fungal peritonitis of 9.8 compared with general population and represents 4.5% of total peritonitis episodes in the PD population \[[@B6]\]. The vast majority of these cases are caused by*Candida* species. Mortality secondary to PD associated peritonitis is organism specific: 28% for fungi, 16% for enteric organisms, and 15% for staphylococcal species. The presence of certain additional factors in PD patients increases the risk for fungal peritonitis. Almost all published series have found an association with both recent antibacterial use and episodes of bacterial peritonitis \[[@B7]--[@B12]\]. When these series are combined, 65 percent of patients had been exposed to antibiotics within 30 days of the onset of fungal peritonitis, and 48 percent had experienced an episode of bacterial peritonitis within the same time frame. Other risk factors include emergency PD, HIV infection, abdominal surgeries, extraperitoneal fungal infections, and environmental exposures. Details of previously reported cases of*H. capsulatum* are also discussed. 2. Case Report {#sec2} ============== A 63-year-old female who was visitor from Veracruz, Mexico, presented to the emergency room with complaints of progressively worsening abdominal pain and distention for three days. She also had fever and altered mentation. Her past medical history was significant for hypertension, diabetes mellitus, hyperlipidemia, and end-stage renal disease. She had been on PD for four years and denied any recent changes in technique. She had two episodes of peritonitis in the past while in Mexico but was unaware of the details of those episodes. Her surgical history was significant for appendectomy, cholecystectomy, and tubal ligation and she denied any recent abdominal procedure. She denied smoking, alcohol intake, or use of recreational drugs. On examination, her blood pressure was 172/85 mm of Hg, pulse 88/min, oral temperature 39.5°C (103.1°F), respiratory rate 14/min, and oxygen saturation on room air 94%. She was lethargic and confused. She had abdominal distention and diffuse tenderness without any rebound or guarding. Her PD catheter exit site was clean and dry. Laboratory studies showed white blood cell count of 14.5 × 10^3^/*μ*L with 87.1% granulocytes, hemoglobin of 6.3 g/dL, and hematocrit of 18.4%. Serum chemistries showed sodium of 130 mmol/L, potassium of 2.7 mmol/L, chloride of 90 mmol/L, bicarbonate of 27 mmol/L, blood urine nitrogen of 30 mg/dL, and creatinine of 7.7 mg/dL. Her liver function tests were within normal limits. Computed tomography of abdomen and pelvis without intravenous or oral contrast showed peritoneal thickening consistent with peritonitis, and there was no evidence of perforation or obstruction ([Figure 1](#fig1){ref-type="fig"}). PD fluid analysis showed white cell count of 2173 per mm^3^ with 96% neutrophils and red blood cells of \<3000 per mm^3^. Blood and PD fluid cultures were sent, and she was empirically treated for bacterial peritonitis with intraperitoneal cefazolin and ceftazidime. PD fluid gram stain revealed budding yeast; blood and PD fluid cultures did not reveal bacterial growth. Given the high suspicion of fungal peritonitis, immediate removal of the PD catheter was discussed with the patient. She chose not to have the catheter removed, leave to Mexico, and get treated by her own nephrologist. Hence oral fluconazole was started for presumed*Candida* peritonitis. However, six days later, the fungal culture \[Mycosel Agar and Brain Heart Infusion Agar\] of the PD fluid grew*H. Capsulatum*. 3. Discussion {#sec3} ============= As previously noted, fungal peritonitis is an uncommon cause of peritonitis in PD patients. There are currently no established guidelines for the diagnosis of fungal peritonitis. The International Society of Peritoneal Dialysis (ISPD) recommends repeating PD fluid cell count at day 3 of culture negative peritonitis and employing special culture techniques for isolating uncommon organisms including fungi. \[[@B13]\]. Isolation of*Histoplasma* from culture may take up to 12 weeks but can happen as early as 1-2 weeks. Nonculture techniques available for diagnosis include molecular techniques like polymerase chain reaction, serological tests like complement fixations, and immunodiffusion tests for precipitins \[[@B14]\]. If yeast is seen on initial gram stain, then prompt antifungal treatment should be started as we did in our patient. Although the mainstay of therapy in the past has been Amphotericin B, its toxicity has frequently precluded its use \[[@B15]\]. Experience with the newer imidazoles/triazoles and flucytosine suggests that these agents are well tolerated and efficacious \[[@B13]\]. Regarding the removal of PD catheter, ISPD guidelines recommend prompt removal once fungal infection is identified. Additionally, an appropriate antifungal agent should be continued to 2 weeks after removing the catheter \[[@B13]\]. There are isolated reports of treating fungal peritonitis without removing the catheter, but with varying degrees of success. However, this should be considered an option only if patient\'s medical condition precludes removal of the catheter \[[@B14]\]. Since peritonitis due to*H. capsulatum* is extremely rare, there are no established guidelines for treatment of this condition. The six reported cases were treated with 6--12-month course of Itraconazole with or without Amphotericin B as noted in [Table 1](#tab1){ref-type="table"} \[[@B16]--[@B20]\]. 4. Conclusion {#sec4} ============= Although*H. capsulatum* peritonitis is extremely rare, morbidity and mortality associated with it are high. Diagnosis requires high degree of suspicion based on geography and occupation. In such patients, if yeast is seen on gram stain it would be prudent to remove the PD catheter and consider Itraconazole as first choice of therapy for extended period of time. Consent ======= The patient left to Mexico immediately after discharge. Her son-in-law, who has the power of attorney over her healthcare issues, provided informed consent over the telephone for publishing this case report. Conflicts of Interest ===================== The authors attest that they do not have any conflicts of interest pertaining to this case report. {#fig1} ###### Management of reported cases of PD patients with *Histoplasma* peritonitis. Cases Treatment regimen Treatment Catheter removed ------------------------------- -------------------------------- --------------------- ------------------ Case 1 \[[@B16]\] Oral Itraconazole 12 months Yes Case 2 & 3 \[[@B17], [@B18]\] Amphotericin B Unknown Yes Case 4 \[[@B18], [@B19]\] Fluconazole and Amphotericin B 1 month and 10 days No Case 5 \[[@B14]\] Oral Itraconazole 6 months Yes Case 6 \[[@B20]\] Oral Itraconazole 12 months Yes [^1]: Academic Editor: Rumeyza Kazancioglu

Introduction ============ One key characteristic of fast-growing solid tumors is the development of intratumoral hypoxia, which promotes malignant tumor progression. Cancer cell migration and invasion play important roles in the metastatic cascade. Cell migration is the transition process from the local, noninvasive confined tumor cells to the migrating, metastatic cancer cells when the cells obtain the ability to dissociate from intracellular adhesions and become motile [@B1],[@B2]. Invasion, on the other hand, is the process by which malignant cells move through the basement membrane and gain access to blood vessels and lymphatic channels. As cancer cell migration and invasion are essential prerequisites for tumor metastasis, identifying agents that effectively block hypoxia-induced cancer migration and invasion would promote development of effective anti-cancer therapeutic approaches. Hypoxia-inducible factor-1 (HIF-1), a master transcriptional factor in response to hypoxia, activates a group of downstream genes that promote tumor malignant progression. HIF-1 is composed of an inducible subunit, HIF-1α and a constitutively expressed subunit, HIF-1ß [@B3]. Heterodimerization of HIF-1α and HIF-1ß is required to form the transcriptionally active HIF-1 that binds to the hypoxic response elements within the promoter regions of target genes [@B4]. HIF-1α degrades quickly under normal oxygen tension [@B5] but it is inducible and stabilized under low oxygen tension (hypoxia) [@B6]. Increased HIF-1α levels have been shown to correlate with decreased patient survival in many cancers including breast cancer (BCa) [@B7]. The activation of HIF-1α stimulates a group of HIF-1α-regulated genes including vascular endothelial growth factor (VEGF) [@B3],[@B8]. As a result, the cells are converted towards malignant progression. HIF-1 activity is mainly dependent on the level of HIF-1α protein, the inducible and regulatory subunit of the HIF-1 dimer complex [@B9], [@B10]. We have previously shown that p16, a tumor suppressor gene and cyclin D kinase inhibitor and a negative cell cycle regulator [@B11], can neutralize the transactivation ability of HIF-1α on its target gene VEGF [@B12]. In this study, we evaluated whether hypoxia stimulates breast cancer migration and invasion, and whether ectopic expression of p16 is capable of modulating hypoxia-mediated cell migration and invasion. We also evaluated whether cocultured stromal cells (fibroblasts) promotes BCa cell invasion and whether p16 inhibits that effect, as well as whether HIF-1α of fibroblasts specifically has a role in promoting cocultured BCa cell invasiveness. Materials and Methods ===================== Cell lines, cell culture conditions, and reagents ------------------------------------------------- Human breast cancer (BCa) cell line MDA-MB-231 was obtained from American Type Culture Collection (ATCC, Manassas, VA). Human breast cancer cell lines LM2, a MDA-MB-231 derivative line that has high lung metastatic ability [@B15], was a generous gift from Dr. J. Massague of Memorial Sloan-Kettering Cancer Center (NY). Murine mammary carcinoma cell line JygMC(A) was from Dr. H. Azuma of Osaka Medical College, Osaka, Japan [@B13]. Mouse MEF fibroblasts expressing wild-type (WT) and knockout (KO) HIF-1α were from Dr. T Seagroves of University of Tennessee Health Science Center (TN). MDA-MB-231 cells were grown in RPM1-1640 (Gibco, Life technologies Corporation, Grand Island, NY) with 10% fetal bovine serum (FBS) (Gibco), LM2 cells were grown in Dulbecco\'s Modified Eagle medium (DMEM) (Cellgro, Mediatech, Inc, Manassas, VA) with 10% FBS. JygMC(A) cells were grown in DMEM (Cellgro) with 10% FBS. Both MEF lines were grown in DMEM medium with 25 mM HEPES (Gibco) with 10% FBS. All cell lines were grown in medium containing 100 units/ml penicillin, and 100 μg/ml streptomycin. The cell cultures were incubated at 37°C either under normoxia (5% CO~2~, 21% O~2~) or hypoxia (5% CO~2~, 1% O~2~, balanced with N~2~) conditions. Generation and screen of stably transfected, Dox inducible (Tet-on) p16-expressing breast cancer cell lines ----------------------------------------------------------------------------------------------------------- The construction of the Tet-on p16 lentiviral system Lenti-Tet-on p16 (pLenti-Tet-p16-pgkpuro and pcFUW-rtTA3-IRES-puro) was described previously [@B12]. Breast cancer cells (MDA-MB-231, LM2, and JygMC(A), respectively) were co-transduced by pLenti-Tet-p16-pgkpuro and pcFUW-rtTA3-IRES-puro at multiplicity of infection (moi) of 10 each, together with 6 μg/ml polybrene (Millipore, Bedford, MA). The stably transfected cells were enriched in medium containing 1 μg/ml puromycin (Clontech, Mountain View, CA). The resultant stably transfected cells are named as MDA/Tet-on-p16, JygMC(A)/Tet-on-p16, and LM2/Tet-on-p16, respectively. For induction of p16 transgene expression in Tet-on-p16 stably transfected cells, 1 μg/ml doxycyline (Dox) (Clontech) was used in the medium for incubation. Immunohistochemistry -------------------- The procedure followed the method as described previously [@B12]. Briefly, for immunohistochemical (IHC) staining, cultured cells were grown on SlideFlasks with bottom detachable slides (Nalge Nunc, Naperville, IL) that could be used for IHC staining directly later. The samples (slides) were first incubated with 1% H~2~O~2~ for 30 min. The samples were incubated with first antibody against human p16 (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) for 16 h at 4^0^C, then by corresponding second antibody and the Universal Elite ABC Kit (Vector Laboratories, Inc., Burlingame, CA) according to the manufacturer\'s protocol. The reaction was visualized with DAB solution (75 mg 3,3\'-Diaminobenzidine and 30 ml 50% H~2~O~2~ in 150 ml PBS) for 3-10 min. Cell migration assay -------------------- The breast cancer MDA-MB-231 cells stably transfected with inducible Tet-on p16 (MDA/Tet-on p16) were incubated in the absence or presence of 1 μg/ml Dox for 72 h. The cells were then harvested and used for the following migration assay. The cell migration was measured by a modified Boyden\'s chamber method using BD Falcon Cell Culture Inserts incorporating polyethylene terephthalate (PET) track-etched membrane with a pore size of 8.0 μm (BD Bioscience, Belgium). The inserts were precoated on the under-surface (between upper and lower chambers) with 10 µg/ml fibronectin (Thermo Fisher Scientific) at 37^0^C for 3 h. Above mentioned harvested cells in suspension of serum-free medium (SFM) with 1% BSA were incubated at 37°C for 30 min under normoxic (5% CO~2~, 21% O~2~) or hypoxic (5% CO~2~, 1% O~2~, balanced with N~2~) conditions. Subsequently, cell suspensions were seeded into the upper chamber of an insert at a density of 10,000 cells per well, and 300 µl serum-free medium with 1% BSA was placed in the lower chamber to act as a chemoattractant in the 24-well cell culture plate. The cells were further incubated at normoxia and hypoxia (see above) at 37°C for 3 h. The inserts were then removed and nonmigrating cells remaining on the upper side of the filter were scraped off. The cells that had migrated to the lower surface of the insert were stained using Giemsa staining solution (Sigma, St. Louis, MO). After extensive washing with water, the migrated cells were counted in five different fields under a microscope at x200 magnification. Migratory activity was expressed as the number of cells (that is, the sum of total cell numbers in five randomly selected fields of view) that migrated to the lower side of the filter. *In vitro* invasion assay with commercially precoated invasion inserts ---------------------------------------------------------------------- The BCa cells were examined by the invasion assay under normoxia by using commercially available invasion inserts. For each Dox inducible BCa/Tet-on p16 cell line, the cells were either incubated in medium with or without 1 μg/ml Dox for 72 h. The cells were then harvested and used for the following *in vitro* invasion assay in the 6-well-plate BD Biocoat Matrigel Invasion Chambers (BD Biosciences Bio-Coat Matrigel Invasion Chamber) according to the manufacturer\'s procedure. Briefly, the chamber was first rehydrated with serum-free medium (SFM) for 2 hr at 37^0^C. After rehydration, the chambers were placed in the lower compartment loaded with medium containing 5% FBS. Meanwhile, the above-mentioned cells were suspended and adjusted to 1.25x10^5^ cells/ml in SFM with or without 1 μg/ml Dox. The cell suspension (2 ml or 2.5x10^5^ cells per well) was immediately added to the upper compartment of the chamber. The cells were then allowed to invade through the matrigel for 22 h at 37^0^C, and the noninvading cells were removed by scrubbing the upper surface with a wet cotton swab. The filters were stained with Diff-Quick stain kit (Dade Behring Inc., Newark, DE), drained and counted under the microscope. *In vitro* invasion assay with self-coated matrigel --------------------------------------------------- The BCa LM2 cells, which express a green fluorescent protein (GFP) gene [@B15], were transfected with Lenti-Tet-on p16 to generate LM2/Tet-on-p16 line. LM2/Tet-on-p16 cells were incubated in the absence or presence of 1 μg/ml Dox for 72 h before the experiment, and the samples with Dox induction were maintained throughout the following assay. All the cells (for simplicity, we used LM2 here referring LM2 or LM2/Tet-on-p16, and MEF referring MEF/HIF-1α WT or MEF/HIF-1α KO) were grown in SFM overnight and then harvested for the following invasion assay. The 24-well size Falcon Cell Culture Inserts (Corning Incorporated, Corning, NY) were precoated on the upper-surface (inside the inserts) with 100 µl matrigel (BD Matrigel, Bedford, MA) per insert and let them solidify at 37^0^C overnight. For experiment with BCa cells alone, cell suspension of 5x10^5^LM2 cells in 100 µl SFM was seeded into the upper chamber of the insert; for experiment with coculture of BCa and fibroblasts, cell suspension of 5x10^5^LM2 cells and 5x10^5^ MEF cells in 100 µl SFM was seeded into the upper chamber of the insert. For the lower chamber of a 24-well cell culture plate, 650 µl SFM with 5% BSA (alternatively, with 5% FBS) was placed. The cells were then incubated at either normoxic or hypoxic conditions (see above) at 37°C for 18 h. The inserts were then removed, the matrigel (containing the unpenetrated cells) in the upper chamber of the inserts was carefully wiped off by using a wet cotton-tipped applicator. The resultant inserts were placed on a clean 24-well plate, and the matrigel-penetrated/invaded BCa LM2 cells on the membrane of the inserts were measured by using the SpectraMax M2 microplate reader (Molecular Devices, Sunnyvale, CA) for counting the GFP fluorescence intensity (RFU). Because LM2 line expresses GFP, this approach would only count the invaded BCa LM2 (GFP-expressing) cells. Results ======= Hypoxia stimulates BCa cell migration and p16 inhibits this hypoxia-induced migration ------------------------------------------------------------------------------------- Cell migration is an important aspect of the tumor metastatic process. HIF-1α was implied, by others\' studies, in stimulation of cancer cell migration [@B16],[@B17] and hypoxia induces HIF-1α [@B14]. Thus, we were interested in analyzing whether hypoxia has effect on breast cancer cell migration and whether p16, a biological agent that appears to neutralize HIF-1α transcriptional activity [@B12], has ability to suppress hypoxia-mediated cancer cell migration. To accurately examine the effects of p16 on BCa cell migration, we designed an inducible p16 expression so that p16 expression and effects can be studied in the same cells with or without inducer. A lentivirus expressing inducible human wild-type (WT) p16 under the control of a Tet-on promoter (Lenti-Tet-on p16) was constructed and stably transfected into MDA-MB-231 cells. The resultant stably infected line (MDA/Tet-on-p16) demonstrated a tightly regulated and inducible p16 expression under inducer Dox treatment (+Dox, top panel) (Fig. [1](#F1){ref-type="fig"}). Importantly, MDA-MB-231 cells, like other BCa cell lines (including JygMC(A) and LM2) we have examined, do not express endogenous p16 protein in the absence of Dox (-Dox) induction (Fig. [1](#F1){ref-type="fig"}, top panel). In other words, MDA/Tet-on-p16 cells in the absence of Dox behave just as control parental MDA-MB-231 cells. MDA/Tet-on-p16 line was subsequently used to detect potential effects that p16 had on cell migration. A modified Boyden\'s chamber assay using transwell precoated with fibronectin (a component of extracellular matrix) onto the insert undersurface of the membrane was used as described in the M & M for the migration measurement. As shown in Fig. [1](#F1){ref-type="fig"} (bottom panel), MDA/Tet-on p16 cells under hypoxia had more than two-fold (\>100%) increase in cell migration than that of normoxia (compare first two columns on the left). Importantly, inducible p16 expression by Dox significantly inhibited this hypoxia-induced cell migration (column 4) with a 36.0% inhibition in comparison to cells under hypoxia without p16 expression (column 2), indicating that p16 effectively inhibits hypoxia-induced migration in MDA-MB-231 cells. In contrast, p16\'s effect on cell migration under normoxia is marginal. It should be noted that the inhibition of migration is *not* due to p16-mediated effect on cell proliferation, as the migration assay was conducted within 3 h and the MDA-MB-231 cell line\'s doubling time is 41.9 h. In addition, Dox treatment *per se* did not have any effect on migration as evidenced by Dox treatment on BCa cells stably transfected with inducible GFP (not shown). These combined results (Fig. [1](#F1){ref-type="fig"}) strongly demonstrate that hypoxia stimulates BCa cell migration and p16 significantly inhibits this hypoxia-induced migration. p16 inhibits invasiveness ability of BCa cells ---------------------------------------------- Invasion is the process by which malignant cells move through the basement membrane and gain access to blood vessels and lymphatic channels. The classic and well-established *in vitro* invasion assay aims to measure cancer cells\' ability to penetrate the matrigel (a resemble mixture of basement membrane matrix). p16\'s effect on cancer cell invasiveness has never been studied before. To investigate the effects of p16 on BCa invasive ability, we used the *in vitro* invasion assay to analyze three BCa lines\' abilities to penetrate matrigel, in the presence or absence of p16 expression, respectively. Briefly, BCa cells stably transfected with inducible Tet-on p16 were incubated with or without 1 µg/ml Dox for 3 days, and then the treated BCa cells were placed on the matrigel invasion chamber (BD Biosciences Bio-Coat Matrigel Invasion Chamber). After 22 hours incubation at 37^0^C, the noninvasive cells were carefully removed from the upper surface of the membrane, and the penetrated cells (i.e., the invasive cells) going through matrigel were counted under the microscope. We found that all three BCa cell lines had significantly reduced invasiveness in the presence of p16 (Dox induced), at a decrease of 85.8% (MDA-MB-231), 82.7% (JygMC(A)), and 80.4% (LM2), compared to their corresponding untreated (p16 non-expressing) counterparts, respectively (Fig. [2](#F2){ref-type="fig"}). Effects of surrounding fibroblasts, and p16\'s neutralization, on cocultured BCa cell invasiveness -------------------------------------------------------------------------------------------------- Emerging evidence implies that tumor-surrounding areas contribute to tumor growth and progression; however, the key attributes from tumor-surrounding cells that clearly display promotion for tumor progression as well as the relevant mechanisms of action remain unknown. To investigate effects of tumor-surrounding cells such as fibroblasts on BCa progression and invasion, we used the *in vitro* invasion assay to analyze effects of MEF fibroblasts on cocultured BCa LM2 cells\' ability to penetrate the matrigel. After 18 hours incubation at 37^0^C of the LM2 alone or cocultured LM2 and MEF on the upper chamber of the insert that were precoated with matrigel (see details of M & M section), the matrigel (containing the noninvasive cells) was carefully removed from the upper surface of the membrane, the invaded cells (i.e., the matrigel-penetrated cells) remaining on the membrane were counted by a fluorescence microplate reader. Because LM2 cells express a GFP reporter gene [@B15], the fluorescence intensity (RFU) readout on the penetrated cells would be reflected only from BCa cells, not MEFs. As shown in Fig. [3](#F3){ref-type="fig"}, by comparing to LM2 cells alone, MEF-cocultured LM2 cells had a 4.3-fold increased invasiveness, demonstrating that surrounding fibroblasts stimulated cocultured BCa cell invasion. To examine whether induced p16 could also inhibit the MEF-stimulated invasiveness, LM2/Tet-on-p16 with or without Dox induction were used for a similar assay. Consistently, p16 induction in the cocultured LM2 cells also significantly decreased invasiveness of the cocultured BCa cells (Fig. [4](#F4){ref-type="fig"}A). To exclude the possibility that Dox incubation *per se* might alter the BCa cell invasiveness, we also used parental LM2 cells cocultured with MEF for the same assay; we did not observe any effect from Dox treatment alone on the change of LM2 cell invasiveness (Fig. [4](#F4){ref-type="fig"}B). Hypoxia and cocultured fibroblasts have additive effects on stimulation of BCa cell invasion ability ---------------------------------------------------------------------------------------------------- To examine whether hypoxia stimulates BCa cell invasiveness, we analyzed the matrigel-penetrating ability of LM2 cells in the presence or absence of cocultured MEF, under hypoxia or normoxia conditions. We found that hypoxia stimulated LM2 invasive ability at either LM2 growing alone (2.7-fold increase, compare column 1 and 2, Fig. [5](#F5){ref-type="fig"}) or coculturing with MEF (1.3-fold increase, compare column 3 and 4, Fig. [5](#F5){ref-type="fig"}). Consistent with what we observed earlier, MEF significantly simulated BCa cell invasion ability, either at normoxic condition (4.4-fold increase, compare column 1 and 3, Fig. [5](#F5){ref-type="fig"}) or hypoxia (2.2-fold increase, compare column 2 and 4, Fig. [5](#F5){ref-type="fig"}). The combination of hypoxia and cocultured MEF had apparent additive effects on stimulation of BCa LM2 invasive ability (5.9-fold increase, compare column 1 and 4). HIF-1α from fibroblasts contributes to BCa cell invasion in coculture --------------------------------------------------------------------- Tumor-surrounding stromal cells (such as fibroblasts) are integral components of the tumor environment that together as an entity plays critical roles in the tumor progression. As both hypoxia and cocultured fibroblasts MEF stimulated BCa LM2 invasiveness (see above), we further investigated whether HIF-1α from tumor-surrounding fibroblasts cells plays a critical role for BCa cell invasion. We used a similar coculture setting as described above, with both HIF-1α WT MEF and HIF-1α KO MEF lines, respectively, to coculture with LM2 cells for the invasion assay. We found that HIF-1α WT MEF had a higher stimulation on cocultured LM2 invasiveness than its HIF-1α KO counterpart did (Fig. [6](#F6){ref-type="fig"}); by compared with LM2 alone, HIF-1α WT MEF had a stimulation of 10.9-fold increase of LM2 invasiveness, whereas HIF-1α KO MEF had 7.1-fold increase. These results implied that at least part (the difference of about 3.8-fold, Fig. [6](#F6){ref-type="fig"}) of the stimulation of LM2 invasion from the cocultured MEF was from HIF-1α effect of the surrounding MEF cells. Discussion ========== Despite the fact that p16 has been studied extensively on its anti-tumor cell proliferation properties and related function as a negative cell-cycle regulator, the link between p16 and hypoxia-stimulated tumor progression remains unclear. Specifically, exploration of p16\'s effects on cancer cell migration and invasion has barely begun. In this report, we investigated novel properties of p16\'s functions on its anti-migration and anti-invasion abilities of BCa cells. We found that p16 is capable of inhibiting hypoxia-induced cell migration and invasion of BCa cells, and suppressing cocultured fibroblast-stimulated invasiveness of BCa cells. We have previously reported that p16 downregulates VEGF gene expression and BCa angiogenesis by neutralizing HIF-1α transactivation on VEGF pathway [@B12],[@B14]. Consistent with our finding, a recent publication [@B18] revealed that p16 appears to repress the migration/invasion abilities of breast stromal fibroblasts and their VEGF-dependent angiogenesis through inhibition of Akt, a component of PI3K/Akt signaling pathway that regulates HIF-1α [@B19],[@B20]. Moreover, silencing HIF-1α in malignant gliomas effectively inhibited cell migration and invasion under hypoxic environment, suggesting that HIF-1α expression induced by hypoxia is an essential event in the activation of glioma cell motility and invasion [@B21]. Furthermore, a recently identified HIF-1α-direct target gene, HEF1 (human enhancer of filamentation 1), was shown to mediate hypoxia-induced migration of colorectal cancer cells [@B22]. HEF1 was originally found as a scaffolding protein but recent data implicated it also acts as a key player in cancer cell migration and invasion [@B22]-[@B25]. Elevated levels of HEF1 were found in various human cancers including BCa [@B26] and its expression proportionally correlates to tumor grade [@B27] and metastasis ability [@B28]. Our unpublished results revealed that p16 inhibited hypoxia/HIF-1α stimulated HEF1 promoter transactivation (not shown). One possibility is that p16 may block the axis of "Hypoxia-HIF-1α-HEF1-migration/invasion" by neutralizing HIF-1α\'s transactivation on HEF1 gene. We are in the process to further elucidate the mechanism of action of p16-mediated inhibition on BCa cell migration and invasion. Although several studies suggested that tumor-surrounding microenvironment contribute to tumor growth and progression, the key attributes from tumor-surrounding cells that clearly display promotion for tumor progression have yet to be specified. While many studies have reported that HIF-1α protein in cancer cells plays an important role in cancer progression and various aspects of the tumor metastatic cascade, no study has yet reported on whether HIF-1α expression from tumor-surrounding cells contributes to tumor progression. We demonstrate here, for the first time, that HIF-1α from cocultured fibroblasts plays a critical role in promoting BCa cell invasion (Fig. [6](#F6){ref-type="fig"}). Consistent with our finding that HIF-1α from tumor-surrounding fibroblasts plays a critical role in promoting tumor cell invasion (Fig. [6](#F6){ref-type="fig"}), our recent data, in a 3-D culture system containing BCa 4T1 cells cocultured with HIF-1α WT or KO fibroblasts, revealed that HIF-1α from fibroblasts plays a critical role under hypoxia for the 3-D BCa cells\' protrusion formation (not shown). Protrusion formation is an established marker for invasive, aggressively malignant tumor type that start to invade the surrounding matrix in 3-D culture [@B29],[@B30]. The increased invasiveness of BCa may be partially due to the increased secreted growth factors (such as VEGF that is upregulated by HIF-1α) in the tumor microenvironment from surrounding HIF-1α-WT fibroblasts. The development of intratumoral hypoxia is a common characteristic for fast-growing solid tumors including breast cancer. The intratumoral hypoxia activates HIF-1α in the tumor cells and its surrounding stromal cells, which in turn stimulate a group of HIF-1α-downstream genes responsible for tumor malignant progression including migration and invasion. Current chemotherapy for BCa treatment, including targeting the HIF-1α pathway, is not ideal due to toxic side effects of chemoagents and cancer resistance to them. One of the more promising approaches is the development of biologically based therapies to thwart the progression of metastatic disease [@B31]. Specifically, a biological therapeutic approach that targets hypoxia-induced malignancy and directly neutralizes HIF-1α activity and its downstream genes mediating oncogenic pathways greatly deserves investigation. This study validates p16 as an effective biological agent to suppress hypoxia-promoted migration and invasion in BCa cells. The results of this study suggest that elevation and restoration of p16 in cancer microenvironment, either by gene therapy or pharmacology approaches, should provide a therapeutic mean to efficiently suppress BCa malignant progression. This research project was supported in part by NIH grant (CA107162) (YL) and University of Tennessee Health Science Center Bridge Funding. Breast cancer : BCa Doxycycline : Dox green fluorescent protein : GFP hypoxia-inducible factor-1alpha : HIF-1α immunohistochemistry : IHC knockout : KO tetracycline-induced regulated promoter : Tet-on promoter vascular endothelial growth factor : VEGF wild-type : WT. {#F1} {#F2} {#F3} {#F4} {#F5} {#F6} [^1]: Competing Interests: The authors have declared that no competing interest exists.

The first case of coronavirus disease 2019 (COVID-19) in Indonesia was reported on March 2, 2020.[@bib1] Since then, the number of people who are infected with SARS-CoV-2 has grown exponentially, a pattern that can usually be observed in an uninhibited outbreak.[@bib2] ^,^ [@bib3] Some experts have stated that the number will continue to soar in the subsequent few weeks if no significant intervention is performed.[@bib4] The government then announced a public health emergency on March 31, 2020.[@bib5] Following the public health emergency announcement, the Indonesian government imposed a temporary ban on all foreign travelers and also published several other regulations.[@bib6] One major regulation was a guideline for large-scale social restriction (Permenkes No. 9 Tahun 2020; Pedoman Pembatasan Sosial Bersakala Besar). The local governments with a high number of cases or deaths in their regions were allowed to impose such restriction with approval from the Indonesian Ministry of Health.[@bib7] Such measures included banning mass gatherings, closing schools and offices, curbing nonessential businesses, and banning public transport.[@bib7] Furthermore, the Indonesian Ministry of Health selected several hospitals as the referral hospitals to handle the emerging infectious disease (KH.01.07/MENKES/169/2020). Dr Sardjito General Hospital was one of the selected hospitals located in the Special Region of Yogyakarta.[@bib8] The hospital, which is also the university hospital for Universitas Gadjah Mada, was chosen because of its full-range intensive care unit and 2 isolation wards for airborne infectious diseases.[@bib9] Finally, to accommodate more patients with COVID-19, the hospital also remodeled a ward as a new isolation ward. Compared with other countries located outside the equatorial zone, the number of COVID-19 cases has been relatively small in Indonesia.[@bib10] However, the outbreak is adding extra burdens on the health care system worldwide, including Indonesia.[@bib11] Neurosurgical service is one of the systems affected.[@bib12] Several articles have discussed how neurosurgical services should be managed during this difficult time[@bib13] ^,^ [@bib14]; however, there is no specific report on neurosurgical services in developing countries. This article focuses on the neurosurgical service in Dr. Sardjito General Hospital, Yogyakarta, Indonesia, as a representative of low- and middle-income countries. Moreover, although the number of COVID-19 cases has remained rather low in Yogyakarta, we believe that hospital preparedness and contingency plans for a pandemic should be formulated, especially for neurosurgery cases, to manage the potential development of this pandemic and possible future outbreaks. Here we review what has been done by related stakeholders to curb the impact of the disease. We also describe our hospital\'s policy for performing neurosurgical services during the COVID-19 outbreak and explore what can be corrected in the future, especially in the setting of low- and middle-income countries. Methods {#sec1} ======= We collected information on the effect of the outbreak in Indonesia from the COVID-19 online database created by the Indonesian Ministry of Health. The government selected the National Institute of Health Research and Development (Badan Penelitian dan Pengembangan Kesehatan Kementerian Kesehatan \[Balitbangkes\]) as COVID-19 national referral laboratory. In addition, the government also selected several regional laboratories as COVID-19 diagnostic laboratories.[@bib15] All facilities had to meet the Biosafety Level 2 standard and have reverse-transcription polymerase chain reaction (RT-PCR) capabilities.[@bib16] By the middle of April 2020, the Indonesian government had tested more than 30,000 samples (i.e., approximately 130 tests per 1 million population).[@bib17] Furthermore, we gathered information from online COVID-19 records generated by the Yogyakarta government to assess the effect of the outbreak in the Special Region of Yogyakarta. By April 13, 2020, the Yogyakarta administration had performed 516 tests, for a rate of approximately 144 tests per 1 million people.[@bib18] The data on confirmed COVID-19 cases from all laboratories throughout Indonesia were then sent to Balitbangkes to be validated and verified.[@bib15] All confirmed cases, which included inpatients, recovered patients, and deaths, had been confirmed by RT-PCR.[@bib15] Finally, the data were sent to the Public Health Emergency Operation Center of the Indonesian Ministry of Health. The data on the number of neurosurgical procedures performed between February 2, 2020, and April 10, 2020, at Dr. Sardjito General Hospital were obtained from the hospital\'s Division of Neurosurgery. Emergency cases were defined as head tumor and cerebrovascular problems with acute deterioration, shunt malfunction, cauda equina syndrome, acute obstructive hydrocephalus, or head and spinal cord trauma. In addition, urgent cases such as non--rapid-onset hydrocephalus, head tumor (with urgency depending on history), or problems with recent onset of upper motor neuron symptoms or signs (e.g., cervical or thoracic myelopathy) were included in the elective surgical planning. Preoperative procedures during the pandemic, including tests to detect COVID-19, are described in the Discussion section. We included data from the previous months to help the reader differentiate the number of surgical procedures performed before and during the outbreak. Moreover, we compared the types of neurosurgical procedures performed at our center before and after the announcement of the first COVID-19 case in Yogyakarta. Finally, we also analyzed the changes in the number of outpatients visiting Dr. Sardjito General Hospital\'s neurosurgical clinic. Results {#sec2} ======= COVID-19 in Indonesia {#sec2.1} --------------------- The number of confirmed cases in Indonesia had skyrocketed in the last few weeks. By the middle of April, the government had tested more than 30,000 samples (i.e., approximately 130 tests per 1 million population) and found more than 4800 confirmed cases, with 459 fatalities ([Figure 1](#fig1){ref-type="fig"} ).[@bib17] ^,^ [@bib19] The highest number of death was in the capital city, Jakarta, which had 2335 infected cases and 241 deaths at the midpoint of April 2020.[@bib20] Moreover, compared with other nations, Indonesia had a considerably higher case fatality rate (i.e., 9.56%).[@bib19] On the other hand, the mortality rate was rather low at roughly 1 death per 588,000 people.[@bib19] Figure 1The confirmed cases in 34 provinces in Indonesia as of April 14, 2020. DKI Jakarta and Special Region of Yogyakarta are highlighted owing to their status as the capital city of Indonesia and the province in which our hospital is located, respectively. COVID-19 in the Special Region of Yogyakarta {#sec2.2} -------------------------------------------- The infection rate in Yogyakarta was not as high as that seen in Jakarta. The first case was detected on March 13, 2020, in a 3-year-old child who had a history of a family trip to Depok, West Java. As of April 13, 2020, there were 55 confirmed cases, including 6 deaths.[@bib18] Sleman, where our hospital is located, was the region with the highest number of positive cases (i.e., 29 cases). Moreover, Yogyakarta has not been declared as an area with local transmission.[@bib21] Nonetheless, it is strongly suggested that each stakeholder remain vigilant. The 4 Phases of the Pandemic at Dr. Sardjito Hospital {#sec2.3} ----------------------------------------------------- We divided our experience during this period into 4 phases to ease the understanding of the nature of the pandemic ([Figure 2](#fig2){ref-type="fig"} ). Phase 1 (March 2--13, 2020) was the period of indigency. There were confirmed cases already in Indonesia during this period; however, knowledge of the disease was limited at that time. As a result, there was no clear diagnosis, and inadequate equipment to manage it. Phase 2 is defined as the period between the initial detection of the outbreak in Yogyakarta and when the government relaxes the strict regulations regarding the outbreak. Early in this phase, the number of cases outside Yogyakarta had soared, and the Indonesian Ministry of Health and the Indonesian Medical Association had instituted several measures in an effort to curb the impact.[@bib22] ^,^ [@bib23] In phase 3, the transition phase, the infection rate has subsided but the end of the outbreak has not been declared. Finally, phase 4 is defined as the period after the outbreak resolved. At the time of this writing, we have not yet entered the last 2 phases; we described them here to explain our expectations of the end of the pandemic.Figure 2Summary of our experiences during the outbreak, described in a four-phase model, in Dr. Sardjito General Hospital. Note that Phase 3 and 4 have not yet passed by our institution. Therefore, the purpose of stating those two phases is to explain our expectations. Number of Neurosurgical Procedures Performed at Dr. Sardjito General Hospital {#sec2.4} ----------------------------------------------------------------------------- Our center\'s staff comprises 6 consultant neurosurgeons and 23 residents capable of performing both emergency and elective procedures.[@bib8] In the last 9 weeks before the first confirmed case in Indonesia, an average of 4 emergency operations were performed each week. However, early in phase 2, there was a drop in the number of emergency operations, to an average of 2.4 per week ([Figure 3](#fig3){ref-type="fig"} ).Figure 3The amount of elective (*solid line*) and emergency (*dashed line*) neurosurgery operations from February 2, 2020 to April 18, 2020 at Dr. Sardjito General Hospital. The data from February 2020 is presented to illustrate the average number of surgical procedures during typical period in Dr. Sardjito General Hospital. The black arrow indicates the time when the Indonesian first case of COVID-19 was announced (the beginning of Phase 1). *Black arrowhead* denotes the date when the first COVID-19 case in Yogyakarta was revealed (Phase 2). Furthermore, an average of 16 elective operations per week were performed during phase 1 ([Figure 3](#fig3){ref-type="fig"}). However, early in phase 2, the number decreased to roughly 9 operations weekly. This was lower than the average of 11.78 elective surgical procedures weekly performed in the last 9 weeks. Tumor resection accounted for the highest number of operative procedures, with 18 cases. In contrast, trauma surgery accounted for the fewest cases, with only 4 procedures ([Figure 4](#fig4){ref-type="fig"} ).Figure 4The number of neurosurgical procedures in Dr. Sardjito General Hospital based on the operative categories during March 2020. Number of Appointments in the Hospital\'s Neurosurgery Outpatient Clinic {#sec2.5} ------------------------------------------------------------------------ The pattern of the number of outpatient visits was similar to that observed in the number of surgical procedures. In the last 8 weeks before phase 1, we had an average of 93.38 outpatients each week ([Figure 5](#fig5){ref-type="fig"} ). However, since early phase 2, there was a slight reduction in outpatient numbers, and since March 20, 2020, the decline has accelerated. Furthermore, we had 12 residents who were eligible (based on year of residency) to provide care in the outpatient clinic. In phase 1, we allocated 4--6 residents each day to provide care in the outpatient clinic. However, in phase 2, we reduced the number of residents to just 2 per day owing to the reduced number of outpatients.Figure 5The amount of patient who came to Dr. Sardjito General Hospital neurosurgery outpatient clinic. The data from January 2020 is presented to illustrate the average number of outpatients during typical period in Dr. Sardjito General Hospital. The *black arrow* indicates the time when the Indonesian first case of COVID-19 was announced (the beginning of Phase 1). *Black arrowhead* denotes the date when the first COVID-19 case in Yogyakarta was revealed (Phase 2). Discussion {#sec3} ========== The COVID-19 Test in Indonesia: The Iceberg Phenomenon {#sec3.1} ------------------------------------------------------ The COVID-19 pandemic in Indonesia is rather challenging. We have a high case-fatality rate yet a low mortality rate, which might reflect the "iceberg" phenomenon. Some experts believe that it is associated with Indonesia\'s low testing rate.[@bib24] Although the number of tests increased in phase 2 compared with phase 1, Indonesia remained among the countries with a low testing rate.[@bib25] Furthermore, evidence shows a high level of SARS-CoV-2 replication in the upper respiratory tract even before symptoms appear.[@bib26] ^,^ [@bib27] Consequently, people who are infected with the virus yet do not have any signs or symptoms may possibly spread the virus to others.[@bib28] This might have significant epidemiologic repercussions, especially in countries with a low testing rate.[@bib19] ^,^ [@bib24] In such countries, asymptomatic carriers cannot be widely detected, which in turn might increase the rate of transmission.[@bib29] Therefore, increasing the effort to test as many people as possible appears to be an essential strategy for reducing the impact of the pandemic.[@bib30] Preoperative Preparation During the COVID-19 Pandemic {#sec3.2} ----------------------------------------------------- In phase 2, we have started to use screening tests for elective and emergency patients to detect possible COVID-19 infection in surgical candidates ([Figure 2](#fig2){ref-type="fig"}). Each test will include a patient history (e.g., previous travel, place of residence, any closed contact with a suspected COVID-19--positive individual), laboratory tests (increased neutrophil-to-lymphocyte ratio and C-reactive protein, decreased lymphocyte count), and chest radiography suggesting COVID-19 infection (i.e., multifocal airspace opacities in the lower lung areas, perihilar and upper lobe abnormalities in the later stage).[@bib31], [@bib32], [@bib33] If the initial tests suggest suspicion for COVID-19, a rapid test and chest computed tomography scan can be used to establish the diagnosis. The rapid test measures the reactivity of SARS-CoV-2 antibody in patients\' blood (i.e., IgG and IgM anti--SARS-CoV-2). For emergency patients, the rapid test can be directly ordered by the neurosurgeons without consulting with the COVID-19 team (i.e., the team assigned to provide specific management for patients with suspected COVID-19). A positive rapid test prompts RT-PCR analysis, the gold standard for diagnosing COVID-19.[@bib34] Adjustment in Surgical Practices and Recent Trends at the Outpatient Clinic {#sec3.3} --------------------------------------------------------------------------- Our status as a tertiary referral hospital results in a high patient load.[@bib8] We have a long list of patients queueing up to undergo operative procedures. In some cases, a tumor patient can wait for up to 6 months for surgery. Consequently, we decided not to eliminate elective surgical procedures during the pandemic. We were aware that this practice is not in line with the recommendation published by the Indonesian Society of Neurological Surgeons suggesting that hospitals postpone all elective procedures.[@bib35] We were afraid that such a policy would lengthen waiting periods and lead to worsening of patients\' conditions, which might further increase the patient load in the future. Later, it became clear that our policy was aligned with a recommendation from Burke et al.,[@bib13] who suggested that eliminating elective surgery is not compulsory until a hospital needs substantial aid from external institutions. Nonetheless, early in phase 2 we reduced the number of elective surgical procedures ([Figure 3](#fig3){ref-type="fig"}). One reason for this reduction was the hospital\'s policy of allocating only 2 spots in the intensive care unit for postoperative neurosurgical patients. This was meant to prepare for a surge of COVID-19 critical patients. Burke et al.[@bib13] recommended that a hospital reduce the number of elective surgeries by 25% when it has a total of 10--99 community cases or 7--16 COVID-19 inpatients. By the time we reduced the number of elective procedures, we had 55 cases in Yogyakarta and 3 COVID-19 inpatients. Another significant issue affecting the number of elective procedures was the poor availability of personal protective equipment (PPE) in our hospital. It is recommended that surgeons who are performing an operative procedure for a suspected patient or confirmed cases use level-3 PPE, including N95 masks or equivalent, coverall gowns, boots, face shields, and other standard surgical protective gear.[@bib36] Unfortunately, the hospital was struggling to provide sufficient PPE, especially in the early phase of the outbreak, a problem that also occurred globally.[@bib37] This issue started to improve in the second week of April 2020. The government and related stakeholders increased their efforts to meet the demand. In addition, there was support from nongovernmental organization and individuals. The improved PPE availability increased the safety of the procedure, which might explain the increased number of elective surgical procedures after the second week of April 2020. As for the emergency cases, the decline could also be seen early in phase 2 ([Figure 3](#fig3){ref-type="fig"}), possibly caused, at least in part, by the government\'s "stay at home" recommendation. This recommendation may cause citizens to hesitate to leave their homes, leading to a reduction in the number of traffic accidents, a cause of emergency neurosurgery.[@bib38] Finally, such hesitation also might inhibit the desire to visit the outpatient clinic, resulting in the drop in outpatient visits ([Figure 5](#fig5){ref-type="fig"}). Consequently, the working hours of staff and residents were also reduced. Importance of a Contingency Plan for the Next Possible Pandemic {#sec3.4} --------------------------------------------------------------- Based on a World Health Organization guideline, every hospital should develop an integrated hospital emergency risk management program. This program should include continuous risk assessment of future threats, a multihazard early warning system, and emergency simulations to prepare for an outbreak.[@bib39] One example is the creation of clear hospital zoning. By treating different groups of patients in a separate zone, the risk of transmission can be reduced.[@bib39] However, there is no exclusive zone dedicated to treating related patients in our hospital. All COVID-19 clinics and wards are not located in the same zone. In phase 2, we also treated all surgical patients in the same building because of limited screening resources. This could result in increased transmission if the prevailing procedure resulted in failure to detect positive cases. Furthermore, a good signage system can also be applied to promote physical distancing and reduce the transmission rate. Therefore, hospital management must clearly plan for the development of a more appropriate infrastructure for facing future pandemics. Conclusion {#sec4} ========== COVID-19 is having a deleterious impact on the stability of global public health, including neurosurgical services. Our hospital in Yogyakarta, Indonesia has been affected by the pandemic, especially in the early phase. Several factors, such as the government\'s slow response, the inadequacy of PPE, and low testing rate, have hindered the management of neurosurgical patients during COVID-19 pandemic. To help us understand the extent of the impact, we used a 4-phase model of the outbreak. In phase 2, when the first case was detected in Yogyakarta, there was a decline in the numbers of both emergency and elective surgical procedures. The number of outpatients was also decreased. Using this model, we also expected to produce a clear strategy for improving the management of neurosurgical patients during the pandemic by reducing the number of surgical procedures and advocating for an adequate supply of PPE to stakeholders. In the future, a comprehensive plan can enhance both utilization and safety of the neurosurgical staff. CRediT authorship contribution statement {#sec5} ======================================== **Wiryawan Manusubroto:** Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing - original draft, Writing - review & editing. **Adiguno Suryo Wicaksono:** Conceptualization, Formal analysis, Methodology, Project administration, Writing - original draft. **Daniel Agriva Tamba:** Conceptualization, Data curation, Investigation, Software, Visualization, Writing - original draft. **Paulus Sudiharto:** Supervision, Validation, Writing - review & editing. **Handoyo Pramusinto:** Formal analysis, Resources. **Rachmat Andi Hartanto:** Resources. **Endro Basuki:** Supervision, Validation, Writing - review & editing. Conflict of interest statement: The authors declare that the article content was composed in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

INTRODUCTION ============ Chlorhexidine is an antiseptic and disinfectant used against a broad of bacteria, viruses and fungi \[[@B1]\]. Since its introduction in 1954, it is used in the hospital settings for medical and surgical products and widely in over-the-counter products \[[@B1][@B2]\]. Many health professionals are unaware of its presence in different products, so it is often a 'hidden' allergen. The most common allergic reactions described to chlorhexidine are delayed reactions (type IV hypersensitivity), T cell mediated, and occur after exposure to the antiseptic for topical use. Contact dermatitis is the most frequent manifestation \[[@B2][@B3][@B4]\]. Immediate reactions (type I hypersensitivity), have also been reported, but much less frequently, and symptoms can range from urticaria to anaphylaxis with a risk of cardiorespiratory arrest and death \[[@B2][@B3][@B4]\]. It has not been described cross reactivity between chlorhexidine and other antiseptic agents \[[@B1]\]. CASE REPORT =========== A 75-year-old male with hypertension receiving beta-blocker and bladder cancer underwent transurethral tumor resection in 2014. Surveillance postsurgical cystoscopy under local anesthesia was performed every 6 months. During the 2nd procedure he developed generalized cutaneous pruritus with no other symptoms with spontaneous resolution after one hour. This reaction was interpreted as allergy to cefoxitin and it was recommended to avoid 2nd generation cephalosporins. Twenty minutes after the 4th cystoscopy, he developed generalized urticaria, oropharyngeal tightening, dyspnea, hypotension (75/40 mmHg) and loss of consciousness with cardiorespiratory arrest. Cardiopulmonary resuscitation was initiated immediately with endovenous administration of adrenaline (1 mg), clemastine (2 mg) and hydrocortisone (200 mg) and orotracheal intubation with invasive ventilation. The patient recovered over the next 2 hours and was extubated on the same day. The patient was referred to the immunoallergology outpatient clinic and a complete workup was performed. Local disinfection and anesthesia were performed with iodopovidone (Betadine, manufacturer, city, country) and lidocaine + chlorhexidine gel (Optilube, manufacturer, city, country). Prophylactic antibiotic therapy was performed only in 2nd procedure (cefoxitin) and ortho-phthalaldehyde (Cidex, manufacturer, city, country) was not used as cystoscope disinfectant. The allergologic investigation revealed negative skin prick test (SPT) to iodopovidone and latex, and negative cutaneous tests (standard concentration \[[@B5]\] to PPL, MDM, amoxicillin, penicillin, cefoxitin). Specific IgE (sIgE) available (latex, penicillin, amoxicillin) were negative. Provocation tests to lidocaine and cefoxitin were negative. SPT to chlorhexidine (2%) was strongly positive (11 mm × 10 mm wheal), with a positive sIgE - 4 kU/L (normal value: \<0.35 kU/L). [Table 1](#T1){ref-type="table"} summarizes the allergologic workup. ###### Allergologic workup carried out in our immunoallergology outpatient clinic  Reagent Skin prick test Intradermal test (standard ENDA concentration) \[[@B5]\] Specific IgE (RV \<0.35 kU/L) Challenge ------------------- ----------------- ---------------------------------------------------------- ------------------------------- --------------- ----------------- Antiseptic agents Iodopovidone Negative Not advised Not available Tolerated Chlorhexidine Positive Not advised 4 kU/L Contraindicated Local anesthetics Lidocaine Negative Not advised Not available Negative (SC) Antibiotics PPL and MDM Negatives Negative \- \- Amoxicillin Negative Negative Negative Not performed Penicillin Negative Negative Negative Not performed Cefaclor Negative Negative Negative Not performed Cefoxitin Negative Negative Not available Negative (IV) Cefazolin Negative Negative Not available Not performed Cefuroxime Negative Negative Not available Not performed Other Latex Negative \- Negative Tolerated ENDA, European Network for Drug Allergy; RV, reference value; SC, subcutaneous; PPL, ; MDM, ; IV, intravenous. We also performed a basophil activation test (BAT) using chlorhexidine digluconate 20% (1062 mg/mL) at 0.05%, 0.005%, 0.005%, and 0.00005% \[[@B6]\]. The basophil population was identified as HLA-DR-CD123+ CD203c+ cells and activation by CD63 expression. BAT was positive at 0.005%, 0.0005%, and 0.00005% with activation of 5.02%, 8.58%, and 11.9% and stimulation index of 3.22, 5.5, and 7.63 respectively ([Fig. 1](#F1){ref-type="fig"}). {#F1} The diagnosis of severe allergic reaction to chlorohexidine was confirmed. The patient was advised to avoid products containing chlorhexidine. Subsequent cystoscopy was uneventful using lidocaine gel as local anesthetic and iodopovidone as disinfectant. Moreover, he was informed to be aware of chlorhexidine as a component of over the counter products and the need to avoid them. DISCUSSION ========== The first case of anaphylaxis to chlorhexidine has been reported in 1984 in Japan \[[@B1][@B3]\]. Although rare, the number of clinical case reports of anaphylaxis (type I hypersensitivity) to this antiseptic is increasing. Odedra et al. \[[@B1]\] published that from 1994 to 2013, 65 case reports of chlorhexidine-related anaphylaxis were diagnosed. The majority was among surgical patients (urology and cardiothoracic) \[[@B6]\]. From 1984 to 2014, 36 cases of perioperative anaphylaxis to chlorhexidine were published \[[@B2]\]. Most reactions have been reported after application of chlorhexidine to damaged skin surfaces (wounds, burns, surgical incision); and to mucous membranes (urethra, eyes, nose) or after insertion of medical devices (central venous catheters, CVC) impregnated with chlorhexidine \[[@B4]\]. The prevalence of perioperative anaphylaxis range from 0.05%--2% and is increasing \[[@B2]\]. True incidence attributed to chlorhexidine is unknown, with several authors suggesting that is rare, but some studies referring incidences ranging from 5.5% to 8.8% \[[@B2]\]. Sharp et al. (Australia, 2016) \[[@B2]\] in a review to chlorhexidine-induced anaphylaxis in surgical patient (total of 68 anaphylactic reactions) showed that most frequent cases occur due to the presence of chlorhexidine in urinary catheter lubricant (n = 30 \[44.12%\]), CVC (n=24 \[35.29%\]) and topical solutions (n=11 \[16.18%\]). It appears to occur more frequently in men with mean age of 58 years, previously reporting a mild cutaneous reaction on chlorhexidine exposure \[[@B1]\]. Patients rarely have history of atopic disease. The clinical presentation is variable. In most cases patients developed erythematous rash/urticarial at the time of reaction and hypotension, with some presenting cardiorespiratory arrest \[[@B1][@B2]\]. Bronchospasm is rarely reported \[[@B1][@B2]\]. Our patient was older than the mean presented, however the reaction occurred during a cystoscopy. This procedure and the severity of the symptoms were similar to the most commonly described. To our knowledge, in the last five years (2014--2018), a total of 24 cases of chlorhexidine-related anaphylaxis were published ([Table 2](#T2){ref-type="table"}). The male gender is the most affected (83%). Mean age was 51 ± 15 years (range, 3--78 years) in agreement with what has already been described. The majority of the diagnosis was established through SPT. Twenty-one patients performed SPT, 20 were positive. The diagnosis in patient with negative SPT was determined by positive provocation test. Fifteen patients performed sIgE and were all positive (mean, 7.12 kU/L; range, 0.04--30 kU/L). Only 3 performed BAT and were positive. ###### Published cases of chlorhexidine-induced anaphylaxis between 2014--2018  Study Country No. of cases Sex Age (yr) SPT sIgE (\<0.35 kU/L) BAT ------------------------------------------- ---------------- -------------- ------ ---------- --------- -------------------- ----- Nakonechna et al., 2014 \[[@B7]\] United Kingdom 6 M 50 NR 30 NR M 78 NR 2.3 NR M 72 Pos 4.4 NR M 73 Pos 3.3 NR M 73 Pos 11.8 NR M 60 Pos 0.69 NR Weng et al., 2014 \[[@B8]\] China 2 M 48 Pos NR NR F 34 Pos NR NR Buergi et al., 2014 \[[@B9]\] Switzerland 1 M 45 Pos 6.1 NR Odedra et al., 2015 \[[@B1]\] United Kingdom 1 M 62 Pos NR NR Rutkowski et al., 2015 \[[@B10]\] United Kingdom 1 M 73 Pos 13.1 NR Hong et al., 2015 \[[@B11]\] Singapore 1 M 66 Pos NR NR Stewart et al., 2015 \[[@B12]\] Australia 1 M 60 Pos Pos NR Chen et al., 2016 \[[@B13]\] United Kingdom 1 \- \- Pos NR NR Wang et al., 2016 \[[@B14]\] Thailand 1 M 54 NR 7.21 NR Teixeira de Abreu et al., 2017 \[[@B15]\] Brazil 1 F 25 Pos NR NR Lasa et al., 2017 \[[@B16]\] Spain 2 M 3 Pos 2.31 Pos M 12 Pos 24.5 Pos Totty et al., 2017 \[[@B17]\] United Kingdom 1 M 70 Pos NR NR Kow et al., 2017 \[[@B18]\] Malaysia 1 M 20 Pos 0.77 NR Postolova et al., 2017 \[[@B19]\] United States 2 M 60 Pos 0.25 (RV=0.1) NR F 29 Pos NR NR Toletone et al., 2018 \[[@B3]\] Italy 1 M 63 Pos 0.04 Pos Gu et al., 2018 \[[@B20]\] China 1 M 57 Neg^\*^ NR NR SPT, skin prick test; BAT, basophil activation test; NR, not reported; Pos, positive; RV, reference value; Neg, . ^\*^The diagnosis was confirmed after the 2nd provocation test. Our review showed that immediate type I allergic reactions to chlorhexidine are increasing, with a mean of 4.8 cases/yr described over the last 5 years, comparing with the 3.25 cases/yr referred in Odedra et al. \[[@B1]\] review over 20 years. This allows us to admit that true incidence of chlorhexidine anaphylaxis is likely to be underestimated in view of its large use as a disinfectant. Undervaluation of previous chlorhexidine reactions increases the risk of a possibly fatal outcome for the patient after re-exposure in future medical-surgical procedures. A prompt referral to a specialist consultation and detailed allergy study is crucial. Detailed history and diagnostic testing allow to confirm the diagnosis of chlorhexidine allergy. **Conflict of Interest:** The authors have no financial conflicts of interest. **Author Contributions:** **Conceptualization:** Mara Fernandes, Tatiana Lourenço, Anabela Lopes, Amélia Spínola Santos, Maria Conceição Pereira Santos.**Data curation:** Mara Fernandes, Tatiana Lourenço.**Investigation:** Mara Fernandes, Tatiana Lourenço, Anabela Lopes, Amélia Spínola Santos, Maria Conceição Pereira Santos.**Supervision:** Maria Conceição Pereira Santos, Manuel Pereira Barbosa.**Validation:** Mara Fernandes, Amélia Spínola Santos.**Writing - original draft:** Mara Fernandes, Tatiana Lourenço, Anabela Lopes, Amélia Spínola Santos, Maria Conceição Pereira Santos.**Writing - review & editing:** Anabela Lopes, Amélia Spínola Santos, Maria Conceição Pereira Santos.

Introduction {#s0005} ============ *Pseudomonas* is a diverse genus known for their ubiquity in the environment and production of secondary metabolites [@bb0005]. Some pseudomonad strains are well-suited to be biocontrol agents [@bb0010], producing a wide range of bioactive metabolites. The general antibiotics produced by *Pseudomonas* include phenazine derivatives, pyoluteorin (Plt), pyrrolnitrin (Prn), hydrogen cyanide (HCN), 2,4-diacetylphloroglucinol (DAPG) and insect toxin [@bb0005]. The ability to release products with antimicrobial activity is the major mechanism by which pseudomonads suppress pathogens. Since the persistent use of chemical pesticides jeopardizes the health of some species like amphibians [@bb0015], biocontrol agents, including plant growth-promoting rhizobacteria (PGPR) such as *Pseudomonas chlororaphis* strains, become the focus of study [@bb0020]. *P. chlororaphis* is an important non-pathogenic biocontrol agent and is studied widely. Many strains show antagonistic activity against a number of disease-causing pathogens, such as *Fusarium oxysporum* f. sp. *radicis-lycopersici* [@bb0025], *Colletotrichum lagenarium*, *Phytophthora capsici*, *Pythium aphanidermatum*, *Pythium ultimum*, *Sclerotinia sclerotiorum*, *F. oxysporum* f. sp. *cucumerinum*, *Corticium sasakii*, *Rhizoctonia solani* [@bb0030] and *Gaeumannomyces graminis* var. *tritici* [@bb0025]. *P. chlororaphis* competitively colonizes the roots, and is distributed worldwide [@bb0030], [@bb0035], [@bb0040]. Additionally, *P. chlororaphis* is well known for its ability to adapt well to the environment using several mechanisms, including the degradation of aromatic compounds [@bb0035], complex regulatory systems [@bb0045], [@bb0050] and metabolism of nitrile compounds [@bb0055]. However, no systematic study of *P. chlororaphis* has been performed. It is important to study the shared and different traits of *P. chlororaphis* at the genomic level to better apply *P. chlororaphis* in agriculture and to promote the production of certain antibiotics. Whole-genome sequencing makes a comparative analysis among strains of *P. chlororaphis* possible. Up until January 2014, four *P. chlororaphis* strains have been whole-genome sequenced. *P. chlororaphis* HT66 was isolated from the rhizosphere of rice in Shanghai, China by our group. This strain shows broad-spectrum resistance to plant pathogens and produces phenazine-1-carboxamide (PCN). *P. chlororaphis* GP72 was isolated from the green pepper rhizosphere by our group, and its genomic information was reported in 2012 [@bb0060]. This strain shows broad-spectrum antimicrobial activity, and can synthesize two phenazine derivatives, phenazine-1-carboxylic acid (PCA) and 2-hydroxyphenazine (2-OH-PHZ). *P. chlororaphis* 30--84 was isolated from the rhizosphere of wheat. Strain 30--84 is regarded as a biocontrol strain because of its ability to control take-all disease. It provides protection mainly by producing phenazines, such as 2-OH-PCA, 2-hydroxyphenazine and PCA [@bb0065]. *P. chlororaphis* O6 was isolated from the rhizosphere of field-grown wheat. It produces phenazines similar to strain 30--84 [@bb0070]. This study first reports the sequencing of the HT66 genome. To identify the shared and divergent genomic characteristics among *P. chlororaphis* strains, we performed a comparative genomic analysis of the four known *P. chlororaphis* strains, HT66, GP72, 30--84 and O6. There are 4833 conserved genes among the four strains and 733 strain-specific genes in the genome of HT66. We focus on their characteristics, such as biocontrol activities, rhizosphere colonization, biosafety, and production of phenazines. Our research provides a theoretical foundation to develop and improve the antagonistic activities of *P. chlororaphis* for agricultural applications, as well as to use *P. chlororaphis* to produce phenazines with high-yield. Results and discussion {#s0010} ====================== General genome features and comparative genomes {#s0025} ----------------------------------------------- The general features of the four *P. chlororaphis* genomes are summarized in [Table 1](#t0005){ref-type="table"}. The genome sizes of HT66, GP72, 30--84 and O6 range from 6.66 to 7.30 Mb, with the number of CDSs ranging from 5869 to 6455. Compared with other pseudomonads, such as *Pseudomonas fluorescens* and *Pseudomonas putida*, these four genomes have higher GC contents [@bb0075], [@bb0080], suggesting that this is a general characteristic of *P. chlororaphis*. The COG database was used to functionally categorize predicted proteins [@bb0085], and we made a comparison of COG categories among the four strains (HT66, GP72, 30--84 and O6). The results are shown in [Fig. 1](#f0005){ref-type="fig"}. The COGs for the four strains showed highly similar distributions, especially COGs F and J, suggesting that the four strains have comparable biological niches. The percentage of genes in COG H was slightly higher in GP72 (approximately 3.95% of the total genes with COG annotations) than in the other three strains (3.77% in HT66, 3.78% in 30--84 and 3.78% in O6), suggesting that genes associated with coenzymes take more proportion in the genome of GP72. We established a phylogenetic tree for completely sequenced representative strains of pseudomonad based on multilocus sequence analysis (MLSA) ([Fig. 2](#f0010){ref-type="fig"}). The tree showed that these four strains of *P. chlororaphis* fall into the same clade. The most closely related pseudomonad species to this clade was *P. fluorescens*, followed by *Pseudomonas syringae*. *P. chlororaphis* may share a more recent common ancestor with *P. fluorescens* than with *P. syringae*. Conserved and specific regions in the genome can be identified through global alignments. BLASTatlas gives an overview of the whole genome homology, and the reference genome can be compared at the gene and/or protein level against many genomes [@bb0090]. In this study, the reference genome of HT66 was compared to the other three query genomes ([Fig. 3](#f0015){ref-type="fig"}). mGenomeSubtractor performs a mpiBLAST based on in silico subtractive hybridization to identify conserved and strain-specific proteins. In this analysis, proteins with homology (H) values of less than 0.42 or more than 0.81 are defined arbitrarily as strain-specific or conserved, respectively [@bb0095]. The degree of protein conservation among HT66 and the other three genomes were determined by blastp based on homology value. The distribution of homology values for the 6455 predicted CDSs from *P. chlororaphis* HT66 compared with the other three genomes is shown in [Fig. 4](#f0020){ref-type="fig"}A. Genes conserved among all of four *P. chlororaphis* isolates comprised 4833 CDSs, representing 74.9%, 79.3%, 82.3% and 77.5% of the coding capacity in HT66, GP72, 30--84 and O6, respectively. Comparisons between HT66 and the other three strains (GP72, 30--84, and O6) revealed that 730 CDSs, 11.3% of the total coding capacity, were HT66 specific (E-value \< 10^− 5^) ([Fig. 4](#f0020){ref-type="fig"}B). Among the 730 genes, we found that 192 genes formed 18 gene clusters. The 192 genes were individually blasted on NCBI (<http://blast.ncbi.nlm.nih.gov/Blast.cgi>). Four clusters showed high homology with *Pseudomonas protegens* Pf-5. One may be related to the metabolism of polysaccharides, one is similar with hemophore and may be responsible for the acquisition of heme, one may represent an *ofa* cluster responsible for the production of orfamide A which was first found in *P. chlororaphis*, and one\'s function remained undetermined. Two additional clusters showed high homology, with one being similar to *P. putida* HB3267, which may be related to phage, and the other having homology with *Pseudomonas* sp. UW4, which has unknown functions. Genomic islands (GIs) {#s0030} --------------------- GIs were identified by IslandViewer [@bb0100]. Since the genome of GP72 contains many contigs, the GI predictions were not reliable (data not shown). The genome of HT66 contains 23 putative GIs ([Fig. 5](#f0025){ref-type="fig"}A), which is more than the other two genomes. The islands range in size from 4008 bp (HGI-5) to 16,094 bp (HGI-18). The largest, HGI-18, contains 10 genes and the smallest, HGI-5, is composed of four genes. Unlike the genome of 30--84 and O6, the GC contents of the putative islands, ranging from 42.65% (HGI-22) to 58.61% (HGI-1), are lower than the average GC content, 62.60%, of the HT66 genome. Compared with the genomes of 30--84 and O6, HT66 has more mobile genetic elements (MGEs). There are five islands (HGI-3, HGI-16, HGI-20, HGI-21 and HGI-23) that contain genes related to putative integrases and two islands (HGI-9 and HGI-14) contain genes encoding putative transposases. Of the 10 genes in HGI-18, five of them (M217_2064, M217_2065, M217_2066, M217_2067 and M217_2068) showed high similarities (84%, 85%, 87%, 89% and 84% at the amino acid level, respectively) with *batOPQRS* genes found in *P. fluorescens* strain BCCM_ID9359. The *bat* gene cluster is responsible for the biosynthesis of kalimantacin/batumin, which has a strong antibacterial activity [@bb0105]. However, the other essential genes in the *bat* cluster are not found in the HT66 genome. Thirteen putative GIs, ranging from 4158 bp (GI-11) to 43,856 bp (GI-6), were identified in 30--84 ([Fig. 5](#f0025){ref-type="fig"}B). The largest, GI 6, is composed of 19 genes, whereas the smallest, GI-11, contains six genes. In the 30--84 genome, nine GIs have lower GC contents, ranging from 43.96% (GI-3) to 59.74% (GI-4), two GIs have higher GC contents, 65.40% for GI-8 and 66.28 for GI-13, and two GIs have analogous GC contents, 61.52% for GI-2 and 60.17% for GI-7, when compared with the average GC content of 62.90%. GI-7 contains a gene (PCHL3084_3079) that encodes a putative transposase and both GI-11 and GI-6 contain genes (PCHL3084_4852 and PCHL3084_3061, respectively) that encode putative integrases. In the genome of O6, 11 putative GIs were identified ([Fig. 5](#f0025){ref-type="fig"}C), with sizes ranging from 4141 bp (OGI-9) to 22,717 bp (OGI-3). The largest GI, OGI-3, contained 13 genes, while only three genes were predicted in the smallest OGI-9. The GC contents of 10 of the islands are lower than the 62.80% average, ranging from 42.92% (OGI-5) to 59.46% (OGI-7). The remaining GI, OGI-11, has a similar GC content, 60.25%, to the average of the O6 genome. OGI-3 contains an integrase gene (PCHLO6_5969) and a transposase gene (PCHLO6_5968), while OGI-6 has two genes (PCHLO6_3566 and PCHLO6_3567) related to transposases. The GIs with MGEs may be able to self-mobilize. The comparative analysis revealed that none of the GIs found in HT66, 30--84 and O6 were in the GP72 genome. Rhizosphere colonization {#s0035} ------------------------ In contrast to the lack of nutrients in soil, the root of plant can secrete a series of compounds to provide rich nutrients for the growth of PGPR. Rhizosphere colonization is important for bacteria to adapt to the nutrient-lack environment. Also rhizosphere colonization is the first step in nearly all interactions between soilborne microorganisms and plants. *Pseudomonas* PGPR strains are regarded as good root colonizers, including the biocontrol bacterium *P. fluorescens* WCS365 and the model bacterium *P. fluorescens* F113 for rhizosphere colonization [@bb0110], [@bb0115]. The major genes and traits involved in colonization competence are identified. Martínez-Granero [@bb0120] reported that variants of *P. fluorescens* F113 with high motility were more competitive in rhizosphere colonization. Flagella play important roles in competitive tomato root tip colonization by *P. fluorescens*. Genes involved in chemotaxis and motility are found in the four genomes. For example, HT66 contains 45 genes related to chemotaxis, including 28 genes encoding methyl-accepting chemotaxis proteins (MCPs). The genome of HT66 contains 42 genes associated with flagellar biosynthesis, including the *fli* (M217_2401--2406), *flh* (M217_2407--2409) and *flg* (M217_5381--5385) operons. There are 40, 42 and 43 genes involved in flagellar biosynthesis in the genomes of GP72, 30--84 and O6, respectively. The *fli*, *flh* and *flg* operons are similarly organized in all four *P. chlororaphis* strains, and show 85% identity at nucleotide level to those located in F113. However, some genes were missing in *P. chlororaphis* strains, such as *flhC*, *flhD* and *fliT*. Whether these differences may affect the ability of motility for *P. chlororaphis* is unclear and remains to be clarified. Root adhesion is just as important as motility for competitive colonization. Pili are appendages on the cell surface that are mainly involved in adhesion. Type IV pili are related to twitching motility and play very important roles in the colonization of plant hosts. Among the *P. chlororaphis* genomes, we identified three putative clusters of type IV pilus biosynthesis genes ([Additional file 1](#ec0005){ref-type="supplementary-material"}), *pilACD*, *pilEYXWV*/*fimUT*, and *pilMNOPQ*, as well as five *pilZ* genes. One cluster, *pilGHIJ*/*chpAC*, was identified in the four genomes and involved in the complex regulatory system for pili biosynthesis. In addition to the biosynthesis of pili, we also found several genes involved in root adhesion in these four *P. chlororaphis* strains, such as genes related to the biosynthesis of filamentous hemagglutinin, hemolysin, lipopolysaccharide O-antigen and alginate ([Additional file 1](#ec0005){ref-type="supplementary-material"}). The functions of these genes have been verified in other species. The genes related to rhizosphere colonization in the four genomes showed a high similarity with each other. This suggests that *P. chlororaphis* may share a similar mechanism in root colonization. Direct plant-growth promotion {#s0040} ----------------------------- Rhizobacteria can directly affect plant growth and development by producing or degrading phytohormones [@bb0125]. Indole-3-acetic acid (IAA) is a well-known plant growth regulator that controls many important plant physiological processes. GP72 and O6 can synthesize IAA via the indole-3-acetamide (IAM) pathway [@bb0130], [@bb0135]. Genes homologous with those encoding tryptophan-2-monooxygenase (*iaaM*, M217_00681/02326) and indoleacetamide hydrolase (*iaaH*, M217_05950/02325/02091/03298) are defective in the HT66 genome, suggesting that HT66 may produce IAA only under certain conditions. 1-Aminocyclopropane-1-carboxylate (ACC) is the precursor of ethylene. The accumulation of ethylene results in the inhibition of root elongation and the acceleration of abscission and senescence [@bb0140]. Putative genes encoding ACC deaminase (M217_04875, MOK_02139, PCHL3084_03963 and PCHLO6_04108) were identified in the four *P. chlororaphis* genomes. ACC deaminase counteracts the plants\' ethylene response by degrading ACC into ammonia and α-ketobutyrate to enhance root growth [@bb0145]. We also found genes related to the catabolism of phenylacetic acid (PAA) (M217_03637--03649, MOK_00348--00336, PCHL3084_02993--02981 and PCHLO6_03084--03072), a plant auxin with antimicrobial activity, in the four genomes. The cluster found in *P. chlororaphis* is similar to the *paa* operon in *P. putida U*, and is involved in the degradation of phenylacetic acid under aerobic conditions. Biocontrol activities {#s0045} --------------------- According to a previous study, the biocontrol abilities of *Pseudomonas* spp. strains play important roles in their capacity to inhibit pathogens. Like other pseudomonad species, *P. chlororaphis* secretes a series of broad spectrum of antibiotics to suppress pathogens [@bb0150]. According to our study, HT66 could suppress *R. solani*, *P. ultimum*, *F. oxysporum* f. sp. *niveum* and the pathogen of Stevia leaf spot disease ([Additional file 2](#f0035){ref-type="fig"}). The putative secondary metabolites were predicted using antiSMASH [@bb0155]. According to our genomic analysis and research in the literature, *P. chlororaphis* produces phenazines, hydrogen cyanide, 2-hexyl-5-propyl-alkylresorcinol (HPR), two siderophores, pyoverdine (Pvd), achromobactin (Acr), *P. fluorescens i*nsecticidal *t*oxin (Fit) and other antibiotics ([Table 2](#t0010){ref-type="table"}). 2-Hexyl-5-propyl-alkylresorcinol (HPR) was reported to show moderate antifungal and antibacterial activities. In the genomes of GP72, 30--84 and O6, a locus similar to the HPR biosynthetic gene cluster (*darABCRS*) described in some other *Pseudomonas* strains [@bb0160], [@bb0165], [@bb0170], [@bb0175] was identified. In the HT66 genome, we only found genes homologous with *darS* and *darR*, which are required to increase the production of HPR [@bb0160]. FitD is related to the potent insecticidal activity of *P. protegens* Pf-5 and CHA0 [@bb0180]. The complete *fit* locus detected in all of four *P. chlororaphis* strains showed a high similarity with that in Pf-5, suggesting that *P. chlororaphis* strains also possess potent insecticidal activity. Orfamide A was found by the genomisotopic approach in Pf-5. It exhibits a role in motility as well as biocontrol activity [@bb0185]. Orfamide A is encoded by an orphan gene cluster composed of *ofaA*, *ofaB* and *ofaC*. This *ofa* cluster was only found in the HT66 genome (M217_04407--04409) among four strains. Interestingly, no mobile elements have been detected near the cluster, thus the mechanism to obtain this cluster remains unknown. The biosynthesis of orfamide A may be conductive to the motility and antimicrobial activity of HT66. The synthesis of siderophores also contributes to the biocontrol activity of *Pseudomonas* and promotes host plant growth. The complete gene cluster for the biosynthesis of Pvd was detected in the four strains. In addition, the four strains also contain putative genes to encode a second siderophore, achromobactin (Acr). HCN is a secondary metabolite produced by some *Pseudomonas* spp., also has biocontrol activity, and the biosynthetic gene cluster (*hcnCBA*) was found in the four *P. chlororaphis* genomes. Prn is an important antibiotic and putative genes involved in the biosynthesis of Prn were detected in the genomes of GP72, 30--84 and O6. *P. chlororaphis* can secrete a variety of secondary metabolites with biocontrol activities. Compared with *P. protegens* Pf-5, *P. chlororaphis* has one advantage to synthesize phenazine derivatives, which have shown impressive antibiosis activities. The presence of a gene cluster for HPR biosynthesis may allow *P. chlororaphis* to exhibit better inhibition to pathogens or competitors. The different antibiotics produced by *P. chlororaphis* enhance the diversity of antibiosis activities. The production of phenazines {#s0050} ---------------------------- Phenazines play important roles in suppressing root diseases caused by pathogens [@bb0190]. There are a variety of phenazine derivatives found in different *Pseudomonas* spp., such as *P. fluorescens* [@bb0195], *Pseudomonas aeruginosa* [@bb0200], and *P. chlororaphis* [@bb0205]. Each genome of the four *P. chlororaphis* strains contains one phenazine biosynthetic gene cluster as shown in [Fig. 6](#f0030){ref-type="fig"}A, but the modifying genes differed among the four genomes. The genome of HT66 contains *phzH*, which encodes an asparagine synthetase that converts PCA to PCN by catalyzing the transaminase reaction [@bb0210], while GP72, 30--84 and O6 contain *phzO*, which encodes an aromatic monooxygenase involved in the hydroxylation of PCA to 2-OH-PCA [@bb0065]. It has been reported that the antifungal activity of PCN was more than 10 times higher than PCA at neutral pH [@bb0025]. Our studies also showed that after 24 h of incubation, only PCN was found in the fermentation broth with HT66, while PCA and 2-OH-PCA or 2-OH-PHZ existed in the fermentation broth with GP72, 30--84 and O6. This suggested that *phzH* was more active than *phzO*. Besides, our studies also showed that the yield of PCN in strain HT66 is obviously higher than the yield of phenazines detected in wild-type of GP72, O6 and 30--84 ([Fig. 6](#f0030){ref-type="fig"}B) [@bb0215], [@bb0220], [@bb0225]. The *P. chlororaphis* strains have high proportions of regulatory genes ([Additional file 3](#ec0010){ref-type="supplementary-material"}), and HT66 contains the greatest number. Our analysis shows that there are 543 putative regulatory genes in the HT66 genome and the proportion of regulatory genes is 8.4%. There were 499 (8.2%), 501 (8.5%) and 526 (8.4%) regulatory genes in the genomes of GP72, 30--84, and O6, respectively. Also, genes reported to relate to the regulation of phenazine production were detected among the four strains, such as *psrA*, *rpoS*, *rpeA* and *lon* proteases [@bb0045], [@bb0230], [@bb0235]. The mutation of *psrA* could threefold increase the phenazine production in PCL1391, and the loss of *rpeA* significantly increased the phenazine production in 30--84 [@bb0230]. It is suggested that we can produce phenazines with high-yield by constructing gene engineering strain of HT66 or other three strains. In addition to *phzH* and *phzO*, there are other modifying genes, such as *phzS* and *phzM*. *phzS*, which encodes a flavin-containing monooxygenase, and *phzM*, which encodes a putative phenazine-specific methyltransferase, are responsible for the conversion of PCA into PYO at certain conditions [@bb0200]. However, pyocyanin (PYO) is involved in pulmonary tissue damage [@bb0240], *P. chlororaphis* is unable to synthesize PYO and makes it more suitable for agricultural applications. Virulence factors {#s0055} ----------------- Like other biosafety strains of *Pseudomonas*, *P. chlororaphis* lacks the key virulence or virulence-related factors. *P. chlororaphis* lacks genomic islands that are homologous with pathogenic islands such as PAPI-1 and PAPI-2 [@bb0245]. Also, genes required for the biosynthesis of phytotoxins (syringomycin, syringopeptin and coronatine) and exoenzymes (cellulases, pectinases and pectin lyases) involved in the degradation of plant cell walls are absent from the genomes of the *P. chlororaphis* strains. No evidence for a type III secretion pathway was found in the genomes of the four *P. chlororaphis* strains. This suggests that *P. chlororaphis* strains are safe for biocontrol applications. Conclusion {#s0015} ========== A comparative genomic analysis of the genomes of HT66, GP72, 30--84 and O6 showed similarities and differences among traits of *P. chlororaphis* strains. It provided new insights into traits involved in the adaption of *Pseudomonas* to environmental niches and in the promotion of plant growth. Our analysis showed that *P. chlororaphis* strains are highly similar in genomic level. Additionally, we analyzed genes related to plant growth promotion. The genomic information indicated that the production of antifungal metabolites differed but all of four strains have one phenazine biosynthesis gene cluster. But the phenazine derivative found in HT66 is PCN whereas the other three strains produce 2-OH-PHZ and PCA. However, only HT66 contains putative genes encoding orfamide A. Also, all of four *P. chlororaphis* strains contain the complete *fit* locus, suggesting that *P. chlororaphis* strains possess potent insecticidal activity. The diversity of antibiotics may allow *P. chlororaphis* to inhibit various pathogens, such as fungi, bacteria and some kinds of insects. Besides, the production of phenazines in HT66 is obviously higher than other strains, and some genes related to the regulation of phenazine biosynthesis have been detected in the four genomes. The analysis of genes contributing to the regulation and biosynthesis of antibiotics may lay the foundation for transforming *P. chlororaphis* to produce high levels of antibiotics. Finally, key virulence or virulence-related factors were absent from the *P. chlororaphis* strains, indicating that *P. chlororaphis* is safe and suitable to be applied in agriculture. Materials and methods {#s0020} ===================== Medium for HT66 and genomic DNA extraction {#s0060} ------------------------------------------ *P. chlororaphis* HT66 was isolated from a rice rhizosphere in Shanghai, China and showed antimicrobial activity to plant pathogenic bacteria. A single colony of HT66 grown on King\'s medium B plate (KB) was inoculated into 5 mL of KB broth and incubated overnight with shaking at 28 °C. Bacterial cells were collected by centrifugation and the genomic DNA was extracted with an Easy Pure Genomic DNA kit (TransGen Biotech) according to the manufacturer\'s instructions. Genome sequencing and annotation {#s0065} -------------------------------- The genome of *P. chlororaphis* HT66 was sequenced using the Illumina Miseq platform (to 40-fold of the sequencing coverage) with paired-end reads. First, a paired-end library was prepared from 4 μg of DNA and subsequently sequenced, generating 590,886 reads in 296,624,772 bp of sequencing data. The data was initially assembled using a Celera Assembler 7.0 and 87 contigs ranging from 112 to 529,941 bp were obtained. 40 scaffolds with genome size of 7.30 Mb ranging from 1906 to 1,134,406 were obtained. The genomes of HT66, 30--84 and O6 were first automatically annotated using the RAST server [@bb0250] and IMG/ER system (<https://img.jgi.doe.gov/cgi-bin/er/main.cgi>) [@bb0255]. The annotations from these two programs were manual curated and combined. Nucleotide sequence accession number {#s0070} ------------------------------------ This Whole Genome Shotgun project of HT66 has been deposited in DDBJ/EMBL/GenBank under the accession number [ATBG00000000](ncbi-wgs:ATBG00000000){#ir0050}. Bioinformatics analysis {#s0075} ----------------------- The genome sequence of HT66 was aligned against other *Pseudomonas* sequences from NCBI\'s database. BLASTatlases were generated using an online tool, GView Server (<https://server.gview.ca/>). Conserved and strain-specific genes were identified based on the homology (H) value (proteins with H values of less than 0.42 or more than 0.81 at E-value \< 10^− 5^ are defined arbitrarily as strain-specific or conserved, respectively) using the mGenomeSubtractor web server (<http://bioinfo-mml.sjtu.edu.cn/mGS/>) [@bb0095]. A comparative genomic analysis of HT66, GP72, 30--84 and O6 was conducted using the IMG website\'s tool, which defined genes with a 60% identity at an E-value \< 10^− 2^ as homologous to those in HT66. The genomic islands were identified using IslandViewer (<http://www.pathogenomics.sfu.ca/islandviewer/query.php>) [@bb0100]. Secondary metabolite production clusters were examined using the antiSMASH program (<http://antismash.secondarymetabolites.org/>) [@bb0155]. The phylogenetic relationships among completely sequenced *Pseudomonas* were determined using the sequences of 1) 16S rRNA and 2) concatenated alignments of 9 highly conserved housekeeping genes: *aroE*, *dnaA*, *guaA*, *gyrB*, *mutL*, *ppsA*, *pyrC*, *recA* and *rpoB*. The multiple-sequence alignments were carried out with ClustalW [@bb0260]. A neighbor-joining tree with 1000 bootstrap replicates was generated using MEGA 6.0 software [@bb0265]. Resistance to plant pathogens {#s0080} ----------------------------- Four normal plant pathogens: *R. solani*, *P. ultimum*, *F. oxysporum* f. sp. *niveum* and the pathogen of Stevia leaf spot disease, were chosen to test the biocontrol activities of HT66. The plant pathogenic bacteria were fully activated on PDA plate. After 3 days of incubation, hyphae block with 8 mm of diameter was added on one side of new PDA plate. The strain HT66 was activated on KB plate and then transferred into new KB broth and incubated to log phase. 10 μL cell suspension was added on filter paper whose diameter also remains to be 8 mm. The distance of the center of the filter paper and hyphae block was 25 mm. The plate was incubated under 28 °C for 5 days. Quantification of phenazine production in HT66 {#s0085} ---------------------------------------------- 400 μL supernatants of 24 h cultures were extracted with 9-times volumes of ethyl acetate and were acidified with adding 20 μL concentrated HCl. Following evaporation of the ethyl acetate under air, phenazines were resuspended in 100% acetonitrile and quantified with HPLC. HPLC was performed with a WondaSil C18-WR column (5 μm; 4.6 × 250 mm, Shimadzu, Japan) and a linear 8 to 60% (vol/vol) gradient of acetonitrile in water with a flow rate of 1 mL/min. UV detection was performed with wavelength scanning at 254 nm. The following are the supplementary data related to this article.Additional file 1Genes related to root colonization among *P. chlororaphis* (table).Additional file 2The biocontrol activities of HT66 strain (figure).Four normal plant pathogens: *R. solani*, *P. ultimum*, *Fusarium oxysporum* f. sp. *niveum* and the pathogen of Stevia leaf spot disease, were chosen to test the biocontrol activities of HT66. (A) *R. solani*; (B) *P. ultimum*; (C) *Fusarium oxysporum* f. sp. *niveum*; (D) the pathogen of Stevia leaf spot disease.Additional file 3Numbers of putative genes involved in regulation in the genomes of four *P. chlororaphis* strains (table). Conflict of interests {#s0090} ===================== The authors declare that they have no competing interests. Authors\' contributions {#s0095} ======================= XHZ, YWC, HBH, HSP and XMS conceived, coordinated and designed the research. WW, XHZ and YWC were responsible for sequencing, finishing and annotating data. YWC, XHZ, XMS and HSP performed experiments and data analyses. XHZ, YWC, HSP and XMS contributed to materials and analysis tools. XHZ, YWC, XMS and HBH drafted the manuscript. All authors read and approved the final manuscript. We are grateful to the sequencing team at Shanghai Personal Biotechnology Co., Ltd. for genome sequencing. Additionally, we would like to acknowledge Dr. Hongyu Ou at Shanghai Jiao Tong University for his assistance with the analysis tools. Our study was funded by the National Natural Science Foundation of China (no. 31270084), the National High Technology Research and Development Program of China (no. 2012AA022107), and the China National Key Basic Research Program (2012CB721005). {#f0005} {#f0010} {#f0015} ![Homology analysis between the *P. chlororaphis* HT66 genome and three other *P. chlororaphis* genomes.\ The mGenomeSubtractor defines coding sequences (CDSs) with a homology (H) value of less than 0.42 as strain-specific, and those with an H value of more than 0.81 as conserved [@bb0095].\ (A) The H-value distribution of 6455 predicted CDSs from HT66 compared with the other three genomes: GP72 (red), 30--84 (green) and O6 (blue).\ (B) Numbers of conserved and strain-specific genes in the genome of HT66 compared with the other three genomes. The total numbers of conserved and strain-specific genes are marked above the bars.](gr4){#f0020} {#f0025} ![Phenazine biosynthesis gene clusters and the production of phenazines in *P. chlororaphis*.\ Phenazine biosynthesis gene clusters are detected in the genomes of *P. chlororaphis* strains. Those genes are shown in different colors. The same color between two strains indicates that the genes are homologous.\ (A1) phenazine biosynthesis gene cluster in strain HT66; (A2) phenazine biosynthesis gene clusters in strains GP72, 30--84 and O6; (B) the production of phenazines in four *P. chlororaphis* strains [@bb0215], [@bb0220], [@bb0225].](gr6){#f0030} ###### General features of the four *P. chlororaphis* strains. HT66 GP72 30--84 O6 ----------------------------------------------------------- -------------- -------------- -------------- -------------- Number of bases 7,298,823 bp 6,663,241 bp 6,665,021 bp 6,977,251 bp G + C (%) 62.60 62.89 62.9 62.8 Protein-coding genes 6455 6091 5869 6236 No. of protein-coding genes with function prediction 5423 5062 5023 5278 No. of protein-coding genes without function prediction 1032 1029 846 949 No. of protein-coding genes connected to KEGG Orthology 3299 3152 3118 3169 No. of protein-coding genes with COGs 4679 4454 4297 4489 No. of protein-coding genes coding signal peptides 719 662 652 664 No. of protein-coding genes coding transmembrane proteins 1509 1444 1400 1453 Coding percentage 88.33% 87.96% 88.12% 88.00% RNA genes 147 88 156 142 rRNA genes (5S rRNA, 16S rRNA, 23S rRNA) 12 (6, 1, 5) 7 (2, 1, 4) 19 (7, 6, 6) 10 (5, 1, 4) tRNA genes 57 61 69 60 Other RNA genes 78 20 68 72 Gene annotations and comparisons were obtained from the IMG database [@bb0255]. ###### Secondary metabolites produced by *Pseudomonas chlororaphis*. HT66 GP72 30--84 O6 --------------------------------------------- ------------------------------------- --------------- --------------- --------------- Hydrogen cyanide (HCN) HCN HCN HCN HCN Phenazine PCN PCA, 2-OH-PCA PCA, 2-OH-PCA PCA, 2-OH-PCA Pyrrolnitrin (Prn) --[a](#tf0005){ref-type="table-fn"} Prn Prn Prn 2-Hexyl-5-propyl-alkylresorcinol -- HPR HPR HPR Orfamide Orfamide A -- -- -- Pyoverdine (Pvd) Pvd Pvd Pvd Pvd Achromobactin (Acr) Acr Acr Acr Acr *P. fluorescens i*nsecticidal *t*oxin (Fit) FitD FitD FitD FitD "--" indicates that the secondary metabolite is not present in the corresponding strain based on previous studies and genomic sequence data.

Full text ========= Progesterone is critical in mammary gland development. Breast cancer evolves from normal tissue through increasingly abnormal cellular changes that include increased expression of progesterone receptor (PR), and PR is an established marker of response to endocrine therapy. PR is expressed as two proteins (PRA and PRB) with different functions, and *in vitro* evidence reveals PRA to inhibit PRB function. This suggests that PRA may repress progesterone action and that the ratio of PRA:PRB may be an important determinant in tissue sensitivity to ovarian steroid hormones. This study examined the expression of PRA and PRB proteins in normal breast tissue (*n* =13) during the menstrual cycle, and in premalignant (*n* =45) and malignant (*n* =39) breast tissues, to determine differences in relative isoform expression.We used dual immunofluorescent histochemistry on formalin-fixed, paraffin-embedded tissue sections using mouse monoclonal antibodies that bind to either PRB or PRA. In most normal breast cases PR staining was confined to scattered epithelial cells expressing equivalent levels of PRA and PRB. However, 50% of cases in the luteal phase (*n* =6) showed reduced PRA expression. In proliferative premalignant lesions without atypia (PDWA, *n* =15), there was a marked increase in intensity and number of cells expressing PR, but inter-cell homogeneity was maintained. Atypical proliferative benign lesions (ADH, *n* =15; DCIS, *n* =15), showed high levels of both PRA and PRB expression with notable inter-cell heterogeneity in relative isoform content. This was also observed in malignant breast tumours (*n* =39). Furthermore, breast tumours expressing an overall predominance of one isoform were associated with features of higher histological grade. In conclusion, our results show a change from inter-cell homogeneity of PRA:PRB in normal tissue to extensive heterogeneity in the malignant state, suggesting a progressive loss of control of relative PRA and B expression that may occur early in cancer development and may eventually be associated with features of poorer prognosis.

By the end of 2018, nearly 8000 complete tailed phage genomes were published online and a further 22,000 partial genomes were stored in databases gathered under the umbrella of the International Nucleotide Sequence Database Collaboration ([@B34]; [@B49]). The classification of this massive group is the formal responsibility of the Bacterial and Archaeal Viruses Subcommittee of the International Committee on the Taxonomy of Viruses (ICTV). In recent years, we (the Subcommittee) have focused on classifying newly described phages into species and genera ([@B41], [@B40]; [@B3]; [@B37]; [@B4]). However, once our attention shifted toward higher order relationships, we found that the ranks currently used in phage taxonomy (species, genus, subfamily, family, and order) are no longer sufficient for the description of phage diversity. The limitation is particularly acute in the case of the order *Caudovirales*---arguably the most abundant and heterogeneous group of viruses ([@B50]; [@B53]; [@B47]). Indeed, the diversity of caudoviruses surpasses that of any other virus taxon. A recent analysis of the gene content of the dsDNA virosphere demonstrated that the global network of dsDNA viruses consists of at least 19 modules, 11 of which correspond to caudoviruses ([@B30]). Each of the eight remaining modules encompasses one or more families of eukaryotic or archaeal viruses. Consequently, each of the 11 caudovirus modules could be considered a separate family. Despite this remarkable diversity, the vast majority of caudoviruses is classified into three families *Myoviridae*, *Podoviridae*, and *Siphoviridae*, which were historically established on morphological features, forming an artificial classification ceiling. These observations prompted us to work on the update of current taxonomic order within the *Caudovirales* order. As an initial step of this major reclassification of the tailed phages, we, the members of the Subcommittee proposed creation of two novel families corresponding to distinct modules revealed in the abovementioned gene-sharing network analyses ([@B30]; [@B11]). The first of these, named *Ackermannviridae*, encompasses phages related to *Salmonella virus ViI* that were formerly assigned to the genus *Viunalikevirus* ([@B2], [@B5]). In the present work, we focus on the second new family, named *Herelleviridae*. The phages belonging to this new family are large myoviruses related to the Bacillus phage SPO1, Staphylococcus phage Twort, Staphylococcus phage K, Listeria phage P100, and Enterococcus phage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\varphi$\end{document}$EF24C. Most of these viruses were previously grouped in the *Spounavirinae* subfamily or recognized as related to it. When this subfamily was first devised ([@B40]), the unifying characteristics of its members included: the hosts belong to the bacterial phylum *Firmicutes*; strictly virulent lifestyle; myovirion morphology (i.e., icosahedral capsid and long contractile tail); terminally redundant, nonpermuted dsDNA genome of 127--157 kbp in length; and "considerable amino acid similarity" ([@B35]). The strictly virulent lifestyle of these viruses has been somewhat disputed ([@B54]; [@B60]) but still remains a rule of thumb for inclusion into the taxon. Since the initial description of the subfamily, the number of its members has grown significantly, and its taxonomic structure has been contested several times ([@B35]; [@B10]; [@B30]; [@B37]; [@B11]; [@B4]). Thus, we wanted not only to delineate a new family but also resolve its internal structure. Unfortunately, there is no one-size-fits-all method for the classification of viruses at all taxonomic ranks. Virus taxonomy has always suffered from the lack of universal marker genes that could be used for phylogenetic reconstruction of the evolutionary relationships. Additionally, differing mutation rates between viral lineages, horizontal gene transfer, and genomic mosaicism limit usefulness of many of the available phylogenetic and phylogenomic methods that have become the gold standard in evolutionary biology ([@B18]; [@B44]). Thus, our strategy for reclassification included a plethora of classification tools that employ very different approaches. Our analyses ranged from coarse-grained, high-throughput, holistic clustering methods where similarity is computed from comparison of all viral genes \[vContact, GRAViTy ([@B11]; [@B7]; [@B8])\] to detailed genome and proteome comparisons \[Victor, Dice, GOAT and Phage Proteomic Tree ([@B52]; [@B45]; [@B44])\] and individual gene phylogenies \[IQtree ([@B46])\]. This multifaceted approach allowed us to gradually descend from the definition of the new family to the study of its internal structure. Interestingly, despite the diversity of the applied methods their results turned out to be complementary and predominantly concordant. All methods painted a robust picture of the new family as a distinct and diverse taxon and supported the same general scheme for its structure ([Table 1](#T1){ref-type="table"}). ###### New classification of the 93 spounaviruses and spouna-like viruses in the new family *Herelleviridae*$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{a}$\end{document}$ Family Subfamily Genus$\documentclass[12pt]{minimal} Species$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{amsmath} \usepackage{wasysym} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{upgreek} \usepackage{mathrsfs} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \setlength{\oddsidemargin}{-69pt} \begin{document} \begin{document} }{}$^{a}$\end{document}$ }{}$^{b}$\end{document}$ ------------------ ------------------- ------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Herelleviridae* *Bastillevirinae* *Agatevirus* *Bacillus virus Agate*, *Bacillus virus Bobb*, *Bacillus virus Bp8pC* (Bp8p-T)     *Bequatrovirus* (formerly *B4virus*) *Bacillus virus AvesoBmore*, *Bacillus virus B4* (B5S), *Bacillus virus Bigbertha, Bacillus virus Riley, Bacillus virus Spock, Bacillus virus Troll*     *Bastillevirus* *Bacillus virus Bastille, Bacillus virus CAM003, Bacillus virus Evoli, Bacillus virus HoodyT*     *Caeruleovirus* (formerly *Bc431virus*) *Bacillus virus Bc431*, *Bacillus virus Bcp1*, *Bacillus virus BCP82*, *Bacillus virus JBP901*     *Nitunavirus* (formerly *Nit1virus*) *Bacillus virus Grass*, *Bacillus virus NIT1*, *Bacillus virus SPG24*     *Tsarbombavirus* *Bacillus virus BCP78* (BCU4), *Bacillus virus TsarBomba*     *Wphvirus* *Bacillus virus BPS13*, *Bacillus virus Hakuna*, *Bacillus virus Megatron* (Eyuki), *Bacillus virus WPh*, *Bacillus virus BPS10C*     Unassigned *Bacillus virus Mater*, *Bacillus virus Moonbeam*, *Bacillus virus SIOphi*   *Brockvirinae* *Kochikohdavirus* *Enterococcus virus ECP3*, *Enterococcus virus EF24C* (phiEFC24C-P2), *Enterococcus virus EFLK1*     Unassigned *Enterococccus virus EFDG1*   *Jasinskavirinae* *Pecentumvirus* (formerly *P100virus*) *Listeria virus A511*, *Listeria virus P100*, *Listeria virus List36*, *Listeria virus LMSP25* (LMTA-57, LMTA-94), *Listeria virus LMTA148*, *Listeria virus LMTA34*, *Listeria virus LP048*, *Listeria virus LP064* (LP-125), *Listeria virus LP083-*2 (LP-124), *Listeria virus AG20*, *Listeria virus WIL1*   *Spounavirinae* *Siminovitchvirus* (formerly *Cp51virus*) *Bacillus virus CP51*, *Bacillus virus JL*, *Bacillus virus Shanette*     *Okubovirus* (formerly *Spo1virus*) *Bacillus virus Camphawk*, *Bacillus virus SPO1*   *Twortvirinae* *Kayvirus* *Staphylococcus virus G1*, *Staphylococcus virus G15*, *Staphylococcus virus JD7*, *Staphylococcus virus K*, *Staphylococcus virus MCE2014*, *Staphylococcus virus P108*, *Staphylococcus virus Rodi*, *Staphylococcus virus S253*, *Staphylococcus virus S25-4*, *Staphylococcus virus SA12*, *Staphylococcus virus Sb1* (676Z, A3R, A5W, Fi200W, IME-SA1, IME-SA118, IME-SA119, IME-SA2, ISP, MSA6, P4W, SA5, Staph1N, Team1)     *Silviavirus* *Staphylococcus virus Remus* (Romulus), *Staphylococcus virus SA11*     *Sepunavirus* (formerly *Sep1virus*) *Staphylococcus virus IPLAC1C*, *Staphylococcus virus SEP1*     *Twortvirus* *Staphylococcus virus Twort*   Unassigned Unassigned *Lactobacillus virus Lb338*   Unassigned Unassigned *Lactobacillus virus LP65*   Unassigned Unassigned *Brochothrix virus A9* $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{a}$\end{document}$Genera were renamed in 2018, taxonomy proposal 2018.007B. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$The species listed here represent the 93 genome data set on which all analyses have been performed. Phage isolates at the subspecies or strain level are indicated between brackets. We emphasize that this reclassification is an essential step in the larger revision of the taxonomy of the order *Caudovirales*. The final goal of our group is a novel system that appropriately accommodates the genomic diversity of prokaryotic viruses and is consistent with taxonomy of eukaryotic viruses ([@B7]; [@B58]). Materials and Methods {#SEC1} ===================== For brevity and clarity's sake, only the basic principles of previously published methods are summarized in the following section. A detailed description of each method used in this study can be found in Supplementary File 1 available on Dryad at [http://dx.doi.org/10.5061/dryad.106q6g6](10.5061/dryad.106q6g6). Creation of the "Herelleviridae" Data Set {#SEC1.1} ----------------------------------------- Genome sequences of known spounaviruses were retrieved from the GenBank or (preferably) RefSeq databases based on literature data, and taxonomic classifications provided by the ICTV and the National Center for Biotechnology Information (NCBI). Records representing genomes of candidate spouna-related viruses were retrieved by searching the same databases with the tBLASTn algorithm ([@B9]) using as queries terminase and major capsid proteins of type isolates of the original subfamily ([@B13]). After manual curation, the search yielded a set of 93 virus genomes (Supplementary Table S1.1 available on Dryad), which were reannotated using PROKKA ([@B55]) and used in the following analyses. To conduct interfamilial comparisons, we compiled an additional genome set including well-described viruses from the ICTV 2016 Master Species List 31V1.1 and Virus Metadata Resource (Supplementary Table S1.2 available on Dryad). All original genome sequences are available from NCBI (accession number information listed in Supplementary Table S1 available on Dryad) and the reannotated genomes are available from Github (github.com/evelienadri/herelleviridae). Definition of the New Herelleviridae Family Within the dsDNA Virosphere {#SEC1.2} ----------------------------------------------------------------------- We examined whether or not the family, *Herelleviridae*, is a clearly distinct group of viruses within the dsDNA phages, by using two cutting-edge virus clustering tools capable of discerning relations even between divergent taxa. Using vConTACT v2.0, we constructed a monopartite network of viral genomes by clustering gene families based on BLAST hits between their protein products as previously described ([@B11]; [@B31]). In this framework, similarities between pairs of genomes were calculated as a function of the shared protein families. The network was visualized with Cytoscape (version 3.5.1; <http://cytoscape.org/>) with genomes sharing more proteins clustered more closely together (detailed information in Supplementary File 1 available on Dryad). The second method used is 'Genome Relationships Applied to Virus Taxonomy' or GRAViTy \[GitHub: Paiewsakun/GRAViTy ([@B7]; [@B8])\]. This framework created a dendrogram of viruses, based on protein profile hidden Markov models of the predicted gene products and genome organization models calculated into a composite generalized Jaccard (CGJ) score representing the difference between two viruses on a scale from 1 to 0 (detailed information in Supplementary File 1 available on Dryad). We also investigated the clustering of the family within the *Caudovirales* order on the VIPtree server ([@B48]), which uses the Phage Proteomic Tree approach described below and detailed in in Supplementary File 1 available on Dryad. Exploration of the Intrafamilial Relationship {#SEC1.3} --------------------------------------------- After demarcation of the family, we proceeded with analysis of its internal structure, using the defined set of 93 genomes described above. In the process, we compared a collection of the classification tools, gathering the phylogenetic signal from the different types of data (whole genome sequences, complete proteomes, marker genes, and gene order). Genome-Based Analyses {#SEC1.4} --------------------- Nucleotide sequence-based grouping of phages was conducted using VICTOR (Virus Classification and Tree Building Online Resource), a Genome-BLAST Distance Phylogeny (GBDP) method ([@B43]; [@B44]). The program calculates intergenomic distances from BLAST+ hits using GBDP (including 100 pseudobootstrap replicates) and used them to infer a balanced minimum evolution tree with branch support via FASTME including subtree pruning and regrafting postprocessing (for details of the algorithm design, see [@B43]; [@B44]). The analysis was conducted under settings recommended for prokaryotic viruses. To reevaluate and interpret results of the VICTOR clustering, we compared the genome sequences using the Gegenees tool with default parameters ([@B15]; [@B6]). The program calculated symmetrical identity (SI) scores for each pairwise comparison based on BLASTn hits and a genome length. To check if the translated local alignment of the whole genomes will be more sensitive to a phylogenetic signal at higher taxonomic ranks, we followed the Dice methodology proposed previously ([@B45]). The Dice score was calculated based on all reciprocal tBLASTx hits between pairs of genomes with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\ge$\end{document}$30% identity, alignment length $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\ge$\end{document}$30 amino acids, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$E$\end{document}$-value $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\le$\end{document}$0.01. Pairs of scores were used to construct a distance matrix, which in turn was converted to the final tree using the BioNJ algorithm ([@B22]). Again, to evaluate and interpret this result, we calculated SI scores between all translated genome sequences using Gegenees. This time, we applied tBLASTx as the alignment algorithm with the other settings left on default values. Proteome-Based Analyses {#SEC1.5} ----------------------- The Phage Proteomic Tree was constructed as described previously ([@B52]). In brief, the protein sequences were extracted and clustered using BLASTp. These clusters were refined by Smith--Waterman alignment using CLUSTALW version 2 ([@B39]). Alignments were scored using open-source PROTDIST from the phylogeny inference package (PHYLIP) ([@B20]). Alignment scores were converted to distances as described in [@B52], and the distances thus obtained were used to generate the final tree using the neighbor joining algorithm. Identification of Protein Clusters {#SEC1.6} ---------------------------------- In order to comprehensively define the gene content in *herellevirus* genomes, we applied two independent, yet complementary methods of identifying orthologous clusters. An initial set of orthologous protein clusters (OPCs) was constructed using the GET_HOMOLOGUES software suite, which utilizes several independent clustering methods ([@B17]). To capture as many evolutionary relationships as possible, a greedy COGtriangles algorithm ([@B36]) was applied with a 50% sequence identity threshold, 50% coverage threshold, and an $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$E$\end{document}$-value cutoff equal to 1e-10. The results were converted into an orthologue matrix with the "compare_clusters" script (part of the GET_HOMOLOGUES suite) ([@B20]). A second method was based on assignment of the genes to a predefined pVOG (prokaryotic Virus Orthologous Group) set described previously ([@B24]) and available at <http://dmk-brain.ecn.uiowa.edu/pVOGs/>. In brief, protein-coding genes in the 93 analyzed genomes were identified using Prodigal V2.6.3 in anonymous mode ([@B29]). Then, the gene products were assigned to the respective orthologue group by HMMsearch ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$E$\end{document}$-value $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$<$\end{document}$10$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{-2}$\end{document}$) against the database of Hidden Markov Models (HMMs) created for every of 9518 pVOG alignments using HMMbuild of HMMer v3.1b2 ([@B21]). Analysis of Gene Synteny {#SEC1.7} ------------------------ To investigate a genomic synteny-based classification signal, we implemented a method developed at the University of Utrecht, a gene order-based metric built on dynamic programming, the Gene Order Alignment Tool (GOAT, Schuller et al.: Python scripts are available on request, manuscript in preparation). The tool used the pVOG assignments described above to generate a synteny profile of every genome (in fact, this pVOGs methodology is integral part of the GOAT pipeline). The algorithm accounted for gene replacements and low similarity between genes by using an all-vs-all similarity matrix between pVOG pairs based on HMM--HMM similarity (HH-suite 2.0.16) ([@B59]). Distant HHsearch similarity scores between protein families were calculated as the average of reciprocal hits and used as substitution scores in the gene order alignment. The GOAT algorithm identified the optimal gene order alignment score between two virus genomes by implementing semiglobal dynamic programming alignment based only on the order of pVOGs identified on every virus genome. To account for virus genomes being cut at arbitrary positions during sequence assembly, the gene order was transmuted at all possible positions and in both sense and antisense directions in search of the optimal alignment score. The optimal GOAT alignment score GAB between every pair of virus genomes A and B was converted to a distance DAB as follows: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$ \mathrm {DAB} = 1 - \frac{GAB + GBA}{GAA + GBB} $\end{document}$$ in which GAB and GBA represent the optimal GOAT score between A and B, and B and A, respectively, while GAA and GBB represent the GOAT scores of the self-alignments of A and B, respectively. This pairwise distance matrix was converted to a tree with BioNJ ([@B22]). Marker Protein Phylogenies {#SEC1.8} -------------------------- Based on the OPC and pVOG clusters defined above, which respectively identified 14 and 38 core protein clusters (Supplementary Table S2 available on Dryad), we chose 10 consistently-predicted protein groups (encoded by genes with well-defined boundaries and without introns) for inclusion as phylogenetic marker. The selected clusters included: DNA helicase cluster, tail sheath protein, two different groups of virion proteins (including the major capsid protein cluster), and six clusters with no known function. The members of these clusters were aligned using Clustal Omega with default parameters ([@B56]). The resulting alignments were analyzed with the IQ-TREE pipeline, which includes the ModelFinder tool that determines the most suitable model of sequence evolution for the alignment, the main algorithm that constructs a maximum-likelihood tree and ultrafast bootstrap (UFBOOT)---an UFBOOT subroutine that calculates the support of the branches ([@B46]; [@B16]; [@B33]; [@B26]). The same program was used to generate the approximation of the "species tree" based on the concatenated alignments of all markers. In this case, the partitioned model of the alignment was also calculated using the ModelFinder module of IQ-TREE and the analysis was run in 100 replicates to select the final tree with best log-likelihood score. Visualization and Comparison of the Results {#SEC1.9} ------------------------------------------- All trees were rooted at Brochothrix phage A9---a phage that consistently appeared as a distant outlier in all obtained topologies (to facilitate comparisons) and visualized using Geneious tree viewer. The taxon coloring and the legend was added using Inkscape 0.92.3 with no distortion of topology, branch lengths, or support. Topological distances between different trees were calculated as Robinson--Foulds metrics ([@B51]) with IQ-TREE and detected differences were visualized as tanglegrams generated using Neighbor Net-based heuristics in Dendroscope 3.5.9 ([@B28]). Results {#SEC2} ======= Definition of the Candidate "Herelleviridae" Family {#SEC2.1} --------------------------------------------------- Recently, several studies have shown the paraphyly of the families constituting the order *Caudovirales* ([@B30]; [@B11]; [@B7]). We created a monopartite network of all dsDNA viruses in the NCBI RefSeq using vConTACT v2.0 ([@B11], Bolduc et al. under revision) showing the phages related to SPO1 as a clearly defined, interrelated cluster ([Fig. 1a](#F1){ref-type="fig"}). The distinctness of the cluster was confirmed with the GRAViTy pipeline ([Fig. 1b](#F1){ref-type="fig"}), which showed that subfamily classifications in the order *Caudovirales* are clustered at the same distance as the new tailed phage family *Ackermannviridae* and as eukaryotic virus families ([@B7]; [@B8]). A further comparison of all dsDNA viruses using the Phage Proteomic Tree method on the VIPTree server showed that myoviruses, siphoviruses, and podoviruses were interspersed with each other, but SPO1-related phages formed a distinct and coherent clade (Supplementary Fig. S1 available on Dryad). These results clearly indicate that the SPO1-related viruses are distinct and form a cohesive group. Based on this evidence, we propose that this group of viruses represents a new family, and we suggest the name *Herelleviridae,* in honor of the 100th anniversary of the discovery of prokaryotic viruses by Félix d'Hérelle. {#F1} Exploration of the Intrafamilial Relationship {#SEC2.2} --------------------------------------------- After delineating the family, we proceeded with the investigation of the relationships between its members. Regardless of the approach applied, we found five clearly-separated clusters interpreted by us as potential subfamilies ([Figs. 1b](#F1){ref-type="fig"}, [2](#F2){ref-type="fig"}, and [3](#F3){ref-type="fig"}, Supplementary Figs. S2--S4 available on Dryad, [Table 1](#T1){ref-type="table"}, Supplementary Table S1 available on Dryad). The first cluster (here suggested to retain the name *Spounavirinae*), groups *Bacillus*-infecting viruses that are similar to Bacillus phage SPO1. The second cluster (*Bastillevirinae*) includes *Bacillus*-infecting viruses that most closely resemble phage Bastille. The third cluster (*Brockvirinae*) comprises viruses of enterococci that are similar to Enterococcus phage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\varphi$\end{document}$EF24C. The fourth cluster (*Twortvirinae*) gathers staphylococci-infected viruses that are similar to Staphylococcus phage Twort. The remaining cluster (*Jasinskavirinae*) consists of viruses infecting *Listeria* that are similar to Listeria phage P100. The classification left three viruses with no genus and subfamily assignment: Lactobacillus phage Lb338, Lactobacillus phage LP65, and Brochothrix phage A9. {#F2} {#F3} Five subfamily-rank clusters can be further subdivided into smaller clades that correspond well with the currently accepted genera ([Table 1](#T1){ref-type="table"}). The evidence supporting this suggested taxonomic reclassification is presented in the following sections. Genome-Based Analyses {#SEC2.3} --------------------- The genome-based analyses used to identify close relationships between phage genomes provide powerful information for species and genus demarcation. We performed an all-against-all BLASTn analysis with Gegenees ([@B6]), revealing that the genomes of several viruses were similar enough to consider them strains of the same species (they shared $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$>$\end{document}$95% nucleotide identity, [Table 1](#T1){ref-type="table"}, Supplementary Table S1, Fig. S2 available on Dryad). We could delineate clear groups with significant nucleotide similarity, proposed as genus-rank taxa, at similarities greater than 50%. Using the BLAST-based phylogenetics framework VICTOR ([@B44]), we were able to confirm that the existing genera form well-supported clades ([Fig. 2a](#F2){ref-type="fig"}). Similar patterns emerged at the translated nucleotide level when the genomes were analyzed using the tBLASTx-based Dice method ([Fig. 2b](#F2){ref-type="fig"}) ([@B45]). An all-against-all comparison at the translated nucleotide level (tBLASTx) with Gegenees showed an overall low level of similarity (15%) within the newly proposed family and allowed us to start delineating the subfamily level at approximately 25% translated genome similarity (Supplementary Fig. S2 available on Dryad). However, the subfamily boundaries were not always clear using these methods. For example, the members of the *Brockvirinae* subfamily shared 20--25% similarity at the translated nucleotide level with the twortviruses and jasinkaviruses. Proteome-Based Analyses {#SEC2.4} ----------------------- As proteome-based analyses rely on genome annotation, they are sensitive to bias introduced by different annotation methods, and the results of such analyses should, therefore, be interpreted with caution. To mitigate this, we reannotated all genomes with the same automated pipeline as described above (M&M, Supplementary File 1 available on Dryad). We inferred a Virus (Phage) Proteomic Tree using only the members of the new family to assess its internal structure ([Fig. 3a](#F3){ref-type="fig"}). This showed clearly-defined clusters at the subfamily and genus rank, but revealed longer than expected branch lengths for phages that had very similar genomes, implying that this method should not be used for fine-grained taxonomic classification. Among 1296 singleton proteins (proteins without recognizable homologues in the analyzed genomes) and 2070 protein clusters defined using the OPC approach, we identified 14 clusters common for all viruses belonging to the new family "*Herelleviridae"* ([Table 2](#T2){ref-type="table"}, Supplementary Table S2 available on Dryad). Classification of the viral proteins using pVOGs showed that 38 pVOGs were shared between all 93 virus genomes, with 14 pVOGs functionally annotated ([Table 2](#T2){ref-type="table"}, Supplementary Table S2 available on Dryad). Upon closer inspection of the gene annotations, we found that these analyses might have been confounded by the presence of introns and inteins in many of the core genes. Indeed, many genes of spounaviruses and related viruses are invaded by mobile introns or inteins ([@B23]; [@B42]). These gaps in coding sequences challenge standard gene prediction tools and introduce additional bias in similarity-based cluster algorithms. Because of these insertions as confounding factors, we used a subset of 10 core genes for further phylogenetic analysis. ###### Core genes with putative annotated functions identified in all 93 herellevirus genomes Putative function of the core gene identified$\documentclass[12pt]{minimal} pVOG/OPC ID Identification method \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{a}$\end{document}$ ----------------------------------------------------------------------------- --------------------------- ----------------------- DnaB-like helicase$\documentclass[12pt]{minimal} VOG0025, OPC6121 OPC, pVOG \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$ Baseplate J-like protein$\documentclass[12pt]{minimal} VOG4691, VOG4644, OPC6132 OPC, pVOG \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$ Tail sheath protein$\documentclass[12pt]{minimal} VOG0067, OPC6142 OPC, pVOG \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$ Terminase large subunit$\documentclass[12pt]{minimal} VOG0051, OPC6160 pVOG \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{c}$\end{document}$ Major capsid protein$\documentclass[12pt]{minimal} VOG0061, OPC6148 OPC, pVOG \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$ Prohead protease VOG4568, OPC6150 pVOG Portal protein VOG4556, OPC6151 OPC, pVOG DNA primase VOG4551 pVOG DNA polymerase I VOG0668, OPC6097 OPC, pVOG RNA polymerase VOG0118 pVOG Recombination exonuclease VOG4575 pVOG Recombination endonuclease VOG0083 pVOG Tail tape measure protein VOG0069 pVOG Tail tube protein VOG0068, OPC6141 OPC, pVOG $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{a}$\end{document}$The full list of protein clusters is available in Supplementary Table S2 available on Dryad (14 core genes identified using OPCs, 38 using pVOGs). $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{b}$\end{document}$Core genes used in concatenated phylogenetic tree. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$^{c}$\end{document}$Omitted in further phylogenetic analyses due to frequent intron invasion and unclear gene borders. OPC = orthologous protein clusters; pVOG = prokaryotic virus orthologous group. Analysis of Gene Synteny {#SEC2.5} ------------------------ Viral genomes are thought to be highly modular, with recombination and horizontal gene transfer potentially resulting in "mosaicism" ([@B32]; [@B38]). By clustering the herelleviruses based solely on the gene order, we investigated plasticity of their genome structure and potential effects of recombination ([Fig. 3b](#F3){ref-type="fig"}). The clustering results proved comparable with results obtained using sequence-based methods, with almost all viruses clustered according the proposed taxa. The potential exception was Bacillus phage Moonbeam ([@B14]), which showed an inversion of the central part of its genome compared with the other herelleviruses. From this overall picture, we can infer that genomic rearrangements leave a measurable evolutionary signal in all lineages, but do not shuffle genomes of related viruses beyond recognition. Thus, we did not observe the high modularity that might be expected with rampant mosaicism. The lack of considerable mosaicism supports recent findings that, at most, about 10% of reference virus genomes have a high degree of mosaicism ([@B11]). Marker Protein Phylogenies {#SEC2.6} -------------------------- We used the amino acid sequences of concatenated marker proteins identified from the OPC analysis ([Table 2](#T2){ref-type="table"}) to generate a phylogenetic tree that is able to identify the evolutionary relationships at the genus and subfamily rank within the new family *Herelleviridae* ([Fig. 4](#F4){ref-type="fig"}). This tree supported all proposed new taxa but was unable to differentiate between the different species. Branches representing subfamilies and genera were particularly well-supported (UFBOOT support above 99%). Additionally, nearly all topologies of single marker trees (Supplementary Fig. S3 available on Dryad) fitted well in the suggested taxonomic structure. The only notable deviation from the proposed classification scheme could be found in the Tail tube protein tree (VOG0068--OPC6141, Supplementary Fig. S3 available on Dryad). It shuffled members of the genus *Silviavirus* into the *Kayvirus* clade and also mixed the genera *Nitunavirus* and *Agatevirus* with unclassified phages. This may indicate that the evolutionary signal contained in this marker is insufficient to resolve related genera. Alternatively, the inconsistencies may be explained by the effect of horizontal gene transfer or convergent evolution introducing additional noise in our data. Regardless of the true reason of this inconsistency, it should be stressed that with a small number of available marker loci, additional sources of phylogenetic signal (e.g., whole genome phylogenies) may be necessary to properly interpret any result. {#F4} Comparison of the Results Obtained Using Different Methods {#SEC2.7} ---------------------------------------------------------- Virus classification methods in general suffer from a low signal-to-noise ratio. This "noise" may be introduced in the data by horizontal gene transfer and differences in mutation rates in different viral lineages. To get a measure of the discrepancies between the methods used above, we calculated the normalized Robinson--Foulds distances (representing the fraction of data partitions that are present only in one of the analyzed trees, Supplementary Table S3 available on Dryad) and created tanglegrams for the visual comparison of topologies (Supplementary Fig. S4 available on Dryad). Trees obtained using different methods differed considerably (normalized Robinson--Foulds metric in range 0.16--0.58) but topological distances between them were comparable to distances between single marker trees (and in most cases smaller, see Supplementary Table S3 available on Dryad). Interestingly, for the herelleviruses, most of the noise becomes averaged at the genus rank, meaning that the grouping at this rank and above remains almost the same regardless of the classification method employed. The only significant discrepancies compared with the proposed taxonomic classification were observed in the GOAT analysis and one single-marker tree (i.e., tail tube protein tree, VOG0068--OPC6141). Both of these deviations concerned a single genus or even unclassified species and they did not follow any commonpattern. Discussion {#SEC3} ========== The rapid expansion of phage genomics and metagenomics has left taxonomy behind. There are more than 8000 publicly-available caudovirus genomes, but only 873 have been officially classified by the ICTV ([@B19]). The remaining genomes are provisionally stashed in the NCBI database within "unclassified" bins attached to the order *Caudovirales* or its associated families ([@B13]; [@B1]; [@B57]). One of the main problems is that the level of sequence divergence is so high that it often leaves no detectable sequence similarity between disparate members of the same order. Thus, not a single reliable phage-specific or even *Caudovirales*-specific marker gene could be defined. In addition, a classification system based on a single marker would be highly prone to instances of horizontal gene transfer. Indeed, there is no commonly recognized general phage classification tool and all of the currently used phylogenetic approaches have their critical limitations as described in this study. For that reason, above the family rank we had to rely on high-throughput network and clustering analyses (vConTACT, GRAViTy, and VipTree) that are capable of discerning the groups of taxa that are comparable, even if phylogenetic signal is sparse. These methods can analyze significant subsets of the viral genomic space in a reasonable time, outcompeting traditional phylogenetic approaches in terms of speed. They are, however, still expensive computationally and need to be recalculated when new data become available ([@B12]). Moreover, these high-throughput methods do not attempt to model the process that gave rise to the observed data, but rather calculate arbitrary distance matrices from local similarities and use them to define groupings. Thus, the relation between the calculated distance and the divergence time remains unclear and the results of these methods should be taken with a grain of salt, especially in less divergent taxa or at the lower taxonomic ranks. After defining the new family *Herelleviridae*, we applied a combination of genome and proteome analyses, gene synteny assessments, and multimarker gene phylogenies to establish its internal taxonomic structure. It has to be stressed that the results of most of these methods should be treated as approximations of phylogenic reconstruction. Many of them suffer from similar methodological drawbacks as the abovementioned high-throughput clustering techniques, lacking proper theoretical support of their algorithms. Only the maximum-likelihood analysis of (a) marker sequence(s) allows for rigorous, statistically sound phylogenetic inference under a well-defined model of sequence evolution. Unfortunately, if the number of available marker loci is small, this method becomes vulnerable to the noise introduced by horizontal gene transfer ([@B18]). More importantly, this approach is heavily influenced by the gene annotation. This may be a crucial disadvantage as the quality of database records is often debatable and computational reannotation of analyzed genomes does not always yield valid, comparable results. On the other hand, these drawbacks can be easily circumvented by methods analyzing whole genome sequences (DICE, VICTOR, BLAST). Obviously, they are annotation-independent and mitigate the effects of horizontal gene transfer by averaging the signal across the total genome length. Unfortunately, if the untranslated nucleotide sequence of the virus is used, rapid decay of the similarity should be expected above the genus rank (e.g., Supplementary Fig. S2 available on Dryad). Above that rank, nucleotide sequence similarities were virtually undetectable, but sequence translations (DICE coefficient) or protein sequences (Phage Proteomic Tree) were still considerably similar. Thus, nucleotide sequence-based approaches capture small differences (e.g., silent mutations) between closely related genomes and may be well suited for species and strain demarcation but gradually lose sensitivity with each consecutive taxonomic rank. To the best of our knowledge, the GOAT algorithm is the only method explicitly aimed at capturing the signal associated with genomic rearrangements in fluid genomes of viruses. Unfortunately, the evolutionary process that is responsible for the observed variations is even less studied than whole genome similarity metrics and we cannot rule out that this algorithm may be disproportionally susceptible to some random rearrangement events. However, it is ideally suited to pinpoint just those kinds of genomic rearrangements and mutations that are missed by other methods. Thus, it can provide unique data on structural dynamics of the studied genomes but in its present form should not be treated as the primary classification tool. Bearing in mind all the advantages and limitations of the classification tools utilized here, and the convergence of their results for the analyzed taxa, we recommend an "ensemble of methods" approach similar to the one we used as a method of choice for the phage taxonomy. We suggest that future classification efforts should implement at least one well established phylogenetic method (e.g., maximum-likelihood analysis of concatenated marker genes/proteins) and at least one whole genome-based annotation-independent method to account for annotation inconsistencies, rearrangements and mosaicism. Additional approaches may be used, especially if methods of choice produce inconclusive or discordant results but should always be used with regard to their limitations. All evidence considered, we suggest that the SPO1-related phages should be removed from the family *Myoviridae* and given a family rank. Hence, we proposed establishing a new family *Herelleviridae*, containing five subfamilies: *Spounavirinae* (*sensu stricto*), *Bastillevirinae*, *Twortvirinae*, *Jasinkavirinae*, and *Brockvirinae*, each comprising the genera listed in [Table 1](#T1){ref-type="table"}. The suggested classification corresponds well with host taxonomy and leaves only 3% of viruses within the new family unassigned. These unassigned viruses may represent clades at the rank of genus or even subfamily that are still undersampled. Removing spounaviruses from the family *Myoviridae* to form the new *Herelleviridae* family is a major change in phage taxonomy. We envisage this detachment from their original taxon will be followed by abolishment of the*Podoviridae*, *Myoviridae*, and *Siphoviridae* and creation of new "phylogenomic" families, based on current subfamily-rank clades, which will faithfully reflect the genetic relationships between bacterial viruses. In our opinion, these changes are necessary to accommodate the observed diversity of tailed phages. It is worth stressing that this change does not remove the historically established caudovirus morphotypes: myovirids forming virions with contractile tails, siphovirids with long noncontractile tails, and podovirids with short noncontractile ones. Nevertheless, by disconnecting morphotype and taxonomy, related clades can be grouped across different morphotypes. Such an approach would solve the problems of the muviruses that are suggested to be classified in the family "*Saltoviridae*" ([@B27]) and potentially the broad set of Escherichia phage lambda-related viruses that are currently distributed among the families *Siphoviridae* and *Podoviridae* ([@B25]). Finally, abolishing the current morphology-based classification of tailed phages will remove the major barrier in classifying phages from metagenomic sequence data. We thank Laura Bollinger, Integrated Research Facility at Fort Detrick, for technical writing services. *Dedication: This article is dedicated to Hans-Wolfgang Ackermann, a pioneer of prokaryotic virus electron microscopy and taxonomy, who died on 12 February 2017, at the age of 80. He was involved in the early stages of this study, and his input is dearly missed*. Supplementary Material ====================== Data available from the Dryad Digital Repository: [http://dx.doi.org/10.5061/dryad.106q6g6](10.5061/dryad.106q6g6). Funding ======= This work was supported by the National Science Centre, Poland (2016/23/D/NZ2/00435) to J.B.; the Netherlands Organization for Scientific Research (NWO) (Vidi 864.14.004) to B.E.D. (MBPS and BED); the US National Science Foundation (DUE-132809 and MCB-1330800) to R.A.E.; the University of Helsinki and Academy of Finland funding for Instruct-FI to H.M.O.; the Chargé de Recherches fellowship from the National Fund for Scientific Research, FNRS, Belgium to A.G.; the EUed Horizon 2020 Framework Programme for Research and Innovation, 'Virus-X' (685778) to F.E.; the Gordon and Betty Moore Foundation Investigator Award (GBMF\#3790) to M.B.S.; the Battelle Memorial Institute's prime contract with the US National Institute of Allergy and Infectious Diseases (NIAID) (Contract number HHSN272200700016I) to J.H.K.; the GOA grant 'Phage Biosystems' of the KULeuven to R.L. the Intramural Research Program of the NIH, National Library of Medicine to J.R.B. and I.T.; and by the Biotechnology and Biological Sciences Research Council Institute Strategic Programme in Gut Microbes and Health BB/R012490/1 and its constituent project BBS/E/F/000PR10353 to E.M.A. The content of this publication does not necessarily reflect the views or policies of the US Department of Health and Human Services or of the institutions and companies affiliated with the authors.

1.. Introduction ================ Methionine is an essential amino acid for animals and is involved in numerous metabolic processes \[[@b1-sensors-10-03562]--[@b4-sensors-10-03562]\]. In addition to being a building block in protein synthesis, methionine, after being transformed into S-adenosylmethionine, serves as a methyl donor in transmethylation reactions involved in the biosynthesis of lipids, biotin, and polyamines \[[@b5-sensors-10-03562]\]. Since methionine cannot be synthesized *de novo* in mammal cells, its supplementation in animal diets is required to provide optimal growth and physiological performance of the animals. Plant proteins, however, are poor in methionine and its optimal level in animal diets is provided by supplementation with crystalline methionine \[[@b6-sensors-10-03562]\] or methionine analogs such as 2-keto-4-(methylthio) butyric acid \[[@b7-sensors-10-03562]\] and hydroxymethionine \[[@b8-sensors-10-03562],[@b9-sensors-10-03562]\]. Therefore, timely and accurate pre-quantification of this amino acid in feed ingredients is necessary to improve cost efficiency of feed formulation and prevent its overdosage. According to Klasing and Austic \[[@b10-sensors-10-03562]\] and Baker \[[@b11-sensors-10-03562]\], excess of individual amino acids due to feed mixing errors can be potentially harmful to the animal, with methionine considered to be the amino acid possessing the highest toxicity. Feed compounds such as cysteine, vitamin B~12~, arginine, choline, and sulfate that are related to methionine metabolism can affect the apparent methionine requirement of animals and additionally complicate the estimation of the optimal dosage of this amino acid in animal diets \[[@b12-sensors-10-03562]\]. Chemical assays including high performance liquid chromatography (HPLC) are commonly used to quantify methionine level in feed ingredients. The analysis, however, involves pretreatment of the samples with performic acid followed by acid digestion \[[@b13-sensors-10-03562],[@b14-sensors-10-03562]\]. The procedure results in a complete protein degradation and liberation of methionine which differs from protein digestion under physiological conditions. Feed-derived methionine, which is available to animals to assimilate, can be more accurately estimated by animal or microbial assays which are considered to correspond more directly to the physiological needs of animals \[[@b15-sensors-10-03562],[@b16-sensors-10-03562]\]. Although considered standard, animal assays are laborious, expensive, and time consuming \[[@b17-sensors-10-03562]--[@b19-sensors-10-03562]\]. The types of animal assays that have been used for quantifying methionine availability have been reviewed extensively by Froelich and Ricke \[[@b18-sensors-10-03562]\] and will not be discussed further in the current review. Microbial assays appear to be easier and more affordable for routine analysis. Rapid development and recent improvements in molecular techniques allow for constructing successful and accurate amino acid biosensors via more precise genetic targeting of specific genes in microbial cells. This review discusses methionine biosynthesis and regulation in *Escherichia coli* and the potential of genetically modifying this microorganism into practical whole cell biosensors for methionine bioavailability quantification. 2.. Microbial Biosensors ======================== Recently, numerous microbial biosensors have been created and used in medical diagnostics, food technology, biotechnology, and environmental monitoring. Microbial biosensors couple a biological element (enzymes, viable or non-viable microbial cells) and a transducer or a device which allows for rapid, accurate and sensitive detection of target analytes \[[@b20-sensors-10-03562],[@b21-sensors-10-03562]\]. Their popularity is due to highly specific selectivity to the substrate of interest, relative inexpensiveness, and portability \[[@b22-sensors-10-03562],[@b23-sensors-10-03562]\]. Versatile microorganisms have proven to be useful in development of biosensors. The bacterium *Vibrio harveyi* and *Mycena citricolor*, a fungal microorganism, demonstrated high sensitivity for detecting cyanide and sodium monofluoroacetate respectively \[[@b24-sensors-10-03562]\]. A microbial biosensor for sensitive, selective, rapid, and direct determination of *p*-nitrophenyl (PNP) -substituted organophosphates was developed based on PNP oxidation metabolic pathway of the *Moraxella* sp. \[[@b25-sensors-10-03562]\]. *Flavobacterium* sp. were employed for development of a biosensor for methyl parathion pesticide \[[@b26-sensors-10-03562]\]. The variety and versatility among microbial species useful in the construction of biosensors for environmental application is more extensively discussed elsewhere \[[@b20-sensors-10-03562],[@b27-sensors-10-03562]\] and will not be further discussed here. In the food industry, microbial biosensors, derived from *Gluconobacter oxydans* and yeast have gained popularity for detecting total sugars, sucrose, and ethanol \[[@b28-sensors-10-03562],[@b29-sensors-10-03562]\]. Respiratory activities of *Gluconobacter oxydans* DSM 2343 cells, immobilized on chitosan, were used in the quantification of glucose. A linear relationship (R^2^ = 0.99) between sensor's response and substrate concentration was achieved in the range of 0.05 to 0.1 mM glucose \[[@b23-sensors-10-03562]\]. By using a microbial biosensor based on immobilized *Saccharomyces ellipsoideus* yeast cells, Rotariu *et al*. \[[@b29-sensors-10-03562]\] were able to determine ethanol concentrations up to 50 mM in alcoholic beverages including two types of beer, vodka, and cognac. The comparison to the chemical assay used for the analyses of the same analyte revealed good correlation (correlation coefficient 0.998) between the biosensor and the spectrometric method. An *Acetobacter pasteurianus*-based biosensor has been proposed as an alternative to chemical methods available for quantifying lactate which is used as an indicator for specific fermentations activities including those of milk, yogurt, and wine \[[@b30-sensors-10-03562],[@b31-sensors-10-03562]\]. *Aeromonas phenologenes*-, *Pseudomonas fluorescens*-, and *Bacillus subtilis*-based biosensors were proposed to serve as alternatives in quantification of amino acids including tyrosine, tryptophan, and glutamate \[[@b32-sensors-10-03562],[@b33-sensors-10-03562]\]. 3.. *E. coli* as a Biosensor ============================ Among all microorganisms, *E. coli* is one of the most highly investigated bacteria for the purposes of biosensor fabrication. It is easy to cultivate, with simple nutritive requirements and rapid growth \[[@b34-sensors-10-03562]\]. *E. coli* is a Gram negative microorganism with very well known genetics which enables the construction of a wide variety of biosensors \[[@b20-sensors-10-03562],[@b21-sensors-10-03562]\]. The complete *E. coli* genome has been sequenced and the information deposited to the National Center for Biotechnology Information (NCBI) \[[@b35-sensors-10-03562]\]. Thus, each DNA sequence of interest is routinely available to the public and can be used for a wide range of potential further genetic manipulations. In promoter-based *E. coli* biosensors, a gene promoter, inducible by the analyte of interest, is fused to a reporter that generates a signal in response to the analyte that can be easily monitored and measured. A strong SOS *E. coli* promoter fused to a *lux* gene resulted in the development of a construct which served in a dose-dependent detection of 6 genotoxic chemicals including mitomycin C, *N*-methyl-*N*-nitro-*N*-nitrosoguanidine, nalidixic acid, dimethylsulfate, hydrogen peroxide, and formaldehyde \[[@b36-sensors-10-03562]\]. An *E. coli* BL21 DE3 (RIL) biosensor strain displayed a specific response and high sensitivity to different aromatic aldehydes. The response was measured by monitoring the fluorescence of a reporter (green fluorescent protein) fused to an alcohol dehydrogenase inducible promoter (*Sso2536adh*) belonging to the archaeon *Sulfolobus solfataricus* \[[@b37-sensors-10-03562]\]. A plasmid-borne transcriptional fusion between the *E. coli* nitrate reductase (*narG*) promoter and the *Photorhabdus luminescens lux* operon was used to generate a modified *E. coli* with a highly bioluminescent phenotype in the presence of nitrate that enabled the detection of nitrate to a level of 5 × 10^−5^ mol L^−1^ (0.3 ppm) \[[@b38-sensors-10-03562]\]. Following the same approach, biosensors for toluene, arsenite and arsenate, and lead have also been generated \[[@b39-sensors-10-03562]--[@b41-sensors-10-03562]\]. In addition to environmental testing and analyses, *E. coli*-based biosensors were found to be useful in the food industry as well. *E. coli* derived β-galactosidase, glucose oxidase, and horseradish peroxidase were immobilized on a glassy carbon electrode to generate a biosensor for quantification of lactose in raw milk \[[@b42-sensors-10-03562]\]. Simultaneous determination of various mono- and disaccharides was performed by a sensor array comprised of bacterial mutants of *E. coli* K12 lacking different transport systems for individual carbohydrates \[[@b43-sensors-10-03562]\]. 4.. *E. coli* as a Biosensor for Amino Acid Bioavailability =========================================================== Amino acids are building monomers in protein synthesis and indicators for protein quality which explains the interest in constructing microbial biosensors for their quantification. Successful whole-cell biosensors for the quantification of threonine, tryptophan, lysine and glutamine have been developed based on *E. coli* auxotrophy for the respective amino acids \[[@b44-sensors-10-03562]--[@b46-sensors-10-03562]\]. Wild type *E. coli* can synthesize all amino acids and does not require their supplementation in media. However, auxotrophic mutants that are defective in the biosynthesis of the amino acid of interest grow in a dose-dependent fashion in response to the external concentration of the amino acid. In addition, *E. coli* is a part of the intestinal microflora of most animals and humans with high similarity in the assimilation of amino acids and peptides which is a necessary prerequisite for the bacterium to serve as a representative biosensor microorganism for these compounds \[[@b47-sensors-10-03562]\]. After pre-treating feed ingredients with pronase and peptidase, Erickson *et al.* \[[@b48-sensors-10-03562]\] obtained a correlation of 0.94 between lysine bioavailability determined by using an *E. coli* lysine auxotroph and previously published chick bioassay data. Indeed, in a direct experimental comparison, the *E. coli* biosensor developed by Chalova *et al.* \[[@b19-sensors-10-03562]\] proved to be as accurate as a chick bioassay for quantitation of bioavailable lysine in diverse feed ingredients and mixtures including soybean meal, cotton seed meal, meat and bone meal, chick starter and finisher, and swine starter. Early efforts for microbial quantification of methionine have also been based on bacterial auxotrophy. Hitchens *et al.* \[[@b45-sensors-10-03562]\] demonstrated that *E. coli* GUC41 could grow on DL-methionine sulfoxide but not on DL-methionine sulfone. The microbiological assay was as accurate as the chemical assay with high correlation coefficients between the two. The microbiological assay values for amino acid content were expressed as percentages of the HPLC values to obtain the bioavailability values. By using *E. coli* ATCC 23798, a methionine auxotroph, Zabala-Díaz *et al.* \[[@b49-sensors-10-03562]\] were able to quantify crystalline methionine added to feed. The feed matrix had negligible influence on the assay and methionine recovery percentages for all supplementation levels ranged from 71 to 80% indicating consistency in the bacterial response to the supplemented methionine. The *E. coli* methionine growth assay has also been miniaturized and adapted to be conducted in microtiter plates where a linear response of the *E. coli* auxotroph to up to 26.8 μM methionine was achieved \[[@b50-sensors-10-03562]\]. In general, this early assay work with *E. coli* methionine auxotrophs supports the feasibility of this approach and potential reliability for routine use. However, there are also limitations with these particular *E. coli* methionine auxotrophs. To the best of our knowledge, *E. coli* methionine auxotrophs used for methionine quantification so far have been generated via chemical mutation \[[@b51-sensors-10-03562]--[@b53-sensors-10-03562]\]. This mutation method is a "hit-or-miss" approach that mutates the organism in random locations, followed by a selection of a certain phenotype. As a result, the mutation is not target specific and various non-methionine related genes can be affected. Revertants or compensatory mutations may occur to abolish the desired functionality \[[@b54-sensors-10-03562]\]. In addition, in the case of methionine, the auxotrophic requirements for this amino acid are not specific and can also be satisfied by a variety of compounds including methioninyl peptides, α-hydroxy methionine, *N*-acetylmethionine, and the α-keto analogue α-keto-λ-methiol butyrate \[[@b55-sensors-10-03562]\]. When a chemically generated *E. coli* methionine auxotroph (ATCC 23798) was used, Froelich *et al*. \[[@b56-sensors-10-03562]\] established no differences based on substrate affinities of an *E. coli* methionine auxotroph to methionine and methionine hydroxy analog, respectively. Estimated maximum growth rate of the *E. coli* auxotroph when grown on both substrates was also found to be similar. Although the *E. coli* methionine auxotroph did not discriminate between methionine and its hydroxyl analog, it appears that both sources are not equally assimilated by animals. While studying the efficacy of both methionine and methionine hydroxyl analog supplementation of pig diets, Shoveller *et al.* \[[@b57-sensors-10-03562]\] established that methionine hydroxyl analog is significantly less bioavailable compared to DL-methionine for protein deposition in growing pigs. Similar observations were made by Feng *et al.* \[[@b58-sensors-10-03562]\] who reported the methionine hydroxyl analog to be 26.8% and 54.4% less effective than methionine for growing pigs with respect to nitrogen retention and plasma urea nitrogen respectively. Therefore, more specific mutagenesis that targets specific gene(s) without altering other metabolic pathways would be a more desirable approach to generate a microbial biosensor for discriminating and quantifying specific forms of methionine. Detailed knowledge about *E. coli*'s genomics and more specifically, the genes involved in methionine biosynthesis and transportation is a prerequisite to accomplish such a goal and is the focus of the discussion in the following sections. 5.. Biosynthesis of Methionine in *E. coli* =========================================== Methionine's carbon skeleton is initially derived from aspartate. The intermediates of this pathway, aspartyl semialdehyde and homoserine, are also used in the synthesis of lysine and threonine. Serine and cysteine are metabolically related to the methionine pathway: serine being the precursor in the synthesis of folate, which is the methyl donor for the synthesis of methionine and cysteine from the precursor of cystathionine, which is intermediate in methionine synthesis \[[@b59-sensors-10-03562]\]. Methionine biosynthesis results from the coupling of homocysteine and a methyl group, but can be accomplished via two distinct pathways \[[@b55-sensors-10-03562],[@b60-sensors-10-03562]\]. The *E. coli* K-12 methionine biosynthesis pathway <http://biocyc.org/ECOLI/organism-summary?object=ECOLI&detail-level=3> has been schematically presented by EcoCys \[[@b61-sensors-10-03562]\], a member of BioCys database collection (<http://biocyc.org/publications.shtml>), and is available via SRI International Pathway Tools, version 13.5 \[[@b62-sensors-10-03562]\]. A summary of the consecutive reactions, participating genes and respective products is given in [Table 1](#t1-sensors-10-03562){ref-type="table"}. The nonfolate branch of the methionine pathway includes *metA, metB, metC, metH, and metK* and the folate branch is comprised of *metF* and *metE* which are all negatively controlled by the *metJ* repressor system \[[@b73-sensors-10-03562]\]. The final methyl transfer is catalyzed by either a B~12~-dependent methyltransferase (*metH* gene product) or a non-B~12~-methyl transferase (*metE* product). The metabolic intermediate, 5-methyltetrahydrafolate, encoded by the *metF*, provides the methyl group for both enzymes to attach. This is a convergence point through which the cells are able to balance the requirement for protein synthesis, methylation reactions, and nucleic acid synthesis on several levels and to regulate the pathway flow of methyl units \[[@b73-sensors-10-03562]\]. Once methionine is formed, it is metabolized to S-adenosylmethionine (AdoMet) in the presence of ATP and AdoMet synthetase, a *metK* gene product \[[@b74-sensors-10-03562]--[@b76-sensors-10-03562]\]. The regulation of the methionine biosynthetic pathway consists of positive and negative feedback mechanisms depending on methionine availability and B~12~. For example, MetE biosynthesis is autoregulated via a negative feedback loop. It can also function as an antagonist of the *metR* gene product, by either interfering directly in the activation mechanism or by repressing *metR* expression \[[@b73-sensors-10-03562],[@b77-sensors-10-03562],[@b78-sensors-10-03562]\]. Alternatively, the activation depends not only on the presence of a functional MetR but also on a coactivator. A functional *metF* gene is required for vitamin B~12~- mediated repression of *metE* gene, and 5-methyltetrahydrofolate may be involved in a negative feedback repression. Inactivation of the *metE* gene allows for accumulation of the methionine intermediates O-succinylhomoserine, cystathionine, homocysteine, and 5-methyltetrahydrafolate \[[@b60-sensors-10-03562],[@b75-sensors-10-03562]\]. The first unique step in bacterial methionine biosynthesis involves the activation of homoserine, which in *E. coli* is accomplished through a succinylation reaction catalyzed by homoserine transsuccinylase (HTS). The activity of this enzyme is closely regulated *in vivo* and therefore represents a critical control point for cell growth and viability \[[@b79-sensors-10-03562]\]. Born and Blanchard \[[@b80-sensors-10-03562]\] cloned homoserine transsuccinylase from *E. coli* and demonstrated that the enzyme generates a complex with the succinyl group of succinyl-CoA before transferring it to homoserine to form the final product, O-succinylhomoserine. The enzyme can be inhibited by iodoacetamide in a pH-dependent manner which suggests the presence of cysteine in the active site that forms a succinyl-cysteine intermediate during enzymatic turnover. In *E. coli*, HTS is not only a key regulator of methionine biosynthesis but also of the bacterial growth at elevated temperatures \[[@b81-sensors-10-03562]\]. *E. coli* growth is impaired at temperatures above 44 °C due to the instability of HTS. According to Biran *et al*. \[[@b79-sensors-10-03562]\], the instability of the protein is determined by the amino-terminal part of the protein, and its removal or substitution by the N-terminal part of beta-galactosidase confers stability. Mordukhova *et al*. \[[@b82-sensors-10-03562]\] reported that two amino acids in the enzyme, namely isoleucine 229 and asparagine 267, are responsible for HTS instability and their substitution leads to stabilization of HTS molecule and improved bacterial growth at elevated temperature. MetA is controlled by the expression of *metJ* \[[@b83-sensors-10-03562]\]. MetE, a zinc-containing monomer, transfers the methyl group of N^5^-methyl-tetrahydrofolic to the thiolate group of homocysteine \[[@b68-sensors-10-03562],[@b75-sensors-10-03562],[@b78-sensors-10-03562]\]. Several mechanisms for repression of *metE* exist. The interaction between methionine as S-adenosylmethionine with MetJ leads to a corepression of *metE* through a negative feedback loop \[[@b75-sensors-10-03562]\]. The absence of MetR can also repress *metE*. MetR and MetE are of similar size and exist relatively in the same location but result from transcription in opposite directions \[[@b77-sensors-10-03562]\]. MetR is a transactivator of both *metE* and *metH* gene \[[@b75-sensors-10-03562],[@b77-sensors-10-03562],[@b78-sensors-10-03562]\]. Homocysteine coactivates both the expression of the *metR* gene and the MetR stimulation of *metE* expression. The vitamin B~12~ -mediated repression of the *metE* gene in *E. coli* requires the B~12~-dependent transmethylase, a product of the *metH* gene. It has been proposed that the MetH-B~12~ holoenzyme complex is involved directly in the repression mechanism \[[@b55-sensors-10-03562],[@b73-sensors-10-03562],[@b75-sensors-10-03562]\]. According to Wu *et al*. \[[@b84-sensors-10-03562]\], however, B~12~-mediated repression of the *metE* gene derives primarily from a loss of MetR-mediated activation due to depletion of the coactivator homocysteine, rather than a direct repression by the MetH-B~12~ holoenzyme. MetH has a higher constant of Michaelis-Menten (K~m~) than MetE, which is compensated by the very strong expression of the *metE* gene \[[@b68-sensors-10-03562]\]. N^5^-methyl-H~4~-folate transfers a methyl unit to the MetH holoenzyme where it is subsequently attached to homocysteine \[[@b55-sensors-10-03562],[@b68-sensors-10-03562],[@b75-sensors-10-03562],[@b78-sensors-10-03562]\]. The cobalamin-independent methyltransferase (MetE) shares no similarity with the sequence of the cobalamin-dependent protein (MetH), suggesting that the two have arisen by convergent evolution \[[@b85-sensors-10-03562]\]. The *metF* gene codes for N^5^-methyl-H~4~-folate and regulates *metE* in an indirect way. N^5^-methyl-H~4~-folate is required for the transfer of a methyl group to the B~12~ within the MetH holoenzyme forming a methyl-B~12~ enzyme; the catalytically active methylated form of the MetH protein regulates *metE* expression \[[@b55-sensors-10-03562],[@b75-sensors-10-03562],[@b77-sensors-10-03562],[@b78-sensors-10-03562]\]. Regulation by MetJ may occur more readily because of the existence of 5 *met* boxes in *metF*'s promotor region making it more sensitive than the other *met* genes to small increases of AdoMet that might occur in B~12~ grown cells \[[@b75-sensors-10-03562]\]. YagD is a third methionine synthase in *E. coli*. YagD is a zinc-dependent methyltransferase with a catalytic mechanism similar to MetH and synthesizes methionine from S-methylmethionine or S-adenosylmethionine and homocysteine. YagD does not contribute to the utilization of methionine sulfoxide as methionine sulfoxide is converted to methionine via reduction. YagD is subject to regulation by the MetJ-S-adenosyl-methionine system \[[@b68-sensors-10-03562]\]. All *met* genes are regulated by MetJ. MetJ protein binds to a specific DNA region, met box, which is present in all *met* genes except *metH*. The met box region is a sequence with dyad symmetry (TGAA . . . TTCA) and produces a helical region containing four leucine residues seven amino acids apart. This motif is called a leucine zipper and has been proposed to play a role in protein dimerization that is required for DNA bindings. MetJ can bind to this region and prevent the transcription of most of the *met* genes (*metA, metBL, metC, metF, metJ, metR,* and *metE*) \[[@b55-sensors-10-03562],[@b59-sensors-10-03562],[@b71-sensors-10-03562],[@b74-sensors-10-03562]\]. The interaction of the MetJ protein with the *met* operator region is markedly enhanced by the presence of AdoMet. 6.. Bacterial Transport of Methionine ===================================== Although *E. coli* prototroph cells are capable of synthesizing methionine *de novo,* they can also acquire external methionine or methionine analogs to satisfy cellular needs for either methionine or sulfur which reflects the high flexibility of the organism under a wide range of environmental conditions. The activity of methionine transport systems in *E. coli* is influenced by the concentrations of both external and internal methionine pools \[[@b86-sensors-10-03562]\]. Cells with increased internal methionine pool or pre-exposed to excess of external methionine exhibit decreased rates of methionine uptake. Conversely, starvation for methionine in a methionine auxotroph can increase the rate of external methionine transport \[[@b86-sensors-10-03562]\]. At least two transport systems for methionine exist in *E. coli*. The high affinity transport system (*metD*) has a K~m~ of approximately 10^−7^ M and is responsible for the uptake of [l]{.smallcaps}- and [d]{.smallcaps}- methionine isomers \[[@b87-sensors-10-03562]\]. MetD is an ABC transporter with Abc the ATPase, YaeE the permease, and YaeC the likely substrate binding protein. The expression of these genes is regulated by [l]{.smallcaps}-methionine and MetJ, the common repressor of the methionine regulon. Interestingly, [l]{.smallcaps}-methionine inhibits the uptake of [d]{.smallcaps}-methionine; however, [d]{.smallcaps}-methionine does little to affect the uptake of the [l]{.smallcaps}-isomer \[[@b88-sensors-10-03562]\]. By performing competition experiments Kadner \[[@b88-sensors-10-03562]\] established that MetD possesses a distinct substrate-binding site for each stereoisomer. The second system (*metP*) is a low affinity system with a K~m~ of approximately 40 μM and can transport [l]{.smallcaps}-methionine but not the [d]{.smallcaps}-isomer \[[@b88-sensors-10-03562],[@b89-sensors-10-03562]\]. By using various deletion mutants, Merlin *et al*. \[[@b87-sensors-10-03562]\] observed that only mutants with active MetD were able to grow on [d]{.smallcaps}-methionine. Methionine can be transported across a concentration gradient (a temperature sensitive uptake process) with the assistance of MetD \[[@b90-sensors-10-03562]\]. The accumulation against a concentration gradient and the temperature influence of the uptake indicates that methionine enters bacterial cells through active transport, which is an energy-dependent process. When starved of methionine, the rate of uptake of methionine is faster than those grown with methionine \[[@b91-sensors-10-03562]\]. Both systems are regulated by the level of the internal methionine pool of the bacterium and differ in affinity by a factor of at least 400-fold \[[@b86-sensors-10-03562],[@b88-sensors-10-03562],[@b91-sensors-10-03562]\]. In *E. coli*, the methionine, transported into the cells, accumulates in the form of AdoMet rather than as free methionine. Active transport can be completely eliminated in the presence of glucose with the presence of azide and fluoride \[[@b86-sensors-10-03562],[@b91-sensors-10-03562]\]. In the absence of glucose, cells could still accumulate methionine less efficiently. The methionine analogs that inhibit uptake required the −S (or Se)-CHS group \[[@b91-sensors-10-03562]\]. The initial rate of uptake of [l]{.smallcaps}-methionine was poorly affected by the addition of α-keto-λ-methiol-butyrate, [d]{.smallcaps}-methionine, or methionine sulfoxide when they were added simultaneously with the substrate. However, methionine transport was reduced in cells exposed to analogs and methionine variations prior to the addition of a substrate \[[@b86-sensors-10-03562]\]. 7.. Genetic Strategies for Construction of *E. Coli* Mutants for Methionine Bioassays ===================================================================================== 7.1.. General Strategies ------------------------ Understanding methionine biosynthesis and transportation in *E. coli* is a prerequisite for constructing accurate and specific microbial biosensors for methionine estimation. An *E. coli* microbial bioassay approach for methionine quantification necessitates using an auxotrophic strain for methionine which is incapable of biosynthesizing this amino acid on its own. As discussed in the previous sections, mutants, currently in use for the purpose of methionine quantification, have been developed and isolated after exposure to a chemical mutagen. However, the disadvantage of the imprecise nature of chemical mutagenesis requires other approaches for generation of mutants that are more specific and efficient. To avoid the hit and miss nature of broad spectrum mutagenesis approaches such as those involving addition of chemical mutagens requires a strategy that targets a specific site on the genome without alteration of the remainder of the genome. Such approaches are more likely to result in phenotypes that are exclusively linked to a specific genetic modification rather than the collective accumulation of several mutations some of which may not be related to the gene(s) of interest. The problem with multiple mutations is not only the risk of unpredictable reversion of the phenotype of interest but a less robust mutant that does not grow as well under the selective conditions required for a particular assay. In the past few years, genetic tools have been developed that harness the utility of biological systems such as transposons that can more directly interact with the bacterial chromosome at specific sites. Transposons are versatile tools for genetic manipulation and analysis. These are DNA sequences that can be mobilized into bacterial chromosome by a recombination process that is catalyzed by an enzyme, called transposase. In contrast to chemical mutagenesis, insertion of a transposon in the bacterial genome causes complete disruption of the gene of interest and results in non-leaky phenotypes that are specifically linked to the mutated gene \[[@b92-sensors-10-03562]\]. This approach is particularly useful where the function of all the genes in the bacterial genome is not known or the biosynthetic pathway of the analyte of interest is complicated or bypassed. Transposon engineering was used by McAdam *et al*. \[[@b93-sensors-10-03562]\] to mutate *Mycobacterium bovis* BCG, a member of the slow-growing *M. tuberculosis* complex. Two auxotrophs for leucine and one for methionine were isolated from the library of transposon insertions and used to study the functionality of the respective genes. The random insertion of transposon Tn4560 into *Streptomyces tendae* ATCC 31160 resulted in identification of six genes involved in the biosynthesis of nikkornycin, a nucleoside-peptide chitin synthase inhibitor \[[@b94-sensors-10-03562]\]. The scope of application of transposon mutagenesis techniques was increased by Kwon and Ricke \[[@b95-sensors-10-03562]\] and Kwon *et al*. \[[@b96-sensors-10-03562]\] when they developed an approach for the identification of transposon location in the bacterial genome based on the amplification of transposon flanking regions using polymerase chain reaction (PCR). Deletion of specific gene(s) which abolishes the biosynthetic capability of the bacteria for certain amino acids is an alternative approach to transposon mutagenesis. This technique is applicable when the sequence of the gene to be deleted is known. Four individual (Δ*glnA1*, Δ*glnA2*, Δ*glnA3*, and Δ*glnA4*) and one triple mutant (Δ*glnA1EA2*) of *Mycobacterium tuberculosis* were generated by deletion to investigate the roles of glutamine synthetase enzymes in the nitrogen metabolism of this specific bacterium \[[@b97-sensors-10-03562]\]. Tryptophan auxotrophy in *Leptospira meyeri* was achieved by deletion of the tryptophan biosynthetic gene *trpE* via homologous recombination \[[@b98-sensors-10-03562]\]. Li and Ricke \[[@b99-sensors-10-03562]\] were able to completely delete *lysA* in *E. coli* K12 by using a linear DNA which contained at both ends 50 bp sequences homologous to upstream and downstream sequences of *lysA*. The *lysA* encoded for diaminopimelic acid decarboxylase and is a key enzyme in lysine biosynthetic pathway in *E. coli* K12 \[[@b99-sensors-10-03562]\]. The recombinant strain behaved as an auxotroph for lysine and was not able to grow in minimal medium without lysine supplementation. The *E. coli* K12 Δ *lysA* growth response to increasing concentrations of lysine was found to be linear, which is a must for the purpose of lysine quantification in feed-derived proteins. In fact, after being converted into a fluorescent biosensor, the strain was successfully used to quantitatively assess lysine in feed ingredients and complete diets \[[@b19-sensors-10-03562]\]. 7.2.. Generating specific *E. coli* Methionine Auxotrophs --------------------------------------------------------- Identifying a specific gene from the biosynthetic pathway of methionine in *E. coli* in which a deletion could result in methionine auxotroph phenotype is not straightforward. Due to the versatility in methionine biosynthetic pathway, only mutations in certain genes in the *met* regulon would result in auxotrophy for methionine and not for any of the pathway intermediates or precursors. For example, *metL* mutants can grow on homoserine while *metBL* mutants can propagate on both cystathionine and homoserine \[[@b101-sensors-10-03562]\]. While studying mutations that influenced the methionine biosynthetic pathway, Mulligan *et al*. \[[@b55-sensors-10-03562]\] observed that deficiencies in *metA* and *metB* resulted in growth requirements for homosysteine and cystathionine, and a mutation in *metE* was overcome by B~12~ supplementation. Insertions in *metL* and *metH* also did not result in methionine auxotrophy. In the same study, *metF* was the only gene which sufficiently abolished biosynthesis functions to ensure a requirement for external methionine for bacterial growth. In *Streptomyces lividans*, the disruption of a gene encoding for 5,10-methylenetetrahydrofolate reductase which was found to be highly homologous to *E. coli* MetF, also resulted in methionine auxotrophy \[[@b102-sensors-10-03562]\]. In contrast to other amino acid mutants of *E. coli* which require the [l]{.smallcaps}-form for growth, *cys* and *met* mutants are capable of using either isomer of cysteine or methionine \[[@b103-sensors-10-03562]\]. [d]{.smallcaps}-Methionine is not bioactive and cannot be directly incorporated into protein biosynthesis. Therefore, the utilization of [d]{.smallcaps}-methionine for [l]{.smallcaps}-methionine is justifiable only if the [d]{.smallcaps}-form of this amino acid is ultimately transformed into [l]{.smallcaps}-methionine. According to Cooper \[[@b104-sensors-10-03562],[@b105-sensors-10-03562]\], conversion of [d]{.smallcaps}- to [l]{.smallcaps}-methionine in *E. coli* is possible and occurs via oxidative deamination and subsequent transamination of the keto-methionine product of the former reaction. By using ultraviolet irradiation, Cooper \[[@b105-sensors-10-03562]\] was able to generate a mutant incapable of growing on [d]{.smallcaps}-methionine. The locus of the [d]{.smallcaps}-methionine utilization was mapped at approximately 2 min away from *lac* region toward threonine and leucine. Therefore, to make an *E. coli* biosensor specific for growth on [l]{.smallcaps}-methionine would require an addition to disabling the [l]{.smallcaps}-methionine biosynthesis genes as well as the genes responsible for conversion of other forms such as [d]{.smallcaps}-methionine transformation to [l]{.smallcaps}-methionine. Another approach may be possible now that the structure and allosteric regulation of the high-affinity *E. coli* methionine ABC transporter is better understood \[[@b106-sensors-10-03562]\] and manipulation of methionine transport may offer a more precise targeting of the relationship between intracellular [l]{.smallcaps}-methionine transport and external concentration of different forms of methionine. 8.. Detection Modes for Methionine Microbial Biosensors ======================================================= As analytical tools, microbial biosensors are genetically engineered to produce a measurable signal in response to the compound of interest. These signals include but are not limited to light emission, reflection, fluorescence, or absorption. Although their function is based on different principles, a common feature is the proportionality between the intensity of the signal and the concentration of target analyte \[[@b107-sensors-10-03562]\]. The choice of an appropriate detection system is an important point since each detection mode possesses advantages and disadvantages which are summarized in [Table 2](#t2-sensors-10-03562){ref-type="table"}. Several detection methods for detecting microbial responses exist for potential implementation in respect to the microbiological assay for methionine quantification. 8.1.. Optical Density --------------------- Measuring optical density (OD) is a common approach to monitor bacterial growth and is thoroughly reviewed by Kavanagh in \[[@b111-sensors-10-03562]\]. The readings provided by the spectrophotometer correlate directly with the concentration of bacteria in the test media. A non-inoculated tube with media is used to calibrate the spectrophotometer as a representative blank or "zero" value. A nutrient medium is inoculated with the *E. coli* bacterial suspension, incubated at 37°C, and the growth response of the test organism is measured hourly. Over time the turbidity/cell number is calculated and ultimately plotted to determine a linear response. The optical density values that constitute the linear slope gradually increase as the concentration of test nutrient increases. The theoretical aspects of photometry have been extensively described elsewhere \[[@b111-sensors-10-03562]\]. Although OD measurements require minimal technical effort and are relatively inexpensive there are several drawbacks for their use in quantifying nutrients from complex oranic matrices such as animal feeds and feed ingredients. To prevent any potential alternations of methionine availability, autoclaving or heat treatment of feed samples should be avoided. Therefore, the primary problem is the contribution of nonspecific microflora growth that results in OD increases, not corresponding to different concentrations of the nutrient being quantified by the assay organism. Erickson *et al*. \[[@b112-sensors-10-03562]\] were able to overcome this with the use of an antibiotic-based selective media which exclusively supported the growth of the *E. coli* lysine bioassay organism. Froelich *et al*. \[[@b113-sensors-10-03562]\] tested a medium containing a cocktail of antibiotics and antifungal agents and demonstrated that they did not alter growth kinetics of a methionine *E. coli* auxotroph in response to various methionine concentrations when compared to growth responses of the same strain of *E. coli* grown without antibiotics. Although the use of antibiotics suppress background microflora sufficiently to allow for short term bioassay measurements eventually background microflora can overcome antibiotic inhibition of growth. Consequently, if detection based on OD alone requires longer assay times, background microflora would need to be eliminated by sterilization of the feed matrix. Sterilization, particularly by thermal processes adds to the uncertainty of the accuracy of the amino acid assay by potentially altering their respective availability and any resulting measurements would no longer reflect the original values of the animal feeds being assayed. 8.2.. β-galactosidase --------------------- The measurement of β-galactosidase expression historically is a well-understood, easily measured and reliable method for examining bacterial genetics and understanding fundamentals of gene regulation \[[@b114-sensors-10-03562]\]. The β-galactosidase enzyme assay has also been used as an indirect method of microbial detection and quantitation and is more sensitive than OD \[[@b108-sensors-10-03562]\]. The *E. coli lac* operon enables the organism to metabolize lactose as a carbon source. This *lac* operon is translated at a constant rate when lactose is present in the media \[[@b115-sensors-10-03562]\]. Therefore, the enzyme concentration of lysed cells can be directly correlated with the total bacterial cell count. β-galactosidase assay was successfully used by Hitchens *et al.* \[[@b45-sensors-10-03562]\] to quantify the bioavailability of cysteine, methionine, threonine, and tryptophan in 17 foods. To overcome the lack of exoproteolytic activities in *E. coli*, the food matrices were enzymatically digested with pronase and further subjected to analysis with the respective auxotrophic bacterial strain. The accuracy of the β-galactosidase assay was evaluated by comparison of the data to the amino acid estimates in the same food-derived proteins obtained by a chemical assay. Spearman rank correlation coefficients for the two methods were significant and found to be as follows: cysteine (0.61), methionine (0.95), threonine (0.64), and tryptophan (0.85). Thus, Hitchens *et al*. \[[@b45-sensors-10-03562]\] and earlier Tuffnell and Payne \[[@b108-sensors-10-03562]\] demonstrated that β-galactosidase biosynthesis correlated to the concentrations of the amino acid needed by the auxotrophic bacterial cells for growth and could be accurately used in the quantification of methionine, tryptophan, and lysine bioavailability. However, the β-galactosidase-based assay requires more steps than an OD assay. It is disruptive and is not appropriate for kinetic studies. More importantly as with OD measurements, there is a risk of nonspecific background microflora contributing to overall β-galactosidase assay response as several organisms possess this enzyme. Therefore, a key requirement for any detection system to be used is that it is sufficiently unique *versus* the typical native microflora already present on the animal feed matrix. 8.3.. Luminescence ------------------ Compared to β-galactosidase, luminescence is a more recent detection approach and has been routinely used in the generation of bacterial biosensors. This method allows the detection of viable cells through a quantum measurement to indirectly enumerate cells. A bioluminescent signal can not only be coupled to bacterial growth response to accurately measure levels of the respective nutrient limiting bacterial growth but is 10- fold more sensitive than OD \[[@b46-sensors-10-03562],[@b109-sensors-10-03562],[@b116-sensors-10-03562]\] and considerably more sensitive than β-galactosidase. While testing the efficiency of both firefly luciferase and β-galactosidase as reporters in developing vaccine virus, Rodrigues *et al*. \[[@b117-sensors-10-03562]\] established that the luciferase assay was 1,000-fold more sensitive than that of β-galactosidase. The limit of detection of luminescence produced by the action of luciferase was found to be approximately one infected cell in a background of a million noninfected cells. In *E. coli*, bioluminescence does not occur and must be acquired via genetic modifications \[[@b118-sensors-10-03562],[@b119-sensors-10-03562]\]. Luminescence is accomplished by the introduction of the *luxAB* genes via plasmid or chromosomal insertion. Cells are subsequently grown in the test media and a chemical reagent is required to induce the bioluminescent phenotype of the inserted sequence. The production of light lasts only minutes (seconds in flash luminescence) before destabilization of the exogenous reagent. This is a shortcoming of luminescence technology and has led to genetic development of longer lasting luminescence and reagent-less requiring strains. The bioluminescence assay response is measured with a flash luminescence luminometer and requires addition of autoinducer \[[@b46-sensors-10-03562],[@b116-sensors-10-03562]\] and therefore, it is not possible to continuously monitor bacterial cell population increases during exponential growth. In order to quantify luminescence, expensive detection devices must be purchased \[[@b46-sensors-10-03562],[@b116-sensors-10-03562]\]. However, Froelich *et al*. \[[@b120-sensors-10-03562]\] demonstrated with a bioluminescent *E. coli* methionine auxotroph that, although the growth kinetics between the transformed strain and a nonplasmid carrying auxotroph were somewhat different, the OD-based standard curves between the two strains were similar. This indicates that even in the absence of available luminescent detection equipment such strains could still be used in a conventional OD-based assay with the advantage being that they could be used for luminescent based assays when the opportunity for using such equipment is made available. 8.4.. Fluorescence ------------------ A similar assay method to luminescence is fluorescence. One advantage of fluorescence over luminescence is that it is less expensive to detect and is a self-contained assay, requiring no additional reagents \[[@b121-sensors-10-03562]\]. Fluorescence occurs naturally in chemicals that resonate (a carbon chain with alternating single and double bonds) \[[@b122-sensors-10-03562]\]. It also occurs in a protein referred to as Green Fluorescent Protein (GFP) that originally was produced in jellyfish (*Aequorea victoria*), but the DNA encoding sequence has been isolated and incorporated via transgenics (the genetic translocation of genes from one species to another, *i.e.*, placing *gfp* from jellyfish into a eukaryotic strain) \[[@b123-sensors-10-03562]\]. Once incorporated, either through transformation of a gene system on a vector or directly incorporated in DNA of the test organism, the fluorescent protein is concomitantly synthesized with other cell proteins. The assay then requires a spectrofluorometer that can excite the engineered organism's new protein and detect emitted light at a different wavelength \[[@b110-sensors-10-03562],[@b124-sensors-10-03562]\]. Just as with OD, the value are recorded over time and graphed linearly over time and concentration. Originally discovered in *Aequorea victoria,* two proteins were found within this jellyfish that had luminescent/fluorescent capabilities. The first was aequorin that emitted blue light with the presence of Ca^++^(luminescence). The second was the green fluorescent protein which when excited could be detected on a fluorometer (fluorescence) \[[@b125-sensors-10-03562]\]. Tsien \[[@b122-sensors-10-03562]\] described the general molecular weight of most GFP forms to be approximately 27 kDa. An advantage of the use of *gfp* as a reporter gene is its structural stability. The eleven beta strands surround and protect the chromophore that is positioned near the geometric center of a "beta can", which protects the chromophore from temperature, acid, and oxidation. Its normal excitation peak is at 395nm with a minor peak at 475nm and emission peaks at 508 nm \[[@b122-sensors-10-03562]\]. It does not require any additional substrates or reagents to fluoresce, and thus sample perturbation and destruction are avoided \[[@b126-sensors-10-03562]\]. Froelich *et al*. \[[@b127-sensors-10-03562]\] successfully transformed an *E. coli* methionine auxotroph with a plasmid encoding for a green fluorescent protein and demonstrated that it could be used to quantify methionine in several representative animal feed ingredients. However, some variation between OD---based measurements and fluorescent measurements were noted suggesting some potential interference with fluorescent measurements. Some artifacts have to be taken into consideration when detecting the GFP chromophore. Media and feeds may contain aromatic and resonating conjugate carbon chains that may also fluoresce. Some of them have emission spectrum overlapping the emission spectrum of GFP (350 to 550 nm). This often leads to low signal to noise ratios, decreased emission intensity and occasionally complete inability to detect the fluorescence emitted by GFP \[[@b128-sensors-10-03562]\]. To correct for this, a simple excitation filter that allows light to pass at wavelengths higher than 350 nm is used in conjunction with an emissions bandpass filter that allows only light with certain wavelength to pass. Heim and Tsien \[[@b129-sensors-10-03562]\] using specific optical filters detected three different forms of GFP simultaneously in samples of viable bacteria. In addition, GFP variants with different emission spectra were created to overcome either the low intensity of the emission signal or the background fluorescence of various compounds. The resulting GFP mutants are characterized with different excitation/emission spectra, brighter fluorescence, higher solubility, and more even distribution throughout the cytoplasm than the wild type \[[@b130-sensors-10-03562]\]. These mutants allow the monitoring of multiple species of bacteria simultaneously in a complex microbial community. However, Patterson *et al*. \[[@b131-sensors-10-03562]\] implied that no single variant was appropriate for all applications but that each of them offers advantages and disadvantages when investigating viable cells. There are some issues associated with fluorescence assay which must be accounted for such as autofluorescence from matrices that naturally fluoresce. By studying autofluorecence capacity of feed ingredients including soybean meal, cottonseed meal, meat and bone meal, Chalova *et al.* \[[@b132-sensors-10-03562]\] observed that hydrolyzed feed proteins in concentrations up to 0.1 mg/ml did not interfere with the fluorescence of Gfpmut3 \[[@b133-sensors-10-03562]\] which was used as a reporter in an *E. coli* whole cell-based lysine biosensor. Nonhydrolyzed soybean meal and cottonseed meal did not exhibit detectable background fluoorescence up to 2 mg/ml. The same authors demonstrated the advantages of the constructed *gfp*-based biosensor in the quantification of bioavailable lysine in the feed samples when contaminated with *E. coli*. The second possible problem is light scatter in the fluorometer. By simply diluting the sample to an OD of 0.1 or less, absorption artifacts and secondary inner filter affects can be avoided. This also prevents light scatter because the density of the sample is lower. Other techniques to lower light scatter should be checked with a blank made from media to determine if scatter is occurring. Finally, the existence of possible quenchers such as other fluorophores which may lower or lose quantum yield can be problematic. This can be corrected with the application of several equations depending on the cause \[[@b110-sensors-10-03562]\]. 9.. Conclusions =============== In conclusion, microbial sensors for methionine quantification in feed and feed ingredients are an alternative to animal assays because they have the advantage of being simpler, more rapid, and cost efficient. Versatile tools in molecular biology combined with current knowledge of *E. coli* genetics favor the generation of appropriate and successful constructs that may serve as methionine biosensors. However, more work needs to be done in understanding the bacterial genome to better target gene(s) that lead to generation of methionine auxotroph exhibiting a single phenotype. The wide variety of available detection modes should facilitate the choice of a reporting system which will contribute to the simplified operation and identification of a biosensor's emitted signal. This review was supported by Hatch Grant H8311 administered by the Texas Agricultural Experimental Station and Texas Advanced Technology program grant \#000517-0220-2001 (Texas Higher Education Board, Austin, TX). C. A. Froelich, Jr. was partially supported by a Texas Public Education Grant (Texas Higher Education College Board, Austin, TX). ###### Summary of genes which participate in methionine biosynthesis and regulation. **Gene** **Product** **Reaction/Function** **Reference** ---------------------------------------- ----------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------ ----------------------------- **Methionine biosynthesis** *metA* homoserine O-transsuccinylase L-homoserine + succinyl-CoA \<==\> O-succinyl-L-homoserine + coenzyme A \[[@b63-sensors-10-03562]\] *metB* Cystathionine ã-synthetase L-cysteine + O-succinyl-L-homoserine \<==\> succinate + L-cystathionine + H+ \[[@b64-sensors-10-03562]\] *metC* cystathionase L-cystathionine + H2O \<==\> pyruvate + ammonia + L-homocysteine + H+ \[[@b65-sensors-10-03562]\] *metH* Cobalamin-dependent tetrahydropteroylglutamate methyltransferase L-homocysteine + 5-methyltetrahydrofolate \<==\> L-methionine + tetrahydrofolate \[[@b66-sensors-10-03562]\] *metE* Cobalamin-independent tetrahydropteroyltriglutamate methyltransferase L-homocysteine + 5-methyltetrahydropteroyltri-L-glutamate \<=\> L-methionine + tetrahydro-pteroyltri-L-glutamate \[[@b67-sensors-10-03562]\] *yagD* homocysteine methyltransferase L-homocysteine + S-adenosyl-L-methionine \<==\> L-methionine + S-adenosyl-L-homocysteine + H+ \[[@b68-sensors-10-03562]\] *metK* methionine adenosyltransferase Catalyzes the formation of the sulfonium compound S-adenosylmethionine from methionine \[[@b70-sensors-10-03562]\] **Methionine biosynthesis regulation** *metR* DNA-binding transcriptional activator, homocysteine-binding Transactivate *metA, metE*, and *metH* \[[@b71-sensors-10-03562]\] *metJ* S-adenosylmethionine transcriptional repressor Represses transcription from associated promoter \[[@b72-sensors-10-03562]\] ###### Detection systems for microbial assays. **Detection Systems** **Characteristics** **Reference** ----------------------- ------------------------------------------------------------------------------------------------ ------------------------------ Optical Density (OD) EconomicalReliableEasy to use \[[@b111-sensors-10-03562]\] â-galactosidase More sensitive than ODRequires more steps \[[@b108-sensors-10-03562]\] Luminescence 10X more sensitive than ODRequires aldehyde to initiate luminescenceExpensive \[[@b109-sensors-10-03562]\] Fluorescence Same advantages as luminescenceLess expensive detectionSelf contained assay: no reagents added \[[@b110-sensors-10-03562]\] [^1]: Current address: Department of Biochemistry and Molecular Biology, University of Food Technologies, Plovdiv, Bulgaria. [^2]: Current address: Department of Biochemistry and Molecular Biology, Louisiana State University Health Science Center, Shreveport, LA 71129, USA.

Introduction {#s1} ============ An ecological trap arises when an artificial habitat is introduced into a natural environment, attracts animals to its vicinity and the subsequent association leads to negative ecological consequences for the animal [@pone.0015646-Battin1]. Animals may prefer an artificial habitat over natural habitats if it mimics the set of ecological cues which signify a good quality habitat, despite other ecological processes rendering the habitat of low quality and leading to poorer reproduction or survival. Robertson and Hutto [@pone.0015646-Robertson1] suggest that ecological traps derive from habitat alteration that operates in one of three ways; (1) increasing the attractiveness of an environment by enhancing the set of cues that animals recognise as attractive; (2) decreasing the suitability of a habitat; or (3) doing both (1) and (2) simultaneously. Alternatively, artificial habitats of high quality, where individuals increase in condition, reproduce better or have improved survival, all of which may ultimately lead to positive population growth rates, act as population sources. Objective testing of whether ecological traps exist is well embedded in the literature concerning terrestrial systems [@pone.0015646-Robertson1], yet few studies have investigated whether they exist in marine environments. Artificial structures that aggregate fish (fish aggregation devices; FADs) have been previously suggested to act as ecological traps by acting as a super-stimulus and misleading fish to make inappropriate habitat selections [@pone.0015646-Hallier1]. Coastal sea-cage fish farms are widespread artificial structures in coastal waters, producing over 2.5 million tons of fish each year [@pone.0015646-FAO1]. They have previously been described as analogous to FADs, attracting and aggregating large assemblages of wild fish in their immediate vicinity [@pone.0015646-Dempster1]. Attraction and aggregation of tons of wild fish to the immediate surrounds of Norway\'s coastal salmon farms [@pone.0015646-Dempster1], [@pone.0015646-Dempster2] meets Robertson and Hutto\'s [@pone.0015646-Robertson1] first condition for the formation of an ecological trap. However, whether the fish farm area is poorer in habitat quality for wild fish than natural adjacent habitats, thus meeting Robertson and Hutto\'s [@pone.0015646-Robertson1] second condition, remains unknown. Relative habitat quality is a key component in determining the extent to which fish farms may act as population sources or ecological traps for wild fish. Along the Norwegian coastline, 1198 coastal sea-cage salmonid farm concessions used 1.2 million tons of fish food to produce 829 000 t in 2008 [@pone.0015646-Norwegian1]. Farming is concentrated in particular fjords, with farms spaced several kilometres apart. Wild saithe are the most abundant species associated with salmon farms within fjord systems [@pone.0015646-Dempster2], [@pone.0015646-Dempster3]. Saithe use farms as a loose network of preferred habitats, moving repeatedly among farms and remaining resident at specific farms for weeks to months [@pone.0015646-Uglem1]. Atlantic cod are also attracted to fish farms in number [@pone.0015646-Dempster2] and may reside in their vicinity for months at a time [@pone.0015646-Uglem2], [@pone.0015646-Uglem3]. Attraction of wild fish to salmon farms is likely to have a range of fitness consequences due to the modified environment fish farms induce, both in the altered trophic network around farms and the close proximity of hundreds of thousands to millions of farmed salmonids. Diet, body fat content, fatty acid composition and parasite loads may all be altered when wild fish closely associate with farms [@pone.0015646-Diamant1], [@pone.0015646-FernandezJover1], [@pone.0015646-FernandezJover2]. Simultaneous analysis of this suite of factors at an extensive number of locations is required to resolve whether farms function as population sources or ecological traps [@pone.0015646-Chalfoun1]. Here, we tested the hypotheses that the diets, indices of condition and parasite loads of cod and saithe associated with salmon farms differed from those of fish present at locations distant from salmon farms. To ensure broad generality of the results, we sampled fish in three intensive fish farming areas along the latitudinal extent of salmon farming in Norway (59°N to 70°N). Materials and Methods {#s2} ===================== Study locations and experimental design {#s2a} --------------------------------------- Saithe and cod were sampled from the three salmon farming areas (Ryfylke 59°N, Hitra 63°N and Øksfjord 70°N) from the same Atlantic salmon (*Salmo salar*) farms and during the same season (summer) as aggregation sizes were determined [@pone.0015646-Dempster2]. Within each salmon farming area, fish were sampled at three farms and two to six non-farm control locations ([Fig. 1](#pone-0015646-g001){ref-type="fig"}). Farm-associated fish were captured within 5 m of cages containing salmon. The number of non-farm locations varied from two to six depending on the area and species of wild fish sampled (Saithe: Ryfylke 2, Hitra 4, Øksfjord 3; Cod: Ryfylke 3, Hitra 6, Øksfjord 3). Control fish were sampled from locations 4 to 20 km distant from the nearest farm ([Fig. 1](#pone-0015646-g001){ref-type="fig"}) to limit the possibility of sampling fish at non-farm locations that had interacted recently with a farm. The 4 km minimum limit was based on telemetry-derived observations of the predominant movements of wild cod and wild saithe [@pone.0015646-Uglem1], [@pone.0015646-Uglem2], [@pone.0015646-Uglem3] in the vicinity of fish farms. {#pone-0015646-g001} All fish were sampled with standardised hook and line fishing gear. Collections by hook and line select for feeding fish, but are more suitable for accurate counts of the number of external parasites than other catch methods such as trawling or gill nets which may remove external parasites through abrasion. Moreover, capture by any other method beside the cages at fish farms is impractical due to possible negative interactions of fishing gear with fish farming structures. Collections were made at each location from June to September 2007 during the period where feed input to salmon farms is high [@pone.0015646-Norwegian1]. Size, diet and condition indices {#s2b} -------------------------------- Upon capture, fish were immediately examined for the presence of external parasites (see parasite sampling section below) and then placed on ice. Fish were weighed and measured to the nearest 0.5 cm (fork length; FL). Each fish was dissected and liver and gonad weights were obtained. Sex for each fish was determined by macroscopic examination of the gonads. In gadoid species, such as Atlantic cod, lipids are stored primarily in the liver [@pone.0015646-Lambert1] making liver weight a measure of spawner quality [@pone.0015646-Marshall1]. Therefore, we calculated three condition indices: body condition, the hepatosomatic index and the gonadosomatic index. Fulton\'s condition index (FCI) was calculated with the formula: FCI = (W/FL^3^)×100, where W = wet weight--stomach content weight and FL = fork length (cm). The hepatosomatic index (HSI) was calculated using the formula: HSI = (LW/W)×100, where LW = liver weight and W = wet weight--stomach content weight. The gonadosomatic index (GSI) was calculated using the formula: GSI = (GW/W)×100, where GW = gonad weight and W = wet weight--stomach content weight. Stomach contents from the foregut were examined and prey species were identified to the lowest taxonomic level possible and weighed. Prey categories were later reduced to 11 for saithe (waste salmon feed, Brachyura, Osteichthyes, Polychaeta, Caridea, zooplankton, Phaeophyceae, Bivalvia (principally *Mytilus* sp.), Ophiuridae, Hydroida (principally *Ectopleura larynx*), and other organic matter) and 13 for cod (waste salmon feed, Brachyura, Osteichthyes, Polychaeta, Caridea, Phaeophyceae, Bivalvia (*Mytilus* sp.), Holothuria, Ophiuridae, Echinoidea, Octopoda, Amphipoda and other organic matter). Parasite sampling {#s2c} ----------------- Fish were examined to estimate the incidence of parasites that may have occurred in increased incidence around fish farms through direct transfer from the farmed salmonids (e.g. mobile sea lice) or through indirect means, such as the modified farm environment increasing the density of con-specific fish or the pool of intermediate hosts available to these parasites, thus increasing their incidence. Immediately upon capture, saithe and cod were examined for the incidence of mobile sea lice (*Caligus* spp.) and attached parasitic copepods (*Clavella* sp.) on all external surfaces, and inside the mouth and gills. In August, 100 mobile sea lice from un-associated (hereafter UA) and farm-associated (hereafter FA) fish were collected in all salmon farming areas to identify the species composition of mobile sea lice. We hypothesised that FA cod and saithe would have elevated levels of *Caligus* compared to UA fish either through direct transfer of adult *Caligus* from caged salmon or elevated levels of *Caligus* larvae in the waters surrounding farms. Gills of cod were examined for the presence and abundance of *Lernaeocera branchialis*, a copepod parasite of cod which invasively attaches to the gills and feeds on blood [@pone.0015646-Khan1]. For *Clavella* sp. and *L. branchialis*, we hypothesised that no differences in infestation levels would be detected between FA and UA fish, as no direct transfer route between salmon farms and wild fish has been established for these parasites. Livers were dissected from both species of fish and inspected for the third stage (L3) larvae of the parasitic nematode *Anisakis simplex* [@pone.0015646-Klimpel1]. Infection intensity was scored on a semi-quantitative scale form 0 to 3: 0 = *A. simplex* absent; 1 = mild infestation; 2 = moderate infestation; and 3 = heavy infestation. We hypothesised that L3 larvae of *A. simplex* would be less abundant in FA than UA fish as high consumption of lost feed at farms would mean lower consumption of natural prey items such as crustaceans, squid and fish, which may contain L3 larvae. Statistical analyses {#s2d} -------------------- As gonadal development was minimal during the non-spawning season sampling period and diets in the non-spawning season are not known to vary among male and female cod and saithe, we pooled the sexes for dietary analyses and analyses of condition. Further, as differences in the incidence of parasites among male and female gadoids have rarely been found [@pone.0015646-Hemmingsen1], and no differences are known for the parasite species investigated here, we pooled the sexes for parasite analyses. Non-parametric multivariate techniques were used to compare dietary compositions among farm and non-farm locations. All multivariate analyses were performed using the PRIMER statistical package. Prior to calculating the Bray-Curtis similarity matrices, the dietary data were pooled across all individuals sampled within each location and month by summing the total weights of prey items within each prey category to reduce the stress of MDS representation. Fourth root transformations were made to weigh the contributions of common and rare dietary categories in the similarity coefficient. Non-metric multidimensional scaling (nMDS) was used as the ordination method. Variables that had more influence on similarities within groups and dissimilarities among groups of locations or depths, determined by ANOSIM (analysis of similarity), were calculated using the SIMPER (similarity percentages) procedure. The ANOSIM permutation test was used to assess the significance of differences among farm and non-farm locations. As diets of both saithe and cod at farms contained feed pellets, we repeated all analyses with this prey category removed to determine if differences in diet among farm and non-farm locations remained significant. To test for differences in fish size (fork length; FL), stomach content weight, FCI, HSI, GSI and the incidence of the various parasites among farm and non-farm locations in each of the three fish farming areas, we used Generalized Linear Models (GLMs). Prior to the GLMs, heterogeneity of variance was tested with Cochran\'s *C*-test. Data were ln(*x*+1) transformed if variances were significantly different at p = 0.05. Comparisons across fish in all size classes were made in each of the three farming areas for cod. To ensure that any differences detected in comparisons were not related to the different sizes of fish in the FA and UA treatments, we used FL as a co-variate in analyses of stomach content weight, condition and parasite loads. For saithe, as HSIs\>10% are indicative of a waste feed dominated diet for several months and wild saithe fed solely on natural diets do not have HSIs\>10% [@pone.0015646-Gjster1], we tested if the incidence of the various parasites differed among FA fish with HSIs\>10%, FA fish with HSIs\<10%, and UA fish with HSIs\<10%. To detect if the parasite loads we detected were related to the body condition (FCI) of wild fish, we applied multiple regression analysis for both cod and saithe. Results {#s3} ======= Size structures of farm-associated and un-associated fish {#s3a} --------------------------------------------------------- In total, 355 FA and 215 UN saithe were captured at sizes ranging from 21.5--108.5 cm fork length (FL) and weights from 0.1--12.5 kg. 171 FA and 178 UA cod were collected at sizes ranging from 28.5--121.0 cm FL and weights from 0.23--18.0 kg. Saithe were captured at all farms, while cod were only available at 8 of the 9 farms (all except one farm at Hitra). Significant differences were detected in mean fork lengths among UA and FA groups for both species in all three farming areas ([Table 1](#pone-0015646-t001){ref-type="table"}). FA saithe were larger than UA saithe at two of the three farming areas (Hitra and Øksfjord), but significantly smaller at Ryfylke. FA cod were significantly larger than UA cod in Ryfylke and Hitra but not Øksfjord, and the magnitude of the difference varied greatly among the areas. 10.1371/journal.pone.0015646.t001 ###### Mean sizes of samples of saithe (*Pollachius virens*) and Atlantic cod (*Gadus morhua*) used to compare diet, condition and parasite loads across farm-associated (FA) and farm unassociated (UA) locations in each of the three Norwegian salmon farming areas. {#pone-0015646-t001-1} FA/UA Ryfylke Hitra Øksfjord ----------------- ------- --------- ----------------- ---------- ----------------- ----- ----------------- ***P. virens*** FA 97 50.1±0.7**^b^** 148 40.2±1.2**^a^** 110 46.2±0.8**^a^** UA 30 54.2±0.9**^a^** 88 34.3±1.1**^b^** 97 43.4±0.8**^b^** ***G. morhua*** FA 13 63.3±4.5**^a^** 89 52.8±1.6**^a^** 65 62.3±2.5 UA 12 46.7±4.4**^b^** 75 45.7±1.9**^b^** 91 58.5±1.5 Superscripts (^a,b^) indicate a significant difference was detected between the FA and UA groups at p\<0.05. Diets of farm-associated and un-associated fish {#s3b} ----------------------------------------------- Saithe captured from non-farm locations had a higher proportion of empty stomachs (31%) and lower average stomach content weight (8.6 g) compared to FA saithe (16%, 20.2 g). For cod, both FA (18%) and UA fish (19%) had similar proportions of empty stomachs, although stomach content weight was higher in FA (32.9 g) than UA fish (23.2 g). 44.3% of saithe and 20% of cod captured around farms had waste feed in their stomachs. Overall, waste feed accounted for 71% (14.2 g) and 25% (8.3 g) of the diet by weight of FA saithe and cod, respectively. The 2-dimensional nMDS plot based on weights of prey groups by location and month revealed clear separation of the diets of FA and UA fish for both saithe ([Fig. 2a](#pone-0015646-g002){ref-type="fig"}) and cod ([Fig. 2b](#pone-0015646-g002){ref-type="fig"}). ANOSIM indicated that differences in diets between FA and UA fish were significant (saithe: *R* ~global~ = 0.69, p = 0.001; cod: *R* ~global~ = 0.45, p = 0.003). When pellets were removed from the analysis, differences in diets between FA and UA fish remained significant (saithe: *R* ~global~ = 0.52, p = 0.01; cod: *R* ~global~ = 0.38, p = 0.02). {#pone-0015646-g002} Diets of FA saithe clustered together, regardless of sampling location and month, while diets of UA saithe were more variable ([Fig. 2a](#pone-0015646-g002){ref-type="fig"}). UA saithe diets were characterised by similar weights of relatively few dietary items. Over 70% of group similarity was accounted for by fish (41.5%), zooplankton (16.8%), crustaceans (8.0%) and ophiuroids (4.5%). Over 80% of similarity in FA saithe diets was due to waste feed (45.7%), fish (14.8%), mussels (10.5%) and zooplankton (10.2%). Dissimilarities in diets between UA and FA saithe were due to large differences in the abundance of a few of the major items (waste feed 32.3% F\>C, fish 14.3% C\>F, zooplankton 10.3% F\>C and mussels 9.9% F\>C). Similarities in UA cod diets were predominantly due to similar weights of fish (39.7%), crabs (24.3%), ophiuroids (9.7%) and crustaceans (6.5%) while similarities in FA cod diets were predominantly due to fish (37.6%), polychaetes (19.6%), pellets (14.6%) and crabs (9.6%). FA cod consumed more waste feed, polychaetes and fish (dissimilarities of 18.9%, 12.1% and 7.8%, respectively) while UA cod consumed more Ophiuridae, crabs and mussels (dissimilarities of 11.5%, 9.9% and 7.8%, respectively). Body, liver and gonad condition of farm-associated and un-associated fish {#s3c} ------------------------------------------------------------------------- FA saithe had significantly higher average FCIs (1.06-1-12 times) than UA fish in all three farming areas ([Fig. 1](#pone-0015646-g001){ref-type="fig"}, [Fig. 3a](#pone-0015646-g003){ref-type="fig"}). Average HSIs were significantly higher in saithe (1.4--1.8 times) collected around farms compared to UA fish at Hitra and Øksfjord ([Fig. 3b](#pone-0015646-g003){ref-type="fig"}). No difference was detected for FA and UA fish sampled from Ryfylke. As the June-September sampling period occurred after the main spawning period for saithe and many of the individuals sampled were less than 2 kg in size and thus likely to be immature, no difference was detected in average GSIs among farm and UA fish in any of the three areas ([Fig. 3c](#pone-0015646-g003){ref-type="fig"}). {#pone-0015646-g003} FCIs, HSIs and GSIs of cod were clearly affected by association with salmon farms. FCIs were consistently 1.06--1.11 times greater in FA than UA cod in all three areas ([Fig. 3d](#pone-0015646-g003){ref-type="fig"}). Similarly, average HSIs varied among the three areas, but were consistently 2.0--2.8 times greater in cod collected around farms compared to UA fish ([Fig. 3e](#pone-0015646-g003){ref-type="fig"}). In contrast to saithe, where average GSIs in FA and UA fish were similar, average GSIs in cod were significantly greater (1.7--4.8 times) in FA than UA cod in all three areas, despite the timing of sampling in the post-spawning period ([Fig. 3f](#pone-0015646-g003){ref-type="fig"}). Parasite loads of farm-associated and un-associated fish {#s3d} -------------------------------------------------------- Significant differences in the abundances of parasites were detected in both directions, with FA or UA fish having greater levels of particular parasites in certain fish farming areas. From the collections in August, two species of mobile sea-lice were identified on both saithe and cod in all three areas: *Caligus elongatus* and *C. curtus*. Significantly higher numbers of sea lice occurred on FA saithe with HSIs\>10 or \<10 compared to UA fish with HSIs\<10 at Ryfylke (2.5 to 3.5 times) and Hitra (3.1 to 3.7 times), but not at Øksfjord ([Fig. 4](#pone-0015646-g004){ref-type="fig"}). *Clavella* sp. abundances were significantly higher in FA saithe with HSIs\>10 or \<10 compared to UA fish at Hitra (1.8 to 2.1 times). FA saithe with HSIs\>10 had 2.6 to 3.6 greater abundances of *Clavella* sp. than both FA saithe with HSIs\<10 and UA fish in Øksfjord. No differences in *Clavella* sp. abundance among the three groups were detected at Ryfylke. For the *Anisakis simplex* index, FA saithe had consistently lower values that UA saithe across the three locations. FA saithe with HSIs\>10 had 1.6 to 2.1 times lower *A. simplex* infestations than FA saithe with HSIs\<10 and UA fish. {#pone-0015646-g004} *Caligus* spp. occurred in abundances 2.4 times higher on FA cod at Øksfjord compared to UA cod, whereas no significant differences between FA and UA cod were detected at Ryfylke and Hitra ([Fig. 5](#pone-0015646-g005){ref-type="fig"}). No significant differences were detected for *Clavella* sp. or *Anisakis simplex* L3 larvae between farm-associated and UA fish in any of the three areas. The gill parasite *Lernaeocera branchialis* occurred in significantly higher abundance (2.8 times) in UA cod than FA cod in Øksfjord, with no difference detected in the other two areas. {#pone-0015646-g005} Multiple regression analysis of parasite loads versus body condition revealed that none on the four species of parasites investigated for cod were significantly related to FCI (F = 1.12, p = 0.35; R^2^ = 0.02; *Caligus* spp.: p = 0.11; *Clavella* sp.: p = 0.17; *L. branchialis*: p = 0.86; A. *simplex*: p = 0.49). For saithe, the multiple regression was significant (F = 9.7, p\<0.001; R^2^ = 0.05), with *Clavella* spp. positively related to FCI (p = 0.003), *Anisakis* sp. strongly negatively related to FCI (p\<0.001), and no relationship evident for mobile sea lice (*Caligus* spp.: p = 0.59). Discussion {#s4} ========== Proxy measures of fitness of farm-associated and un-associated fish {#s4a} ------------------------------------------------------------------- We have demonstrated that proxy measures of fitness (FCI, HSI, abundances of specific parasites and diet) of wild saithe and cod caught in close association with salmon farms differ significantly from their counterparts captured distant from farms. These effects are likely to be general across the spatial extent of salmon farming in Norway (59°N--70N°) and apply to a substantial pool of fish aggregated around farms. Dempster et al. [@pone.0015646-Dempster2] conservatively estimated that over 12000 tons of wild fish, principally saithe and cod, were aggregated at Norway\'s 1198 salmon farms on any given day in summer based on video-derived estimates of aggregations at the same 9 farms investigated here. Conclusions derived from this study are therefore based upon these abundance estimates, but are limited to the summer months during which samples were taken. Salmon farms clearly increased the amount of food consumed by closely associated saithe and cod, indicating a strong trophic link between farms and wild fish. Stomachs of FA saithe contained more than twice the amount of food by weight than UA fish with stomach content weight similarly elevated in FA cod (1.4 times). Food pellets are high in fish proteins and oils and thus provide a high energy source of feed [@pone.0015646-Tacon1], although with distinctly different fatty acid distributions from natural diets [@pone.0015646-FernandezJover1]. While waste feed dominated diets of FA saithe and cod, the composition of dietary items still differed among FA and UA fish when waste pellets were removed from analyses, indicating that the availability of other types of prey differed between farm and non-farm locations. Salmon farms are known to have modified meio- and macro-fauna communities [@pone.0015646-Kutti1] and modified fish assemblages [@pone.0015646-Dempster2] compared to control locations, which likely contributed to the dietary differences. The increased body and liver condition observed in FA saithe and cod is likely linked to the trophic subsidy that farms provide. Livers are the principal lipid and thus energy stores in gadoids [@pone.0015646-Lambert1]. High HSIs are indicative of high total lipid energy, which is known as a direct proxy to egg production in gadoid fish [@pone.0015646-Marshall1]. Moreover, lipid energy reserves 3--4 months prior to spawning are the best proxy for fecundity [@pone.0015646-Skjraasen1]. In this context, association with fish farms throughout summer and autumn could increase the fecundity of saithe and cod, which spawn in early spring, even if these fish migrate away from farms months prior to spawning. While fecundity, in terms of egg numbers or size, may increase through FA fish having high energy reserves, the composition of stored lipids in FA saithe and cod may differ from those of UA fish which consume a natural diet (Fernandez-Jover et al. unpubl. data). This may effect egg quality as farm-feeds contain low proportions of highly unsaturated fatty acids (HUFAs) and arachidonic acids, which are key to fertilization rates and egg quality [@pone.0015646-Salze1]. If the waste-feed dominated diet alters the fatty acid composition of saithe and cod livers and has a negative effect upon egg quality during vitellogenesis, the increased condition evident in FA fish may not translate to a proportional increase in spawning success. Experimental manipulations of wild saithe and cod fed diets containing different proportions of waste feed for various durations and the subsequent evaluation of the effect this has on egg and larval quality are required to determine the extent of this potentially negative effect. Some parasites were found in elevated abundances in FA fish. We hypothesised that mobile sea-lice would occur in higher abundances on FA fish due to direct transfer or greater infestation levels as larvae occurred in greater abundance. This was the case for saithe at Hitra and Ryfylke and cod in Øksfjord. Similarly the attached copepod *Clavella* sp. was detected in elevated abundances in FA saithe at Hitra and Øksfjord. In contrast to the mobile sea-lice and *Clavella* sp. loads, the gill parasite *Lernaeocera branchialis* and the internal parasite *Anisakis* simplex were only ever detected in lower levels in FA fish. For *L. branchialis*, significant differences between FA and UA fish were only detected in Øksfjord, where UA cod had higher levels. Significantly lower *A. simplex* infections occurred in FA saithe with HSIs\>10 in all three farming areas, suggesting that the longer-term residence at salmon farms required to generate an HSI\>10 [@pone.0015646-Gjster1] plays an important role in reducing the level of *A. simplex* infection. The strong trophic link between saithe and fish farms, with saithe diets containing \>70% by weight of lost feed pellets which are free of *A. simplex* A3 larvae, reduces the amount of potential hosts of A3 larvae such as small fish and crustaceans that saithe consume [@pone.0015646-Klimpel1]. Elevated levels of *Caligus* spp. and *Clavella* sp. detected in FA fish may have had detrimental effects upon condition. Limited information exists to assess the threshold levels at which *Caligus* spp. and *Clavella* sp. infestations cause reductions in condition in cod and saithe, although heavy infestation of *Clavella adunca* can produce a moderate reduction in cod condition [@pone.0015646-Hemmingsen1]. However, mobile sea-lice infestations of gadoids were generally close to the range of those typically recorded in Norwegian fjord and coastal waters (1 to 2 *C. elongatus* gadoid^−1^; [@pone.0015646-Heuch1]). *L. branchialis* is considered the most serious metazoan parasite of wild cod [@pone.0015646-Khan1], [@pone.0015646-Hemmingsen1] and can cause mortality, loss of condition and affect reproductive output. Similarly, heavy *Anisakis simplex* infestation has the capacity to reduce the condition of wild gadoids [@pone.0015646-Hemmingsen1]. The reduction of both of these parasites in FA fish at some locations was therefore likely to have led to increased average condition compared to control fish. However, multiple regression analyses revealed that farm-modified parasite loads did not have major effects on the somatic condition of cod. For saithe, *Clavella* sp. abundance was positively correlated with condition, while the *A. simplex* infestation index was strongly negatively correlated with condition. Regardless of these relationships, body condition was significantly higher for FA fish than UA fish for both cod and saithe across all farming locations. As the body condition index integrates all factors that influence the condition of a fish over its recent life history, including the effects of parasites upon condition, our data suggests that the trophic subsidy that farms provide elevates body condition such that any effects on condition related to modified parasite loads were negligible in comparison. In addition to the parasite species investigated here, gadoid fish such as *Gadus morhua* and *Pollachius virens* are infected by over 100 pathogens and parasites, at least 20 of which may be directly transferred among salmonids and gadoids [@pone.0015646-Bricknell1]. These include some of the most significant diseases prevalent in salmon aquaculture, including *Vibrio anguillarum*, salmonid alphavirus and infectious pancreatic necrosis virus [@pone.0015646-Graham1], [@pone.0015646-Wallace1]. If these pathogens are enhanced in wild fish aggregated at fish farms they could negatively affect condition and survival; further research is required in this field. Fish farms: ecological traps or population sources for wild fish? {#s4b} ----------------------------------------------------------------- In contrast to the detrimental effects of salmon farming detected at the population-level for wild salmonids (sea lice: [@pone.0015646-Krkoek1], [@pone.0015646-Costello1], [@pone.0015646-Krkoek2]; escapes: [@pone.0015646-Thorstad1], [@pone.0015646-Jensen1]), we did not detect significant negative effects of the co-occurrence of wild saithe and cod with salmon farms. The diet and condition data indicate that wild saithe and cod benefited from their associations with salmon farms through access to greater amounts of food which translated to enhanced condition. While *Caligus* spp. and *Clavella* sp. loads were elevated at some farming locations compared to controls and *Anisakis* sp. and *Lernaeocera branchialis* loads were lowered at some farming locations compared to controls, it appears that any effects these modified parasite loads may have had on the condition of wild cod and saithe were overridden by the trophic subsidy that farms supply. The results provide no evidence that salmon farms act as ecological traps for wild cod and saithe that aggregate in their vicinity, provided that: 1) the modified fatty acid distributions and elevated organohalogen levels in fat stores in livers that results from a fish farm modified diet [@pone.0015646-Bustnes1], [@pone.0015646-FernandezJover3] does not negatively affect physiological processes, vitellogenesis or egg and larval quality; 2) salmon farms do not amplify any of the numerous pathogens not investigated here that salmonids and gadoids share [@pone.0015646-Bricknell1]; and 3) that attraction to farms does not disrupt natural spawning migrations or behavior. Future research should seek to discern the effects of both salmon and cod farms during the spawning season for cod resident in fjords containing farms, as a range of different effects are possible during this period, including mass spawning of farmed cod in cod farms [@pone.0015646-Jrstad1] and possible avoidance of fjords containing salmon farms by spawning cod [@pone.0015646-Bjrn1]. As saithe and cod condition is enhanced by farms, an opportunity exists to protect wild fish around salmon farms where they are aggregated and vulnerable to fishing, thus enabling farms to act as a population source. Presently, fishing adjacent to salmon farms occurs [@pone.0015646-Maurstad1], although the importance of this activity to overall catches is unknown. Stocks of fjord cod in southern Norwegian waters, in particular, are depressed due to chronic overfishing [@pone.0015646-Berg1]. Therefore, to ensure farms do not act as ecological traps for cod via increased fishing mortality alone, restrictions on the fishing of cod in the vicinity of farms could be introduced. Spatial protection from fishing would allow an opportunity for the enhanced condition that cod and saithe generate due to their association with salmon farms to translate to enhanced spawning success. Fish farms have recently become targets of significant fishing pressure [@pone.0015646-Akyol1] in other coastal ecosystems, thus the principle of restricting fishing around farms to ensure they do not function as ecological traps may be broadly applicable. As large, multi-specific aggregations of wild fish aggregate around coastal fish farms wherever they occur [@pone.0015646-Dempster1], [@pone.0015646-Dempster2], we predict that significant conservation benefit would be derived through the protection of tens of thousands of tons of wild coastal fish in high spawning condition if this measure were implemented worldwide. We thank Dr Bengt Finstad (NINA) who assisted in identifying mobile sea lice. We thank the salmon farming companies Marine Harvest, Greig and Lerøy Midnor for access to farm locations throughout the study. **Competing Interests:**SINTEF Fisheries and Aquaculture is a private, not-for-profit research institute. It is not a commercial funder of this research. The authors have declared that no competing interests exist. **Funding:**Funding was provided by the Norwegian Research Council Havet og kysten program to the CoastACE project (no: 173384). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: TD PSJ IU. Performed the experiments: TD PSJ DFJ JBS RN PAB IU. Analyzed the data: TD. Contributed reagents/materials/analysis tools: TD. Wrote the paper: TD.

1. Introduction =============== Type 1 diabetes (T1D) is an incurable autoimmune disease that destroys the pancreas's ability to naturally produce insulin, a hormone necessary for the absorption and storage of glucose. Failure to adequately monitor and manage blood glucose and insulin levels can lead to major vascular complications and organ damage from hyperglycemia \[ [@JR0047-1] \] on the one hand and seizures, loss of consciousness, and death from hypoglycemia on the other \[ [@JR0047-2] \]. In the United States, over a million adults live with T1D, and recent estimates are that over 200,000 youth (individuals under the age of 20) live with it as well \[ [@JR0047-3] \]. While the United States Food and Drug Administration has recently approved an "artificial pancreas" device that will automatically manage blood sugar and release insulin into the bloodstream of a person with diabetes \[ [@JR0047-4] \], its relative newness, cost, and the requirement that patients be at least 14 years old limit its accessibility to a substantial portion of the T1D population. An alternative that some diabetics have pursued is the creation of their own insulin management system using resources such as OpenAPS (Open Artificial Pancreas System, openaps.org). To date, users of OpenAPS have self-reported improved glycohemoglobin levels and increases in the amount of time that they were in a desired blood-sugar range. These same users observed, however, that the amount of technical know-how and efforts needed to create and maintain their self-developed artificial pancreas system are likely substantial obstacles that will limit broader adoption \[ [@JR0047-5] \]. As such, diabetes management for most individuals with T1D involves frequent manual blood tests to measure the amount of sugar in one's blood coupled with the manual administration of insulin multiple times during the day. In some cases, devices such as insulin pumps or continuous glucose monitors (CGM) may be used. Such devices typically obtain and store several days' worth of specific data about how much insulin was administered and the blood sugar level readings for a patient over certain time intervals. Regardless of whether readings and administrations are done manually with a glucometer or with a pump or CGM, there are many measures being obtained by a person with diabetes even during a single day. What happens to those measures depends on the individual collecting the information and the technological infrastructure that they use, with some data being stored briefly mentally and others on proprietary cloud systems. This current paper is a case study showing the comparative realizations that were made by a T1D-affected family upon examination of aggregated blood sugar information obtained using two data collection methods. One method was manual testing involving regular finger pricking and use of a separate glucometer accompanied by systematic handwritten logging. The other involved use of a CGM device. The diabetic individual whose data are examined and shared in this report is the third author (Chris), who explicitly asked (and with consent of his parents, as he was 9 years old at the time of this writing) that his identity be disclosed. 2. Objectives ============= The primary objective of this report is to illustrate what regularities are detected in aggregate visualizations of different blood glucose data collection methods of end users. In this case, the end users are the child with diabetes and his parents. One data collection method involved manual data collection and recording over the course of three and a half years, and the other involved data collected by a CGM. With this family, manual data collection turned out to exhibit blind spots due to daily family routines and transition to formal schooling that were only fully realized when data were aggregated. On the other hand, automated data collection created more data but tended to invite only examinations of data from preceding hours but not aggregate analysis for recurring daily patterns. 3. Methods ========== The quantitative data for this report were obtained by the second and third author and other family members prior to any plans for a research study. As such, this is a post hoc analysis of existing data. Since those who collected data are co-authors and written records exist as a paper trail, we are able to describe the data collection process with some accuracy. The diabetic individual, Chris, had been diagnosed at approximately 20 months of age. His parents (including his father, the second author), began to prepare journals of Chris's blood sugar data from the time of his diagnosis (August, 2010). The written log initially included records of how much of which insulin type was administered and when along with a detailed specification of what foods were eaten and the equivalent number of carbohydrates that would lead to an increase in blood glucose. The log (see ► [Figure 1](#FI0047-1){ref-type="fig"} ) served as the sole record of Chris's blood sugar information in the family, and served as a boundary object \[ [@JR0047-6] \] around which Chris, his parents, and ultimately health care practitioners, would discuss recent trends observed over past days and make necessary adjustments. The records that were manually recorded by Chris's family were considered to be quite thorough relative to many other patient families and were important resources for both the parents and health care providers to examine when planning meals, activities, and insulin dosage. {#FI0047-1} The journals changed in structure and content over time as carbohydrate and serving size information was internalized by family members, and they stopped logging that information. Eventually, only blood sugar levels and times of day were recorded. As part of another study related to educational and learning activities that take place by way of self-quantification in and out of the classroom \[ [@BR0047-7] -- [@JR0047-9] \], these journals were introduced to the first author. With the consent of Chris and his parents, the glucose measurements were digitized by the first author and research assistants. One month of data (March, 2013) was missing. In total, there were 7,437 entries from all the journals from August 2010 to February 2014. Details about the numbers of readings are provided in ► [Table 1](#TB0047-1){ref-type="table"} . ###### Number of digitized handwritten glucose data measurements per year. **Year** **2010** **2011** **2012** **2013** **2014** ------------------------------------------ ---------- ----------- ----------- ----------- -------------------- **Total Glucose Entries** 617 2595 2184 1899 142 **Time Period** 4 months 12 months 12 months 11 months 1 month and 5 days **Estimated number of readings per day** 5.14 7.21 6.07 5.75 4.06 Family circumstances changed in February, 2014 that made continued handwritten journaling difficult (February, 2014). Blood sugar readings were still obtained but not retained in a permanent record. In summer of 2014, Chris's parents were able to purchase a CGM device and switched to it as the primary source and repository of blood glucose information. None of the CGM data were collected with the intention of performing a pre-planned self-experiment or to be used for third party analysis. The CGM data were obtained through a Dexcom G4 Platinum (► [Figure 2](#FI0047-2){ref-type="fig"} ), which was configured to obtain a blood glucose reading every five minutes. The Dexcom unit was linked to Chris's parents' smart-phones so that they could be alerted to extraordinary deviations from desired blood sugar ranges (typically between 80 and 180 mg/dL). Because several dozen data points were accessible to Chris and his parents on an ongoing basis, these blood sugar data were not manually recorded by any individual. The information available when they were checking immediate blood sugar levels and the previous few hours was deemed adequate for routine in-the-moment decision making on their part with respect to deciding on Chris's insulin dosage. {#FI0047-2} Due to some misunderstandings about how and for how long data were stored by the device, the readings from the Dexcom were not downloaded to be viewed or retained separately by the family until October of 2016, when the family switched endocrinologists (due to retirement of their longstanding provider) and were explicitly requested by the new attending endocrinologist to export the Dexcom data and bring specific reports to their first consultation. For the purposes of this paper, 14 days of CGM data are considered. That is the equivalent of 6,720 blood sugar data points. This number of days is considered for two reasons. One is that after preparing and reviewing an aggregate visualization of these fourteen days, Chris's family noticed clear evidence of a regularity that they wanted to change for Chris's benefit. Thus, those 14 days were prior to any intervention that was meant to mitigate the detected regularity. Secondly, this number of data points is comparable to the number of data points obtained in the manual record, albeit over a different time scale and density. By comparing roughly equal numbers of manually and CGM collected data points, we feel that the density of sampling and the inferences made from large numbers of visualized data points becomes foregrounded in our analysis rather than the absolute number of measurements. In addition to the aforementioned numerical data, the first author met with Chris and his family on three occasions for 1--2 hours at a time to conduct open interviews with them about their data collection and experience with tracking and managing Chris's blood sugar. Two interviews were audiorecorded at Chris's home. One interview was recorded with handwritten notes as it took place in a public venue. These interviews form the basis of determining the family's response and impressions of the data visualized from these two tracking methods. These were also supplemented with some additional qualitative records such as email correspondence between the parents and the first author. The overarching study and the interview procedures described had been approved by an institutional review board, and all necessary consent and assents were obtained in writing. 4. Results ========== 4.1 Observed Features of Manually Collected Data Visualized ----------------------------------------------------------- A plot prepared in Tableau of all manually obtained glucose readings over time of day across all days in the journals appears in ► [Figure 3](#FI0047-3){ref-type="fig"} . A secondary set of plots for years 2010--2013 (2014 was excluded because it was far fewer data points) showing the number of readings during each fifteen minute time interval (dots) and the computed average blood sugar level given all records during the associated 15 minute time interval in the associated year (line) is provided in ► [Figure 4](#FI0047-4){ref-type="fig"} . {#FI0047-3} {#FI0047-4} From the audio and video records of the family's examination of all aggregated manually obtain data, it was clear to Chris's family that breakfast, lunch, and dinner were all heavily sampled times. This was reflected by the dark bars in the times roughly corresponding to 8:00 AM, 12:00 PM, and 7:00 PM. The evenings after 7:00 PM and up to roughly 11:00PM, when both parents were home, had more samples obtained than other four-hour time blocks including the time between breakfast and lunch and the time between lunch and dinner. They recognized that this was because of post-dinner snacking and the family was conscientious about monitoring blood sugar immediately before and after eating and in preparation of sleep. Chris's morning readings obtained around breakfast time tended to have higher blood glucose levels than later ones, which they retroactively attributed to having just woken up and having generally more physically active afternoons with recreational activities such as preschool soccer programs. There were not as many, but still several, readings in the middle of the night between midnight and 7:00 AM. As Chris's parents described, if there were unusual readings prior to bedtime or if they had an intuitive hunch that it was possible for his blood sugar to be too low, they would do a prick test themselves in the middle of the night. Chris's parents believe that there may have been more tests than reflected in ► [Figure 2](#FI0047-2){ref-type="fig"} and [Figure 3](#FI0047-3){ref-type="fig"} , but since it was the middle of the night, they recalled that they would overlook recording the middle of the night blood sugar readings especially if they were within what they considered to be an acceptable range. Based on their examination of ► [Figure 3](#FI0047-3){ref-type="fig"} , Chris's family did not feel there were any clear glucose level trends that could be ascertained. Since the data were separated by years and by times during the day, they were aware that outlier sensitivity could make some days' readings appear high. Thus, it was not obvious there was a regularity in blood sugar levels. However, they did notice that in 2013 there were two times in the plot when their appeared to be a substantial increase in the number of blood sugar readings relative to other years. One of those times was about 10:30 AM, which in 2013 had 48 readings. In other years it had at most just a few. The other increase was at 8:00 PM which had 112 readings whereas in previous years there would usually be between 70--80 readings. Upon reflection, Chris's family realized that 2013 was the year that Chris began Kindergarten. For over three months, his daily schedule became structured around the school schedule, and 10:30 AM corresponded to the time just before his school's morning recess. That was a consistent set time for a blood sugar reading at school. The 8:00 PM readings were attributable to a strict bedtime that was followed because of needing to wake up at a certain time for Kindergarten. Chris was the oldest child in the family and thus the first to have their home schedule affected by a public school schedule. 4.2 Observations from CGM Collected Data Visualized --------------------------------------------------- With the Dexcom CGM data, the family reported relying most often on the numerical display on the CGM device and the ongoing reports from the Dexcom mobile device app (► [Figure 2](#FI0047-2){ref-type="fig"} ). As parents of multiple children and one parent working outside of the home full-time and pursuing graduate studies part time, the just-in-time access to immediate information was considered the most pressing and useful. Chris's family had access to the Dexcom Clarity software, which produced data visualizations like the one shown in ► [Figure 5](#FI0047-5){ref-type="fig"} and [Figure 6](#FI0047-6){ref-type="fig"} . However, these were unused until they were switched to a new endicronologist who asked specifically for certain reports from the Clarity software. {#FI0047-5} When the aggregated two weeks of Dexcom blood sugar data were examined as prepared in the Clarity software, the family noticed two clear, consistent, and surprising regularities (► [Figure 5](#FI0047-5){ref-type="fig"} ). One was that there was a steady increase in blood sugar levels in the morning between breakfast and lunch. As the previous manual sampling did not heavily sample during that time period, this regularity was not previously detected. This was consistent with another visualization of daily data provided by the Clarity visualization tool (► [Figure 6](#FI0047-6){ref-type="fig"} ). Together, this morning increase which they believe had been present for quite some time since they did not perceive major noteworthy changes to Chris's daily insulin, dietary, or physical activity routines between when they had stopped maintaining paper journals and when they began to use the CGM. This led to an immediate change in when insulin would be administered in the morning with the hope of making that blood sugar increase go away. {#FI0047-6} The other regularity that was noted by Chris's family was that the nighttime glucose levels were relatively high compared to awake hours. Chris's parents retroactively attributed this to a cautious stance of theirs in which they tended to prefer he run high rather than low at night, as falling into a coma and already being asleep when that happens is a major concern for people with diabetes. Again, night time was not heavily sampled or recorded by the manual method. These relatively high blood sugar levels were not a regularity that the family had monitored, and they have since been re-evaluating how to help Chris manage his blood sugar at night. 5. Discussion ============= Through this juxtaposition of comparable numbers of blood sugar measurements, we observe that different features of youth diabetic blood sugar measurements that were detected and interpreted differently by the family. First, in the aggregated view of the manual data recordings, periods of frequent and regular sampling were revealed. Sampling was driven largely by eating, formal school schedules, and both parents being in immediate proximity to Chris. This sampling approach inadvertently created data blind spots for the family. The continuous glucose monitoring visualizations revealed that some of these blind spots were consequential. There were two periods of time (early morning prior to awakening and the hours between breakfast and lunch) that were previously undersampled through manual approaches, and those times showed higher blood sugar levels than what Chris or his parents desired or expected. This led to a change in treatment regimen, discussed generally with their endocrinologist and refined through daily family practice, that was informed by this new visualization. However, that new CGM visualization was only obtained and reflectively examined when the family was called upon by a new doctor to provide a particular report using the Clarity tool. What was much more common and more accessible for the moment-to-moment decision making by individual members was independent examination of the most recent numerical readouts from Chris's CGM and mobile apps that only showed a partial window of the typical day. In retrospect, the family reported that the ease and frequency of immediate and abundant data access seemingly contributed to a sense that they knew more about what was happening with Chris's blood sugar already at any given moment. Until they switched endocrinologists, they were not otherwise being cued to produce aggregate reviews of his data, and operated off of partial (but to their view, seemingly abundant) blood sugar information. Thus, while they had the ability to examine blind spots in detail, there was little prompting that necessitated such examinations amidst the hustle and bustle of daily family life. As the opportunity for new technologies and more data obtained through a range of biosensors beyond CGM become increasingly available to aid in the management of chronic conditions like T1D \[ [@JR0047-10] \], such considerations for how such data will be routinely reviewed and used by families remain important for improving youth health management. 6. Conclusions ============== In conclusion, there are two major lessons embedded in this case study. One is that day-to-day routines and childhood school day schedules can affect how data are obtained for a T1D youth. This became apparent in this case with a family that diligently maintained a multiyear written record of blood sugar test results. The aggregation of these years of records revealed blind spots within a systematically collected personal data archive. The other lesson is that while new digital tools might have embedded features and accessories that allow families to inspect and review aggregated blood sugar data, there is no assurance that such features will be organically used. Studies of user-technology interaction consistently demonstrate that just because functionality exists, users may not know about or access it \[ [@BR0047-11] \] and the addition of just a few extra cognitive actions -- in this case, exporting data, switching to different view settings, and recognizing that there were times that were previously overlooked -- can implicitly dissuade pursuit of new routines \[ [@BR0047-12] \]. Considering these features went unused by a highly motivated family who took the time and effort to systematically record years of bloodsugar data and otherwise educate themselves and others about T1D, this suggests that those interested in medical information and personal information management should consider when and how data collection and reflection can be better motivated so that these blind spots are better examined in situ. Automated tools and resources such as artificial pancreas systems or CGM technology may be attractive because they have the potential of obtaining and storing large amounts of frequently obtained data. However, better prompting or cueing for examination of data visualization features should be encouraged for the collected data to be made maximally informative to families and their healthcare providers. This could take the form of automatic reports being generated and appearing as alerts in their applications or in creating expectations of standard data visualization forms that are to be shared between patients and healthcare providers. Designers and tool developers should also be aware that users can mistakenly take large amounts of recent data to be alone sufficient when views of data at multiple time scales would be more informative for better health management. The authors acknowledge the assistance of Joel Drake, Liam Fischback, and Kourtney Schut in assisting with data digitization and the Thurston family for sharing their data.

1. Introduction {#sec1} =============== According to the commentary on the 2008 WHO classification of lymphoid tumors, follicular lymphoma *in situ* (FLIS) refers to lymph nodes with a background of hyperplastic germinal centres harbouring distinct areas with Bcl-2 overexpression in centroblasts and centrocytes \[[@B1]\]. In FLIS, there is a B-cell population with immunophenotypic and genotypic features of follicular lymphoma; however, these B-cells are exclusively localised to germinal centres in morphologically reactive lymph nodes \[[@B2]\]. FLIS was first recognised in 2002 \[[@B3]\]; however, it is not presently clear cut whether this condition is a precursor to full blown follicular lymphoma (FL) \[[@B4]\]. In a series of 34 cases of FLIS identified by Jegalian et al., six had prior or concurrent FL and five had FLIS composite with another lymphoma. Of patients with negative staging at diagnosis and available followup (twenty-one patients), only one developed FL \[[@B5]\]. We present a case recently diagnosed in a middle-aged African lady which to our knowledge is the first reported occurrence in the West African subregion of the world. 2. Case Report {#sec2} ============== A 48-year-old civil servant presented with axillary lymphadenopathy of insidious onset, discovered on routine mammography. The lymph node was excised and sent for histological analysis. She had neither clinically obvious enlarged lymph nodes elsewhere nor a previous history of lymphadenopathy. Histological examination was done with the aid of immunohistochemistry. The predominant abnormality on H&E examination was a reactive change characterised by follicular hyperplasia, sinus histiocytosis, and expansion of the interfollicular T-cell zones with increased numbers of inter-follicular dendritic cells associated with patchy aggregates of melanophages. These features were considered indicative of a dermatopathic lymphadenopathy by some of the consultant pathologists. However, some others disagreed and felt there was some subtle evidence of lymphoma. This resulted in the use of immunohistochemistry. Immunohistochemistry done in our laboratory showed CD20 positivity in B-cell areas of the lymph node and follicular germinal centres which were CD10 and Bcl-2 positive. Due to our limited immunohistochemistry experience and the few immunohistochemistry panels at our disposal, the blocks were sent to the Department of Cellular Pathology in Queen\'s Hospital, Rom Valley Way, Romford, Essex, UK. Their analysis revealed secondary follicles with germinal centres that were variably colonised by small CD10 and Bcl-6 positive cells that also overexpressed Bcl-2. The involved follicles had a low proliferation fraction as determined by Ki67 immunohistochemistry. S100 stained the increased numbers of interfollicular dendritic cells (see [Figure 1](#fig1){ref-type="fig"}). These features indicated an intrafollicular neoplasia/*in situ* follicular lymphoma. 3. Discussion {#sec3} ============= Follicular lymphoma (FL) is the second most common non Hodgkin lymphoma (NHL) in the Western world \[[@B6]\]. It has an average incidence of 2.6 per 100,000 and median age in the 6th decade \[[@B6]\] and is slightly more common amongst females \[[@B7]\]. In FLIS, the enlarged lymph nodes are usually incidental findings, and the patient has no generalised lymphadenopathy \[[@B4]\]. However, there may be a coexisting FL in the same lymph node or in other nodes as has been reported in a few cases \[[@B5], [@B8]\]. In the index patient, the lymph nodes were discovered on routine mammography. In the series of lymph nodes affected by *in situ* follicular lymphoma analysed by Jegalian et al. \[[@B5]\], they found a majority of females (56%) and a peak age of occurrence between the fifth and sixth decades. This report corresponds with the clinical profile of our index patient who is female and aged 48 years. Follicular lymphoma is a mature B-cell neoplasm thought to be derived from follicular centre B lymphocytes. The lymphoid cells express the immunophenotypic markers associated with germinal centre B-cells, including CD10 and Bcl-6 \[[@B9]\]. In FL, the *Bcl-2* gene on chromosome 18 is merged with the immunoglobulin heavy gene locus on chromosome 14 (t 14:18) (q32;q21) \[[@B2]\]. This results in constitutive activation of the *Bcl-2* gene which is antiapoptotic and leads to accumulation of follicular centre B-cells which may otherwise have died through apoptosis. Bcl-2 is not expressed in normal follicle centre cells. Inappropriate expression of the *Bcl-2* oncogene has long been believed to be the initial event in malignant transformation to FL \[[@B9]\]. Immunohistochemical staining for Bcl-2 protein also provides a very useful tool for distinguishing between reactive and neoplastic lymphoid follicles as normal germinal centres almost never express Bcl-2 protein \[[@B10], [@B11]\]. To diagnose FLIS, the following criteria are used \[[@B4], [@B5], [@B8], [@B12]\]: preserved nodal architecture with open sinuses and preserved paracortical regions. On H&E sections, the follicles appear to be reactive. Immunohistochemistry shows follicles positive for CD10, Bcl-6, and CD20. Bcl-2 positive germinal centre cells are confined to the germinal centres of the follicles, do not replace the entire follicle centre, and are not seen in the interfollicular regions or elsewhere in the lymph node. The involved follicles also have a lower proliferation rate with ki67 than the adjacent reactive follicles. Molecular studies, for example, fluorescence *in situ* hybridization analysis for t(14:18), are only necessary for cases in which immunohistochemistry findings are ambiguous  \[[@B12]\]. In the index case, the immunohistochemistry results were not ambiguous and diagnosis of FLIS was made without FISH. FLIS has been reported in association with other conditions such as nonlymphoid malignancies and Crohn\'s disease \[[@B12]\]. In these reports, the FLIS was an incidental finding. FLIS may transform to follicular lymphoma. In the case series reviewed by Jegalian et al. \[[@B5]\], one out of thirty-four patients reviewed eventually developed follicular lymphoma at the same site. Bonzheim et al. \[[@B2]\] have, however, proposed a clonal evolution from FLIS to manifest follicular lymphoma. This finding has not been supported by other studies with long-term follow-up of patients already diagnosed with FLIS \[[@B3], [@B5]\]. The index patient, a year after this diagnosis, is presently on followup and has not presented with symptoms or signs of a follicular lymphoma. *In situ* FL needs to be differentiated from cases of partial involvement of a lymph node by a follicular lymphoma (PFL). In PFL, the architecture of the lymph node is altered when compared to the preserved architecture for FLIS. Also, in FLIS, the germinal centres stain strongly for Bcl-2 and CD10, while in PFL these markers show variable intensity \[[@B5]\]. Other criteria adopted by a panel of experts during the workshop on "early lesions in lymphoid neoplasms" organised by the European Association of Hematopathology in Uppsala, Sweden (2010) include the following: the follicular size is normal in FLIS while being expanded in PFL, the follicles are widely scattered in FLIS while they are grouped together in PFL, the follicular cuff is intact in FLIS while it is attenuated in PFL, and FLIS is composed of almost pure centrocytes while PFL is composed of centrocytes with few centroblasts \[[@B13]\]. 4. Conclusion {#sec4} ============= This report highlights a hitherto unreported entity in the West African subregion of the world and highlights the need for immunohistochemistry in the diagnosis of lymph node pathology---a resource which is very limited in this environment. Learning Points {#sec5} =============== FLIS is a rare entity.This is the first case reported in this subregion.Immunohistochemistry is very relevant in lymph node pathology---a resource which is scarcely available in this environment. The authors declare that there is no conflict of interests. Orah Nnamdi Obumneme carried out the literature research and prepared the draft paper. Akinde Ralph interpreted the H&E sections. Igbokwe Uche interpreted the immunohistochemistry. Irurhe Nicholas did the mammography. Banjo Adekunbiola edited the final paper. {#fig1} [^1]: Academic Editors: A. Rajput and A. N. Walker

Introduction {#s1} ============ Chinese herbal medicines (CHMs) were the main treatment method used in ancient times by the Chinese to combat disease. As early as the Qin and Han Dynasty (around 221 BCE to 220 CE), *Sheng Nong\'s Herbal Classic* recorded 365 medicines. By the time of the Ming Dynasty (1368--1644), the number of CHMs listed in the book of *Compendium of Materia Medica* had increased to 1892. Most herbal medicines in such publications have been used constantly throughout medical history and are still applied in practice today. For example, according to *Sheng Nong\'s Herbal Classic, Coptis chinensis* Franch. was found to relieve abdominal pain and diarrhea and this herb is still widely used in China for the treatment of diarrhea or dysentery. Further, *Panax notoginseng* (Burk.) F. H. Chen, a traditional herb was initially used to stop bleeding, promote blood circulation and ease pain, was recorded in the *Compendium of Materia Medica*. It is now commonly used in cases of trauma and cardiovascular, and cerebrovascular diseases. A recent meta-analysis further demonstrated that several *P. notoginseng* preparations are beneficial for patients with unstable angina pectoris (Song et al., [@B121]). Despite the positive effects of CHMs, little is known about their effective constituents, bioactive ingredients, and mechanisms of action. Therefore, in addition to the clinical applications mentioned in classic texts, understanding the specific active ingredients and clarifying the mechanisms of action of these compounds would facilitate the improved application of CHMs. The discovery of the drug artemisinin best illustrates the importance of CHM to the world (Tu, [@B133]), inspiring the notion that, through study of their bioactive ingredients, CHMs can help people around the world to conquer life-threatening diseases. Non-coding RNA molecules (ncRNAs), which mainly comprise miRNA, lncRNA, and circRNA, do not encode proteins; however, as the most abundant class of RNA (at least 90%) (Sana et al., [@B113]), ncRNAs have important functions in gene regulation and are involved in pathological processes contributing to many diseases (Batista and Chang, [@B4]; Memczak et al., [@B97]; Zhang et al., [@B177]), particularly cancer, and cardiovascular and nervous system diseases. Moreover, circRNA and lncRNA act as competitive endogenous RNAs (ceRNA), which are natural miRNA sponges that influence miRNA-induced gene silencing via miRNA response elements (Tay et al., [@B127]). Thus, complex regulatory networks exist, comprising circRNA, lncRNA, miRNA, and target genes. Unraveling of this complexity has laid the foundation for a comprehensive understanding of the pathology and treatment of diseases influenced by gene regulatory networks, rather than only core disease-related genes (Boyle et al., [@B9]). Excitingly, recent studies (Feng et al., [@B30]; Tian F. et al., [@B128]; Zhou Y. et al., [@B196]) have revealed that some miRNA, lncRNA, circRNA, and ceRNA crosstalk can be regulated by bioactive ingredients from CHMs, which often have multiple targets ([Table 1](#T1){ref-type="table"}). By influencing regulatory mechanisms, including pro-apoptosis (Feng et al., [@B30]), anti-proliferation and anti-migration (Liu T. et al., [@B83]), anti-inflammation (Fan et al., [@B28]), anti-atherosclerosis (Han et al., [@B42]), anti-infection (Liu et al., [@B79]), anti-senescence (Zhang J. et al., [@B178]), and suppression of structural remodeling (Liu L. et al., [@B81]), these ingredients exert protective functions in cancer, cardiovascular disease, nervous system disease, inflammatory bowel disease, asthma, infectious diseases, and senescence-related diseases. ###### Detailed information on bioactive ingredients targeting ncRNAs. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **No**. **Name of ingredient** **Original plant** **ncRNA target** **Gene/protein/pathway** **Disease** --------- ------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------- 1 Berberine *Coptis chinensis* Franch., *Berberis soulieana* Schneid, *Berberis poiretii* Schneid., *Berberis vernae* Schneid, *Berberis wilsoniae* Hemsl., and *Platycladus orientalis* (Linn.) Franco. miR-99a\~125b (Feng et al., [@B30])\ RAC1, NFκB1, MYC, JUN, and CCND1 (Feng et al., [@B30])\ Multiple myeloma (Luo et al., [@B89]; Feng et al., [@B30]; Yin et al., [@B166]) miR-21 (Luo et al., [@B89])\ IL6/STAT3, PDCD4 (Luo et al., [@B89])\ miR-19a/92a (Yin et al., [@B166]) P53 signaling pathway (Luo et al., [@B89])\ P53, ERB, and MAPK signaling pathways (Feng et al., [@B30]) miR-23a (Wang N. et al., [@B141]) NEK6, P53, P21, GADD45α (Wang N. et al., [@B141]) Hepatocellular carcinoma (Wang N. et al., [@B141]) miR-152\ DNMT1, DNMT3A, DNMT3B, CDK4\ Colorectal cancer (Su et al., [@B123]; Huang et al., [@B48]) miR-429\ (Su et al., [@B123]; Huang et al., [@B48])\ miR-29a (Huang et al., [@B48])\ Pin1-β-catenin-cyclin D1 signaling pathway (Su et al., [@B123]) miR-296-5p (Su et al., [@B123]) miR-34a\ CDK4, CyclinD1, CyclinE, CDK2 (Yang L. H. et al., [@B159]) Melanoma (Yang L. H. et al., [@B159]) miR-154\ miR-26a\ miR-124 (Yang L. H. et al., [@B159]) miR-93 (Chen et al., [@B17]) miR-93/PTEN/AKT signaling pathway (Chen et al., [@B17]) Ovarian cancer (Chen et al., [@B17]) miR-203 (You et al., [@B167]) BCL-w (You et al., [@B167]) Gastric cancer (You et al., [@B167]) miR-27a miR-27b (Wu et al., [@B154]) PPAR-γ (Wu et al., [@B154]) Obesity (Wu et al., [@B154]) lncRNA MRAK052686 (Yuan et al., [@B172]) NRF2 (Yuan et al., [@B172]) Steatotic liver (Yuan et al., [@B172]; Li C. H. et al., [@B69]) miR-373 (Li C. H. et al., [@B69]) EGR1, AKT1\ AKT-mTOR-S6K signaling pathway (Li C. H. et al., [@B69]) miR-29a-3p (Mao et al., [@B94]) IRS1 (Mao et al., [@B94]) Insulin resistance (Mao et al., [@B94]) 2 Artesunate *Artemisia annua* L. lncRNA UCA1\ lncRNA UCA1/miR-184/BCL-2 axis (Zhou S. et al., [@B194]) Prostate cancer (Zhou S. et al., [@B194]) miR-184 (Zhou S. et al., [@B194]) 3 Triptolide/ Triptonide *Tripterygium wilfordii* Hook. f miR-21 (Li et al., [@B74]) Caspase-3 and 9, PTEN (Li et al., [@B74]) Non-small cell lung cancer (Li et al., [@B74]) 227 miRNAs (Reno et al., [@B109]) Focal adhesion kinase (Reno et al., [@B109]) Lung cancer (Reno et al., [@B109]) miR-17-92\ c-MYC, BIM, PTEN, and P21 (Li S. G. et al., [@B71]) Hepatocellular carcinoma (Li S. G. et al., [@B71]) miR-106b-25 (Li S. G. et al., [@B71]) lncRNA THOR (Wang et al., [@B143]) IGF2BP1, Myc, IGF2, and Gli1 (Wang et al., [@B143]) Nasopharyngeal carcinoma (Wang et al., [@B143]) 4 Ailanthone *Ailanthus altissima* (Mill.) Swingle miR-21 (Yang P. et al., [@B160]) caspase 3, caspase 9, Beclin-1, LC3-II, p62, and cyclin D1\ Vestibular schwannomas (Yang P. et al., [@B160]) Ras/Raf/MEK/ERK and mTOR pathways (Yang P. et al., [@B160]) 5 Cordycepin/Soya-cerebroside *Cordyceps militaris* miR-21 (Yang et al., [@B157]) PTEN, Akt (Yang et al., [@B157]) Renal cell carcinoma (Yang et al., [@B157]) miR-432 (Liu S. C. et al., [@B82]) MCP-1\ \ AMPK and AKT signaling pathways (Liu S. C. et al., [@B82]) Osteoarthritis\ (Liu S. C. et al., [@B82]) 6 Tubeimoside I *Bolbostemma paniculatum* (Maxim) Franquet miR-126-5p (Shi et al., [@B118]) VEGF-A/VEGFR-2/ERK signaling pathway (Shi et al., [@B118]) Non-small cell lung cancer (Shi et al., [@B118]) 7 Oridonin *Rabdosia rubescens* (Hemsl.) Hara 105 miRNAs (Gui et al., [@B37]) / Laryngeal cancer (Gui et al., [@B37]) 8 Curcumin *Curcuma longa* L. miR-208 (Guo H. et al., [@B38]) CDKN1A (Guo H. et al., [@B38]) Prostate cancer (Guo H. et al., [@B38]; Cao et al., [@B11]; Liu W. L. et al., [@B84]; Zhang et al., [@B176]) miR-145\ CCND1, CDK4, OCT4, CD44, and CD133 (Liu W. L. et al., [@B84]) lncRNA-ROR (Liu W. L. et al., [@B84]) miR-770-5p\ DLK1-DIO3 (Zhang et al., [@B176]) miR-1247 (Zhang et al., [@B176]) miR-143 (Cao et al., [@B11]) PGK1, FOXD3 (Cao et al., [@B11]) miR-98 (Liu W. L. et al., [@B84]) LIN28A, MMP 2, MMP9 (Liu W. L. et al., [@B84]) Lung cancer (Tang et al., [@B126]; Liu W. L. et al., [@B84]) miR-186\* (Tang et al., [@B126]) / miR-203 (Saini et al., [@B112]) AKT2, SRC (Saini et al., [@B112]) Bladder cancer (Saini et al., [@B112]) miR-33b (Sun et al., [@B124]) XIAP (Sun et al., [@B124]) Gastric cancer (Sun et al., [@B124]) miR-192-5p (Jin et al., [@B57]) PI3K/AKT signaling pathway (Jin et al., [@B57]) Non-small cell lung cancer (Jin et al., [@B57]) miR-7 (Ma et al., [@B92]) SET8 (Ma et al., [@B92]) Pancreatic cancer (Ma et al., [@B92]) miR-30c (Lu et al., [@B87]) MTA1 (Lu et al., [@B87]) Paclitaxel-resistant non-small-cell lung cancer (Lu et al., [@B87]) miR-29b-1-5p (Zhou S. et al., [@B194]) PPARG, RRM2, SRSF1, EPAS1, MAPK, mTOR, PI3K-AKT, AMPK, TNF\ Adriamycin-resistant breast cancer (Zhou S. et al., [@B194]) RAS signaling pathways (Zhou S. et al., [@B194]) lncRNA AK294004 (Wang Q. et al., [@B142]) Cyclin D1(Wang Q. et al., [@B142]) Radioresistant nasopharyngeal carcinoma (Wang N. et al., [@B141]) miR-146a (Wu et al., [@B152])\ NF-κB signaling (Wu et al., [@B152])P38 (Li et al., [@B73]) Glioblastoma (Wu et al., [@B152]; Li et al., [@B73]) miR-378 (Li et al., [@B73]) miR-34a (Guo et al., [@B39]) BCL-2, BMI-1 (Guo et al., [@B39]) Breast cancer (Guo et al., [@B39]; Zhou et al., [@B195]) miR-29b-1-5p\ DDIT4, EPAS1, VEGFA, RPS14, and DCDC2 (Zhou et al., [@B195]) miR-29b-3p\ miR-6068\ miR-6790-5p\ miR-4417 (Zhou et al., [@B195]) miR-122\ FGF2, MMP2, VEGF, HGF, TF, FVII (Zhang S. et al., [@B184]) Hepatocellular Carcinoma (Guo Y. et al., [@B40]; Liu W. L. et al., [@B84]) miR-221 (Zhang S. et al., [@B184]) lncRNA AK125910 (Guo Y. et al., [@B40]) miR-155 (Ma F. et al., [@B91]) TNF-α, IL-6\ LPS-induced inflammatory response (Ma F. et al., [@B91]) PI3K/AKT pathway (Ma F. et al., [@B91]) miR-17-5p (Tian L. et al., [@B129]) WNT signaling pathway effector TCF7l2 (Tian L. et al., [@B129]) Adipogenic differentiation (Tian L. et al., [@B129]) 9 Shikonin *Lithospermum erythrorhizon* Sieb. et Zucc. miR-106b (Huang and Hu, [@B47]) miR-106b/PTEN/AKT/mTOR signaling pathway (Huang and Hu, [@B47]) Endometrioid endometrial cancer (Huang and Hu, [@B47]) miR-128 (Wei et al., [@B148]) BAX (Wei et al., [@B148]) Breast cancer (Wei et al., [@B148]) miR-143 (Liu et al., [@B80]) BAG3 (Liu et al., [@B80]) Glioblastoma (Liu et al., [@B80]) 10 Paeoniflorin *Paeonia lactiflora* Pall. miR-16 (Li W. et al., [@B72]) MMP-9 (Li W. et al., [@B72]) Glioma (Li W. et al., [@B72]) 11 Honokiol *Magnolia grandiflora* miR-34a (Avtanski et al., [@B3]) STAT3\ Breast tumor (Avtanski et al., [@B3]) WNT1-MTA1-β-catenin signaling (Avtanski et al., [@B3]) 12 Schisandrin B *Schisandra sphenanthera* Rehd. et Wils. miR-150\ miR-150/ lncRNA BCYRN1/ cell proliferation axis (Zhang X. Y. et al., [@B185]) Asthma (Lu et al., [@B85]) lncRNA BCYRN1 (Zhang X. Y. et al., [@B185]) 13 Resveratrol Grapes, blueberries, *Morus alba* L.,\ lncRNA MALAT1 (Ji et al., [@B52]) c-MYC, MMP-7\ Colorectal cancer (Ji et al., [@B52]; Karimi Dermani et al., [@B60]) *Polygonum cuspidatum* Sieb. et Zucc. and *Rubus idaeus* L. WNT/β-catenin signaling (Ji et al., [@B52]) miR-200c (Karimi Dermani et al., [@B60]) Vimentin, ZEB-1, E-cadherin (Karimi Dermani et al., [@B60]) miR-221 (Wu and Cui, [@B151]) NF-κB, TFG (Wu and Cui, [@B151]) Melanoma (Wu and Cui, [@B151]) miR-21\ P53, PTEN, EGFR, STAT3, COX-2, NF-κB\ Glioma (Wang G. et al., [@B136]) miR-30a-5p\ PI3K/AKT/mTOR pathway (Wang G. et al., [@B136]) miR-19 (Wang G. et al., [@B136]) miR-155 (Ma C. et al., [@B90]) TNF-α, IL-6, MAPKs, STAT1/STAT3, and SOCS1 (Ma C. et al., [@B90]) LPS-induced inflammatory response (Ma C. et al., [@B90]) miR-663\ JUNB, JUND, activator protein-1 (Tili et al., [@B130]) Malignancies (Tili et al., [@B130]) miR-155 (Tili et al., [@B130]) miR-96 (Bian et al., [@B6]) BAX (Bian et al., [@B6]) Hypoxia/ischemia-induced brain injury (Bian et al., [@B6]) miR-13\ CREB, BDNF (Zhao et al., [@B190]) Alzheimer\'s disease (Zhao et al., [@B190]) miR-124 (Zhao et al., [@B190]) 14 Soybean Isoflavones *Glycine max* (Linn.) Merr. miR-29a\ TRIM68, PGK-1 (Li et al., [@B75]) Prostate cancer (Li et al., [@B75]) miR-1256 (Li et al., [@B75]) 15 Matrine *Sophora flavescens Ait* miR-19b-3p (Wei et al., [@B149]) PTEN (Wei et al., [@B149]) Melanoma (Wei et al., [@B149]) 16 Corylin *Psoralea corylifolia* Linn lncRNA GAS5 (Chen et al., [@B14]) / Hepatocellular carcinoma (Chen et al., [@B14]) 17 Tanshinone IIA/Magnesium lithospermate B *Salvia miltiorrhiza* Bge. miR-155\ TLR4, MyD88, GM-CSF, sICAM-1, CXCL-1, MIP-1α, TNF-α, IL-1β, COX-2\ LPS-induced inflammation (Fan et al., [@B28]) miR-147\ TLR4-NF-κB pathway (Fan et al., [@B28]) miR-184\ miR-29b\ miR-34c (Fan et al., [@B28]) miR-146b\ CRP, ox-LDL, IL-1β, IL-6, IL-12, TNF-α, CCL-2, CD40, and MMP-2 (Xuan et al., [@B156]) Atherosclerosis (Xuan et al., [@B156]) miR-155 (Xuan et al., [@B156]) miR-133 (Zhang et al., [@B180]) MAPK ERK1/2 (Zhang et al., [@B180]) Hypoxia (Zhang et al., [@B180]) miR-1 (Shan et al., [@B116]; Zhang et al., [@B187]) SRF, Kir2.1 (Shan et al., [@B116]) Arrhythmias post-AMI (Shan et al., [@B116]) Cx43, SRF, MEF2,\ Myocardial infarction (Zhang et al., [@B187]) P38 MAPK signal pathway (Zhang et al., [@B187]) miR-107 (Yang et al., [@B164]) GLT-1, glutamate (Yang et al., [@B164]) Cerebral I/R injury (Yang et al., [@B164]) 18 Baicalin *Scutellaria baicalensis* Georgi miR-191a (Wang L. et al., [@B140]) ZO-1 (Wang L. et al., [@B140]) Inflammatory bowel disease (Wang L. et al., [@B140]) miR-294 (Wang J. et al., [@B146]) c-jun and c-fos (Wang J. et al., [@B146]) Inhibition of proliferation (Wang J. et al., [@B138]) 19 Cinnamaldehyde *Cinnamomum cassia* Presl. miR-21\ TNF-α, IL-1β, IL-6, AKT, mTOR, COX2 (Qu et al., [@B107]) Ulcerative colitis (Qu et al., [@B107]) miR-155 (Qu et al., [@B107]) has-circ-0043256, miR-1252 (Tian F. et al., [@B128]) has-circ-0043256/miR-1252/ITCH axis\ Non-small cell lung cancer (Tian F. et al., [@B128]) WNT/β-catenin pathway (Tian F. et al., [@B128]) 20 Geniposide *Gardenia jasminoides Ellis* miR-145 (Su et al., [@B122]) IL-6, TNF-α, MCP-1,\ Inflammation in cardiomyocyte (Su et al., [@B122]) MEK/ERK pathway (Su et al., [@B122]) 21 Carvacrol/ Thymol *Origanum vulgare* L. or wild bergamot miR-155\ TLR2, TLR4, SOCS1, SHIP1 (Khosravi and Erle, [@B61]) Asthma (Khosravi and Erle, [@B61]) miR-146a\ miR-21 (Khosravi and Erle, [@B61]) 22 3-acetyl-11-keto-β-boswellic acid *Boswellia serrata* miR-155 (Sayed et al., [@B114]) P-IκB-α, carbonyl protein, SOCS-1 (Sayed et al., [@B114]) Neuroinflammation (Sayed et al., [@B114]) 23 Sinapic acid *Sinapis alba* L. LncRNA MALAT1 (Han et al., [@B42]) ET-1, IL-1β, ASC, NRLP3, Caspase-1 (Han et al., [@B42]) Diabetic atherosclerosis (Han et al., [@B42]) 24 Polydatin *Polygonum cuspidatum* Sieb. et Zucc. miR-214 (Zhou et al., [@B192]) Blood glucose, ALT, AST, TC, TG, LDL-C, HDL-C, MDA, SOD (Zhou et al., [@B192]) Atherosclerosis with liver dysfunction (Zhou et al., [@B192]) 25 Ampelopsin *Ampelopsis grossedentata* miR-21 (Yang D. et al., [@B158]) eNOS, DDAH1, NO, and ADMA (Yang D. et al., [@B158]) Endothelial dysfunction (Yang D. et al., [@B158]) miR-34a (Kou et al., [@B64]) SIRT1-mTOR signal pathways (Kou et al., [@B64]) Neurodegenerative diseases (Kou et al., [@B64]) 26 Icariine *Epimedium brevicornu* Maxim. miR-34c (Liu et al., [@B79]) RUNX2, JNKs, and p38,\ Bacteria-induced bone loss diseases (Liu et al., [@B79]) NF-kB pathways (Liu et al., [@B79]) miR-21 (Li J. et al., [@B70]) PTEN, RECK, Caspase-3, and BCL-2 (Li J. et al., [@B70]) Ovarian cancer (Li J. et al., [@B70]) 27 Ginsenosides *Panax ginseng* C. A. Mey. miR-15b (Chan et al., [@B13]) IP-10 (Chan et al., [@B13]) H9N2/G1 infection (Chan et al., [@B13]) 28 Salidroside *Rhodiola rosea* L let-7c\ p53, transcription factor CREB\ Senescence (Zhang J. et al., [@B178]) let-7e\ \ miR-3620\ \ miR-411\ \ miR-24-2-5p\ \ miR-485-3p (Zhang J. et al., [@B178]) AKT/mTOR signaling (Zhang J. et al., [@B178]) 29 Phlorizin *Acanthopanax senticosus* (Rupr. et Maxim.) Harms miR-135b (Choi et al., [@B24]) p63, PCNA, integrin α6, integrin β1, and type IV collagen (Choi et al., [@B24]) Skin aging (Choi et al., [@B24]) 30 Osthole *Cnidium monnieri* (L.) Cuss. miR-107 (Xiao et al., [@B155]) Aβ, BACE1, and LDH (Xiao et al., [@B155]) Alzheimer\'s disease (Xiao et al., [@B155]) 31 Panax Notoginseng Saponins *Panax notoginseng* (Burk.) F. H. Chen. miR-29c (Liu L. et al., [@B81]) Collagen (Col) 1a1, Col1a2, Col3a1, Col5a1, FBN1, TGFβ1 (Liu L. et al., [@B81]) Myocardial injury and fibrosis (Liu L. et al., [@B81]) miR-146b-5p (Wang J. et al., [@B146]) / Oxidative damage (Wang J. et al., [@B146]) miR-34a (Lai et al., [@B66]) miR-34a/SIRT1/p53 pathway (Lai et al., [@B66]) Senescence (Lai et al., [@B66]) miR-18a (Yang Q. et al., [@B162]) CD34, VWF (Yang Q. et al., [@B162]) Tumor complicated with myocardial ischemia (Yang Q. et al., [@B162]) miR-222 (Yang Q. et al., [@B161]) p27 and PTEN\ Lewis lung carcinoma (Yang Q. et al., [@B161]) Met/miR-222 axis (Yang Q. et al., [@B161]) 32 Tetrandrine *Stephania tetrandra* S. Moore. miR-27b\ VEGFC, BCL2L12, COL4A3, FGFR2 (Ning et al., [@B103]) Hypertrophic scar (Ning et al., [@B103]) miR-125b (Ning et al., [@B103]) 33 Leonurine *Leonurus artemisia* (Laur.) S. Y. Hu F. miR-1 (Lu et al., [@B86]) ANP, ET-1, p38 MAPK, p-p38 MAPK, myocyte enhancer factor 2, β-myosin heavy chain, and α-myosin heavy chain protein (Lu et al., [@B86]) Cardiomyocyte hypertrophy (Lu et al., [@B86]) 34 Calycosin/\ *Astragalus membranaceus* (Fisch.) Bunge. miR-375 (Wang Y. et al., [@B145]) ER-α and Bcl-2 and RASD1 (Wang Y. et al., [@B145]) Cerebral I/R injury (Wang Y. et al., [@B145]) Astragaloside IV/Total flavonoids lncRNA EWSAT1 (Kong et al., [@B63]) / Nasopharyngeal carcinoma (Kong et al., [@B63]) miR-34a (Zhang C. et al., [@B174]) LDHA, MCT1, MCT4, HIF-1α, CD147, TIGAR and p53 (Zhang C. et al., [@B174]) Gastric carcinoma (Zhang C. et al., [@B174]) miR-378\ / Viral myocarditis (Wan et al., [@B134]) miR-378\* (Wan et al., [@B134]) 35 Paeonol *Paeonia suffruticosa* Andr. and *Cynanchum paniculatum* (Bunge) Kitagawa miR-1 (Zhang and Xiong, [@B179]) / Ischemic arrhythmia (Zhang and Xiong, [@B179]) 36 Salvianolic acid A *Salvia miltiorrhiza* Bge. miR-101 (Yu D. S. et al., [@B168]) tight junction proteins, HO-1, p-caveolin-1, ZO-1, occluding, Nrf2, and Cul3 (Yu D. S. et al., [@B168]) Spinal cord injury (Yu D. S. et al., [@B168]) miR-3686\ MDR1 (Chen et al., [@B15]) Lung cancer (Chen et al., [@B15]) miR-4708-3p\ miR-3667-5p\ miR-4738-3p (Chen et al., [@B15]) 37 Andrographolide *Andrographis paniculata* (Burm. f.) Nees miR-222-3p\ signaling pathways of miRNAs in cancer, MPAKs, and focal adhesion (Lu et al., [@B85]) Hepatoma tumor (Lu et al., [@B85]) miR-106b-5p\ miR-30b-5p\ miR-23a-5p (Lu et al., [@B85]) 38 Puerarin *Radix Puerariae Lobatae* miR-22 (Wang L. et al., [@B139]) caveolin-3, amphiphysin-2, and junctophinlin-2 (Wang L. et al., [@B139]) Cardiovascular diseases (Wang L. et al., [@B139]) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Methodology {#s2} =========== The bioactive ingredients of CHMs and their interactions with ncRNA targets are the subject of intensive and rapidly expanding research. In this study, we undertook a comprehensive review of this research. The PubMed database was searched using the terms: "(ncRNA) AND herbal medicine", "(((miRNA) OR lncRNA) OR circRNA) AND herbal medicine," "(((miRNA) OR lncRNA) OR circRNA) AND active ingredient," "(((miRNA) OR lncRNA) OR circRNA) AND Chinese herb," "(((miRNA) OR lncRNA) OR circRNA) AND natural agent," "(((miRNA) OR lncRNA) OR circRNA) AND natural compound," or "(((miRNA) OR lncRNA) OR circRNA) AND traditional Chinese medicine." In addition, the China National Knowledge Infrastructure (CNKI) was also searched with terms as follows: "FT = 'Chinese herbal medicine' AND SU = 'ncRNA' NOT (TI = 'Review' OR TI = ' Progress\' OR TI = 'Overview' OR TI = 'Current situation')," "FT = 'Chinese herbal medicine' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = ' Progress\' OR TI = 'Overview' OR TI = 'Current situation')," "FT = 'Active ingredient' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = 'Progress' OR TI = 'Overview' OR TI = 'Current situation')," "FT = 'Natural compound' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = 'Progress' OR TI = 'Overview' OR TI = 'Current situation')," "FT = \'Natural ingredient' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = 'Progress' OR TI = 'Overview' OR TI = 'Current situation')," "FT = 'Traditional Chinese medicine extract' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = 'Progress' OR TI = 'Overview' OR TI = 'Current situation')," or "FT = 'Traditional Chinese medicine' AND (SU = 'lncRNA' OR SU = 'miRNA' OR SU = 'circRNA') NOT (TI = 'Review' OR TI = 'Progress' OR TI = 'Overview' OR TI = 'Current situation')." "FT" means full text; "SU" means subject; "TI" means title. Articles included in "Guide to Core Journals of China," "Chinese Science Citation Database" and "Chemical Abstracts" simultaneously, were selected to ensure high quality of literature. According to the above searching method, English and Chinese original articles related to bioactive ingredients of CHMs and any ncRNA (miRNA, lncRNA, or circRNA) were selected manually. Pro-apoptosis Effects of CHMs {#s3} ============================= Apoptosis is programmed cell death, which is a normal physiological process of cells. Imbalance of apoptosis is closely associated with various diseases, particularly cancer. Proteins that inhibit apoptosis are over-expressed in various cancers, and are considered to be related to tumorigenesis and chemotherapy resistance (Mohamed et al., [@B100]); therefore, the induction of apoptosis is a promising method for cancer management (Fulda and Vucic, [@B31]). Recently, several bioactive ingredients of CHMs have been reported to promote apoptosis by targeting miRNA, lncRNA, or ceRNA crosstalk, indicating their potential as complementary therapies for cancer. Berberine --------- Berberine \[BBR, 9,10-Dimethoxy-2,3-(methylenedioxy)-7,8,13,13a-Tetrahydroberbinium; [Chem. 1](#T2){ref-type="table"}\] is an isoquinoline alkaloid extracted from the roots of several species including: *Coptis chinensis* Franch., *Berberis soulieana* Schneid., *Berberis poiretii* Schneid., *Berberis vernae* Schneid., *Berberis wilsoniae* Hemsl., and *Platycladus orientalis* (Linn.) Franco. These herbs are considered to have antipyretic and detoxification effects, based on the theory of traditional Chinese medicine (TCM), and were mainly used to treat diseases of the digestive and urinary systems, such as diarrhea, ulcer, jaundice, and urinary infection, as well as dermatological diseases, including eczema. Newly reported researches (Luo et al., [@B89]; Chai et al., [@B12]) have found the anti-cancer activity of two extracts from *Coptis chinensis* Franch. one of which is BBR. It\'s indicated that BBR affects apoptotic pathways in various cancers through regulation of multiple miRNAs. For example, BBR exerts significantly protective effects in multiple myeloma (MM) through targeting several miRNAs. BBR down-regulates the expression of miR-99a\~125b, miR-17\~92, miR-106\~25 (Feng et al., [@B30]), and miR-21 (Luo et al., [@B89]), thereby influencing the P53, ERB, and MAPK signaling pathways, leading to acceleration of apoptosis and growth inhibition. Moreover, BBR suppresses MM cell viability through down-regulating miR-19a/92a expression (Yin et al., [@B166]). Further, BBR can up regulate miR-23a in hepatocellular carcinoma (HCC) (Wang N. et al., [@B141]), as well as miR-152, miR-429, and miR-29a in colorectal cancer (Huang et al., [@B48]) in a P53-dependent manner. By inducing NEK6 inhibition and transcriptional activation of the P53-associated tumor suppressor genes, P21 and GADD45α, BBR induces cell death, G2/M cell cycle arrest, and tumor growth suppression in HCC cells; in contrast, miR-23a inhibition can attenuate these BBR-mediated functions (Wang N. et al., [@B141]). Besides, BBR also regulates cell cycle and inhibits cell proliferation in melanoma A375 cells through promoting miR-34a, miR-154, miR-26a, and miR-124 expression, as well as suppressing target genes CDK4, CyclinD1, CyclinE, and CDK2 (Yang L. H. et al., [@B159]). Additionally, BBR can enhance cellular sensitivity to chemotherapeutic drugs, helping to address the problem of drug resistance. In ovarian and gastric cancers, BBR can enhance cisplatin sensitivity by regulating the expression of miR-93 (Chen et al., [@B17]) and miR-203 (You et al., [@B167]), respectively, thus inducing apoptosis. Furthermore, there is evidence that BBR can function together with other compound. It has a synergistic antiproliferative effect on colorectal cancer when combined with NVP-AUY922. The potential mechanism underlying this phenomenon was reported to be suppression of CDK4 and induction of miR-296-5p-mediated inhibition of Pin1-β-catenin-cyclin D1 signaling, resulting in cell growth arrest (Su et al., [@B123]). Interestingly, BBR also has protective effects against obesity, steatotic liver and insulin resistance. It inhibits cell viability, cell differentiation, and triglyceride content in a dose- and time-dependent manner, through marked induction of miR-27a and miR-27b; while miR-27a and miR-27b inhibitors can counteract this repressive function of BBR (Wu et al., [@B154]). Steatotic liver results from disordered lipid metabolism, where lncRNA MRAK052686, NRF2 (Yuan et al., [@B172]), and miR-373 (Li C. H. et al., [@B69]) are down-regulated. BBR can reverse the abnormal expression of these genes in steatotic liver, thereby inhibiting the AKT-mTOR-S6K signaling pathway and preventing the development of hepatic steatosis (Li C. H. et al., [@B69]). In addition, through downregulating miR-29a-3p in insulin resistant HepG2 cells, BBR can increase the mRNA and protein expression of IRS1, leading to regulation of the insulin receptor signaling pathway protein (Mao et al., [@B94]). Based on the above results, it is clear that BBR possesses different pharmacological effects via targeting miRNAs. Particularly, it shows obvious advantages for treatment of MM and digestive cancers, mainly through its activities in promotion of apoptosis and inhibition of cell growth. More importantly, its direct anticancer properties are strongly associated with P53 signaling. Therefore, BBR presents as promise for potential future use in cancer treatment. Artesunate ---------- Artesunate (ART, dihydroartemisinin-12-alpha-succinate; [Chem. 2](#T2){ref-type="table"}) is a sesquiterpene lactone extracted from the leafy portions of the Chinese herb, *Artemisia annua* L. It\'s the semisynthetic derivative of artemisinin which is widely known to be a natural antimalarial medicine (Tu, [@B133]). Recently, the function of ART in cancer therapy, by targeting lncRNA and promoting cell apoptosis, was newly identified. The lncRNA, UCA1, is up-regulated in prostate cancer tissues and positively correlated with poor prognosis (Zhou Y. et al., [@B196]). ART significantly decreased the expression of lncRNA UCA1, thereby regulating the downstream miR-184/BCL-2 axis, inducing apoptosis, and inhibiting metastatic ability. Furthermore, these protective effects could be reversed by overexpression of lncRNA UCA1, indicating that it is a target of ART (Zhou Y. et al., [@B196]). Hence, ART exhibits anticancer properties through regulating the ceRNA crosstalk of the lncRNA UCA1/miR-184/BCL-2 axis in prostate cancer. Nevertheless, additional evidence to support these findings is lacking and the stability of this regulatory network requires validation. Triptolide/Triptonide --------------------- Triptolide (TP, (3bs,4as,5as,6r,6ar,7as,7bs,8as,8bs)-6-hydroxy-6a-isopropyl-8b-methyl-3b,4,4a,6,6a,7a,7b,8b,9,10-decahydrotrisoxireno\[6,7:8a,9:4b,5\]phenanthro\[1,2-c\]furan-1(3h)-one; [Chem. 3](#T2){ref-type="table"}) and Triptonide (TN,(3bS,4aS,5aS,6aS,7aS,7bS,8aS,8bS)-6a-isopropyl-8b-methyl-3b,4,4a,7a,7b,8b,9,10-octahydrotrisoxireno\[6,7:8a,9:4b,5\]phenanthro\[1,2-c\]furan-1,6(3H,6aH)-dione; [Chem. 4](#T2){ref-type="table"}) are both diterpene lactone components originated from *Tripterygium wilfordii* Hook. f. (TwHf) that has traditionally been used for treatment of rheumatoid arthritis (RA). A recent study has revealed that TwHf exerts its anti-rheumatic effects through regulation of miR-146a, which is over-expressed in patients with RA and negatively correlated with prognosis. TwHf treatment could significantly decrease miR-146a expression. Moreover, miR-146a could be used as a predictor of patient clinical response to TwHf (Chen Z. Z. et al., [@B21]). Researches about TP and TN broaden the traditional application and generate new pharmacological effect for cancer treatment. TP has been found to exert anticancer activities in lung cancer. It can induce apoptosis and suppress proliferation through inhibiting miR-21 and increasing expression of phosphatase and tensin homolog (PTEN) protein in non-small cell lung cancer. Moreover, miR-21 upregulation could reverse the effect of TP on cell viability and PTEN (Li et al., [@B74]). Furthermore, TP treatment is also reported to regulate 227 miRNAs and markedly decrease the migration, invasion, and metastasis of lung cancer cells (Reno et al., [@B109]). In addition, TP can promote apoptosis and suppress cell proliferation in hepatocellular carcinoma, potentially via inhibition of miR-17-92 and miR-106b-25, in a c-MYC-dependent manner, leading to an increase of BIM, PTEN, and P21 levels (Li S. G. et al., [@B71]). However, TN shows significant therapeutic advantages in human nasopharyngeal carcinoma (NPC). It promotes NPC apoptosis and cell cycle arrest, as well as inhibition of cell migration and invasion, without toxicity to nasopharyngeal epithelial cells. This anti-cancer activity is attributed to suppression of lncRNA THOR, followed by downregulation of IGF2BP1 mRNA targets involving Myc, IGF2, and Gli1. Furthermore, lncRNA THOR knockout enhances the protection of TN on NPC; while lncRNA THOR overexpression reverses TN-induced treatment in cells. *In vivo*, TN administration also obviously impedes subcutaneous NPC xenograft growth in mice. Similarly, lncRNA THOR knockout inhibits xenograft growth (Wang et al., [@B143]). Therefore, TP and TN are attractive candidate chemotherapeutic agents against the above cancers. With regard to TP, PTEN is an important target; while TN possesses anti-cancer activity *in vitro* and *in vivo* through regulating lncRNA THOR/IGF2BP1 signaling. Nevertheless, as the extracts from TwHf with general toxicity (Chen et al., [@B16]; Luo et al., [@B88]), the effectiveness and safety of TP and TN require additional confirmation. Ailanthone ---------- Ailanthone (AIL, Picrasa-3,13(21)-diene-2,16-dione, 11,20-epoxy-1,11,12-trihydroxy-, (1-beta,11-beta,12-alpha)-; [Chem. 5](#T2){ref-type="table"}) is a water-soluble quassinoid extracted from the root bark of *Ailanthus altissima* (Mill.) Swingle. Traditionally, the root bark was used for improvement of itching, bleeding and diarrhea in TCM theory. However, AIL has been newly found to possess anti-tumor activity in different tumors (Chen Y. et al., [@B20]; Peng et al., [@B105]; Yang P. et al., [@B160]). Among those, inhibitory effect of human vestibular schwannomas (VSs) induced by AIL is correlated with miRNA. A research has demonstrated that AIL cleaves caspase 3 and caspase 9, promotes Beclin-1, LC3-II accumulation, and decreases p62, cyclin D1 expression, thus increasing apoptotic cell rate. The upstream mechanism may be suppression of miR-21, and the Ras/Raf/MEK/ERK and mTOR pathways, leading to apoptosis and autophagy in AIL-treated cells. In addition, miR-21 overexpression can attenuate the regulation of AIL on Ras/Raf/MEK/ERK and mTOR pathways, as well as apoptosis and autophagy, indicating miR-21 can be the treatment target of AIL in VSs (Yang P. et al., [@B160]). Cordycepin ---------- Cordycepin (COR, 9-(beta-D-3′-Deoxyribofuranosyl)adenine; [Chem. 6](#T2){ref-type="table"}) is the main bioactive ingredient of *Cordyceps militaris*, a precious CHM. The medicinal herb has immunity-strengthening effect, and has been already used as a health care product in clinical practice. Modern researches broaden the application of COR in various cancers (Wang et al., [@B135]; Liang et al., [@B77]; Yu X. et al., [@B170]). Specifically, treatment of COR for renal cell carcinoma (RCC) is attributed to regulation of miR-21 and PTEN phosphatase. It\'s indicated that COR down-regulates miR-21 expression and Akt phosphorylation, yet promotes PTEN phosphatase in RCC Caki-1 cells, resulting in induction of apoptotic cell death and suppression of cell migration. Furthermore, miR-21 mimic or PTEN siRNA can markedly abolish the above effects induced by COR (Yang et al., [@B157]). Therefore, it\'s confirmed that COR possesses pro-apoptosis and anti-migration function through regulating miR-21/PTEN axis. In addition, soya-cerebroside ([Chem 7](#T2){ref-type="table"}), another extracts from *Cordyceps militaris* is demonstrated to be anti-inflammatory for osteoarthritis (OA). It suppresses AMPK and AKT signaling pathways, and then promotes miR-432 expression in OA synovial fibroblasts, leading to inhibition of monocyte chemoattractant protein-1 (MCP-1), monocyte migration and infiltration, as well as cartilage degradation (Liu S. C. et al., [@B82]). As a result, soya-cerebroside exerts protective effect for OA partially via regulating miR-432, MCP-1, AMPK and AKT pathways; while in this study a clear functional relationships among those factors are not reported. Tubeimoside I ------------- TubeimosideI (TBMSI, nosyl\]-β-D-glucopyranosyl\]oxy\]-2,23-dihydroxy-,28-(O-β-D-xylopyranosyl-(1 → 3)-O-6-deoxy-α-L-mannopyranosyl-(1 → 2)-α-L-arabinopyranosyl)ester, intramol. ester, \[2β,3β(S),4α\]- Tubeimoside TUBEIMOSIDE A(P); [Chem. 8](#T2){ref-type="table"}) is the main triterpenoid saponin originated from *Bolbostemma paniculatum* (Maxim) Franquet. which has detoxification and detumescent activities. Recent studies have revealed the pharmacological action of TBMS1 as a potential anti-cancer agent (Wang et al., [@B144]; Gu et al., [@B36]). A research demonstrated that TBMS1 can promote apoptosis, and attenuate migration, invasion of non-small cell lung cancer cells. The underlying mechanism is attributed to upregulation of miR-126-5p, followed by inactivation of VEGF-A/VEGFR-2/ERK signaling pathway. MiR-126-5p inhibitor can reverse the downregulated VEGF-A and VEGFR-2 induced by TBMS1 treatment; moreover, both miR-126-5p inhibitor, and VEGF-A, VEGFR-2 overexpression upregulate the mRNA expression and phosphorylation of MEK1 and ERK. Significantly, apoptosis, migration and invasion of TBMS1-treated cells can be reversed by either miR-126-5p inhibitor or ERK activator (Shi et al., [@B118]). From the above results, it can be concluded that miR-126-5p/VEGF-A/VEGFR-2/ERK signaling is the protective pathway of TBMS1 for cancer therapy. Oridonin -------- Oridonin (ORI, (14R)-7-alpha,20-Epoxy-1-alpha,6-beta,7,14-tetrahydroxykaur-16-en-15-one, [Chem. 9](#T2){ref-type="table"}) is a ent-kaurane diterpenoid compound mainly originated from *Rabdosia rubescens* (Hemsl.) Hara. Traditionally, the herb was convinced to have the effect of detoxification, circulation promotion and pain relief in China. Currently, ORI is illustrated to participate in the treatment of several tumors via different regulatory pathways. It\'s reported that human laryngeal cancer cell is accelerated to apoptosis after ORI treatment through inhibiting EGFR signaling. Similarly, EGFR suppression increased ORI-induced apoptosis by the promotion of oxidative stress, and activation of intrinsic and extrinsic apoptotic pathways (Kang et al., [@B59]). Moreover, 105 miRNAs are involved in the regulation of ORI-treated pancreatic cancer (Gui et al., [@B37]). Therefore, it\'s possible that miRNAs involve in the anti-cancer activity of ORI; however, whether EGFR is the downstream target of miRNAs deserves more researches. Anti-proliferation and Anti-migration Effects of CHMs {#s4} ===================================================== Abnormal cell proliferation is involved in the pathogenesis of many diseases. In particular, the proliferation and invasion of cancer cells are primary contributors to poor patient outcomes (Gao et al., [@B33]). In addition, asthma is also associated with the cell proliferation and migration in airway smooth muscle (Zhao et al., [@B189]). Hence, suppression of cell proliferation and migration are critical methods for treatment of these diseases. Excitingly, some bioactive ingredients of CHMs have been found to inhibit the proliferation and migration of both cancer and asthma cells through targeting miRNA, lncRNA, or ceRNA crosstalk. Curcumin -------- Curcumin (CUR, (1E,6E)-1,7-Bis(4-hydroxy-3-methoxyphenyl)hepta-1,6-diene-3,5-dione; [Chem. 10](#T2){ref-type="table"}) is a phenolic compound extracted from *Curcuma longa* L., which was traditionally used as painkiller in rheumatism and other bone and joint diseases. Recent studies have found that CUR can also act as an anticancer agent, via miRNA and lncRNA targets. CUR inhibits miR-208 and activates expression of the cell cycle suppressor, CDKN1A, resulting in dose-dependent suppression of prostate cancer cell proliferation (Guo H. et al., [@B38]). Further, CUR can significantly increase miR-143 and decrease PGK1 expression, while ectopic expression of FOXD3 can enhance the regulatory effect of CUR on miR-143, thereby inhibiting the proliferation and migration of prostate cancer cells (Cao et al., [@B11]). Further studies reveal that CUR also acts on human prostate cancer stem cells (HuPCaSC). CUR treatment increases the expression of miR-145 and decreases levels of lncRNA-ROR, the cell cycle proteins CCND1, CDK4, and the stem cell markers OCT4, CD44, and CD133. The tumorigenicity of these cells is thereby significantly reduced through inhibition of their proliferation, invasion, and cell cycle arrest (Liu T. et al., [@B83]). Moreover, expression levels of miR-770-5p and miR-1247 in the DLK1--DIO3 imprinted gene cluster were significantly up-regulated, leading to suppression of HuPCaSC proliferation and invasion *in vitro* (Zhang et al., [@B177]). CUR also promotes the expression of miR-98 in lung cancer, thus inhibiting cell growth and migration (Liu W. L. et al., [@B84]). By reducing miR-186^\*^ expression, it induces apoptosis and decreases cell viability in lung cancer cells as well (Tang et al., [@B126]). Furthermore, CUR inhibits both proliferation and accelerates apoptosis in bladder, gastric, non-small cell lung, pancreatic cancers, and hepatic carcinoma via the up-regulation of miR-203 (Saini et al., [@B112]), miR-33b (Sun et al., [@B124]), miR-192-5p (Jin et al., [@B57]), miR-7 (Ma et al., [@B92]), and lncRNA AK125910 (Guo Y. et al., [@B40]), respectively. CUR has also been reported to increase the sensitivity of non-small-cell lung cancer (Lu et al., [@B87]), breast cancer (Zhou S. et al., [@B194]), and nasopharyngeal carcinoma (Wang Q. et al., [@B142]) to chemotherapy drugs by targeting ncRNAs including miR-30c, miR-29b-1-5p, and lncRNA AK294004, respectively, along with their downstream genes. Moreover, CUR can exert synergistic effects in combination with other compounds, to suppress cell proliferation and invasion and induce apoptosis in glioblastoma (Wu et al., [@B152]), breast cancer (Guo et al., [@B39]), and hepatocellular carcinoma (Zhang S. et al., [@B184]). In glioblastoma, miR-378 was found to promote the anticancer effect of CUR by regulating p38 expression, demonstrating the mutual interaction of miRNA and CUR (Li et al., [@B73]). Furthermore, CUR is reported to exert anti-inflammatory effects (Ma F. et al., [@B91]) and to inhibit adipogenic differentiation (Tian L. et al., [@B129]). Notably, as liposome technology is a good method for targeting drug delivery system that can solve the solubility problems of poorly soluble drugs (Allen and Cullis, [@B2]). A research has used this technology to produce CUR-loaded liposome, increasing solubility and oral bioavailability of CUR, as well as reducing first pass effect of hepar. This drug combination can also promote sensitivity of breast cancer cells to chemotherapy, through regulating different miRNAs of miR-29b-1-5p, miR-29b-3p, miR-6068, miR-6790-5p, and miR-4417, as well as their target genes involving DDIT4, EPAS1, VEGFA, RPS14, and DCDC2 (Zhou et al., [@B195]). These data demonstrate that CUR can suppress cell proliferation, growth, and metastasis in various cancers by targeting ncRNAs. In particular, CUR has obvious advantages for the treatment of prostate cancer through its regulation of cancer and cancer stem cells. Moreover, the synergistic effects of CUR with other chemotherapies provide new alternative strategies to address drug resistance. Excitingly, structural improvement of CUR not only ensures its anti-cancer effect, but also promotes the bioavailability. Shikonin -------- Shikonin (SHK, 5,8-dihydroxy-2-((1R)-1-hydroxy-4-methyl-3-penten-1-yl)-1,4-naphthalen-edione; [Chem. 11](#T2){ref-type="table"}) is a naphthoquinone derivative compound. SHK is extracted from the root of the natural herbal medicine, *Lithospermum erythrorhizon* Sieb. et Zucc. This plant was generally used to treat rash, pox, measles, and urticaria in TCM. Modern studies have discovered broader applications for this compound in cancer, by revealing its anti-proliferation function, which is reported to be associated with targeting of miRNAs. SHK can inhibit proliferation and promote apoptosis by modulating the miR-106b/PTEN/AKT/mTOR signaling pathway in endometrioid endometrial cancer (Huang and Hu, [@B47]). Moreover, SHK inhibits the proliferation of breast cancer cells through down-regulation of tumor-derived exosomal miR-128 (Wei et al., [@B148]). In addition, the anticancer activity of SHK in glioblastoma is enhanced by miR-143 by reducing the expression of the anti-apoptosis regulator, BAG3, which is a functional target of miR-143 (Liu et al., [@B80]). Overall, the regulatory relationships between SHK and miRNAs are mutual. SHK could target miR-106b and miR-128 in endometrioid endometrial cancer and breast cancer to prevent cell proliferation. Further, miR-143 expression influences the anticancer activity of SHK in glioblastoma. Finally, the results reviewed above demonstrate that the anti-proliferation activity of SHK in cancers can be attributed to its interactions with miRNAs. Paeoniflorin ------------ Paeoniflorin\[PF,5b-((Benzoyloxy)methyl)tetrahydro-5-hydroxy-2-methyl-2,5-methano-lH-3,4-dioxacyclobuta(cd)pentalen-1a(2H)-yl-beta-D-glucopyranoside; [Chem. 12](#T2){ref-type="table"}\] is the main active ingredient of *Paeonia lactiflora* Pall., which was commonly used to regulate blood circulation and relieve pain in TCM theory. Recent investigations have revealed roles for PF in vasodilation (Goto et al., [@B35]), anti-inflammation (Chen et al., [@B19]; Hu et al., [@B45]), microcirculation improvement (Zhou et al., [@B193]), anti-oxidation (Chen et al., [@B18]), and anti-cancer (Wang et al., [@B137]) activities. Specifically, PF exhibits protective activity in glioma via suppression of cell proliferation and promotion of apoptosis. The potential underlying mechanism may involve upregulation of miR-16 and downregulation of matrix metalloproteinase-9 (MMP-9), which are differentially expressed in glioma tissues and cells compared with healthy controls (Li W. et al., [@B72]). This result lays the foundation for treatment of cancer using PF; however, supporting evidence is insufficient and more investigations are needed. Honokiol -------- Honokiol (HNK, 5,3′-Diallyl-2,4′-dihydroxybiphenyl; [Chem. 13](#T2){ref-type="table"}) is a bioactive polyphenol isolated from *Magnolia grandiflora*. Although the flower was traditionally valued as ornamental, it contains the phenolic ingredient, HNK, which has been shown to have antimicrobial activity (Clark et al., [@B25]). A recent study discovered that HNK has anti-tumor activity; it can markedly inhibit the growth, invasion, and migration of breast cancer cells, and breast-tumor-xenograft growth induced by leptin. HNK promotes the expression of miR-34a, and inhibits WNT1-MTA1-β-catenin signaling, through suppression of STAT3 phosphorylation and recruitment of STAT3 to the promoter of miR-34a (Avtanski et al., [@B3]). Hence, HNK has demonstrated a protective effect on breast cancer in a diet-induced-obese mouse model with high leptin levels and could serve as a new endocrine therapy drug for patients with obesity-related breast cancer accompanied by negative estrogen and progesterone receptors; however, the research described above was limited to animal experiments, and further evidence in humans is required, thus clinical trials are warranted to further investigate HNK. Schisandrin B ------------- SchisandrinB (SchB, 7-dimethyl-ethoxy-stereoisomer;benzo(3,4)cycloocta(1,2-f)(1,3)benzodioxole,5,6,7,8-tetrahydro-1,2,3,13-tetram; [Chem. 14](#T2){ref-type="table"}) is a type of lignan, extracted from *Schisandra sphenanthera* Rehd. et Wils. The original fruit was commonly used to relieve symptoms of cough, gasp, abnormal sweating, nocturnal emission, thirst, and palpitations, under TCM theory. Although it was widely used to treat various diseases in ancient China, its specific target and underlying mechanism of action were unclear. A recent study of Sch B provided information about the involvement of ncRNA. Sch B may increase the expression of miR-150 and subsequently reduce levels of the lncRNA BCYRN1 in airway smooth muscle cells (ASMCs) of asthmatic rats. By regulating these two ncRNAs, Sch B suppresses the proliferation, viability, and migration of ASMCs; therefore, the study generated evidence that partially explains the mechanism underlying the activity of Sch B against asthma (Zhang X. Y. et al., [@B185]). Moreover, Sch B can mediate ceRNA crosstalk between miR-150 and lncRNA BCYRN1, further establishing an miR-150/lncRNA BCYRN1/cell proliferation axis; however, as a new regulatory mechanism influencing asthma, the stability of the ceRNA crosstalk requires further investigation. Resveratrol ----------- Resveratrol (RES, 3,4′,5-trihydroxystilbene; [Chem. 15](#T2){ref-type="table"}) is a natural phenol stilbenoid that is mainly found in food, including the skin of grapes and blueberries, and several CHMs, including *Morus alba* L., *Polygonum cuspidatum* Sieb. et Zucc., and *Rubus idaeus* L. It is considered to protect individuals from cardiovascular diseases, as well as dietary and metabolic diseases (Bradamante et al., [@B10]; Baur et al., [@B5]; Lagouge et al., [@B65]). Recently, its anticancer properties have also been evaluated by researchers and RES has been used as a dietary supplement (Garvin et al., [@B34]; Kalra et al., [@B58]; Roy et al., [@B111]). RES can down-regulate the lncRNA, MALAT1, and up-regulate miR-200c, as well as inhibiting WNT/β-catenin signaling, leading to suppression of cell invasion, metastasis, and migration in colorectal cancer (Ji et al., [@B52]; Karimi Dermani et al., [@B60]). Moreover, by significantly decreasing oncogenic miR-221 and regulating NF-κB and TFG, RES exerts inhibitory effects on melanoma cells, both *in vitro* and *in vivo* (Wu and Cui, [@B151]). In glioma, RES inhibits cell proliferation, arrests the cell cycle in S phase, and induces apoptosis *in vitro*, through down-regulation of miR-21, miR-30a-5p, and miR-19, as well as regulating their targets, including P53, PTEN, EGFR, STAT3, COX-2, NF-κB, and the PI3K/AKT/mTOR pathway (Wang G. et al., [@B136]). RES also has anti-inflammatory effects. It can reduce expression of miR-155 and promote that of its target gene, suppressor of cytokine signaling 1 (SOCS1), leading to subsequent inhibition of the inflammatory factors, TNF-α, IL-6, MAPKs, and STAT1/STAT3 (Ma C. et al., [@B90]). Interestingly, by increasing miR-663 expression, RES down-regulates miR-155, thus acting as both an anti-inflammatory and an anticancer agent (Tili et al., [@B130]). Furthermore, RES exhibits neuroprotective effects. It promotes miR-96 and inhibits its target gene, BAX, resulting in prevention of oxygen/glucose deprivation/re-oxygenation-induced apoptosis and brain damage, while this protective function can be reversed by miR-96 inhibitor (Bian et al., [@B6]). In Alzheimer\'s disease, RES also improves long-term memory formation and induction of long-term potentiation of hippocampus CA1 neurons, through down-regulation of miR-134 and miR-124, and up-regulation of CREB and BDNF (Zhao et al., [@B190]). Therefore, RES is a potential therapeutic agent against cancers, cerebral ischemia, Alzheimer\'s disease, and other inflammatory conditions. Soybean Isoflavones ------------------- Soybean isoflavones (SIF, 3-phenyl-4h-1-benzopyran-4-one; [Chem. 16](#T2){ref-type="table"}) are extracted from *Glycine max* (Linn.) Merr. They act as phytoestrogens in mammals and have been used as dietary supplements. SIF are associated with breast cancer (Douglas et al., [@B27]; Takagi et al., [@B125]). Recently, they have also been demonstrated to suppress cell growth and invasion in prostate cancer. A potential mechanism underlying the anti-prostate cancer activity of SIF is its promotion of miR-29a and miR-1256, leading to down-regulation of TRIM68 and PGK-1 by inhibiting methylation of the miR-29a and miR-1256 promoters (Li et al., [@B75]). Nevertheless, as a controversial ingredient with weak estrogen-like properties, the influence of SIF on hormone-receptor-positive cancers has caused widespread concern. Therefore, research is needed to determine the effectiveness and safety of SIF in the context of different cancers. Matrine ------- Matrine (MAT, (7aS,13aR,13bR,13cS)-Dodecahydro-1H,5H,10H-dipyrido\[2,1-f:3′,2′,1′-ij\](Memczak et al., [@B97]; Song et al., [@B121])naphthyridin-10-one; [Chem. 17](#T2){ref-type="table"}) is the main alkaloid extract from *Sophora flavescens Ait* which was commonly used for diseases of dysentery, eczema and jaundice in China. Modern pharmacological research shows that MAT has protective activity in melanoma, as evidenced by inhibition of proliferation and invasion, and promotion of apoptosis in melanoma cell lines. By downregulating miR-19b-3p expression, MAT increases the protein and mRNA expression of PTEN, a direct target of miR-19b-3p. Similarly, miR-19b-3p downregulation can imitate the effect of MAT; while PTEN silencing reverses the protection induced by MAT (Wei et al., [@B149]). As a result, MAT can exert anti-cancer activity in melanoma via regulating miR-19b-3p/PTEN axis. Corylin ------- Corylin (CL, 3-(2,2-dimethylchromen-6-yl)-7-hydroxychromen-4-one; [Chem. 18](#T2){ref-type="table"}) is the flavonoid compound extracted from *Psoralea corylifolia* Linn. In TCM practice, *Psoralea corylifolia* Linn. was often used for degenerative bone and joint diseases. Newly reported studies have revealed its application in inflammation (Kim et al., [@B62]; Hung et al., [@B51]) and cancer (Chen et al., [@B14]). The anti-cancer activity induced by CL is related to upregulation of tumor suppressor lncRNA GAS5 and its downstream anticancer pathways activation. As a result, the proliferation, migration, and invasiveness, as well as epithelial-mesenchymal transition are all inhibited in hepatocellular carcinoma cells. Moreover, lncRNA GAS5 silencing can attenuate the above inhibitory effect of CL. In an animal experiment, CL is observed to obviously retard tumor growth as well, with no significant physiological toxicity (Chen et al., [@B14]). Taken together, lncRNA GAS5 may act as the treatment target of CL in hepatocellular carcinoma; however, specific downstream gene of lncRNA GAS5 still needs further study. Anti-inflammatory Effects of CHMs {#s5} ================================= Inflammation is a common pathological process involved in many diseases, including coronary heart disease, inflammatory bowel disease, myocarditis, asthma, and neuroinflammatory disorder (Harrington, [@B43]; Robinson et al., [@B110]; Mahajan et al., [@B93]); however, both non-steroidal anti-inflammatory drugs and immunosuppressive agents have clear side effects (Shah and Gecys, [@B115]; Ahmad et al., [@B1]). Consequently, safe and effective anti-inflammatory drugs for the treatment of the basic pathologies underlying the above diseases are still needed. Several bioactive ingredients of CHMs are reported to target miRNA or ceRNA crosstalk, thereby exerting anti-inflammatory effects. Tanshinone IIA -------------- Tanshinone IIA (Tan IIA, Phenanthro \[1, 2-b\]furan-10, 11-dione, 6, 7, 8, 9-tetrahydro-1, 6, 6-trimethyl; [Chem. 19](#T2){ref-type="table"}) is a lipophilic diterpenoid extracted from the root of *Salvia miltiorrhiza* Bge. Under TCM theory, the original herb is considered to promote blood circulation. Recent studies have illustrated that Tan IIA has cardioprotective activity (Shang et al., [@B117]; Feng et al., [@B29]) and injection of sodium Tan IIA sulfonate has been widely used as an adjunctive therapy for cardiovascular diseases in China (Yu M. L. et al., [@B169]). A potential mechanism underlying its inhibition of inflammation (Pan et al., [@B104]; Cheng et al., [@B22]), and an upstream regulator, is miRNA. Tan IIA can reduce the expression levels of cytokines, chemokines, and acute-phase proteins, including TLR4, MyD88, GM-CSF, sICAM-1, CXCl-1, and MIP-1α. Moreover, it significantly inhibits the mRNA expression levels of IL-1β, TNF-α, and COX-2, thereby suppressing lipopolysaccharides (LPS) -induced activation of the TLR4-NF-κB pathway. Furthermore, expression of miR-155, miR-147, miR-184, miR-29b, and miR-34c is also reduced by Tan IIA, and these may be upstream regulators in anti-inflammation processes (Fan et al., [@B28]). In addition, by down-regulation of miR-146b and miR-155, Tan IIA significantly reduces the levels of inflammatory factors, including CRP, ox-LDL, IL-1β, IL-6, IL-12, TNF-α, CCL-2, CD40, and MMP-2, thereby exerting protective functions in atherosclerosis induced by *Porphyromonas gingivalis* (Xuan et al., [@B156]). Another study indicated that Tan IIA can also inhibit apoptosis caused by hypoxia. Through increasing miR-133 expression and activating the stress-induced protein kinase, MAPK ERK1/2, Tan IIA enhances resistance to hypoxic exposure in neonatal cardiomyocytes (Zhang et al., [@B180]). Treatment with Tan IIA has also been illustrated to reverse the abnormal expression of miR-1, SRF, and MEF2, and participates in suppression of the p38 MAPK signaling pathway, restoring declined I(K1) current density and Kir2.1 and Cx43 protein levels, thus lowering the incidence of arrhythmogenesis and mortality after myocardial infarction, and improving cardiac function (Shan et al., [@B116]; Zhang et al., [@B187]). These results provide a partial explanation for the anti-inflammatory and anti-hypoxia activity of Tan IIA via miRNAs in cardiovascular diseases; in particular, miR-155 may be a specific target of Tan IIA in inflammation. Additionally, an aqueous extract from *Salvia miltiorrhiza* Bge., named magnesium lithospermate B (MLB, magnesium (2R)-3-(3,4-dihydroxyphenyl)-2-\[(E)-3-\[(2S,3S)-2-(3,4-dihydroxyphenyl)-3-\[(2R)-3-(3,4-dihydroxyphenyl)-1-oxido-1-oxopropan-2-yl\]oxycarbonyl-7-hydroxy-2,3-dihydro-1-benzofuran-4-yl\]prop-2-enoyl\]oxypropanoate; [Chem. 20](#T2){ref-type="table"}), has neuroprotective effect in ischemia/reperfusion (I/R) injury. I/R injury can lead to miR-107 upregulation, glutamate transporter 1 (GLT-1) suppression and glutamate accumulation, increasing neurological deficit score, infarct volume and cellular apoptosis (Yang Z. B. et al., [@B165]). MLB treatment improves I/R-induced cerebral injury through reversing the abnormal expressions of miR-107, GLT-1 and glutamate (Yang et al., [@B164]). The above results may help to throw light on the underlying mechanisms of Tan IIA and MLB for the treatment of cardiovascular and cerebrovascular diseases from the perspective of miRNA; however, it should be noted that the pharmacological action of *Salvia miltiorrhiza* Bge. is not limited to ncRNAs (Zhu et al., [@B197]). Baicalin -------- Baicalin (BA, 7-D-glucuronic acid-5,6-dihydroxy-flavone; [Chem. 21](#T2){ref-type="table"}) is a flavone glycoside extracted from *Scutellaria baicalensis* Georgi, which was commonly applied for the treatment of respiratory and digestive diseases in CHM. The traditional treatment effects may be related to regulation of inflammatory responses. TNF-α stimulation promotes the expression of miR-191a, causing downregulation of ZO-1 mRNA and protein. BA pretreatment could reverse the effects of ZO-1 and miR-191a expression induced by TNF-α, leading to improved viability and migration of rat small intestine epithelial cells. Furthermore, knockdown of miR-191a expression significantly increased BAL-induced ZO-1 protein expression, thereby enhancing the protective effect of BA on cell motility (Wang L. et al., [@B140]). These data suggest that miR-191a may be an upstream target of BA in the treatment of inflammatory bowel disease; moreover, the therapeutic effects of BA can also be influenced by miR-191a. Also, another research illustrates the proliferative inhibition of mouse embryonic stem cells induced by baicalin. Baicalin suppresses the expression of miR-294, c-jun and c-fos; while miR-294 overexpression could significantly reverse the above effect of baicalin, indicating miR-294 may be the treatment target (Wang J. et al., [@B138]). Therefore, BA exerts anti-inflammatory and anti-proliferative effects by targeting miRNAs and emerges as a potential treatment agent for digestive disease. Further, as BA remains one of the most frequently used medicines for the treatment of cough and phlegm, the activity of BA in respiratory disease warrants similar studies. Cinnamaldehyde -------------- Cinnamaldehyde (CA, 3-phenylprop-2-enaldehyde; [Chem. 22](#T2){ref-type="table"}) is a conjugated aromatic aldehyde extracted from the bark of the Chinese herb, *Cinnamomum cassia* Presl. According to TCM theory, the traditional plant can enhance the function of "yang qi" (a substance with excitatory function in TCM) and is often used to relieve symptoms of weakness. Recent studies have broadened the application of this preparation to the treatment of cerebrovascular diseases, ulcerative colitis, and cancer (Zhao et al., [@B188]; Tian F. et al., [@B128]; Qu et al., [@B107]), where it acts by exerting anti-inflammatory or ncRNA regulatory functions. CA improves symptoms of weight loss, disease activity index, and infiltration of inflammatory cells, by decreasing the levels of pro-inflammatory cytokines, including TNF-α, IL-1β, and IL-6, as well as the NLRP3 inflammasome, miR-21, and miR-155, in both colon tissue and macrophages. Moreover, levels of reactive oxygen species were also reduced, along with the phosphorylation of AKT, mTOR, and COX2 proteins. Further experiments revealed similar suppression of IL-1β and IL-6 in response to miR-21 or miR-155 inhibitors, demonstrating that these inflammatory factors are positively regulated by miR-21 or miR-155 (Qu et al., [@B107]). As a result, CA suppresses the miR-21/miR-155/IL-1β/IL-6 axis to exert its protective function in ulcerative colitis. CA also has anti-cancer activity through regulation of ceRNA crosstalk and can suppress cell proliferation and induce apoptosis in non-small cell lung cancer. Through upregulation of has-circ-0043256 and ITCH expression, CA inhibits the WNT/β-catenin pathway, while this function can be partially abolished by miR-1252, indicating that miR-1252 may participate in has-circ-0043256-related regulation. Moreover, has-circ-0043256 knockdown can reverse the effects of CA on cells (Tian F. et al., [@B128]). Consequently, has-circ-0043256/miR-1252/ITCH crosstalk may contribute to the anticancer activity of CA. Geniposide ---------- Geniposide (GEN, methyl (1S,4aS,7aS)-7-(hydroxymethyl)-1-\[(2S,3R,4S,5S,6R)-3,4,5-trihydroxy-6-(hydroxymethyl)oxan-2-yl\]oxy-1,4a,5,7a-tetrahydrocyclopenta\[c\]pyran-4-carboxylate; [Chem. 23](#T2){ref-type="table"}) is derived from *Gardenia jasminoides Ellis*, a traditional antipyretic and detoxifying CHM. A recent research reported GEN has effects of anti-inflammatory and cardiomyocyte protection in LPS-injured H9c2 cells. It up-regulates miR-145 expression, inhibits pro-inflammatory factors of IL-6, TNF-α, and MCP-1, and then suppresses the MEK/ERK pathway, thus promoting cell viability and inhibiting apoptosis. Moreover, miR-145 inhibitor could reverse the above protective function induced by GEN pretreatment (Su et al., [@B122]). Therefore, GEN becomes a potential therapeutic agent for myocarditis in practice by targeting miR-145 and anti-inflammation in cardiomyocyte. Carvacrol/Thymol ---------------- Carvacrol (Car, 5-Isopropyl-2-methylphenol; [Chem 24](#T2){ref-type="table"}) and Thymol (Thy, 2-Isopropyl-5-methylphenol; [Chem. 25](#T2){ref-type="table"}) are isolated from the essential oil of *Origanum vulgare* L. or wild bergamot. They are isomers and belong to monoterpenoid phenol. Traditionally, *Origanum vulgare* L. was applied for the treatment of cold and heatstroke. Bergamot can relieve pain and vomit under TCM theory. A research further expanded the applicable scope of these herbs by studying the two bioactive ingredients. Car/Thy can suppress the allergic inflammation in asthma by regulating miRNAs and inflammatory factors. In chitin-induced model, expression levels of miR-155, miR-146a and miR-21, promotor of pro-inflammatory cytokines, are upregulated. Furthermore, SOCS1 and SHIP1, targets of miR-155 and negative regulators of TLR-mediated inflammation, are demonstrated to be inhibited by chitin. However, Car/Thy treatment can reverse the abnormal expressions of TLR2, TLR4, SOCS1, SHIP1, and miR-155, miR-146a, miR-21 (Khosravi and Erle, [@B61]). These results preliminarily establish the relationships between anti-inflammation effect of Car/Thy and miRNAs; but the specific target and corresponding regulatory network are not reported regrettably. Boswellic Acids --------------- Boswellic acids are extracted from oleo-gum-resin of *Boswellia serrata*, a traditional CHM with promoting blood circulation and pain relief function. Boswellic acids contain different ingredients, among which 3-acetyl-11-keto-β-boswellic acid (AKBA, (3R,4R,6aR,6bS,8aR,11R,12S,14bS)-3-acetyloxy-4,6a,6b,8a,11,12,14b-heptamethyl-14-oxo-1,2,3,4a,5,6,7,8,9,10,11,12,12a,14a-tetradecahydropicene-4-carboxylic acid; [Chem. 26](#T2){ref-type="table"}) possesses the most potent anti-inflammatory activity (Siddiqui, [@B120]; Sayed et al., [@B114]). AKBA can attenuate the behavioral dysfunction in LPS-induced neuroinflammation, similarly with that effect of dexamethasone. Moreover, AKBA lowers expression of miR-155, P-IκB-α, and carbonyl protein, and increases contents of normal cytokine and SOCS-1, resulting in effects of anti-apoptotic and anti-amyloidogenic (Sayed et al., [@B114]). Therefore, regulation of miR-155 and downstream protein helps to reveal the possible mechanism underneath AKBA\'s positive role in neuroinflammatory disorders; however, the specific target deserves more verification. Anti-atherosclerosis Effects of CHMs {#s6} ==================================== Atherosclerosis is the basic pathology underlying coronary artery disease, cerebral infarction, and other vascular diseases (Pothineni et al., [@B106]; Li Q. et al., [@B76]). Medicines with anti-atherosclerosis activities are therefore highly significant for the prevention and treatment of these diseases. Statins are currently the main drugs used against atherosclerosis; however, when taken for long periods of time, they risk impairing liver function and causing muscle lysis, particularly in elderly patients (Guyton, [@B41]; Ramakumari et al., [@B108]). Therefore, better drugs with relatively few side effects are needed and CHMs represent a good resource in this context. Recent studies have identified three bioactive ingredients of CHMs as regulators of atherosclerosis through targeting miRNA or lncRNA. Sinapic Acid ------------ Sinapic Acid (SA, 3,5-dimethoxy-4-hydroxycinnamic acid; [Chem. 27](#T2){ref-type="table"}) is the bioactive ingredient isolated from seeds of the Chinese herb, *Sinapis alba* L. The seeds were commonly used to treat cough, phlegm, limb numbness, and chronic abscess. A recent study reported protective effects of SA in atherosclerosis, which helped to partially explain the original application of the seed to treat limb numbness. The lncRNA MALAT1 is significantly up-regulated in rats with diabetic atherosclerosis and low-dose SA treatment can suppress this abnormal expression. Subsequently, pyroptotic death of bone marrow derived macrophages is inhibited, accompanied by decreased expression of ET-1 and IL-1β, and the pyroptotic proteins, ASC, NRLP3, and Caspase-1 (Han et al., [@B42]). Hence, SA exerts anti-inflammatory activity and prevents pyroptosis, thus exerting anti-atherosclerosis effects, through targeting lncRNA MALAT1; however, the efficacy and safety of SA as a potential treatment agent require verification by additional studies. Polydatin --------- Polydatin (PLD, 3,4,5-trihydroxystilbene-3-beta-monoglucoside; [Chem. 28](#T2){ref-type="table"}), also known as Piceid, is the bioactive ingredient from *Polygonum cuspidatum* Sieb. et Zucc. This herb was traditionally used for the treatment of jaundice and cough. However, PLD has also been found to exert protective effects against cardiac hypertrophy (Zhang et al., [@B181]), insulin resistance, and hepatic steatosis (Zhang et al., [@B182]). Further, a recent study has revealed the underlying regulatory action of PLD in atherosclerosis with liver dysfunction. The findings indicated that PLD treatment can markedly lower increased blood glucose, serum ALT, AST, TC, TG, and LDL-C in mice with high-fat diet. Simultaneously, changes in HDL-C, MDA, SOD, and miR-214 were also improved in liver tissue (Zhou et al., [@B192]). This study indicates that PLD can be therapeutically effective in complex diseases by regulating various factors. In addition, PLD shows great potential as a complement to treatment for statin-induced liver damage via its anti-atherosclerosis and liver protection properties; however, the above study only reported the expression levels of various factors induced by PLD, rather than systematically studying the relationships between miR-214 and its target genes. Therefore, further in-depth investigations are required in the future. Ampelopsin ---------- Ampelopsin \[(2r,3r)-3,5,7-trihydroxy-2-(3,4,5-trihydroxyphenyl)chroman-4-one; [Chem. 29](#T2){ref-type="table"}\], alsocalled dihydromyricetin (DHM), is the main flavonoid compound from *Ampelopsis grossedentata*. The original herb has effects of detoxification, anti-inflammatory and analgesic, commonly used as a dietary supplement. DHM is now demonstrated to impede atherosclerotic process by regulating endothelial dysfunction (Yang D. et al., [@B158]), and exert anti-aging effect against neurodegenerative diseases (Kou et al., [@B64]). It inhibits miR-21 expression and then improves endothelial dysfunction induced by TNF-α, accompanied by suppression of abnormal expression of eNOS, DDAH1, NO, and ADMA, as well as improvement of tube formation and migration. Furthermore, miR-21 blockade can produce similar effects with DHM treatment; while miR-21 overexpression abolishes the above protection. Additionally, improvement of endothelial dysfunction can be reversed by a non-specific NOS inhibitor, indicating DHM ameliorates vascular endothelial function and inhibits atherosclerosis by targeting miR-21-mediated DDAH1/ADMA/NO signal pathway (Yang D. et al., [@B158]). Another study reveals that miR-34a is upregulated in D-gal-induced brain aging rats; while DHM management can inhibit the abnormal expression. Moreover, DHM suppresses apoptosis and ameliorates impaired autophagy of neurons in D-gal-injured hippocampus tissue, by up-regulating SIRT1 and down-regulating mTOR signal pathways (Kou et al., [@B64]). Therefore, DHM possesses anti-aging effect partially through regulating miR-34a-mediated SIRT1-mTOR signal pathway, showing important role for the treatment of neurodegenerative diseases. From the above results, it can be seen that DHM exerts not only anti-atherosclerosis effect, but also anti-aging function by targeting miRNAs and downstream signaling pathways. Anti-infection Effects of CHMs {#s7} ============================== Antibiotics and antiviral drugs are basic treatments for infectious diseases; however, a deteriorating situation caused by antibiotic abuse, drug-resistance, and viral mutations is shifting the focus of research attention to other therapeutic and complementary drugs for treatment of these conditions (Miyoshi et al., [@B99]; Jiang et al., [@B55]). To date, two bioactive ingredients of CHMs have been found to contribute to the treatment of infection through regulation of miRNAs. These substances can protect the human body from pathological damage, although they do not directly induce pathogen resistance. Icariine -------- Icariine \[ICA, 2-(4′-methoxyphenyl)-3-rhamnosido-5-hydroxyl-7-glucosido-8-(3′-methyl-2-butylenyl)-4-chromanone; [Chem. 30](#T2){ref-type="table"}\] is the main bioactive flavonoid glucoside extracted from *Epimedium brevicornu* Maxim. Under TCM, this herb was considered to nourish "yang qi" and generally applied for treatment of osteoarticular and reproductive diseases. ICA can suppress osteoclast bone resorption and bone loss, indicating great potential for use as a treatment agent for both aseptic loosening and bacteria-induced bone loss (Zhang et al., [@B186]; Liu et al., [@B79]). Specifically, ICA can restore LPS-induced bone loss, without obvious cytotoxicity. This product can down-regulate expression of the osteogenic inhibitor, miR-34c, while it up-regulates levels of the key transcription factor, RUNX2, thereby inducing osteogenic differentiation and mineral nodule formation. Moreover, miR-34c overexpression can reverse these effects of ICA. Additionally, ICA markedly suppresses LPS-induced activation of JNK, p38, and NF-kB pathways, leading to therapeutic effects in diseases causing bacteria-induced bone loss, such as osteomyelitis and septic arthritis (Liu et al., [@B79]). Interestingly, ICA also exhibits anticancer activity in ovarian cancer via down-regulation of miR-21 expression. This subsequently induces PTEN, RECK, and Caspase-3 activity, while BCL-2 protein expression is inhibited, leading to suppression of cell proliferation and increased apoptosis (Li J. et al., [@B70]). Based on these findings, miR-34c appears to facilitate the mechanisms underlying the role of ICA in infectious bone loss. Furthermore, the identification of miR-21 suggests a potential new application of ICA in cancer therapy. Therefore, there is promise that additional currently unknown functions of this medicinal herb could be determined by studying ncRNA and related regulatory networks. Ginsenosides ------------ Ginsenosides (GS,(3S,5R,8R,9R,10R,14R,17S)-17-(2-hydroxy-6-methylhept-5-en-2-yl)-4,4,8,10,14-pentamethyl-2,3,5,6,7,9,11,12,13,15,16,17-dodecahydro-1H-cyclopenta\[a\]phenanthren-3-ol; [Chem. 31](#T2){ref-type="table"}), also referred to as panaxosides, are a class of natural steroid glycosides and triterpene saponins. These products include various active components, such as ginsenoside Re, Rg, Rh, Rb, and Rc. GS are mainly isolated from *Panax ginseng* C. A. Mey., a valuable herb with nourishing effects and a long history of use in ancient China. At present, GS products are not only used to promote health, but also for their activity as immune regulators in many diseases (Jiang Z. et al., [@B56]; Shin et al., [@B119]; Yu X. et al., [@B171]). GS exert a cytoprotective effect, thereby promoting cell viability on avian influenza H9N2/G1 infection. During this process, the expression of miR-15b was up-regulated, while production of IP-10 was markedly inhibited. Furthermore, cytometry and TUNEL analyses indicated that ginsenoside Re prevents apoptosis and DNA damage in human endothelial cells caused by H9N2/G1 (Chan et al., [@B13]). These results are inconsistent with the traditional concept that GS is only suitable for treatment of sub-optimal health status or chronic diseases and greatly expand the potential for application of GS for treatment of acute infectious diseases in the future. Anti-senescence Effects of CHMs {#s8} =============================== Cell senescence is the irreversible state in which cells undergo cycle arrest responding to various factors (Watanabe et al., [@B147]). It participates in biological processes involving embryonic development, wound healing and aging, closely relating to organismal aging and diseases, and thus arousing widespread concerns in researchers (Watanabe et al., [@B147]; de Magalhães and Passos, [@B26]). Currently, several bioactive ingredients of CHMs have been found to act positive roles in anti-senescence. Salidroside ----------- Salidroside \[SAL, 2-(4-Hydroxyphenyl)ethyl beta-D-glucopyranoside; [Chem. 32](#T2){ref-type="table"}\] is the main bioactive extract from *Rhodiola rosea* L. with effects of nourishing "yang qi" and promoting blood circulation under TCM theory. Modern pharmacological study further revealed that the medicinal herb not only exerts anti-fatigue ability, but also improves resistance to hypoxia (Li et al., [@B68]). Moreover, SAL has been supported to possess anti-senescence activity. The potential mechanism is related to regulation of multiple miRNAs expression. Through upregulating let-7c, let-7e, miR-3620, and decreasing expression of miR-411, miR-24-2-5p and miR-485-3p in the aging cells, SAL participates in several pathways involving p53, transcription factor CREB and AKT/mTOR signaling (Zhang J. et al., [@B178]). As is known that both let-7 and mTOR are aging-related (Marasa et al., [@B95]; Wu et al., [@B153]), and the former factor can directly inhibit the expression of the latter (Marcais et al., [@B96]). Therefore, it\'s possible that SAL possesses anti-senescence effect by regulating let-7 and mTOR; however, the predicted regulatory relationship requires more validation in the future. Phlorizin --------- Phlorizin (PZ, Phloretin-2′-O-beta-glucoside; [Chem. 33](#T2){ref-type="table"}) is the main active ingredient of *Acanthopanax senticosus* (Rupr. et Maxim.) Harms which is a traditional CHM with functions.of nourishing and enhancing strength. PZ is convinced to exert effects of anti-fatigue, learning improvement and immune-enhancing (Huang et al., [@B49]). Researches further reported that PZ can act as a promising agent for skin aging (Zhai et al., [@B173]; Choi et al., [@B24]). By promoting epidermal cell proliferation and self-renewal, PZ thickens epidermis to maintain skin structure and resistance to aging. Moreover, PZ increases expression of p63 and proliferating cell nuclear antigen (PCNA), as well as integrin α6, integrin β1 and type IV collagen. Particularly, the mRNA of type IV collagen is also increased and possibly regulated by downregulation of miR-135b (Choi et al., [@B24]). As a result, miR-135b/type IV collagen axis may be the underlying regulatory mechanism of anti-senescence induced by PZ. Osthole ------- Osthole (Ost, 7-methoxy-8-(3-methyl-2-butenyl)-2H-1-benzopyran-2-one; [Chem. 34](#T2){ref-type="table"}) is mainly extracted from *Cnidium monnieri* (L.) Cuss. which was commonly used for nourishing "yang qi" and relieving itching in TCM practice. Current pharmacological researches newly revealed that Ost can improve memory, delay senescence and resist cell damage in Alzheimer\'s disease (AD) (Hu et al., [@B46]; Zheng et al., [@B191]). As it\'s clear that beta-amyloid peptide (Aβ) is the critical pathology of AD, inhibition of Aβ deposition thereby becomes an important treatment strategy for the disease (Wilcock et al., [@B150]). Ost was reported to enhance cyclic AMP response element-binding protein (CREB) phosphorylation and then inhibit Aβ cytotoxicity on neural cells (Hu et al., [@B46]). Further mechanism study indicates that it upregulates miR-107, and then promotes cells viability of neuron, resulting in suppression of the protein expression of Aβ and BACE1, as well as LDH (Xiao et al., [@B155]). Therefore, Ost exertes obvious neuroprotective effect through targeting miR-107 and impeding Aβ deposition, presenting as a potential treatment agent for neurogenic aging and neurodegenerative disease. Inhibitory Effects of CHMs on Structural Remodeling {#s9} =================================================== Structural remodeling is an important factor that can impede the normal functions of tissues and organs. It is also the main pathological change during the late stages of various diseases, making poor prognosis and difficult treatment (Bijkerk et al., [@B7]; Bittencourt et al., [@B8]; Zhuang et al., [@B198]). Encouragingly, three CHM ingredients have been demonstrated to exert protective effects on this pathological change through regulation of miRNAs. Panax Notoginseng Saponins -------------------------- Panax notoginseng saponins (PNS, notoginsenoside-fe, 98%; [Chem. 35](#T2){ref-type="table"}) are a chemical mixture extracted from the root of *Panax notoginseng* (Burk.) F. H. Chen. According to TCM theory, the traditional herb can simultaneously promote blood circulation and prevent bleeding; therefore, it was commonly used to treat coronary artery disease, stroke, gastrointestinal bleeding, irregular menstruation, and bruises. Currently, several PNS preparations, including xuesaitong injections and xuesetong capsules, are widely used to treat cardiovascular diseases (Song et al., [@B121]). The improvement in cardiac prognosis caused by PNS has been attributed to its regulation of miRNAs and suppression of structural remodeling. PNS was reported to increase expression of the anti-fibrotic factor, miR-29c, which is clearly reduced in mice with isoproterenol-induced myocardial fibrogenesis, leading to downregulation of its target genes: collagen (Col) 1a1, Col1a2, Col3a1, Col5a1, Fbn1, and TGFβ1, thus exerting protective effects against myocardial injury and fibrosis (Liu L. et al., [@B81]). In addition, PNS has obvious resistance to H~2~O~2~-induced oxidative damage, showing anti-apoptosis activity in vascular endothelial cells (VECs) by suppressing miR-146b-5p expression (Wang J. et al., [@B146]). Moreover, notoginsenoside R1, one main component of PNS, can delay the process of senescence in VECs by regulating miR-34a/SIRT1/p53 pathway (Lai et al., [@B66]). As a result, through repairing VECs damages, PNS inhibits vascular pathological process. PNS also has an active role in tumors complicated by myocardial ischemia where paradoxical treatment strategy existed. PNS and its major components, Rg1, Rb1, and R1, are implicated in tissue-specific regulation of angiogenesis, and can inhibit tumor growth, as well as attenuating myocardial ischemia. The potential underlying mechanism may be the down-regulation of miR-18a and vascular markers (CD34 and vWF) in tumor, with simultaneous up-regulation of these factors in heart (Yang Q. et al., [@B162]). Notably, by modulating Met/miR-222 axis, and then increasing target genes of tumor suppressor p27 and PTEN expression, PNS selectively inhibits the survival of lewis lung carcinoma cells and attenuates tumor growth in mice (Yang Q. et al., [@B161]). Based on the above results, PNS appears to exert its cardioprotective function by preventing fibrosis, improving vascular endothelium, and promoting angiogenesis in the heart. Simultaneously, considering the cardioprotection and anti-tumor effects through targeting miRNAs, PNS is especially suitable for patients with heart disease and tumor. Tetrandrine ----------- Tetrandrine (TET, 6,6′,7,12-tetramethoxy-2,2′-dimethyl-1 beta-berbaman; [Chem. 36](#T2){ref-type="table"}) is a bisbenzylisoquinoline alkaloid extracted from *Stephania tetrandra* S. Moore. The medicinal plant was often used for treatment of rheumatism and edema under the theories of TCM. Notably, the source plant should not be confused with *Radix Aristolochiae Fangchi* which causes nephrotoxicity, despite their similar Chinese names. A recent study identified a new pharmacological effect of TET in treatment of anti-hypertrophic scarring. The underlying mechanism was suggested to be repression of DNA and collagen synthesis in scar-derived fibroblasts (Liu et al., [@B78]). Furthermore, by upregulating miR-27b and downregulating miR-125b, TET influenced the expression of putative targets, including VEGFC, BCL2L12, COL4A3, and FGFR2, predicted to contribute to several scar and wound healing-related signaling pathways (Ning et al., [@B103]). Consequently, TET has therapeutic potential for the inhibition of skin tissue hyperplasia after wounding or surgery. Leonurine --------- Leonurine (LEO, 4-Guanidino-n-butyl syringate; [Chem. 37](#T2){ref-type="table"}) is the alkaloid compound from *Leonurus artemisia* (Laur.) S. Y. Hu F, which was commonly used for gynecological diseases in TCM. Newly reported results have indicated the cardioprotective effect by studying LEO activity, namely anti-atherosclerosis (Jiang T. et al., [@B54]), anti-oxidation (Gao et al., [@B32]) and resistant to cardiomyocyte hypertrophy (Lu et al., [@B86]). LEO treatment can significantly reduce the surface area of hypertrophic cardiomyocytes, decrease the content of atrial natriuretic peptide (ANP), endothelin-1 (ET-1), p38 mitogen-activated protein kinase (p38 MAPK), phosphorylated p38 MAPK (p-p38 MAPK), myocyte enhancer factor 2 and β-myosin heavy chain. Moreover, it also up-regulates the expression levels of α-myosin heavy chain protein and miR-1. Thus, by upregulating miR-1 expression and then inhibiting the activation of p38MAPK signaling pathway, LEO may inhibit AngII-induced cardiomyocyte hypertrophy and structural remodeling (Lu et al., [@B86]). Other Effects of CHMs {#s10} ===================== Bioactive ingredients of CHMs can exert various protective effects through targeting different miRNA, lncRNA, or circRNA. Besides the above mentioned mechanisms, some ingredients also play positive roles in anti-I/R injury, anti-arrhythmia, recovery of blood-spinal cord barrier, and promotion of cardiac differentiation by targeting lncRNA and miRNA. Calycosin/Astragaloside IV -------------------------- Calycosin (CAL, 7,3′-dihydroxy-4′-methoxyisoflavone; [Chem. 38](#T2){ref-type="table"}) is a natural phytoestrogen derived from *Astragalus membranaceus* (Fisch.) Bunge. which can nourish "yang qi" and was commonly used for cardiovascular and cerebrovascular diseases in TCM practice. It\'s indicated that the neuroprotection effect of CAL is related to miRNA. CAL markedly improves the infarcted volume, brain water content, and neurological deficit in cerebral I/R injury rats, by upregulating miR-375, ER-α and Bcl-2, and inhibiting RASD1 expression (Wang Y. et al., [@B145]). Regrettably, a systematic mechanism of miR-375 and those downstream targets has not been revealed in this study. Moreover, CAL also possesses positive role in anti-cancer, enriching the pharmacological effect and application of *Astragalus membranaceus* extracts (Tseng et al., [@B131]; Kong et al., [@B63]). It\'s demonstrated that CAL significantly impedes lncRNA EWSAT1 expression in nasopharyngeal carcinoma (NPC), followed by influenced downstream factors and pathways, leading to inhibitory growth. Furthermore, lncRNA EWSAT1 overexpression can reverse CAL-induced effect, indicating lncRNA EWSAT1 act as the specific target of CAL promisingly (Kong et al., [@B63]). Additionally, Astragaloside IV (ASIV, 3-O-beta-D-xylopyranosyl-6-O-beta-D-glucopyranosylcycloastragenol; [Chem. 39](#T2){ref-type="table"}), another bioactive compound of *Astragalus membranaceus* (Fisch.) Bunge., can ameliorate precancerous lesions of gastric carcinoma (PLGC) markedly. It lowers mRNA and protein expressions of LDHA, MCT1, MCT4, HIF-1α, and CD147, as well as increasing TIGAR and p53 content. Furthermore, ASIV treatment promotes miR-34a expression. As a result, ASIV improves abnormal glycolysis and dysplasia possibly via regulation of miR-34a/LDHA pathway (Zhang C. et al., [@B174]). Interestingly, the total flavonoids of *Astragalus membranaceus* (Fisch.) Bunge. (TFA) can improve heart function damaged by viral myocarditis. By upregulating the expression of miR-378 and miR-378^\*^ in cardiomyocytes infected with coxsackie B3 virus, TFA may inhibit cardiac hypertrophy and improve prognosis (Nagalingam et al., [@B102]; Wan et al., [@B134]). Therefore, it can be speculated that the heart protection of TFA is attributed to inhibition of myocardial pathology by regulating miR-378 and miR-378^\*^. From the above results, it can been seen that although CAL, ASIV and TFA are extracted from the same herb, they have different targets and are applicable for distinct diseases, demonstrating the study necessity of identified ingredients and targets from CHMs. Paeonol ------- Paeonol (PAE, 2′-Hydroxy-4′-methoxyacetophenone; [Chem. 40](#T2){ref-type="table"}) is the main bioactive ingredient of *Paeonia suffruticosa* Andr. and *Cynanchum paniculatum* (Bunge) Kitagawa. The two herbs promoted blood circulation and could be used for cardiovascular diseases in TCM practice. Further pharmacological research shows PAE significantly reduces the incidence of ischemic arrhythmia in rats, including lowered frequency of ventricular premature beat, ventricular tachycardia and ventricular fibrillation. Moreover, it markedly decreases infarct size of myocardium. The potential treatment target is miR-1 which is inhibited by PAE in cardiomyocytes (Zhang and Xiong, [@B179]). Nevertheless, the above study only reveals the possible regulatory relationship between PAE and miR-1, which needs more verification and further identifies the downstream target gene of miRNA. Salvianolic Acid A ------------------ Salvianolic acid A (Sal A, (R)-3-(3,4-Dihydroxyphenyl)-2-(((E)-3-(2-((E)-3,4-dihydroxystyryl)-3,4-dihydroxyphenyl) acryloyl)oxy)propanoic acid; [Chem. 41](#T2){ref-type="table"}) is derived from water-soluble phenolic compound of *Salvia miltiorrhiza* Bge. It has protective effects of anti-IR injury (Jiang et al., [@B53]; Yang et al., [@B163]), recovery of neurological function (Yu D. S. et al., [@B168]) and anti-cancer activities (Chen et al., [@B15]; Lu et al., [@B85]), the latter two pharmacological actions of which are attributed to miRNA regulation. It\'s reported that Sal A significantly increases expression of tight junction proteins and HO-1, and decreases p-caveolin-1 and apoptosis-related proteins, resulting in recovery of blood-spinal cord barrier integrity after spinal cord injury (SCI). Furthermore, HO-1 inhibitor can attenuate the regulation of ZO-1, occluding, and p-caveolin-1 by Sal A. The underlying target and mechanism may be upregulation of miR-101 which promotes expression of nuclear factor erythroid 2-related factor 2 (Nrf2) and HO-1. Conversely, miR-101 inhibitor accelerates the permeability of rat brain microvascular endothelial cells, and the protein of Cul3 by targeting its mRNA. As a result, Sal A improves neurological function after SCI through targeting miR-101/Cul3/Nrf2/HO-1 signaling pathway (Yu D. S. et al., [@B168]). Another study indicated that Sal A can also down-regulate the expression of multidrug resistance gene MDR1 in lung cancer, thereby emerging as a new treatment agent for lung cancer resistance. The potential mechanism may be related to up-regulation of 4 miRNA expressions including miR-3686, miR-4708-3p, miR-3667-5p, and miR-4738-3p (Chen et al., [@B15]). This study attempts to find the upstream target of Sal A against MDR1 from perspective of post-transcriptional regulation; but current result cannot directly confirm the regulatory correlation between the 4 miRNAs and MDR1. Thus, more and deeper experiments are urgently needed in the future. Andrographolide --------------- Andrographolide (Andro, 17-hydro-9-dehydro-andrographolide; [Chem. 42](#T2){ref-type="table"}) is a diterpenoid lactone compound derived from *Andrographis paniculata* (Burm. f.) Nees, a natural anti-bacterial and anti-viral CHM. A recent study further demonstrates that Andro can inhibit hepatoma tumor growth. It promotes the expression of 22 miRNAs, but declines that of other 10 miRNAs in a xenograft mouse tumor model *in vivo*. Among those upregulated miRNAs, miR-222-3p, miR-106b-5p, miR-30b-5p, and miR-23a-5p are confirmed in cell experiments *in vitro*. Functional analysis reveals that those miRNAs are mainly involved in signaling pathways of miRNAs in cancer, MPAKs and focal adhesion. Moreover, 24 target genes involved in the above signaling pathways are illustrated to be consistent with miRNAs expression (Lu et al., [@B85]). As a result, Andro prevents hepatoma tumor growth partially through regulating miRNA profile; whereas the specific target and underlying mechanism still need deeper study. Puerarin -------- Puerarin (PUE, 7,4\'-Dihydroxy-8-C-glucosylisoflavone; [Chem. 43](#T2){ref-type="table"}) is the main active ingredient extracted from *Radix Puerariae Lobatae* which improved symptoms of fever, neck stiffness, thirst and diarrhea in ancient China. Modern researches reveal the cardiovascular protection of PUE for myocardial infarction (Zhang et al., [@B183]) and arrhythmia (Zhang et al., [@B175]). The active effects may be attributed to promotion of cardiac differentiation (Cheng et al., [@B23]; Wang L. et al., [@B139]). PUE upregulates expression of caveolin-3, amphiphysin-2 and junctophinlin-2, and then ameliorates myofibril array and sarcomeres formation, accompanied by increased t-tubules development in the embryonic stem cell-derived cardiomyocytes. Moreover, PUE suppresses the upstream regulatory factor of caveolin-3, namely miR-22, indicating miR-22/caveolin-3 axis may be the underlying mechanism of cardiomyogenesis induced by PUE (Wang L. et al., [@B139]). Conclusion {#s11} ========== CHM has long been a powerful weapon used by Chinese people to combat disease. Over thousands of years, practitioners of CHM have accumulated a wealth of knowledge which is used to prevent and cure diseases. The theoretical concepts of TCM act as the basis for scientific research into CHM today, including identification of the bioactive ingredients and underlying mechanisms of CHMs that could be of benefit internationally---a gift from the Chinese people to the world (Tu, [@B132], [@B133]). Chinese pharmacologist, Youyou Tu, and her team discovered the highly effective and low-toxicity bioactive ingredient "artemisinin" from *Artemisia annua* L. (also referred to as "qinghao" in Chinese), inspired by, "*A Handbook of Prescriptions for Emergencies"* (written around 317--420 CE), thus making an outstanding contributions to the global treatment of malaria (Tu, [@B133]). Consequently, we became convinced that study of the bioactive ingredients of CHMs is an effective way to reveal their potential mechanisms of action and further broaden their clinical application. By conducting the above comprehensive review, we found that bioactive ingredients of CHMs can play positive roles in treatment of cancer, cardiovascular, nervous system, respiratory, digestive, infectious, and senescence-related diseases. Through targeting various miRNAs, lncRNAs, circRNAs, or ceRNA crosstalk, these ingredients exert protective effects, including pro-apoptosis, anti-proliferation and anti-migration, anti-inflammation, anti-atherosclerosis, anti-infection, anti-senescence, and anti-structural remodeling. Some miRNAs, including miR-21, miR-34a, miR-34c, miR-155, miR-29a, miR-203, miR-27b, miR-184, and miR-143, contributed to the treatment mechanisms of more than one bioactive ingredient of CHMs. In particular, miR-21 was identified as targeted and regulated by BBR (Luo et al., [@B89]), TP (Li et al., [@B74]), RES (Wang G. et al., [@B136]), CA (Qu et al., [@B107]), ICA (Li J. et al., [@B70]), AIL (Yang P. et al., [@B160]), Car/Thy (Khosravi and Erle, [@B61]), DHM (Yang D. et al., [@B158]), and COR (Yang et al., [@B157]), especially in its anti-cancer activities, indicating that this miRNA is stably targetable and responsive to the pharmacological effects of various CHMs. Moreover, miR-155 was associated with inflammatory responses and could be inhibited by CUR, RES, Tan IIA, CA, Car/Thy and AKBA, in inflammatory-related diseases (Tili et al., [@B130]; Fan et al., [@B28]; Khosravi and Erle, [@B61]; Ma F. et al., [@B91]; Xuan et al., [@B156]; Qu et al., [@B107]; Sayed et al., [@B114]). Thus, it is highly likely that miR-155 could represent a new treatment target for anti-inflammation. In addition, three complex ceRNA crosstalk networks were discovered to function in the therapeutic mechanisms of ART, Sch B, and CA. Specifically, ART regulates the lncRNA UCA1/miR-184/BCL-2 axis, to inhibit prostate cancer (Zhou S. et al., [@B194]), while the has-circ-0043256/miR-1252/ITCH axis was involved in the treatment of non-small cell lung cancer by CA (Tian F. et al., [@B128]). The miR-150/ lncRNA BCYRN1 axis was targeted by Sch B treatment, leading to suppression of cell proliferation in asthma (Zhang X. Y. et al., [@B185]). All of these complex networks provide foundations for in-depth understanding and broader application of CHMs in the near future. The interactions between bioactive ingredients of CHMs with ncRNA targets are the subject of intensive and rapidly expanding research. This has helped to reveal the treatment mechanisms underlying the activities of CHMs and offers promising complementary and alternative treatments for diseases, based on scientific research. Although some previous reviews have revealed the increasing importance of bioactive ingredients (Huang et al., [@B50]; Lelli et al., [@B67]) or CHMs (Hong et al., [@B44]) in the treatment of diseases by targeting ncRNA. However, most of them mainly focused on one kind of ncRNA (particularly miRNA), or are limited to a specific disease (mostly cancer) (Mohammadi et al., [@B101]; Mirzaei et al., [@B98]); rather than overall and comprehensive ncRNA targets, ceRNA crosstalk and corresponding mechanisms. As a result, we consider it\'s necessary to make a systematic review about the treatment mechanisms of bioactive ingredients from CHMs by targeting miRNA, lncRNA, and circRNA. From our review, it can be seen that studies are currently in initial and exploratory phases, and several critical problems remain. First, various individual ncRNA molecules are targets of CHM bioactive ingredients; however, recent results are far from sufficient to allow understanding of the complex regulatory interactions between circRNA, lncRNA, miRNA, and mRNA in the treatment of diseases. Second, the metabolism of drugs in single cell lines and animals may differ from that in the human body; therefore, results based on basic research require further verification in clinical trials. Third, each CHM generally contains numerous ingredients and a TCM clinical prescription often consists of several CHMs; therefore, multiple targets and ceRNA crosstalk must occur and the study of classic TCM formulae will further complicate the picture. In conclusion, ncRNAs are potential targets of CHMs and understanding of ceRNA crosstalk has helped to reveal the complex mechanisms underlying multi-target and multi-level regulation of bioactive ingredients from CHMs. Therefore, CHM ingredients represent new and promising choices for future alternative disease treatments. Author Contributions {#s12} ==================== JW and JL designed the study. YD, HC, and JG conducted searches and extracted the data. YL and YD analyzed the data. YD wrote the manuscript. Conflict of Interest Statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This study was supported by the National Natural Science Foundation of China (No. 81673847; No. 81473561). ###### Chem. 1--43 Chemical formulae of bioactive ingredients from CHMs. ------------------------------------- -------------------------------------------- \ \ *Chem. 1 Berberine* *Chem. 2 Artesunate* \ \ *Chem. 3 Triptolide* *Chem. 4 Triptonide* \ \ *Chem. 5 Ailanthone* *Chem. 6 Cordycepin* \ \ *Chem. 7 Soya-cerebroside* *Chem. 8 Tubeimoside I* \ \ Chem. 9 Oridonin Chem. 10 Curcumin \ \ Chem. 11 Shikonin Chem. 12 Paeoniflorin \ \ Chem. 13 Honokiol Chem. 14 Schiscandrin B \ \ Chem. 15 Resveratrol Chem. 16 Soybean isoflavones \ \ Chem. 17 Matrine Chem. 18 Corylin \ \ Chem. 19 Tanshinones UA Chem. 20 Magnesium lithospermate B \ \ Chem. 21 Baicalin Chem. 22 Cinnamaldehyde \ \ Chem. 23 Geniposide Chem. 24 Carvacrol \ \ Chem. 25 Thymol Chem. 26 3-acetyl-11-keto-a-boswellic acid \ \ Chem. 27 Sinapic acid Chem. 28 Polydatin \ \ Chem. 29 Ampelopsin Chem. 30 Icariine \ \ Chem. 31 Ginsenosides Chem. 32 Salidroside \ \ Chem. 33 Phlorizin Chem. 34 Osthole \ \ Chem. 35 Panax Notoginseng Saponins Chem. 36 Tetrandrine \ \ Chem. 37 Leonurine Chem. 38 Calycosin \ \ Chem. 39 Astragaloside IV Chem. 40 Paeonol \ \ Chem. 41 Salvianolic acid A Chem. 42 Andrographolide \ Chem. 43 Puerarin ------------------------------------- -------------------------------------------- [^1]: Edited by: Chandravanu Dash, Meharry Medical College, United States [^2]: Reviewed by: Zijian Zhang, Texas Tech University Health Sciences Center El Paso, United States; Yong Xu, First Hospital of Shanxi Medical University, China [^3]: This article was submitted to Pharmacogenetics and Pharmacogenomics, a section of the journal Frontiers in Pharmacology [^4]: †These authors have contributed equally to this work

Takeuchi H, Savitzky AH, Ding L, et al. Evolution of nuchal glands, unusual defensive organs of Asian natricine snakes (Serpentes: Colubridae), inferred from a molecular phylogeny. Ecol Evol. 2018;8:10219--10232. 10.1002/ece3.4497 1. INTRODUCTION {#ece34497-sec-0001} =============== In the 20th Century, many biologists were focused on commonalities among taxa, as represented by studies using model organisms (Alberts et al., [2008](#ece34497-bib-0001){ref-type="ref"}). On the other hand, appreciating the diversity of life and its evolutionary origins has been another essential pursuit in biology (Rosenzweig, [1995](#ece34497-bib-0037){ref-type="ref"}; Whittaker, [1972](#ece34497-bib-0051){ref-type="ref"}). Because evolution of novel phenotypic characters, such as wings of birds and mammary glands of mammals, can facilitate the diversification of a lineage (Wagner & Lynch, [2010](#ece34497-bib-0049){ref-type="ref"}), investigation of the evolutionary history of such novel characters can provide basic information that clarifies the processes underlying species diversification. Snakes (Serpentes) comprise a distinct monophyletic taxon within the Squamata (Pyron, Burbrink, & Wiens, [2013](#ece34497-bib-0033){ref-type="ref"}), including over 3,500 species that are distributed on all continents except Antarctica (Wallach, Williams, & Boundy, [2014](#ece34497-bib-0050){ref-type="ref"}). In spite of their seemingly uniform appearance, snakes exhibit prominent morphological and ecological diversity (Greene, [1997](#ece34497-bib-0011){ref-type="ref"}; Lillywhite, [2014](#ece34497-bib-0021){ref-type="ref"}) and have often evolved novel organs that serve particular ecological functions. A well‐known example of a novel defensive structure is the rattle of rattlesnakes, which is used to warn potential predators of the snakes' venomous bite (Greene, [1997](#ece34497-bib-0011){ref-type="ref"}). The rattle evolved once in the ancestor of extant rattlesnakes (Castoe & Parkinson, [2006](#ece34497-bib-0005){ref-type="ref"}; Greene, [1997](#ece34497-bib-0011){ref-type="ref"}), and it has been lost secondarily in some island populations, where selection for defense is reduced in the absence of mammalian predators (Martins, Arnaud, & Murillo‐Quero, [2008](#ece34497-bib-0024){ref-type="ref"}; Rowe, Farrell, & May, [2002](#ece34497-bib-0038){ref-type="ref"}). The nuchal gland system is another example of a novel defensive structure that has evolved in snakes (Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}). Nuchal glands were originally described in a Japanese natricine snake, *Rhabdophis tigrinus* (Figure [1](#ece34497-fig-0001){ref-type="fig"}; Nakamura, [1935](#ece34497-bib-0031){ref-type="ref"}). The organs, which superficially resemble secretory structures, are embedded in the dermal layer of the dorsal skin of the neck. The nuchal glands of *R. tigrinus* contain cardiotonic steroid toxins known as bufadienolides (Hutchinson et al., [2007](#ece34497-bib-0019){ref-type="ref"}), which are sequestered from toads consumed as prey and can be redeployed as a defensive mechanism (Hutchinson et al., [2007](#ece34497-bib-0019){ref-type="ref"}). The glands of some other species also contain bufadienolides (Mori et al., unpublished). Ontogenetically, the nuchal glands are of mesodermal origin (Fukada, [1958](#ece34497-bib-0010){ref-type="ref"}; Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}), which is different from any other skin glands of terrestrial vertebrates, all of which arise from ectoderm (Savitzky et al., [2012](#ece34497-bib-0039){ref-type="ref"}). The glands lack a secretory epithelium and consist of a homogeneous population of fluid‐filled cells surrounding a dense aggregation of capillaries. There is no central lumen or duct, and the glands simply rupture through the skin to expel their fluid contents when the snake is under predatory attack (Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}). {#ece34497-fig-0001} Nuchal glands and the structurally similar nucho‐dorsal glands (which extend the full length of the body; Smith, [1938](#ece34497-bib-0041){ref-type="ref"}) are currently known in 17 species of Asian Natricinae (Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}; Mori, Jono, Ding, et al., [2016](#ece34497-bib-0028){ref-type="ref"}). Hereafter, we refer to all such structures as nuchal glands, for simplicity. No other animals have been reported to possess organs similar in their structural details to the nuchal glands. The 17 species that possess such glands belong to three nominal genera, *Balanophis*,*Macropisthodon*, and *Rhabdophis*. Interestingly, *Macropisthodon* and *Rhabdophis* also include species that do not have nuchal glands (Table [1](#ece34497-tbl-0001){ref-type="table"}). This distribution might indicate the occurrence of (a) multiple independent origins of these unusual organs, (b) their secondary loss, and/or (c) improper generic assignment of some species. ###### A species list for the three nominal genera, *Balanophis*,*Macropisthodon*, and *Rhabdophis* Species Glands Source -------------------------------- -------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ***Balanophis ceylonensis*** P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***Macropisthodon flaviceps*** A/P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***M. plumbicolor*** P Mori, Jono, Takeuchi, Ding et al. ([2016](#ece34497-bib-0030){ref-type="ref"}) and Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) *M. rhodomelas* P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***M. rudis*** A Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) and Takeuchi and Mori ([2012](#ece34497-bib-0045){ref-type="ref"}) ***Rhabdophis adleri*** P Mori, Jono, Ding et al. ([2016](#ece34497-bib-0028){ref-type="ref"}) *R. akraios* U Doria, Petri, Bellati, Tiso and Pistarino ([2013](#ece34497-bib-0006){ref-type="ref"}) *R. angelii* U Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) *R. auriculatus* U Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) *R. barbouri* U Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) ***R. callichromus*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) and Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) *R. chrysargoides* U Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) ***R. chrysargos*** A Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. conspicillatus*** A Mori, Jono, Takeuchi and Das ([2016](#ece34497-bib-0029){ref-type="ref"}) ***R. formosanus*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) and Takeuchi, Ota, Oh and Hikida ([2012](#ece34497-bib-0046){ref-type="ref"}) ***R. guandongensis*** U Zhu, Wang, Takeuchi and Zhao ([2014](#ece34497-bib-0053){ref-type="ref"}) ***R. himalayanus*** P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. lateralis*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) and Takeuchi et al. ([2012](#ece34497-bib-0046){ref-type="ref"}) ***R. leonardi*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) *R. lineatus* U Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) ***R. murudensis*** A/P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}), Smith ([1938](#ece34497-bib-0041){ref-type="ref"}), and Steubing and Lian ([2002](#ece34497-bib-0044){ref-type="ref"}) ***R. nigrocinctus*** P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. nuchalis*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}), Mori, Jono, Takeuchi, Ding et al. ([2016](#ece34497-bib-0030){ref-type="ref"}), and Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. pentasupralabialis*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) and Mori, Jono, Takeuchi, Ding et al. ([2016](#ece34497-bib-0030){ref-type="ref"}) *R. spilogaster* A Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. subminiatus*** P Smith ([1938](#ece34497-bib-0041){ref-type="ref"}) ***R. swinhonis*** A/P Mao and Chang ([1999](#ece34497-bib-0023){ref-type="ref"}) and Hsiang, Li and Yang (2009) ***R. tigrinus*** P Mori et al. ([2012](#ece34497-bib-0027){ref-type="ref"}) and Nakamura ([1935](#ece34497-bib-0031){ref-type="ref"}) Species included in the analyses of this study are shown by bold. P, A, and U indicate present, absent, and unknown, respectively. Our study strongly suggests that *Balanophis* and *Macropisthodon,* except *M. rudis,* belong to *Rhabdophis*. John Wiley & Sons, Ltd To infer the evolutionary history of the nuchal glands, we investigated the molecular phylogenetic relationships among Eurasian natricine species, including all but one of the species that have hitherto been reported to possess such glands (Table [1](#ece34497-tbl-0001){ref-type="table"}). Our phylogeny is based on partial sequences of the oocyte maturation factor Mos (Cmos) gene, the recombination‐activating gene 1 (RAG1), and the mitochondrial cytochrome b (MT‐CYB) gene, for a total of 2.7 kbp. Several recent phylogenetic studies of snakes have either focused on or included a number of Asian natricine species (Figueroa, Mckelvy, Grismer, Bell, & Lailvaux, [2016](#ece34497-bib-0009){ref-type="ref"}; Guo et al., [2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"}; Pyron, Kandambi et al., [2013](#ece34497-bib-0034){ref-type="ref"}). However, no previous study has addressed the evolution of the nuchal glands. Furthermore, our sampling of species and populations of *Macropisthodon* and *Rhabdophis* is much greater than that of previous studies. Specifically, our main purpose was to answer three questions: (a) Have the nuchal glands originated only once, or have they arisen multiple times independently among natricine snakes? (b) Do the species of *Macropisthodon* and *Rhabdophis* that lack such glands represent the secondary loss of those structures? (c) Are any of the species lacking nuchal glands incorrectly assigned to *Macropisthodon* or *Rhabdophis*? 2. MATERIALS AND METHODS {#ece34497-sec-0002} ======================== A total of 122 sequences of natricine snakes and three sequences of outgroup taxa were used for phylogenetic analyses (Appendix [1](#ece34497-app-0001){ref-type="app"}). Of those, 54 sequences were obtained from GenBank. Because our preliminary analysis suggested that the sequence data for *Rhabdophis adleri* registered in GenBank were incorrectly identified, we did not use the GenBank data for that species. The other 68 sequences were obtained by the following methods. In each sample, total DNA was extracted from liver, skeletal muscle, or tail tips, which had been preserved in 99.5% ethanol or in freezers, using the DNeasy Tissue Kit (Qiagen). The Cmos, RAG1, and MT‐CYB regions were amplified with a PCR System GeneAmp 2700 Thermal Cycler (Applied Biosystems), using an Ex Taq Polymerase Kit (Takara Bio Inc.) and primers S77/S78 for Cmos (Lawson, Slowinski, Crother, & Burbrink, [2005](#ece34497-bib-0020){ref-type="ref"}), R13/R18 for RAG1 (Groth & Barrowchlough, [1999](#ece34497-bib-0012){ref-type="ref"}), and L14910/H16064 for MT‐CYB (Burbrink, Lawson, & Slowinski, [2000](#ece34497-bib-0004){ref-type="ref"}). The thermocycling schedule for the polymerase chain reaction (PCR) was identical to that described by these previous studies. Before sequencing, unincorporated primers were removed from the PCR products using polyethylene glycol precipitation. Cycle sequencing reactions were performed with the Big Dye Terminator Cycle Sequence Ready Reaction Kit, version 3.1 (Applied Biosystems), using the same primers as for PCR. The samples purified by ethanol precipitation were sequenced with a 3130xl Genetic Analyzer (Applied Biosystems). All fragments were sequenced for both forward and reverse sense. We assembled them using the GAP 4 program (Staden, [1996](#ece34497-bib-0042){ref-type="ref"}). Using CLUSTAL X (Thompson, Gibson, Plewniak, Jeanmougin, & Higgins, [1997](#ece34497-bib-0048){ref-type="ref"}), 125 sequences were aligned. Identical sequences from different specimens were treated as single units so that 114 sequences were recognized. To infer the phylogeny, we employed Maximum Likelihood (ML) using combined sequences (Cmos + RAG1 + MT‐CYB) and Bayesian inference (BI) using the sequence of mitochondrial DNA (MT‐CYB). For both data sets, the most appropriate pattern of sequence evolution was selected by applying the Bayesian Information Criterion (BIC; Schwarz, [1978](#ece34497-bib-0040){ref-type="ref"}), using MEGA5 (Tamura et al., [2011](#ece34497-bib-0047){ref-type="ref"}). We set the rate categories of discrete gamma rate heterogeneity as eight for ML and BI. Reliability of the ML tree was assessed by calculating bootstrap probability (BP; Felsenstein, [1985](#ece34497-bib-0008){ref-type="ref"}), with 1,000 replications. The BI tree was constructed using BEAST version 1.8 (Drummond & Rambaut, [2007](#ece34497-bib-0007){ref-type="ref"}), employing a single Markov chain Monte Carlo (MCMC) run for 50 million generations, sampled every 1,000 generations, and excluding the first 5 million generations as burn‐in. Convergence of the chains to the stationary distribution was checked by visual inspection, using TRACER version 1.6 (Rambaut, Suchard, Xie, & Drummond, [2007](#ece34497-bib-0036){ref-type="ref"}). To estimate divergence times, we employed Bayesian relaxed‐clock dating, using BEAST version 1.8. Because no fossils of *Balanophis*,*Macropisthodon*, or *Rhabdophis* are known, we set the following calibration points: 30 Mya (*SD* = 0.115) at the crown of natricine snakes, 22 Mya (*SD* = 0.15) at the crown of the genus *Natrix*, and 16 Mya (*SD* = 0.15) at the crown of the genus *Thamnophis* (Guo et al., [2012](#ece34497-bib-0016){ref-type="ref"}). 3. RESULTS {#ece34497-sec-0003} ========== The final alignment of three gene fragments consisted of 2,767 aligned base pairs. Of those, 787--1,149 bp were from MT‐CYB (114 taxa), 259--689 bp were from Cmos (86 taxa), and 855--929 bp were from RAG1 (21 taxa). The most appropriate model under the BIC was the GTR + G + I model for the data sets of both the ML and BI trees. The ML and BI trees were almost identical in topology. The ML tree (−In *L* = −35078.3994) is shown in Figure [2](#ece34497-fig-0002){ref-type="fig"}. A consensus tree from the ML and BI analyses is shown in Figure [3](#ece34497-fig-0003){ref-type="fig"}, along with the BP values from ML and the posterior probability (PP) value from BI at each node (shown only for BP ≥ 70% in ML and PP ≥ 0.90 in BI). The main difference between the ML and BI trees is the status of *Rhabdophis chrysargos*. Unlike the ML tree, the Bl tree supported monophyly of *R. chrysargos* + *R. conspicillatus* + 3 species of *Xenochrophis* (Figure [3](#ece34497-fig-0003){ref-type="fig"}a). {ref-type="app"}). Status of nuchal or nucho‐dorsal glands of our three focal genera (*Rhabdophis, Macropisthodon, and Balanophis*) is indicated by blue (present), red (absent), purple (present/absent), and green (unknown; see also Table [1](#ece34497-tbl-0001){ref-type="table"}). The photographs have been digitally modified for clarity. Photograph of *Balanophis ceylonensis* by Udaya Chanaka](ECE3-8-10219-g002){#ece34497-fig-0002} {ref-type="table"})](ECE3-8-10219-g003){#ece34497-fig-0003} Monophyly of Natricinae was strongly supported by the PP value. Within this subfamily, monophyly of the New World taxa (the Thamnophiini), and the Old World taxa *Natrix*,*Sinonatrix*,*Hebius*, and *Amphiesma *+* Xenochrophis *+* Atretium *+* Rhabdophis* + *Macropisth odon* (except *M. rudis*) + *Balanophis* clades were highly supported. Of the latter clade, a subclade of *Rhabdophis* (except *R. chrysargos* and *R. conspicillatus*) + *Macropisthodon* (except *M. rudis*) + *Balanophis* was separated from the remainder with strong support (Figure [2](#ece34497-fig-0002){ref-type="fig"}b). The average estimated divergence time of this subclade was 19.18 Mya (16.28--22.16 in 95% credible ranges). Hereafter, we refer to this subclade as the nuchal gland clade (NGC). Within this clade, *Macropisthodon plumbicolor* first diverged from the other species. The latter include *Rhabdophis subminiatus, R. murudensis* + *Macropisthodon flaviceps*,*R. himalayanus *+* Balanophis ceylonensis, R. tigrinus* + *R. lateralis* + *R. formosanus*, and a large group including *R. adleri* + *R. callichromus + R. nigrocinctus + R. swinhonis* + *R. guangdongensis* + *R. nuchalis* + *R. leonardi* + *R. pentasupralabialis* (with \>90% support in BP and/or 0.9 in PP). The latter clade comprises two subclades: *R. adleri* + *R. callichromus + R. nigrocinctus* and *R. swinhonis* + *R. guangdongensis* + *R. nuchalis* + *R. leonardi* + *R. pentasupralabialis*. Several nominal species exhibit substantial population structuring. *Rhabdophis subminiatus* exhibits strong differentiation between Laos/Vietnam and Thailand samples, and *R. nuchalis* consists of a number of population segments and is paraphyletic with respect to both *R. leonardi* and *R. pentasupralabialis*. 4. DISCUSSION {#ece34497-sec-0004} ============= Although differing in some details, recent molecular phylogenetic analyses of the Natricinae (Figueroa et al., [2016](#ece34497-bib-0009){ref-type="ref"}; Guo et al., [2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"}; Pyron, Burbrink et al., [2013](#ece34497-bib-0033){ref-type="ref"}; Pyron, Kandambi et al., [2013](#ece34497-bib-0034){ref-type="ref"}), including ours, agree on the general pattern of relationships among the major lineages. A basal dichotomy separates the subfamily into two major clades. One includes the entire North American natricine fauna (the Thamnophiini) and its sister group, the Eurasian genus *Natrix*. Those two, in turn, are sister to a clade containing the Asian genera *Opisthotropis* and *Sinonatrix*. A clade containing two endemic Sri Lankan genera, *Aspidura* and *Haplocerus*, is variously recovered as sister to this North American--Eurasian clade (Pyron, Burbrink et al., [2013](#ece34497-bib-0033){ref-type="ref"}; Pyron, Kandambi et al., [2013](#ece34497-bib-0034){ref-type="ref"}) or as the most basal branch of the natricine clade (our study, but with weak support). The other major clade of natricines is almost entirely Asian, the sole exception being a monophyletic group of three African genera (*Afronatrix, Natriciteres,* and *Lycognathophis*, the latter not included in our analysis). The African clade is variously recovered as sister to, or embedded within, the much larger Asian radiation. The relationships among the Asian taxa display varying topologies among recent analyses, as taxon sampling within this group has improved. Consistent with other recent studies (Guo et al., [2014](#ece34497-bib-0017){ref-type="ref"}), we recover a monophyletic genus *Hebius*, distant from *Amphiesma stolatum*, as well as a polyphyletic *Xenochrophis*, some related to *Atretium* and others close to *Rhabdophis* and *Macropisthodon*. These results engender confidence in our analysis of the relationships within the NGC. 4.1. Evolution of the nuchal glands {#ece34497-sec-0005} ----------------------------------- Our results show that all species that possess nuchal glands belong to a single, strongly supported clade (NGC). Therefore, based on the principle of parsimony, we infer that the common ancestor of this clade possessed nuchal glands. We find no evidence of multiple, independent origins of the glands. Thus, interspecific differences in the distribution and morphology of the glands, such as the occurrence of nucho‐dorsal glands along the entire length of the body in *M. plumbicolor* and several species of *Rhabdophis* (Mori, Jono, Ding et al., [2016](#ece34497-bib-0028){ref-type="ref"}; Mori, Jono, Takeuchi, & Das, [2016](#ece34497-bib-0029){ref-type="ref"}; Smith, [1938](#ece34497-bib-0041){ref-type="ref"}) and the presence of elongate, nonsacculated glands accompanied by scaleless areas of skin in *M. rhodomelas* (not included in our analysis), *M. flaviceps*, and *B. ceylonensis* (Smith, [1938](#ece34497-bib-0041){ref-type="ref"}), are considered to represent alternative morphologies that arose after a single evolutionary origin of the nuchal gland system. Further study of the morphological details is needed to clarify the process of glandular diversification within this clade. Among species currently included in *Rhabdophis* and *Macropisthodon*,*R. chrysargos*,*R. conspicillatus*, and *M. rudis* have been reported to lack nuchal glands (Table [1](#ece34497-tbl-0001){ref-type="table"}; Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}; Mori, Jono, Takeuchi, & Das, [2016](#ece34497-bib-0029){ref-type="ref"}). *Macropisthodon rudis* is only distantly related to the NGC (see below), and *R. conspicillatus* and *R. chrysargos* also belong to clades outside the NGC. Thus, the absence of the nuchal glands in these species does not constitute secondary loss. Rather, it appears that they have simply retained the ancestral condition of the absence of integumentary defensive glands. *Rhabdophis swinhonis* has been reported to lack nuchal glands (Table [1](#ece34497-tbl-0001){ref-type="table"}; Mao & Chang, [1999](#ece34497-bib-0023){ref-type="ref"}). However, in contrast to *R. conspicill atus* and *R. chrysargos*, our analysis shows that this species occupies a position within the NGC. This strongly suggests that *R. swinhonis* has secondarily lost the nuchal glands. However, Hsiang, Li, and Yang ([2009](#ece34497-bib-0018){ref-type="ref"}) noted the presence of nuchal glands in this species. If both observations are correct, there are two possible interpretations: either the occurrence of intraspecific variation or the presence of two distinct but cryptic species. Whichever is true, the deeply nested position of *R. swinhonis* within the NGC implies the recent or ongoing secondary loss of the glands in at least some populations. Intraspecific variation in the presence of the nuchal glands also has been described in *R. murudensis* and *M. flaviceps* (Table [1](#ece34497-tbl-0001){ref-type="table"}; Smith, [1938](#ece34497-bib-0041){ref-type="ref"}; Mori et al., [2012](#ece34497-bib-0027){ref-type="ref"}). In our analysis, both species are recovered within the NGC. Therefore, as with *R. swinhonis*, the nuchal glands of *R. murudensis* and *M. flaviceps*, if accurately described in the literature, might be in a transitional stage of secondary loss or these nominal species may contain closely related cryptic species. We estimate that the common ancestor of the NGC arose 19.18 Mya. This is only slightly later than the date of 23--24 Mya shown by Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"}, Figure [2](#ece34497-fig-0002){ref-type="fig"}) for the origin of *Rhabdophis*, suggesting that nuchal glands arose at or soon after the origin of this genus. 4.2. Taxonomy {#ece34497-sec-0006} ------------- Our analysis requires a reevaluation of the taxonomic status of the genera *Balanophis* and *Macropisthodon*. The validity of the monotypic genus *Balanophis* (Smith, [1938](#ece34497-bib-0041){ref-type="ref"}) has been controversial. Malnate ([1960](#ece34497-bib-0022){ref-type="ref"}) recognized the species as *Rhabdophis ceylonensis*, and McDowell ([1961](#ece34497-bib-0025){ref-type="ref"}) supported his position. Figueroa et al. ([2016](#ece34497-bib-0009){ref-type="ref"}) found the species nested within *Rhabdophis*, as sister to *R. himalayanus*, and despite stating in the text (p. 21) that they declined to synonymize the genera, they recognized the species as *R. ceylonensis* in their figure 7a. Our analysis also strongly supports a sister relationship between *B. ceylonensis* and *R. himalayanus*, and thus, we formally propose that *Balanophis* be synonymized with *Rhabdophis*. Our analysis includes three of the four currently recognized species of *Macropisthodon* (Wallach et al., [2014](#ece34497-bib-0050){ref-type="ref"}), no two of which are recovered as each other\'s closest relative. When the genus was described by Boulenger ([1893](#ece34497-bib-0003){ref-type="ref"}), most other natricine snakes were treated as members of the genus *Tropidonotus*. Stejneger ([1907](#ece34497-bib-0043){ref-type="ref"}) placed *Tropidonotus* in the genus *Natrix*, where it remained until Malnate ([1960](#ece34497-bib-0022){ref-type="ref"}) divided *Natrix* sensu lato into six genera, resurrecting *Rhabdophis* Fitzinger, 1843. Malnate suggested that *Macropisthodon* might later prove not to be distinct from *Rhabdophis*, but the overreliance on characters of the maxillary dentition had precluded its earlier inclusion in *Natrix* and presumably influenced Malnate\'s decision to retain the genus. In our analysis, the type species of *Macropisthodon*,*M. flaviceps*, is strongly supported as sister to *R. murudensis*. Figueroa et al. ([2016](#ece34497-bib-0009){ref-type="ref"}) show the fourth species, *M. rhodomelas*, nested well within *Rhabdophis*. Therefore, we synonymize *Macropisthodon* with *Rhabdophis*. Thus, it is presently reasonable to include all species belonging to the NGC within *Rhabdophis*, the type species of which is *R. subminiatus*. However, partitioning of this morphologically diverse clade should be considered in the future. The divergent position of *Macropisthodon rudis*, which lacks nuchal glands and is recovered as distant from the NGC, supports the resurrection of the monotypic genus *Pseudoagkistrodon* (Van Denburgh 1909), as suggested by Wallach et al. ([2014](#ece34497-bib-0050){ref-type="ref"}). Although recent studies have differed in the exact placement of this species (Guo et al., [2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"}), no analysis with sufficient taxon sampling of Asian natricines has placed it close to *Rhabdophis*. The taxonomic status of "*R". conspicillatus* and *"R". chrysargos*, which lie just outside the NGC, remains to be determined. Our analysis suggests that *Rhabdophis* contains several undescribed species. Substantial genetic divergence occurs within *R. nigrocinctus, R. swinhonis, R. nuchalis*, and especially *R. subminiatus*. A comprehensive analysis of this complex subclade, including both morphological and molecular studies, will be necessary before this group can be reliably partitioned. 5. CONCLUSIONS {#ece34497-sec-0007} ============== Our analysis indicates that the nuchal and nucho‐dorsal glands, as a group, have evolved only once among Asian natricine snakes. The absence of the nuchal glands in some nominally congeneric species, such as *M. rudis, R. conspicillatus,* and *R. chrysargos*, reflects old classifications based on phenetic analysis of morphological characters. All of those species lie outside the single clade that possesses the defensive glands. However, a few species within the nuchal gland clade (*M. flaviceps, R. murudensis,* and *R. swinhonis*) may represent a transitional stage in the secondary loss of the glands. Clarification of the developmental origin of these unique organs is likely to provide insight into how these neomorphic structures have arisen, diversified, and may subsequently be disappearing in a few species. The nuchal glands are fruitful subjects for investigating the evolution of novel biological systems that involve the complex interplay of morphology, physiology, ecology, and behavior. AUTHOR CONTRIBUTIONS {#ece34497-sec-0009} ==================== Hirohiko Takeuchi designed and performed research, analyzed data, and wrote the paper. Alan H. Savitzky designed research and wrote the paper. Li Ding designed and performed research in China. Anslem de Silva performed research in Sri Lanka. Indraneil Das performed research in Malaysia. Tao Thien Nguyen performed research in Vietnam. Tein‐Shun Tsai performed research in Taiwan. Teppei Jono performed research in China and analyzed data in Japan. Guang‐Xiang Zhu performed research in China. Dharshani Mahaulpatha performed research in Sri Lanka. Yezhong Tang designed and performed research in China. Akira Mori designed and performed research and wrote the paper. DATA ACCESSIBILITY {#ece34497-sec-0010} ================== DDBJ accessions [LC325298](LC325298)--[LC325357](LC325357), [LC325746](LC325746)--[LC325803](LC325803), and [LC326011](LC326011)--[LC326031](LC326031) (DNA sequences). We thank K. Kurita and T. Okamono for further instruction in phylogenetic analyses and Y. Kojima for assistance with executing phylogenetic analyses. We are indebted to H. Ota and N. Kuraishi for providing tissue samples, D.‐E. Lin, C. Li, Y.‐W. Yeh, and C.‐A. Tu for assistance in collecting tissue samples*,* Q. Chen for field assistance, U. Chanaka for providing a photograph, and the Vice Chancellor and Dean, University of Sri Jayewardenepura for allowing the maintenance of live snakes and for providing laboratory facilities. We are grateful to the Council of Agriculture, Executive Yuan, Taiwan for permits to collect snakes in Taiwan and for the export permit to Kyoto University (Permit No. AGF4X104070012); and to the Department of Wildlife Conservation, Sri Lanka for permits to collect snakes in Sri Lanka and for the export permit to Kyoto University. This study was supported by grants from the Japan--China Joint Research Project (2014--2016) between the Japan Society for the Promotion of Science (JSPS) and National Natural Science Foundation of China (NSFC, 31411140033) and from JSPS Scientific Researches C (26440213) and B (17H03719). This research was partially supported by project grant DTDLXH.19/15 and BSTMV.08/16‐19 to NTT, by the Niche Research Grant Scheme from the Ministry of Higher Education, Government of Malaysia (NRGS/1087/2013(01); IA010200‐0708‐0007) to ID, by a grant from National Natural Science Foundation of China (NSFC, 31401959) to GSZ, and by grants from NSFC (Fund for Young International Scientists: 31450110454), Chinese Academy of Sciences Fellowship for Young International Scientists (Y5C3051100), and JSPS (16J06675) to TJ. This study was also supported in part by the Global COE Program A06 to Kyoto University from MEXT and by funding provided to AHS by Utah State University. 1.1. Accession numbers and their localities (countries) for all DNA sequence data used in the phylogenetic analyses in this study. Individuals with an asterisk indicate identical sequences within the species, and thus have the same accession number. Names (and No.) in the species column correspond to those shown in Figure [1](#ece34497-fig-0001){ref-type="fig"} {#ece34497-sec-0011} =========================================================================================================================================================================================================================================================================================================================================================================== SpeciesIndividual No.CountryAccession no. of GenBankReferencesCyt.bC‐mosRag‐1*Afronatrix anoscopus*ROM19842Liberia[AF420073](AF420073)[AF471123](AF471123)[EU402832](EU402832)Lawson et al., [2005](#ece34497-bib-0020){ref-type="ref"}, de Queiroz, Lawson, and Lemos‐Espinal [2002](#ece34497-bib-0035){ref-type="ref"}, and Wiens et al. ([2008](#ece34497-bib-0052){ref-type="ref"})*Amphiesma stolatum*\_1HT0548China[LC325319](LC325319)[LC325765](LC325765)--This study*Amphiesma stolatum*\_2HT0798Sri Lanka[LC325347](LC325347)[LC325793](LC325793)[LC326030](LC326030)This study*Amphiesma stolatum*\_3GP2213China[KJ685693](KJ685693)[KJ685643](KJ685643)[KJ685585](KJ685585)Guo et al. ([2014](#ece34497-bib-0017){ref-type="ref"})*Aspidura guentheri*RAP0437Sri Lanka[KC347472](KC347472)[KC347380](KC347380)[KC347418](KC347418)Pyron, Kandambi et al. ([2013](#ece34497-bib-0034){ref-type="ref"})*Atretium schistosum*\_1HT0799Sri Lanka[LC325348](LC325348)[LC325794](LC325794)--This study*Atretium schistosum*\_2--Sri Lanka[KC347487](KC347487)[KC347383](KC347383)[KC347421](KC347421)Pyron Kandambi et al. ([2013](#ece34497-bib-0034){ref-type="ref"})*Atretium yunnanensis*GP842China[JQ678448](JQ678448)[JQ281787](JQ281787)[KJ685602](KJ685602)Guo et al. ([2014](#ece34497-bib-0017){ref-type="ref"})*Balanophis ceylonensis*\_1HT0785Sri Lanka[LC325339](LC325339)[LC325785](LC325785)[LC326026](LC326026)This study*Balanophis ceylonensis*\*\_2HT0786Sri Lanka[LC325339](LC325339)----This study*Balanophis ceylonensis*\_3HT0787Sri Lanka[LC325340](LC325340)[LC325786](LC325786)--This study*Haplocercus ceylonensis*RS145Sri Lanka[KC347478](KC347478)[KC347401](KC347401)[KC347438](KC347438)Pyron, Kandambi et al. ([2013](#ece34497-bib-0034){ref-type="ref"})*Hebius atemporale*HT0550China[LC325320](LC325320)[LC325766](LC325766)--This study*Hebius craspedogaster*HT0801China[LC325350](LC325350)[LC325796](LC325796)--This study*Hebius ishigakiensis*HT0800Japan[LC325349](LC325349)[LC325795](LC325795)--This study*Hebius khasiense*HT0679Vietnam[LC325327](LC325327)[LC325773](LC325773)--This study*Hebius octolineatus*HT0586China[LC325321](LC325321)[LC325767](LC325767)--This study*Hebius pryeri*HT0340Japan[LC325312](LC325312)[LC325758](LC325758)--This study*Hebius vibakari*\_1HT0274Japan[LC325309](LC325309)[LC325755](LC325755)--This study*Hebius vibakari*\_2HT0277Japan[LC325310](LC325310)[LC325756](LC325756)--This study*Macropisthodon flaviceps*HT0809Malaysia[LC325355](LC325355)[LC325801](LC325801)--This study*Macropisthodon plumbicolor*\_1HT0782Sri Lanka[LC325336](LC325336)[LC325782](LC325782)[LC326025](LC326025)This study*Macropisthodon plumbicolor*\_2HT0783Sri Lanka[LC325337](LC325337)[LC325783](LC325783)--This study*Macropisthodon plumbicolor*\_3HT0784Sri Lanka[LC325338](LC325338)[LC325784](LC325784)--This study*Macropisthodon rudis*\_1HT0339China[LC325311](LC325311)[LC325757](LC325757)[LC326016](LC326016)This study*Macropisthodon rudis*\_2GP1266China[JQ687452](JQ687452)[JQ687434](JQ687434)[KJ685566](KJ685566)Guo et al. ([2014](#ece34497-bib-0017){ref-type="ref"})*Natriciteres olivacea*--Congo[AF471058](AF471058)[AF471146](AF471146)--Lawson et al. ([2005](#ece34497-bib-0020){ref-type="ref"})*Natrix maura*\_1--Spain[AY866530](AY866530)----Guicking, Lawson, Joger and Wink ([2006](#ece34497-bib-0015){ref-type="ref"})*Natrix maura*\_2--Tunisia[AY487682](AY487682)----Guicking, Joger and Wink ([2008](#ece34497-bib-0013){ref-type="ref"})*Natrix maura*\_3--Italy[AY487683](AY487683)----Guicking et al. ([2008](#ece34497-bib-0013){ref-type="ref"})*Natrix natrix*\_1--Spain[AY866536](AY866536)----Guicking et al. ([2006](#ece34497-bib-0015){ref-type="ref"})*Natrix natrix*\_2--France[AY866537](AY866537)----Guicking et al. ([2006](#ece34497-bib-0015){ref-type="ref"})*Natrix tessellata*\_1--Iran[AY487574](AY487574)----Guicking et al. ([2006](#ece34497-bib-0015){ref-type="ref"})*Natrix tessellata*\_2--Iran[AY487575](AY487575)----Guicking, Joger and Wink ([2009](#ece34497-bib-0014){ref-type="ref"})*Natrix tessellata*\_3--Bulgaria[AY866533](AY866533)----Guicking et al. ([2006](#ece34497-bib-0015){ref-type="ref"})*Nerodia cyclopion*--USA[AF402909](AF402909)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia erythrogaster*--USA[AF402912](AF402912)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia fasciata*--USA[AF402910](AF402910)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia floridana*--USA[AF402911](AF402911)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia rhombifer*--USA[AF402915](AF402915)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia sipedon*--USA[AF402913](AF402913)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Nerodia taxispilota*--USA[AF402914](AF402914)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Opisthotropis cheni*GP383China[GQ281779](GQ281779)[JQ687441](JQ687441)[KJ685595](KJ685595)Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"})*Opisthotropis guangxiensis*GP746China[GQ281776](GQ281776)[JQ687447](JQ687447)--Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"})*Opisthotropis lateralis*GP646China[GQ281782](GQ281782)[JQ687445](JQ687445)--Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"})*Opisthotropis latouchii*GP647China[GQ281783](GQ281783)[JQ687446](JQ687446)--Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"})*Opisthotropis typica*HT0794Malaysia[LC325343](LC325343)[LC325789](LC325789)[LC326028](LC326028)This study*Pseudoxenodon macrops* (Out group)\_1HT0646China[LC325323](LC325323)[LC325769](LC325769)--This study*Pseudoxenodon macrops* (Out group)\_2HT0802Malaysia[LC325351](LC325351)[LC325797](LC325797)--This study*Regina grahami*--USA[AF402918](AF402918)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Regina rigida*\_1--USA[AF402919](AF402919)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Regina rigida*\_2CAS:HERP:165994USA[AF471052](AF471052)[AF471120](AF471120)--Lawson et al. ([2005](#ece34497-bib-0020){ref-type="ref"})*Regina septemvittata*--USA[AF402917](AF402917)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Rhabdophis adleri*\_1HT0831China[LC325356](LC325356)[LC325802](LC325802)--This study*Rhabdophis adleri*\_2HT0832China[LC325357](LC325357)[LC325803](LC325803)--This study*Rhabdophis callichromus*\_1HT0654Vietnam[LC325324](LC325324)[LC325770](LC325770)--This study*Rhabdophis callichromus*\_2HT0674Vietnam[LC325325](LC325325)[LC325771](LC325771)[LC326020](LC326020)This study*Rhabdophis chrysargos*HT0342Malaysia[LC325313](LC325313)[LC325759](LC325759)[LC326017](LC326017)This study*Rhabdophis conspicilatus*HT0791Malaysia[LC325342](LC325342)[LC325788](LC325788)[LC326027](LC326027)This study*Rhabdophis formosanus*\_1HT0033Taiwan[LC325304](LC325304)[LC325750](LC325750)--This study*Rhabdophis formosanus*\*\_2HT0031Taiwan[LC325304](LC325304)----This study*Rhabdophis formosanus\**\_3HT0030Taiwan[LC325304](LC325304)----This study*Rhabdophis guangdongensis*SYSr000018China[KF800930](KF800930)[KF800920](KF800920)--Zhu et al. ([2014](#ece34497-bib-0053){ref-type="ref"})*Rhabdophis himalayanus*\_1HT0847China[LC325299](LC325299)[LC325746](LC325746)[LC326011](LC326011)This study*Rhabdophis himalayanus*\*\_2HT0848China[LC325299](LC325299)----This study*Rhabdophis himalayanus*\*\_3HT0849China[LC325299](LC325299)----This study*Rhabdophis himalayanus*\_4CAS224420Myanmar[KF800929](KF800929)[KF800919](KF800919)--Zhu et al. ([2014](#ece34497-bib-0053){ref-type="ref"})*Rhabdophis lateralis*\_1HT0855China[LC325302](LC325302)----This study*Rhabdophis lateralis*\_2GP613China[JQ687444](JQ687444)[GQ281785](GQ281785)[KJ685600](KJ685600)Guo et al. ([2014](#ece34497-bib-0017){ref-type="ref"})*Rhabdophis leonardi*\_1HT0851China[LC325300](LC325300)[LC325747](LC325747)[LC326012](LC326012)This study*Rhabdophis leonardi\*\_*2HT0852China[LC325300](LC325300)----This study*Rhabdophis leonardi\*\_*3HT0853China[LC325300](LC325300)----This study*Rhabdophis leonardi*\_4RDQ200905367China[KF800932](KF800932)[KF800922](KF800922)--Zhu et al. ([2014](#ece34497-bib-0053){ref-type="ref"})*Rhabdophis murudensis*HT0788Malaysia[LC325341](LC325341)[LC325787](LC325787)--This study*Rhabdophis nigrocinctus*\_1HT0253Thailand[LC325307](LC325307)[LC325753](LC325753)[LC326015](LC326015)This study*Rhabdophis nigrocinctus*\_2HT0343Thailand[LC325314](LC325314)[LC325760](LC325760)--This study*Rhabdophis nigrocinctus*\_3HT0845China[LC325298](LC325298)----This study*Rhabdophis nuchalis*\_1HT0701China[LC325333](LC325333)[LC325779](LC325779)[LC326022](LC326022)This study*Rhabdophis nuchalis*\_2HT0803China[LC325352](LC325352)[LC325798](LC325798)--This study*Rhabdophis nuchalis*\_3HT0807China[LC325353](LC325353)[LC325799](LC325799)[LC326031](LC326031)This study*Rhabdophis nuchalis*\_4HT0854China[LC325301](LC325301)[LC325748](LC325748)--This study*Rhabdophis nuchalis*\_5SICAU090001China[KF800925](KF800925)[KF800935](KF800935)--Zhu et al. ([2014](#ece34497-bib-0053){ref-type="ref"})*Rhabdophis pentasupralabialis*\_1HT0699China[LC325331](LC325331)[LC325777](LC325777)--This study*Rhabdophis pentasupralabialis*\_2HT0700China[LC325332](LC325332)[LC325778](LC325778)[LC326021](LC326021)This study*Rhabdophis pentasupralabialis*\_3HT0808China[LC325354](LC325354)[LC325800](LC325800)--This study*Rhabdophis subminiatus*\_1HT0267Laos[LC325308](LC325308)[LC325754](LC325754)--This study*Rhabdophis subminiatus*\_2HT0344Thailand[LC325315](LC325315)[LC325761](LC325761)--This study*Rhabdophis subminiatus*\_3HT0345Thailand[LC325316](LC325316)[LC325762](LC325762)--This study*Rhabdophis subminiatus*\_4HT0680Vietnam[LC325328](LC325328)[LC325774](LC325774)--This study*Rhabdophis swinhonis*\_1HT0021Taiwan[LC325303](LC325303)[LC325749](LC325749)--This study*Rhabdophis swinhonis*\_2HT0717Taiwan[LC325334](LC325334)[LC325780](LC325780)[LC326023](LC326023)This study*Rhabdophis swinhonis*\*\_3HT0716Taiwan[LC325334](LC325334)----This study*Rhabdophis swinhonis*\*\_4HT0718Taiwan[LC325334](LC325334)----This study*Rhabdophis swinhonis*\*\_5HT0719Taiwan[LC325334](LC325334)----This study*Rhabdophis tigrinus*\_1HT0098Japan[LC325305](LC325305)[LC325751](LC325751)[LC326013](LC326013)This study*Rhabdophis tigrinus*\_2HT0177Japan[LC325306](LC325306)[LC325752](LC325752)[LC326014](LC326014)This study*Sibynophis subpunctatus* (Out group)RAP0491Sri Lanka[KC347471](KC347471)[KC347411](KC347411)[KC347449](KC347449)Pyron, Kandambi et al. ([2013](#ece34497-bib-0034){ref-type="ref"})*Sinonatrix aequifasciata*\_1HT0678Vietnam[LC325326](LC325326)[LC325772](LC325772)--This study*Sinonatrix aequifasciata*\_2HT0681Vietnam[LC325329](LC325329)[LC325775](LC325775)--This study*Sinonatrix aequifasciata*\_3GP357China[JQ687430](JQ687430)[JQ687440](JQ687440)--Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"})*Sinonatrix annularis*GP889China[JQ687431](JQ687431)[JQ687449](JQ687449)[KJ685604](KJ685604)Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"})*Sinonatrix percarinata*GP956China[JQ687433](JQ687433)[JQ687451](JQ687451)[KJ685607](KJ685607)Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"})*Storeria dekayi*CAS:HERP:196039USA[AF471050](AF471050)[AF471154](AF471154)--Lawson et al. ([2005](#ece34497-bib-0020){ref-type="ref"})*Thamnophis butleri*--USA[AF402923](AF402923)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis couchii*--USA[AF402936](AF402936)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis cyrtopsis*--USA[AF402924](AF402924)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis elegans*--USA[AF402925](AF402925)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis godmani*--Mexico[AF420135](AF420135)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis marcianus*--USA[AF402926](AF402926)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis ordinoides*--USA[AF402927](AF402927)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis proximus*----[AF402928](AF402928)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis radix*--USA[AF402934](AF402934)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis sirtalis*\_1----[AF402929](AF402929)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Thamnophis sirtalis*\_2----[AF402930](AF402930)----Alfaro and Arnold ([2001](#ece34497-bib-0002){ref-type="ref"})*Trachischium monticola*GP1487China[JQ687428](JQ687428)[JQ687435](JQ687435)[KJ685570](KJ685570)Guo et al. ([2012](#ece34497-bib-0016){ref-type="ref"}, [2014](#ece34497-bib-0017){ref-type="ref"})*Xenochrophis asperrimus*\_1HT0797Sri Lanka[LC325346](LC325346)[LC325792](LC325792)--This study*Xenochrophis asperrimus*\_2--Sri Lanka[KC347480](KC347480)[KC347414](KC347414)[KC347451](KC347451)Pyron, Kandambi et al. ([2013](#ece34497-bib-0034){ref-type="ref"})*Xenochrophis flavipunctatus*HT0682Vietnam[LC325330](LC325330)[LC325776](LC325776)--This study*Xenochrophis maculatus*HT0720Malaysia[LC325335](LC325335)[LC325781](LC325781)[LC326024](LC326024)This study*Xenochrophis piscator*\_1HT0347Thailand[LC325317](LC325317)[LC325763](LC325763)[LC326018](LC326018)This study*Xenochrophis piscator*\_2HT0371Vietnam[LC325318](LC325318)[LC325764](LC325764)--This study*Xenochrophis piscator*\_3HT0796Sri Lanka[LC325345](LC325345)[LC325791](LC325791)--This study*Xenochrophis trianguligerus*HT0795Malaysia[LC325344](LC325344)[LC325790](LC325790)[LC326029](LC326029)This study*Xenochrophis vittatus*\_1HT0615Indonesia[LC325322](LC325322)[LC325768](LC325768)[LC326019](LC326019)This study*Xenochrophis vittatus\**\_2HT0527Indonesia[LC325322](LC325322)----This study

{.35} {.36} {.37} {.38} {.39} {.40}

1. Introduction {#sec1} =============== In cardiac anesthesia BIS monitoring is increasingly used to monitor anesthesia depth as well monitoring cerebral ischemia, which may be particularly important during cardiopulmonary bypass (CPB) \[[@B1], [@B2]\] and cardiopulmonary resuscitation \[[@B3], [@B4]\]. We describe a patient who, while being monitored with BIS, suffered a transient ischemic attack (TIA) of the brainstem. 2. Case Report {#sec2} ============== A 76-year-old man was scheduled for a three-vessel coronary artery grafting (CAG) using CPB. His medical history included hypertension, a minor inferior myocardial infarction, amaurosis fugax, surgical resection of a vocal cord carcinoma, and an eyelid correction. Physical examination showed a 70 kg, 1.69 m tall patient with a blood pressure of 170/70 mmHg and a pulse of 52. A bruit was heard over the heart, both carotid arteries and femoral arteries consistent with an aortic valve sclerosis. Cardiac ultrasound showed a hypokinetic inferior left ventricular wall, a good left and right ventricular function and an aortic valve sclerosis with minor aortic valve insufficiency. Coronary angiogram showed occlusion of the RCA, 90% occlusion of the LAD and 70% occlusion of the RCX. Patient took acetylsalicylic acid, chlortalidone, amlodipine, atorvastatin, metoprolol, isosorbide mononitrate, and isosorbide dinitrate. The patient was scheduled for a coronary bypass and premedicated with paracetamol 1000 mg and midazolam 7.5 mg, orally. Once in the operating theatre the patient was prepared for the operation with an 14 G IV infusion and a 20 G arterial cannula in the left radial artery as well as standard monitoring with a 5 lead ECG, pulsoximetry, and a noninvasive blood pressure band on the right arm. A BIS Quatro Sensor (XP) was placed and the monitor (Aspect Medical Systems, Inc. model A-2000 BIS Monitor) started. During this period the patient was alert and communicative. Without provocation, the patient suddenly complained of dizziness, stopped breathing, became unresponsive, and his eyes deviated upwards en laterally. The BIS monitor then gave its first reading of 60. We initially ventilated the patient by mask and gave naloxone to rule out an accidental sufentanil bolus. Subsequently the patient was intubated without medication. We requested an emergency neurological consultation. During this period the blood pressure and pulse remained stable at around 170/85 with a pulse of 55. Neurological examination showed a Glasgow Coma Score (GCS) of E-1, M-1, and V-T: eyes closed, no motor response to painful stimuli, and no sounds. He had pinpoint pupils and pupillary reactions showed minimal constriction to light. Corneal reflexes were present while oculocephalic reflexes were absent. Both eyes were deviated upwards and to the left. Tendon reflexes were increased on the left side, with bilateral extensor plantar responses. We performed a CT-scan, which showed central and cortical atrophy with subthalamic and periventricular hypodensities consistent with older vascular damage. The CT-angiogram showed generalized atherosclerotic changes in all brachiocephalic vessels, especially in the common and internal carotid arteries and a slight stenosis of the origin of the left vertebral artery. After the CT-scan, we restarted the BIS monitoring which still showed the BIS value at about 50 to 60. Eleven minutes after resuming BIS monitoring the patient opened his eyes, started breathing and responding to voice commands, and BIS values increased to around 80--85. The patient was extubated and wanted to know the result of the operation. On neurological examination, the patient was alert and responded adequately. There was a bilateral downbeat nystagmus in downgaze. Tendon reflexes remained increased on the left side, while both plantar responses were flexor. Based on the neurological examination, the transient nature of the neurological deficit, and the CT-scan, we concluded that the patient had suffered a transient ischemic attack of the brainstem. During the whole event which lasted 102 minutes the patient had remained hemodynamically stable. He was transferred to the neurological intensive care where the rest of his stay remained unremarkable, and he recovered without any neurological deficit. 3. Discussion {#sec3} ============= The primary reason to use an EEG monitor in anesthesia is to prevent awareness. Indeed, individuals suffering from this experience can have serious mental disorders. Awareness in a general population was found to be just below 0,2% and in a high-risk population just below 1% \[[@B5], [@B6]\]. Two large scale prospective trials (SAFE trial and B-Aware trial \[[@B5], [@B6]\]) demonstrated a reduction of 80% in the incidence of awareness using the BIS monitor. Furthermore, BIS and other EEG devices can be used to titrate anesthetics towards a desired level of hypnosis, with the aim to prevent exaggerated plasma concentration, which could lead to hemodynamic instability and prolonged awakening. There are some clinical conditions where BIS is unusually low. These are conditions were the cerebral functions are impaired by hypoperfusion, ischemia, hypoglycaemia, and hypothermia \[[@B7]\]. Patients with a neurological disease can present with unusual BIS values and anticonvulsant drugs may reduce BIS values. BIS may also detect microembolic injury \[[@B8]\]. In our case, BIS monitoring coincided with the clinical findings of reduced cerebral activity and later the return of consciousness. The BIS value is derived from two frontal leads. As such it will primarily monitor frontal lobe cortical electrical activity and indirectly frontal cerebral perfusion. As there was no change in blood pressure and pulse rate we think that the overall cerebral perfusion pressure was not reduced during the ischemic attack. Therefore, together with the diagnosis of brainstem TIA, we think that the BIS value is decreased by another mechanism then a perfusion disturbance. The cerebral cortex receives extensive afferent projections from brainstem nuclei. There is a long list of pathological brain conditions known to affect the EEG. Ischaemia is one of these conditions. Ischaemia of the lower brainstem (with a clinical picture characterized by coma, respiratory abnormalities, and pinpoint pupils) results in diffuse low-voltage activity and bilateral slowing of the EEG although a posterior alpha rhythm may be preserved. Hence a logical explanation for the significant decrease in the BIS signal observed in our patient may be the decrease in frontal cortical activity during ischemia of the brainstem. However, people using the BIS should be aware that the simplifying algorithm built in to the monitor limits its diagnostic possibilities and there are many factors, unknown to the user, which can alter the BIS value. The BIS monitoring did not lead to a change in treatment policy. Brainstem TIAs are supposedly a rare occurrence. The timing of our patients attack was quite spectacular: 20 minutes earlier and he would have been found asphyxiated in his bed and his death would most likely have been attributed to a coronary event. Thirty seconds later he would have been under anesthesia and subjected to hypothermia, hypotension, and full heparinization from the CPB. We can only speculate what the neurological outcome would have been. If the neurological outcome had been bad, it would have been labeled as a complication of the CPB. It begs the question how often the brainstem TIAs really happen in our cardio vascular-compromised population. 4. Summary {#sec4} ========== This 76-year-old patient suffered a brainstem TIA before cardiac surgery. The TIA was registered on BIS and resulted in a drop in BIS to a value of 60. When consciousness returned spontaneously, the BIS increased to 85. We believe that the lack of input from the brainstem to the frontal cortex resulted in the reduced cortical electrical activity as registered with the BIS. [^1]: Academic Editor: Michael G. Irwin

Introduction {#Sec1} ============ In manufacturing industries, industry 4.0 and digital transformation are interrelated fields that both motivate the development of digital twins. *Industry 4.0* is a concept attracting much research and development over the last decade, including reference models \[[@CR1]\], applications \[[@CR2]\], standards \[[@CR3]\] and supporting methods \[[@CR4]\]. A core idea of Industry 4.0 is to connect physical devices (e.g., manufacturing systems and the objects they produce), digital components (e.g. ERP or MES systems) and human actors along production processes for the sake of seamless integration and continuous monitoring and control \[[@CR5]\]. *Digital twins* (DT) support this core idea and can be defined as "a dynamic virtual representation of a physical object or system across its lifecycle, using real-time data to enable understanding, learning and reasoning" \[[@CR6]\]. *Digital transformation*, in general, denotes adopting digital technologies, such as industry 4.0 related technologies or DTs, in the digitalization of an organization's business model and its operations (cf. Sect. [2.3](#Sec5){ref-type="sec"}). Several researchers emphasize the importance of industry 4.0 for digital transformation \[[@CR21]\] or, vice versa, that digital transformation motivates the implementation of industry 4.0 \[[@CR20]\]. However, DTs as an element of digital transformation or digital transformation as driver for DT development are not included in the aforementioned work. A literature analysis (see Sect. [5](#Sec11){ref-type="sec"}) confirmed that digital twin research predominantly focuses on technological questions of DT design and operations. So far, organizational and business model related aspects of DTs are only sparsely covered in research which motivated this paper. In response to this, the paper's objective is *to investigate how DT solutions are integrated into organizational structures and business models of manufacturing enterprises, and what motivates the development of DT from a digital transformation perspective.* Enterprise Modeling (EM) is a versatile approach and is able to tackle various organizational design problems by means of multi-perspective conceptual modeling. EM captures organizational knowledge about the motivation and business requirements for designing IS \[[@CR7]\]. Hence it has the potential of capturing and representing the organizational motivation for DT design. A key aspect of operating and managing DTs is to configure and adjust them according to the situational changes in operations. Capability Management, and in particular Capability Driven Development (CDD), has been proven applicable for managing information systems (IS) in changing context \[[@CR10]\]. E.g., CDD supports generation of monitoring dashboards from models that include context elements, measurable properties, KPIs as well as rule-based based adjustments based on context data. In concrete terms, the goal of this paper is *to analyze the suitability of EM and capability management for the purpose of supporting the development and management of DTs from an organizational perspective.* We have chosen the 4EM and CDD methods for the purpose of this study because they have already established integration mechanisms between themselves and with other modeling languages. The rest of the paper is structured as follows. Section [2](#Sec2){ref-type="sec"} gives background to EM, CDD, and digital transformation. Section [3](#Sec6){ref-type="sec"} describes our research approach. Section [4](#Sec7){ref-type="sec"} presents two case studies. Section [5](#Sec11){ref-type="sec"} summarizes the main requirements for developing Industry 4.0 solutions found in literature. Section [6](#Sec12){ref-type="sec"} discusses the requirements from Sects. [4](#Sec7){ref-type="sec"} and [5](#Sec11){ref-type="sec"} with respect to CDD. Section [7](#Sec15){ref-type="sec"} discusses an example of a capability model for the purpose of DT development. Section [8](#Sec16){ref-type="sec"} provides concluding remarks. Background {#Sec2} ========== Enterprise Modeling and 4EM {#Sec3} --------------------------- EM is the process of creating an enterprise model that captures all the enterprise's aspects or perspectives that are required for a given modeling purpose. An enterprise model consists of a set of interlinked sub-models, each of them focusing on a specific perspective, like, processes, goals, concepts, actors, rules, IS components. 4EM \[[@CR7]\] is a representative of the Scandinavian strand of EM methods. At its core is participatory stakeholder involvement and the modeling process is usually organized in the form of facilitated workshops. 4EM shares many underlying principles of the, so called, multi-perspective, approaches that recommend analyzing organizational problems from various perspectives, e.g. AKM \[[@CR12]\] and MEMO \[[@CR11]\]. 4EM consists of six interconnected sub-model types for modeling a specific aspect or perspective of the enterprise -- Goals Model, Business Rules Model, Concepts Model, Business Process Model, Actors and Resources Model, as well as Technical Components and Requirements Model. 4EM also supports integration with other modeling languages and methods by allowing to define new inter-model relationships between the 4EM components and components of the modeling language to be integrated. Capability Driven Development {#Sec4} ----------------------------- In \[[@CR10]\] the concept of *capability thinking* and a method to capability management are introduced. It is an organizational mindset that puts capabilities in focus of the business model and IS development. Capability thinking emphasizes that capabilities are not self-emergent, instead they should be planned, implemented, controlled, and adjusted. In doing so they need to be addressed from the perspectives of (1) vision (e.g. goals and KPIs), (2) enterprise designs such as processes and IS architectures, (3) situation context incl. measurable properties, as well (4) best practices such as process variants and patterns for dealing with context changes. Capability as a concept allows reasoning about these four aspects of the business in an integrated way because enterprises need to know how to realize the business vision and designs as well as what needs to be changed depending on real-life situations. The definition of *capability is the ability and capacity that enables an enterprise to achieve a business goal in a certain context* \[[@CR10]\]. Successful implementation of capability thinking will lead to *capability management* as a systematic way to plan, design, develop, deploy, operate, and adjust capabilities. CDD is a method supporting the four perspectives of capability thinking. CDD consists of a number of method components each focusing on a specific task of the capability cycle, such as Capability Design, Context Modeling, Patterns and Variability Modeling, Capability Adjustment Algorithm Specification, as well as method extensions for dealing with certain business challenges such as supporting business process outsourcing and managing service configuration with the support of open data \[[@CR17]\]. Digital Transformation {#Sec5} ---------------------- In scientific literature, digital transformation often is discussed in the general context of digitalization and considered the most complex digitalization phase \[[@CR13]\]. Its focus is on the disruptive social and economic consequences which, due to the potential of digital technologies to substantially change markets, lead to new technological application potentials and the resulting changes in economic structures, qualification requirements for employees and working life in general. \[[@CR14]\] proposes to distinguish between transformation of the value proposition and the value creation when analysing and planning digital transformation. These two "dimensions" can be divided into different steps of digitalization which form the prerequisite for the next step. In \[[@CR15]\] we have proposed the steps for the dimensions of operations and product digitization. In the operations dimension, the steps are (1) replacing paper documents with digital representations, (2) end-to-end automated processing of this digital representation within a process and (3) integration of relevant processes within the enterprise and with partners. On the product dimension, the departure point for digitization are physical products without built-in information or communication technology. Digitization steps are (1) to enhance the product/service by providing complementary services (maintenance information, service catalogs) without actually changing it, (2) to extend functionality and value proposition of products by integration of sensors and actuators, and (3) redefinition of the product or service which leads to a completely new value proposition. A completed digital transformation requires all three steps in both dimensions. Research Approach {#Sec6} ================= This study is part of a research program aiming to provide methodological and tool support for organizations in dynamic contexts, e.g., supporting the process of digital transformation and capability management. It follows the five stages of Design Science research \[[@CR16]\], namely, problem explication, requirements definition, design and development of the design artifact, demonstration, as well as evaluation. This study concerns the first two steps for the design artifact supporting DT design and management from an organizational perspective. This part of our research started from the following research question which is based on the motivation presented in Sect. [1](#Sec1){ref-type="sec"}: *RQ: In the context of digital transformation, how are digital twin initiatives emerging and what are the driving forces for starting implementation projects?* The research method used for working on this research question is a combination of literature study and descriptive case study. Based on the research question, we identified industrial cases of digital transformation suitable for studying the origin of DT developments, i.e. we performed qualitative case studies in order to obtain relevant and original data (see Sect. [4](#Sec7){ref-type="sec"}). Qualitative case study is an approach to research that facilitates exploration of a phenomenon within its context using a variety of data sources. This ensures that the subject under consideration is explored from a variety of perspectives which allows for multiple facets of the phenomenon to be revealed and understood. Within the case studies, we used three different perspectives, which at the same time represent sources of data: we analyzed documents about business models, products, manufacturing process of the companies; we performed workshops targeting digital transformation and DTs as part thereof; and we interviewed domain experts. Yin \[[@CR18]\] differentiates case studies explanatory, exploratory and descriptive. The case studies in Sect. [4](#Sec7){ref-type="sec"} are considered descriptive, as they describe the phenomenon of initiating DT development and the real-life context in which it occurs. Based on the results of the case studies, primarily case study requirements to DT development, we selected research areas with relevant work for these requirements and analyzed the literature in these areas. The purpose of the analysis was to find existing approaches and methods for modeling DT and how they are integrated into the business. This work limits the focus on DTs in manufacturing, although they can also be used in other application fields. To summarize,The case studies explore whether business models and organizational context are really relevant from industrial perspective. We focus on the early phases of DT realization, i.e. decision making and specification,A literature study explores whether existing research work covers modeling approaches for business models and organizational context of DT. Industrial Case Studies {#Sec7} ======================= Case Study A: Producer of Pumps {#Sec8} ------------------------------- Company A is a medium-sized manufacturer of different kinds of pumps and pumping technologies, e.g. swimming pool pumps, sewage pumps, industrial pumps for heavy environments or ship pumps. Company A is well-established on the international market with a market share of more than 50% in some segments. Although its business is stable and developing well, the management decided to explore new service opportunities and business models applying digital technologies. More concretely, the idea of the company's product management is to integrate sensors into pumps and transmit the information to the back-office by using a datalink. This idea can be classified as converting the pumps into smart connected products or Internet-of-Things (IoT) devices. The opportunity for data collection at company A emerged when the it agreed to start a study on digital transformation options. The study so far had two workshops at the company's headquarter and several interviews. The first workshop was directed to the management with a focus on clarifying general steps of digital transformation, possible procedures and aspects of the enterprise to be considered. The second workshop was directed towards identifying concrete digital transformation options and potential ways of implementation. For our research question, the second workshop and the preparatory interviews were the most relevant and will be in focus of the analysis. One purpose of the preparatory interviews for the second workshop was to understand the current situation of IoT and sensor integration into the company's products. The key expert here was the research and development manager. Before the interview, guidelines consisting of a list of questions and aspects to explore were prepared. The interview took 30 min and was conducted by one researcher; notes were taken. As a preparation of the digital transformation workshop, the participants were selected to include all relevant departments of company A (product development, production, marketing, sales & distribution, and services) and members of top and middle management. All eight participants were informed in beforehand about the purpose of the workshop and importance of their participation. The workshop included the collection and clustering of new product and services ideas from the participants, joint definition of priorities, and development of a business model prototype for the top three product/service ideas. The workshop was documented in photo documentation of collected ideas and clusters, written documentation of the business model prototypes. Notes were taken to capture additional information regarding ideas and the business model. The product manager stated as one of the motivations for the workshop: "Our datalink device is nearly ready. It captures data and puts them into our own cloud. So far, we only capture data about malfunction or energy consumption that is anyhow visible on the pump's display. But we do not have a good idea, how to do business with this data. And we probably need more sensors." Among the top innovation ideas were (a) smart pumps and (b) pumping as a service, which the workshop participants both related to the topic of digital twins. When discussing the smart pump, the sales representative explained: "We think that our bigger customers want to have control if our pumps do what they are supposed to do in their installations. Some of them call it the digital twin. This would help us to sell pumps to them. We have to use or develop sensors that deliver this kind of information." Pumping as a service aims at selling the functionality of the pump instead of the pump as physical device which would lead to a service agreement where the company is paid for pumped cubic meters or hours of pumping. One of the participants remarked to this idea: "For this, we need full control what is happening with the pump. So, we need something a bit like a digital twin, but for our internal purposes." When developing the business model prototype for pumping as a service, most of the discussion time was spent on organizational issues within the company: "where does all the information from our pumps arrive, how do we make sense out of it and how do we organize the reaction?" For the smart pumps, the discussion was more about "how do we integrate our pumps in the DT system of our customer and what kind of sensors do we need?" Furthermore, the development department mentioned "We would need to know what technical basis our customers use for their DTs and what interfaces we have to provide. But most of our customers have no real answers to these questions. Sometimes we get the impression that they simply don't know." Case Study B: Tool Produces for Automotive Industry {#Sec9} --------------------------------------------------- Company B is a subsidiary of a major automotive manufacturer responsible for producing tools for the metal parts of chassis production, such as roofs, doors, side panels, etc. These tools, called (press) forms, are developed individually for each car model variant in an iterative process of casting, milling and/or welding, and polishing. Active components, such as cutters, hydraulic springs or punchers are integrated into the actual form. Putting the forms into operation in a press shop requires a try-out phase to fine-tune the forms' precision. Company B is doing the largest share of its business with the automotive manufacturer. It also serves other automotive and truck suppliers. Due to its unique specialization on forms for a specific metal, company B is well-positioned in the market. However, its management aims to increase efficiency and flexibility in the business model to be prepared for possible future market changes. This case study emerged when company B decided to investigate radical digital innovation focusing on disruptive ways of working or technologies instead of gradual optimization or increase in efficiency. A workshop was planned to investigate the potential for radical innovation concerning the possibilities for drastic and seemingly unrealistic changes, like, reduction of production time for forms to 10% of the current value, no setup time of the production system or internal logistics requiring no staff. Preparation and execution of the workshop was similar to what was described for the first case study: the selected participants represented all relevant departments of the company (design, production, logistics, procurement, human resources, economics, service and customer care), mostly represented by the head of the unit or senior experts. All ten participants were informed beforehand about purpose of the workshop, the need to think "out-of-the-box" and the importance of their participation. The workshop included the collection and clustering of radical transformation of products and of operations, joint clustering and definition of priorities. Based on the priorities, an initial evaluation of the top three options for radical transformation of products and the top three transformations in operations was done. The content of the workshop was documented in photo documentation of collected ideas and clusters, written documentation of the evaluation results, and notes. The workshop was conducted by two researchers: one facilitator and one note taker. In this paper analyzes the documented content. For radical transformation of internal operations, one of the clusters identified was named "digital twin of the own factory". The primary intention was to always have a real-time status of all resource in the own production system including facilities, parts and staff. For the radical transformation of the products, one of the clusters was the DT of each individual form on the customers' site. It is expected that a fully digitalized and automated press shop would need full control and real-time monitoring of the complete production flow and all components of the press shop. In this regard, the workshop participants discussed how to set up the cooperation with press manufacturers and logistics companies to discuss standards for the DT. Case Study Analysis {#Sec10} ------------------- The case study requirements (CSR) are derived from analysis of the two case studies and are presented in the following with a short motivation from the cases. *CSR1: DTs have to support the goals and business models of the company.* Both companies did not have any ongoing DT activity before the start of the digital transformation initiative. Once the workshops explicated ideas for service and business models that demand DT-like functionality, DTs were seriously considered and finally selected for implementation. In both cases, the primary goal is not to implement DT per se but to provide services or create a platform which can be facilitated by DTs. *CSR2: DTs are part of operations in an enterprise* -- *either to support manufacturing execution in the own or the client's production, or to facilitate value*-*added services on the customer side, like, e.g. predictive/preventive maintenance.* The concept of DT and envisioned functionality appeared in the use cases in different shapes: a) the DTs of the company's products installed at the clients' sites for the purpose of offering services depending on (real-time) data supply and monitoring (e.g., pumping-as-a-service requires monitoring of the pumps installed at the clients' industrial facility), b) the DT for the control of a facility possibly integrating various components (e.g. the DT of the manufacturing facility of company B or the DT of a ship which has a pump of company A installed), and c) the combination of a) and b), i.e. the company's product monitored in a client's facility. E.g., the form of company B with remote monitoring for purposes of preventive maintenance and local monitoring for optimizing production in the press shop. Options a) and b) require different information to be aggregated, displayed, and monitored. *CSR3: What aspect of reality has to be represented in a DT depends on the organizational integration and the intended business model of the company.* CSR1 sates that DTs must be supporting a company's business model. When implementing business models, this means that the digital twin has to provide the information about status or operations of the product required for the value creation underlying the business model. E.g. in case A, pumping-as-a-service requires to capture the performance of a pump to be able to invoice the provided hours or pumped volume, the energy consumption of the pump, and the status of lubricants to avoid problems in the service. *CSR4: Identification of features and parameters that have to be visible in the DT can be supported by developing business model prototypes and investigating organizational integration.* In both case studies, the options for new DT-based services were subject to an initial feasibility study. This study started from the definition of what service has to be provided for the customer, what information and functionality are required for the services (i.e., specification of features and parameters) and how this information is processed and used in the enterprise to deliver the service (i.e. the organizational processes). *CSR5: Component developers request a better methodical and technical integration of DTs (platform) development and component development.* In particular in case B, the case study company made clear that the development of a smart form would require collaboration with the manufacturer of the press for implementing the vision of a smart press shop. In case A, a similar request emerged when discussing the integration of pumps in complex systems, like, e.g. a cruising ship. Both cases showed the need for technical agreements (interfaces and platforms) and methodical agreements with the digital twin provider. *CSR6: Business models and organizational processes are subject to continuous improvement and so are DT features and parameters.* During development of business model prototypes, in both cases a kind of roadmap for stepwise implementing and extending services and business model was discussed, and the actual prototype intended to cover only the first stage. An expectation was expressed that the first stage would have to be extended based on the feedback of the customer and lessons learned from operations. With respect to modeling support, our recommendation is to explicitly model organizational context and business models as preparation of the DT design. Requirements for Digital Twin Design from Literature {#Sec11} ==================================================== DTs are usually designed and operated in the context of industry 4.0. In the field of production systems, there is a substantial amount of work on DTs. In the context of this paper, the intersection of digital transformation and DT as industry 4.0 solution is most relevant. Mittal et al. \[[@CR20]\] investigated what manufacturing small and medium-sized enterprises (SME) need to successfully adopt industry 4.0. As a result, 16 specific requirements for SME were identified including smart manufacturing solutions for specialized products, which includes DTs. Schumacher et al. \[[@CR21]\] proposed a maturity model for assessing Industry 4.0 readiness and identify nine dimensions of maturity and 62 maturity items in their Industry 4.0 Maturity Model. The maturity items include technology and product related aspects, like digitalization of products, product integration into other systems, and DTs. Considering the objective of this research our primary focus is on supporting the fit of the DT to the organization's needs in the industry 4.0 program, which, as discussed previously, can be supported by modeling. There have been several investigations of the needs for modeling support for industry 4.0. Hermann et al. \[[@CR9]\] present four main principles of industry 4.0, namely:*Interconnection* supporting various aspects of communication between actors, such as human to human, human to machine, and machine to machine.*Information transparency* requires supporting the identification and linking of various data types and sources, e.g. sensor data, process execution data, and factory designs, which in essence leads to DTs. A part of this task is the creation and monitoring the surrounding environment and situational properties related to the factory, i.e. the application context needs to be modelled and monitored. Some of the context information might also be needed in advance which requires using the means of predictive data analytics. All data needs to be presented to participants in the industry 4.0 design, depending on the criticality and relevance.*Decentralized decisions*. The design should be able to combine local as well as global information to support decentralized and autonomous decision making.*Technical assistance*. The decentralized decision making needs to be supported by assistance IS that are able to aggregate and visualize content in various formats suitable for different application contexts. Wortmann et al. \[[@CR8]\] report on a systematic literature review and in terms of the expected benefits for modeling for industry 4.0 puts forward the following: reducing time (development time, time-to-market), reducing costs (of development, integration, configuration), improving sustainability, and improving international competitiveness. This is in line of what are the general intentions of allying development methods and tools. In the context of industry 4.0 modeling addresses cyber aspects, physical aspects, or cyber-physical aspects of which the latter is the least researched and for which the least number of contributions have been elaborated. Wortmann et al. indicate that the current trends include methods for modeling digital representation, failure handling, human factors, information management, integration, process, product, configuration validation and verification, as well as visualization. The areas of product modeling, validation and verification, and information management attracting the most attention right now. Human factors and visualization are addressed by considerably fewer contributions. However, this study focused mostly on methods that have proven useful for IS design and development, and these methods do not support a holistic view on design that integrates organizational and human aspects with the more common IS aspects. The analysis of the current state of modeling for the purpose of designing industry 4.0 solutions, including DTs, calls for a number of areas of advancements, as follows. Concerning *modeling and model management*:Support for integrated multi-perspective views on all aspects of, such as, business and organizational, IS architecture, implementation, and operation at runtime.Integration of different artifact kinds such as models, 3D drawings. In this regard, Wortmann et al. call for the integration of models in the engineering, deployment process, and operation processes. To achieve alignment, the integration should start with the business design and requirements for the engineering process.Supporting design models with runtime data and, consequently, extracting models that can be used in later design iterations from runtime data. Using runtime data for the purpose of assessing the performance of designs, especially reusable designs that are applied in several operational installations. Concerning *adaptation and adjustment:*Support for adaptation and adjustment of the solution according to changing business goals and requirements as well as application context.The solution should have built-in means for runtime adaptations that do not require re-design and re-deployment. Concerning *continuous lifecycle management:*Supporting visualizations of runtime data in design models, e.g. by specifying what data should be presented and in what format. Management dashboards and presentation views can be generated from models.Support of the management of the complete lifecycle including design and runtime. With respect to the latter, \[[@CR8]\] discuss the possibility of adopting the DevOps principles for developing industry 4.0 solutions. The proposed vision for such a lifecycle is similar to the CDD process \[[@CR10]\], discussed in Sect. [6](#Sec12){ref-type="sec"}. Analysis {#Sec12} ======== Discussing the Requirements from Literature and Case Studies {#Sec13} ------------------------------------------------------------ First, we will discuss the requirements from Sect. [5](#Sec11){ref-type="sec"} and how the three main topics of (1) modeling and model management, (2) adaptation and adjustment; and (3) continuous lifecycle management, can be addressed by EM and CDD. This will be followed by a discussion of the case study requirements. *Modeling and model management.* It calls for multi-perspective views to integrate various aspects and artefacts of the organizational design and align them with the DT design. Multi-perspective EM methods, such as 4EM, are suitable for supporting this. The organizational aspects need to be linked with capabilities and DT designs. A part of this task would be modeling the context information that affects the operation of the DT. Since DTs are operated continuously the runtime data allows the assessment and improvement of the design models. E.g. the Context Modeling component supports the design by officering a set of measurable context elements that are already available in the context platform. CDD includes a component for Reuse of Capability Designs supported by a pattern repository that captures pattern performance data over time \[[@CR19]\]. This information is valuable for new as well as for improving the existing designs. *Adaptation and adjustment.* DTs need to be operated continuously, in various situations, and according to various business models. It can also be expected that these change under the lifetime of a DT. In this respect EM can be used for capturing the business dimensions of change, and CDD components for Capability Design and Context Modeling are to be used for capturing changes in the application context. Components for Reuse of Capability Designs and for Runtime Adjustments can be used to specify automatic adjustments or reconfigurations of the solutions including the DTs. *Continuous lifecycle management pertain* to two key aspects. First, visualization of operational and contextual data at runtime and then using this data and information to create new business models and DT designs as well as to change the existing ones. Part of the CDD method is generation of capability monitoring applications from capability design models and context models. Similarly, a monitoring dashboard for DTs can be generated from capability models, because it allows specifying KPIs and context elements together with their calculation from measurable properties that can be assembled from various data sources -- internal application data as well as external environment data. Concerning the second aspect, the lifecycle support, CDD is focusing on capability design and context-based adjustment of IS. To include DT designs in capability designs would need having a more explicit integration with EM as well as dedicated tasks for designing the functionality of DTs. This would imply that the DT is designed together with the capability as a solution to a business goal. Such an approach would contribute to ensuring that the DT fits the business design. The CDD method is also supported by a tool environment which would be needed to monitor context and runtime data, calculate KPI values, and, if necessary, to trigger adjustment algorithms.Table 1.Requirements from case studies supported by the CDD method componentsCDD method componentsCSRs*Enterprise ModelingCapability DesignContext ModelingReuse of Capability DesignsRuntime Del. AdjustmentsCSR1*Captures the business motivationDT design linked to the requiring parts of the enterprise modelCaptures reusable solutions, e.g. DT capabilities*CSR2*Captures the business motivationLinks design to the business motivationContext monitoring with model driven dashboardsSpecify operational adjustments of DTs*CSR3*Captures the business modelLinks the capability driven DT to the business modelModel the context data for monitoringSupports management of reusable components*CSR4*Captures the business motivationCapability driven DT and allows model drivenContext monitoring with model driven dashboards*CSR5*Management of reusable artifacts, incl. their performance*CSR6*Captures the business motivationCapability-based development of DT functionalityModeling of the DT usage context, generation of dashboardsContext driven adjustment at runtime The requirements elicited in the case studies are to a large extend addressing similar issues to the requirements from literature. Table [1](#Tab1){ref-type="table"} summarizes the CDD support. The requirements from literature and the case studies point to the need for the extension of the DT design with the aspects of business motivation and lifecycle management. The following modeling artifacts and practices contribute to this purpose:Enterprise models to capture the business models. Later they can be linked with the DT design models repressing the technical details of the DT.Capability design models to represent the more detailed designs of the DT.Context models to show the dependence on local and global data in the environment as well as to adjustments of the DTs and their monitoring dashboards.The capability design models and the enterprise models need to be linked to establish the business motivation and fit of the DT.Capability designs and context models should be used for generating dashboards for DT management. Key data types that have the potential of being useful here are context data, KPI, historical data about performance of reusable components.The models used need to be reasonably open and extendable in order to be able to incorporate additional perspectives of modeling. Supporting the Continuous Way of Working {#Sec14} ---------------------------------------- Concerning requirements CSR5 and CSR6, they can be supported by the CDD's method components as discussed in the previous section, but they also call for the establishment of a new way of working. It needs to support the core tasks of development and management of efficient DTs, such as, capturing the business motivation, design of the DT, and delivery and operation of the DT. The CDD process, which shares similarities with the DevOps principle of continuous development and operation, has the potential of being adapted for this purpose. Wortmann et al. \[[@CR8]\] also call for this kind of approach to DT development and operation. The case study requirements suggest that to make the DTs more fitting to the business model, explicit focus should be on the issues such as business goals, processes, and integration with the IS architecture. These are issues suitable for EM. Figure [1](#Fig1){ref-type="fig"} proposes a DT development and management lifecycle that incorporates three sub-cycles -- EM, DT Design, and Delivery and Operation. The internal steps and tasks in the sub-cycles follow the established procedures in \[[@CR7]\] for EM and in \[[@CR10]\] for Design and Operation. The following artifacts support the transition between the sub-cycles (grey arrows in Fig. [1](#Fig1){ref-type="fig"}): EM provides explicated knowledge about the business motivation for the DT in the form of enterprise models.Capability design provides (1) capability based digital twin design that are executable in the sense that they are integrated with the physical twins, and (2) generates the monitoring applications for digital twin management from the context model.The delivery and operation sub-cycle provides data types (e.g. context element types, measurable properties, KPIs) of available data used at runtime of the digital twin. This allows extending the existing designs as well as selecting existing and obtainable data in new designs. The Design provides best practices and reusable components on which the EM sub-cycle can base new business developments. Fig. 1.An overview of the envisioned capability-driven cycle of management digital twins. Feasibility Demonstration {#Sec15} ========================= Figure [2](#Fig2){ref-type="fig"} illustrates the feasibility of the CDD use with a fragment of a capability model consisting of goals, capabilities, and context modeling elements. The digital transformation workshop at company A identified an option to develop a pump-as-a-service product. When prioritizing the options, this option was top rated and, hence, converted into Goal 1 to develop pumping-as-a-service. It was refined into three sub-goals aiming at low maintenance pumps (1.1), possibility for real-time monitoring (1.2) and development of a preventive maintenance service (1.3). KPIs were set for all three sub-goals. The goal model is shown on the right side of Fig. [2](#Fig2){ref-type="fig"} and follows the 4EM notation.Fig. 2.A capability model for operating product as a service. From a capability perspective, pump-as-a-service can be considered in more general terms as product-as-a-service capability if company A wants to offer other physical products as a service. The core sub-capabilities of product-as-a-service are real-time monitoring a product at the client site (which motivates the digital twin) and a reliable product without downtimes. The capabilities are visible in the center of the model. The left side shows the context elements and the measurable properties on which they are based, in which context sets they are included, and their relation to capabilities. These context elements are calculated from the measurable properties and monitored once the capability is implemented. This also specifies the data to be provided by the DT. An example for a context element is the total energy consumption of the pump measured by the energy consumption of the motor and the energy recuperation achieved by the installed power converter. This context element is required for providing the product as a service (as part of the cost structure) and also for evaluating when to trigger preventive maintenance. From this model the CDD Environment would be able to generate a monitoring dashboard for Capability 1.2. It would display energy consumption, product status, and lubricant level as runtime properties of the application for operational monitoring. For a more strategic view on the capability fulfillment of KPI3 and KPI4 also need to be monitored. For brevity reasons, context and KPI calculations as well as the operational processes linked to the capabilities are not shown in the model. Concluding Remarks and Future Work {#Sec16} ================================== The starting point for our work was the analysis of two industrial cases on how digital twin initiatives emerge and what the driving forces for starting implementation projects are. An observation from both cases is that digital transformation and development of new business model options are a motivation and driving force of DT implementation. Both case studies resulted from the need of companies to explore innovative products or services based on digital technologies, embedded into their operational processes and structures. DTs are considered as a way of integrating innovative products/services into the operational context, which leads to requirements for DT functionality and implementation. In summary, we see clear support for our conjecture that DT have to be integrated into the organizational structures and business models of manufacturing enterprises. Furthermore, we analyzed requirements for DT development from literature. Requirements pose a number of issues concerning a model-based design and management with a particularly strong emphasis on establishing a good fit between the business issues and DT-based solutions. This is an area to which EM and CDD have the potential of contributing. In this regard we have proposed an integrated lifecycle and discussed how capability-based DT designs could be used. The initial feasibility demonstration gives reason to the assumption that this approach is promising and should be pursued. Concerning future work, we plan to investigate to what extent existing technically motivated DT implementations are used for new services or products and cause digital transformation in the enterprise and how the design of the DT features can be included in the capability design. We also aim to establish a development environment for the proposed way of working by integrating components of the CDD Environment with the modeling support by the ADOxx tool.

Background {#Sec1} ========== The concept of the autonomy of a child in the context of healthcare is both complex and challenging globally. In South Africa the controversy surrounding children and autonomy has come into sharp focus since the promulgation of the Children's Act 38 of 2005 (hereinafter referred to as "the Children's Act" or "the Act"). Much of the debate revolves around the concept of maturity and the child's developing capacity to consent. The process of development generally concerns progressive advances from one state, usually primitive or simple, toward another, usually more complex or advanced. Where this process typically terminates is what is colloquially (and formally) understood as maturity. There are various dimensions of maturity including emotional, biological, cognitive and social. However, for the sake of firmer pertinence to our research question we concern ourselves herein with cognitive and social maturity as they bear directly on Western liberalism and African communitarianism, the former often emphasising rationality and individualism and the latter; sociality of persons albeit not refuting the significance of other dimensions of maturity such as the emotionality of the deciding subject in decision-making. We can thus conceive of a child as a developing *person* with evolving capacities like autonomy, mental (decisional) capacity and capacity to assume responsibility. Notwithstanding possession of capacities, we must first plainly conceive of a child as a *person*; a human being. Although this plain conception of a child is indeed attractive as it admits no prejudice toward children as rights-holders, a fuller and more adequate definition is required to define when a subject becomes a person[1](#Fn1){ref-type="fn"} and at what age we should consider a person no longer to be regarded as a child but rather an adult \[[@CR1]\]. The Constitution of the Republic of South Africa (hereinafter referred to as "the Constitution") aligns itself with the African Charter on the Rights and Welfare of the Child (ACRWC) and the United Nations Convention on the Rights of the Child (UNCRC) \[[@CR2], [@CR3]\] with regard to exalting children as independent legal actors -- as stipulated in the Act[2](#Fn2){ref-type="fn"}, and in defining which persons are entitled to provisions entailed therein. It provides that a child is a *person* under the age of 18 years[3](#Fn3){ref-type="fn"} \[[@CR2], [@CR4]\]. It also follows that a person is considered to have attained majority at this age. Furthermore, under the Children's Act, a child is considered a rights-holder, not merely a property or extension of her parents or an object of adult concern \[[@CR2], [@CR5]\]. Children are indeed persons with an evolving capacity for individual autonomy \[[@CR6]\] hence deserve the right to express their views freely in matters affecting them[4](#Fn4){ref-type="fn"}. The relevant sections in the Children's Act attend to the various jurisdictions of a child but of particular interest to us is section 129 of the Children's Act which pertains to the consent of children to medical treatment. Section 129 expressly dictates the prerequisites for the medical treatment of a child and stipulates as follows: '(2) A child may consent to his or her own medical treatment or to the medical treatment of his or her child if-the child is over the age of 12 years; andthe child is of sufficient maturity and has the mental capacity to understand the benefits, risks, social and other implications of the treatment.' \[[@CR4]\] In the past, when the Child Care Act 75 of 1983 was still in effect, only children above the age of 14 years could consent to medical treatment \[[@CR7]\]. What necessitated law reform was a realisation of a number of shortcomings experienced with the Child Care Act[5](#Fn5){ref-type="fn"} and a need to fully acknowledge children as rights-holders. A lower threshold for age of consent was thus seen as a means to promote access to health services, promote participation of children in health decisions affecting them in accordance with international trends \[[@CR7], [@CR8]\]. Over the years there has been mounting empirical evidence suggesting lowering age thresholds for decisional capacity in children. For example it has been demonstrated that children below 12 years can make well considered decisions \[[@CR9]\] and that children as young as nine years old can understand issues pertinent to decision making in clinical trials \[[@CR10]\] however the statutory age of consent to medical treatment as stipulated in various countries appears arbitrary as it varies from 12 to 19 years \[[@CR11]\]. A child contemplated under the Children's Act must satisfy two requirements before accessing medical treatment on his or her own, that is, without parental, guardian, or care-giver's consent being required. The first requirement is that the child must have reached 12 years of age to consent. The second requirement is that the child must have 'sufficient maturity' and decisional capacity to understand the 'benefits, risks, social and other implications of the treatment.'[6](#Fn6){ref-type="fn"} \[[@CR12]\] However, there are a few deficiencies in this section of the Act with regard to definitions, regulations and sufficient descriptions \[[@CR8]\]. Firstly, the Act does not provide a definition regarding what ought to be considered medical treatment. Hence, for the purposes of this article, we define medical treatment as a non-invasive intervention usually in the form of a drug[7](#Fn7){ref-type="fn"}. Secondly, the Act also does not provide a definition of sufficient maturity. Hence, we will comprehend that the Act infers by 'sufficient maturity' a degree of cognitive development that affords a child the kind of engagement necessary in decision-making comparable to that of fully developed persons, namely*,* adults[8](#Fn8){ref-type="fn"}. We will provide an alternative rendition of 'sufficient maturity' in the course of this article. Thirdly, there is no provision in the Act specifying how the health practitioner ought to assess a child's decisional capacity. This is compounded by the fact that there currently is no standard objective tool for assessing the decisional capacity of children \[[@CR9]\]. Moreover, considering that South Africa is a culturally diverse country \[[@CR5]\] another concern with regard to the implementation of the Act involves the potential consequence of conferring (autonomy) rights on children without commensurate responsibilities to their community[9](#Fn9){ref-type="fn"}. For '\[i\]n the African context, for example, individual autonomy is of smaller status than the pursuit of the communal good'[10](#Fn10){ref-type="fn"} \[[@CR5]\]. In view of this, it appears the conferring of autonomy rights on children without cementing their reciprocal duties erodes interdependent relations between the child and his or her community \[[@CR13], [@CR14]\]. Our research question may be posited as follows: given the newest developments in child law as regards the conception of the child and his or her participation in society, how may appeals to different moral theories (African communitarianism and Western liberalism) aid in finding better and alternative means of determining *how* and *by whom* decisions about medical treatment of the child should be made? Perhaps there has not been a time better suited to address questions of this nature than today given the near-universal advocacy for children's rights and the resurgence of activism and scholarly criticism against old hegemonic conceptions such as the status of children in civil society, person and personhood and so forth. In advancing forth our argument we first assert children as rights-holders, give an overview of the doctrine of informed consent and the principle of respect for individual autonomy and the legal conception of a person in the setting of the Constitution; discuss African communitarianism with regard to its notion of a person and personhood and the child and the implications thereof in the consent of children to medical treatment. And in pursuit of a case-specific definition for sufficient maturity we appeal to the notion of capacity for responsibility. Lastly, in view of both legal liberal and African communitarian moral vantages we conclude by giving due attention to the enquiry whether a 12 year old is of sufficient maturity to consent to medical treatment with the conviction that no moral theory should be assigned an absolute (moral) value *a priori*, that is, antecedent to the context within which it is to be observed and/or contemplated. Discussion {#Sec2} ========== Asserting children as rights-holders {#Sec3} ------------------------------------ As a point of departure, a child is a developing person. When he or she obtains decisional capacity of such degree that affords him or her the kind of engagement necessary in decision-making comparable to that of fully developed persons, *viz.* adults, we will comprehend that as what is inferred by the Act as 'sufficient maturity'. It appears to follow from this that a child with sufficient maturity ought to be equally afforded autonomy rights in decision-making, including medical treatment as is the case in adults. For '\[c\]onferring rights on children is viewed as '*recognising their moral equality with adults, thereby underscoring the moral worth of all human beings, irrespective of their situation*.' (emphasis added) \[[@CR12]\], and by autonomy rights we understand broadly those entitlements persons have which allow them the freedom of involvement in matters affecting them as members of civil society, be they public or private (also referred to herein as participatory rights); different perhaps to rights in general which are often conceived as entitlements persons have plainly by virtue of being persons . Having said that, do these rights also extend to those children who do not possess sufficient maturity and/or decisional capacity? The Act is unambiguous on this issue. Where a child is judged to lack sufficient maturity and decisional capacity to understand the benefits, risks, social and other implications of the treatment the authorisation of his or her consent devolves on the parent, guardian or caregiver. However, this question highlights a central problem in many rights theories as to what we mean by the notion of rights and who qualifies to be a rights-holder \[[@CR15]\]? For if the conferring of participatory rights is contingent on possession of certain dispositions or traits such as capacity, degree of maturity, age, condition of dependency and so on as some commentators might argue then holding such rights indeed becomes exclusionary and further, fails dismally in serving the very groups it was purposed to protect \[[@CR12], [@CR15]\]. Hence, we maintain: if we are to truly recognise the moral equality of children with adults we ought to grant that capacity of whatever kind need not be the arbitrating principle on the conferring of rights on children. Admissibly, as Mosikatsana observes, '\[t\]he difficulty with granting children rights is that their physical, emotional, and intellectual immaturity cause dependence on adults to assist children in exercising those rights' \[[@CR13]\], but, as O Neill writes (as cited by Mosikatsana) the fact that children 'cannot claim their rights for themselves...is no reason for denying them rights. Rather it is reason for setting up institutions that can monitor those who have children in their charge and intervene to enforce rights.' \[[@CR13]\] Therefore, as convenient a notion as sufficient maturity and decisional capacity may appear, they do confine our discourse on the rights of the child to the exclusion of others and their claims. Moreover, children have moral status (or moral worth) plainly by virtue of being humans or persons (these terms are used interchangeably in this paper). It would indeed appear morally unsound, let alone 'morally monstrous' \[[@CR16]\], for one to argue that children have lower moral status compared to adults as it also appears unlikely that one can indeed justify it with sound moral reasoning. It is rather best assuming a value theory that does not in any manner legitimise preferences to the acquisition of certain capacities in the development of persons \[[@CR17]\] in order for us to arrive upon the conclusion of equal rights and moral status of all persons plainly by virtue of their humanity. And by humanity we broadly refer to the totality of universal potentialities, qualities and dispositions which both constitute and distinguish us as persons, whatever these may entail. Here the conferring of rights is then premised solely on the notion of humanity and not on some other contingent condition. (This argument equally applies to the entities enumerated in the following developmental continuum: 'blastocyst, zygote, embryo, foetus, neonate, baby, infant, child, minor, adolescent, adult' \[[@CR16]\].) This attribution of rights founded plainly on the notion of humanity is also apparent in the preamble of the UNCRC which recognises the 'inherent dignity and...equal... rights of all members of the human family' \[[@CR12], [@CR18]\]. It follows from this that children are indeed rights-holders for the same reasons we recognise in adults (that is, their humanity) and thus should be afforded equal rights as adults including participatory rights. However, to exercise participatory rights requires autonomy -- a capacity that is acquired over time through the process of development. It is obvious that certain age-groups will lack this capacity and thus may not have the commensurate wherewithal to exercise participatory rights in decision making \[[@CR19]\]. Although there seems little contention to assert this, it need not necessarily follow that persons judged deficient of such capacity be stripped of that right completely as the argument herein advanced is that the affordment of participatory rights should not be predicated on the basis of capacity to exercise a right but rather on the existence of fundamental human interests that deserve protection from prejudicial forces. In contending the notion of capacity in the setting of the "rights talk" Federle writes:"Children clearly have been disadvantaged by a rights theory premised upon capacity. The incapacities of children and their concomitant need to be protected from themselves and others permit the state to restrict the activities of children in ways that would be impermissible in the case of adults. Furthermore, these incompetencies suggest that the rights children do have are somehow different, less fundamental, and more easily overridden by paternalistic concerns for the safety and well-being of children. Consequently, the courts have authorized significant restrictions on the liberty interests of children as legitimate protective measures. Nevertheless, our laws may subject children to selective and discriminatory laws with concomitantly greater restrictions on their liberty than would be sanctioned in the case of adults. \[[@CR15]\]" It is for these reasons that we argue that children are rights-holders regardless of whether or not they possess the wherewithal to exercise these rights. The doctrine of informed consent {#Sec4} -------------------------------- The doctrine of informed consent holds that persons are their own sovereign and should thus be allowed to make the final decision on affairs concerning them providing that the elements required for informed consent (or informed refusal) \[[@CR20]\] have been satisfied. These elements include:Competence;Disclosure of information;Understanding and appreciation of information disclosed;Voluntariness in decision-making;Ability to express a choice \[[@CR5]\]. In view of the above it may safely be declared that informed consent has occurred when a competent person has received a thorough disclosure, understands and appreciates the disclosure, acts voluntarily, and consents to the intervention \[[@CR19]\]. We briefly elaborate on these in the following accounts. Competence[11](#Fn11){ref-type="fn"} simply refers to the ability to perform a task \[[@CR20]\]. It is task and context-specific and changes over time. By convention, age and decisional capacity are thought to be the chief elements that constitute competence. Albeit several competence assessment tools for children have been devised by various authors e.g. Hopkin's Competency Test, Competency Questionnaire-Child Psychiatric and the Competency Questionnaire-Pediatric Outpatient Modified Version, there currently exists no standard objective tool to assess a child's competence to consent to medical treatment \[[@CR9], [@CR10]\]. This inclines assessors of competence (health practitioners) to make judgements based on subjective assessments. A patient's competence is influenced by their experience with a medical condition, hospitalisation, family relationships and social roles and development \[[@CR21]\]. Furthermore, '\[i\]t is a legal obligation for health practitioners to disclose relevant information to their patients regarding:The patient's health condition (except when disclosure of information would be contrary to the patient's best interest)Available diagnostic and treatment optionsRisks, benefits, costs and consequences attached with each optionThe option of non-treatment, that is, informed refusal and its implications.' \[[@CR5]\] The patient should also attach significance to the information disclosed. 'The process of consent should also be conducted in a language that the patient understands and in a manner that considers the patient's level of literacy. This is especially so with children.' \[[@CR5]\]. In addition, for informed consent to be valid it must be voluntary, that is, the patient must not be influenced by other individuals either by coercion, persuasion or manipulation \[[@CR5], [@CR19]\]. Lastly the patient's choice to treatment or non-treatment may be expressed orally, in writing or may be implied, that is, tacit consent \[[@CR19], [@CR22]\]. Capacity for responsibility {#Sec5} --------------------------- A deciding subject, in this instance a child, ought not to only consider given choices but also accept the prospective responsibilities involved. And to '*accept responsibility means to be able to be held accountable for whatever decisions are taken, on the basis of the assumption that reasons can be provided, that they have been thought through, and even though they might be fallible.*' \[[@CR23]\]. That is, a deciding child must also have the capacity for responsibility for that particular choice decided upon, whatever this may entail. Capacity for responsibility therefore refers to a deciding subject's ability to deal with the likely outcomes of his or her decision. Whilst we grant that a person need not possess capacity of any kind to have moral status and constitutional rights (human dignity, privacy, freedom), as we established above, we argue that a deciding subject *must* then necessarily possess or be facilitated insofar as it is practically possible to possess the commensurate wherewithal for responsibility to account for that particular choice decided upon. In view of this, we arrive at our ultimate definition of 'sufficient maturity':"*A child has sufficient maturity to consent to medical treatment insofar as he or she can independently demonstrate (or be facilitated either by aids or a helper as far as it is practically possible in that given setting to possess) the commensurate wherewithal required to assume responsibility for that specific decision.*" To clarify this definition, let us make an example: a child patient is newly diagnosed with type I diabetes mellitus and it is required that she consents to using insulin injections as her treatment. To determine whether she has sufficient maturity to consent to using insulin injections the health practitioner must consider, among other factors, whether the child would be able to take the chronic medication as frequently as prescribed. A child who has previous experience with a chronic illness like asthma may be presumed to already possess the capacity to assume the responsibility of taking chronic medication. Those children whom it is believed cannot demonstrate the forgoing capacity in order to assume responsibility can be facilitated to attain this capacity. In the case where a child patient refuses treatment we advise that recourse be made to the best interest principle. A child (or adult) who fails this definition of sufficient maturity may be considered incompetent to make a decision. The Constitution on autonomy and the legal conception of a person {#Sec6} ----------------------------------------------------------------- Human dignity is expressly enumerated in the Bill of Rights Chapter of the Constitution as a human right that deserves respect and protection. It is a foundational value that 'informs the interpretation of other specific rights' \[[@CR24]\]. Albeit some authors, such as Jordaan \[[@CR24]\], claim that one of the fundamental elements of human dignity include the capacity for *autonomy* whether understood as free-will or rational deliberation \[[@CR25]\], we maintain throughout this paper that human dignity in general denotes a universal, and objective value *inherent* to all human *persons* notwithstanding capacity*.* The notion of *autonomy* is derived from the Greek expressions: '*autos*' -- self, and '*nomos*' -- law, referring to a self-legislating agent \[[@CR19], [@CR24], [@CR25]\]. Autonomy is a constitutional value defined by the Courts as 'the ability to regulate one's own affairs, even to one's own detriment' \[[@CR24]\]. Implicit in this juridical definition is the acknowledgement of autonomy as a developmental phenomenon. This is inferred by the term "ability" implying that autonomy is an evolving capacity that is, *acquired* in the process of human development. According to the provisions of the UNCRC and the ACWRC, a child has autonomy rights. The Children's Act first defines a child as a *person* below the age of 18 years and further specifies in section 129 which children can fully exercise autonomy rights in the setting of consent to medical treatment (as dealt with above) \[[@CR4]\] It is plain from the forgoing definition of a child that rights are ascribable only to *persons* not things[12](#Fn12){ref-type="fn"}. According to Black's Law dictionary a natural *person* considered in juridical contexts is a human being; a legal entity with rights and duties that deserve protection and respect \[[@CR26]\]. However, it still remains unclear as to what is truly meant by the notion of *person* or *human being*; what potentialities, qualities and dispositions declare us as persons and thereby entitle us to constitutional rights (e.g. autonomy rights), and duties in general. We acknowledge the import of such a definition as a desideratum not only in juridical but also in philosophico-ethical contexts with regard to moral status and abortion. African communitarianism on autonomy, the conception of a person and the consent of a child to medical treatment {#Sec7} ---------------------------------------------------------------------------------------------------------------- "'Amidst gathering talk of human rights and civil society, of the celebration of autochthony and authenticity, the version of an African Renaissance arises to counter the rampant excesses of European modes of being-in-the world' \[[@CR27]\]" Communitarianism is a moral theory concerned with the pursuit of the communal good. It expressly repudiates individual autonomy (and liberal moral theory) and exalts community. In this theory, individual rights become docile whilst duties owed by a member to his or her community are held to be of great import, and communal values such as mutual reciprocity, collective loyalties and solidarity are endorsed \[[@CR19], [@CR28]\]. The consideration of a *person* has always been at the centre of consternation in this moral theory. The problem can be stated as follows: is a person wholly embedded in a communal matrix of interrelations and interdependencies without the concession of individual autonomy as radical communitarians insist or does one retain his or her individual merits like autonomy within a community as moderate communitarians argue?[13](#Fn13){ref-type="fn"} \[[@CR19], [@CR28]\]. African societies generally uphold communal values (African communitarianism), of those, the highest weight is assigned to relationships shared within a community \[[@CR19], [@CR28], [@CR29]\], and to human life (vitality). Thus, a *person* has the duty to preserve the continuity of such relationships by pursuing the communal good, whatever this may entail. In traditional African thought a person exists as an extended entity embedded within a communal matrix of interrelations and interdependencies, owing much to the relational nature of human beings. Thus, a person is regarded as an ontological and epistemological reference thereof \[[@CR28], [@CR29]\]. This concept of a person is no better expressed than in John Mbiti's coinage of the African ethos:"'I am because we are; and since we are, therefore I am' \[[@CR30]\]" Personhood in the African ethos is thought to be *acquired* through a process of incorporation into the community \[[@CR29]\] and this involves executing one's duties owed to the community. And we may add here that this requires a good measure of social maturity. Personhood in this view is something that one can indeed fail. Moreover, in this view a child is not considered as a *person* as *it* is yet to fulfil *its* duties to attain personhood \[[@CR29]\]. This however raises an important question: How can we acknowledge the rights of children (as we asserted elsewhere) if we cannot conceive of them as full persons? To answer this we appeal to an alternative interpretation of the notion of human dignity established upon the African communitarian value for vitality \[[@CR25]\] as opposed to autonomy and declare this as follows: a child (or being) has human dignity thereby human rights insofar as he or she has vitality[14](#Fn14){ref-type="fn"}. In truth, however, African communitarianism[15](#Fn15){ref-type="fn"} is premised upon a duty-based system; not naturally perceived nor experienced as being oppressive to the individual since the individual himself or herself realises his or her interests as being consonant with the pursuit of the communal good and sees nothing else outside this \[[@CR31]\]. He or she therefore finds little sense in "going on" about individual rights that seemingly conflict with the harmony of interrelations and interdependencies shared within the very community whence he or she derives self-worth and security with regard to individual welfare. In healthcare where informed consent is a necessary ethical and legal requirement to solicit from a patient before performing an indicated medical intervention a patient from an African communitarian society may often wish to consult with his or her community to make a decision \[[@CR32], [@CR33]\]. This derives from the fact that in African communitarian societies the best interests of all persons, not only the child, are determined by the community based on the communal value-system. Hence important decisions are arrived upon through collective discussions, often in the presence of elders from the community since their wisdom is highly regarded concerning (moral) decision-making to guard the interest of the community. Where consent to the medical treatment of a child (or person) is concerned it is likely that the community from which the child belongs will collectively decide on this. It appears therefore that the African value system is indeed in conflict with the law which permits a child 12 years or older to make an autonomous decision regarding his/her medical treatment. Conclusion {#Sec8} ========== It seems reasonable to suggest that we move away from a general age of consent toward more individualised, context-specific approaches in determining the maturity of a child patient to consent to medical treatment. Conferring rights upon children based on capacity (autonomy rights and decisional capacity in this instance) alone may be myopic at best. Hence, we subscribe to the minimal notion that where a child is able to express his or her will based on an established value system and rationality they ought to have their views taken seriously in decisions pertaining to their medical treatment. We also suggested an alternative definition of sufficient maturity which can be used to determine whether a child patient is indeed competent to consent to medical treatment or not without unfairly discriminating against children based on their perceived incapacities. The proposed definition places emphasis on the prospective duties that the decision-maker ought to be able to execute consequent to his or her decision. In South Africa conflict exists between law and the African value system. In view of both legal liberal and African communitarian moral theories it is however plain that no one theory can account for how we ought to conceive of a child and his or her freedom to consent to medical treatment antecedent to the context within which the child is raised. Hence, we argue that no moral theory should be assigned an absolute (moral) value *a priori*, that is, antecedent to the context within which it is to be observed and/or contemplated and propose that in the case of a child who belongs to an African communitarian society decision-making with respect to consent to her medical treatment ought to involve the child's community (included here are the child's parents/guardians/caregivers) insofar as it is possible provided that the best interests of the child are awarded priority. ACRWC : African Charter on the Rights and Welfare of the Child UNCRC : United Nations Convention on the Rights of the Child Alternatively, when does personhood or childhood begin? The Children's Act 38 of 2005 is the most significant statute in South Africa entailing provisions and protections of the constitutional rights of the child. Among these rights are those pertaining to the participation of children in health treatment decisions. See Article 2 of the ACRWC and Article 1 of the UNCRC. South Africa ratified the ACRWC in 2000 and the UNCRC in 1995. See section 10 of the Children's Act. Some of these shortcomings included difficulties finding the caregivers or foster parents of orphans or child-headed households when required to consent to medical treatment thereby access health services. This requirement derives from *the Gillick v West Norfolk and Wisbech Area Health Authority* case and is commonly referred to as the "*Gillick* competence test" \[[@CR8]\]. However, we also maintain that regardless of whether an intervention is indeed medical or not recourse should always be made to the principle of respect for human dignity as enshrined in the Constitution. It still remains to be proven whether such degree of maturity can indeed be clearly defined, measured and objectively validated with the support of empirical evidence. The term "community" will be used broadly to also denote family as a communal unit. The communal good involves preserving the continuity of communal interrelations and interdependencies. In this article we use the terms *competence* and *capacity* interchangeably. A calabash for example cannot conceivably have rights (or duties) insofar as any rational person cannot conceive of it a person, or rather, insofar as it is not a person notwithstanding its aesthetic, economic and social value. For the sake of brevity we will not engage this enquiry any further, saving it for another occasion. Vitality may be thought as a being's ability (or potentiality thereof) to 'exhibit a superlative degree of health, strength, growth, reproduction, creativity, vibrancy, activity, self-motion, courage and confidence, with a lack of life force being constituted by the presence of disease, weakness, decay, barrenness, destruction, lethargy, passivity, submission, insecurity and depression' \[[@CR25]\]. The terms "African communitarianism", "African context" and "traditional African thought" are employed in the general sense where that which we denote as "African" refers to whatever it is that relates to the continent's indigenous cultures, people and heritage. Often it is asked how can one speak broadly of an African context, thought or ethos and so given the diversity within the continent. To answer this question let us imagine for instance there being culture *p* (*p* being one of the indigenous cultures of Africa). Although culture *p* may not be a microcosm of the African's mode of being in the world, it may be said that it shares certain commonalities with other indigenous cultures *mutatis mutandis* to permit reference to our observations in *p* as "African", especially if one considers African cultures forming intersecting lines or partly overlapping circles; where they intersect or overlap we may speak of specific (ethnocultural) universals which one may broadly denote as African. The authors would like to thank the Medical Protection Society for having hosted the annual Ethics Essay Writing Competition for health sciences students at Stellenbosch University for the past 5 years. As a winning entry in the competition in 2013, WG was awarded a prize of R5000 at the MPS Annual Conference 'Ethics for All' hosted at the Cape Town International Convention Centre (CTICC) in December 2013. Authors' contributions {#FPar1} ====================== WG conducted an in-depth literature review on the topic and composed the article in the form of an essay. He subsequently redrafted it into its article form. SK reviewed all drafts of the article, clarified conceptual issues and provided comments from her experience as a paediatrician and ethicist. KM proposed the topic of this paper, edited all drafts and gave guidance in composing the article. Competing interests {#FPar2} =================== The authors declare that they have no competing interests.

For addressing the question of cardiovascular importance of hypoglycemia, it is important to clarify its context. First, hypoglycemia is a result of treatment of hyperglycemia by oral insulin secretagogues or insulin. Chronic hyperglycemia usually expressed by HbA~1c~ level is considered a risk factor for cardiovascular disease, although this epidemiological association does not necessarily mean the existence of causal association, so the possibility cannot be excluded that HbA~1c~ may be only a marker of atherosclerotic vascular disease. Thus, in the present review the evidence related to hyperglycemia and hypoglycemia as factors contributing to the development of cardiovascular events will be discussed and the main following issues will be addressed: The relationship of hyperglycemia to cardiovascular disease will be documented based on analysis of epidemiological and clinical interventional studies. Furthermore, the evidence will be summarized that hypoglycemic episodes contribute to the development of cardiovascular events in patients with type 2 diabetes treated by hypoglycemia-inducing drugs. Finally, it will be demonstrated how the conclusions from the described studies translated in practical recommendations for personalized treatment of type 2 diabetes. Is hyperglycemia related to cardiovascular disease? {#s2} =================================================== The evidence about a relationship between hyperglycemia and cardiovascular disease comes from epidemiological studies and epidemiological post hoc analyses of clinical trials. For consideration of a biological variable, e.g., HbA~1c~, as a cardiovascular risk factor, it is important to analyze its relationship with cardiovascular disease also outside the diabetic range. The epidemiological study European Prospective Investigation into Cancer in Norfolk (EPIC-Norfolk) included 4,662 men and 5,570 women. Relative risks for cardiovascular disease (nonfatal or fatal coronary heart disease and strokes) adjusted for age and risk factors were calculated after 6-year follow-up period. An increase in HbA~1c~ of 1% (11 mmol/mol) was associated with relative risk for cardiovascular disease of 1.21 (95% CI 1.13--1.29 for males and 1.11--1.31 for females; *P* \< 0.001). Moreover, the increased risk associated with diabetes seemed to be mediated entirely through HbA~1c~ level, since diabetes was no longer a significant predictor when HbA~1c~ was included into multivariate model ([@B1]). Very similar results were found by another large prospective epidemiological study---Atherosclerosis Risk in Communities (ARIC)---which included 11,092 adults without history of diabetes or cardiovascular disease. After 15-year follow-up, an increase in HbA~1c~ of 1% (11 mmol/mol) was associated with hazard ratio (HR) of 1.19 (1.11--1.27) for coronary heart disease and 1.34 (1.22--1.48) for stroke ([@B2]). Epidemiological analysis from the UK Prospective Diabetes Study (UKPDS) showed a similar association. A reduction in HbA~1c~ by 1% (11 mmol/mol) was associated with a 14% decrease in fatal and nonfatal myocardial infarction (MI) (*P* \< 0.0001), as well as 12% decrease in fatal and nonfatal stroke (*P* = 0.035). The relationship between HbA~1c~ and incidence of cardiovascular end points was linear to the level of HbA~1c~ of 5.5% (37 mmol/mol) ([@B3]). On the other hand, an epidemiological analysis from the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) study showed that within the range of HbA~1c~ studied (5.5--10.5%; 37--91 mmol/mol), there was evidence for a threshold effect: While for microvascular events this value was 6.5% (48 mmol/mol), for macrovascular events and death the threshold was 7% (53 mmol/mol). Above this threshold, the risks increased significantly so that every 1% (11 mmol/mol) higher HbA~1c~ was associated with a 40% higher risk of microvascular events (*P* \< 0.0001), a 38% higher risk of macrovascular outcomes (*P* \< 0.0001), and a 38% higher risk of all-cause mortality (*P* \< 0.0001) ([@B4]). Epidemiological analysis of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) study showed that 1% (11 mmol/mol) increase in average HbA~1c~ during 3.4 years' duration of the study was associated with 22% increase in mortality (*P* = 0.0001). Interestingly, the relationship between mortality and HbA~1c~ was linear in the range of 6--9% (42--75 mmol/mol) only in the intensively treated group (*P* \< 0.0001), while no significant relationship (*P* = 0.17) was observed in the standard treatment group ([@B5]). Does the reduction of high blood glucose lead to a cardiovascular benefit? {#s3} ========================================================================== Studies in newly diagnosed patients with type 2 diabetes. {#s4} --------------------------------------------------------- ### University Group Diabetes Program (UGDP). {#s5} The first study to approach this question in patients with type 2 diabetes was the UGDP. This study included 1,027 patients and was statistically underpowered (with \~200 patients in each treatment category group: placebo, tolbutamide, phenphormin, insulin standard, or insulin variable regimens) to detect beneficial effect of any treatment modality. The first analysis, published in 1970, showed that despite better glycemic control, a significantly higher cardiovascular mortality was observed in a group treated by tolbutamide in comparison with placebo and both insulin regimens ([@B6]). Further analysis from UGDP showed that patients treated with tolbutamide had significantly higher incidence of fatal MI in comparison with patients on placebo (*P* = 0.01), while patients on variable insulin regimen had borderline significantly higher incidence of fatal MI (*P* = 0.06) compared with patients treated with placebo. There was no difference in incidence of nonfatal MI events among the four groups ([@B7]). With respect to the incidence of hypoglycemia in UGDP, the number of patients who had glucose levels \<50 mg/dL was zero for placebo, four for tolbutamide, three for standard insulin regimen, and five for variable insulin regimen. ### The UK Prospective Diabetes Study (UKPDS). {#s6} The UKPDS study included 4,203 patients with newly diagnosed type 2 diabetes. The main results of the UKPDS study were published in 1998 in two articles. UKPDS 33 reports results of 3,867 patients with newly diagnosed type 2 diabetes who were randomized to intensive glycemic control policy with sulfonylureas or insulin or to conventional treatment policy---primarily with diet. More drugs were added in both groups of patients if fasting plasma glucose was ≥15 mmol/L. The patients in the intensive group had median HbA~1c~ of 7.0% (53 mmol/mol) during 10-year follow-up, while patients in the conventional group achieved median HbA~1c~ of 7.9% (63 mmol/mol) ([@B8]). While significant risk reduction by 12% (*P* = 0.029) in the incidence of any diabetes-related end point in the intensive treatment group was observed, nonfatal and fatal MI incidence was reduced by 16% with a borderline significance (*P* = 0.052) ([@B8]). Major hypoglycemic episodes defined as the mean proportion of patients per year with one or more episode occurred with chlorpropamide (1.0%), glibenclamide (1.4%), insulin (1.8%), and diet (0.7%). Interestingly, after 10-year poststudy follow-up as more events occurred, risk reductions for MI (15%, *P* = 0.01) and all-cause mortality (13%, *P* = 0.007) became significant ([@B9]). The results of subgroup analysis of 1,704 overweight patients with type 2 diabetes randomized to intensive treatment by metformin or sulfonylurea/insulin or to conventional treatment were published separately ([@B10]). Patients treated primarily by intensive metformin treatment had a median HbA~1c~ level of 7.4% (57 mmol/mol) during the follow-up, while patients in the conventional treatment group had median HbA~1c~ level of 8.0% (64 mmol/mol). Patients allocated to metformin compared with the conventional group had significantly reduced risk for diabetes related death by 42% (*P* = 0.017), as well as for fatal/nonfatal MI by 39% (*P* = 0.01). Patients allocated to metformin had lower risk for all-cause mortality (*P* = 0.021) and for stroke (*P* = 0.032) compared with patients allocated to insulin or sulfonylurea. Major hypoglycemic episodes occurred in 0.6% patients/year treated with metformin ([@B10]). One of the explanations of lower cardiovascular preventive effect of sulfonylurea or insulin treatment in comparison with metformin in UKPDS might be that metformin-treated patients had lower incidence of severe hypoglycemic episodes. ### Outcome Reduction with an Initial Glargine Intervention (ORIGIN). {#s7} The ORIGIN study included a total of 12,537 participants, among whom 88% had diabetes and 12% had prediabetic dysglycemias. Patients were assigned either to insulin glargine or to standard care treatment. After the median follow-up of 6.2 years, there was no significant difference in rates of cardiovascular outcomes between the study groups. Rates of severe hypoglycemia were higher in the glargine-treated group (1.00 vs. 0.31/100 person-years) ([@B11]). Studies in patients with long-term duration of diabetes and macrovascular disease. {#s8} ---------------------------------------------------------------------------------- ### Prospective Pioglitazone Clinical Trial in Macrovascular Events (PROactive). {#s9} The PROactive study included 5,238 patients with previous macrovascular disease. The interventional study group patients were given pioglitazone in addition to the previous treatment. This resulted in an on-study difference of HbA~1c~ level by 0.6% (7 mmol/mol) between the pioglitazone-treated and control groups. The patients on pioglitazone had nonsignificantly reduced incidence of a widely defined primary end point (the composite of all-cause mortality, nonfatal MI, stroke, acute coronary syndrome, vascular interventions in the coronary or leg arteries, and amputations above ankle) by 10% (*P* = 0.095). The incidence of a more commonly used (in the other studies) main secondary end point (total mortality, nonfatal MI, and stroke), which was not predefined in the design of the study, was significantly reduced by 16% (*P* = 0.027) in the pioglitazone-treated patients. Symptoms compatible with hypoglycemia arose in 28% on pioglitazone and 20% on placebo (*P* \< 0.0001) ([@B12]). ### ACCORD. {#s10} In the ACCORD trial, 10,251 patients were randomized to receive intensive glucose-lowering treatment aiming for HbA~1c~ \<6% (42 mmol/mol) or standard diabetes treatment targeting HbA~1c~ level in the range 7.0--7.9% (53--63 mmol/mol). No specific treatment was requested in either of the study groups, and multiple drug combinations were allowed to achieve the defined target. In the intensive treatment group a median HbA~1c~ of 6.4% (46 mmol/mol) and in the standard treatment group a median of 7.5% (58 mmol/mol) were achieved, respectively. The study was prematurely stopped after 3.5 years of follow-up in 2008 because of an observed 22% significant increase in all-cause mortality (*P* = 0.04) and 35% increase in cardiovascular mortality (*P* = 0.02) in patients with intensive glycemic control ([@B13]). The primary end point of the study (nonfatal MI, nonfatal stroke, or death from cardiovascular causes) was nonsignificantly reduced in the intensive treatment group by 10% (*P* = 0.16). Significant reduction in the incidence of nonfatal MI by 24% (*P* = 0.004) was observed in the intensive therapy group. The subgroup analysis revealed a significantly more beneficial effect on primary end point reduction in the intensive treatment group in the patients without previous cardiovascular disease and with better diabetes control with HbA~1c~ \<8%, (64 mmol/mol). Hypoglycemia requiring medical assistance was three times more frequent in the intensive therapy group in comparison with standard therapy (10.5 vs. 3.5%, *P* \< 0.001) ([@B13]). ### ADVANCE. {#s11} In the ADVANCE trial, 11,140 patients were randomized to intensive treatment defined as use of gliclazide along with other drugs with a target of HbA~1c~ \<6.5% (48 mmol/mol) or standard treatment. The standard treatment strategy was based on local guidelines. Median follow-up of patients was 5 years. A nonsignificant 6% reduction in the incidence of macrovascular events---nonfatal MI, nonfatal stroke, and death from cardiovascular causes---was observed ([@B14]). In contrast with the ACCORD trial, no significant increase in all-cause or cardiovascular mortality was observed. Subgroup analysis suggested that there might be a more pronounced effect on primary end point reduction in the subgroup of patients with no history of macrovascular disease. However, the test of heterogeneity between the groups with and without history of macrovascular disease was not significant. Severe hypoglycemia was much less frequent than in the ACCORD study. However, it was more common in the intensive control group than in the standard control group (2.7 vs. 1.5%, *P* \< 0.001) ([@B14]). ### Veterans Affairs Diabetes Trial (VADT). {#s12} The VADT had a design similar to those of the ACCORD and ADVANCE trials. A total of 1,791 patients were randomized to intensive diabetes treatment aiming for HbA~1c~ \<6% (42 mmol/mol) and to standard treatment aiming for HbA~1c~ \<9% (75 mmol/mol). The goal for HbA~1c~ between-group difference was 1.5% (17 mmol/mol). The on-treatment median HbA~1c~ was 6.9% (52 mmol/mol) for the intensive-treatment group and 8.4% (68 mmol/mol) for the standard treatment group ([@B15]). The primary end point was any major cardiovascular event (a composite of MI, stroke, death from cardiovascular disease, congestive heart failure, surgery for vascular disease, inoperable coronary disease, and amputation for ischemic gangrene). After the median follow-up of 5.6 years, a nonsignificant reduction in primary end point in the intensive therapy group by 12% (*P* = 0.14) was observed. Incidence of none of the end points included in the primary end point did not differ significantly between the study groups. Similarly to the ACCORD and ADVANCE studies, significantly more episodes of hypoglycemia were reported in the intensive therapy group than in standard therapy (*P* \< 0.001) ([@B15]). In a subgroup of 301 patients, coronary artery calcium (CAC) was measured by computed tomography. Those with low CAC, i.e., less extensive calcified coronary atherosclerosis, had significant benefit from glucose-lowering treatment (HR 0.08 \[95% CI 0.01--0.77\]; *P* = 0.03), while in the patients with CAC \>100 no significant benefit of treatment was observed ([@B16]). Studies in patients with type 2 diabetes and recent MI. {#s13} ------------------------------------------------------- ### Diabetes Mellitus, Insulin Glucose Infusion in Acute Myocardial Infarction 2 (DIGAMI 2). {#s14} The hypothesis that insulin treatment in the postinfarction period prolongs survival of patients was tested in DIGAMI 2 study, which was performed in Scandinavian countries, the Netherlands, and U.K. and included 1,253 patients with type 2 diabetes. Three treatment strategies were compared: Group 1 included patients in whom insulin-glucose infusion was followed by long-term insulin-based regimen. Group 2 included patients who received insulin-glucose infusion followed by standard glucose control, while group 3 had routine metabolic management according to local practice both in hospital and during the posthospitalization period. The median study duration was 2.1 years. After 24 h of hospitalization, blood glucose was significantly reduced in both groups with insulin-glucose infusion to 9.1 mmol/L, while in group 3 it was reduced to 10.0 mmol/L. Hypoglycemia \<3 mmol/L with and without symptoms was more frequent during the initial 24 h in groups 1 and 2 than in group 3. Long-term follow-up data on hypoglycemia incidence were not published in this study. By the end of follow-up, HbA~1c~ levels were reduced in all three groups similarly by 0.5% (6 mmol/mol) to final 6.8% (51 mmol/mol) ([@B17]). Difference in mortality between groups 1 (23.4%) and 2 (22.6%) was the primary end point of the study, and this difference was not statistically significant. The difference in mortality between group 1 and group 3 (19.3%), which was the secondary end point of the study, also was not significant. There were no significant differences in the incidence of reinfarctions or strokes among all three study groups ([@B17]). ### Hyperglycemia and Its Effect After Acute Myocardial Infarction on Cardiovascular Outcomes in Type 2 Diabetes Mellitus (HEART2D). {#s15} HEART2D enrolled 1,115 patients with type 2 diabetes and acute MI. Patients were followed on average 2.7 years. The study was designed to compare two treatment strategies: the first strategy was based on use of basal insulin, while the second strategy aimed to achieve the lowest possible postprandial glucose by use of prandial insulins. Patients in the prandial group experienced 174 events, and patients in the basal group experienced 181 events, with HR of 0.98 (95% CI 0.80--1.21). Secondary analyses included various combinations of cardiovascular outcomes, with hard end points such as cardiovascular death, MI, or stroke being of major interest. The groups did not show any difference with respect to these individual outcomes or combinations of outcomes ([@B18]). The two treatment groups had similar HbA~1c~ throughout the trial: 7.7% (61 mmol/mol) vs. 7.8% (62 mmol/mol). Patients in the prandial group had on average lower postprandial blood glucose, while patients in the basal strategy group had lower fasting/premeal blood glucose. However, the difference in postprandial blood glucose between the groups was smaller (7.8 vs. 8.6 mmol/L; *P \<* 0.01) than anticipated (2.5 mmol/L) in the study design. The incidence of severe hypoglycemia was similar throughout the trial (prandial group vs. basal group 12.9 vs. 9.5%, respectively; *P* = 0.071), while the incidence of nocturnal hypoglycemia was significantly higher in the basal group than in the prandial group (10.6 vs. 6.1%, *P* = 0.007) ([@B18]). ### Bypass Angioplasty Revascularization Investigation in Type 2 Diabetes (BARI 2D). {#s16} The study included 2,368 patients with type 2 diabetes and coronary disease who were assigned to undergo either prompt revascularization with intensive medical therapy or intensive medical therapy alone. Intensive medical therapy was achieved by either insulin sensitization or insulin provision. At 5 years, there was no statistically significant difference in the rate of survival between insulin sensitization and insulin provision groups (88.2 vs. 87.9%). Incidence of severe hypoglycemia was significantly higher in the insulin provision group (9.2 vs. 5.9%, *P* = 0.003) ([@B19]). Meta-analyses of the studies aiming for intensive glycemic control. {#s17} ------------------------------------------------------------------- After publication of the results of three large studies in 2008, several meta-analyses were performed to assess cardiovascular benefits of glucose-lowering treatment. These meta-analyses combined in their majority the results of five trials: UKPDS, PROactive, ACCORD, ADVANCE, and VADT. Although their results slightly differed with respect to evaluated end points, the reduction of HbA~1c~ by an average of 0.9% (10 mmol/mol) was shown to reduce incidence of major cardiovascular events by \~10% and of nonfatal MI by \~15%. No significant effect, either beneficial or deleterious was shown on incidence of stroke and both cardiovascular and total mortality ([@B20]--[@B23]). The only group-level meta-analysis combined data from UKPDS, ACCORD, ADVANCE, and VADT. Subgroup analysis showed that beneficial effect on reduction of major cardiovascular events was shown only in diabetic patients without history of macrovascular disease (HR 0.84 \[0.75--0.94\]; *P* value for group difference of 0.04). Overall, the intensively treated groups had also significantly---approximately 2.5 times---increased risk of severe hypoglycemia ([@B24]). Is hypoglycemia a risk factor for cardiovascular disease? {#s18} ========================================================= The counterintuitive results of the ACCORD study led to several retrospective analyses of data that tried to explain the role of severe hypoglycemia in increased cardiovascular mortality in the intensively treated group. This analysis showed that the participants with at least one episode of severe hypoglycemia requiring assistance had almost twice as high mortality (6.9 vs. 4.1%) than subjects without a hypoglycemic event. Surprisingly, this risk appeared to be higher in the standard group than in the intensive group. Thus, the investigators concluded that previous severe hypoglycemia was not responsible for the difference in mortality rates between the study groups ([@B25]). More recent analysis showed that the frequency of hypoglycemic episodes also did not explain increased mortality in the intensively treated group in ACCORD ([@B26]). Similar analysis performed on the data from the ADVANCE study showed that severe hypoglycemia was associated with significant increase in risks of major macrovascular events (HR 2.88 \[95% CI 2.01--4.12\]), death from cardiovascular disease (2.68 \[1.72--4.19\]), and all-cause mortality (2.69 \[1.97--3.67\]); *P* \< 0.001 for all comparisons) ([@B27]). A meta-regression analysis indicated three significant predictive factors for cardiovascular mortality in intensively treated groups: incidence of severe hypoglycemia, baseline BMI, and the duration of diabetes ([@B20]). Conclusions {#s19} =========== Data from physiological studies showed that severe hypoglycemia or repeated episodes of milder hypoglycemia might lead to sudden arrhythmic death, MI, or stroke predominantly in patients with preexisting macrovascular disease ([@B28]). Epidemiological studies indicated that reduction of HbA~1c~ by 1% (11 mmol/mol) should lead to reduction of major cardiovascular events by \~20%. In reality, in two studies (UGDP and ACCORD) the use of drugs causing hypoglycemia was associated with an increased cardiovascular mortality. There is also substantial evidence, mainly from the observational studies, that mortality of patients on sulfonylureas is higher than of patients on metformin ([@B29],[@B30]). In the other randomized trials mentioned in this review, there was either no effect of intensive glucose lowering on reduction of cardiovascular events or the effect was smaller than expected. The observed reduction in the incidence of cardiovascular events based on meta-analyses of the most important clinical trials was \~10%. Whether this lack of expected effect could be assigned to the increased incidence of hypoglycemia in intensively treated patients is not clear. Other factors such as an increase in body weight, specific side effects of individual antidiabetes drugs, or further nonidentified factors may contribute to hypothesized lack of effect. Based on the knowledge from the above-mentioned studies related to cardiovascular benefit of decreasing hyperglycemia and taking into the account the cardiovascular risk of hypoglycemia, a rational treatment approach was defined in the recent years leading to creation of personalized guidelines. In general, the treatment goal in diabetes is to achieve HbA~1c~ \<7% (53 mmol/mol) ([@B31]). More stringent goals (HbA~1c~ 6.0--6.5%; 42--48 mmol/mol) might be considered in patients with short disease duration, long life expectancy, and no significant cardiovascular disease if this can be achieved without significant hypoglycemia or other adverse effects of treatment. Less stringent HbA~1c~ goals (7.5--8.0%; 58--64 mmol/mol) might be appropriate for patients with a history of severe hypoglycemia, limited life expectancy, advanced complications, and extensive comorbid conditions ([@B31]--[@B37]). This publication is based on the presentations from the 4th World Congress on Controversies to Consensus in Diabetes, Obesity and Hypertension (CODHy). The Congress and the publication of this supplement were made possible in part by unrestricted educational grants from Abbott, AstraZeneca, Boehringer Ingelheim, Bristol-Myers Squibb, Eli Lilly, Ethicon Endo-Surgery, Janssen, Medtronic, Novo Nordisk, Sanofi, and Takeda. See accompanying article, p. S264. This work was supported by research grants from Ministry of Education, Science, Research and Sport of Slovak Republic VEGA 1/0112/11 and VEGA 1/0340/12 and from the Slovakian Agency for Research and Development (APVV-0134-11). I.T. received speaking and/or consulting fees from Boehringer Ingelheim, Eli Lilly, Novo Nordisk, Sanofi, Servier, and Worwag Pharma. No other potential conflicts of interest relevant to this article were reported. I.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Introduction {#s1} ============ As the longest and most architecturally complex cells in the body, neurons face the unique challenge of regulating membrane and protein levels in distal compartments. Neurons have highly elaborate dendritic arbors. These dendrites possess synapses, points of contact where electrochemical transmission of information occurs. Most of the excitatory synapses are situated on dendritic spines, tiny protrusions with a head and neck comprising a geometry that is essential for shaping electrical signals ([@bib55]; [@bib30]; [@bib53]; [@bib26]; [@bib29]; [@bib54]) and providing biochemical compartmentalization ([@bib28]; [@bib6]; [@bib14]). For synapses to function appropriately, the levels of receptor proteins at the postsynaptic density must also be finely tuned. Synapses are often located hundreds of micrometers away from the neuronal cell body. Adding to this spatial problem is the challenge of regulating protein abundance on the membrane in a temporally precise manner, as demanded by fast-acting processes such as synaptic potentiation. Integral membrane proteins destined for the cell surface are canonically thought to be synthesized in the somatic rough endoplasmic reticulum, transported to the Golgi apparatus, and then secreted into the plasma membrane via exocytosis. It is now known that many proteins are translated locally in dendrites, a highly regulated process essential for normal development and plasticity ([@bib49]; [@bib10]; [@bib25]). Endoplasmic reticulum (ER) extends into dendrites, forming a continuous tubular network with regions of varying structural complexity and occasional entry into spines ([@bib48]; [@bib15]; [@bib16]). Together with endosomes, the ER is perfectly positioned to provide a local source of membrane and integral membrane proteins, such as glutamate receptors. However, the Golgi apparatus is absent in most distal dendrites. This puzzling observation has been resolved by recent work demonstrating that dendritic and somatic protein trafficking are highly segregated, and that glutamate receptors are trafficked through a specialized Golgi apparatus-independent pathway from the dendritic ER to the plasma membrane via recycling endosomes ([@bib9]). Structural changes in ER contribute to normal synaptogenesis during development and maturation ([@bib16]). The involvement of this system in activity-induced synaptogenesis is unknown. Long-term potentiation (LTP), the long-lasting enhancement of synaptic strength due to repetitive activity, is thought to underlie learning and memory. This process has been studied extensively in the hippocampus, a key brain region responsible for new memory formation. Insertion of glutamate receptors from an extrasynaptic reserve pool into the postsynaptic compartment is required for LTP in hippocampal area CA1 ([@bib23]). LTP is also accompanied by structural changes in dendritic spines ([@bib8]; [@bib2]). In the young rat hippocampus, LTP produces new dendritic spines ([@bib52]), contrasting with adult rat hippocampus where new spine outgrowth is stalled in favor of synapse enlargement ([@bib7]; [@bib3]). While Golgi apparatus-independent trafficking has not been studied directly in the context of lasting LTP, recycling endosomes (RE) are known to supply AMPA receptors ([@bib40]), and recycling endosome exocytosis is required for spine formation and growth shortly after the induction of LTP ([@bib41]). Expanded knowledge about the involvement of Golgi apparatus-independent pathways in developmental synaptic plasticity could provide new targets for rescuing dysregulated synaptogenesis in cases of profound developmental disorders ([@bib18]). Here, three-dimensional reconstruction from serial section electron microscopy (3DEM) revealed morphological changes in SER and endosomal compartments 2 hr following the induction of LTP. The findings are consistent with the involvement of the Golgi-bypass secretory system in supporting synaptic plasticity in the developing hippocampus. Results {#s2} ======= An acute within-slice experimental protocol ([@bib52]) was used to compare the effects of TBS and control stimulation on subcellular membranous compartments in dendrites. In brief, two stimulating electrodes were positioned \~800 µm apart with a recording electrode halfway in between them in CA1 stratum radiatum of P15 rat hippocampus in one slice from each of two animals ([Figure 1A](#fig1){ref-type="fig"}). Baseline responses were collected from both electrodes. TBS was delivered at one stimulating electrode and control stimulation was delivered at the other stimulating electrode, counterbalanced in position relative to CA3 for each experiment. There was a significant increase in the field excitatory postsynaptic potential (fEPSP) slope immediately after TBS ([Figure 1B,C](#fig1){ref-type="fig"}). Slices were fixed 120 min later. EM image volumes were collected from tissue on a diagonal \~120 µm below and to the side of each stimulating electrode. Segments of spiny dendrites, synapses, and all subcellular membrane compartments were reconstructed in three dimensions (see Materials and methods for details). ![Within-slice experimental design and electrophysiological outcome.\ (**A**) Illustration of an acute slice from a P15 rat hippocampus with a recording electrode (rec.) in the middle of CA1 stratum radiatum between two bipolar stimulating electrodes (S1 and S2). S1 and S2 are separated by 600-800 µm. The two experiments were counterbalanced for which of the two electrodes delivered TBS or control stimulation. Tissue samples collected for 3DEM were located \~120 µm beneath and to the side of the stimulating electrodes. D.G., dentate gyrus; Sub., subiculum. (**B**) Representative waveforms from control (CON, blue) and TBS (LTP, red) sites. Each waveform is the average of the final 10 responses to each stimulating electrode obtained for the last 20 min before delivery of TBS at *time 0* (light color) and for 20 minutes before the end of the experiment at 120 min after TBS (dark color). The stimulus intensity was set at population spike threshold to activate a large fraction of the axons in the field of each stimulating electrode. The positive deflection in the post-TBS waveform at \~3-4 ms reflects synchronous firing of pyramidal cells with LTP. (**C**) Changes in the slope of the field excitatory postsynaptic potential (fEPSP), expressed as a percentage of the average baseline response to test-pulses, were recorded for 20 min before delivery of TBS at *time 0* (red) or control stimulation (blue). Responses were recorded for n=2 slices for 120 min after the first TBS train, then fixed and processed for 3DEM as described in Methods. Error bars are SEM. Adapted from [@bib52] where it was originally published under a CC BY-NC-ND 4.0 license <https://creativecommons.org/licenses/by-nc-nd/4.0/>).\ 10.7554/eLife.46356.003Figure 1---source data 1.Excel spreadsheet containing the raw numbers that generated the graphs and waveforms for these experiments.](elife-46356-fig1){#fig1} Limited entry of SER into dendritic spines {#s2-1} ------------------------------------------ Consistent with previous reports on hippocampal dendrites ([@bib48]; [@bib15]), the SER formed an anastomosing network throughout the dendritic shaft with occasional entry into a subset of dendritic spines ([Figure 2A](#fig2){ref-type="fig"}; see [Figure 2---figure supplement 1](#fig2s1){ref-type="fig"} for all analyzed dendrites reconstructed with SER). While the dendritic spine density more than doubled 2 hr following TBS, a similar increase in the occurrence of SER in spines did not occur ([Figure 2B,C](#fig2){ref-type="fig"}). {#fig2} Spines with small synapses, as measured by the surface area of the postsynaptic density (PSD) (\<0.05 µm^2^), accounted for the LTP-induced increase in spine density ([Figure 2B](#fig2){ref-type="fig"}). This difference was not present at earlier times, and the small spines more than tripled in density by 2 hr post induction of LTP, suggesting that most of this population comprised newly formed spines ([@bib52]). There were no significant effects on SER content in these small spines; not in frequency of spine-localized SER ([Figure 2D](#fig2){ref-type="fig"}), average SER volume ([Figure 2E](#fig2){ref-type="fig"}), nor average SER surface area ([Figure 2F](#fig2){ref-type="fig"}). Since the occurrence of SER did not keep pace with the increase in small spines, the most parsimonious interpretation is that the LTP-induced new spines did not acquire SER. In contrast, while the incidence of SER entry into spines with larger synapses (PSD area ≥0.05 µm^2^) did not change ([Figure 2G](#fig2){ref-type="fig"}), there was however a decrease in the average volume ([Figure 2H](#fig2){ref-type="fig"}) and surface area ([Figure 2I](#fig2){ref-type="fig"}) of SER in these spines. The spine apparatus is an organelle comprising cisterns of SER laminated with electron dense plates that may serve Golgi functions in spines ([@bib24]; [@bib47]; [@bib44]). Consistent with previous observations ([@bib48]; [@bib15]), the spine apparatus appeared in only one dendrite in each condition (data not shown), suggesting that this structure is not central to the activity-induced spinogenesis at this age. Overall, these results reveal that SER entry into dendritic spines is limited and does not scale up with rapid synaptogenesis following LTP at P15. Reduced complexity in shaft SER after LTP {#s2-2} ----------------------------------------- Previous work demonstrated in cultured neurons that local zones of ER complexity produce ER exit sites and compartmentalize membrane proteins near the base of dendritic spines ([@bib16]). Consistent with this finding, SER was inhomogeneously distributed across spiny and aspiny regions of the dendrites in both control and LTP conditions. SER appeared as small circular profiles on some sections, and swollen cisternae with bridging elements on other sections ([Figure 3A](#fig3){ref-type="fig"}). In 3D reconstruction, the primarily tubular structure of SER in aspiny regions and the expanded SER in spiny regions of the dendrite become apparent ([Figure 3B](#fig3){ref-type="fig"}). Following LTP, there was a trend towards reduced shaft SER surface area ([Figure 3C](#fig3){ref-type="fig"}) that reached statistical significance with reduced shaft SER volume ([Figure 3D](#fig3){ref-type="fig"}) when quantified across the total length of the dendritic segments. The SER complexity was estimated by summing the dendritic shaft SER cross-sectional areas in each section, assigning the value to the spiny or aspiny segments, and summing across their independent lengths ([@bib16]). This measure of SER complexity was greater in spiny than aspiny segments under both conditions yet was significantly reduced in both the aspiny and spiny regions following LTP relative to the control condition ([Figure 3E](#fig3){ref-type="fig"}). Considering the prior work, this outcome suggests that SER resources may have contributed to the spine outgrowth by 2 hr following the induction of LTP. {ref-type="fig"}. Aspiny segments consist of two or more sections (\>100 nm) of no spine origins. Spiny segments had at least one spine and were surrounded by aspiny segments. Scale cube is 0.5 µm on each side. (**C**) No statistically significant differences between control and LTP conditions were found in surface area of SER in the dendritic shaft (ANOVA F(~1,12~)=3.8833, p=0.07228). Control n = 8, LTP n = 8 dendrites. (**D**) Volume of dendritic SER network was reduced in the LTP relative to control conditions (ANOVA F(~1,12~)=6.4397, p=0.02605, η^2^ = 0.35). Control n = 8, LTP n = 8 dendrites. (**E**) Summed cross-sectional area of SER tubules and cisterns as a measure of changes in complexity. More SER on spiny than aspiny sections within both control (hnANOVA F(~1,1432~) = 51.672, p\<0.00000, η^2^ = 0.034; spiny n = 493, aspiny n = 955 sections) and LTP conditions (hnANOVA F(~1,324~)=17.535, p=0.00003, η^2^ = 0.013; spiny n = 714, aspiny n = 626 sections). Reduced SER complexity with LTP for both spiny (hnANOVA F(~1,1191~) = 51.745, p\<0.00000, η^2^ = 0.019; Control n = 493, LTP n = 714 sections) and aspiny sections (hnANOVA F(~1,1565~) = 29.991, p\<0.00000, η^2^ = 0.042; Control n = 955, LTP n = 626 sections) relative to control. Bar graphs show mean ± S.E.M. Control (CON, blue) and TBS (LTP, red).\ 10.7554/eLife.46356.008Figure 3---source data 1.Excel spreadsheets containing the raw numbers that generated the graphs in each part of this figure along with the summary of statistics.](elife-46356-fig3){#fig3} Identifying the dendritic trafficking network {#s2-3} --------------------------------------------- Recent work has shown that SER participates in a local, Golgi apparatus-independent secretory trafficking pathway through recycling endosomes in dendrites ([@bib9]). Recycling endosomes have been identified as transferrin receptor-positive membrane compartments in dendrites by immuno-EM ([@bib41]). Other work found that non-SER subcellular components endocytose BSA-conjugated gold particles from the extracellular space ([@bib15]). Together these findings suggest that while these two compartments interact, the SER is not an endocytic structure. Here we considered the possibility that the endosome-based satellite system was also mobilized during LTP. Once the continuous network of SER was reconstructed, the non-SER compartments could be identified as distinct terminating structures. Endosomal subtypes were classified as depicted in [Figure 4A](#fig4){ref-type="fig"} ([@bib15]; [@bib41]; [@bib17]; [@bib51]). Coated pits, coated vesicles, and large vesicles were treated as one category of primary endocytic structures. Sorting complexes and recycling complexes were treated as functionally separate categories. Whorls, free multivesicular bodies, lysosomes, and autophagosomes were classified as degradative structures. Detailed descriptions based on EM morphology follow. ![Identification of endosomal compartments.\ (**A**) Model of the dendritic endosomal pathway. Clathrin-coated pits (CPs) invaginate, becoming clathrin-coated vesicles (CVs) and large vesicles (LVs) after coat shedding. Large vesicles fuse to form tubules, recycling complexes (RCs), and sorting complexes (SCs) with a multivesicular body (MVB). From here, the sorted material may be sent to the plasma membrane via small vesicles (SVs) that pinch off coated tips of tubules. MVBs may serve as exosomes (Exo) or primary lysosomes, that are more darkly stained than exosomes due to the acidic cytomatrix of lysosomes (adapted from [@bib15]). Sample electron micrographs illustrate (**B**) recycling complex (pink arrow) and small vesicles (purple arrow), (**C**) clathrin-coated pit (orange arrow), (**D**) sorting complex (light blue arrows point to multivesicular body (MVB) in the center and tubules around it), (**E**) amorphous vesicle (green arrow), (**F**) lysosome (black arrow), and (**G**) whorl (black arrow). Scale bar in (**G**) is 0.5 µm for all images.\ 10.7554/eLife.46356.016Figure 4---source data 1.Excel spreadsheets containing details of the locations of each object in [Figure 4](#fig4){ref-type="fig"}.](elife-46356-fig4){#fig4} Tubules were cylindrical in shape with a smooth outer membrane, uniform diameter, and a dark, grainy interior. When two or more tubules occurred in proximity, they were categorized as a recycling complex ([Figure 4B](#fig4){ref-type="fig"}; [Figure 4---figure supplement 1](#fig4s1){ref-type="fig"}, [Figure 4---video 1](#fig4video1){ref-type="video"}). Vesicles were distinguished from tubules by examining adjacent sections. Small vesicles (40--60 nm diameter, [Figure 4B](#fig4){ref-type="fig"}; [Figure 4---figure supplement 1](#fig4s1){ref-type="fig"}) and large vesicles (60--95 nm diameter) had a smooth outer membrane and ended within 1--2 sections. Coated pits were omega-shaped invaginations surrounded by clathrin coats ([Figure 4C](#fig4){ref-type="fig"}; [Figure 4---figure supplement 2](#fig4s2){ref-type="fig"}). Coated vesicles had a clathrin coat, were free-floating in the cytoplasm. Occasionally, clathrin-coated buds were observed on the ends of tubules. Multivesicular bodies (MVB) contained a variable number of internal vesicles. When a multivesicular body was found surrounded by tubules, the grouping was categorized as a sorting complex ([Figure 4D](#fig4){ref-type="fig"}; [Figure 4---figure supplement 3](#fig4s3){ref-type="fig"} and [Figure 4---video 2](#fig4video2){ref-type="video"}). Future work might reveal some MVBs to be exosomal compartments ([@bib1]; [@bib43]). Amorphous vesicles had a smooth membrane, an electron-lucent interior, and an irregular shape ([Figure 4E](#fig4){ref-type="fig"}; [Figure 4---figure supplement 4](#fig4s4){ref-type="fig"}). Lysosomes were spherical structures with a homogeneous, electron-dense interior enclosed by one membrane and measuring 70--150 nm in diameter ([Figure 4F](#fig4){ref-type="fig"}; [Figure 4---figure supplement 5](#fig4s5){ref-type="fig"}). Lysosomes were classified as degradative structures. A MVB was considered to be a primary lysosome, namely a degradative structure, when found alone and containing vesicles or pieces of membrane in a dark matrix ([@bib42]; [@bib22]; [@bib38]; [@bib15]). Whorls had multiple convoluted membranes spanning many sections, had a single point of entry into the dendrite, and were classified as degradative structures ([Figure 4G](#fig4){ref-type="fig"}; [Figure 4---figure supplement 6](#fig4s6){ref-type="fig"}; [Figure 4---video 3](#fig4video3){ref-type="video"}). All non-degradative structures were classified as constructive for the quantitative analyses presented next. Constructive endosomes occurred more frequently in spines after LTP {#s2-4} ------------------------------------------------------------------- Endosomal structures occurred in the dendritic shafts and a subset of spines ([Figure 5A](#fig5){ref-type="fig"}; see [Figure 5---figure supplement 1](#fig5s1){ref-type="fig"} for all analyzed dendrites reconstructed with constructive endosomes). Overall, endosomal frequency did not change significantly across conditions within dendritic shafts ([Figure 5B](#fig5){ref-type="fig"}); however, when analyzed by subtype the occurrence of recycling complexes was increased ([Figure 5B](#fig5){ref-type="fig"}). Similarly, there was no significant effect of LTP relative to the control condition on endosomal distribution to aspiny or spiny dendritic segments. ![Increased occurrence of endosomes in small spines after LTP.\ (**A**) Sample serial EM sections and representative 3D reconstructed dendrites illustrate the distribution of endosomal compartments from control and LTP conditions. Dendrites are yellow, synapses are red, and color-coded arrows point to endosome-containing spines. The color-coded key in the lower left corner indicates amorphous vesicles (AV), recycling complexes (RC), coated pits (CP), coated vesicles (CV), large vesicles (LV), sorting complexes (SC), small vesicles (SV) and degradative structures (DEG); these abbreviations apply also to the graphs. Vesicles are represented as 100 nm spheres (AV, CP, CV, LV, and SV). The other structures (RC, SC, DEG) are reconstructed in 3D to scale. (**B**) Endosomal structures in dendritic shafts (\#/µm) with relative distributions to aspiny and spiny segments in control (CON) and LTP conditions. Overall, shaft endosomes (hnANOVA F(~1,293~)=0.93104, p=0.33539), degradative structures (hnANOVA F(~1,293~)=0.47789, p=0.48993) or constructive endosomal compartments (Constr. = all minus degradative; hnANOVA F(~1,293~)=0.62167, p=0.43107) did not differ between LTP and control conditions or segment locations. Recycling complexes (RC) were greater in the LTP than control dendritic shafts (hnANOVA F(~1,293~)=6.4920, p=0.01135, η^2^ = 0.022), but no significant differences occurred in the other categories: amorphous vesicles (hnANOVA F(~1,293~)=1.5092, p=0.22025); small vesicles (hnANOVA F(~1,\ 293~)=1.1699, p=28031); coated pits, coated vesicles, and large vesicles (hnANOVA F(~1,293~)=0.89152, p=0.34584); and sorting complexes (hnANOVA F(~1,293~)=0.45286, p=0.50151). (For control (CON) n = 151 aspiny + spiny segments and for LTP n = 158 aspiny + spiny segments.) (**C**) More dendritic spines contained endosomes along the dendrites in the LTP than the control condition (ANOVA F(~1,12~)=18.047, p=0.00113, η^2^ = 0.60), an effect that was carried by spines with PSD areas less than 0.05 µm^2^ (ANOVA F(~1,12~)=23.642, p=0.00039, η^2^ = 0.66) but not in spines with PSD area [\>]{.ul}0.05 µm^2^ (ANOVA F(~1,12~)=0.84714, p=0.37550). (**D**) Stability in percentage of spines containing endosomes following TBS (ANOVA F(~1,12~)=.72158, p=0.41225). (**E**) Among spines with PSD area less than 0.05 µm^2^, the increase in occupancy of endosomes was due to more with coated pits, coated vesicles, and large vesicles (ANOVA F(~1,12~)=4.94433, p=0.046140, η^2^ = 0.29), recycling complexes (ANOVA F(~1,12~)=11.009, p=0.00613, η^2^ = 0.48), and more with small vesicles (ANOVA F(~1,12~)=5.2575, p=0.04072, η^2^ = 0.30). No significant changes in spine occupancy occurred for amorphous vesicles (ANOVA F(~1,12~)=1, p=0.33705), sorting complexes (ANOVA F(~1,12~)=1, p=0.33705), or degradative structures (ANOVA F(~1,12~)=0.46689, p=0.5074). Bar graphs show mean ± S.E.M. (For **C--E**), Control (CON, n = 8 full dendrite reconstructions) and LTP (n = 8 full dendrite reconstructions).\ 10.7554/eLife.46356.023Figure 5---source data 1.Excel spreadsheets containing the raw numbers that generated the graphs in each part of this figure along with the summary of statistics.](elife-46356-fig5){#fig5} In contrast, there was a substantial increase in the occurrence of dendritic spines with endosomes, an effect that was confined to spines with small PSD areas (\<0.05 µm^2^, [Figure 5A,C,D](#fig5){ref-type="fig"}). Furthermore, this increase in spines involved constructive endocytic compartments (including coated pits, coated vesicles, large vesicles, recycling complexes, and small vesicles), with no significant effects on the rare occurrence of spines with amorphous vesicles, sorting complexes, or degradative structures ([Figure 5E](#fig5){ref-type="fig"}; see [Figure 5---figure supplement 2](#fig5s2){ref-type="fig"} for all analyzed dendrites reconstructed with degradative endosomes). These data suggest that the non-canonical secretory trafficking contributes locally in support of spines added 2 hr following the induction of LTP at P15. Discussion {#s3} ========== These results provide several advances towards understanding mechanisms of enduring LTP in the developing hippocampus. A population of spines that increased in density by 2 hr after the induction of LTP relative to control stimulation had small synapses and mostly lacked SER. Spines with larger synapses were unchanged in density and retained SER in similar proportions under both conditions. The distribution of SER along the dendritic shaft was non-uniform, with greater abundance and complexity in spiny than aspiny regions under control and LTP conditions. However, the shaft SER was reduced in volume and complexity after LTP. In conjunction, there was an LTP-related increase in endosomal structures confined to the small, presumably newly formed spines. This elevation involved constructive endocytic, recycling, and exocytic structures in the small spines. In contrast, no differences occurred between control and LTP conditions in the frequency or locations of the degradative structures. These data are from two animals using the within-slice paradigm to control for between-slice variance. The stimulating electrodes were positioned such that the sampling of dendrites was counter-balanced with respect to position from the CA3 axons that were stimulated. Dendrites were matched for caliber to avoid the confound that thicker dendrites have more spines per micron. Future work will be needed to determine whether these findings generalize beyond the medium caliber dendrites and position within the dendritic arbor, and to other slice and LTP induction paradigms. The findings suggest a model in which local Golgi apparatus-independent secretory trafficking adds and prepares new spines for subsequent plasticity ([Figure 6](#fig6){ref-type="fig"}). TBS induces LTP via the insertion of glutamate receptors from recycling endosomes and lateral diffusion ([@bib37]; [@bib13]). By 5 min (early LTP), there is a temporary swelling of spines and recycling endosomes are recruited into the spines; however the PSD is not enlarged at this early timepoint suggesting receptors are inserted into pre-existing slots ([@bib40]; [@bib33]; [@bib41]; [@bib7]; [@bib35]; [@bib52]). By two hours (late LTP), shaft SER decreases as it contributes membrane and proteins via ER exit sites to the formation of new spines, which have silent synapses lacking AMPAR. Constructive endosomes are recruited to the new spines and provide a reserve pool of receptors that are in position for rapid insertion of AMPAR upon subsequent potentiation. {#fig6} Effects of LTP on SER and spines {#s3-1} -------------------------------- Previous work has shown that integral membrane proteins rapidly diffuse throughout tubular SER and become confined in regions where the SER is more complex, having branches between tubules and distended cisternae ([@bib16]). As spine density increases across development so too does SER complexity, leading to decreased mobility of ER membrane cargo with age. SER complexity was measured as the summed cross-sectional area to capture the local variation. SER and spine density were positively correlated where more dendritic spines clustered locally. Using the same methods, we found SER volume and complexity were greater in spiny than aspiny regions and were reduced in conjunction with TBS-induced spinogenesis along these P15 dendrites. This result suggests that the membrane lost from SER in the shaft could have been used to build new spines after LTP. In adult hippocampal area CA1, LTP produced synapse enlargement at the expense of new spine outgrowth ([@bib7]; [@bib3]; [@bib12]). SER is a limited resource, entering only 10--20% of hippocampal dendritic spines ([@bib48]; [@bib15]; [@bib12]). Spines containing SER are larger than those without SER, and in adults 2 hr after induction of LTP the SER was elaborated into a spine apparatus in spines with enlarged synapses ([@bib12]). Spines clustered around the enlarged spines and local shaft SER remained complex, whereas distant clusters had fewer spines than control dendrites and lost local shaft SER. These findings suggest that mature dendrites support a maximum amount of synaptic input and strengthening of some synapses uses resources that would otherwise be targeted to support spine outgrowth, even in adults. At P15, CA1 dendrites have less than one-third mature synaptic density, which will nearly reach adult levels in another week ([@bib32]). These findings suggest that P15 may well be an age when synaptogenesis predominates over the growth of existing synapses, which may account for the spinogenesis response to LTP. At P15, SER was also restricted to a small number of spines, and like adults the few spines that had SER were larger than those without SER ([@bib12]). However, at P15, most of the small, presumably newly formed spines did not contain SER. Similar to adults, shaft SER was reduced in complexity and volume, but at P15 the redistribution was apparently targeted only to the plasma surface, rather than elaboration of the spine apparatus and growth of potentiated spines, as in adults ([@bib12]). These findings suggest that synapse growth occurs where synapses had already been activated or previously potentiated, and few of those existed at P15 prior to the induction of LTP. Thus, resources were available for spine outgrowth to dominate. Future work is needed to learn when the shaft SER recovers, and when this recovery becomes necessary for additional synaptogenesis or synapse enlargement as the animals mature. SER regulates intracellular calcium ion concentration ([@bib50]). Regulation of postsynaptic calcium levels is necessary for the expression of synaptic plasticity ([@bib34]; [@bib36]), hence the presence of SER could be important for signaling cascades associated with LTP and stabilization of AMPA receptors at potentiated synapses ([@bib4]). Consistent with this, spines with larger synapses tended to contain SER, and were maintained at stable density post-TBS. However, it might be of some concern that calcium regulation is disrupted by the reduction in SER volume in both adult and P15 hippocampal dendritic shafts by 2 hr after induction of LTP. The reduction in SER volume was by no means complete, and instead likely reflects the multiple roles of SER in membrane and protein trafficking in addition to the regulation of calcium. That a substantial amount of shaft SER remains well after the induction of LTP, supports the hypothesis that SER is a dynamically regulated resource at both ages. Role of satellite secretory system in synaptogenesis and subsequent plasticity {#s3-2} ------------------------------------------------------------------------------ Dendrites support local processing and secretory trafficking of newly synthesized cargo independent of a Golgi apparatus ([@bib9]). Secretory cargo passes from the ER to ER-Golgi intermediate compartments (ERGICs) into recycling endosomes en route to the plasma membrane. While molecular understanding of this pathway has been achieved, the spatial organization of the responsible organelles has been nebulous. Recycling endosomes were seen about 25% of spines on cultured neurons that also contained synaptopodin, a marker for the ER-derived spine apparatus ([@bib9]). This finding suggested that recycling endosomes might receive newly synthesized cargo directly from a spine apparatus. However, at P15, only one spine apparatus was found in each of the control and TBS conditions, suggesting that recycling endosomes derive from alternate recycling organelles in the dendritic shaft. Previously, this satellite secretory system has only been studied in neurons under baseline conditions in culture. Here, we provide the first evidence that this specialized secretory system locally supports spine formation during plasticity. Synaptogenesis at P15 does not precede the expression of LTP, as evidenced by a lack of added spines at 5 min following TBS ([@bib52]). The magnitude of potentiation following the initial TBS is constant across time, so the added small spines at 2 hr after the induction of LTP are likely to be functionally silent. Hence, the newly added spines could be viewed as a form of heterosynaptic plasticity that readies the neurons for subsequent potentiation. In support of this hypothesis, a second bout of TBS delivered 90 min after the first TBS produces substantial additional potentiation at this age ([@bib11]). Many of the added small spines contained endosomes at 2 hr after the initial induction of LTP. These endosomes might be interpreted as a heterosynaptic mechanism for long-term depression, namely internalizing receptors from pre-existing spines. However, since most of the endosomal structures occupied the added small spines and were of a constructive nature, they could instead be available to convert the new silent synapses to active synapses after a later bout of potentiation. Such a mechanism would support the establishment of functional circuits as the young animals learn and begin to form memories. Materials and methods {#s4} ===================== -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Reagent type\ Designation Source or reference Identifiers Additional\ (species) or\ information resource ----------------------------------------------------- ---------------------------------- ------------------------------ -------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------- Strain, strain background (Rattus norvegicus, male) Long-Evans Rat Charles River Charles River strain\# 006; RRID:[RGD_2308852](https://scicrunch.org/resolver/RGD_2308852) Chemical compound, drug Potassium ferrocyanide Sigma-Aldrich Cat\# P3289 Chemical compound, drug Osmium tetroxide Electron Microscopy Sciences Cat\# 19190 Chemical compound, drug Uranyl acetate Electron Microscopy Sciences Cat\# 22400 Chemical compound, drug LX-112 embedding kit Ladd Research Industries Cat\# 21210 Chemical compound, drug Lead nitrate Ladd Research Industries Cat\# 23603 Chemical compound, drug Pioloform F Ted Pella Cat\# 19244 Software, algorithm Igor Pro 4 WaveMetrics <https://www.wavemetrics.net/> Software, algorithm Reconstruct [@bib20] Executable and manual: <http://synapseweb.clm.utexas.edu/software-0> Source at:<https://github.com/orgs/SynapseWeb/teams/reconstruct-developers> Software, algorithm STATISTICA 13 Academic Tibco <https://onthehub.com//statistica/> Other Tissue slicer Stoelting Cat \# 51425 Other Vibratome Leica Biosystems VT1000S Other Ultramicrotome Leica Biosystems UC6 Used with a Diatome Ultra35 knife Other SynapTek Grids Ted Pella Cat\# 4514 or 4516 Other Diffraction grating replica Electron Microscopy Sciences Cat\# 80051 Other Transmission electron microscope JEOL JEM-1230 Other Harris Lab wiki Harris Lab <https://wikis.utexas.edu/display/khlab/> This wiki site hosts experimental methods used for this paper and updates. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Animals {#s4-1} ------- Hippocampal slices (400 µm) were rapidly prepared from P15 male Long-Evans rats (RRID:[RGD_2308852](https://scicrunch.org/resolver/RGD_2308852), n \> 100, including the initial test experiments and slices used in prior work for the 5 min and 30 min time points; [@bib52]). For the 2 hr time point reported here, one slice each from two rats met the strict physiology and ultrastructural criteria for inclusion as outlined below. All procedures were approved by the University of Texas at Austin Institutional Animal Care and Use Committee and were followed in compliance with NIH requirements for humane animal care and use (Protocol number 06062801). All rats were of comparable features indicative of health at the time they were taken for experimentation. Preparation and recording from acute hippocampal slices {#s4-2} ------------------------------------------------------- Rats were decapitated and the left hippocampus was removed and sliced into 400 µm thick slices from the middle third of the hippocampus at a 70° traverse to the long axis using a tissue chopper (Stoelting, Wood Dale, IL). Hippocampal slices were kept room temperature (\~25°C) in artificial cerebrospinal fluid (ACSF) bubbled with 95% O~2~/5% CO~2~ ([@bib5]). ACSF consisted of 116.4 mM NaCl, 5.4 mM KCl, 3.2 mM CaCl~2~, 1.6 mM MgSO~4~, 26.2 NaHCO~3~, 1.0 mM NaH~2~PO~4~, and 10 mM D-glucose at pH 7.4. Slices were immediately transferred to nets on top of wells containing ACSF at the interface of humidified O~2~ (95%) and CO~2~ (5%). Dissection and slice preparation took less than 5 min. The slices were kept at 32°C for approximately 3 hr *in vitro* prior to recording ([@bib19]). Two concentric bipolar stimulating electrodes (100 µm diameter, Fred Haer, Brunswick, ME) were positioned \~300--400 µm on either side of a single glass extracellular recording electrode in the middle of stratum radiatum for independent activation of subpopulations of synapses ([@bib46]; [@bib39]; [@bib7]). The recording electrode was a glass micropipette filled with 120 µM NaCl. After initial recovery period, stable baseline recordings were obtained from both sites for a minimum of 40 min. Extracellular field potentials (fEPSPs) were obtained with custom designed stimulation data acquisition protocols using Igor software (WaveMetrics, Lake Oswego, OR). fEPSPs were estimated by linear regression over 400 µs along maximal initial slope (mV/ms) of test pulses of 100 µs constant, biphasic current. Stimulus intensity was set to evoke 1/2 maximum fEPSP slope based on a stimulus/response curve for each experiment and was held constant for the duration of the experiment. TBS-LTP paradigm {#s4-3} ---------------- Theta burst stimulation (TBS) was used to induce LTP. TBS was administered by one stimulating electrode as one episode of eight trains 30 s apart, each train consisting of 10 bursts at 5 Hz of 4 pulses at 100 Hz. The control stimulating electrode delivered one pulse every 2 min. Stimulations were alternated between the TBS-LTP and the control electrode once every two minutes with a 30 s interval between electrodes. In order to counterbalance across experiments, control and TBS-LTP electrode positions were interchanged between the CA3 and subicular side of the recording electrode ([Figure 1A](#fig1){ref-type="fig"}). Physiological responses were monitored for 120 min after the first train of TBS ([Figure 1B,C](#fig1){ref-type="fig"}) and then rapidly fixed, as described below. Fixation and processing for 3DEM {#s4-4} -------------------------------- One slice from each animal was fixed and processed for electron microscopy 2 hr after induction of LTP. Only slices with good physiology were used, defined as a gradually inclining I/O curve in response to incremental increases in stimulus intensity for both stimulating electrodes, a stable baseline response at both stimulating electrodes unchanged at the control site post LTP-induction, and a significant increase in fEPSP slope that was immediately induced by TBS and was sustained for the duration of the experiment. Within a few seconds of the experiment's end, electrodes were removed and slices were immersed in fixative (6% glutaraldehyde and 2% paraformaldehyde in 100 mM cacodylate buffer with 2 mM CaCl~2~ and 4 mM MgSO~4~), microwaved at full power (700 W microwave oven) for 10 s to enhance penetration of fixative ([@bib31]), stored in the fixative overnight at room temperature, rinsed three times for 10 min in 100 mM cacodylate buffer, and embedded in 7% low melting temperature agarose. They were then trimmed, leaving only the CA1 region that contained the two stimulating electrodes. They were mounted in agarose and vibra-sliced into 70 µm thick slices (VT1000S, Leica, Nusslock, Germany). Vibra-slices were kept in a 24-well tissue culture dish and examined under a dissecting microscope to locate the vibra-slices containing indentations from the stimulating electrodes. The vibra-slices with the indentations due to the stimulating electrodes and two vibra-slices on either side of these indentations were collected and processed in 1% OsO~4~ and 1.5% potassium ferrocyanide in 0.1M cacodylate buffer for 5--10 min, rinsed five times in buffer, immersed in 1% OsO~4~ and microwaved (1 min on, 1 min off, 1 min on) twice with cooling to 20°C in between, and rinsed five times in buffer for two minutes and then twice in water. They were then dehydrated in ascending concentrations of ethanol (50%, 70%, 90%, and 100%) with 1--1.5% uranyl acetate and microwaved for 40 s at each concentration. Finally, slices were transferred through room temperature propylene oxide, embedded in LX-112 (Ladd Research, Williston, VT), and cured for 48 hr at 60°C in an oven ([@bib27]). Slices with high-quality preservation, defined as dendrites with evenly spaced microtubules, well-defined mitochondrial cristae, and well-defined PSDs that were not thickened or displaced from the membrane, were selected for analysis. The region of interest was selected from middle of the CA1 stratum radiatum and 120--150 µm beneath the air surface, then cut into 150--200 serial sections. The sections were mounted on Pioloform-coated slot grids (Synaptek, Ted Pella, Redding, CA). The sections were counterstained with saturated ethanolic uranyl acetate, then Reynolds lead citrate ([@bib45]) for five minutes each, and then imaged with a JEOL JEM-1230 transmission electron microscope with a Gatan digital camera at 5000X magnification along with a diffraction grating replica for later calibration (0.463 µm cross line EMS, Hatfield, PA or Ted Pella). Imaging was conducted blind to condition. 3D reconstructions and measurements of dendrites {#s4-5} ------------------------------------------------ A random five-letter code was assigned to each series of images for the experimenter to be blind to the original experimental conditions during data collection. Reconstruct software (freely available at <http://www.synapseweb.clm.utexas.edu>; [@bib20]) was used to calibrate pixel size and section thickness, align sections, and trace dendrites, SER, endosomes, and PSD. The diffraction grating replica imaged with each series was used to calibrate pixel size. Cylindrical diameters method was used to calculate section thickness ([@bib21]). Calculated section thicknesses ranged from 46 to 63 nm. Dendrites selected for analysis were chosen based on their orientation (cross-sectioned or radial oblique) and matched for diameter. Microtubule count was used as a measure of dendritic caliber (6--22 MTs) as this range under control condition showed no differences in spine density. All dendrites chosen for the analysis were completely reconstructed. The z-trace tool in Reconstruct was used to measure dendrite lengths across serial sections of each analyzed dendrite. Four dendrites were sampled from each condition (control or TBS-LTP) in each animal, resulting in a total of 16 dendritic segments from four EM series. Each analyzed dendritic segment traversed over 100 serial sections. In total, 173 µm of dendritic length was sampled. Identification and quantification of subcellular compartments {#s4-6} ------------------------------------------------------------- The process of tracing, reviewing, and curating dendrites, synapses, and subcellular objects was confirmed by three scientists (Kulik, Watson, and Harris) and conducted blind as to condition. On the rare occasions where there was disagreement, we met to arrive at a consensus based on the 3D structures; hence all objects were eventually provided a confirmed identification as outlined below. Dendrites and PSDs were traced and dimensions were quantified as previously described ([@bib52]). SER was identified on the basis of its characteristic morphology of tubules with dark staining membrane, occasional flattened cisternal distensions with a wavy membrane and clear lumen, and continuity across sections within each reconstructed dendrite. Once SER was completely traced, the remaining membrane-bound intracellular compartments were traced and their identity was assigned on the basis of morphology, as described in Results. Criteria used to differentiate endosomes included: 1) Continuity across sections: vesicles appear on single sections; tubules span multiple sections and then terminate; SER is continuous across sections throughout the entire dendrite; MVBs and tubules form a sorting complex when found on continuous sections; 2) Geometry: small and large vesicles are spherical, while amorphous vesicles are not; tubules have a uniform diameter across sections; SER has a highly variable profile across sections; MVBs have an unmistakable outer membrane surrounding multiple internal vesicles, and MVBs have tubules attached when part of a sorting complex; 3) Dimensions: small vesicles are 40--60 nm in diameter; large vesicles are 60--95 nm in diameter; 4) Electron density: amorphous vesicles and SER have a clear lumen; tubules and MVBs have a dark, grainy interior; lysosomes have a very dark, electron-dense interior. Only spines that were entirely contained within the series were used for the analyses of subcellular compartments. In this way, we avoided possible undercounting of compartments that may have entered a portion of an incomplete spine outside the series. Spines were considered to contain a subcellular structure when it entered the head or neck of the spine, but not if it was only at the base of a spine. The frequency of occurrence was calculated as the total number of occurrences of objects divided by the length of dendrite in microns. The 3D visualization of dendrites and subcellular structures was achieved with Reconstruct. The 3D reconstructions from serial EMs allowed us to calculate volumes and surface areas of objects and to assess SER and endosome distribution within dendrites. Statistical analyses {#s4-7} -------------------- The statistical package STATISTICA (version 13.3; TIBCO, Palo Alto, CA) was used for all analyses. There were two conditions represented in each animal: control (CON), and LTP at 120 min following TBS. In this study, eight control dendrites (four from each animal) and 8 LTP dendrites (four from each animal) were analyzed. One-way ANOVAs were run on all density (\#/µm) data involving one measurement per dendrite, in which case n = number of dendrites. Hierarchical nested analysis of variance (hnANOVAs) were run when many measures were obtained from each dendrite (e.g. SER volume per spine, PSD area etc.). In this case, n = total spines, as each spine was considered separately. In hnANOVAs dendrite was nested in condition and experiment, and experiment nested in condition to account for inter-experiment variability. Results of the one-way ANOVAs and hnANOVAs are reported as (F~(df\ condition,\ df\ observations)~=F value, P value) where df is degrees of freedom presented for condition and error. In hnANOVAs degrees of freedom are further decreased by one for each dendrite. Absolute p values are reported for each test. Statistical tests are reported in the figure legends. Data in bar graphs is plotted as mean ± SEM. Significant P values are indicated by asterisks above the bars. Significance was set at p\<0.05. The effect sizes for significant differences are also presented in the figure legends as η^2^ (which was determined as SS~condition~/SS~(condition\ +\ error)~, where SS = sum of squares determined in Statistica for each analysis). We have provided the raw images, Reconstruct trace files, and analytical tables in the public domain at Texas Data Repository: DOI: <https://doi.org/10.18738/T8/5TX9YA>. Caveats {#s4-8} ------- One might be concerned that these data arise from two animals. We note that these experiments are within-slice experiments, namely the control and LTP sites are from independent locations within the same slice from two different animals. Based on numerous preliminary experiments, we found that this approach greatly reduces variation due to slice preparation, *in vitro* conditions, and subsequent processing for electron microscopy when comparing the control and LTP outcomes. We also note that enhanced statistical power came from the large number of synapses and spines tested using the hierarchical nested ANOVA design with dendrite nested in condition by animal ([Figures 2E,F,H,I](#fig2){ref-type="fig"} and [3E](#fig3){ref-type="fig"}). In this way, degrees of freedom are adjusted for animal and dendrites, and outcomes are tested to ensure that no one dendrite or animal carried the findings. In addition, we had power to detect changes using multifactor ANOVAs for measurements that involved one measure per dendrite (\#/µm listed on the y axes of [Figures 2B--D, G](#fig2){ref-type="fig"}, [3C--D](#fig3){ref-type="fig"} and [5B--E](#fig5){ref-type="fig"}). Given the extremely time-consuming nature of the imaging and 3DEM analysis, additional animals and slices were not included. Source data files (Named Figures 1-5--source data 1 in each legend) {#s4-9} ------------------------------------------------------------------- There is one source data file for each of [Figures 1](#fig1){ref-type="fig"}--[5](#fig5){ref-type="fig"} that contains Excel spreadsheets with the object locations in the Reconstruct trace files (provided in the public domain) for EMs. These files also contain the raw numbers that generated graphs in each part of each figure along with the summary of statistics. Funding Information =================== This paper was supported by the following grants: - http://dx.doi.org/10.13039/100000002National Institutes of Health NS21184 to Kristen M Harris. - http://dx.doi.org/10.13039/100000002National Institutes of Health R01NS074644 to Kristen M Harris. - http://dx.doi.org/10.13039/100000002National Institutes of Health R01MH095980 to Kristen M Harris. - http://dx.doi.org/10.13039/100000002National Institutes of Health R01MH104319 to Kristen M Harris. - http://dx.doi.org/10.13039/100000001National Science Foundation NeuroNex 1707356 to Kristen M Harris. - http://dx.doi.org/10.13039/100000002National Institutes of Health F32 MH096459 to Deborah J Watson. We thank Robert Smith and Elizabeth Perry for technical support in the ultramicrotomy; Heather Smith and Patrick Parker for their contributions in some of the dendrite analyses; and Patrick Parker for editorial comments. We thank Graeme W Davis for his support of YDK during the writing of this manuscript. Additional information {#s5} ====================== No competing interests declared. Conceptualization, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Writing---original draft, Writing---review and editing. Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing---review and editing. Conceptualization, Formal analysis, Validation, Investigation, Methodology, Writing---review and editing. Data curation, Validation, Methodology, Writing---review and editing. Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing---original draft, Project administration, Writing---review and editing. Animal experimentation: All procedures were approved by the University of Texas at Austin Institutional Animal Care and Use Committee and were in compliance with NIH requirements for humane animal care and use. Protocol number (06062801). All rats were of comparable features indicative of health at the time they were taken for experimentation. Additional files {#s6} ================ 10.7554/eLife.46356.025 Data availability {#s7} ================= The relevant image series files and numerical data have been provided. In addition, the program Reconstruct is freely available from <http://synapseweb.clm.utexas.edu/>, and can be used to image and visualize the raw trace files. We have provided the raw images, Reconstruct trace files, and analytical tables in the public domain at Texas Data Repository: DOI: <https://doi.org/10.18738/T8/5TX9YA>. The following dataset was generated: KulikYDWatsonDJCaoGKuwajimaMKristenM Harris2019Raw images, Reconstruct trace files, and analytical tablesTexas Data Repository10.18738/T8/5TX9YA 10.7554/eLife.46356.029 Decision letter Helmstaedter Moritz Reviewing Editor Max Planck Institute for Brain Research Germany In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included. Thank you for submitting your article \"Structural plasticity of dendritic secretory compartments during LTP-induced synaptogenesis\" for consideration by *eLife*. Your article has been reviewed by Eve Marder as the Senior Editor, a Reviewing Editor, and three reviewers. The reviewers have opted to remain anonymous. The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission. The reviewers agreed that the manuscript contains a set of very valuable data about structural changes in the setup of spines undergoing plasticity in young animals, with a particular focus on the smooth endoplasmatic reticulum and endocytic compartments. The reviewers however raised two key concerns, which we request to be addressed in the revised manuscript: \- The reviewers are concerned about the underlying statistics that give rise to the conclusions drawn. In particular, the number of animals, slices, and reconstructed dendrites used are both unclear from the manuscript and suspected to be very small. Here, the reviewers request clarification and (if the low n is in fact confirmed) addition of further data to exclude inter-individual variability as a source of the observed effects. \- The methodological description is considered improvable by two reviewers who request better definition of concepts such as \"new spine\". For data annotations that may be considered subjective, the independent annotation by multiple experts may offer a way to provide classifications with confidence intervals. Please see the detailed comments below for the context of these two key requests. All other comments can be treated as recommendations for the revision. *Reviewer \#1:* This manuscript by Kulik et al. describes detailed ultra-structural analysis of dendritic spines after LTP using 3D reconstruction of serial section electron microscopy. They used relatively young animals, in which stage LTP induces spinogenesis. In particular, they focused on intracellular membrane structures including smooth endoplasmic reticulum (SER) and various endosomes. They observed that spine density almost doubled 2 hours after LTP induction. Several differences in intracellular membrane structure in LTP area and control area are described. First, they observed less fraction of small spines after LTP contain SER, while large spines have reduced SER content. Next, they observed SER in dendritic shafts decreases their volume after LTP. Finally, spines containing various endosomes increased after LTP. Overall, the manuscript contains potentially important data describing anatomical changes induced by LTP in young animals. However, some of their interpretation, for example their definition of \"new spines\" and their model of SER/endosome structure at the early time point, appears to be not well validated by the measurement. More serious issue is that it is not clear how slice-to-slice and animal-to-animal variation can be taken into account. It appears that only two slices (not clear if they are from the same animal or different animals) are used in their analysis. The conclusion may not be generalizable to all animals. Essential revisions: 1\) Figure 1: The title \"New small spines induced by TBS contained no SER while existing large spines had reduced SER content\" sounds to be misleading. It is not described how they conclude that these are \"new\" spines. Also, I don\'t see any evidence suggesting \"no SER\" in \"new spines\" from the figure, as there are some SER in small spines. 2\) Figure 5: \"Increased endosomal activity\" may be misleading, as they are not measuring the activity but the distribution of endosomes. 3\) Statistics: It appears that the entire data is based on only two slices (Figure 1:. Is it from two animals?) Perhaps any statics would not work well with n=2. As LTP varies fairly a lot from slice to slice and from animal to animal, this raises a question of whether the conclusion can be generalizable. 4\) Figure 6: It is not clear how they come up with the model of SER structure and endosome structure at \"early LTP\", as the measurement only at 2 hours. *Reviewer \#2:* In this study, the authors conducted 3D EM reconstruction of CA1 dendrites after TBS LTP and concentrated on measuring organelles such as ER and endocytic compartments. Using this method, the authors make a few new observations: 1\) New spines that are formed after LTP do not contain SER and existing spines lose some SER. 2\) After LTP SER in dendritic shafts is reduced. 3\) After LTP, increased endocytic compartments were observed in small spines. Based on these observations the authors suggest that new spines observed after LTP are supported by recycling endosomes rather than SER. The apparent increase in endocytic structures after LTP is intriguing, although I have some concerns on how these structures are classified. Essential revisions: 1\) The identification of the specific endocytic structures in Figure 4 and Figure 5 relies solely on morphology, and based on a limited set of images it is unclear how reliable the distinction can be made between recycling endosomes and endosomes that may be heading to a lysosome/degradation pathway. The authors list a few papers as explanation of how these structures were classified, but the description of how the dendrites were annotated is vague. The authors need to provide a lot more detail on exactly how these structures were classified and how reliable is the distinction between similar-looking endocytic structures. Based on the current data presented, the conclusion that small spines are mostly supported by recycling endosomes is not strongly supported. 2\) The 3D reconstruction of one time point after LTP makes it hard to infer dynamics from a static snapshot. While the reconstruction is extremely consuming, the authors need to discuss caveats and alternative interpretations of the data in the discussion section. For example, can the authors rule out heterosynaptic LTD or other homeostatic mechanisms that may rely on endocytosis of receptors? *Reviewer \#3:* The paper by Kulik et al., examines the changes to the ultrastructure within dendrites of the P15 hippocampus after a stimulation protocol that produces long term potentiation. This piece of work is complementary to a number of other studies that have looked at similar changes in the adult and for this reason the paper it will be of some interest. The study uses serial section EM to make detailed reconstructions of segments of dendrites that included the intracellular features such as smooth endoplasmic reticulum and endosomes. The analysis looked at dendritic spines, and the immediate piece of parent dendrite, as well as stretches of dendrite that bore no spines and compared their contents. The stimulation protocol produced a significant increase in the number of dendritic spines, but not of spines that contained SER. The larger spines, however, considered as being the more permanent ones, had a reduction in their amount of SER, although still present. The authors conclude that these changes indicate that SER is not an organelle that is pivotal to the production of new spines induced by LTP. Measurements of the amount of SER in the parent dendrite show a decrease. These changes led the authors to explore the intracellular ultrastructure in more detail and analyze the presence of different organelles. These they classed as either belonging to a degradative class such as lysosomes and multivesicular bodies, or constructive such as coated pits, vesicles, and endosomes. This analysis shows that the constructive elements were present only in the new (small) spines. The study is well explained and illustrated. The quality of the electron microscopy is high and typical of this laboratory with considerable experience in this field of neuronal plasticity. The figures are well laid out and the descriptions are clear, as are the results and discussion parts. The Materials and methods section fully describes the procedures used, including the details of the analysis. However, it would be useful if there is a clear explanation of how the different dendrites were sampled and grouped. It appears that two slices were used from two rats, and these produced 16 dendrites divided into two groups. It's also stated that \'four 3DEM series were sampled\'. Are 4 sets of serial images used? Or is it four different sets of sections from four different vibra-slices? It's also not clear how many microns in dendritic length are sampled in total. Each dendrite from the 16 traverses over 100 serial sections. It would be useful to understand what sort of sampling has been done in this study. For the analysis of the SER that is shown in Figure 3. The authors describe that there is less SER in the shaft after the LTP. There are less volume and less surface area as well as less cross-sectional surface area. As SER can have either a tubular or flattened appearance, I wonder whether these changes are due either to a simple alteration of the shape of the SER or a retraction of branches. Clearly, these are highly complex shapes, but it\'s not clear to me the significance of these results. Could a mere volume change account for the result rather than a removal of parts of the reticulum? 10.7554/eLife.46356.030 Author response > The reviewers agreed that the manuscript contains a set of very valuable data about structural changes in the setup of spines undergoing plasticity in young animals, with a particular focus on the smooth endoplasmatic reticulum and endocytic compartments. > > The reviewers however raised two key concerns, which we request to be addressed in the revised manuscript: > > Key concern 1: The reviewers are concerned about the underlying statistics that give rise to the conclusions drawn. In particular, the number of animals, slices, and reconstructed dendrites used are both unclear from the manuscript and suspected to be very small. Here, the reviewers request clarification and (if the low n is in fact confirmed) addition of further data to exclude inter-individual variability as a source of the observed effects. Subsection "Caveats" has been added to the end of the Materials and methods section. We note that these experiments are within-slice experiments, namely the control and LTP sites are from independent locations within the same slice from two different animals. Based on numerous preliminary experiments, we found that this approach greatly reduces variation due to slice preparation, in vitro conditions, and subsequent processing for electron microscopy when comparing the control and LTP outcomes. We also note that enhanced statistical power came from the large number of synapses and spines tested using the hierarchical nested ANOVA design with dendrite nested in condition by animal (Figure 2E,F,H,I, Figure 3E). In this way, degrees of freedom are adjusted for animal and dendrites, and outcomes are tested to ensure that no one dendrite or animal carried the findings. In addition, we had power to detect changes using multifactor ANOVAs for measurements that involved one measure per dendrite (\#/µm listed on the y axes of Figure 2B-D, 2G, Figure 3C-D, Figure 5B-E). Given the extremely time-consuming nature of the imaging and 3DEM analysis, additional animals and slices were not included. (See subsection "Caveats") We have expanded the Materials and methods section to clarify the choice of animals, slices, reconstructed dendrites, spines, subcellular organelles, and statistics for these experiments. Regarding these choices the relevant text occurs at lines: Subsection "Animals", subsection "Fixation and processing for 3DEM", subsection "3D reconstructions and measurements of dendrites", subsection "3D reconstructions and measurements of dendrites" and subsection "Statistical analyses". We have provided the F values, degrees of freedom, p values, and n's in each Figure legend. Where there were significant differences (p\<0.05) we have added effect sizes (η^2^) and described how they were calculated in lines subsection "Caveats". We have added supplemental figures of complete 3D reconstructions of all the analyzed dendrites, arranged by condition and spine densities and illustrating the SER composition (Figure 2---figure supplement 1) or endosome composition (Figure 5---supplement 1 and Figure 5---supplement 2). We have provided data source files for all of the figures. We have prepared a site to release the raw images, reconstruct trace files, and analytical tables in the public domain at Texas Data Repository, DOI: https://doi.org/10.18738/T8/5TX9YA, which is not yet public, but will be upon acceptance of this paper. > Key concern 2: The methodological description is considered improvable by two reviewers who request better definition of concepts such as \"new spine\". For data annotations that may be considered subjective, the independent annotation by multiple experts may offer a way to provide classifications with confidence intervals. Response regarding dynamic language such as "new spine": To address this concern, we have revised the wording in the Abstract, Results section and Discussion section to reflect our interpretations, and we added phrases like "LTP compared to control condition" to make clear that we have not actually watched the new spines form or other structures change, but made the interpretation. For example, since there were three times as many small spines in the LTP as the control condition, we interpret this outcome to mean that a subpopulation of spines was added by 2 hours after the induction of LTP. We clarified these distinctions throughout the manuscript as follows: Abstract only the last sentence uses "new" reflecting our model and interpretation. Introduction restate the interpretation from the prior literature. Results section provide an explicit explanation of how we arrived at the parsimonious explanation that some are new spines. The Discussion section opens with a further explanation of how we arrived at the conclusion that there are new spines and puts the conclusions in the context of our model in Figure 6 and beyond. Response regarding "subjective" annotations: We added a new subsection "Identification and Quantification of subcellular compartments". We revised subsection "Identifying the dendritic trafficking network" to provide a more detailed description of the compartments including the revised Figure 4 legend. New supplemental figures are provided that contain 3D reconstructions of all the dendrites to illustrate the SER in dendrites from the control and LTP conditions (Figure 2---figure supplement 1). We added a diagram in Figure 4A that describes the definition of each endosome compartment, based on prior work using endocytosis of gold particles (from Cooney et al., 2002). We added supplemental serial section images for each of the example endocytic compartments shown in Figure 4B-G and elaborated in Figure 4---figure supplement 1, Figure 4---figure supplement 2, Figure 4---figure supplement 3, Figure 4---figure supplement 4, Figure 4---figure supplement 5, Figure 4---figure supplement 6, which also include movies through serial sections when objects occupied more than 4 serial sections Figure 4---video 1, Figure 4---video 2, Figure 4---video 3. As mentioned above, we are also prepared to release the raw images, reconstruct trace files, and analytical tables in the public domain. Reviewer \#1: > \[...\] Overall, the manuscript contains potentially important data describing anatomical changes induced by LTP in young animals. However, some of their interpretation, for example their definition of \"new spines\" and their model of SER/endosome structure at the early time point, appears to be not well validated by the measurement. > > More serious issue is that it is not clear how slice-to-slice and animal-to-animal variation can be taken into account. It appears that only two slices (not clear if they are from the same animal or different animals) are used in their analysis. The conclusion may not be generalizable to all animals. > > Essential revisions: > > 1\) Figure 1: The title \"New small spines induced by TBS contained no SER while existing large spines had reduced SER content\" sounds to be misleading. It is not described how they conclude that these are \"new\" spines. Also, I don\'t see any evidence suggesting \"no SER\" in \"new spines\" from the figure, as there are some SER in small spines. The new title reads: Figure 2: The limited occupancy of spines by SER does not increase during spinogenesis in the LTP condition. > 2\) Figure 5: \"Increased endosomal activity\" may be misleading, as they are not measuring the activity but the distribution of endosomes. Revised title reads: Figure 5: Increased occurrence of endosomes in small spines after LTP. > 3\) Statistics: It appears that the entire data is based on only two slices (Figure 1: Is it from two animals?) Perhaps any statics would not work well with n=2. As LTP varies fairly a lot from slice to slice and from animal to animal, this raises a question of whether the conclusion can be generalizable. This comment is addressed above under Key concern \#1. > 4\) Figure 6: It is not clear how they come up with the model of SER structure and endosome structure at \"early LTP\", as the measurement only at 2 hours. This comment is addressed in the revised Discussion section, indicating that the earlier time points are deduced from the prior literature, more completely cited now and moved from the figure legend to the primary text of the Discussion section. Once the model is presented, then the subsequent sections have been revised to justify these interpretations -- please see the revised Discussion section. Reviewer \#2: > In this study, the authors conducted 3D EM reconstruction of CA1 dendrites after TBS LTP and concentrated on measuring organelles such as ER and endocytic compartments. Using this method, the authors make a few new observations: > > 1\) New spines that are formed after LTP do not contain SER and existing spines lose some SER. > > 2\) After LTP SER in dendritic shafts is reduced. > > 3\) After LTP, increased endocytic compartments were observed in small spines. > > Based on these observations the authors suggest that new spines observed after LTP are supported by recycling endosomes rather than SER. The apparent increase in endocytic structures after LTP is intriguing, although I have some concerns on how these structures are classified. > > Essential revisions: > > 1\) The identification of the specific endocytic structures in figure 4 and Figure 5 relies solely on morphology, and based on a limited set of images it is unclear how reliable the distinction can be made between recycling endosomes and endosomes that may be heading to a lysosome/degradation pathway. The authors list a few papers as explanation of how these structures were classified, but the description of how the dendrites were annotated is vague. The authors need to provide a lot more detail on exactly how these structures were classified and how reliable is the distinction between similar-looking endocytic structures. Based on the current data presented, the conclusion that small spines are mostly supported by recycling endosomes is not strongly supported. > > 2\) The 3D reconstruction of one time point after LTP makes it hard to infer dynamics from a static snapshot. While the reconstruction is extremely consuming, the authors need to discuss caveats and alternative interpretations of the data in the discussion. For example, can the authors rule out heterosynaptic LTD or other homeostatic mechanisms that may rely on endocytosis of receptors? Please see the complete response to Key concerns \#1 and \#2 above and revisions to the description of Figure 6 and the Discussion section that are also in response to these comments. Specifically, regarding "heterosynaptic LTD and other homeostatic mechanisms", we have added a few statements in the Discussion section, which describe a potential concern about calcium regulation if SER is diminished. We also discuss heterosynaptic LTD and explain why we opted for an interpretation that the constructive endosomes would have a positive impact on subsequent potentiation in the developing hippocampus, rather than reflect heterosynaptic LTD. Reviewer \#3: > \[...\] The study is well explained and illustrated. The quality of the electron microscopy is high and typical of this laboratory with considerable experience in this field of neuronal plasticity. The figures are well laid out and the descriptions are clear, as are the results and discussion parts. > > The Materials and methods section fully describes the procedures used, including the details of the analysis. However, it would be useful if there is a clear explanation of how the different dendrites were sampled and grouped. It appears that two slices were used from two rats, and these produced 16 dendrites divided into two groups. It's also stated that \'four 3DEM series were sampled\'. Are 4 sets of serial images used? Or is it four different sets of sections from four different vibra-slices? It's also not clear how many microns in dendritic length are sampled in total. Each dendrite from the 16 traverses over 100 serial sections. It would be useful to understand what sort of sampling has been done in this study. We have modified the methods as described under Key concern \#1 above -- specific to this inquiry please see subsection "3D reconstructions and measurements of dendrites" where we indicate that the total analyzed length was 173 µm. Also please note as indicated above that the n values are now given in each Figure legend. We also spell out that we compared the dendrites of comparable calibers, because spine density varies with dendrite caliber -- see subsection "3D reconstructions and measurements of dendrites". > For the analysis of the SER that is shown in Figure 3. The authors describe that there is less SER in the shaft after the LTP. There are less volume and less surface area as well as less cross-sectional surface area. As SER can have either a tubular or flattened appearance, I wonder whether these changes are due either to a simple alteration of the shape of the SER or a retraction of branches. Clearly, these are highly complex shapes, but it\'s not clear to me the significance of these results. Could a mere volume change account for the result rather than a removal of parts of the reticulum? Yes, it is possible that a reduction in volume without a change in reticulum could have a similar impact. In fact, as shown in Figure 3, there was a lower SER volume and surface area in the LTP relative to the control condition. To address local complexity, we used the strategy developed in the Cui-Wang et al., 2012 paper, by summing the cross-sectional area of the SER profiles on a section by section basis. In this way, we were able to detect local variation and decrease in the combined volume and reticulum fractions and compare aspiny to spiny segments of the dendrite. We have clarified this description in subsection "Reduced complexity in shaft SER after LTP" regarding Figure 3 and added further comments regarding SER complexity in the Discussion section. [^1]: Department of Biochemistry and Biophysics, Kavli Institute for Fundamental Neuroscience, University of California, San Francisco, San Francisco, United States. [^2]: QPS, LLC Pencader Corporate Center, Newark, United States.

(To access the full article, please see PDF)

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Low temperature is a major constraint on the growth, geographical distribution, and yield of some plants. Cold resistance of many plants\[[@pone.0188514.ref001]--[@pone.0188514.ref004]\], e.g. *Eucalyptus nitens*, *Miscanthus*, *Medicago sativa* and *North American Rhododendron* can be improved by prior exposure to a period of low, nonfreezing temperatures, which known as cold acclimation (CA) \[[@pone.0188514.ref005]--[@pone.0188514.ref007]\]. For instance, CA improves the tolerance of North American Rhododendron from -7°C to -53°C \[[@pone.0188514.ref004]\]. Up to date, CA is a key strategy to increase the physiological adaptation of tea plants to low temperatures \[[@pone.0188514.ref008]\]. During CA, many physiological and biochemical processes are altered in plants. Those processes include the cytoskeleton rearrangement as an integrating system perceiving the signals \[[@pone.0188514.ref009]\], accumulated membrane phospholipids and modifications in lipid composition of different organelles. For example, the proportion of MGDG (monogalactosyldiacylglycerols) was decreased and the proportion of DGDG (digalactosyldiacylglycerols) was increased in the chloroplast in Rye \[[@pone.0188514.ref010]--[@pone.0188514.ref014]\]. Moreover, plants introduce the accumulation of antifreeze proteins and cryoprotectants like soluble sugars and proline \[[@pone.0188514.ref015]--[@pone.0188514.ref016]\]. The increased synthesis of soluble sugars, including sucrose, glucose, raffinose, and fructose, contributes directly to membrane stabilization in *Alcantarea imperiali* \[[@pone.0188514.ref017]\], and *Camellia sinensis* \[[@pone.0188514.ref008]\]. The raised content of proline in *Triticum aestivum* \[[@pone.0188514.ref018]\], *Arabidopsis* \[[@pone.0188514.ref019]\] and *Camellia sinensis* \[[@pone.0188514.ref008]\] was also observed during CA. Antioxidant metabolism is known to improve the scavenging activity of reactive oxygen species (ROS) and maintain redox balance during CA \[[@pone.0188514.ref020]\]. During CA, a high ratio of abscisic acid (ABA) to gibberellin content has been shown to increase freezing tolerance in some woody taxa \[[@pone.0188514.ref021]\]. Upon cold stress, the expression of various cold-regulated (COR) genes are induced to protect plants \[[@pone.0188514.ref022]\]. The expression of COR genes is regulated by both the CBF (C-repeat-binding factor)-mediated ABA-independent pathway and the bZIP (basic region/leucine zipper)-mediated ABA-dependent pathway \[[@pone.0188514.ref023]\]. CBF transcription factors regulate \~12% of the cold-responsive transcriptome \[[@pone.0188514.ref024]\]. ICE1 (inducer of CBF expression 1)-CBF-COR cold-response pathway in plants is critical for configuring cold-induced transcriptomic changes \[[@pone.0188514.ref025]--[@pone.0188514.ref026]\]. Genes of the ICE1-CBF cold-response pathway have been reported in woody and herbaceous plants \[[@pone.0188514.ref027]--[@pone.0188514.ref029]\]. Studies have shown that the cascade regulation of *ICE1*, *CBF*, and *COR* is the main pathway for cold acclimation \[[@pone.0188514.ref030]--[@pone.0188514.ref031]\]. In *Arabidopsis*, *ICE1* express constitutively and is not responsive to cold stress, whereas ICE1 undergoes sumoylation to become functionally active \[[@pone.0188514.ref032]\]. Three CBFs (*CBF1-3*) were found in *Arabidopsis*. *CBF1* and *-CBF3* positively regulates the downstream CBF-target genes, while *CBF2* negatively regulates them \[[@pone.0188514.ref033]\]. Wang at al. \[[@pone.0188514.ref034]\] found that the ICE1-CBF-COR pathway was conserved in tea plants. To date, several COR genes have been discovered in tea plant including one *CsICE1* (FE861156), two *CsCBFs*, designated as *CsCBF1* (EU563238), *CsCBF2* (KC702795), and three dehydrin homologs designated as *CsDHN1* (GQ228834.1), *CsDHN2* (FJ436978) and *CsDHN3* (KY270880) \[[@pone.0188514.ref034]--[@pone.0188514.ref036]\]. Several studies revealed the key enzymes' activities during sugar synthesis, and associated genes expression during CA in plants. Sucrose is synthesized in the cytosol by the sucrose-phosphate synthase (SPS) and degraded by either sucrose synthase or invertase (INV) into a monosaccharide or derivative \[[@pone.0188514.ref037]\]. Raffinose synthase (RS) for raffinose synthesis was also explored in recent researches upon cold stress \[[@pone.0188514.ref038]\]. Yue et al. \[[@pone.0188514.ref008]\] analyzed the expression patterns of 32 genes during the natural CA in tea plant (var. *sinensis* cv. *Longjing43*) and found that expression of *CsSPS*, *CsINV5 and CsRS2* was significantly induced. To date, it is known that the proline biosynthesis is catalyzed by P5C synthase (P5CS) and P5C reductase (P5CR) in plants \[[@pone.0188514.ref039]--[@pone.0188514.ref040]\]. Another key enzyme in the proline synthesis pathway is Ornithine-D-aminotransferase (OAT) \[[@pone.0188514.ref041]\]. Degradation of proline is catalyzed by Pro-dehydrogenase (ProDH) and P5C-dehydrogenase (P5CD) \[[@pone.0188514.ref042]\]. In tea plants, the sequences of *CsP5CS* (KJ143742.1), *CsOAT* (KJ641844.1) and *CsP5CR* (KY368574), *CsP5CDH*(KY368572) and *CsProDH* (KY368573) are available at NCBI (<https://www.ncbi.nlm.nih.gov/>). Tea plants (*Camellia sinensis* (L.) O. Kuntze), one of the important economic wooden plants in the world, are mainly grown in subtropical and tropical regions. Two basic classes of varieties can be classified as var. *assamica*, a quick-growing tree well suited to tropical climates, and var. *sinensis*, a slower-growing bush that can withstand colder climates than *assamica* \[[@pone.0188514.ref043]--[@pone.0188514.ref044]\]. Tea plants are vulnerable to cold injury during winter such as in East Asia (China, Japan), especially in northern China. Recent studies have explored the response of tea plants to cold stress and natural CA \[[@pone.0188514.ref045]--[@pone.0188514.ref048]\]. However, a comparative study on cold resistance between cold-resistant and cold-susceptible cultivars has not been reported yet. The present study was conducted to explore the molecular mechanism of cold resistance by treating the cold-resistant *camellia var*. *sinensis* CV. *Shuchazao* (SCZ) and cold-susceptible *camellia var*. *assamica* CV *Yinghong9* (YH9) under CA and de-acclimation (DA). We found difference in biochemical changes, including EL50 (temperature leading to 50% tissue damages due to leakage of electrolyte), Fv/Fm (maximum quantum yield of PSII photosystems), sugars and proline. Then we examined the expression of 14 genes related to these biochemical changes. Comparison of gene expression and study of biochemical changes in the responses to cold in two tea cultivars led to our finding of the difference in cold tolerance. Our results indicated that the increased expression of *CsCBF1* and *CsDHNs* coupling with the accumulation of sucrose has played a role in conferring higher cold resistance in tea cultivar SCZ. The results provide understanding in biochemical and gene regulatory mechanisms of cold resistance in tea plants. Materials and methods {#sec002} ===================== Plant material {#sec003} -------------- The clone cuttings of *Camellia sinensis* cv. *Shuchazao* and *Camellia sinensis* var. *assamica* cv.*Yinghong9* were obtained from the Dechang Tea Plantation in Anhui (116° 56\' 24\'\' E, 31° 27\' N) and the Tea Research Institute of Guangdong Academy of Agricultural Sciences (113° 22\' 48\'\' E, 24° 10\' 12\'\' N), China, respectively. One-year-old cutting-propagated plants were transferred to a growth chamber with temperature cycles of 25°C at day time and 20°C at night time, 12 h photoperiod, and 70% relative humidity for one month. Subsequently, they were subjected to varying degrees of cold acclimation and de-acclimation. Ten well-grown one-year-old tea plants were collected and used as non-acclimation (NA). The following cold acclimation (CA) treatments were applied in this study: CA1 was conducted by exposing SCZ and YH9 to low temperature (10/4°C, day/night temperature) for 7 days. Afterwards, CA2 was conducted by exposing SCZ and YH9 to lower temperatures (4/0°C, day/night temperature) for another 7 days. Lastly, the plants were exposed to normal temperature (25/20°C, day/night temperature) for 7 days for de-acclimation. At each time point, the leaves were collected, immediately frozen in liquid nitrogen and stored at −80°C until use. Three biological replicates were conducted. Electrolyte leakage assay {#sec004} ------------------------- Relative electrolyte leakage was measured to evaluate the cell membrane damage as described with some modifications \[[@pone.0188514.ref049]\]. Briefly, after washing with distilled deionized water, the leaf pieces were obtained using a puncher from leaves after each treatment. After subsequently exposed to -2°C, -4°C, -6°C, -8°C, and -10°C for 12 hours, samples were placed in glass bottles containing 20 mL of distilled deionized water. The electrical conductivity of the solution (L1) was determined using a conductivity meter STARTER 3100C (Ohaus; America) at 25°C. The solutions were then heated to 100°C for 30 min and the final electrical conductivity (L2) was determined after cooling to 25°C. The REC (relative electrical conductivity) was calculated as L1/L2×100%. Fv/Fm {#sec005} ----- Mature leaves (from third to fifth leaves) of tea cultivar SCZ and YH9 were carefully clamped in the middle part of the leaves, avoiding the main vein and then dark-adapted in leaf clips for 20 min prior to measurement. Chlorophyll fluorescence parameters Fm and Fo were measured by OS-30P modulated fluorometer (Opti-Sciences, USA) and Fv was obtained using Fv = Fm-Fo \[[@pone.0188514.ref050]\]. Ten biological replicates were performed for the experiment. Measurement of proline content {#sec006} ------------------------------ Proline contents in SCZ and YH9 were measured by the colorimetric assay according to Bates method with some modifications \[[@pone.0188514.ref051]\]. Briefly, approximately 0.5 g leaves of SCZ and YH9 were ground into fine powder in liquid nitrogen. The powder was immediately resuspended in 5 mL of 4% sulfonic acid and sonicated for 30 min. The mixture was subsequently centrifuged for 30 min at 12000 rpm and the supernatant was collected. 2 mL supernatant, 2 mL glacial acetic acid and 3 mL ninhydrin reagent (2.5% \[w/v\] ninhydrin, 60% \[v/v\] glacial acetic acid, and 40% 6 M phosphoric acid) were added, mixed and heated to 100°C for 40 min. After cooling down to room temperature, 5 mL toluene was added and the absorbance was measured at 520 nm using an UV Spectrophotometer (U-5100, Hitachi). Measurements of soluble carbohydrates {#sec007} ------------------------------------- Contents of soluble carbohydrates of fructose, sucrose, glucose, raffinose and trehalose in the leaves of tea cultivar SCZ and YH9 were measured by High Performance Liquid Chromatography (HPLC) (Agilent, America). The samples were prepared following the protocol as previously described with some modifications \[[@pone.0188514.ref052]\]. Briefly, approximately 0.5 g leaves of SCZ and YH9 were weighed and ground in liquid nitrogen, and 10 ml of distilled water was added immediately. After heating at 100°C for 1 h, the mixture was subsequently centrifuged for 10 min at 12000 rpm and the supernatant was collected. The aqueous phase was collected and dried on a rotary evaporator. It was then resuspended in distilled water and filtered through a 0.22 μm filter membrane prior to HPLC analysis. Standard of fructose, sucrose, glucose, raffinose and trehalose were purchased from Sangon Biotech. Co. (Shanghai, China) RNA extraction and real-time quantitative PCR analysis {#sec008} ------------------------------------------------------ Total RNAs were extracted from leaves of tea cultivar SCZ and YH9 with RNA prep Pure Plant kit (Tiangen, Beijing, China). The total RNAs were reverse transcribed into first-strand cDNA with PrimeScript Reagent Kit (TaKaRa, Dalian, China) and the cDNAs obtained were used as templates for PCR amplification with specific primers. Gene-specific primers ([Table 1](#pone.0188514.t001){ref-type="table"}) were used for real-time quantitative RT-PCR. The glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene was used as an internal reference gene \[[@pone.0188514.ref053]\] and the relative expression was calculated using the 2^ΔCt^ method \[[@pone.0188514.ref054]\]. Each reaction contained 12.5 μL of SYBR^®^Premix Ex Taq^™^II (Tli RNaseH Plus; TaKaRa, Dalian, China), 2 μL cDNA, and 1 μL 10 μM gene-specific primers in a final volume of 25 μL. All reactions were carried out using the CFX96^™^ Real-Time System (Bio-Rad, USA) using a two-step method: 95°C for 3 min; 40 cycles of 95°C for 10 s, 62°C for 30 s. 10.1371/journal.pone.0188514.t001 ###### Genes and corresponding primers used for the RT-qPCR experiments. {#pone.0188514.t001g} Gene name GenBank Accession No. Primer Primer sequence (5\'--3\') ----------- ------------------------------- --------- ------------------------------- GAPDH GE651107 Forward `TTG GCA TCG TTG AGG GTC T` Reverse `CAG TGG GAA CAC GGA AAG C` CsICE1 FE861156 Forward `ATG TTT TGT AGC CGC AGA C` Reverse `GCT TTG ATT TGG TCA GGA TG` CsCBF1 EU563238 Forward `AGA AAT CGG ATG GCT TGT GT` Reverse `TTG TCG TCT CAG TCG CAG TT` CsCBF2 KC702795 Forward `CAC AGC CTG CTC ATC ACT` Reverse `ACC ACT GCC ACA ATC TG` CsDNH1 GQ228834.1 Forward `ACA CCG ATG AGG TGG AGG TA` Reverse `AAT CCT CGA ACT TGG GCT CT` CsDNH2 FJ436978 Forward `ACT TAT GGC ACC GGC ACT AC` Reverse `CTT CCT CCT CCC TCC TTG AC` CsDNH3 KY270880 Forward `TCC ACA TCG GAG GCC AAA AG` Reverse `AAC CCT CCT TCC TTG TGC TC` CsSPS KF696388 Forward `ACC TGG AGG CGA TTC TGG ATG` Reverse `TTC CAA ATC CGC CAG CAC ATA` CsRS2 KP053395 Forward `CGG TTT GGC GCT TAC TCT TC` Reverse `TCT CCT CTT CTG CAA CCG GA` CsINV5 KP053402 Forward `AGT CTT GCC CCT TGA TGT CG` Reverse `AAC CAA ACG GTC CAA GAG CA` CsP5CS KJ143742.1 Forward `AGG CTC ATT GGA CTT GTG ACT` Reverse `CAT CAG CAT GAC CCA GAA CAG` CsOAT KJ641844.1 Forward `GCG GTT AAT CAG GGA CAT` Reverse `ACA CCT TCG GCA CCA GTA` CsP5CR KY368574 Forward `TAG GGG AGG CGG CAT CAG TT` Reverse `ACC CCT CCA TCA GCC AAA GC` CsP5CDH1 KY368572 Forward `TGC TGA TGG GAA GAC GAT` Reverse `GCC GAG CAC TTT TGA CCA CT` CsProDH KY368573 Forward `CAA AAC CCA AAT CCA ACC G` Reverse `TCC TCC TCA CTA CCC CCA AC` Primer design {#sec009} ------------- The primers were designed against the sequence of genes which is retrieved from Genbank using the listed accession number ([Table 1](#pone.0188514.t001){ref-type="table"}). The software Primer Premier 5 (Premier Biosoft International, Palo Alto, California, USA) was used to designed specific primers ([Table 1](#pone.0188514.t001){ref-type="table"}) and the primers were then synthesized by Sangon Biotech Co. (Shanghai, China). We checked the specify of the primers and which produced one peak in melting curve, indicating a single amplicom of target gene. Then we used these primers for the level of transcript (Figures A-O in [S1 Text](#pone.0188514.s001){ref-type="supplementary-material"}). qPCR products have been sequenced and the results evaluated using the DNAman computer software version 5.0 (Lynnon Biosoft) (Figures A-O in [S2 Text](#pone.0188514.s002){ref-type="supplementary-material"}). Statistical analysis {#sec010} -------------------- EL50 was calculated by logistic equations. Statistical analyses were performed using DPS and Prism5, GraphPad Software. The results were expressed as mean value ± standard error (SE). Different letters indicate significant differences to Duncan's multiple range tests with *P* \< 0.05. Results {#sec011} ======= Cold acclimation induces difference freezing tolerance in tea cultivars {#sec012} ----------------------------------------------------------------------- To investigate the cold tolerance, we selected two tea cultivars SCZ and YH9 which are known as with high and low cold tolerance, respectively ([Fig 1](#pone.0188514.g001){ref-type="fig"}). SCZ has been planted in cool areas in middle and warm areas in south China while cultivar YH9 in warm areas only in south China. Tea cultivar SCZ has smaller leaf than cultivar YH9 ([Fig 1](#pone.0188514.g001){ref-type="fig"}). We treated the one-year-old plants clonally propagated from cuttings of these two tea cultivars SCZ and YH9 with cold treatments CA1, CA2 and DA in growth camber to measure the physiological responses, membrane damage and chlorophyll content. We first treated the tea cultivar SCZ under 10/4°C (day/night) for 28 days and checked the change Fv/Fm and total sugar contents in a time course manner. We found that 7 days of cold treatment is enough to detect significant changes ([S3 Text](#pone.0188514.s003){ref-type="supplementary-material"}). Therefore, we used 7 days of treatment in this study. Our results showed that SCZ leaves remain green while those of YH9 became reddish brown after all treatments ([Fig 2A](#pone.0188514.g002){ref-type="fig"}). We further examined the electrolyte leakage which reflects cell membrane damage in cold by using EL50 analysis. As shown in [Fig 2B](#pone.0188514.g002){ref-type="fig"}, the EL50 had significant difference between SCZ and YH9, and the EL50 were -5.7°C and -2.3°C in SCZ and YH9 before cold treatment, respectively. Cold treatment CA1 treatment resulted in reduced EL50 values for both SCZ and YH9 ([Fig 2B](#pone.0188514.g002){ref-type="fig"}). Further cold treatment CA2 led to a more reduction of EL50 in SCZ to -9.4°C, while the EL50 value of YH9 remained unchanged compared with CA1. Lower EL50 represents less leakage. Thus, this result suggested higher cold tolerance in cultivar SCZ. The Fv/Fm value of both cultivars was relatively consistent in the range of 0.80\~0.85 before treatment while both SCZ and YH9 displayed similar lower Fv/Fm values (*P* \< 0.05) after CA1 treatment ([Fig 2C](#pone.0188514.g002){ref-type="fig"}). CA2 treatment further reduced the Fv/Fm value but the value of YH9 reduced more than SCZ (*P* \< 0.05) ([Fig 2C](#pone.0188514.g002){ref-type="fig"}). The lower Fv/Fm suggests less chlorophyll, which explains the observed reddish color in YH9 after treatment. After DA treatment, the ratios of Fv/Fm in the two cultivars returned to the normal level ([Fig 2C](#pone.0188514.g002){ref-type="fig"}), which supported the Fv/Fm change in chlorophyll was caused by cold treatment and was reversible. {#pone.0188514.g001} {#pone.0188514.g002} Effect of CA and DA on soluble sugars accumulation in SCZ and YH9 {#sec013} ----------------------------------------------------------------- Sugar accumulation is known to have both osmotic and non-colligative functions, as it stabilizes cell membrane during cold acclimation and enhances freezing resistance in plants \[[@pone.0188514.ref007], [@pone.0188514.ref055]\]. As shown in [Fig 3](#pone.0188514.g003){ref-type="fig"}, the total sugar content and the sucrose level in both cultivars were significantly increased under CA condition, with a higher increase observed in SCZ. Relative to NA, the total sugar content in SCZ leaves was increased 2.39-fold after CA2, while the total sugar content in YH9 was increased 1.83-fold after CA2 ([Fig 3A](#pone.0188514.g003){ref-type="fig"}). Furthermore, sucrose content in SCZ was increased 2.56-fold after CA1 and reached 3.26-fold after CA2 relative to NA (*P* \< 0.05). In contrast, the sucrose content in YH9 leaves was increased 2.0-fold after CA1 and remained constant after CA2 relative to CA1 ([Fig 3B](#pone.0188514.g003){ref-type="fig"}). In addition, CA1 and CA2 also induced a moderate increase in glucose and fructose contents in SCZ and YH9 (*P* \< 0.05) ([Fig 3C and 3F](#pone.0188514.g003){ref-type="fig"}). Differently, CA1 and CA2 induced a little accumulation of raffinose in SCZ (*P* \< 0.05), while only a small accumulation of trehalose was observed in YH9 under CA2 (*P* \< 0.05) ([Fig 3D and 3E](#pone.0188514.g003){ref-type="fig"}). After DA, individual sugar content was decreased by a varying degree ([Fig 3](#pone.0188514.g003){ref-type="fig"}). {#pone.0188514.g003} Effect of CA and DA on proline accumulation between SCZ and YH9 {#sec014} --------------------------------------------------------------- As the proline is a multi-functioned osmotic protective substance involved in cold tolerance \[[@pone.0188514.ref039]\], we measured the changes in proline content in SCZ and YH9 during CA and DA ([Fig 4](#pone.0188514.g004){ref-type="fig"}). As shown in [Fig 4](#pone.0188514.g004){ref-type="fig"}, the proline levels in both cultivar SCZ and YH9 were increased by 2.27-fold and 4.9-fold during CA1, respectively, and accumulated to similar contents. The study also revealed that the proline content reached the peak in CA1, and afterwards gradually decreased in both SCZ and YH9 ([Fig 4](#pone.0188514.g004){ref-type="fig"}). {#pone.0188514.g004} Effect of CA and DA on the gene expression of the ICE1-CBF pathway {#sec015} ------------------------------------------------------------------ The transcription of *CBFs* is regulated by ICE1 protein, which binds to the DRE/CRT cis-elements in the promoter regions of CORs ([Fig 5A](#pone.0188514.g005){ref-type="fig"}). *CBFs* play a central role in integrating the activation of multiple components of the CA respond to chilling and freezing stress in plants \[[@pone.0188514.ref031], [@pone.0188514.ref056]\]. This study did not observe significant changes in *CsICE1* transcription in SCZ and YH9 during CA and DA ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). The expression of Cs*CBF1* was significantly increased during CA, and reached approximately 282-fold and 29-fold in CA2, in SCZ and YH9, respectively. The transcript level of *CsCBF1* during CA was 7.1--9.5 folds higher in SCZ than that of YH9 (*P* \< 0.05) ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). In contrast, the transcript level of *CsCBF2* was increased in YH9 but remained unaffected in SCZ during CA ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). For COR genes, the transcription of *CsDNH1*, *CsDNH2* and *CsDNH3* in both cultivars were increased significantly during CA and rapidly decreased following DA. Furthermore, the transcript levels of *DHNs* were higher in SCZ than in YH9 ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). Specifically, *CsDNH3* transcript level in SCZ was dramatically increased by 68.7-fold in CA2, while it was only increased by 9.2-fold in YH9 ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). {#pone.0188514.g005} Effect of CA and DA on the transcription of sucrose- and raffinose-related genes {#sec016} -------------------------------------------------------------------------------- The *CsSPS* and *CsINV5* are responsible for sucrose accumulation and converting to other saccharides ([Fig 6A](#pone.0188514.g006){ref-type="fig"}). The transcripts levels of these two genes were checked in both cultivars during CA and DA. We found that the transcript level of *CsSPS* increased in tea cultivar SCZ but reduced in YH9 through all CA stages, especially in stage CA1 after seven days of CA ([Fig 6B](#pone.0188514.g006){ref-type="fig"}). However, *CsINV5* expression was decreased at CA1 after seven days of CA but increased by three folds (*P* \< 0.05) at CA2 after 14 days in both cultivars. *CsINV5* expression in cultivar SCZ increased by three folds at CA2 compared with less than one fold in YH9 ([Fig 6B](#pone.0188514.g006){ref-type="fig"}). Raffinose synthase gene *CsRS2*, responsible for synthesizing of raffinose, expression was also increased during CA in variety SCZ and then decreased following DA, while it had no distinct changes during CA in YH9 ([Fig 6B](#pone.0188514.g006){ref-type="fig"}). This suggested that the genes responsible for sucrose and raffinose accumulation acted positively to regulate the corresponding sugar accumulation in cold treatment. {#pone.0188514.g006} Effect of CA and DA on proline biosynthesis and degradation {#sec017} ----------------------------------------------------------- We illustrated the proline synthesis and degradation in plants in schematic diagram ([Fig 7](#pone.0188514.g007){ref-type="fig"}). The expression patterns of these proline-associated enzyme encoding genes were studied in SCZ and YH9 over the treatments of CA and DA. *CsP5CS* and *CsP5CR* were up-regulated at CA2 in SCZ after 14-days cold treatment, and their levels were increased by 1.4-fold and 3.0-fold, respectively. By the contrary, no changes in transcription of *CsP5CS* were observed in YH9 over the treatments of CA and DA. While *CsP5CR* was up-regulated in CA1 in YH9. This suggested that long time induced *CsP5CR* in stage CA2 may be responsible for higher cold tolerance. In both cultivars, the transcription of *CsOAT* was decreased during CA treatments and returned to the normal level after DA treatment. In addition, the transcript level of *CsOAT* in SCZ was similar to that in YH9 ([Fig 7](#pone.0188514.g007){ref-type="fig"}). *CsP5CDH* was down-regulated in SCZ during CA1, whereas was unaffected by the CA treatments in YH9 ([Fig 7](#pone.0188514.g007){ref-type="fig"}). The transcription of *CsProDH* in both cultivars was decreased during CA treatments ([Fig 7](#pone.0188514.g007){ref-type="fig"}). {#pone.0188514.g007} Discussion {#sec018} ========== Low temperature has been a major constraint for tea plantation \[[@pone.0188514.ref047]\]. During natural CA, tea plants can increase their tolerance to cold weather and survive the winter \[[@pone.0188514.ref008], [@pone.0188514.ref045]\]. In the present study, we demonstrated that plants of either cold-resistant or cold-sensitive tea cultivars can enhance their freezing tolerance due to the treatment of CA in experimental conditions, with a stronger freezing tolerance developed in SCZ than in YH9 ([Fig 2B](#pone.0188514.g002){ref-type="fig"}). To understand the mechanisms underlying such differential cold tolerance between SCZ and YH9, we investigated differences between the two cultivars in the physiological and molecular processes that were known to induce cold tolerance in other plants. Our results showed that SCZ exhibited a higher accumulation level of soluble sugars, particularly sucrose than YH9, during cold acclimation ([Fig 3](#pone.0188514.g003){ref-type="fig"}). The increased expression of both *CBF1* and its targets *DHNs* could contribute to cold tolerance ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). These findings may further elucidate how cold-resistant tea plants can induce strong freezing tolerance in winter. Through the process of CA, cold resistance was steadily induced in SCZ, while it was induced in YH9 at a slower rate ([Fig 2B](#pone.0188514.g002){ref-type="fig"}). Similar to our results, both cold-resistant Medicago (*M*. *falcate*) and cold-susceptible Medicago (*M*. *truncatula*) could enhance their freezing tolerance by CA at 4°C \[[@pone.0188514.ref057]\]. However, Pennycooke et al. \[[@pone.0188514.ref058]\] found that the CA-induced freezing tolerance occurred only in cold-resistant plants, but not in cold-susceptible plants. Under cold stress, the inhibition of chlorophyll synthesis and chloroplast formation can lead to reduced Fv/Fm \[[@pone.0188514.ref059]\]. Bonnecarrère et al. \[[@pone.0188514.ref016]\] also used the Fv/Fm to identify cold-resistant rice between two japonica genotypes under identical cold stress. Our results indicated that Fv/Fm fell lower in YH9 than in SCZ during CA2 ([Fig 2C](#pone.0188514.g002){ref-type="fig"}), suggesting that Fv/Fm, combined with EL50, could be used to evaluate freezing tolerance in cold-resistant and cold-susceptible tea plants. Soluble sugar accumulation during CA is positively correlated with freezing tolerance in plants \[[@pone.0188514.ref007]\]. Sucrose was found as a dominant component of enhanced soluble sugars in giant reed and Medicago during CA \[[@pone.0188514.ref057], [@pone.0188514.ref060]\]. But in *Arabidopsis*, the accumulation of glucose is largely responsible for the increased level of soluble sugars during cold acclimation and sucrose is the second most abundant sugar in CA \[[@pone.0188514.ref061]\]. In our study, the content of sucrose, glucose, raffinose, and fructose were increased during CA in SCZ and YH9, and the total sugar content was higher in SCZ than in YH9 ([Fig 3](#pone.0188514.g003){ref-type="fig"}). In other studies, greater accumulation of sucrose in cold resistant Medicago, wheat, and maize was found to be responsible for higher freezing tolerance \[[@pone.0188514.ref057], [@pone.0188514.ref062]--[@pone.0188514.ref063]\]. Trehalose accumulation conferred tolerance to cold stress serving as an osmolyte or protein/membrane protectant by acting as scavengers for ROS to alleviate oxidative damage to the membranes \[[@pone.0188514.ref064]\]. Therefore, the accumulation of sucrose, as the major sugar, in cold resistant tea plants could play an essential role in conferring higher freezing tolerance in tea plant. Trehalose accumulation was not observed in SCZ during the CA, although the trehalose content was induced slightly in YH9, yet lower than in SCZ ([Fig 3E](#pone.0188514.g003){ref-type="fig"}). Our data showed that raffinose contents in the two cultivars were very similar during CA ([Fig 3D](#pone.0188514.g003){ref-type="fig"}). However, raffinose was found to be not essential for basic freezing tolerance or for cold acclimation of *A*. *thaliana* \[[@pone.0188514.ref038]\]. Thus, the role of raffinose in cold resistance in tea plant may be not essential too. Yue et al. \[[@pone.0188514.ref008]\] reported the content of total sugars and several specific sugars including sucrose, glucose and fructose were constantly elevated in *Longjing43* tea leaves during nature acclimation. While Shen et al. \[[@pone.0188514.ref047]\] reported the raffinose, maltose, glucose and fructose were all more abundant in HuangShanzhong tea leaves during nature acclimation. During natural cold acclimation, a series of sugar-related genes, including *CsSPS*, *CsRS2*, and *CsINV5*, are up-regulated in the tea plant (cv. *Longjing43*) \[[@pone.0188514.ref008]\], which suggests that these genes might be responsible for sugar accumulation. Under the controlled cold treatment and DA in our study, the transcription of the *CsSPS*, *CsINV5*, and *CsRS2* during CA was also up-regulated in SCZ but remained unchanged in YH9 ([Fig 6](#pone.0188514.g006){ref-type="fig"}). This demonstrated that the expression regulation of *CsSPS*, *CsINV5*, and *CsRS2* during CA are cultivar specific. Therefore, these genes' expression can be used as cultivar specific cold resistance indicator for tea breeding. Free proline has been reported to accumulate in many plants in response to biotic and abiotic stresses, acting as a compatible solute against osmotic stress \[[@pone.0188514.ref039]\]. Free proline was one of the indicators used to identify dehydration 'resistant' wheat genotypes from 'sensitive' ones \[[@pone.0188514.ref065]\]. Kumar and Yadav \[[@pone.0188514.ref066]\] reported that enhanced proline could increase the tolerance of tea bud to cold stress. Similarly, tea cultivars *'Zhuyeqi'* (drought-susceptible) and *'Ningzhou 2*' (drought-tolerant) could be distinguished due to their differential proline contents under drought stress \[[@pone.0188514.ref067]\]. Our data showed that proline content was significantly increased in SCZ and YH9 during CA1, but no difference in the total proline content was found between the two cultivars ([Fig 4](#pone.0188514.g004){ref-type="fig"}). Therefore, we propose that proline content has effects on the abiotic resistance of tea plants and that the accumulated proline was not a key factor for conferring cold-resistant in tea plants. However, proline concentration is correlated with cold-resistance in giant reed and spring canola \[[@pone.0188514.ref060], [@pone.0188514.ref068]\]. According to Delauney \[[@pone.0188514.ref069]\], under abiotic stress, proline was accumulated by the glutamate biosynthesis pathway. Our result showed that the *OAT*, responsible for glutamie acid scmialdehyd, transcript level decreased in both cultivars ([Fig 7](#pone.0188514.g007){ref-type="fig"}). In accordance with Delauney \[[@pone.0188514.ref069]\], it was glutamate, not ornithine, which could likely be the main precursor for proline biosynthesis in tea plants during CA. The transcription of *CsP5CS* and *CsP5CR* for proline biosynthesis was higher in SCZ than in YH9, and the transcripts of *CsProDH* and *CsP5CDH* for proline degradation differed between two cultivars ([Fig 7](#pone.0188514.g007){ref-type="fig"}). It showed that the transcription level of the related genes was not consistent with metabolic changes and further enzymatic assays are required to elucidate the proline biosynthesis mechanisms. To date, ICE and CBF genes are known to play key roles in cold tolerance. The transcription of *CsCBF1*, *not CsICE1*, was induced at 4°C \[[@pone.0188514.ref034]\]. With CA treatment, our results consistently showed that there was not change in expression of *CsICE1* in both cold resistant and susceptible tea cultivars. We found that the transcription of *CsCBF1* was significantly up-regulated by CA, and remained high level until DA ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). A higher expression change of *CsCBF1* was found in cold resistant cultivar SCZ than cold susceptible YH9 ([Fig 5B](#pone.0188514.g005){ref-type="fig"}), which may explain the difference in cold resistance in the two tea cultivars. A similar result was also found in Medicago and Jatropha \[[@pone.0188514.ref057], [@pone.0188514.ref060]\]. However, Pan et al. \[[@pone.0188514.ref070]\] reported a contradictory finding that the cold-susceptible rice had a much strong transcription of *CBFs* than cold-resistant rice. This might be caused by the species difference. Our data found a higher up-regulation of *CsCBF2* transcription in YH9 than SCZ during CA ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). *AtCBF2* is a negative regulator of *AtCBF1* in *Arabidopsis*. We hypothesized that that *CsCBF2* in tea plant might also be a negative regulator of *CsCBF1*. In this case, the lowered transcription of *CsCBF1* can be explained by the suppression of the increased transcription of *CsCBF2* in YH9. The high level of *CsCBF1* transcription in SCZ was a result of low transcription of *CsCBF2*. Further investigation of the suppression would be the priority in future study. In addition, *CsCBF1* in SCZ and YH9 had the same ORF (Open Reading Frame) without Intron. We speculate that SNP and INDEL may be present in the promoter region of *CsCBF1*, and regulate the transcription of *CsCBF1* between resistant and susceptible species. *DHNs* (COR genes), a subgroup of the late embryogenesis abundant protein family, can act as a cryoprotectant and molecular chaperone as well as an anti-oxidant. *DHNs* can be induced by cold stress, and their transcription is correlated with freezing tolerance \[[@pone.0188514.ref036], [@pone.0188514.ref071]\]. In this study, the transcripts of *CsDHN1*, *CsDHN2*, and *CsDHN3* were accumulated at higher levels in SCZ than YH9 during CA ([Fig 5B](#pone.0188514.g005){ref-type="fig"}). Similar results were found in Loquat (Eriobotrya japonica), where seven dehydration genes were up-regulated under low-temperature stress, with significantly higher transcription observed in cold resistance than cold-susceptible cultivars \[[@pone.0188514.ref071]\]. The higher levels of *CsDHN1*, *CsDHN2*and *CsDHN3* may result in higher amount of dehydrin proteins, thus protecting SCZ from dehydration under freezing stress. Of course, other mechanisms may involve in the cold acclimation in tea plant. One of them is that the PLD (Phospholipase D) pathway, which responses to freezing and plays key roles in conferring higher cold resistance \[[@pone.0188514.ref011]\]. PLDs, lipid catabolism enzymes, are activated by a fall in temperature, and the expression levels are found to increase during cold stress \[[@pone.0188514.ref072], [@pone.0188514.ref073]\]. Phosphatidic acid (PA), a catalyzed production of phospholipase D (PLD), involves in many cellular processes, including cell signaling, vesicular trafficking and membrane remodeling \[[@pone.0188514.ref012], [@pone.0188514.ref072]\]. Cold acclimation also affects cell lipid composition, which in favor of the maintenance of plasma membrane functionality and fluidity \[[@pone.0188514.ref010], [@pone.0188514.ref074]\]. In particular, the proportion of unsaturated fatty acids making up the phospholipids is increased \[[@pone.0188514.ref074]\]. A substantial increase in linoleic acid (C18:2) has been reported for cold acclimated Solanum commersonii plants, a potato wild species able to increase freezing tolerance. While the freezing susceptible species, Solanum tuberosum, was an increase of C18:3 \[[@pone.0188514.ref075]\]. Fatty acids unsaturation is controlled by a transcriptional regulation of key desaturase genes. The cotyledons of cold acclimated plants produced a high-fold increase in delta 12 desaturase FAD2-3 (FAD2-3) expression compared with non-acclimated plants \[[@pone.0188514.ref013]\]. For tea plant (Cs var. sinensis), both CsFAD7 and CsFAD8 were cloned, and CsFAD8 genes has a high expression in cold resistant cultivar than susceptible cultivar \[[@pone.0188514.ref048]\]. Due to a fact that the cold acclimation may not apply to some plants such as crop wheat, it is worthy to investigate how the lipid metabolism is regulated under cold in tea and whether it is correlated with expression level of fatty acid related genes in cold in our future study using RNA-seq and metabolomics strategy. In summary, the data presented here have demonstrated the difference of physiological, biochemical, and gene expression levels explained the difference in cold tolerance in cold-resistant tea cultivar SCZ and cold-susceptible tea cultivar YH9. These findings have contributed a better insight into the molecular mechanisms that underly cold tolerance in tea plants. Supporting information {#sec019} ====================== ###### Melt curve and Melt peak of *genes*. (DOCX) ###### Click here for additional data file. ###### Comparative sequence analysis among GenBank accession and qPCR products of SCZ and YH9 of *genes*. (DOCX) ###### Click here for additional data file. ###### Effects of CA on freezing tolerance of SCZ. (DOCX) ###### Click here for additional data file. This work was supported by the National Natural Science Foundation of China \[grant number: 31270729\], the Special Innovative Province Construction in Anhui Province (15czs08032), the Central Guiding the Science and Technology Development of the Local (2016080503B024). CA : cold acclimation CBF : *C-repeat/dehydration-responsive element binding factor* CsINV5 : *Invertase gene* CsOAT : *Ornithine-D-aminotransferase* CsP5CDH : *P5Cdehydrogenase* CsP5CR : *1-pyrroline-5-carboxylate reductase* CsP5CS : *Δ -1-Pyrroline-5-carboxylate synthase* CsProDH : *Proline dehydrogenase* CsRS2 : *Raffinose synthase gene* CsSPS : *Sucrose phosphate synthase* DA : de-acclimation DHN : *Dehydrin* EL50 : Temperature leading to 50% tissue damages due to leakage of electrolyte Fv/Fm : maximum quantum yield of PSII photosystems SCZ : *Camellia sinensis* cv. *Shuchazao* YH9 : *Camellia sinensis* var. *assamica* cv.*Yinghong9* [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Pes cavus is an increase of normal plantar concavity, where the anterior and posterior weight-bearing areas of the foot are brought closer together. A wide spectrum of foot deformities includes a plantarflexed first ray, forefoot pronation and adduction, and hindfoot varus or high calcaneal pitch.^[@bibr1-2058-5241.2.160077],[@bibr2-2058-5241.2.160077]^ Cavovarus deformity can be classified according to the severity of malalignment ranging from a subtle and flexible cavovarus foot to a severe and fixed cavovarus deformity. There are many aetiologies of unequal frequency that account for cavovarus foot deformities. Traumatic causes are rare (improperly treated fracture or subluxation of the tarsal bones or scarring from a burn of the sole of the foot). Cavovarus deformity has been long associated with neurological disease such as cerebral palsy, Charcot-Marie-Tooth (CMT) disease or other hereditary sensory and motor neuropathies (myelodysplasia, Friedreich ataxia, etc). CMT disease results from defects in the genetic code for the protein of the peripheral myelin sheath and is classified into subtypes varying in progression. CMT IA is the most common form including peripheral nerve myelin degeneration and decreased motor nerve conduction. In most cases, the disease process is progressive rather than static; therefore, the deformities worsen and surgical treatment must be considered to prevent the progression to a fixed and symptomatic deformity.^[@bibr3-2058-5241.2.160077]^ However, in recent years, a mild variation of the cavovarus deformity has been increasingly observed to exist without an identifiable underlying deficit.^[@bibr4-2058-5241.2.160077]^ In our experience, this primary pes cavus (idiopathic) is diagnosed by elimination in more than half the cases and most authors believe that it is the consequence of a latent neurological disorder. Thus, neurological disorders must be looked for in the family history and clinical and electrophysiologica evaluation of the patient is necessary to eliminate any very subtle neurological lesion. Patho-anatomy {#section1-2058-5241.2.160077} ============= There are several types of pes cavus, depending on the site of the deformity. Some authors have divided the deformity into posterior, anterior or mixed cavus which includes both deformities.^[@bibr5-2058-5241.2.160077]^ The most frequent anterior pes cavus is characterised by lowering of the forefoot in plantarflexion ([Fig. 1](#fig1-2058-5241.2.160077){ref-type="fig"}). In total pes cavus, the increase of the slope of the forefoot involves the whole of the metatarsal range, whereas in medial pes cavus, it decreases from the medial to the outer side which causes pronation of the forefoot. {#fig1-2058-5241.2.160077} The posterior cavus or calcaneocavus is characterised by an isolated high calcaneal pitch of greater than 30° related to a weakness of the gastrocnemius muscle leading to a calcaneus deformity of the hindfoot. The exact cause in the cavus foot is a longstanding issue, and both intrinsic and extrinsic muscle imbalance may play a role in the final deformity.^[@bibr6-2058-5241.2.160077]^ An imbalance between the antagonistic muscles, in particular the peroneus longus and tibialis anterior, is often listed as a cause.^[@bibr7-2058-5241.2.160077]^ Manoli et al^[@bibr8-2058-5241.2.160077]^ consider the primary deforming force to be the plantarflexed first metatarsal, which is thought to be a result of peroneus longus overaction. Relative weakness of the peroneus brevis and tibialis anterior muscles with strong tibialis posterior and peroneus longus muscles cause plantar flexion of the first metatarsal bone and varus of the hindfoot.^[@bibr9-2058-5241.2.160077]^ Recruiting extensor hallucis longus and extensor digitorum longus as secondary ankle dorsiflexors will lead to 'cock-up' deformity of the hallux and clawtoe deformity of the lesser toes. To allow the toe pulp to touch the ground, the flexor muscles of the toes contract, producing clawing of the toes, which is also aggravated by a deficiency of the interosseous muscles. Clawing of the toes accentuates the slope of the metatarsals due to the exaggerated pressure on the metatarsal heads, which in turn increases the tension in the plantar aponeurosis. Additional contracture of the plantar fascia will accentuate the windlass mechanism and further depress the metatarsal heads. Hindfoot varus is described as being forefoot or hindfoot driven. In forefoot-driven varus, excessive plantarflexion of the first metatarsal and supination of the midfoot leads to the hindfoot moving into varus, whereas hindfoot-varus-driven is related to simple varus malalignment of the heel ([Fig. 2](#fig2-2058-5241.2.160077){ref-type="fig"}). {#fig2-2058-5241.2.160077} Hindfoot varus also increases the risk of damage to the lateral structures of the foot and ankle. Thus, peroneal tendons can suffer as a consequence of the hindfoot varus^[@bibr10-2058-5241.2.160077]^ but may also be responsible for the hindfoot varus in cases with relative weakness or paralysis of one or both. The relationship between the varus heel and chronic instability has been well documented; moreover, the heel varus overloads the lateral structures of the foot and ankle and may lead to varus ankle arthritis. Because of hindfoot inversion, the Achilles tendon will shift medially and act as a secondary invertor. Furthermore, patients with cavus feet often have tight calves and a short and tight gastrocnemius leading to increase of the plantar pressures in the forefoot and the plantar fascia and act as a deforming hindfoot inverting force. Radiographic evaluation {#section2-2058-5241.2.160077} ======================= Plain film radiographs are essential in surgical planning, not only to identify the site of the deformity but also to quantify the degree of correction that is required and to decide whether to perform an osteotomy or an arthrodesis. The apex of the deformity can vary. Usually the deformity is located in the mid-foot at the transverse tarsal articulation or at the naviculocuneiform joint.^[@bibr1-2058-5241.2.160077]^ Weight-bearing radiographs of the foot include at least three views: 1. A lateral view of the weight-bearing ankle and foot allows the cavus to be demonstrated and measured. 2. A frontal view of the ankle (Meary view or Salzman view) demonstrates the frontal deformity of the hindfoot.^[@bibr11-2058-5241.2.160077]^ 3. A dorsoplantar view of the forefoot shows adduction of the forefoot and opening of the metatarsal plate. Numerous geometric measurements have been proposed on lateral weight-bearing radiographs to quantify cavus deformity ([Fig. 3](#fig3-2058-5241.2.160077){ref-type="fig"}). In France, the angle of the medial arch is widely used (Djian-Annonier angle) and in pes cavus foot it is less than 120°. A Hibb's angle (angle between the long axis of the calcaneum and first metatarsal) of more than 45° indicates cavus.^[@bibr12-2058-5241.2.160077]^ {#fig3-2058-5241.2.160077} The intersection point between the first metatarsal axis and the sagittal axis of the talus corresponds to the apex of the deformity which is important when considering osteotomies. The cavus foot is defined as a Meary's angle (the angle between the long axes of the talus and first metatarsal) greater than 5°. In posterior cavus foot, the calcaneal pitch angle is greater than 30°. An associated equinus deformity of the ankle is characterised by a tibio-talar angle greater than 105°. On the lateral view, a stacking effect may be observed because the first and medial metatarsals tend to be at a greater inclination---the lateral metatarsals are more horizontal---while the talus appears to be flattened due to the rotation of the talus in the coronal and saggital planes. The medial cuneiform to fifth metatarsal base distance is increased and the fibula appears in a more posterior position related to external rotation of the lower limb. Thickening or fractures of the base of the fifth metatarsal may result from mechanical overload on the lateral border of the foot. Frontal weight-bearing views are very important to assess to the hindfoot alignment during mid-stance ([Fig. 4](#fig4-2058-5241.2.160077){ref-type="fig"}). Evaluation of the hindfoot varus tilt is therefore essential. For this, a frontal view with a Coleman 'block test' is necessary to evaluate the correctability of this hindfoot varus and appreciate the amount of correction of hindfoot varus achieved after surgical intervention. {#fig4-2058-5241.2.160077} Stress views are a useful adjunct in radiographic evaluation to address tibio-talar instability and approach for the reducibility of this talar tilt. The ankle radiographs may also show medially localised or generalised arthritic changes with narrowing and osteophyte formation of the tibiotalar joint. In our experience, MRI or ultrasound are rarely used except to address a severe peroneal tendinopathy when a lateral ligament reconstruction is considered.^[@bibr13-2058-5241.2.160077]^ Also, CT scan or more often single photon-emission computed tomography combined with CT are helpful to assess the hindfoot joints for evidence of arthritis and to specify the site and severity of the degeneration or the presence of an associated lesion. Clinical examination {#section3-2058-5241.2.160077} ==================== Clinical examination is the key to successful management of pes cavus, especially in subtle cavovarus. The aim is to confirm the presence and rigidity of the cavovarus deformity but also to identify any underlying neurological disease. In any case, an evaluation of the entire lower limb is mandatory and calf-wasting or hypertrophy should be noted. A complete neurological examination of both the upper and the lower limbs is needed. To detect any muscular imbalance, a full examination of all muscle groups should be performed for power and graded between 1 and 5 on the Medical Research Council (MRC) scale. Neurological investigations are best performed by a neurologist and electrodiagnostic studies can be considered to confirm hereditary motor sensory neuropathies. Patients can present with a wide range of complaints; among them, pain is the main reason for consultation. Metatarsalgia is most often observed in the anterior cavus foot and talalgias in posterior pes cavus. Local tenderness when wearing shoes and callosities related to clawed toes are common complaints. Flat-heeled shoes are poorly tolerated and high heels are more comfortable. Walking is disturbed by cramp and dull aching in the calf by contraction of the muscle of the plantar arch during prolonged walking. Instability of the ankle leads to repeated sprains in varus. The clinical examination of the foot should begin with the evaluation of the patterns of wear affecting the heels of the shoes, especially on the lateral side in hindfoot varus. Much information can be obtained from mere observation of the weight-bearing posture of the foot (high arch, metatarsus adductus, clawing of the toes, callosity under the first or fifth metatarsal heads, varus heel, prominence or posterior position of the lateral malleolus). In the cavovarus foot, it is not unusual to find marked callosities under the first and fifth metatarsal heads. Passive and active range of movement of the ankle, hindfoot and forefoot should be noted and stability tested. Variations in muscle balance should be tested with resisted inversion and eversion. Hindfoot varus is confirmed through the 'peek-a-boo' heel sign, first described by Manoli et al in 1993, which is the clinical condition whereby the heel is visible on the medial side when viewing the patient from the front with the feet in neutral rotation ([Fig. 5](#fig5-2058-5241.2.160077){ref-type="fig"}). {#fig5-2058-5241.2.160077} The Coleman 'block test' should be performed to ascertain whether hindfoot varus is correctable or not.^[@bibr14-2058-5241.2.160077]^ If the hindfoot varus remains, then the deformity is fixed. However, if the hindfoot corrects to physiologic valgus, then the deformity is flexible and driven by the forefoot deformity. Frequently, the hindfoot varus partially corrects, and it is important to see the magnitude of the heel correction beyond neutral or a varus position. Some authors have suggested manoeuvres for evaluating hindfoot flexibility by placing the patient in a prone position with the knee flexed at 90°. In this position, the foot is allowed to move freely without the influence of the first ray and hindfoot manipulation is easily performed, allowing determination of rigidity. To assess the presence of an isolated gastrocnemius tightness, the Silfverskiöld test is performed by comparing the range of ankle dorsiflexion with the knee in flexion and in extension ([Fig. 6](#fig6-2058-5241.2.160077){ref-type="fig"}).^[@bibr15-2058-5241.2.160077]^ Knee flexion relaxes the gastrocnemius but leaves soleus tension unaffected and a large range of dorsiflexion with the knee flexed means isolated gastrocnemius tightness. Without improvement of ankle dorsiflexion with the knee flexed at 90°, gastrocnemius contracture is diagnosed. This aspect should be addressed at the time of surgery. {#fig6-2058-5241.2.160077} At the end of this clinical investigation, we must separate subtle and severe cavus feet.^[@bibr8-2058-5241.2.160077]^ The subtle cavus foot is easy to misdiagnose because patients often present with symptoms relating to forefoot overload, and most of the time they exhibit a flexible hindfoot with a subtle varus on standing frontal examination. Examination typically reveals a high arch and lesser metatarsalgia is a common complaint. Stress fractures of the lesser metatarsals and fractures of the fifth metatarsal are common. Ankle instability, recurrent sprains and lateral pain are common presenting symptoms. This may be due to lateral overload caused by the hindfoot varus but may also be due to lax lateral ankle ligaments.^[@bibr16-2058-5241.2.160077]-[@bibr18-2058-5241.2.160077]^ Excessive loading on the lateral side of the ankle may lead to peroneal tendon symptoms including tendinopathy, tears, subluxations or dislocations. Anteromedial impingement between talar and tibial spurs has been described and appears to be more common in those with subtle cavus feet. In the hindfoot, the symptoms may include Achilles tendinopathy or plantar fasciitis. By contrast, a severe cavus foot is often the end-product of a longstanding deformity associated with a plantarflexed first ray, or sometimes even severe plantarflexion through the entire midfoot, an increased calcaneal pitch and especially neuromuscular foot imbalance (such as is seen with CMT). This typically includes a fixed heel varus that no longer corrects with Coleman block testing. Treatment {#section4-2058-5241.2.160077} ========= Non-surgical management {#section5-2058-5241.2.160077} ----------------------- A large number of patients with milder symptoms associated with a cavus deformity can be treated successfully by conservative means. In most cases, a reducible deformity can be corrected using a custom orthosis which produces reduced pain and instability.^[@bibr19-2058-5241.2.160077]^ The aim of conservative treatment is to re-align the hindfoot correctly to offload the lateral border of the foot and to overcome the gastrocnemius tightness. The type of orthotic chosen depends on the Coleman 'block test'. In a forefoot-driven cavus with a supple hindfoot, correction of the plantarflexed first ray will allow the hindfoot varus to correct and a first ray recess associated with a metatarsal bar and lateral forefoot post are frequent enough. Furthermore, in front of a hindfoot-driven cavus, the appropriate orthosis includes a lateral hindfoot-to-midfoot heel wedge with a first metatarsal recess and minimal or absent medial arch support. To treat equinus, a gastrocnemius-stretching programme should be initiated and the heel may be slightly elevated. In addition, ankle instability is treated with proprioception training and an ankle support brace worn during exercise. Surgical treatment {#section6-2058-5241.2.160077} ------------------ Surgery is considered if conservative treatments fail to control the symptoms but operative treatment should only be considered in carefully selected patients. The aim of surgery is to achieve a foot that is plantigrade, mobile and pain-free. In any case, surgical treatment should leave the foot in a normal position or slightly overcorrected, because an iatrogenic flat foot is better tolerated than a residual cavus deformity. A wide variety of procedures for the treatment of cavovarus foot deformities have been described including soft-tissue release or lengthening and tendon transfers, hindfoot or midfoot osteotomy, or arthrodesis. Soft-tissue procedures {#section7-2058-5241.2.160077} ---------------------- Initially, the deformities are flexible and reversible but if the muscle imbalance remains the foot becomes stiffer and less adaptable. Joint preservation and decreased deterioration of the deformities can be obtained by balancing of the affected muscles; therefore, various soft-tissue procedures are prefered while the deformity is flexible and corrective arthrodesis or osteotomy must be used when the deformity becomes rigid. Soft-tissue release alone is no longer applicable in fixed deformities in adults. The surgical options to adress idiopathic cavovarus and neuromuscular deformity are not significantly different, but severe rigid cavus deformities are most frequently observed in neurological diseases leading to midfoot correction or triple arthrodesis. In subtle cavovarus foot (idiopathic cavus deformity), the equinus deformity and fixed forefoot deformity are addressed first and a valgus osteotomy should be performed if required.^[@bibr20-2058-5241.2.160077]^ The Achilles tendon must be carefully assessed. If a global gastroc-soleus contracture is present, an Achilles tendon lengthening using a triple hemisection is performed. In some cases, an isolated gastrocnemius tightness is addressed by using a gastrocnemius recession (Strayer or Barouk technique).^[@bibr21-2058-5241.2.160077]^ Achilles tendon lengthening is justified when the heel is in varus, but if the heel is in a neutral position, the increased calcaneal pitch is secondary to Achilles weakness and tendon lengthening should be avoided. To the extent that plantar fascia retraction contributes to the high medial arch, the need for a plantar fascia release is debatable. A subcutaneous plantar fascia release may benefit patients with minimal deformity. In severe fixed pes cavus a complete release as recommanded by Steindler may be advised.^[@bibr22-2058-5241.2.160077]^ Tendon transfers can be used when there is an identifiable muscle imbalance, especially in younger patients with a flexible deformity. It is recommended to transfer only muscle tendon units with a power of MRC 4 or MRC 5.^[@bibr23-2058-5241.2.160077]-[@bibr25-2058-5241.2.160077]^ Tendon transfers are also important after osteotomies for preventing recurrence of the deformity and should be done within the same surgery. Most authors recommend using peroneus longus (PL) to brevis (PB) tendon transfer in subtle cavovarus foot to correct forefoot pronation, reduce the first ray plantarflexion and re-inforce the weak eversion of the hindfoot.^[@bibr26-2058-5241.2.160077]^ Tibialis posterior tendon transfer is more commonly used in hereditary sensory motor neuropathy to weaken the deforming power and strengthen deficient functions of the anterior tibial tendon. Furthermore, this transfer reduces the recruitment of the long toe extensors in assisting in ankle dorsiflexion. Transfer onto the second cuneiform improves neutral dorsiflexion whereas insertion on the cuboid re-inforces a weaker eversion of the foot. Osteotomies {#section8-2058-5241.2.160077} ----------- Dorsal wedge osteotomy of the first metatarsal may be an effective way to decrease the medial forefoot plantarflexion deformity if the pre-operative Coleman 'block test' has confirmed a forefoot-driven pes cavus.^[@bibr27-2058-5241.2.160077]^ The first metatarsal is exposed and a dorsal wedge is excised around 2 cm from the tarsometatarsal joint. For severe first ray deformity with poor correctability, a first tarso-metatarsal dorsiflexion arthrodesis is preferred, which produces a higher degree of forefoot cavus correction than metatarsal osteotomy. If the Meary's angle is markedly increased with poor passive correctability, this dorsal metatarsal osteotomy will be insufficient and a midfoot dorsal wedge osteotomy should be considered. Midfoot dorsal wedge osteotomy is performed at the apex of the deformity and aims to realign the axes of the talus and first metatarsal.^[@bibr28-2058-5241.2.160077]^ Anterior tarsectomy, described by Cole and Meary,^[@bibr29-2058-5241.2.160077],[@bibr30-2058-5241.2.160077]^ is centred on the naviculo-cuneiform space and the cuboid and consists of a dorsomedial closing-wedge osteotomy that predominantly affects the medial rays ([Fig. 7](#fig7-2058-5241.2.160077){ref-type="fig"}).^[@bibr31-2058-5241.2.160077]^ This midfoot dorsal-wedge osteotomy allows correction in the frontal, sagittal and coronal planes without compromising the tarsal inversion/eversion and dorsoplantar motion of the foot. In our experience, one-third of patients were not satisfied with the outcome because of residual mild to moderate pain; moreover, only 20° to 25° of tarsometatarsal correction can be obtained using Cole's osteotomy leaving, in some cases, residual cavus deformity ([Fig. 8](#fig8-2058-5241.2.160077){ref-type="fig"}).^[@bibr32-2058-5241.2.160077]^ {#fig7-2058-5241.2.160077} {#fig8-2058-5241.2.160077} Various other types of osteotomy may be considered. Japas^[@bibr33-2058-5241.2.160077]^ proposed an inverted V osteotomy with two limbs crossing the cuboid on the outer side and the cuneiform bones on the medial arch. We have no experience with this procedure. Wilcox and Weiner^[@bibr34-2058-5241.2.160077]^ proposed intra-cuneiform osteotomy and other authors considered tarsometatarsal resection or metatarsal osteotomy but these procedures do not allow greater corrections to be made.^[@bibr35-2058-5241.2.160077]^ These metatarsal osteotomies lie distal to the apex of the cavus deformity leaving a residual dorsal bony prominence proximally in addition to an inadequate correction of the frontal plane deformity. Hindfoot surgery {#section9-2058-5241.2.160077} ---------------- The Coleman 'block test' and hindfoot alignment view are useful to assess the need for corrective hindfoot surgery. When the hindfoot varus is reducible and forefoot-driven, a valgus calcaneal osteotomy is not required. Conversely, if the hindfoot varus is flexible or if there is residual varus after midfoot surgery, a calcaneal osteotomy must be considered. In case of any doubt, an overcorrection is better than an undercorrection. Various extra-articular osteotomies have been described to correct the hindfoot alignment.^[@bibr36-2058-5241.2.160077]^ In our experience, Dwyer's osteotomy^[@bibr37-2058-5241.2.160077]^ may not provide sufficient correction for severe deformity and the correction is only in one plane. Also, with lateral sliding calcaneal osteotomy there is a limit to the amount of translation that can be obtained. Moderate hindfoot varus associated with subbtle cavovarus is easily adressed with a Dwyer or a lateralising calcaneal osteotomy but severe deformity may require more complex osteotomies. In those cases, the authors usually perform a Z-osteotomy described by Malerba and DiMarchi which removes a lateral wedge of varying thickness that allows multiplanar correction of the calcaneal deformity ([Fig. 9](#fig9-2058-5241.2.160077){ref-type="fig"}).^[@bibr38-2058-5241.2.160077]^ ![Operative view showing a lateral bone wedge osteotomy described by Malerba and DiMarchi.^[@bibr38-2058-5241.2.160077]^](eor-2-221-g009){#fig9-2058-5241.2.160077} In our experience, calcaneal osteotomy in isolated calcaneal pitch (without varus) greater than 30° associated with forefoot cavus is useless because midfoot dorsal wedge osteotomy provides a spontaneous reduction of calcaneal slope. If the talus is tilted in the mortise on the anteroposterior (AP) ankle view, the cavovarus foot is at risk of osteoarthritis related to increased point loading.^[@bibr39-2058-5241.2.160077]^ It is essential to look for a lateral ligament complex deficiency and a reciprocal deltoid ligament tightness which requires lateral ligament reconstruction associated with medial deltoid ligament release. In order for these ligament reconstruction procedures to be effective and long-lasting, the hindfoot varus deformity should be re-aligned by bony correction before the ligament reconstruction and the patient must be aware that there is a significant rate of failure and further surgery may be required in case of progression of their arthritis.^[@bibr40-2058-5241.2.160077]^ Associated toe deformities {#section10-2058-5241.2.160077} -------------------------- Appropriate osteotomies of the midfoot provide some degree of correction of the claw toe deformities and flexible correctible claw toes may improve spontaneously after midfoot osteotomies. If the correction is not sufficient and the toes remain mobile, flexor to extensor tendon transfers (Girdlestone transfer) may be performed to improve the re-alignment of the toes. Once toe deformities have become fixed, a proximal interphalangeal joint arthrodesis or excision arthroplasty is required. In some cases, clawing of the hallux requires a modified Jones procedure with a transfer of the extensor hallucis longustendon through the first metatarsal neck and arthrodesis of the interphalangeal joint.^[@bibr41-2058-5241.2.160077]^ Various others procedures can be performed if there is associated pathology, guided by clinical assessment and imaging. Ankle arthroscopy allows debridment and removal of a large antero-medial osteophytes or anterior ankle synovitis. Repair is recommended for peroneal tendons that have damage to more than 50% of their cross-sectional area. An occasional Jones fracture may require fixation. ### Rigid cavovarus deformity {#section11-2058-5241.2.160077} In adult patients presenting with rigid cavovarus deformity, the management is challenging and fusions and osteotomies represent the mainstay of treatment once the foot has lost its reducibility ([Fig. 10](#fig10-2058-5241.2.160077){ref-type="fig"}). Severe rigid cavus foot requires complex midfoot osteotomies to correct three-dimensional multiplanar deformity. Results of a plantigrade foot obtained by an arthrodesis are better than joint-sparing surgery leading to an inadequate and hence painful correction. However, many authors consider that it is preferable to fuse the minimum number of joints possible in order to maintain as much function as possible. {#fig10-2058-5241.2.160077} In many cases, a triple arthrodesis is indicated,^[@bibr42-2058-5241.2.160077]^ but soft-tissue balancing, by means of tendon transfers, must frequently be included in the correction to maintain a successful result over time. In the absence of the peroneus longus-to-brevis transfer the recurrence of first metatarsal plantarflexion deformity may occur. Likewise, saving the posterior tibial tendon leads to progression of heel varus and causes adductus at the talonavicular level.^[@bibr43-2058-5241.2.160077]^ Midtarsal tarsectomy was recommended by Imhäuser and Steinhäusser.^[@bibr44-2058-5241.2.160077]^ The operation centres on the midtarsal joint space and removes a dorso lateral bone wedge, which allows the cavus and adduction of the forefoot to be corrected. If the cavus is marked, the bone wedge can include removal of the whole of the navicular bone and also encroach on the cuneiform bones. Long-term follow-up studies have shown a high incidence of osteoarthritis of the remaining foot joints after this triple arthrodesis. In order to preserve ankle motion, some authors considered that if less than one-third of the ankle joint is affected, then re-alignment allows a mobile joint relatively free of symptoms. If more than one-third of the ankle joint has degenerative changes the results are less predictable and a tibiotalar fusion could be advised. In conclusion, recent literature indicates that adult cavovarus foot deformities should be commonly addressed with joint preservation osteotomies and adjunctive soft-tissue procedures and less with triple arthrodesis, especially in subtle cavovarus foot. Many surgical options are available to achieve good results and the order in which surgery is performed is important. The fixed forefoot deformity is addressed first and a valgising osteotomy should be performed if required. Any bony correction must be in conjunction with a soft-tissue balancing procedure and residual toe deformities are corrected as a final step. Treatment strategies should be individualised but the difficulty in obtaining a plantigrade and balanced foot using conservative surgery must not be forgotten. **ICMJE Conflict of Interest Statement:** None. Funding ======= No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.

Related literature {#sec1} ================== For the triclinic polymorph of the title compound, see: Chekhlov (2007[@bb3]). Experimental {#sec2} ============ {#sec2.1} ### Crystal data {#sec2.1.1} H~3~O^+^·ClO~4~ ^−^·C~20~H~24~O~6~*M* *~r~* = 478.9Monoclinic,*a* = 8.6586 (1) Å*b* = 26.7718 (3) Å*c* = 19.1518 (2) Åβ = 100.0011 (10)°*V* = 4372.05 (8) Å^3^*Z* = 8Cu *K*α radiationμ = 2.09 mm^−1^*T* = 124 K0.26 × 0.18 × 0.13 mm ### Data collection {#sec2.1.2} Oxford Diffraction Xcalibur Atlas Gemini ultra diffractometerAbsorption correction: multi-scan (*CrysAlis RED*; Oxford Diffraction, 2008[@bb4]) *T* ~min~ = 0.098, *T* ~max~ = 1.00036184 measured reflections6865 independent reflections5283 reflections with *I* \> 3σ(*I*)*R* ~int~ = 0.048 ### Refinement {#sec2.1.3} *R*\[*F* ^2^ \> 2σ(*F* ^2^)\] = 0.051*wR*(*F* ^2^) = 0.124*S* = 2.076865 reflections595 parametersH atoms treated by a mixture of independent and constrained refinementΔρ~max~ = 0.49 e Å^−3^Δρ~min~ = −0.32 e Å^−3^ {#d5e700} Data collection: *CrysAlis CCD* (Oxford Diffraction, 2008[@bb4]); cell refinement: *CrysAlis RED* (Oxford Diffraction, 2008[@bb4]); data reduction: *CrysAlis RED*; program(s) used to solve structure: *SIR2002* (Burla *et al.*, 2003[@bb2]); program(s) used to refine structure: *JANA2006* (Petříček *et al.*, 2006[@bb5]); molecular graphics: *DIAMOND* (Brandenburg & Putz, 2005[@bb1]); software used to prepare material for publication: *JANA2006* and *publCIF* (Westrip, 2010[@bb6]). Supplementary Material ====================== Crystal structure: contains datablocks global, I. DOI: [10.1107/S1600536810048622/hb5736sup1.cif](http://dx.doi.org/10.1107/S1600536810048622/hb5736sup1.cif) Structure factors: contains datablocks I. DOI: [10.1107/S1600536810048622/hb5736Isup2.hkl](http://dx.doi.org/10.1107/S1600536810048622/hb5736Isup2.hkl) Additional supplementary materials: [crystallographic information](http://scripts.iucr.org/cgi-bin/sendsupfiles?hb5736&file=hb5736sup0.html&mime=text/html); [3D view](http://scripts.iucr.org/cgi-bin/sendcif?hb5736sup1&Qmime=cif); [checkCIF report](http://scripts.iucr.org/cgi-bin/paper?hb5736&checkcif=yes) Supplementary data and figures for this paper are available from the IUCr electronic archives (Reference: [HB5736](http://scripts.iucr.org/cgi-bin/sendsup?hb5736)). This work was supperted by the institutional research plan No. AVOZ10100521 of the Institute of Physics, the project Praemium Academiae of the Academy of Sciences of the Czech Republic and the Czech Ministry of Education, Youth and Sports, Project MSM 4977751303. Comment ======= The crystal structure of dibenzo-18-crown-6 hydronium perchlorate was previously published by A.N.Chekhlov (2007). The published structure determined at room temperature is triclinic(space group P-1, a = 8.582 Å, b = 10.486 Å, c = 26.293 Å, α = 79.45°,β = 82.00° and γ = 79.36°, V =2272.5 Å) with asymmetric unit consisting of two independent molecules of macrocycle with complexed hydronium ions. The neutrality of the compound is ensured by two perchlorate anions. The data of crystal structure, presented in this paper, were collected at room temperature (testing stage) and at 120 K (final data collection). We found the complex monoclinic, *P*2~1~/*c* space group, with unit-cell parameters a = 8.6535 Å, b = 26.7823 Å, c = 19.1707 Å, β = 99.9987° and doubled unit cell volume V = 4372.05 Å^3^. The difference between both structures is in their system of hydrogen bonds. In Chekhlov\'s structure, the hydronium ion is held by three hydrogen bonds inside the crown cavity. In presented structure, hydronium ion and crown-ether form only two hydrogen bonds. The third hydrogen atom of hydronium ion is shared with perchlorate anion which makes it to point out of the cavity. This hydrogen bond causes that the perchlorate anions are not disordered as it was observed in Chekhlov\'s structure. Consequently, sharp maxima in difference Fourier map could be used for localizing hydrogen positions in both oxonia cations (Fig 3) and the found hydrogen positions could be refined without restraints. The distance between hydronium and oxygen atoms in macrocycles are 1.637 Å (O21---H1···O3) and 1.864 Å (O21---H9···O5) for one crownether molecule and 1.895 Å (O22---H10···O11) and 1.661 Å (O22---H8···O9) for the other one. The length of hydrogen bond between hydronium ion and perchlorate is 1.732 Å (O21---H7···O17) and 1.687 Å (O22---H6···O14). The distances between hydrogen atoms and oxygen atoms in hydronium correspond to the extent of their participation in hydrogen bonding system: O---H distance close (but still longer) to the standard value 0.983 Å has been only found for the weakest hydrogen bond O22---H10···O11. For stronger hydrogen bonds O---H distance becomes significantly longer, taking the maximum value 1.29 (4) for O22---H6···O14. The O---H and corresponding H···O distances for oxonia are summarized in Table 2. The hydronium ions are enclosed in the crown-ether cavities by phenyl ring of neighbouring molecules. This arrangement is stabilized due to CH-π interactions between phenyl rings and CH~2~ groups of crownether (the distance between centroid of phenyl ring C11→C16 and H37*b* in ethylen group is 2.989 Å and between centriod of phenyl ring C31→C36 and H17*a* in ethylen group is 2.870 Å) and due to the face-to-edge orientation of phenyl rings (distance between the centriod of phenyl ring C21→C26 and H13 of phenyl ring C11→C16 is 3.207 Å and between the centriod of phenyl ring C1→C6 and H33 in phenyl ring C31→C36 is 3.004 Å). Experimental {#experimental} ============ Dibenzo-18-crown-6, perchloric acid and acetonitrile were purchased by Fluka. Crystals were prepared by slow evaporation of equimolar mixture of dibenzo-18-crown-6 (0.05*M*) and perchloric acid (0.05*M*) in acetonitrile to yield colourless prisms of the title compound. Figures ======= {#Fap1} {#Fap2} {#Fap3} Crystal data {#tablewrapcrystaldatalong} ============ ----------------------------------- ---------------------------------------- H~3~O^+^·ClO~4~^−^·C~20~H~24~O~6~ *F*(000) = 2016 *M~r~* = 478.9 *D*~x~ = 1.455 Mg m^−3^ Monoclinic, *P*2~1~/*c* Cu *K*α radiation, λ = 1.5418 Å Hall symbol: -P 2ybc Cell parameters from 20984 reflections *a* = 8.6586 (1) Å θ = 3.3--62.5° *b* = 26.7718 (3) Å µ = 2.09 mm^−1^ *c* = 19.1518 (2) Å *T* = 124 K β = 100.0011 (10)° Prism, colourless *V* = 4372.05 (8) Å^3^ 0.26 × 0.18 × 0.13 mm *Z* = 8 ----------------------------------- ---------------------------------------- Data collection {#tablewrapdatacollectionlong} =============== ------------------------------------------------------------------------------ -------------------------------------- Oxford Diffraction Xcalibur Atlas Gemini ultra diffractometer 6865 independent reflections Radiation source: Enhance Ultra (Cu) X-ray Source 5283 reflections with *I* \> 3σ(*I*) mirror *R*~int~ = 0.048 Detector resolution: 10.3784 pixels mm^-1^ θ~max~ = 62.6°, θ~min~ = 3.3° rotation method data acquisition using ω scans *h* = −9→9 Absorption correction: multi-scan (*CrysAlis RED*; Oxford Diffraction, 2008) *k* = −30→28 *T*~min~ = 0.098, *T*~max~ = 1.000 *l* = −22→21 36184 measured reflections ------------------------------------------------------------------------------ -------------------------------------- Refinement {#tablewraprefinementdatalong} ========== ------------------------------------- --------------------------------------------------------------------------------- Refinement on *F*^2^ 198 constraints *R*\[*F*^2^ \> 2σ(*F*^2^)\] = 0.051 H atoms treated by a mixture of independent and constrained refinement *wR*(*F*^2^) = 0.124 Weighting scheme based on measured s.u.\'s *w* = 1/\[σ^2^(*I*) + 0.0016*I*^2^\] *S* = 2.07 (Δ/σ)~max~ = 0.022 6865 reflections Δρ~max~ = 0.49 e Å^−3^ 595 parameters Δρ~min~ = −0.32 e Å^−3^ 0 restraints ------------------------------------- --------------------------------------------------------------------------------- Special details {#specialdetails} =============== --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Refinement. The refinement was carried out against all reflections. The conventional *R*-factor is always based on *F*. The goodness of fit as well as the weighted *R*-factor are based on *F* and *F*^2^ for refinement carried out on *F* and *F*^2^, respectively. The threshold expression is used only for calculating *R*-factors *etc*. and it is not relevant to the choice of reflections for refinement.All the H atoms were discernible in difference Fourier maps and could be refined to reasonable geometry. Despite of it the H atoms bonded to carbon atoms were constrained to ideal positions. The O---H distances and angles in hydronium ions were not restrained. The isotropic temperature parameters of hydrogen atoms were calculated as 1.2\**U*~eq~ of the parent atom.The program used for refinement, Jana2006, uses the weighting scheme based on the experimental expectations, see \_refine_ls_weighting_details. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å^2^) {#tablewrapcoords} ================================================================================================== ------ ------------- -------------- -------------- -------------------- -- *x* *y* *z* *U*~iso~\*/*U*~eq~ Cl1 0.25806 (7) 0.16885 (2) 0.35587 (3) 0.0248 (2) Cl2 0.74151 (7) 0.41184 (2) 0.38383 (4) 0.0261 (2) O1 1.0665 (2) 0.50098 (6) 0.15146 (9) 0.0252 (6) O2 1.2316 (2) 0.46140 (6) 0.26441 (9) 0.0238 (6) O3 1.1821 (2) 0.35876 (6) 0.28737 (9) 0.0270 (6) O4 0.9437 (2) 0.30634 (6) 0.19288 (9) 0.0249 (6) O5 0.7465 (2) 0.35281 (6) 0.09651 (9) 0.0271 (6) O6 0.8039 (2) 0.45403 (6) 0.06355 (9) 0.0243 (6) O7 0.5802 (2) 0.08959 (6) 0.11052 (9) 0.0263 (6) O8 0.7525 (2) 0.12376 (6) 0.22540 (9) 0.0243 (6) O9 0.7056 (2) 0.22401 (6) 0.27262 (9) 0.0268 (6) O10 0.4558 (2) 0.28100 (6) 0.19535 (10) 0.0267 (6) O11 0.2643 (2) 0.24193 (6) 0.08853 (9) 0.0284 (6) O12 0.3196 (2) 0.14350 (6) 0.03615 (10) 0.0266 (6) O13 0.4098 (2) 0.17956 (8) 0.39605 (12) 0.0404 (7) O14 0.2617 (2) 0.17463 (8) 0.28118 (10) 0.0400 (8) O15 0.1434 (2) 0.20252 (7) 0.37496 (10) 0.0327 (7) O16 0.2159 (2) 0.11841 (7) 0.36877 (11) 0.0348 (7) O17 0.7348 (2) 0.40931 (8) 0.30779 (10) 0.0399 (8) O18 0.6961 (2) 0.46121 (6) 0.40192 (10) 0.0298 (6) O19 0.8976 (2) 0.40152 (8) 0.41890 (12) 0.0452 (8) O20 0.6341 (2) 0.37595 (7) 0.40378 (11) 0.0345 (7) O21 0.9470 (3) 0.41200 (9) 0.20320 (14) 0.0495 (9) O22 0.4740 (3) 0.17637 (9) 0.17481 (13) 0.0478 (9) C1 1.2210 (3) 0.51595 (9) 0.16739 (14) 0.0213 (8) C2 1.2909 (3) 0.54843 (9) 0.12647 (14) 0.0249 (9) C3 1.4465 (3) 0.56198 (10) 0.14712 (14) 0.0273 (9) C4 1.5327 (3) 0.54337 (10) 0.20921 (15) 0.0272 (9) C5 1.4644 (3) 0.50926 (9) 0.25017 (14) 0.0251 (9) C6 1.3092 (3) 0.49536 (9) 0.22946 (14) 0.0218 (8) C7 1.3267 (3) 0.43350 (10) 0.32048 (13) 0.0243 (9) C8 1.2279 (3) 0.39347 (10) 0.34457 (14) 0.0271 (9) C9 1.0956 (3) 0.31719 (9) 0.30858 (14) 0.0277 (9) C10 1.0523 (3) 0.28207 (10) 0.24743 (13) 0.0251 (9) C11 0.8679 (3) 0.27715 (9) 0.13818 (14) 0.0235 (9) C12 0.8904 (3) 0.22609 (10) 0.13211 (14) 0.0263 (9) C13 0.8027 (3) 0.19963 (10) 0.07643 (15) 0.0295 (10) C14 0.6942 (3) 0.22368 (10) 0.02681 (16) 0.0315 (10) C15 0.6734 (3) 0.27535 (10) 0.03163 (15) 0.0300 (9) C16 0.7594 (3) 0.30152 (9) 0.08723 (14) 0.0249 (9) C17 0.6510 (3) 0.37942 (10) 0.03977 (14) 0.0264 (9) C18 0.6510 (3) 0.43353 (10) 0.05990 (14) 0.0267 (9) C19 0.8080 (3) 0.50524 (9) 0.08415 (14) 0.0264 (9) C20 0.9729 (3) 0.52437 (10) 0.09165 (14) 0.0251 (9) C21 0.7327 (3) 0.07209 (9) 0.12468 (14) 0.0224 (8) C22 0.7959 (3) 0.03910 (10) 0.08188 (14) 0.0268 (9) C23 0.9502 (3) 0.02291 (10) 0.10120 (15) 0.0295 (9) C24 1.0392 (3) 0.04011 (10) 0.16306 (15) 0.0285 (9) C25 0.9785 (3) 0.07432 (9) 0.20588 (15) 0.0262 (9) C26 0.8251 (3) 0.09029 (9) 0.18699 (14) 0.0232 (8) C27 0.8501 (3) 0.14721 (9) 0.28477 (13) 0.0237 (8) C28 0.7525 (3) 0.18269 (10) 0.31932 (14) 0.0272 (9) C29 0.6189 (3) 0.26055 (9) 0.30496 (15) 0.0277 (9) C30 0.5650 (3) 0.30146 (10) 0.25339 (14) 0.0258 (9) C31 0.3730 (3) 0.31430 (10) 0.14860 (14) 0.0247 (9) C32 0.3859 (3) 0.36568 (10) 0.15492 (15) 0.0291 (10) C33 0.2916 (3) 0.39643 (11) 0.10659 (15) 0.0336 (10) C34 0.1876 (4) 0.37606 (11) 0.05210 (16) 0.0362 (10) C35 0.1769 (4) 0.32457 (10) 0.04384 (16) 0.0329 (10) C36 0.2687 (3) 0.29363 (9) 0.09150 (14) 0.0247 (9) C37 0.1739 (3) 0.22021 (10) 0.02603 (14) 0.0287 (9) C39 0.3192 (3) 0.09096 (9) 0.04725 (14) 0.0269 (9) C38 0.1685 (3) 0.16499 (10) 0.03607 (15) 0.0291 (9) C40 0.4813 (3) 0.07089 (10) 0.04808 (14) 0.0255 (9) H2 1.231405 0.561714 0.083496 0.0299\* H3 1.494768 0.584392 0.118191 0.0328\* H4 1.63932 0.55383 0.224244 0.0326\* H5 1.525077 0.49554 0.292542 0.0302\* H12 0.966406 0.209052 0.166353 0.0316\* H13 0.81811 0.164292 0.072615 0.0354\* H14 0.632921 0.205091 −0.011045 0.0378\* H15 0.599728 0.292466 −0.003542 0.036\* H22 0.733316 0.027309 0.038698 0.0321\* H23 0.994135 −0.000095 0.071569 0.0354\* H24 1.144758 0.028339 0.17691 0.0342\* H25 1.042704 0.086749 0.248246 0.0314\* H32 0.459801 0.380137 0.192675 0.0349\* H33 0.299694 0.432064 0.111515 0.0403\* H34 0.121755 0.397456 0.019413 0.0435\* H35 0.105429 0.310504 0.004858 0.0395\* H7a 1.364134 0.455426 0.359401 0.0292\* H7b 1.413162 0.41857 0.302888 0.0292\* H8a 1.287265 0.376258 0.384371 0.0325\* H8b 1.135985 0.408146 0.357785 0.0325\* H9a 1.002067 0.329164 0.323545 0.0332\* H9b 1.159082 0.299958 0.347238 0.0332\* H10a 1.144849 0.272934 0.229166 0.0301\* H10b 1.004393 0.252726 0.263034 0.0301\* H17a 0.545754 0.366738 0.033141 0.0317\* H17b 0.694493 0.375793 −0.002769 0.0317\* H18a 0.578391 0.451419 0.02519 0.0321\* H18b 0.6189 0.436793 0.105205 0.0321\* H19a 0.772303 0.508381 0.128674 0.0317\* H19b 0.741144 0.524343 0.04869 0.0317\* H20a 1.01435 0.516511 0.049624 0.0302\* H20b 0.973457 0.559875 0.098682 0.0302\* H27a 0.932884 0.165285 0.268531 0.0285\* H27b 0.893741 0.122157 0.318336 0.0285\* H28a 0.661042 0.165685 0.329117 0.0326\* H28b 0.812898 0.194786 0.362846 0.0326\* H29a 0.684642 0.274275 0.345999 0.0333\* H29b 0.529575 0.244913 0.319251 0.0333\* H30a 0.514058 0.326949 0.27639 0.031\* H30b 0.653423 0.315137 0.235919 0.031\* H37a 0.222792 0.22752 −0.014119 0.0344\* H37b 0.069301 0.233425 0.018916 0.0344\* H39a 0.247671 0.07545 0.009571 0.0323\* H39b 0.287414 0.084002 0.091848 0.0323\* H38a 0.13522 0.157791 0.080279 0.0349\* H38b 0.094171 0.150572 −0.001568 0.0349\* H40a 0.478954 0.035064 0.049777 0.0306\* H40b 0.519586 0.082359 0.006772 0.0306\* H1 1.043 (4) 0.3869 (13) 0.2320 (19) 0.0594\* H6 0.384 (4) 0.1733 (12) 0.2216 (18) 0.0574\* H7 0.855 (4) 0.4148 (12) 0.2446 (19) 0.0594\* H8 0.572 (4) 0.2039 (12) 0.2057 (18) 0.0574\* H9 0.893 (4) 0.3913 (13) 0.156 (2) 0.0594\* H10 0.421 (4) 0.1965 (13) 0.1302 (19) 0.0574\* ------ ------------- -------------- -------------- -------------------- -- Atomic displacement parameters (Å^2^) {#tablewrapadps} ===================================== ----- ------------- ------------- ------------- -------------- -------------- -------------- *U*^11^ *U*^22^ *U*^33^ *U*^12^ *U*^13^ *U*^23^ Cl1 0.0258 (3) 0.0217 (3) 0.0262 (3) −0.0007 (3) 0.0022 (3) 0.0018 (3) Cl2 0.0255 (3) 0.0245 (3) 0.0287 (4) 0.0012 (3) 0.0055 (3) −0.0003 (3) O1 0.0219 (10) 0.0248 (10) 0.0276 (10) −0.0008 (8) 0.0004 (8) 0.0063 (8) O2 0.0254 (10) 0.0211 (9) 0.0240 (10) −0.0010 (8) 0.0020 (8) 0.0068 (8) O3 0.0385 (11) 0.0212 (10) 0.0217 (10) −0.0086 (8) 0.0057 (8) −0.0004 (8) O4 0.0308 (11) 0.0193 (9) 0.0222 (10) 0.0028 (8) −0.0023 (8) 0.0007 (8) O5 0.0345 (11) 0.0176 (9) 0.0258 (10) 0.0015 (8) −0.0042 (8) 0.0004 (8) O6 0.0259 (10) 0.0161 (9) 0.0295 (10) 0.0001 (8) 0.0010 (8) −0.0017 (8) O7 0.0255 (10) 0.0250 (10) 0.0268 (10) 0.0005 (8) −0.0003 (8) −0.0064 (8) O8 0.0269 (10) 0.0213 (9) 0.0236 (10) 0.0012 (8) 0.0014 (8) −0.0051 (8) O9 0.0396 (11) 0.0189 (9) 0.0227 (10) 0.0054 (8) 0.0074 (8) 0.0015 (8) O10 0.0341 (11) 0.0181 (9) 0.0257 (10) −0.0008 (8) −0.0009 (8) −0.0012 (8) O11 0.0358 (11) 0.0207 (10) 0.0259 (10) 0.0012 (8) −0.0022 (9) −0.0016 (8) O12 0.0273 (10) 0.0207 (10) 0.0309 (11) 0.0011 (8) 0.0027 (8) −0.0005 (8) O13 0.0249 (11) 0.0429 (13) 0.0482 (14) −0.0092 (9) −0.0080 (10) −0.0016 (10) O14 0.0500 (14) 0.0495 (13) 0.0231 (11) 0.0046 (10) 0.0135 (10) 0.0041 (9) O15 0.0371 (12) 0.0303 (11) 0.0304 (11) 0.0104 (9) 0.0048 (9) −0.0022 (9) O16 0.0360 (12) 0.0173 (10) 0.0503 (13) −0.0042 (8) 0.0053 (10) 0.0068 (9) O17 0.0541 (14) 0.0455 (13) 0.0241 (11) −0.0078 (10) 0.0185 (10) −0.0074 (9) O18 0.0388 (12) 0.0193 (10) 0.0300 (11) 0.0057 (8) 0.0027 (9) −0.0035 (8) O19 0.0232 (11) 0.0461 (13) 0.0626 (16) 0.0111 (10) −0.0032 (10) 0.0050 (11) O20 0.0379 (12) 0.0254 (10) 0.0423 (12) −0.0069 (9) 0.0127 (10) 0.0065 (9) O21 0.0451 (14) 0.0437 (14) 0.0543 (16) −0.0028 (11) −0.0068 (12) 0.0096 (12) O22 0.0459 (14) 0.0455 (14) 0.0467 (15) 0.0057 (11) −0.0070 (12) −0.0087 (11) C1 0.0202 (13) 0.0173 (13) 0.0258 (14) 0.0009 (11) 0.0026 (11) −0.0028 (11) C2 0.0291 (15) 0.0199 (14) 0.0259 (15) 0.0013 (12) 0.0056 (12) 0.0015 (11) C3 0.0301 (16) 0.0248 (15) 0.0276 (15) −0.0035 (12) 0.0066 (12) 0.0018 (12) C4 0.0256 (15) 0.0225 (14) 0.0334 (16) −0.0006 (12) 0.0052 (12) 0.0005 (12) C5 0.0280 (15) 0.0214 (14) 0.0255 (15) 0.0006 (12) 0.0031 (12) −0.0002 (11) C6 0.0247 (14) 0.0153 (13) 0.0262 (14) −0.0015 (11) 0.0064 (11) −0.0005 (11) C7 0.0290 (15) 0.0231 (14) 0.0192 (14) 0.0018 (12) −0.0005 (11) 0.0014 (11) C8 0.0343 (16) 0.0257 (14) 0.0207 (15) −0.0046 (12) 0.0030 (12) −0.0009 (12) C9 0.0345 (16) 0.0233 (14) 0.0244 (15) −0.0069 (12) 0.0029 (12) 0.0047 (12) C10 0.0299 (15) 0.0184 (13) 0.0255 (15) −0.0004 (11) 0.0010 (12) 0.0065 (11) C11 0.0291 (15) 0.0203 (14) 0.0213 (14) −0.0042 (12) 0.0052 (12) −0.0005 (11) C12 0.0325 (16) 0.0212 (14) 0.0259 (15) −0.0002 (12) 0.0066 (12) 0.0030 (12) C13 0.0392 (17) 0.0189 (14) 0.0313 (16) 0.0001 (12) 0.0090 (13) 0.0005 (12) C14 0.0364 (17) 0.0267 (16) 0.0297 (16) −0.0060 (13) 0.0016 (13) −0.0059 (13) C15 0.0343 (17) 0.0247 (15) 0.0284 (16) −0.0012 (13) −0.0018 (13) 0.0011 (12) C16 0.0299 (15) 0.0177 (14) 0.0264 (15) −0.0018 (11) 0.0030 (12) 0.0013 (11) C17 0.0280 (15) 0.0235 (14) 0.0252 (15) 0.0006 (12) −0.0023 (12) 0.0052 (12) C18 0.0226 (14) 0.0254 (15) 0.0296 (16) −0.0002 (12) −0.0026 (12) 0.0029 (12) C19 0.0281 (15) 0.0183 (14) 0.0305 (16) 0.0036 (11) −0.0014 (12) −0.0019 (12) C20 0.0309 (15) 0.0191 (14) 0.0236 (14) 0.0045 (12) 0.0000 (12) 0.0029 (11) C21 0.0215 (14) 0.0190 (13) 0.0267 (15) 0.0018 (11) 0.0042 (11) 0.0031 (11) C22 0.0329 (16) 0.0234 (15) 0.0235 (15) −0.0008 (12) 0.0038 (12) −0.0005 (12) C23 0.0333 (16) 0.0263 (15) 0.0299 (16) 0.0043 (13) 0.0084 (13) 0.0002 (12) C24 0.0262 (15) 0.0244 (15) 0.0351 (17) 0.0018 (12) 0.0060 (13) 0.0004 (12) C25 0.0303 (15) 0.0222 (14) 0.0255 (15) −0.0013 (12) 0.0030 (12) 0.0010 (12) C26 0.0268 (15) 0.0179 (13) 0.0250 (15) −0.0008 (11) 0.0046 (12) −0.0015 (11) C27 0.0261 (15) 0.0216 (14) 0.0213 (14) −0.0035 (11) −0.0020 (11) 0.0003 (11) C28 0.0348 (17) 0.0242 (14) 0.0215 (14) 0.0006 (12) 0.0021 (12) 0.0011 (12) C29 0.0341 (16) 0.0234 (14) 0.0263 (15) 0.0016 (12) 0.0070 (12) −0.0051 (12) C30 0.0298 (15) 0.0207 (14) 0.0262 (15) −0.0025 (12) 0.0027 (12) −0.0057 (11) C31 0.0306 (16) 0.0230 (14) 0.0223 (14) 0.0015 (12) 0.0091 (12) 0.0019 (11) C32 0.0374 (17) 0.0227 (15) 0.0296 (16) −0.0008 (12) 0.0125 (13) −0.0023 (12) C33 0.0499 (19) 0.0211 (15) 0.0322 (17) 0.0052 (14) 0.0139 (15) 0.0014 (13) C34 0.0502 (19) 0.0265 (16) 0.0329 (17) 0.0116 (14) 0.0096 (15) 0.0059 (13) C35 0.0414 (18) 0.0289 (16) 0.0277 (16) 0.0040 (13) 0.0039 (14) 0.0001 (13) C36 0.0320 (15) 0.0178 (14) 0.0248 (14) 0.0015 (12) 0.0067 (12) −0.0001 (11) C37 0.0306 (16) 0.0287 (15) 0.0240 (15) 0.0015 (12) −0.0027 (12) −0.0041 (12) C39 0.0288 (15) 0.0221 (14) 0.0284 (15) −0.0032 (12) 0.0010 (12) −0.0019 (12) C38 0.0244 (15) 0.0307 (15) 0.0300 (16) 0.0023 (12) −0.0014 (12) −0.0050 (13) C40 0.0315 (15) 0.0212 (14) 0.0221 (14) −0.0003 (12) −0.0001 (12) −0.0044 (11) ----- ------------- ------------- ------------- -------------- -------------- -------------- Geometric parameters (Å, °) {#tablewrapgeomlong} =========================== ------------------- ------------- ------------------- ----------- Cl1---O13 1.4313 (19) C13---C14 1.375 (4) Cl1---O14 1.445 (2) C13---H13 0.96 Cl1---O15 1.435 (2) C14---C15 1.400 (4) Cl1---O16 1.4315 (19) C14---H14 0.96 Cl2---O17 1.449 (2) C15---C16 1.380 (4) Cl2---O18 1.4385 (19) C15---H15 0.96 Cl2---O19 1.428 (2) C17---C18 1.499 (4) Cl2---O20 1.434 (2) C17---H17a 0.96 O1---C1 1.379 (3) C17---H17b 0.96 O1---C20 1.428 (3) C18---H18a 0.96 O2---C6 1.372 (3) C18---H18b 0.96 O2---C7 1.443 (3) C19---C20 1.499 (4) O3---C8 1.439 (3) C19---H19a 0.96 O3---C9 1.438 (3) C19---H19b 0.96 O4---C10 1.434 (3) C20---H20a 0.96 O4---C11 1.379 (3) C20---H20b 0.96 O5---C16 1.392 (3) C21---C22 1.381 (4) O5---C17 1.435 (3) C21---C26 1.403 (3) O6---C18 1.423 (3) C22---C23 1.392 (4) O6---C19 1.426 (3) C22---H22 0.96 O7---C21 1.383 (3) C23---C24 1.376 (4) O7---C40 1.435 (3) C23---H23 0.96 O8---C26 1.378 (3) C24---C25 1.392 (4) O8---C27 1.438 (3) C24---H24 0.96 O9---C28 1.436 (3) C25---C26 1.382 (4) O9---C29 1.437 (3) C25---H25 0.96 O10---C30 1.437 (3) C27---C28 1.499 (4) O10---C31 1.374 (3) C27---H27a 0.96 O11---C36 1.385 (3) C27---H27b 0.96 O11---C37 1.435 (3) C28---H28a 0.96 O12---C39 1.423 (3) C28---H28b 0.96 O12---C38 1.429 (3) C29---C30 1.494 (4) C1---C2 1.379 (4) C29---H29a 0.96 C1---C6 1.409 (3) C29---H29b 0.96 C2---C3 1.386 (4) C30---H30a 0.96 C2---H2 0.96 C30---H30b 0.96 C3---C4 1.382 (4) C31---C32 1.384 (4) C3---H3 0.96 C31---C36 1.405 (4) C4---C5 1.400 (4) C32---C33 1.392 (4) C4---H4 0.96 C32---H32 0.96 C5---C6 1.384 (4) C33---C34 1.368 (4) C5---H5 0.96 C33---H33 0.96 C7---C8 1.493 (4) C34---C35 1.389 (4) C7---H7a 0.96 C34---H34 0.96 C7---H7b 0.96 C35---C36 1.378 (4) C8---H8a 0.96 C35---H35 0.96 C8---H8b 0.96 C37---C38 1.493 (4) C9---C10 1.498 (4) C37---H37a 0.96 C9---H9a 0.96 C37---H37b 0.96 C9---H9b 0.96 C39---C40 1.500 (4) C10---H10a 0.96 C39---H39a 0.96 C10---H10b 0.96 C39---H39b 0.96 C11---C12 1.388 (4) C38---H38a 0.96 C11---C16 1.393 (3) C38---H38b 0.96 C12---C13 1.391 (4) C40---H40a 0.96 C12---H12 0.96 C40---H40b 0.96 O13---Cl1---O14 109.58 (13) O6---C19---C20 109.4 (2) O13---Cl1---O15 110.37 (12) O6---C19---H19a 109.4711 O13---Cl1---O16 109.51 (12) O6---C19---H19b 109.4709 O14---Cl1---O15 108.66 (12) C20---C19---H19a 109.4713 O14---Cl1---O16 108.79 (13) C20---C19---H19b 109.4711 O15---Cl1---O16 109.90 (12) H19a---C19---H19b 109.4972 O17---Cl2---O18 108.66 (12) O1---C20---C19 109.0 (2) O17---Cl2---O19 109.47 (14) O1---C20---H20a 109.4717 O17---Cl2---O20 108.77 (12) O1---C20---H20b 109.4714 O18---Cl2---O19 109.90 (12) C19---C20---H20a 109.4707 O18---Cl2---O20 109.54 (12) C19---C20---H20b 109.4714 O19---Cl2---O20 110.47 (12) H20a---C20---H20b 109.9286 C1---O1---C20 116.4 (2) O7---C21---C22 124.3 (2) C6---O2---C7 116.22 (19) O7---C21---C26 115.7 (2) C8---O3---C9 111.9 (2) C22---C21---C26 120.0 (2) C10---O4---C11 117.51 (19) C21---C22---C23 120.1 (2) C16---O5---C17 116.21 (19) C21---C22---H22 119.9264 C18---O6---C19 111.29 (19) C23---C22---H22 119.9263 C21---O7---C40 117.2 (2) C22---C23---C24 119.5 (3) C26---O8---C27 116.48 (19) C22---C23---H23 120.231 C28---O9---C29 111.6 (2) C24---C23---H23 120.2312 C30---O10---C31 117.10 (19) C23---C24---C25 121.0 (3) C36---O11---C37 116.52 (19) C23---C24---H24 119.4897 C39---O12---C38 111.8 (2) C25---C24---H24 119.4882 O1---C1---C2 124.2 (2) C24---C25---C26 119.5 (2) O1---C1---C6 115.8 (2) C24---C25---H25 120.227 C2---C1---C6 120.0 (2) C26---C25---H25 120.2259 C1---C2---C3 120.3 (2) O8---C26---C21 115.7 (2) C1---C2---H2 119.8743 O8---C26---C25 124.6 (2) C3---C2---H2 119.8742 C21---C26---C25 119.7 (2) C2---C3---C4 120.3 (3) O8---C27---C28 109.1 (2) C2---C3---H3 119.8663 O8---C27---H27a 109.4717 C4---C3---H3 119.866 O8---C27---H27b 109.4708 C3---C4---C5 120.0 (2) C28---C27---H27a 109.4709 C3---C4---H4 120.0034 C28---C27---H27b 109.4711 C5---C4---H4 120.0038 H27a---C27---H27b 109.8839 C4---C5---C6 119.9 (2) O9---C28---C27 109.5 (2) C4---C5---H5 120.0572 O9---C28---H28a 109.4715 C6---C5---H5 120.0574 O9---C28---H28b 109.4712 O2---C6---C1 115.6 (2) C27---C28---H28a 109.4707 O2---C6---C5 124.9 (2) C27---C28---H28b 109.4711 C1---C6---C5 119.6 (2) H28a---C28---H28b 109.4873 O2---C7---C8 108.6 (2) O9---C29---C30 109.9 (2) O2---C7---H7a 109.4716 O9---C29---H29a 109.4711 O2---C7---H7b 109.471 O9---C29---H29b 109.4712 C8---C7---H7a 109.4713 C30---C29---H29a 109.4714 C8---C7---H7b 109.4716 C30---C29---H29b 109.4712 H7a---C7---H7b 110.3703 H29a---C29---H29b 109.0288 O3---C8---C7 109.0 (2) O10---C30---C29 108.7 (2) O3---C8---H8a 109.4704 O10---C30---H30a 109.4715 O3---C8---H8b 109.4715 O10---C30---H30b 109.4713 C7---C8---H8a 109.4715 C29---C30---H30a 109.4711 C7---C8---H8b 109.4715 C29---C30---H30b 109.4714 H8a---C8---H8b 109.952 H30a---C30---H30b 110.27 O3---C9---C10 109.7 (2) O10---C31---C32 124.3 (2) O3---C9---H9a 109.471 O10---C31---C36 116.3 (2) O3---C9---H9b 109.4715 C32---C31---C36 119.4 (2) C10---C9---H9a 109.4714 C31---C32---C33 120.1 (2) C10---C9---H9b 109.4705 C31---C32---H32 119.973 H9a---C9---H9b 109.2636 C33---C32---H32 119.9722 O4---C10---C9 109.2 (2) C32---C33---C34 120.2 (3) O4---C10---H10a 109.4713 C32---C33---H33 119.8784 O4---C10---H10b 109.4713 C34---C33---H33 119.8778 C9---C10---H10a 109.4712 C33---C34---C35 120.3 (3) C9---C10---H10b 109.4714 C33---C34---H34 119.8372 H10a---C10---H10b 109.7634 C35---C34---H34 119.8381 O4---C11---C12 124.5 (2) C34---C35---C36 120.1 (3) O4---C11---C16 116.1 (2) C34---C35---H35 119.9488 C12---C11---C16 119.3 (2) C36---C35---H35 119.9493 C11---C12---C13 120.0 (2) O11---C36---C31 115.8 (2) C11---C12---H12 120.0012 O11---C36---C35 124.4 (2) C13---C12---H12 119.9988 C31---C36---C35 119.8 (2) C12---C13---C14 120.5 (3) O11---C37---C38 108.6 (2) C12---C13---H13 119.7378 O11---C37---H37a 109.4712 C14---C13---H13 119.7376 O11---C37---H37b 109.4723 C13---C14---C15 119.8 (2) C38---C37---H37a 109.4709 C13---C14---H14 120.0818 C38---C37---H37b 109.4708 C15---C14---H14 120.0821 H37a---C37---H37b 110.3319 C14---C15---C16 119.6 (2) O12---C39---C40 109.2 (2) C14---C15---H15 120.1961 O12---C39---H39a 109.4712 C16---C15---H15 120.1939 O12---C39---H39b 109.4706 O5---C16---C11 115.6 (2) C40---C39---H39a 109.4708 O5---C16---C15 123.7 (2) C40---C39---H39b 109.4716 C11---C16---C15 120.7 (2) H39a---C39---H39b 109.7323 O5---C17---C18 108.1 (2) O12---C38---C37 110.4 (2) O5---C17---H17a 109.4708 O12---C38---H38a 109.471 O5---C17---H17b 109.471 O12---C38---H38b 109.4713 C18---C17---H17a 109.4713 C37---C38---H38a 109.4714 C18---C17---H17b 109.4714 C37---C38---H38b 109.4718 H17a---C17---H17b 110.8097 H38a---C38---H38b 108.4738 O6---C18---C17 110.1 (2) O7---C40---C39 107.8 (2) O6---C18---H18a 109.4711 O7---C40---H40a 109.4719 O6---C18---H18b 109.4712 O7---C40---H40b 109.4709 C17---C18---H18a 109.4709 C39---C40---H40a 109.4714 C17---C18---H18b 109.4714 C39---C40---H40b 109.4709 H18a---C18---H18b 108.827 H40a---C40---H40b 111.0734 ?---?---?---? ? ------------------- ------------- ------------------- ----------- Hydrogen-bond geometry (Å, °) {#tablewraphbondslong} ============================= ------------------------------------------------------------------------------------------------------------- Cg1, Cg2, Cg3 and Cg4 are the centroids of the C31--C36, C11--C16, C21--C26 and C1--C6 rings, respectively. ------------------------------------------------------------------------------------------------------------- ----------------------- ---------- ---------- ----------- --------------- *D*---H···*A* *D*---H H···*A* *D*···*A* *D*---H···*A* O21---H1···O3 1.14 (3) 1.64 (3) 2.763 (3) 168 (3) O21---H1···O4 1.14 (3) 2.39 (3) 2.835 (3) 101 (2) O21---H7···O17 1.22 (4) 1.73 (4) 2.945 (4) 171 (3) O22---H8···O9 1.20 (3) 1.66 (3) 2.802 (3) 157 (3) O22---H8···O10 1.20 (3) 2.29 (3) 2.837 (3) 104.4 (19) O21---H9···O4 1.09 (3) 2.40 (3) 2.835 (3) 102 (2) O21---H9···O5 1.09 (3) 1.87 (3) 2.910 (3) 159 (3) O21---H9···O6 1.09 (3) 2.47 (4) 2.967 (3) 107 (2) O22---H10···O11 1.05 (3) 1.90 (3) 2.840 (3) 149 (3) O22---H10···O12 1.05 (3) 2.34 (3) 2.895 (3) 112 (2) C5---H5···O18^i^ 0.96 2.52 3.479 (3) 177 C8---H8b···O19 0.96 2.55 3.416 (3) 150 C15---H15···O13^ii^ 0.96 2.42 3.369 (3) 169.02 C20---H20b···O16^iii^ 0.96 2.43 3.165 (3) 134 C35---H35···O15^ii^ 0.96 2.59 3.278 (4) 129 C38---H38b···O19^iv^ 0.96 2.50 3.447 (3) 168 C17---H17a···Cg1 0.96 2.87 3.704 (3) 146 C37---H37b···Cg2^v^ 0.96 2.99 3.825 (3) 146 C13---H13···Cg3 0.96 3.20 4.070 (3) 150 C33---H33···Cg4^v^ 0.96 3.00 3.899 (3) 156 ----------------------- ---------- ---------- ----------- --------------- Symmetry codes: (i) *x*+1, *y*, *z*; (ii) *x*, −*y*+1/2, *z*−1/2; (iii) −*x*+1, *y*+1/2, −*z*+1/2; (iv) *x*−1, −*y*+1/2, *z*−1/2; (v) *x*−1, *y*, *z*. ###### Hydrogen-bond geometry (Å, °) *Cg*1, *Cg*2, *Cg*3 and *Cg*4 are the centroids of the C31--C36, C11--C16, C21--C26 and C1--C6 rings, respectively. *D*---H⋯*A* *D*---H H⋯*A* *D*⋯*A* *D*---H⋯*A* ----------------------- ---------- ---------- ----------- ------------- O21---H1⋯O3 1.14 (3) 1.64 (3) 2.763 (3) 168 (3) O21---H1⋯O4 1.14 (3) 2.39 (3) 2.835 (3) 101 (2) O21---H7⋯O17 1.22 (4) 1.73 (4) 2.945 (4) 171 (3) O22---H8⋯O9 1.20 (3) 1.66 (3) 2.802 (3) 157 (3) O22---H8⋯O10 1.20 (3) 2.29 (3) 2.837 (3) 104.4 (19) O21---H9⋯O4 1.09 (3) 2.40 (3) 2.835 (3) 102 (2) O21---H9⋯O5 1.09 (3) 1.87 (3) 2.910 (3) 159 (3) O21---H9⋯O6 1.09 (3) 2.47 (4) 2.967 (3) 107 (2) O22---H10⋯O11 1.05 (3) 1.90 (3) 2.840 (3) 149 (3) O22---H10⋯O12 1.05 (3) 2.34 (3) 2.895 (3) 112 (2) C5---H5⋯O18^i^ 0.96 2.52 3.479 (3) 177 C8---H8*b*⋯O19 0.96 2.55 3.416 (3) 150 C15---H15⋯O13^ii^ 0.96 2.42 3.369 (3) 169.02 C20---H20*b*⋯O16^iii^ 0.96 2.43 3.165 (3) 134 C35---H35⋯O15^ii^ 0.96 2.59 3.278 (4) 129 C38---H38*b*⋯O19^iv^ 0.96 2.50 3.447 (3) 168 C17---H17*a*⋯*Cg*1 0.96 2.87 3.704 (3) 146 C37---H37*b*⋯*Cg*2^v^ 0.96 2.99 3.825 (3) 146 C13---H13⋯*Cg*3 0.96 3.20 4.070 (3) 150 C33---H33⋯*Cg*4^v^ 0.96 3.00 3.899 (3) 156 Symmetry codes: (i) ; (ii) ; (iii) ; (iv) ; (v) .

1. Introduction {#S1} =============== Congenital coronary artery anomalies are defined as a coronary pattern that is found in less than 1% of the general population, with a prevalence ranging from 0.3%−5.6% \[[@R1]\]. In few types of the anomalies, there is an association with sudden death and premature coronary disease \[[@R2]\]. Congenital absence of left main coronary artery (LMCA) and anomalous origins of left anterior descending artery (LAD) and left circumflex artery (LCX) arising from right sinus of Valsalva is rarely reported. Here we are presenting a 62-year-old male who presented with non-ST NSTEMI who found to have anomalous absence of left main coronary artery and anomalous origins of left anterior descending artery and left circumflex artery from right sinus of Valsalva. 2. Report of the Case {#S2} ===================== 62 years old man with past medical history of hypertension, dyslipidemia and type II diabetes mellitus presented with acute chest pain. The pain started suddenly, pressure like, at the left side, 9/10 in intensity, radiates to left arm and was associated abdominal discomfort, nausea and diaphoresis. The pain was relieved by sublingual nitroglycerin. Electrocardiography showed tall positive T waves at inferior leads ([Figure 1](#F1){ref-type="fig"}). His troponin was initially 0.5 ng/L then increased after 4 hours to 1.2 ng/L. He was started on aspirin, clopidogrel and heparin. Transthoracic echocardiography showed ejection fraction estimated to be 60% without wall motion abnormality. He was taken for cardiac catheterization, which showed 95% occlusion of proximal left circumflex artery (LCX) and 60% occlusion of distal Left anterior descending artery (LAD). LCX and LAD were originated from right coronary cusp ([Figure 2](#F2){ref-type="fig"}). He was treated with drug-eluting stent for proximal LCX ([Figure 3](#F3){ref-type="fig"}). He was discharged on aspirin and ticagrelor in a stable medical condition. 3. Discussion {#S3} ============= Coronary heart disease one of the leading causes of death in developed countries \[[@R3]\]. Congenital coronary artery anomalies are relatively rare entities with a prevalence that is reported to be approximately 0.3%-5.6%. The variation in prevalence is likely explainable by the diversity of the study populations \[[@R1]\]. Studies suggest that congenital coronary artery anomalies are the second most common cause of sudden death in young athletes, likely due to premature coronary disease \[[@R2]\]. Based on predisposition to CAD, congenital coronary artery anomalies are classified into 3 major groups: first: anomalies that predispose to ischemia such as coronary vessels originating from right atrium; second: anomalies that do not predispose to ischemia, as in anomalous origin of right coronary artery from posterior sinus of Valsalva; third: anomalies that might predispose to ischemia as in congenital absence of LAD \[[@R4]\]. It is usually diagnosed incidentally by CT coronary angiography, interventional coronary angiography, and in some severe and unusual cases, by transthoracic or transesophageal echocardiography, particularly in pediatric populations \[[@R4]\]. In our case, the patient has multiple anomalies: congenital absence of the left main coronary artery (LMCA) which was reported in 0.41%−0.67% of the cases \[[@R5]\]. It is, in most of the cases, clinically benign. However, an association with myocardial infarction and syncope were reported \[[@R6]\]. The second one is anomalous origins of left anterior descending artery (LAD) and left circumflex artery (LCX) from right sinus of Valsalva, were report separately as 0.03% and 0.032% respectively. The combination of both is extremely rare. The clinical significance of this anomaly is according to the course of the LAD. The possibilities of LAD course are: pre-pulmonic anterior to the right ventricular outflow tract which rarely causes ischemia; retro-aortic posterior to the aortic root, as in our case, usually benign; inter-arterial between the aorta and pulmonary artery, often associated with unfavorable outcomes; trans-septal subpulmonic course, which rarely causes ischemia; and retro-cardiac in the posterior atrioventricular groove, which predisposes to coronary disease \[[@R6]\]. The other aspect of the anomaly management is technical difficulties of coronary angiography and percutaneous coronary intervention due to unexpected positions and difficult angles. 4.. Conclusion {#S4} ============== We reported a case of combined anomalies of the coronary arteries including absence of Left Main Coronary Artery with Anomalous Origin of Left Anterior Descending and Left Circumflex Arteries that presented with NSTEMI. This combination of anomalies is exceedingly rare. While congenital coronary artery anomalies are quite rare, these entities could result in acute coronary syndrome with technical difficulties during percutaneous coronary interventions. Our case report highlights the need to keep that possible diagnosis of congenital coronary anomalies in mind while managing patients with acute myocardial infarction. This work is supported, in part, by the efforts of Dr. Moro O. Salifu M.D., M.P.H., M.B.A., M.A.C.P., Professor and Chairman of Medicine through NIH Grant number S21MD012474. {#F1} {#F2} {#F3}

Introduction ============ The epidemic of diabetes is worsening. The Centers for Disease Control estimates that over 23 million Americans suffer from diabetes, with an incidence of over 1.4 million new cases of diabetes every year.[@b1-vhrm-5-225] As the prevalence of diabetes increases, there is greater concern for the vascular complications that accompany diabetes. Insulin remains the most potent medication available to treat diabetes, and arguably to prevent diabetic complications. Some advocate using insulin much sooner in the course of diabetes.[@b2-vhrm-5-225] On the other hand, others believe that insulin may actually be harmful in obese type 2 diabetic subjects because it increases body fat, which may exacerbate insulin resistance.[@b3-vhrm-5-225],[@b4-vhrm-5-225] This review explores the interplay between insulin and the vascular system with special emphasis on the rapid-acting insulin analogs. Because of the heterogeneity of the literature, this paper will refer to vascular events defined very loosely, including peripheral events, cardiovascular events, cerebrovascular events, and death due to vascular causes. Is insulin harmful? =================== The relationship between insulin and vascular events is debatable. One study has associated higher serum insulin levels as an independent predictor of cardiovascular events. The Helsinki Policeman study was a retrospective study of 22-year mortality data of 970 non-diabetic men without coronary artery disease. The authors found an increase in cardiovascular mortality among subjects with higher serum insulin levels.[@b5-vhrm-5-225] A larger study, the Paris Protection Study, followed 7246 non-diabetic men without coronary artery disease for an average of 63 months and concluded that higher fasting plasma insulin levels were an independent risk factor for the development of coronary artery disease.[@b6-vhrm-5-225] Although these studies may suggest that elevated circulating insulin is a direct cause of vascular disease, this does not seem to be true. Subjects with insulin producing neoplasms do not have an increase in clinically overt atherosclerotic disease.[@b7-vhrm-5-225] This suggests that factors other than hyperinsulinemia are responsible for an increased risk of vascular disease. Insulin resistance is often associated with hypertension, lipid abnormalities, and obesity, all of which are thought to contribute much more than hyperinsulinemia to the development of vascular disease.[@b8-vhrm-5-225] In fact, some have proposed that insulin resistance is another symptom associated with the metabolic syndrome and cardiovascular disease, rather than the root cause of cardiovascular disease. Cardiologists performing a population study on 322 healthy adults have described an "insulin gradient."[@b9-vhrm-5-225] These researchers noted a direct correlation between body weight and blood pressure and insulin levels. The heavier their subjects were, the higher the blood pressure and insulin level. Thus, the higher cardiovascular rates seen with higher insulin levels may be caused by an increase in other risk factors such as hypertension, rather than insulin itself. Such a relationship is often ignored in articles that claim an association between insulin resistance and cardiovascular events. For example, in the Veteran Affairs-HDL Intervention Trial (VA-HIT), the authors concluded that "the occurrence of a new cardiovascular event was dependent on ... the presence or absence of insulin resistance."[@b10-vhrm-5-225] However, it is interesting to note that the group with insulin resistance had an average body mass index of over 31 while those without insulin resistance had an average body mass index of only 27.[@b10-vhrm-5-225] Furthermore, there was no reporting of blood pressure values. Although it is possible that higher levels of insulin increased the rate of cardiovascular events in this population, it is equally plausible than other factors such as hypertension were responsible. This then begs the question of whether hypertension is a byproduct of hyperinsulinemia. Although no unequivocal data to answer this question exist, multiple experiments in dogs have shown this not to be the case. There were no pressor effects noted in normal dogs infused with insulin, or in dogs with a 70% reduction in kidney mass on a high salt diet.[@b11-vhrm-5-225] Interestingly, chronic hyperinsulinemia actually caused a reduction in total peripheral vascular resistance as well as arterial pressure.[@b11-vhrm-5-225] This decrease in vascular resistance disappeared when the dogs were made obese via a high-fat diet.[@b11-vhrm-5-225] This series of experiments suggests that in dogs, hyperinsulinemia does not cause hypertension. Whether or not this translates into humans remains to be seen. However, a cross-sectional relational study found obesity to be an independent risk factor for left ventricular hypertrophy, but not insulin resistance or fasting insulin levels.[@b12-vhrm-5-225] These data seem to suggest that obesity and its complications are responsible for cardiovascular risk, not merely high insulin levels. Although there is a theoretical concern of a mitogenic effect of insulin analogs, this concern appears limited to the long-acting insulin glargine.[@b13-vhrm-5-225] Glargine appears to be a more potent stimulus of DNA synthesis in human osteosarcoma cell lines than the native insulin molecule; the rapid-acting analogs appear to be equivalent to regular insulin.[@b14-vhrm-5-225] Some have postulated that mitogenic potency is related to the half-life of the receptor--ligand binding complex, which would explain why the rapid-acting analogs do not appear to have as much theoretical mitogenic effect as the long-acting analog glargine.[@b15-vhrm-5-225] Does intensive diabetes therapy with insulin improve vascular events? ===================================================================== Recently, there have been several large prospective studies examining the relationship between the treatment of hyperglycemia and vascular complications. The first major study to associate a decrease in vascular events with glycemic control was the United Kingdom Prospective Diabetes Study (UKPDS). This prospective observational study included 4585 type 2 diabetic subjects. It concluded that each 1% reduction in mean HbA~1c~ was associated with a risk reduction of 21% for deaths related to diabetes, 14% for myocardial infarction, and 37% of microvascular complications.[@b16-vhrm-5-225] These patients were followed for another 10 years without diabetic treatment manipulation by the researchers. Despite the difference in HbA~1c~ disappearing after one year, the group intensively treated initially with insulin or sulfonylurea still had a 24% risk reduction for microvascular disease, a 15% in risk reduction for myocardial infarction, and a 13% reduction for death from any cause.[@b17-vhrm-5-225] Because patients initially treated with metformin also had risk reductions, the authors of this study concluded that this legacy effect was not the result of insulin, but rather a possible reduction in advanced glycation end products from the initial intensive treatment of hyperglycemia.[@b18-vhrm-5-225] The Diabetes Control and Complications Trial (DCCT) and the subsequent Epidemiology of Diabetes Interventions and Complications (EDIC) illustrated that intensive insulin therapy in type 1 diabetic patients reduced the major diabetic complications of neuropathy, nephropathy, and retinopathy.[@b19-vhrm-5-225] Intensive insulin therapy was associated with less cardiovascular disease as evidenced by decreased intima-media thickness and lower coronary artery calcium accumulation.[@b20-vhrm-5-225] Intensive insulin treatment also decreased the risk of any cardiovascular disease by 42% and the risk of non-fatal stroke, myocardial infarction, or cardiovascular death by 57% in type 1 diabetic subjects.[@b21-vhrm-5-225] Intensive insulin treatment was also shown to be beneficial to those with the worst vascular disease. Diabetic patients who suffered an acute myocardial infarction had an absolute reduction in mortality of 11% when treated with intensive insulin therapy.[@b22-vhrm-5-225] Other studies have been less convincing. The recent Action in Diabetes and Vascular Disease (ADVANCE) trial was aimed to address the effect of intensive glucose control on vascular outcomes in subjects with type 2 diabetes. The researchers randomly assigned 11,140 type 2 diabetic patients to standard glucose control or intensive glucose control using to a HbA~1c~ of 6.5% or less.[@b23-vhrm-5-225] Although the first line drug was a sulfonylurea, 40.5% of subjects in the intensive arm and 24.1% of subjects in the standard arm ended up on insulin.[@b23-vhrm-5-225] After a median of 5 years of follow-up, the researchers found that intensive glucose control led to a reduction in nephropathy (4.1 versus 5.2%), but had no effect on retinopathy, major macrovascular events, or death from cardiovascular causes.[@b23-vhrm-5-225] The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial also attempted to address the relationship between vascular events and glucose control in patients with type 2 diabetes. One distinguishing feature of this trial is that all subjects had to have either established cardiovascular disease or additional cardiovascular risk factors. Therefore, it was aimed for secondary instead of primary prevention of vascular disease. 10,251 subjects were randomly assigned to intensive therapy (HbA~1c~ \< 6%) or standard therapy (HbA~1c~ 7.0%--7.9%).[@b24-vhrm-5-225] This portion of the study (a blood pressure arm is still ongoing) was discontinued after a mean of 3.5 years of follow up due to an increase in total mortality (257) in the intensive group as opposed to the standard group (203).[@b24-vhrm-5-225] Interestingly, the number of events in the primary outcome (non-fatal myocardial infarction, non-fatal stroke, or death from cardiovascular cause) was actually lower in the intensive group (352) as compared to the standard group (371).[@b24-vhrm-5-225] This calls into debate whether the increase in deaths was due to vascular complications, or if the increase would have panned out had the trial not ceased prematurely. Thus far, there has been no concrete explanation for the increase in mortality seen in the intensive arm. However, some postulate that hypoglycemia may have contributed to the cardiovascular events.[@b25-vhrm-5-225] If so, then theoretically, in comparison to regular insulin, the lower hypoglycemic rate associated with the rapid-acting analogs could provide a cardiovascular benefit.[@b26-vhrm-5-225] A third trial, the Veterans Affairs Diabetes Trial, also examined the effects of intensive blood glucose control on cardiovascular events. The intensive goal in this trial was defined by a goal HbA~1c~ of \< 7.5% and a standard arm goal of HbA~1c~ \< 8.3%, while controlling lipids and blood pressure in both arms.[@b27-vhrm-5-225] This trial was conducted at 20 different VA centers with 1791 patients with a mean follow-up of 6.25 years. In this study, there was no significant reduction in cardiovascular events in the intensive arm.[@b28-vhrm-5-225] Thus, the effect of aggressive control of blood glucose on cardiovascular events remains debatable. UKPDS and DCCT/EDIC have shown that tight control in younger patients in recently diagnosed disease was beneficial, whereas ADVANCE, ACCORD, and VADT have shown that in older patients with long-standing disease the benefits of tight glucose control are not as convincing. Diabetes control may very well have different effects on different cohorts. How might insulin reduce vascular events? ========================================= Insulin's beneficial effects on the vascular system are likely due to actions on the endothelium (see [Table 1](#t1-vhrm-5-225){ref-type="table"}). Insulin has been shown to be a potent vasodilator by causing an increased production of nitric oxide.[@b29-vhrm-5-225] Insulin induces the expression of nitric oxide synthetase, which converts arginine to nitric oxide.[@b30-vhrm-5-225] Separate experiments then illustrated that nitric oxide levels increase in a dose dependent manner in human vein endothelial cells as well as human aortic endothelial cells.[@b31-vhrm-5-225],[@b32-vhrm-5-225] These observations were translated in humans when researchers infused insulin into normal subjects' arteries, and found that blood flow increases in a dose dependent manner with insulin.[@b29-vhrm-5-225] However, in type 2 diabetic as well as non-diabetic obese patients, the vasodilatory effect of insulin is blunted.[@b33-vhrm-5-225],[@b34-vhrm-5-225] Whether this decrease in vasodilation is due to insulin resistance or if insulin resistance is a byproduct of a decrease in skeletal muscle perfusion and insulin-mediated glucose uptake remains to be seen. In addition to effects on vascular tone, diabetes also has an inflammatory component, characterized by endothelial dysfunction and elevated adhesion molecules.[@b35-vhrm-5-225] Insulin helps to counteract this inflammation, by suppressing the expression of adhesion molecules, allowing monocytes to couple with endothelial cells, releasing monocyte chemoattractant protein-1 (MCP-1). This in turn attracts more monocytes which then intermingle with low-density lipoprotein cholesterol, forming foam cells.[@b29-vhrm-5-225] Insulin also reverses the increase in the adhesion molecule e-selectin seen in type 2 diabetic subjects.[@b36-vhrm-5-225] Along with suppressing MCP-1 and e-selectin, insulin also suppresses nuclear factor kappa B (NF-kB).[@b37-vhrm-5-225] NF-kB is a central figure in inflammation, responsible for the production of cytokines, enzymes, and adhesion molecules.[@b38-vhrm-5-225] When human aortic endothelial cells were incubated with insulin, intranuclear NF-kB binding activity and MCP-1 mRNA expression were both suppressed.[@b37-vhrm-5-225] This decrease in MCP-1 was replicated in vivo in obese subjects, in whom the translocation of NF-kB was inhibited by insulin infusion.[@b39-vhrm-5-225] Insulin also decreases NADPH oxidase, plasma tissue factor, and matrix metalloproteinase.[@b29-vhrm-5-225] Insulin has been shown to decrease the inflammatory marker C-reactive protein in critically ill patients.[@b40-vhrm-5-225] In patients suffering from an acute myocardial infarction, insulin has anti-inflammatory and pro-fibrinolytic effects as evidenced by a blunting of the increase in high-sensitivity C-reactive protein, serum amyloid A, and plasminogen activator inhibitor-1.[@b41-vhrm-5-225] Are the effects of insulin a class effect? ========================================== It is unclear if the effects of insulin are unique to a certain formulation or are class effects that extend to the rapid-acting insulin analogs. [Table 1](#t1-vhrm-5-225){ref-type="table"} summarizes the experimental evidence of the various formulations of short-acting insulin. One type of rapid-acting insulin is lispro (Humalog^®^, Eli Lilly and Company). Lispro is a human insulin analog manufactured from *Escherichia coli* with the amino acids at positions 28 and 29 reversed.[@b42-vhrm-5-225] This reversal of lysine and proline allows the insulin to be absorbed much faster than regular insulin from subcutaneous tissue. Lispro has a peak action of 30 to 90 minutes.[@b42-vhrm-5-225] Because of its fast onset, lispro is often used as a bolus insulin to address postprandial glucose excursions. Like the other rapid-acting analogs, it can be used alone in an insulin pump, or in combination with a longer acting insulin formulation. It is marketed by itself as well as pre-mixed with lispro protamine, a crystallized form of lispro made by combining lispro with protamine sulfate, lengthening its duration of action.[@b43-vhrm-5-225] Clinical guidelines and algorithms for the use of rapid-acting analogs in diabetic patients are beyond the scope of this article. There are some data to show that lispro is beneficial to the vascular system. Specifically, lispro improves microvascular blood flow in postprandial type 1 diabetic patients.[@b44-vhrm-5-225] In this study, 20 non-diabetic patients and 20 diabetic patients had their skin microvascular blood flow measured by laser Doppler flux every 30 minutes after a standardized test meal. The researchers found that microvascular blood flow was impaired in type 1 diabetic patients given regular insulin, but when given insulin lispro, the microvascular blood flow improved to mimic the non-diabetic subject. This is thought to be due to the treatment of acute hyperglycemia, which stimulates the production of free radicals and augments thrombin formation.[@b44-vhrm-5-225] Hyperglycemia is also thought to create an increase in adhesion molecules and endothelin, while decreasing levels of nitric oxide.[@b45-vhrm-5-225] Because lispro has a faster onset of action than regular insulin, it reduces these postprandial excursions better, and is more likely to counteract the increase in adhesion molecules and decrease in nitric oxide.[@b46-vhrm-5-225] Similar to lispro, glulisine (Apidra^®^, Sanofi-Aventis) has also been shown to have beneficial effects on surrogate vascular markers. Glulisine is a rapid-acting insulin homologous with regular human insulin save for the amino acid asparagine at position B3 replaced with lysine and the lysine at position B29 substituted with glutamic acid.[@b47-vhrm-5-225] This analog is produced by recombinant DNA using *Escherichia coli*, and it is employed to lower blood glucose in a similar fashion as lispro. In a microvascular blood flow study similar to the one described above using lispro and regular insulin, glulisine was compared to regular insulin in 15 type 2 diabetic subjects after a liquid meal challenge. Serial laser Doppler fluxometry was then performed. Like the results in the lispro study, glulisine-reated subjects had higher postrandial insulin levels, lower glucose excursions, and higher microvascular blood flow.[@b48-vhrm-5-225] Again, without any head to head comparison studies, it is impossible to say that the short-acting insulin analogs are equal. Yet, the fact that both glulisine and lispro had similar effects in a similarly designed study suggests a class effect benefit from treating postprandial glucose excursions. Although a similar study using laser fluxometry has not been published using aspart (NovoLog^®^, Novo Nordisk Inc.), there have been other studies examining the effect of aspart on vascular risk factors. Aspart is similar in composition to regular human insulin, save for the substitution of the amino acid proline with aspartic acid in position B28.[@b49-vhrm-5-225] It is produced using recombinant DNA technology using *Saccharomyces cerevisiae* (baker's yeast) and is used to treat hyperglycemia in a similar fashion as lispro and glulisine. Despite the similar function and onset of action, the data on aspart's effects on the vascular system are not as clear as with the other rapid-acting analogs. Twenty-one patients with insulin-treated type 2 diabetes were given neutral protamine Hagedorn (NPH) insulin and pre-prandial regular insulin or aspart for 6 weeks.[@b50-vhrm-5-225] At the end of the study, although there was no statistical difference in hemoglobin A~1c~ (HbA~1c~), the aspart group had a significantly lower postprandial (90 minutes after a meal) average glucose (7.9 mmol/L versus 9.3 mmol/L).[@b50-vhrm-5-225] Despite this decrease in glucose excursions, there were no statistical differences in markers of vascular risk (fasting lipid profile, apolipoproteins, fibrinogen, plasminogen activator inhibitor-1, E-selectin, or homocysteine). However this study has several limitations to preclude the conclusion that treating postprandial glucose excursions has no effect on vascular risk. This study's conclusion may reflect a lack of power, brief study duration, or even a difference between the effects of distinct insulin analogs on vascular risk. Does treatment of postprandial glucose help? ============================================ The rapid-acting insulins' superiority lies in their ability to treat postprandial glucose and excursions. Is such an ability clinically relevant? Fortunately, several studies highlight how postprandial excursions and high glucose variability affect the vascular system. One such study involved type 1 diabetic patients who underwent myocardial perfusion studies while either on a euglycemic or hyperglycemic hyperinsulinemic clamp. The myocardial perfusion reserve was significantly decreased when subjects were hyperglycemic.[@b51-vhrm-5-225] Just as chronic diabetes is associated with inflammation, acute hyperglycemia has also been shown to cause an increase in proinflammatory cytokines such as tumor necrosis factor-alpha, interleukin (IL)-6, IL-1 beta, and IL-8, which leads to vascular inflammation.[@b52-vhrm-5-225] This increase in cytokines was counteracted by insulin infusion and a return to normoglycemia.[@b52-vhrm-5-225] By mitigating hyperglycemia with a faster onset of action than regular insulin, some have argued that rapid-acting insulin may have favorable cardiovascular effects.[@b53-vhrm-5-225],[@b54-vhrm-5-225] One argument is that using premixed lispro combinations lower postprandial glucose better than regular insulin combinations, and therefore may reduce CVD risk.[@b54-vhrm-5-225] When lispro was added to bedtime NPH insulin and compared to twice-daily NPH in type 2 diabetic subjects, the lispro-treated group experienced lower postprandial glucose, lower HbA~1c~, lower triglycerides, low-density lipoprotein cholesterol (LDL), and higher high-density lipoprotein cholesterol (HDL).[@b53-vhrm-5-225] The authors surmised that this favorable milieu might have a positive cardiovascular effect. In addition, a recent study was performed comparing the long-acting insulin detemir (Levemir^®^, Novo Nordisk Inc.) to the rapid-acting aspart. It found that patients using a prandial bolus of insulin aspart had a lower HbA~1c~ than those treated with basal detemir.[@b55-vhrm-5-225] These two studies would suggest that the postprandial excursions can drive overall blood glucose control, and that by treating this variability, vascular events could be mitigated. Reducing glycemic variability could also lessen glycemic events due to fewer episodes of hypoglycemia. A recent study of 100 type 1 diabetic patients followed for 11 years found a strong correlation between glycemic variability and hypoglycemic unawares.[@b56-vhrm-5-225] It is possible that using insulin analogs could reduce glycemic variability and hypoglycemic episodes, whereby overall glucose control and vascular events are improved. However, this theory can only be answered with a prospective long-term study with an extremely large sample size using rapid-acting insulin. On the other hand, specifically targeting postprandial glucose may not have a clinically significant vascular effect. The hyperglycemia and its effect after acute myocardial infarction on cardiovascular outcomes in patients with type 2 diabetes mellitus (HEART2D) attempted to randomize 1355 subjects with type 2 diabetes and acute myocardial infarction to either pre-meal insulin lispro with basal NPH insulin, (with a target 2-hour postprandial blood glucose of \< 135 mg/dL) or basal insulin only (NPH insulin twice daily, insulin glargine once daily, or pre-mixed human insulin \[70% NPH/30% regular\] twice daily) targeting fasting.[@b57-vhrm-5-225] Although both groups ended up with similar HbA~1c~ values (7.6%), and the lispro-treated group had statistically fewer postprandial excursions, there was no significant difference in cardiovascular events between the two groups and the study was stopped for "futility."[@b58-vhrm-5-225] These data would suggest that glycemic variability does not increase cardiovascular events, although the sample size of the study was relatively small and it experienced a high drop out rate. Conclusion ========== In summary, the UKPDS has shown that treating diabetes to a goal HbA~1c~ \< 7% will decrease some vascular events. Intensively treating diabetes to \< 6% is debatable and difficult given present treatment modalities. Individualized goals should be designed between physician and patient. In meeting these goals, insulin has been shown to be beneficial by not only by treating hyperglycemia, but also by providing vasodilatory and anti-inflammatory effects. Although not proven, these effects likely extend to the rapid-acting analogs. With its faster onset of action and lower postprandial glucose excursions, rapid-acting insulin appears superior to regular insulin, and could possibly reduce vascular events further. **Disclosures** The authors declare no conflicts of interest. ###### Comparison of rapid-acting insulin analogs with regular insulin **Regular insulin** -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Time of onset 30--60 minutes   Peak action 50--120 minutes   Duration of action 5--8 hours   Effects on vascular system     Improved outcomes in diabetic subjects with decreased        Intima-media thickness       Cardiovascular risk       Risk of non-fatal stroke       Risk of myocardial infarction       Risk of cardiovascular death       Mortality post-acute myocardial infarction     Vasodilation     Increases        Nitric oxide     Suppresses        Adhesion molecules       E-selectin       MCP-1       Nf-kB       NADPH oxidase       Plasma tissue factor       Matrix metalloproteinase       C-reactive protein       Serum amyloid A       Plasminogen activator inhibitor-1 **Lispro (Humalog^®^)**   Time of onset 5--15 minutes   Peak action 30--90 minutes   Duration of action 3--5 hours   Difference from regular insulin     Produced from *Escherichia coli*     Lysine and proline at positions 28 and 29 transposed     Lower postprandial glucose   Effects on vascular system     Improved postprandial blood flow **Glulisine (Apidra^®^)**   Time of onset 5--15 minutes   Peak action 34--91 minutes   Duration of action 55--149 minutes   Difference from regular insulin     Produced from *Escherichia coli*     Asparaginase at B3 replaced with lysine and lysine at B29 replaced with glutamic acid     Lower postprandial glucose   Effects on vascular system     Improved blood flow postprandially **Aspart (Novolog^®^)**   Time of onset 5--15 minutes   Peak action 40--50 minutes   Duration of action 3--5 hours   Difference from regular insulin     Produced from *Saccharomyces cervisiae*     Proline replaced with aspartic acid at B28     Lower postprandial glucose   Effects on vascular system     No changes in fasting lipid profile, apolipoproteins, fibrinogen, plasminogen activator inhibitor-1, E-selectin, or homocysteine when compared to regular insulin. **Note:** Time values assume subcutaneous bolus injection.

Introduction {#Sec1} ============ A hexanucleotide repeat expansion mutation in *chromosome 9 open reading frame 72* (*C9orf72*) is the most common known genetic cause of amyotrophic lateral sclerosis (ALS), a fatal neurodegenerative disease characterized by the progressive loss of corticospinal, brainstem and spinal motor neurons. A *C9orf72* mutation underlies approximately 40% of familial and 5% of sporadic ALS cases^[@CR1]--[@CR3]^. This hexanucleotide expansion is also found in approximately 10% of cases of a second neurodegenerative disease, frontotemporal dementia (FTD), a common cause of dementia in middle-aged patients^[@CR3]--[@CR5]^. Mutations in *C9orf72* are a rare risk factor for several additional neurologic and psychiatric disorders including Alzheimer's disease, Parkinson's disease, Huntington's disease phenocopy patients, multiple system atrophy, depressive pseudodementia, bipolar disorder, and schizophrenia^[@CR6]--[@CR16]^. However, it is not known how this genetic mutation leads to these cell type-specific neurodegenerative disorders. Both loss-of-function and gain-of-function mechanisms have been proposed to mediate *C9orf72*-linked ALS^[@CR17],\ [@CR18]^. Several studies have demonstrated decreased *C9orf72* transcript and protein levels in patients with ALS and FTD^[@CR19]--[@CR23]^. Deletion of the mouse ortholog of the *C9orf72* gene (3110043O21Rik, referred to here as *C9orf72*) has been reported to shorten lifespan and induce modest motor deficits in some, but not all mouse models, and cause profound dysregulation of the immune system^[@CR24]--[@CR29]^. Gain-of-function toxicity of the repeat expansion induced by sense and anti-sense RNA transcripts, as well as dipeptide proteins generated through repeat-associated non-ATG (RAN) translation, are also thought to contribute to neurodegeneration^[@CR17],\ [@CR18]^. Several transgenic mouse lines recently developed using patient-derived gene constructs demonstrate that *C9orf72* repeat expansions induce age-dependent accumulation of RNA foci and dipeptide repeat proteins, along with neurodegeneration and behavioral abnormalities that at least partially recapitulate human disease^[@CR26],\ [@CR28],\ [@CR30],\ [@CR31]^. Although the complex mechanisms underlying *C9orf72*-related disease have not been resolved, understanding the expression pattern of *C9orf72* in the central nervous system (CNS) is not only important for understanding the pathogenesis of ALS, but is also relevant to the wide spectrum of *C9orf72*-associated diseases. Although the signature of ALS is the loss of corticospinal, brainstem, and spinal motor neurons, multiple cell types have been shown to contribute to the pathogenesis of the disease. Non-neuronal cells including oligodendrocytes, astrocytes, and microglia are also critical players in the pathogenesis of ALS^[@CR17]^. Although some progress has been made in understanding the cell-type specific expression of *C9orf72* ^[@CR26],\ [@CR28],\ [@CR32]^, a comparison of the distribution of *C9orf72* expression across different neuronal and glial cell types in relevant regions of the brain and spinal cord is still lacking. Whether *C9orf72* promoter activity is specifically enriched in affected corticospinal neurons, spinal motor neurons, or oligodendrocytes in regions implicated in ALS is not yet fully understood. Here, we systematically mapped the promoter activity of the mouse ortholog of *C9orf72* in a genetically engineered strain of mice containing a targeted *LacZ* insertion under the control of the *C9orf72* native promoter. Through quantitative comparisons among different types of neurons and glial cells labelled with retrograde neuronal tracers and cell type-specific markers, we demonstrate that mouse *C9orf72* promoter activity, although widespread throughout the brain and spinal cord, is specifically enriched in corticospinal and spinal motor neurons and in oligodendrocytes, subsets of cells known to undergo degeneration in ALS, in regions affected by ALS. In contrast, *C9orf72* promoter activity was detected in only a small percentage of microglial cells and even fewer astrocytes. Thus, these data suggest that, despite widespread expression, *C9orf72* promoter activity reflects the patterns of degeneration typically seen in this disease, consistent with direct cell autonomous toxicity. Results {#Sec2} ======= The distribution of *C9orf72* promoter activity and cellular density are highly correlated in the CNS {#Sec3} ----------------------------------------------------------------------------------------------------- Mice have a single gene, 3110043O21Rik (here referred to as *C9orf72*), which is orthologous to human *C9orf72* ^[@CR32]^. To determine the distribution of *C9orf72* promoter activity in the central nervous system, we generated chimeric mice using several mouse embryonic stem cell lines heterozygous for an allele with a *LacZ* insertion in the *C9orf72* locus generated by the Knock Out Mouse Project (KOMP)^[@CR29],\ [@CR32]--[@CR35]^. The *LacZ* insertion results in deletion of exons 2--6 of the mouse *C9orf72* gene, producing a knockout allele (Supplementary Fig. [1a](#MOESM1){ref-type="media"}). Analysis of RNA-sequencing (RNA-seq) data from mice generated using the same targeting cassette and embryonic stem (ES) cell background indicates that the transcript structure is largely maintained between wild type (WT) and *C9orf72* ^*LacZ*/+^ mice^[@CR28]^ (Supplementary Fig. [1b](#MOESM1){ref-type="media"}). Although WT transcripts have two alternative starts, exon 1a or 1b, while the *LacZ* allele appears to use only exon 1a, there is no reported evidence for differential usage of exon 1a and 1b among different cell types indicating that the *LacZ* reporter likely reflects the pattern of the wild type gene expression. The heterozygous mice have a normal phenotype until six months of age, after which a fraction of *C9orf72* ^*LacZ*/+^ mice exhibit an age-dependent decrease in survival, with approximately 20% of the heterozygotes dead by 600 days^[@CR29]^. We therefore used young, six to eight-week-old heterozygotes to assess the distribution of *LacZ* as a reporter for *C9orf72* promoter activity. We confirmed that *C9orf72* promoter activity was not limited to areas known to degenerate in ALS and FTD. X-gal staining to assess regions of β-galactosidase (β-gal) expression (encoded by *LacZ*) revealed widespread promoter activity throughout the brain and spinal cord. The regions with the most intense X-gal signals in the brain corresponded to regions with high cell density, such as the dentate gyrus of the hippocampus and the granular layer of the cerebellum (Fig. [1a](#Fig1){ref-type="fig"}). Regions with the weakest signals corresponded to areas with low cell densities such as the molecular layer of the cerebellum and the corpus callosum (Fig. [1a](#Fig1){ref-type="fig"}). We also observed broad X-gal staining in primary motor cortex (Fig. [1a](#Fig1){ref-type="fig"}) and in the spinal cord (Fig. [1b](#Fig1){ref-type="fig"}). These results suggest that the distribution of *C9orf72* promoter activity largely follows cellular density across brain regions. To more directly test this hypothesis, we compared the distribution of cells stained with the nuclear marker, DAPI, and an antibody specific to β-galactosidase (β-gal; Supplementary Fig. [1c](#MOESM1){ref-type="media"}) and found that the distribution of β-gal signal correlated with the overall cellular density across layers in primary motor cortex, primary somatosensory cortex, and spinal cord grey matter (Fig. [1c-d](#Fig1){ref-type="fig"}). Together, these results indicate that *C9orf72* promoter activity is widely distributed in the CNS, consistent with that reported in previous studies^[@CR26],\ [@CR27],\ [@CR32]^, and largely correlates with overall cellular density.Figure 1The distribution of *C9orf72* promoter activity follows overall cellular density in the central nervous system. (**a**) The overall distribution of *C9orf72* promoter activity in a parasagittal brain section from a *C9orf72* ^*LacZ/*+^ mouse revealed using X-gal staining. *C9orf72* promoter activity was high in brain regions with high cell density (dentate gyrus of the hippocampus, cerebellar granule cell layer, two left panels), low in regions with low cell density (cerebellar molecular cell layer and corpus callosum, middle panels), and medium in motor cortex (rightmost panel). (**b**) The overall distribution of *C9orf72* promoter activity in a coronal section of the lumbar spinal cord. X-gal staining was seen in both the dorsal and ventral horns (left and right panels) (**c**) Representative images showing the distribution of immunofluorescence staining for β-gal in primary motor (top) and somatosensory (bottom) cortex of *C9orf72* ^*LacZ/*+^ mice concurrently labelled with the nuclear stain, DAPI. The intensities of the β-gal (red) and DAPI (blue) signals were summed along the horizontal axis in the region indicated by the white boxes (n = 3 mice) and the intensity values plotted (mean ± SEM). (**d**) A similar analysis was performed for lumber spinal cord (*n* = 3 mice). Scale bars: (**a**) 1 mm, insets 100 µm, (**b**) 200 µm, insets 100 µm and **(c,d)** 300 µm. *The C9orf72* promoter is active in both excitatory and inhibitory neurons {#Sec4} -------------------------------------------------------------------------- Although at a macroscopic level, *C9orf72* promoter activity was correlated with overall cellular density throughout the brain and spinal cord, enrichment or reduction of *C9orf72* promoter activity in specific regions of the neocortex or spinal cord or within specific cell types in each region would go undetected at this level of analysis. Therefore, we compared the distribution of *C9orf72* promoter activity in neurons in primary motor cortex, a cortical region which undergoes degeneration in ALS, as well as primary somatosensory cortex (see Methods). *C9orf72* promoter activity was observed in more than 80% of the neurons labelled with antibodies to NeuN in layer 5 (L5) of primary motor cortex, which contains corticospinal neurons known to degenerate in ALS (Fig. [2a,b](#Fig2){ref-type="fig"}). Unexpectedly, neurons in layer 2/3 (L2/3) of primary somatosensory cortex, which consists primarily of excitatory corticocortical projection neurons, and L5 of primary somatosensory cortex, exhibited a similar distribution of *C9orf72* promoter activity (Fig. [2a,b](#Fig2){ref-type="fig"}). Due to the high density of small β-gal-positive puncta in our samples, it is possible that we detected *C9orf72* promoter activity in such a large number of neurons spuriously. To assess whether the distribution of the β-gal signal was specific to these neurons, we compared the percentage of β-gal-positive neurons detected before and after inverting the β-gal channel relative to the NeuN channel in each analyzed image. Following channel inversion, the percentage of β-gal-positive neurons significantly decreased, indicating that the relationship between the distribution of β-gal-positive puncta and the distribution of neurons is greater than would be expected by chance (Supplementary Fig. [2](#MOESM1){ref-type="media"}). These data demonstrate that the overall distribution of *C9orf72* promoter activity in neurons is similar between primary motor cortex and primary sensory cortex when specific cell types are not taken into account.Figure 2*C9orf72* promoter activity is widespread in inhibitory neurons. (**a**) Representative images of layer 5 (L5) of primary somatosensory cortex (S1, *top*) and primary motor cortex (M1, *bottom*) from *C9orf72* ^*LacZ/*+^ mice stained for NeuN (green) and β-galactosidase (β-gal; red). High magnification images show β-gal-positive puncta within NeuN-positive neurons (*rightmost panels*). Scale bars: 50 μm and 10 μm. (**b**) The percentage of neurons containing β-gal puncta in layer 2/3 and L5 of somatosensory cortex and L5 of motor cortex (S1 L2/3: 362 of 429 cells, 84.4%; S1 L5, 499 of 583 cells, 85.6%; M1 L5: 425 of 523 cells, 81.3%; *n* = 3 mice for each group, *p* = 0.1378, Chi-Square test). (**c**) Representative images of layer 1 (L1) of somatosensory cortex stained for NeuN (green) and β-gal (red). High magnification images showing β-gal-positive puncta within L1 neurons (*rightmost panels*). Scale bars: 50 μm and 5 μm. (**d**) The percentage of L1 neurons containing β-gal puncta (26 of 32 cells, 81.3%, *n* = 3 mice). (**e**) Representative images of L5 of motor cortex stained for parvalbumin (PV, green) and β-gal (red). High magnification images show β-gal puncta within a PV interneuron (*rightmost panels*). Scale bars: 50 μm and 10 μm. (**f**) The percentage of PV interneurons containing β-gal puncta in L5 of motor cortex and somatosensory cortex (M1 L5: 36 of 51 cells, 70.6%; S1 L5: 32 of 37 cells, 86.5%; *n* = 3 mice for each group, *p* = 0.0789, Chi-Square test). Although *C9orf72* transcripts have been identified in excitatory cortical projection neurons and spinal motor neurons^[@CR26],\ [@CR32]^, whether *C9orf72* promoter activity is found in inhibitory neurons is not known. To test for *C9orf72* promoter activity in inhibitory neurons, we first analysed neurons in layer 1 (L1) of the cortex, which are overwhelmingly GABAergic^[@CR36],\ [@CR37]^. We found that β-gal-positive puncta were detected in more than 80% of L1 inhibitory interneurons (Fig. [2c-d](#Fig2){ref-type="fig"}). Similarly, we found that *C9orf72* promoter activity was detected in approximately 70 to 85% of cortical parvalbumin-expressing (PV) inhibitory interneurons, the most common type of interneurons in the cortex^[@CR37]^ (Fig. [2e,f](#Fig2){ref-type="fig"}), demonstrating that *C9orf72* promoter activity is not limited to glutamatergic cortical projection neurons. Taken together, these data demonstrate that *C9orf72* promoter activity is also widespread in inhibitory interneurons. *C9orf72* promoter activity is enriched in corticospinal and spinal motor neurons {#Sec5} --------------------------------------------------------------------------------- The hallmark of ALS is progressive degeneration of corticospinal and spinal motor neurons. A previous study reported that the majority of layer 5 (L5) cortical neurons exhibiting *C9orf72* promoter activity were CTIP2-positive^[@CR32]^, suggesting that sub-cerebral projection neurons, which include the corticospinal neurons (CSNs) that degenerate in ALS, express *C9orf72*. To further test this hypothesis, we compared the distribution of *C9orf72* promoter activity in retrogradely labelled CSNs and in unlabeled NeuN-positive neurons intermingled within L5 of motor cortex (Fig. [3a,b](#Fig3){ref-type="fig"}). A significantly higher percentage of CSNs exhibited β-gal-positive puncta than unlabeled neighboring neurons, indicating that *C9orf72* promoter activity is specifically enriched in corticospinal neurons (Fig. [3c,d](#Fig3){ref-type="fig"}; *p* \< 0.0001, Chi-Square test). Similarly, a significantly higher percentage of choline acetyltransferase (ChAT)-immunoreactive spinal motor neurons exhibited detectable *C9orf72* promoter activity than either ChAT-immunonegative neurons intermingled within the same region of the ventral horn or ChAT-immunonegative neurons in the dorsal horn of the spinal cord (Fig. [3e,f](#Fig3){ref-type="fig"}; *p* \< 0.0001, Chi-Square test). Together, these results indicate that *C9orf72* promoter activity is specifically enriched in both retrogradely labelled corticospinal neurons and cholinergic spinal motor neurons, neuronal types that are specifically vulnerable in ALS.Figure 3*C9orf72* promoter activity is enriched in corticospinal and spinal motor neurons. (**a**) Schematic of the experimental design. Green fluorescent retrograde tracers were injected into the contralateral cervical spinal cord to retrogradely label corticospinal neurons in *C9orf72* ^*LacZ/*+^mice. (**b**) Retrogradely labelled corticospinal neurons are shown in layer 5 (L5) of primary motor cortex. Scale bar: 100 μm. (**c**) Representative images of L5 of motor cortex showing NeuN-positive neurons (cyan), retrogradely labelled corticospinal neurons (green) and β-gal staining (red). High magnification images show β-gal puncta within a corticospinal neuron (*rightmost panels*). Scale bars: 50 μm and 5 μm. (**d**) The percentage of unlabeled NeuN-positive neurons and of corticospinal neurons (CSNs) containing β-gal puncta in L5 of motor cortex (Unlabeled NeuN-positive neurons: 1357 of 1771 cells, 76.6%; CSNs: 132 of 137 cells, 96.4%; *n* = 5 mice for each group, *p* \< 0.0001, Chi-Square test). Unlabeled NeuN-positive neurons and retrogradely labelled corticospinal neurons were analyzed from the same images. (**e**) Representative images of the dorsal (*top*) and ventral horn (*bottom*) of the spinal cord immunostained for NeuN (green), Choline acetyltransferase (ChAT; cyan), and β-gal (red). High magnification images show β-gal puncta within NeuN-positive neurons in the dorsal horn (*top right)* and a ChAT-positive neuron in the ventral horn (*bottom right*). Scale bars: 50 μm and 10 μm. (**f**) The percentage of ChAT-negative and ChAT-positive neurons containing β-gal puncta in the dorsal and ventral horn of spinal cord (dorsal neurons: 734 of 1050 cells, 69.9%; ventral ChAT-negative neurons: 61 of 78 cells, 78.2%; ventral ChAT-positive neurons: 113 of 114 cells, 99.1%; *n* = 5 mice for each group; *p* \< 0.0001, Chi-Square test). *C9orf72* promoter activity in Purkinje cells and cerebellar granule cells {#Sec6} -------------------------------------------------------------------------- *C9orf72*-mediated disease is notable for RNA foci and neuronal inclusions found in the cerebellum^[@CR38]--[@CR41]^. RAN-translated polypetides stemming from the hexanucleotide repeat expansion have been detected in Purkinje cells and granule cells^[@CR40]^. Furthermore, prior studies have detected high levels of *C9orf72* expression in the cerebellum^[@CR23],\ [@CR42]^. Here we assessed *C9orf72* promoter activity in both Purkinje cells and granule cells in the cerebellum. Consistent with the X-gal staining showing intense signal in the cerebellum (Fig. [1a](#Fig1){ref-type="fig"}), we detected *C9orf72* promoter activity in a large percentage of both Purkinje cells and cerebellar granule cells (Fig. [4](#Fig4){ref-type="fig"}). Overall, 41.2% of Purkinje cells and 55.8% of granule cells had detectable *C9orf72* promotor activity (*p* = 0.0116, Chi-Square test), indicating that *C9orf72* promoter activity is relatively enriched in the granule cell layer. These data provide additional evidence for cell-type specific enrichment of *C9orf72* promoter activity in distinct cell types in affected brain regions in *C9orf72*-mediated disease.Figure 4*C9orf72* promoter activity in Purkinje cells and granule cells of the cerebellum. (**a**) Representative images of the cerebellum showing calbindin-positive Purkinje cells (green), NeuN-positive granule cells (cyan), and β-gal staining (red). Scale bar: 100 μm. (**b**) Representative images of calbindin-positive Purkinje cells (green) and β-gal (red). Scale bars: 50 μm and 10 μm. (**c**) The percentage of Purkinje cells containing β-gal puncta (35 of 85 cells, 41.2%, *n* = 3 mice). (**d**) Representative images of NeuN-positive granule cells (green) and β-gal (red). Scale bars: 10 μm and 2 μm. (**e**) The percentage of granule cells containing β-gal puncta: (306 of 548 cells, 55.8%; *n* = 3 mice). Cell-type and region specific regulation of *C9orf72* promoter activity in glial cells {#Sec7} -------------------------------------------------------------------------------------- Glial cells play important roles in the pathophysiology of ALS. Astrocytes undergo reactive changes in ALS, although their role in the initiation and progression of *C9orf72*-mediated ALS remains to be elucidated^[@CR17],\ [@CR43]--[@CR45]^. While a prior study reported that spinal cord astrocytes did not express *C9orf72* ^[@CR32]^, RNA sequencing experiments detected low levels of *C9orf72* transcripts in mouse^[@CR28],\ [@CR46]^ and human astrocytes^[@CR47]^. Consistent with the RNA sequencing data, we found that approximately 20% of S100β-positive astrocytes in the spinal cord had detectable *C9orf72* promoter activity (Fig. [5a-c](#Fig5){ref-type="fig"}). Interestingly, almost no S100β-positive astrocytes in the cortex exhibited detectable *C9orf72* promoter activity (Fig. [5a-c](#Fig5){ref-type="fig"}). These data indicate that far fewer astrocytes than neurons express *C9orf72*, and that *C9orf72* promoter activity is not only regulated in a cell-type specific manner but also in a region dependent way.Figure 5Differential *C9orf72* promoter activity in cortical and spinal cord astrocytes. (**a**) Representative images of layer 5 (L5) in primary motor cortex (M1, *top*) and the ventral horn of the spinal cord (SC, *bottom*) of *C9orf72* ^*LacZ/*+^ mice immunostained for S100β-positive astrocytes (green), Olig2-positive oligodendrocyte lineage cells (cyan), and β-gal (red). Asterisks indicate S100β-positive Olig2-positive oligodendrocyte lineage cells and S100β-positive motor neurons which were excluded from the analysis. High magnification images show an astrocyte in motor cortex which lacks β-gal puncta (*top right*) and astrocytes in the spinal cord which contain β-gal puncta (*bottom right*). Scale bars: 50 μm and 10 μm. Percentage of astrocytes containing β-gal puncta in L5 of primary motor and primary somatosensory cortex (**b**) L5 motor cortex: 1 of 92 cells, 1.1%, L5 somatosensory cortex, 1 of 98 cells, 1.0%, *n* = 3 mice for each group, *p* = 0.9641, Chi-Square test) and spinal cord (**c**) dorsal spinal cord: 15 of 84 cells, 17.9%, ventral spinal cord: 16 of 78 cells, 20.5%, *n* = 3 mice for each group, *p* = 0.6676, Chi-Square test). Recent studies have demonstrated that *C9orf72* plays an important role in regulating the immune system^[@CR24],\ [@CR25],\ [@CR28]^ and have reported high levels of *C9orf72* transcript expression in bulk cellular populations enriched for microglia^[@CR28],\ [@CR46]^. In contrast, a prior study detected *C9orf72* promoter activity in only a small subset of microglia in the anterior horn of the spinal cord^[@CR32]^. Similarly, we found that only a small proportion of Iba1-immunoreactive microglia in the cortex and spinal cord had detectable *C9orf72* promoter activity (Fig. [6a--c](#Fig6){ref-type="fig"}). These findings suggest that the level of expression of *C9orf72* varies substantially across the population of microglial cells in the brain and spinal cord, with only a few cells expressing high levels and the majority expressing low or undetectable levels.Figure 6*C9orf72* promoter activity is detected in few microglia. (**a**) Representative images of layer 5 (L5) of primary motor cortex (M1, *top*) and the ventral horn of the spinal cord (SC, *bottom*) of *C9orf72* ^*LacZ/*+^ mice immunostained for Iba1-positive microglia (green) and β-gal (red). High magnification images show microglia which do not contain β-gal puncta in the motor cortex (*top right*) and the spinal cord (*bottom right*). Scale bars: 50 μm and 10 μm. The percentage of microglia containing β-gal puncta in L5 of the primary motor and primary somatosensory cortex (**b**) L5 of somatosensory cortex: 16 of 127 cell, 16.0%; L5 of motor cortex: 22 of 160 cells, 15.2%; *n* = 7 mice for each group, *p* = 0.7749, Chi-Square test) and the spinal cord (**c**; dorsal spinal cord: 14 of 106 cells, 13.2%; ventral spinal cord: 2 of 34 cells, 5.9%, *n* = 3 mice for each group, *p* = 0.2427, Chi-Square test). Oligodendrocytes provide metabolic support critical for neuronal health particularly for long axons. The degeneration of oligodendrocytes seen in ALS patient tissue and in mouse models of the disease suggests that these cells play an important role in the selective vulnerability of specific neuronal populations in ALS^[@CR48]--[@CR51]^. Removal of the ALS mutation *SOD1* ^*G37R*^ from the oligodendrocyte lineage significantly delays disease onset and extends the lifespan of ALS model mice^[@CR52]^, providing further support for an important contribution of oligodendrocyte dysfunction in the disease. Although *C9orf72* expression has been reported in oligodendrocytes^[@CR26],\ [@CR28]^, the distribution of *C9orf72* promoter activity in the cortex and spinal cord is still poorly understood. We found that more than 25% of CC1-immunoreactive oligodendrocytes in layer 5 of primary somatosensory cortex and of primary motor cortex exhibited *C9orf72* promoter activity. Interestingly, we detected β-gal puncta in a significantly greater percentage of CC1-positive oligodendrocytes in motor cortex than primary somatosensory cortex (Fig. [7a,b](#Fig7){ref-type="fig"}; *p* = 0.014, Chi-Square test), indicating enrichment of *C9orf72* expression in cortical regions which undergo the most profound degeneration in ALS. An even greater percentage of oligodendrocytes in the spinal cord exhibited detectable *C9orf72* promoter activity (Fig. [7a,c](#Fig7){ref-type="fig"}); the highest percentage of β-gal-positive, CC1-positive oligodendrocytes was found in the ventral white matter of the spinal cord, where almost 65% of oligodendrocytes were positive for β-gal (Fig. [7c](#Fig7){ref-type="fig"}; *p* \< 0.0001, Chi-Square test). *C9orf72* promoter activity was also observed in approximately 20--35% of cortical NG2-immunoreactive oligodendrocyte precursor cells (OPCs; Fig. [7d,e](#Fig7){ref-type="fig"}). More than 50% of OPCs of the spinal cord also exhibited *C9orf72* promoter activity (Fig. [7d,f](#Fig7){ref-type="fig"}). These data indicate that *C9orf72* promoter activity is detected in a much larger percentage of oligodendrocytes and OPCs than in astrocytes and microglia, and that *C9orf72* promoter activity is specifically enriched in oligodendrocytes in regions thought to degenerate in ALS.Figure 7*C9orf72* promoter activity in oligodendrocytes and OPCs is enriched in cortical regions associated with neurodegeneration in amyotrophic lateral sclerosis. (**a**) Representative images of layer 5 (L5) of the primary motor cortex (M1, top), ventral horn of the spinal cord (SC, middle), and the ventral white matter of the spinal cord (SC WM, bottom) of *C9orf72* ^*LacZ/*+^ mice immunostained for CC1-positive oligodendrocytes (green) and β-gal (red). High magnification images show colocalization of β-gal puncta within oligodendrocytes (right panels).The percentage of oligodendrocytes containing β-gal puncta in L5 of primary somatosensory (S1) and motor (M1) cortex (**b**) S1 L5: 38 of 167 cells, 26.0%; M1 L5: 82 of 234 cells, 35.0%; n = 5 mice for each group, p = 0.0140, Chi-Square test) and spinal cord (**c**) dorsal spinal cord: 209 of 533 cells, 39.2%; ventral spinal cord: 191 of 525 cells, 36.4%; spinal cord white matter: 283 of 438 cells, 64.6%, n = 3 mice for each group, p \< 0.0001 Chi-Square) (**d**) Representative images of L5 of primary motor cortex (M1, top) and the ventral horn of the spinal cord (bottom) immunostained for NG2-positive oligodendrocyte precursor cells (OPCs; green) and β-gal (red). High magnification images show colocalization of β-gal puncta within oligodendrocytes (right panels). The percentage of OPCs containing β-gal puncta in L5 of primary motor (M1) and primary somatosensory (S1) cortex (**e**) S1 L5: 21 of 92 cells, 22.6%; M1 L5: 33 of 92 cells, 36.5%; n = 5 mice for each group, p = 0.0520, Chi-Square test) and spinal cord (**f**) dorsal spinal cord: 39 of 75 cells, 52.0%; ventral spinal cord: 44 of 65 cells, 67.7%; n = 3 mice for each group, p = 0.0594, Chi-Square test). Scale bars: 50 μm and 10 μm. Discussion {#Sec8} ========== Genes that are mutated in ALS are widely expressed in the CNS, creating unique vulnerabilities in distinct cell types. To understand how these mutations alter cellular physiology we must define both the tissue and cellular expression patterns of these mutant genes. Here, we show that although the distribution of *C9orf72* promoter activity follows overall cellular density, there is striking enrichment in neuronal and glial cell types that degenerate in ALS. *C9orf72* promoter activity was detected in a significantly higher percentage of both corticospinal and spinal motor neurons than in neighboring neuronal cell types. Similarly, we demonstrate that *C9orf72* promoter activity was enriched in oligodendrocytes in cortical regions associated with ALS pathology compared to unaffected areas. In contrast to neurons and oligodendrocytes, we detected *C9orf72* promoter activity in few astrocytes and microglia^[@CR32]^. Together, these data indicate that the pattern of degeneration seen in ALS reflects the distribution of *C9orf72* promoter activity, suggesting that cell autonomous effects in these populations of neurons and glia may account for their loss. The widespread expression of *C9orf72* promoter activity we observed in this Neo-deleted KOMP mouse is consistent with the overall expression patterns reported in recent studies of the KOMP mouse^[@CR26],\ [@CR32]^ as well as *C9orf72* RNA expression data^[@CR26]--[@CR28],\ [@CR53]^, and indicates that the distribution of *C9orf72* promoter activity grossly follows cellular density throughout the CNS. However, an analysis of specific cell types demonstrated that widely varying fractions of cells expressed *C9orf72*. For example, a large majority of both inhibitory and excitatory neurons exhibited *C9orf72* promoter activity while a small minority of astrocytes did. The widespread expression of *C9orf72* by interneurons which we demonstrate here is particularly interesting, given their pivotal role in regulating neuronal excitability and brain states, and dysfunction of cortical inhibitory neurons may contribute to the cortical hyperexcitability seen in patients with *C9orf72* disease^[@CR54],\ [@CR55]^. Our data also indicate that *C9orf72* promoter activity is enriched in cell types and brain regions that undergo degeneration in ALS. A higher percentage of corticospinal neurons express *C9orf72* than either neighboring neurons in motor cortex or neurons in L5 of somatosensory cortex. Similarly, a higher percentage of oligodendrocytes in L5 of motor cortex express *C9orf72* than in L5 of somatosensory cortex. We also found that approximately 25--50% of oligodendrocytes and OPCs in the cortex and spinal cord expressed *C9orf72*. Importantly, we show that a greater fraction of oligodendrocytes in L5 of motor cortex expresses *C9orf72*, providing further evidence that cell types that degenerate in ALS express *C9orf72* more widely than other cell types in the CNS. Together with the evidence of widespread degeneration of oligodendrocytes in ALS patient tissue and mouse models^[@CR48]--[@CR51]^, and the finding that removal of the *SOD1* ^*G37R*^ ALS mutant transgene selectively from oligodendrocyte lineage cells significantly delays disease onset and extends the lifespan of ALS model mice^[@CR52]^, these data support the hypothesis that oligodendroglia are critical primary players in the selective degeneration of CNS regions in ALS. The pathological mechanisms associated with expression of mutated *C9orf72* may confer selective vulnerability on these cell types during the course of ALS and may help to explain the regional pattern of degeneration observed in ALS. A hallmark of *C9orf72*-mediated disease is cerebellar pathology including RNA foci and neuronal inclusions^[@CR38]--[@CR41]^. Prior studies have also detected high levels of *C9orf72* expression in the cerebellum^[@CR23],\ [@CR42]^. Consistent with this work, we detected *C9orf72* promoter activity in both Purkinje cells and cerebellar granule cells. Interestingly, a higher percentage of granule cells exhibited detectable *C9orf72* promoter activity than neighboring Purkinje cells, contributing to the intense staining we detected in the granule cell layer. Although there are marked neuropathological findings in the cerebellum in *C9orf72*-mediated disease and high levels of *C9orf72* expression, the relationship between these findings and any neurodegeneration remains unclear^[@CR38],\ [@CR56]^. In contrast to the enrichment of *C9orf72* promoter activity that we detected in neurons and oligodendrocyte lineage cells, we found that very few astrocytes and microglia exhibited detectable *C9orf72* promoter activity. Furthermore, there were striking regional differences in expression. For example, almost no cortical astrocytes exhibited *C9orf72* promoter activity while approximately 20% of the astrocytes in the spinal cord expressed *C9orf72*. These results are consistent with analyses of *C9orf72* transcript expression, although an early study did not detect *LacZ* expression in spinal cord astrocytes in a similar mouse line^[@CR24],\ [@CR26],\ [@CR28],\ [@CR32],\ [@CR46]^. Recent studies have demonstrated high levels of *C9orf72* expression in enriched microglia cell populations^[@CR26],\ [@CR28],\ [@CR46]^. This result is surprising in view of the small percentage of spinal cord microglia previously found to exhibit *C9orf72* promoter activity in the KOMP mouse^[@CR32]^. Not only did we also detect *C9orf72* promoter activity in only a small fraction of microglia in the spinal cord, we found that few microglia in the cortex exhibited detectable *C9orf72* promoter activity. It is possible that *C9orf72* transcript levels may depend on the activation state of a microglial cell, and that *C9orf72* expression is rapidly upregulated when cells are dissociated from the brain for gene expression analysis. Recent work demonstrating that immune challenge of monocytes increases *C9orf72* expression hints at this possibility, although unchallenged monocytes express higher levels of *C9orf72* than microglia from fetal human brain tissue^[@CR23]^. Alternatively, there may be widely varying levels of expression in different microglial cells, with a small fraction of microglial cells expressing high levels of *C9orf72* leading to high levels of transcript detection in bulk populations. Understanding the patterns of *C9orf72* expression in combination with the distribution of the protein in different cell types represents an essential step in elucidating the mechanisms of dysfunction in CNS cells in the spectrum of *C9orf72*-associated diseases. Several studies have demonstrated decreased *C9orf72* transcript and protein levels in patients with ALS and FTD^[@CR19]--[@CR23]^, suggesting that haploinsufficiency may contribute to ALS pathogenesis. Our analysis of the patterns of *C9orf72* expression suggests that the cell types that undergo degeneration in ALS, including corticospinal neurons, spinal motor neurons and oligodendrocytes, would be most affected by any *C9orf72* haploinsufficiency. However, the mechanisms underlying the range of phenotypes following *C9orf72* deletions in mouse models remains to be elucidated^[@CR24]--[@CR29]^, and understanding this variability remains an essential step for uncovering the cellular mechanisms underlying ALS and other *C9orf7*2-linked diseases. Methods {#Sec9} ======= Generation of *C9orf72*^*LacZ/*+^ mice {#Sec10} -------------------------------------- All experimental procedures were approved by the Johns Hopkins Animal Care and Use Committee and conducted in accordance with the guidelines of the National Institutes of Health and the Society for Neuroscience. The mice carrying a targeted deletion of *C9orf72* and insertion of *LacZ* were generated as described^[@CR29]^. In brief, several embryonic stem cell lines harboring an insertion that replaced exons 2--6 of the mouse *C9orf72* gene (3110043O21Rik) with *LacZ* (National Institutes of Health Knockout Mouse Project) were used for blastocyst injection to generate chimeric mice which were then selected for germline transmission. The original embryonic stem cells had the genetic background of C57BL/6N-Atm1Brd and the derived mice were maintained on the C57BL/6 background. Male mice bearing the targeted allele were crossed with *Sox2-Cre* recombinase transgenic female mice (Jackson Laboratory, 008454), maintaining the C57BL/6 background, to remove the LoxP-flanked neomycin selection cassette in all progeny. Subsequent breeding eliminated the Sox2-Cre transgene from our mouse line. Six to eight-week-old Neo-deleted *C9orf72* ^*LacZ/*+^ heterozygous mice were used for all experiments. The genotyping primers were the following: gaatggagatcggagcacttatgg (wild-type, forward), gccttagtaactaagcttgctgccc (wild-type, reverse), gcacaagctatgttcatttgg (KO, forward), gactaacagaagaacccgttgtg (KO, reverse). Immunohistochemistry and X-galactosidase staining {#Sec11} ------------------------------------------------- Mice were deeply anesthetized with sodium pentobarbital (100 mg/kg) and perfused transcardially with 4% paraformaldehyde in 0.1 M sodium phosphate buffer. Brain and spinal cord tissue was isolated and postfixed in this solution overnight at 4 °C, then washed in phosphate buffer. Spinal cord tissues were cryoprotected in 30% sucrose, and sectioned at 35 µm thickness on a cryostat. Brain sections, 35 μm thick, were prepared using a vibratome (VT-1000S, Leica). To obtain sections of motor cortex, the brain was mounted in coronal orientation on a 15° ramp prior to cutting slices. To obtain sections of somatosensory cortex, the brain was mounted in parasagittal orientation on a 30° ramp prior to cutting slices. Cerebellar slices were obtained in the parasagittal orientation, after mounting the cerebellum on a flat block. Free-floating sections were permeabilized with 0.3% Triton X-100 in 0.1 M sodium phosphate buffer for at least 5 min and then blocked with 0.3% Triton X-100 and 5% normal donkey serum in 0.1 M sodium phosphate buffer (blocking solution) for at least 1 h at room temperature. When performing immunohistochemistry for calbindin, 5% normal goat serum was also added. Sections were then incubated with primary antibodies prepared in blocking solution overnight at 4 °C and then incubated with secondary antibodies in blocking solution for at least 2 h at room temperature. Primary antibodies used included the following: rabbit anti-β-galactosidase (1:5000; gift from Dr. Joshua Sanes, Harvard University, Cambridge, MA), guinea pig anti-NG2 (1:30,000; Dr. Dwight Bergles, Johns Hopkins University, Baltimore, MD), mouse anti-APC (CC1; 1:50; Millipore, Cat. No. OP80), guinea pig anti-Olig2 (1:20,000; gift from Dr. Ben Novitch, University of California, Los Angeles, CA), mouse anti-S100β (1:400; Sigma, Cat. No. S2532), goat anti-Iba1 (1:250; Novus, Cat. No. NB100--1028), mouse anti-NeuN (1:500; Millipore, Cat. No. MAB377), mouse anti-Parvalbumin (1:300; Swant, Cat. No. PV235), goat anti-ChAT (1:500; Millipore, Cat. No. AB144P) and chicken anti-calbindin D28 (1:100, EnCor Biotechnology Inc., Cat. No. CPCA-Calb). Secondary antibodies used, all raised in donkey except for the goat anti-chicken secondaries, included the following: Alexa Fluor 488-, 546-, and 647- as well as Cy2-, Cy3-, and Cy5-conjugated secondary antibodies to rabbit, guinea pig, mouse, and goat (1:2000; Invitrogen and Jackson ImmunoResearch). For X-galactosidase staining, sections were incubated in 1 mg/mL X-gal (Invitrogen) in a solution of 5 mM potassium ferricyanide (Sigma), 5 mM potassium hexacyanoferrate(II) trihydrate (Sigma), and 2 mM magnesium chloride (Sigma) in PBS for 24 h at 37 °C. Corticospinal neuron identification with retrograde neuronal tracers {#Sec12} -------------------------------------------------------------------- To identify corticospinal neurons, retrograde neuronal tracers were injected into the spinal cord. Mice were anesthetized with ketamine (50 mg/kg), dexmedetomidine (25 μm/kg) and the inhalation anesthetic, isoflurane (0.5--3%). A small laminectomy was performed over the left cervical cord. After removing one vertebra (C5-C6), 100--200 nl of retrograde neuronal tracer, either green Retrobeads (Lumafluor) or Alexa Fluor 488-conjugated cholera toxin B (Invitrogen), were pressure injected through a glass pipet (20--30 μm tip, inner diameter, Drummond), 500 μm lateral to the midline of the exposed spinal cord at a depth of 1000 μm. Buprenorphine (0.05 mg/kg) was administered to all animals post-operatively as an analgesic. Mice were sacrificed 7--12 days after tracer injections and brain sections were cut and processed as described above. Image acquisition and analysis {#Sec13} ------------------------------ To assess the overall distribution of *C9orf72* promoter activity, fluorescence images were collected on an AxioImager M1 microscope (Carl Zeiss). Intensity plots for DAPI, β-gal, and ChAT were made by averaging the signal intensity along the horizontal axis using ImageJ software (NIH). To analyze the distribution of *C9orf72* promoter activity within cell types, fluorescence images were acquired with an LSM 510 Meta confocal microscope (Carl Zeiss). Stacks of confocal images (0.3 µm *z*-interval) were imported into Imaris software for three-dimensional analysis. Surface renderings of NeuN-positive, ChAT-positive, PV-positive, calbindin-positive, Iba1-positive, CC1-positive, NG2-positive, and S100β-positive/Olig2-negative cells were created, the locations of β-gal-positive puncta were marked using the Spots function, and any cell with one or more β-gal-positive puncta within the three-dimensional rendering was counted as exhibiting *C9orf72* expression. As the cerebellar granule cells were very tightly packed, there were occasionally small groups of NeuN-positive neurons that could not be clearly segmented into individual neurons. These groups were eliminated from the analysis. To determine whether the distribution of β-gal puncta related to the underlying cellular distribution, the channel containing the β-gal signal was flipped horizontally while leaving the cell type-specific channels unchanged. The same counting procedure was then applied to this new configuration. Analysis of gene expression {#Sec14} --------------------------- The transcript analysis was performed using RNA-sequencing (RNA-seq) data from a previously published study^[@CR28]^. The RNA-seq data were derived from spinal cords of 3-month old wild-type and *C9orf72* (3110043O21Rik) knockout mice (three animals from each group). These knockout mice and the independent line analyzed in the present study were generated from a common source of mouse embryonic stem cells (DEPD00552, the Mouse Biology Program, [www.mousebiology.org](http://www.mousebiology.org)). The RNA-seq reads were aligned to mouse genome build mm10 using HISAT2 using standard parameters^[@CR57]^. The BAM files were then visualized using integrated genome viewer^[@CR58]^. The Sashimi plot was generated to visualize the RNA alignment as well as the splice junctions with a minimum junction coverage of 2. Statistics {#Sec15} ---------- The Chi-Square test was used to determine whether there was a significant difference in the expected frequencies of β-gal immunoreactive cells between cell types or between different regions. For the flipped image test, a paired *t-*test was used after confirming the normality of data distribution with the Shapiro Wilk normality test. *p* \< 0.05 was considered to be statistical significant, and the asterisk was used to mark the statistical significance on the graphs, \* for *p* \< 0.05, \*\* for *p* \< 0.01, and \*\*\* for *p* \< 0.001. Electronic supplementary material ================================= {#Sec16} Supplementary information Abraham J. Langseth and Juhyun Kim contributed equally to this work. **Electronic supplementary material** **Supplementary information** accompanies this paper at doi:10.1038/s41598-017-05864-2 **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank Dr. Joshua R. Sanes (Harvard University) for providing the anti-β-gal antibody, and Karen Conchina and Justin Chu for their assistance in the generation and maintenance of the mice. This work was supported by Target ALS (D.E.B., S.P.B.), the Robert Packard Center for ALS Research (J.W., S.P.B.), the U.S. Department of Defense (J.W.) and the National Institutes of Health (J.W.: NS074324 and NS089616; S.P.B.: NS098819; D.E.B.: NS050274). J.K. was supported by a National Research Foundation of Korea Fellowship (NRF‐2011‐357‐E00005). A.J.L. was supported by a National Multiple Sclerosis Society Postdoctoral Fellowship (FG 20114-A-1). S.P.B. is supported by a Klingenstein-Simons Fellowship in the Neurosciences. J.W., D.E.B., and S.P.B. conceived of the project. A.J.L., J.K., D.E.B., and S.P.B. designed the study. J.E.U., Y.S. and J.W. generated the mice. A.J.L. and J.K. acquired and analysed the anatomical data. H.-Y. H. analyzed the RNA-sequencing data. A.J.L., J.K., D.E.B., and S.P.B. interpreted the data and drafted the manuscript. J.E.U. and J.W. contributed to the writing of the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

{#sp1 .457}

Introduction {#Sec1} ============ Superparamagnetic iron oxide nanoparticles (SPION) are widely used for biomedical applications, including magnetic resonance imaging (MRI), magnetic particle imaging (MPI), magnetic fluid hyperthermia (MFH), separation of biomolecules, and targeted drug and gene delivery \[[@CR1]--[@CR3]\]. This widespread list of applications not only results from the magnetic properties of SPION, but also from the capability of synthesizing them in different sizes and shapes. For all of the above applications, SPION should ideally have a high magnetization value, a size below 100 nm and a narrow size distribution \[[@CR4], [@CR5]\]. SPION are typically based on Fe~3~O~4~ and/or Fe~2~O~3~. They can be synthesized using various methods, such as co-precipitation \[[@CR5], [@CR6]\], thermal decomposition \[[@CR7]\], sol--gel \[[@CR8]\], microemulsion \[[@CR9]\], hydrothermal \[[@CR10]\], and electrochemical synthesis \[[@CR11]\]. The co-precipitation technique is among the most successful, most commonly employed and most cost-effective methods for high-yield synthesis. However, strategies are needed to overcome the most important limitation of this method, i.e. the very broad particle size distribution of the resulting SPION mixture \[[@CR5], [@CR6]\]. In this study, we describe a straightforward, easily implementable and broadly applicable centrifugation protocol to obtain relatively monodisperse SPION from a polydisperse starting mixture prepared using the co-precipitation technique. As a result of their refined size distribution, the obtained optimized SPION dispersions showed substantially improved performance in MRI, MPI and MFH compared to the crude starting formulation, as well as to commercial SPION products, such as Resovist® and Sinerem®. In this context, it is important to keep in mind that not the centrifugation protocol per se, but the eventual development of a SPION formulation with a very well-defined size and with a very narrow size distribution (and its consequent more optimal use for diagnostic and therapeutic purposes) is the objective of our work. Thus far, no systematic study has been published on SPION size-isolation via sequential centrifugation, and no systematic analysis is available in which the performance of five size-isolated SPION sub-fractions (and clinically/commercially relevant controls) is head-to-head compared in MRI, MPI and MFH setups. Results and discussion {#Sec2} ====================== SPION preparation and size-isolation {#Sec3} ------------------------------------ Prototypic citrate-coated SPION were prepared via the standard co-precipitation technique, under nitrogen atmosphere \[[@CR5], [@CR6]\] (see "[Experimental](#Sec12){ref-type="sec"}" section for details). Based on this highly polydisperse starting batch, which we refer to as the "crude sample", five sequential rounds of centrifugation were performed to obtain much more monodispersed SPION subfractions. To this end, as depicted schematically in Fig. [1](#Fig1){ref-type="fig"}, the crude sample was transferred into 1.5 ml Eppendorf tubes and centrifuged at 14,000 rpm for 20 min. The resulting 1 ml of supernatant was collected and referred to as the "C1 sample". Subsequently, 0.1 ml of the bottom compartment in the Eppendorf tube that contained the largest nanoparticle fraction was resuspended in water. The obtained dispersion was then again centrifuged, the top 1 ml was collected as the "C2 sample", and the bottom 0.1 ml was again resuspended and re-centrifuged. These steps were sequentially repeated to obtain five fractions of relatively monodisperse SPION samples. These fractions are referred to as C1--C5. The crude starting mixture, Resovist® and Sinerem® are referred to as C, R and S, respectively. Multiple systematic experiments were performed to identify the optimal centrifugation speeds and times for obtaining monodispersed SPION with well-defined sizes. The optimum conditions for size-isolation are presented in Fig. [1](#Fig1){ref-type="fig"}. The production efficiencies of the size-isolated fractions C1, C2, C3, C4 and C5 were approximately 7, 29, 23, 18 and 11%, respectively.Fig. 1SPION size-isolation via sequential centrifugation. Schematic overview of the centrifugation protocol to obtain monodispersed SPION with different hydrodynamic diameters from a crude mixture of polydisperse SPION. The polydisperse SPION sample (C) was transferred into 1.5 ml Eppendorf tubes and centrifuged at 14,000 rpm for 20 min. The resulting 1 ml of supernatant was collected (C1). 0.1 ml of the bottom compartment in the Eppendorf tube was resuspended in water and again centrifuged, and the top 1 ml was collected (C2). These steps were repeated multiple times, with optimized centrifugation times and speeds, to obtain three additional fractions of monodisperse SPION samples (C3--C5). The different fractions were subsequently analyzed for magnetic resonance imaging (MRI), magnetic particle imaging (MPI) and magnetic fluid hyperthermia (MFH) performance, and compared to the crude sample (C), to Resovist® and to Sinerem® Despite the large number of previous publications describing the synthesis of iron oxide nanoparticles, the tools and technologies for their size separation are relatively limited. Techniques employed to control average particle size and polydispersity can be based on the use of magnetic/electric fields, porous media, and mass- and density-based purification \[[@CR12]--[@CR14]\]. Fortin and colleagues for instance synthesized citrate-coated nanocrystals of maghemite and cobalt ferrite by alkaline co-precipitation, and size-sorted the nanoparticles by successive electrostatic phase separation \[[@CR15]\]. Magnetic field-flow fractionation (MFFF) uses a homogeneous external magnetic field applied orthogonal to the direction of flow, to achieve efficient separation of particles \[[@CR12]\]. Nonmagnetic size-exclusion chromatography (SEC) is another frequently used method for size separation of iron oxide nanoparticles. The fractions separated by SEC and MFFF have similar size distributions. However, MFFF is faster and has a higher capacity \[[@CR12], [@CR16]\]. In addition to the above techniques, differential magnetic catch-and-release (DMCR) has recently been established to size-sort magnetic nanoparticles. DMCR, like MFFF, relies on an external magnetic field to separate magnetic species \[[@CR17]\]. High-gradient magnetic separation (HGMS) is a column flow method used to isolate iron oxide nanoparticles from a nonmagnetic medium \[[@CR18]\]. Capillary electrophoresis (CE) is used for the separation of colloidal nanoparticles in an electric field. CE requires specialized equipment, because of the high electric field. Electrical field-flow fractionation (ElFFF) separates iron oxide nanoparticles based on their size and electrophoretic mobility but without the drawbacks of CE \[[@CR12], [@CR16]\]. As compared to the above techniques, the here presented centrifugation method is somewhat more time- and labor-intensive, but it is also easier to perform and more broadly applicable, because it does not require specialized equipment. Particle size, size distribution and surface charge {#Sec4} --------------------------------------------------- Figure [2](#Fig2){ref-type="fig"} shows the results obtained using TEM, DLS and NTA on the size and size distribution of the SPION formulations prepared and evaluated in this study. The reported TEM values which correspond to the average size were calculated on the basis of manually measuring at least 100 randomly chosen particles, using Image SP Viewer software. The average core sizes of the samples C1, C2, C3, C4 and C5 were 7.7 ± 1.6, 10.6 ± 1.8, 13.1 ± 2.2, 15.6 ± 2.8 and 17.2 ± 2.1 nm, respectively (Fig. [2](#Fig2){ref-type="fig"}a, b). This indicates that all five fractions are superparamagnetic, as SPION typically present superparamagnetic behavior when their core size is below 20 nm \[[@CR5]\]. The corresponding average hydrodynamic diameters obtained by DLS-based on intensity---for the five samples were 26.3 ± 1.2, 49.4 ± 1.1, 64.8 ± 2.1, 82.1 ± 2.3 and 114.6 ± 4.4 nm (Fig. [2](#Fig2){ref-type="fig"}c). The average sizes obtained using NTA were comparable to the values observed in DLS (Fig. [2](#Fig2){ref-type="fig"}d). The numerical values corresponding to the results presented in Fig. [2](#Fig2){ref-type="fig"}b--d are provided in Additional file [1](#MOESM1){ref-type="media"}: Table S1. The fact that the TEM sizes are smaller than those obtained via DLS and NTA can be explained by keeping in mind that DLS and NTA measure the hydrodynamic diameter of the citrate-coated SPION in aqueous solution incorporating surface-bound water layers in their measurement, while TEM determines the actual core size of dried nanoparticle formulations.Fig. 2Effect of sequential size-isolation on SPION size and size distribution. **a** TEM images and size distributions obtained by TEM. **b**--**d** Analysis of nanoparticle size obtained using TEM, DLS and NTA. **e** Polydispersity indices (PDI) assessed using DLS for the crude (C), C1--C5, Resovist® (R) and Sinerem® (S) samples. Results represent average ± standard deviation The results obtained using DLS, NTA and TEM demonstrate that both the core size and the hydrodynamic diameter gradually increase upon employing our centrifugation protocol. In this regard, it is important to note that from C1 to C5, the increase in hydrodynamic diameter (DLS) is much larger than the increase in core size (TEM). Equally important is the notion that the polydispersity indices (PDI) obtained from DLS confirmed that the samples C1--C5 have a much narrower size distribution than the crude sample, and also than Resovist® and Sinerem®. The PDI for the crude sample, for Resovist® and for Sinerem® were 0.28 ± 0.04, 0.26 ± 0.05 and 0.20 ± 0.04, respectively, while for C1--C5, all PDI's were approximately 0.10 (Fig. [2](#Fig2){ref-type="fig"}e). The size distribution results obtained by TEM are in good agreement with this (see the insets in Fig. [2](#Fig2){ref-type="fig"}a and the data presented in Fig. [2](#Fig2){ref-type="fig"}e). Based on these results, it is concluded that our sequential centrifugation protocol is highly useful for achieving relatively monodisperse SPION formulations. Consequently, it is considered to be a useful alternative to more complex synthetic methods to obtain relatively uniform SPION, such as thermal decomposition, which requires very high temperatures and which critically depends on efficient and tailored means for surface modification to eventually obtain water-dispersible SPION formulations \[[@CR7]\]. We also determined the zeta potential for the differently sized iron oxide nanoparticle samples (Additional file [1](#MOESM1){ref-type="media"}: Figure S1). The results confirm the expected highly negatively surface charge for all size-isolated fractions (C1--C5), which contributes to their high colloidal stability. SPION biocompatibility {#Sec5} ---------------------- Almost all SPION formulations were found to be biocompatible. Additional file [1](#MOESM1){ref-type="media"}: Figures S2--S4 document the observed cytotoxicity for the crude, C1--C5, Resovist® and Sinerem® samples studied by XTT, LDH and ROS assays. XTT analysis at iron concentrations of 0.1 and 1.0 mM showed no significant differences in the viability of NIH3T3 cells upon incubation with the samples C1--C5 as compared to Resovist® and Sinerem®. Interestingly, at iron concentrations of 5 and 10 mM, XTT-based viability assessment indicated that all monodispersed samples except for C1 had an even higher biocompatibility than Resovist® and Sinerem® (Additional file [1](#MOESM1){ref-type="media"}: Figure S2). The XTT findings were confirmed using the LDH assay (Additional file [1](#MOESM1){ref-type="media"}: Figure S3). At iron concentrations of 0.1 and 1 mM, no changes in NIH3T3 membrane damage were noted for C1--C5 as compared to Resovist® and Sinerem®, while at iron concentrations of 5 and 10 mM, LDH values (and membrane damage) were lower than for Resovist® and Sinerem® (again except for the smallest-sized batch C1). In line with this, analysis of ROS production in NIH3T3 cells showed that there was no significant change in the ROS content of cells exposed to the monodispersed samples C1--C5 as compared to the crude sample, Resovist® and Sinerem® (Additional file [1](#MOESM1){ref-type="media"}: Figure S4). Together, these results demonstrate that all monodispersed samples except for C1 have negligible toxicity. The higher cytotoxicity associated with the smallest particles is assumed to result from a more rapid and more extensive cellular uptake, as well as from a relatively larger surface area \[[@CR19]--[@CR21]\]. SPION stability in physiological media {#Sec6} -------------------------------------- All size-isolated SPION samples showed excellent stability in DI water (see columns 4 and 5 of Additional file [1](#MOESM1){ref-type="media"}: Table S1; demonstrating stable dispersion up to 6 months). This can be attributed to the highly negatively charged surface of the SPION. All SPION formulations also showed high colloidal stability in physiological media, i.e. in fetal bovine serum (FBS) and in bovine serum albumin (BSA). The monitoring of the samples by visual inspection up to 24 h implied the absence of aggregation of SPION (see Additional file [1](#MOESM1){ref-type="media"}: Figures S5a and S6a). In line with this, the hydrodynamic diameters and PDI obtained using DLS for 2, 6 and 24 h of incubation in physiological media did not show significant changes in size and size distribution (see Additional file [1](#MOESM1){ref-type="media"}: Figures S5b, c, S6b, c and Table S1). In good agreement with our findings, Yu et al. synthesized two different types of SPION with different surface coatings: tetramethylammonium hydroxide-coated SPION (T-SPION) and citrate-coated SPION (C-SPION). The C-SPION showed robust stability in biological media, while T-SPION aggregated rapidly in all media evaluated \[[@CR22]\]. Magnetic properties {#Sec7} ------------------- Field-dependent magnetization analysis of the C1--C5 samples showed no discernible hysteresis, demonstrating that they are superparamagnetic (Fig. [3](#Fig3){ref-type="fig"}a). For biomedical applications, iron oxide nanoparticles with superparamagnetic behavior are preferred, because in case of superparamagnetic materials, the magnetization drops to zero after removing the applied magnetic field. This implies that due to lack of coercive forces or remanence, it keeps the nanoparticles from sticking together, avoiding aggregation and the formation of clots in the blood stream, which could lead to serious adverse events \[[@CR23]\].Fig. 3Magnetic characterization. **a** Field-dependent magnetization at 5 K. **b** Saturation magnetization at 5 K of the crude SPION mixture (C), the size-isolated samples C1--C5, Resovist® (R) and Sinerem® (S). Results were normalized to Fe content The saturation magnetizations (M~s~) of samples were very high, indicating an excellent magnetic response to the magnetic field (Fig. [3](#Fig3){ref-type="fig"}b). Three important observations were obtained by these analyses: first, comparing the M~s~ values of the samples C2 and C3 at 5 K and 30 kOe (73.8 and 82.5 emu/g, respectively) to those of Resovist® and Sinerem® (53.1 and 28.8 emu/g, respectively) illustrates the good magnetic properties of C2 and C3. Second, the M~s~ values for C2 and C3 are approximately three-quarters of the M~s~ value of bulk magnetite, which is \~ 100 emu/g at 5 K and 30 kOe \[[@CR24]\]. Third, the magnetization reaches 94% of its maximum value for C2 and 93% of its maximum value for C3 in magnetic fields as low as 5 kOe, underlining the suitability of these samples for the envisaged applications. Field-cooled (FC) magnetization measurements were also carried out, in an applied field of 1000 Oe, at temperatures ranging from 5 to 300 K. As shown in Additional file [1](#MOESM1){ref-type="media"}: Figure S7, the FC curves demonstrate only a very little decrease with temperature for all nanoparticle samples tested, and the results obtained are in good agreement with those of saturation magnetization analyses. Both MRI and MPI rely on the use of magnetic nanoparticles with strong saturation magnetization, high magnetic susceptibility and no coercivity. Similarly, also for MFH, the amount of saturation magnetization should be as high as possible, to guarantee efficient heating under an alternating magnetic field \[[@CR23]\]. Saturation magnetization of SPION depends not only on core size, but also on other parameters, such as size distribution, type of coating, chemical composition (with magnetite being better than maghemite) and crystalline structure. Generally, a larger particle size results in higher saturation magnetization values and in a better performance in MRI, MPI and MFH. However, when the particle size is too large, magnetic nanoparticles become ferromagnetic and the saturation magnetization drops, which is undesired for biomedical applications. For the C1--C5 samples, field-dependent magnetization analysis revealed that all fractions are in the superparamagnetic range. Increasing the size gradually approaches ferromagnetic behavior, explaining the somewhat lower saturation magnetization values for C4 and C5 as compared to C2 and C3. Also, the low saturation magnetization for C4 and C5 compared to C2 and C3 could be explained on the basis of a more polycrystalline structure of the samples. Conversely, it is important to keep in mind that smaller-sized nanoparticles are typically preferred in vivo, e.g. because they can more readily exploit vascular leakiness in tumors and at sites of inflammation, and because they allow for deeper target tissue penetration. These considerations exemplify that it is crucial to identify the optimal size for the anticipated biomedical application \[[@CR25], [@CR26]\], and they underline the importance of developing tools, such as the centrifugation protocol presented here, to prepare SPION formulations with distinct sizes and with low polydispersity. Another important thing to keep in mind is that sometimes the saturation magnetization is found to be lower than expected. This reduction in magnetic performance of the nanoparticles can be attributed to the existence of a \"magnetically dead layer\" on their surfaces. Because of this magnetically dead layer, the magnetic diameter is smaller than the physical diameter, sometimes by several nanometers. Saturation magnetization is proportional to the magnetic diameter, not physical diameter \[[@CR27]--[@CR29]\]. As an example to illustrate this, Unni and colleagues synthesized two series of iron oxide nanoparticles with a similar diameter of 21 nm via thermal decomposition; the MS value was 17 emu/g for one nanoparticle, and 74 emu/g for the other \[[@CR27]\]. Kemp et al. produced monodisperse magnetite nanoparticles with diameters in the range between 15 and 30 nm by thermolysis and they varied oleic acid ratios for size control. With increasing particle size, there was no clear trend in saturation magnetization (sometimes increasing and sometimes decreasing) \[[@CR28]\]. Such irregularities were also observed by Baaziz et al. for iron oxide nanoparticles with diameters between 4 and 28 nm \[[@CR29]\]. The lower MS values for the samples C4 and C5 as compared to C2 and C3 can be explained by taking the above notions into account. Magnetic resonance imaging {#Sec8} -------------------------- All SPION samples showed excellent performance as contrast agent for magnetic resonance imaging (MRI). Figure [4](#Fig4){ref-type="fig"} and Additional file [1](#MOESM1){ref-type="media"}: Figures S8--10 show T~1~- and T~2~-weighted MR images and quantification of key MRI parameters for the crude, C1--C5, Resovist® and Sinerem® samples \[i.e. relaxivities (r~1~, r~2~), relaxation rates (1/T~1~, 1/T~2~) and relaxivity ratios (r~2~/r~1~)\]. Figure [4](#Fig4){ref-type="fig"} indicates that all newly prepared samples, i.e. both the monodisperse and the polydisperse SPION, have transverse relaxivities (r~2~) greater than Resovist® and Sinerem®. Interestingly, while the crude starting mixture and Resovist® were both highly polydisperse, the r~2~ value of the former was found to be two times higher than that of the latter.Fig. 4Magnetic resonance imaging of size-isolated SPION. MRI of the crude, C1--C5, Resovist® and Sinerem® samples upon characterization on a 3 T clinical scanner. **a** T~1~- and T~2~-weighted MR images of the samples at a concentration of 0.01 mM. MR images for other SPION concentrations are provided in Additional file [1](#MOESM1){ref-type="media"}: Figure S8. **b** and **c** Longitudinal (r~1~) and transversal (r~2~) relaxivities of the samples in water. Values represent average ± standard deviation of three independent samples After sequential centrifugation, the r~2~ values of the monodisperse SPION gradually increased up until the third round of centrifugation. The C3 sample with 13.1 ± 2.2 nm core size possessed the most optimal MRI capabilities, with an r~2~ value of 434 mM^−1^ s^−1^. It produced 3.3 and 5.5 times more contrast in T~2~-weighted imaging than Resovist® (130 mM^−1^ s^−1^) and Sinerem® (79 mM^−1^ s^−1^), respectively. A number of studies have demonstrated that the core size, the size distribution and the magnetization of SPION are key factors influencing the transverse relaxation rate (1/T~2~) \[[@CR15], [@CR30]\]. The trend for the r~1~ values for the samples C1--C5 was found to be similar to that observed for the r~2~ values. The efficiency of a T2 contrast agent relies on the r2/r1 ratio in addition to the r2 value \[[@CR31]\]. In this context, it is important to note that for all size-isolated samples, it can be concluded that there is a specific enhancement of the r~2~/r~1~ ratio in comparison to Resovist® and Sinerem® (Additional file [1](#MOESM1){ref-type="media"}: Figure S10), confirming the suitability of these samples for T~2~-weighted MR imaging. Saraswathy and colleagues synthesized citrate-coated iron oxide nanoparticles with a similar coating and with a similar core size as C3 sample. They employed this SPION formulation for in vivo magnetic resonance imaging of liver fibrosis. The values for r~1~ and r~2~ were 2.69 and 102 mM^−1^ s^−1^, respectively \[[@CR32]\]. Comparing the r~2~/r~1~ value of their formulation (i.e. 37.9) to that of our C3 sample (i.e. 84.4) exemplifies the usefulness and the potential added value of our sequential size-isolation protocol. Smolensky et al. investigated the effect of multiple parameters, including particle size and shape, temperature and magnetic field strength, on the longitudinal and transverse relaxivities of iron oxide nanoparticles. According to their findings, r~2~ values increased linearly with increasing core size (from 4.9 to 18 nm), while r~1~ values remained relatively constant for particles with core sizes larger than 8 nm \[[@CR33]\]. Surface coating and nanoparticle aggregation are also highly important parameters. Blanco-Andujar and coworkers studied the evolution of r~2~ with SPION aggregate size \[[@CR34]\]. In case of small clusters, nanoparticles are homogeneously dispersed in water and protons can readily diffuse between the magnetic cores. Under these conditions, r~2~ values gradually increase with hydrodynamic diameter (up to approx. 80 nm). At a size of 80--90 nm, there is no further increase in r~2~. If the size exceeds 90 nm, r~2~ values start to decrease with increasing size, due to reductions in surface accessibility and proton exchange rate. This trend is in line with our results, showing reductions in r~2~ values when the hydrodynamic diameter goes beyond 70 nm (r~2~ values for C4 and C5 are 398 and 350 mM^−1^ s^−1^, respectively, as compared to 434 mM^−1^ s^−1^ for C3). Magnetic particle imaging {#Sec9} ------------------------- SPION are important tracer materials for magnetic particle imaging (MPI). MPI is a novel and increasingly popular hot-spot imaging technique that can be employed to visualize magnetic nanoparticles with very high temporal and spatial resolution. MPI is able to provide real-time 3D imaging information on the localization and concentration of magnetic nanoparticles, and it can be employed for multiple medical imaging applications \[[@CR35]\]. The potential usefulness of MPI strongly depends on the availability of size-optimized SPION to generate high quality images. As a matter of fact, MPI contrast generation critically depends on both SPION size and size distribution, since both parameters strongly affect the magnetization response. Resovist® was originally developed as a contrast agent for MRI. In recent years, it has also been extensively employed for MPI, because of its large magnetic moment. At the moment, Resovist® is the most extensively employed SPION formulation for MPI. From TEM images, it is known that Resovist® mainly consists of particles with an average core diameter of 5.8 ± 2.5 nm, many of which are agglomerated in clusters (Fig. [2](#Fig2){ref-type="fig"}a). It is assumed that these aggregates, which are formed by small elementary particles, are responsible for its good MPI performance \[[@CR26]\]. However, the MPI performance of Resovist® still leaves significant room for improvement. As result of this, in recent years, ever more scientists have started to work on the development of better SPION formulations for MPI \[[@CR26], [@CR36]\]. Figure [5](#Fig5){ref-type="fig"}a shows the MPI signal-to-noise (SNR) values of the different SPION formulations used in this study, obtained at the 4th harmonic frequency of the drive field. It also shows the full width at half maximum (FWHM) values, and the hysteresis loss determined from the point spread function (PSF) measurements. To allow for a quantitative comparison, it is generally considered to be sufficient to read the SNR at one harmonic frequency. This is typically the 4th harmonic frequency (Fig. [5](#Fig5){ref-type="fig"}a). Additional file [1](#MOESM1){ref-type="media"}: Figure S11 shows the SNR values for other harmonic frequencies. To compare the MPI performance of the different samples, SNR values were normalized to the iron concentration inside the probe volume. The normalized SNR values for C2 and C3 were found to be much higher than for all other samples. At the 4th harmonic frequency, the normalized SNR for C2 was 2.3 and 7.0 times higher than for Resovist® and Sinerem®, respectively. In addition, FWHM and hysteresis loss analysis showed that C2 and C3 were almost as good as Resovist®. Lower FWHM and hysteresis loss values refer to a higher achievable spatial resolution and to a lower spatial displacement in MPI, respectively.Fig. 5Magnetic particle imaging of size-isolated SPION. **a** Key MPI parameters including normalized signal-to-noise ratios (SNR) of the samples at the 4th harmonic of the MPI drive field as well as full width at half maximum (FWHM) measurements and hysteresis loss analyses of the samples were obtained using magnetic particle spectroscopy (MPS; which is comparable to a zero-dimensional MPI acquisition without the superimposed gradient field measurements). **b** MPI images reconstructed based on "E" shaped phantoms filled with the crude sample, C2 and Resovist®. **c** The intensity line profiles of the red marked lines through the phantoms in **b** are shown. The line profiles show the voxel intensity along the marked line and demonstrate a doubling of signal intensity for C2 in comparison to Resovist® To exemplify the MPI imaging capabilities of our size-isolated SPION, we fabricated two phantoms. One was an E-shaped phantom (Fig. [5](#Fig5){ref-type="fig"}b), serving as a somewhat more complex structure, made up of single tracer-filled dots of 0.5 mm. The other phantom was V-shaped (Additional file [1](#MOESM1){ref-type="media"}: Figure S12a), and consisted of single dots with a diameter of 0.5 mm with an increasing distance between them (2, 3, 4, 5 and 6 mm). Both phantoms were filled with the crude starting mixture, with the C2 sample and with Resovist®, making sure that the iron concentrations were identical. Figure [5](#Fig5){ref-type="fig"}c and Additional file [1](#MOESM1){ref-type="media"}: Figure S12b show the line profiles of the voxel intensities along the red marked lines for the E and V phantoms, respectively. It can be seen that the lowest and the highest intensities are obtained with the crude and the C2 sample, respectively. The C2 sample produced signal intensities more than two times higher than those of Resovist®. From the MPI parameter analysis as well as from the MPI phantom experiments it can, therefore, be concluded that the C2 (and to a lesser extent also the C3) formulation is a useful alternative for Resovist® and suitable contrast agent for MPI. Magnetic fluid hyperthermia {#Sec10} --------------------------- Hyperthermia is a treatment modality in which cancerous tissue is exposed to a supernormal temperature. Cancer cells die as soon as temperatures exceed 42 °C, while normal cells can survive under these conditions \[[@CR37]\]. Hyperthermia can be generated using radiofrequency, ultrasound and microwave energy, as well as using magnetic fluid hyperthermia (MFH). In MFH, increased temperatures are created by applying a sinusoidally alternating magnetic field (AMF). When SPION are exposed to an AMF, heat is generated to release the magnetic energy consumed for the alignment of the magnetization of the magnetic particles in the direction of the applied magnetic field. In principle, three mechanisms are responsible for heat dissipation, which can act separately or simultaneously, depending on the nanoparticle properties: (1) hysteresis power loss, originating from the irreversibility of the magnetization process, (2) Néel relaxation, conditioned by the rotation of the magnetic moments of the particles, and (3) friction losses due to Brownian rotation of the magnetic particles as a whole. As a result of these three mechanisms, SPION and magnetic temperature gradually increase in an AFM until a saturation temperature is achieved \[[@CR37], [@CR38]\]. In a cellular environment, however, SPION are immobilized inside lysosomes and form agglomerates \[[@CR39], [@CR40]\]. This leads to partial blocking of the above-mentioned Brownian relaxation and to a drop in heating efficiency. In consequence, depending on the mechanism responsible for heat generation for a specific nanoparticle type, the in vivo hyperthermia performance could significantly decrease \[[@CR30]\]. Figure [6](#Fig6){ref-type="fig"}a depicts the time--temperature curves for the monodisperse SPION batches C1-C5, as well as for the crude sample C, Resovist® and Sinerem® in a low-frequency AMF. The iron concentration of all samples was 9 mM and the dispersant media was DI water. For all size-isolated samples except for C1, the required time for increasing the temperature from 37 to 42 °C (t~H~) was lower than for Resovist® and Sinerem®. In this context, a shorter t~H~ time reflects a better heating performance and contributes to shorter AMF application times in hyperthermia-based cancer treatment. The shortest t~H~ value was achieved using C3, having a core size of 13 nm. For this sample, the time to increase the temperature from 37 to 42 °C was 128 s, which was approximately 3 times faster than for Resovist® (t~H~ = 374 s).Fig. 6Magnetic fluid hyperthermia using size-isolated SPION. **a** Time--temperature curves obtained upon exposing the crude, C1--C5, Resovist® and Sinerem® samples to an alternating magnetic field (AMF). The frequency and amplitude of the AMF were 186 kHz and 46 kA m^−1^, respectively. The iron concentration was 9 mM for all samples. A Box-Lucas curve was fitted to each data set. **b** Difference between initial and maximum temperatures after 30 min of field exposure (ΔT~rise~). **c** Specific absorption rate values (SAR; calculated on the basis of Additional file [1](#MOESM1){ref-type="media"}: Equations S5, S9). Values represent average ± standard deviation of three separate experiments In addition to t~H~, the specific absorption rate (SAR) is an important quantitative parameter to determine the suitability of SPION formulations for MFH. From Additional file [1](#MOESM1){ref-type="media"}: Equations S7 and S8, it can be deduced that the SAR is directly proportional to ΔT~rise~ which is defined as the difference between the maximum temperature reached during AMF exposure and the initial temperature (in this specific case 37 °C). Comparing the ΔT~rise~ and the SAR values of the different formulations shows that the samples with a higher ΔT~rise~ have a higher SAR and consequently a better MFH performance (Fig. [6](#Fig6){ref-type="fig"}b, c). For the C3 sample, the SAR was approximately 2.5 times higher than for Resovist®. This indicates that the magnetic power absorbed per unit mass of the C3 sample in the presence of an AMF is \~ 2.5 times higher than that of Resovist®. This high SAR value is expected to be due to a high saturation magnetization arising from individual magnetic anisotropy. Higher SAR values are beneficial from a clinical point of view, as they allow for lower SPION dosing to achieve similar hyperthermia efficacy. A wide range of SAR values have been reported in the literature for diverse colloidal SPION formulations. SAR values strongly depended on the mean size and monodispersity of SPION, structural and magnetic properties, and the frequency and amplitude of the magnetic field. In the majority of cases, SAR values in the range between 4 and 100 W/g were achieved for commercially available SPION dispersions \[[@CR41]\]. For some customized formulations, higher SAR values have been reported. Bakoglidis and colleagues, for instance, synthesized spherical oleic acid-coated SPION with core sizes between 5 and 18 nm by thermal decomposition, and subjected them to MFH, showing maximal performance for 10 nm, with a SAR of 230 W/g. They used hexane as the dispersion medium to maintain a stable suspension of the nanoparticles \[[@CR42]\]. For the size-isolated C3 sample, we observed an SAR of 350 W/g, which exceeds this previously reported value by more than 50%. This notion indicates that upon simple and straightforward size-isolation via sequential centrifugation, SPION formulations with optimal performance for biomedical applications can be readily obtained. Conclusion {#Sec11} ========== We here present a centrifugation protocol to obtain SPION with well-defined sizes (hydrodynamic diameter: 26.3 ± 1.2, 49.4 ± 1.1, 64.8 ± 2.1, 82.1 ± 2.3 and 114.6 ± 4.4 nm; and core size: 7.7 ± 1.6, 10.6 ± 1.8, 13.1 ± 2.2, 15.6 ± 2.8 and 17.2 ± 2.1 nm) and with a very narrow size distribution (PDI below 0.1) from a polydisperse starting mixture prepared via the co-precipitation technique. The samples obtained upon the 2nd and 3rd round of centrifugation, which had a core size of 10.6 ± 1.8 and 13.1 ± 2.2 nm, and a hydrodynamic diameter of 49.4 ± 1.1 and 64.8 ± 2.1 nm, were found to be optimal for MRI, MPI and MFH application, with an up to 3.3-, 3.3- and 7-fold improved performance as compared to the crude starting mixture, Resovist® and Sinerem®, respectively. Our results demonstrate that simple and straightforward size-isolation helps to improve the performance for biomedical application. Experimental {#Sec12} ============ SPION synthesis {#Sec13} --------------- Eight mmol of ferric chloride was dissolved in DI water and mixed for 5 min under mechanical stirring. Subsequently, 4 mmol of ferrous chloride tetrahydrate was added to the solution and mixed for a further 5 min at room temperature. The pH of the solution was adjusted to 11.0 by adding of 1 M aqueous ammonia solution drop-wisely and it was stirred at 25 °C for 30 min under nitrogen atmospher. The formed black-colored iron oxide particles were decanted using a permanent magnet and washed at least three times with DI water. Afterwards, a specific amount of 0.1 M hydrochloric acid was added to the particles and sonicated for 10 min. Following, the citrate solution was added to the mixture and was stirred at 80 °C for 2 h. The citrate-coated polydisperse particles were separated by the use of a permanent magnet and then resuspended in DI water. Finally, the suspension was passed through a 0.2 µm filter to remove the big particles. Additional synthetic details are provided in Additional file [1](#MOESM1){ref-type="media"}. SPION characterization {#Sec14} ---------------------- The prepared SPION were subjected to several systematic analyses, to assess their properties and performance. The particle size and the size distribution of the crude sample, of the C1--C5 subfractions, and of Resovist® and Sinerem® were investigated by multiple different sizing techniques, including dynamic light scattering (DLS), nanoparticle tracking analysis (NTA) and transmission electron microscopy (TEM). The zeta potential values of the nanoparticles in aqueous solution were measured using a Zetasizer Nano-ZS (Malvern Instruments, Malvern, UK). The iron concentration of the respective samples was measured using the 1,10-phenanthroline assay \[[@CR43]\]. We also evaluated the cytotoxicity of the samples. This was done via 2,3-bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxanilide (XTT), lactate dehydrogenase (LDH) and reactive oxygen species (ROS) assays at multiple different iron concentrations, ranging from 0.1 to 10 mM. The colloidal stability of all size-isolated samples was investigated in two physiologically relevant media. These were fetal bovine serum (FBS), which is the most widely used serum-supplement for the in vitro cell culture, and bovine serum albumin (BSA). Colloidal stability was analyzed upon incubation in FBS and BSA for 2, 6 and 24 h, via visual inspection and DLS analysis. Measurements of magnetic properties, including field-dependent magnetization, saturation magnetization (M~s~) and field-cooled (FC) magnetization, were performed using a Quantum MPMS-5XL SQUID magnetometer. Additional characterization details are provided in Additional file [1](#MOESM1){ref-type="media"}. SPION application {#Sec15} ----------------- MRI experiments were performed on a 3T clinical MR scanner (Philips Achieva, Best, The Netherlands) and images were acquired using SENSE-flex-M coil (Philips Achieva, Best, The Netherlands). From MRI tests, R~1~ and R~2~ relaxation rates and corresponding r~1~ and r~2~ relaxivities were calculated \[[@CR44]\]. MPI measurements were performed using the Philips pre-clinical demonstrator system and relevant parameters of the SPION were determined including the signal-to-noise ratio (SNR) and the full width at half maximum (FWHM) of the point spread function (PSF). In order to evaluate hyperthermia performance, a custom-build setup (Trumpf Hüttinger, Freiburg, Germany) was employed and the heating efficiency of the different SPION formulations under an alternating magnetic field (AMF) was quantified using the specific absorption rate (SAR), which provides a measure of the magnetic power absorbed per unit mass of the magnetic material (see Additional file [1](#MOESM1){ref-type="media"} for more details). Supplementary information ========================= {#Sec19} **Additional file 1: Figure S1.** Zeta potential analysis of the crude, C1-C5, Resovist® and Sinerem® samples. **Figure S2.** Cell viability of NIH3T3 cells treated with the samples with various concentrations ofSPION for 4 h according to XTT assay. The data were normalized to control value (SPION-freemedia), which was set as 100% cell viability. Experiments were performed at different concentrationsof SPION in the range of 0.1 to 10.0 mM. Values represent means ± standard deviations of fiveidentical experiments made in three replicates. **Figure S3.** LDH leakage of NIH3T3 cells treated with the samples with various concentrations ofSPION for 4 h according to the manufacturer's instructions. Experiments were done at differentconcentrations of SPION in the range of 0.1 to 10.0 mM. Values represent mean ± standard deviationof five identical experiments made in three replicates. **Figure S4.** ROS generated in NIH3T3 cells incubated with the samples with various concentrations ofSPION to the control cells (SPION-free media) after 24 h treatment. Experiments were done atdifferent concentrations of SPION in the range of 0.1 to 5 mM. Data represent mean ± standarddeviation of three identical experiments made in three replicates. **Figure S5.** Colloidal stability of the samples in undiluted FBS monitored by visual inspection andDLS. Visual inspection indicated no aggregation up until 24 h. In line with this, size and PDI obtainedby DLS also showed no significant changes at 24 h. The iron concentration for all the samples was 5mM. The FBS size according to DLS was 19.7±1.5 nm which is very close to hydrodynamic diameterof C1. Also, FBS is polydisperse and has PDI of 0.49±0.05. These two notions explain the high PDI forC1 in FBS. **Figure S6.** Colloidal stability of the samples in 4 wt% BSA in DI water. Visual inspection showed noaggregation at 24 h. Also, size and PDI obtained by DLS showed no important differences in theirvalues at 24 h. The iron concentration for all the samples was 5 mM. **Figure S7.** Temperature-dependent magnetization at 1000 Oe of the crude SPION mixture (C), thesize-isolated samples C1-C5, Resovist® (R) and Sinerem® (S). Results were normalized to Fe content. **Figure S8.** T1- and T2-weighted MR images of the crude, C1-C5, Resovist® and Sinerem® samples atdifferent concentrations from 0.005 to 0.05 mM. **Figure S9.** Longitudinal (1/T1; a) and transverse (1/T2; b) relaxation rates of the crude, C1-C5,Resovist® and Sinerem® samples as a function of concentration of Fe. The straight lines represent thelinear fit to the experimental data. The relaxivities r1 and r2 were calculated as the slope of the linesfitted to the experimental data. Values represent average of one experiment made in three replicates. **Figure S10.** Relaxivity ratios (r~2~/r~1~) for the crude, C1-C5, Resovist® and Sinerem® samples. **Figure S11.** Normalized SNR values of the samples from the 4th up to the 10th harmonic of the MPIdrive field. **Figure S12.** Magnetic particle imaging of size-isolated SPION. (a) MPI images reconstructed basedon "V" shaped phantoms filled with the crude sample, C2 and Resovist®. (b) The intensity line profilesof the red marked lines through the phantoms in panel (a) are shown. The line profiles show the voxelintensity along the marked line and demonstrate a doubling of signal intensity for C2 in comparison toResovist®. **Table S1.** Overview of the results obtained in the size analyses performed using TEM, DLS and NTA.The different SPION formulations were evaluated in different media and upon different storage times. **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s12951-020-0580-1. The authors acknowledge technical support by Marek Weiler. SMD, FK and TL conceived the project. SMD, DC, MD, KR, PV, MS, NG, UE, MB and JL performed the experiments. JM, SB, IS, PK, BHS, VS, FK and TL supervised data collection and interpretation. SMD drafted the manuscript. All authors revised the manuscript. All authors read and approved the final manuscript. The authors gratefully acknowledge financial support by the European Research Council (ERC: Starting Grant 309495 (NeoNaNo) and Proof-of-Concept Grants 680882 (CONQUEST) and 813086 (PIcelles), by the German Research Foundation (DFG: GRK2375 (grant 331065168), SFB/TRR57, SFB1066), and by RWTH Aachen University (ERS Prep Fund PFLS009). Data and materials will be made available upon request. We ensure adherence to relevant ethical standards. We agree to the publication policies of Journal of Nanobiotechnology. The authors declare that they have no competing interests.

(Current Biology *24*, 2018--2024; September 8, 2014) A reader has brought to our attention a labeling error in Figure 2 of this manuscript as originally published online and in print: the two arrows pointing upward under panels D4 and D5 should have been labeled "touch." This error has now been corrected in Figure 2 of the article online. The authors apologize for any confusion this error may have caused.

1. Introduction {#sec1-nanomaterials-08-00083} =============== Cancer is one of the serious concerns around the world and it is one of the main causes of death worldwide \[[@B1-nanomaterials-08-00083]\]. In a press release by the World Health Organization in 2014, it was reported that cancer accounted for 8.2 million deaths worldwide in 2012 with lung, breast and colorectal cancers identified as most common occurrences worldwide in the year 2012 \[[@B2-nanomaterials-08-00083]\]. This was further supported by a press release by the World Health Organization in 2015. According to the Cancer statistics 2017, in the United States alone, about 1,688,780 new cancer cases and 600,920 cancer deaths are projected to occur \[[@B3-nanomaterials-08-00083]\]. Current cancer treatments rely on radiation and chemotherapeutic agents that work by killing rapidly dividing cells in the body. The main drawback of conventional chemotherapy is the adverse effects on the body as it cannot deliver selective action specifically to the cancer cells, thus damaging the surrounding normal healthy cells or rapidly dividing healthy cells such as the cells of gastrointestinal tract, bone marrow, hair follicles, causing issues like cardiac, hepatic, pulmonary, renal and gastrointestinal toxicities \[[@B4-nanomaterials-08-00083],[@B5-nanomaterials-08-00083],[@B6-nanomaterials-08-00083]\]. Drug delivery systems offer numerous advantages over tradition chemotherapy such as targeted delivery at disease site, sustained release leading to prolonged bioavailability, lower dosage requirement and improved drug solubility among others \[[@B7-nanomaterials-08-00083],[@B8-nanomaterials-08-00083],[@B9-nanomaterials-08-00083],[@B10-nanomaterials-08-00083],[@B11-nanomaterials-08-00083],[@B12-nanomaterials-08-00083]\]. Significant impact has been made by the application of nanotechnology in medicine for theranostic agents development which can diagnose and cure the diseases simultaneously \[[@B13-nanomaterials-08-00083],[@B14-nanomaterials-08-00083]\]. Variety of nanocarriers have been designed and successfully applied for the delivery of the therapeutic agents such as graphene oxide, polymers-based delivery systems, layered double hydroxides, gold nanoparticles, multifunctional nanoparticles and iron oxide magnetite nanoparticles \[[@B8-nanomaterials-08-00083],[@B15-nanomaterials-08-00083],[@B16-nanomaterials-08-00083],[@B17-nanomaterials-08-00083],[@B18-nanomaterials-08-00083],[@B19-nanomaterials-08-00083],[@B20-nanomaterials-08-00083],[@B21-nanomaterials-08-00083]\]. Polymer coated iron oxide magnetite nanoparticles (Fe~3~O~4~) have received much of the attention as the novel cancer therapeutics vectors because of their unique properties such as ease of preparation, easily scalable production, sustained release properties, high encapsulation capacity, biocompatibility with normal cells and tissues, easier surface modification and stable magnetic nature \[[@B9-nanomaterials-08-00083],[@B22-nanomaterials-08-00083],[@B23-nanomaterials-08-00083],[@B24-nanomaterials-08-00083],[@B25-nanomaterials-08-00083],[@B26-nanomaterials-08-00083]\]. All of these characteristics make iron oxide magnetite nanoparticles (Fe~3~O~4~) an ideal candidate for cancer delivery vectors. However, the large surface area to volume ratio and the dipole-dipole attraction causes the agglomeration of nanoparticles, hence the need for surface polymer modification. The magnetic nanocarriers need to be stable in normal saline and water at neutral pH for biological, medical diagnostic and therapeutic applications \[[@B23-nanomaterials-08-00083],[@B27-nanomaterials-08-00083]\]. To avoid agglomeration, the surface of iron oxide magnetite nanoparticles is coated with polymer which also helps in sustained release and better stability in physiological conditions. The polymer poly (ethylene-glycol) (PEG) has been widely applied in drug delivery and is being utilized as protective layer for the nanoparticles. The monomer unit of PEG contains both polar oxygen and two methylene group which are non-polar. This dual polarity makes PEG to be soluble in variety of polar and non-polar solvents and has been widely used to improve aqueous solubility of hydrophobic drugs \[[@B28-nanomaterials-08-00083],[@B29-nanomaterials-08-00083]\]. Gallic acid (3,4,5-trihydroxybenzoic acid) is a bioactive compound found in plants and foods such as white tea, witch hazel and it has been reported to possess antioxidant, anti-inflammatory, anticancer properties and is also known for its protective activity on normal cells which makes them pivotal for cancer therapy \[[@B30-nanomaterials-08-00083]\]. In this study we have redesigned an anticancer nanocomposite formulation of Gallic acid loaded on iron oxide magnetite nanoparticles coated with polyethylene glycol (Fe~3~O~4~-PEG-GA) with improved drug loading and better sustained release properties and was characterized by X-ray diffraction (XRD), dynamic light scattering (DLS), in vitro cytotoxicity assay and drug loading quantification. In previous study we tested Gallic acid nanocomposite formulation \[(P-Fe~3~O~4~-PEG-GA) (in formula P stands for previous)\] against MCF-7, a breast cancer cell line with the IC~50~ value of 11.61 ± 0.12 µg/mL and human normal lung fibroblast cells MRC-5 was used as a model for normal cell in which more than 80% cell viability was observed after 72 h incubation \[[@B31-nanomaterials-08-00083]\]. In this study we tested the free drug GA, empty nanocarrier and the anticancer nanocomposite formulation Fe~3~O~4~-PEG-GA against A549 human lung carcinoma cells, *HT29* human colon adenocarcinoma cell line, repeated on MCF-7 breast cancer cells and normal 3T3 cells for incubation period of 24, 48 and 72 h. 2. Results {#sec2-nanomaterials-08-00083} ========== 2.1. Physicochemical Characterization {#sec2dot1-nanomaterials-08-00083} ------------------------------------- ### 2.1.1. X-ray Diffraction (XRD) Analysis {#sec2dot1dot1-nanomaterials-08-00083} [Figure 1](#nanomaterials-08-00083-f001){ref-type="fig"}a shows the XRD patterns of iron oxide magnetite nanoparticles (Fe~3~O~4~) alone, poly ethylene glycol (PEG) and anticancer nanocomposite Fe~3~O~4~-PEG-GA. Iron oxide magnetite nanoparticles (Fe~3~O~4~) showed the six characteristics peaks ascribed to Brag reflections due to (220), (311), (400), (422), (511), and (440) and these peaks can be observed at 2*θ* = 30.16°, 35.95°, 43.34°, 54.17°, 57.27° and 62.98° respectively. The pure polymer PEG showed two main characteristics high intensity peaks at about 2*θ* = 19.3° and 23.5°. The pure gallic acid (GA) has been reported to show many peaks between the 2*θ* of 10--50° as reported previously \[[@B31-nanomaterials-08-00083],[@B32-nanomaterials-08-00083]\]. In the XRD patterns of the nanocomposite Fe~3~O~4~-PEG-GA characteristic peaks of iron oxide magnetite nanoparticles (Fe~3~O~4~), PEG and of the pure drug peaks are present with slight lesser intensity. The presence of characteristic peaks of Fe~3~O~4~, PEG and GA in the final anticancer nanocomposite confirms the successful formation of the nanocomposite Fe~3~O~4~-PEG-GA. ### 2.1.2. Dynamic Light Scattering (DLS) {#sec2dot1dot2-nanomaterials-08-00083} The size of the anticancer nanocomposite Fe~3~O~4~-PEG-GA was determined using Zetasizer by dynamic light scattering (DLS). The sample was dispersed in water and sonicated for 15 min and then analyzed with Zetasizer. The sample was found to have narrow size distribution between 5 and 12 nm with average particle size of 10 nm as shown in [Figure 1](#nanomaterials-08-00083-f001){ref-type="fig"}b. This size distribution is much smaller compared to previously reported (P-Fe~3~O~4~-PEG-GA) which had wide distribution 20--50 nm with average size of 31.44 nm \[[@B31-nanomaterials-08-00083]\]. ### 2.1.3. In vitro Release Studies {#sec2dot1dot3-nanomaterials-08-00083} Release behavior of GA from the nanocomposite Fe~3~O~4~-PEG-GA was conducted in human body simulated buffer saline (PBS) solution of pH 7.4 (human blood pH) and in pH 4.8 (intracellular lysosomal pH) as shown in [Figure 1](#nanomaterials-08-00083-f001){ref-type="fig"}c. For the release studies 10 mg of the nanocomposite was put in 10 mL PBS solution of pH 7.4 and pH 4.8 in thermostat at 37 °C with constant shaking. At different time points 3 mL aliquot was taken out and replaced with new buffer of either solution of pH 7.4 and pH 4.8 respectively and analyzed for the percentage release using UV-Vis spectrophotometer (Waltham, MA, USA). The release of GA was found to be sustained in both physiological pHs (7.4 and 4.8) and took 200 h (about 8 days) for the complete release in both conditions as shown in [Figure 1](#nanomaterials-08-00083-f001){ref-type="fig"}c. The release profile of GA from the nanocomposite Fe~3~O~4~-PEG-GA is much more sustained compared to previously designed nanocomposite P-Fe~3~O~4~-PEG-GA which initially showed burst release with more than 40% drug released in less than 2 h \[[@B31-nanomaterials-08-00083]\]. Moreover, the release profile of free drug GA has been reported to be extremely fast which took less than 2 min for the complete release \[[@B31-nanomaterials-08-00083]\]. This suggests that nanocomposite designed (Fe~3~O~4~-PEG-GA) has much better sustain release profile compared to free drug and the previously designed nanocomposite (P-Fe~3~O~4~-PEG-GA). ### 2.1.4. Drug Loading Percentage Quantification {#sec2dot1dot4-nanomaterials-08-00083} Percentage loading of the GA in the nanocomposite Fe~3~O~4~-PEG-GA was determined with (UV-Vis) spectrophotomer (Waltham, MA, USA) GA acid was extracted (deloaded) from the nanocomposite Fe~3~O~4~-PEG-GA by putting 10 mg of it in 50 mL of 1 molar phosphate buffer saline (PBS) solution, followed by sonication for an hour and kept in thermostat with constant shaking at 37 °C for 10 days. After that sample was filtered and filtrate was subjected to quantification of GA. For quantification different concentrations of GA standards were prepared e.g., 25, 50, 100, 150 and 200 ppm and were analyzed and correlation coefficient (*r*^2^) was found to be 0.9918. After that different filtrate was analyzed three times and loading was found to be 35%. The percentage loading of GA in this redesigned nanocomposite is much higher (35%) compared to our previously reported designed nanocomposite of Fe~3~O~4~-PEG-GA in which loading percentage of GA was found to be very lower i.e., 7%. The improved loading can be ascribed to solvent used to dissolve GA i.e., (80% Methanol:20% water). ### 2.1.5. Cytotoxicity on 3T3 Fibroblast Cells {#sec2dot1dot5-nanomaterials-08-00083} All the samples e.g., designed anticancer nanocomposite (Fe~3~O~4~-PEG-GA), empty nanocarrier (Fe~3~O~4~-PEG) and free dug GA was tested against 3T3 fibroblast cells for cytotoxicity evaluation using their gradient concentrations i.e., (0.47, 0.94, 1.88, 3.75, 7.5, 15 and 30 µg/mL) and were incubation for 24, 48 and 72 h as shown in [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}a--c respectively. Cell viability studies revealed that free drug GA, empty carrier and the designed nanocomposite (Fe~3~O~4~-PEG-GA) were found to be biocompatible with 3T3 cell as the percentage cell viability was found to be about 80% even after 24, 48 and 72 h incubation at maximum concentration of 30 μg/mL as shown in [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}a--c. The MTT results showed that neither GA, Fe~3~O~4~-PEG nor Fe~3~O~4~-PEG-GA caused toxicity to 3T3 cells at all time points. As we previously reported nanocomposite (P-Fe~3~O~4~-PEG-GA) was also found to be biocompatible with MRC-5 human normal lungs cells with percentage cell viability of 100% after 72 h incubation \[[@B31-nanomaterials-08-00083]\]. These results indicate the high biocompatibility of all the samples. ### 2.1.6. Anticancer Assays {#sec2dot1dot6-nanomaterials-08-00083} The free drug GA, empty nanocarrier (Fe~3~O~4~-PEG) and designed magnetite nanocomposite formulation (Fe~3~O~4~-PEG-GA) were tested against different cancer cell line namely lung cancer cell (A549), breast cancer cell (MCF-7) and colon cancer cell (HT-29) lines to determine their anticancer properties. In our previous studies we had tested the nanocomposite (P-Fe~3~O~4~-PEG-GA) against breast cancer cell (MCF-7) in this study we repeated the assay on breast cancer cell (MCF-7) as the percentage drug loading of GA is different than previously reported (P-Fe~3~O~4~-PEG-GA). Doxorubicin, a chemotherapeutic drug, has been studied extensively and its IC~50~ values towards the same cancer cell lines have been reported to be 0.33 ± 0.03, 0.05 ± 0.01 and 0.58 ± 0.01 µg/mL for HT-29, MCF-7 and A549 respectively \[[@B33-nanomaterials-08-00083]\].These IC~50~ values of Doxorubicin were used as a reference for the positive control in this study. ### 2.1.7. Anticancer Activity against Lung Cancer Cell (A549) {#sec2dot1dot7-nanomaterials-08-00083} [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}d--f shows the percentage cell viability of A549 lung cancer cells treated with higher concentrations i.e., 6.25, 12.5, 25, 50, 100 and 200 μg/mL of the free drug GA, empty nanocarrier Fe~3~O~4~-PEG and the nanocomposite Fe~3~O~4~-PEG-GA. Previous anticancer studies involving A549 used a higher range of concentration of GA as well, ranging from a minimum of 10 µg/mL up to 500 µg/mL \[[@B34-nanomaterials-08-00083],[@B35-nanomaterials-08-00083],[@B36-nanomaterials-08-00083]\]. All the samples with above concentrations were incubated for 24, 48 and 72 h with the A549 lung cancer cells as shown in [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}d--f respectively. The designed anticancer nanocomposite Fe~3~O~4~-PEG-GA showed better anticancer activity (IC~50~ 37.49 µg/mL) compared to free drug GA (IC~50~ 56.49 µg/mL). The effective IC~50~ concentration (i.e., actual amount of GA based on percentage drug loading) present in the nanocomposite 13.12 µg/mL is much lower than IC~50~ of whole nanocomposite 37.4 µg/mL. So in reality, effective IC~50~ of nanocomposite 13.121 µg/mL compared to IC~50~ free GA (i.e., 56.49 µg/mL) much lower against lung cancer cells A549. [Table 1](#nanomaterials-08-00083-t001){ref-type="table"} shows the IC~50~ of free drug GA, and the nanocomposite Fe~3~O~4~-PEG-GA compared to Doxorubicin. ### 2.1.8. Anticancer Activity against Breast Cancer Cell MCF-7 Cells {#sec2dot1dot8-nanomaterials-08-00083} All the samples i.e., free drug GA, empty nanocarrier (Fe~3~O~4~-PEG) and anticancer nanocomposite (Fe~3~O~4~-PEG-GA) were tested against breast cancer cell MCF-7 cells to screen their anticancer activity. Different gradient concentrations i.e., 0.47, 0.94, 1.88, 3.75, 7.5, 15 and 30 μg/mL of all the samples were incubated for 24, 48 and 72 h with breast cancer cell MCF-7 cells and results are shown in [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}g--i. A range of cytotoxicity in a time and dose-dependent manner was observed. The IC~50~ value of the nanocomposite Fe~3~O~4~-PEG-GA 7.28 μg/mL and its effective IC~50~ 2.548 μg/mL are much lower than free drug GA IC~50~ 21.35. The IC~50~ of the this redesigned nanocomposite Fe~3~O~4~-PEG-GA nanocomposite in this in vitro study against breast cancer cell MCF-7 cells is much lower than our previously reported nanocomposite P-Fe~3~O~4~-PEG-GA \[[@B31-nanomaterials-08-00083]\]. The better anticancer activity of nanocomposite can be attributed to the nanosize and sustained release properties of the anticancer nanocomposite. [Table 1](#nanomaterials-08-00083-t001){ref-type="table"} shows the IC~50~ vales of the nanocomposite and free drug GA against Doxorubicin. ### 2.1.9. Anticancer Activity against Colon Cancer Cell (HT-29) {#sec2dot1dot9-nanomaterials-08-00083} The free drug GA, empty nanocarrier and the nanocomposite Fe~3~O~4~-PEG-GA were tested against colon cancer cell (HT-29) using different concentrations i.e., 0.47, 0.94, 1.88, 3.75, 7.5, 15 and 30 μg/mL and incubated for 24, 48 and 72 h as shown in [Figure 2](#nanomaterials-08-00083-f002){ref-type="fig"}j--l respectively. The IC~50~ of free drug GA was found to be 15 μg/mL. The IC~50~ of the nanocomposite Fe~3~O~4~-PEG-GA was found 4.85 μg/mL and the effective IC~50~ of the nanocomposite Fe~3~O~4~-PEG-GA in 4.85 μg/mL was calculated to be 1.697 μg/mL based on 35% GA loading. The IC~50~ values are given in [Table 1](#nanomaterials-08-00083-t001){ref-type="table"}. 3. Discussion {#sec3-nanomaterials-08-00083} ============= In this study we have redesigned our previously reported an anticancer nanocomposite formulation of Gallic acid loaded on iron oxide magnetite nanoparticles coated with polyethylene glycol (Fe~3~O~4~-PEG-GA). The XRD patterns of the nanocomposite Fe~3~O~4~-PEG-GA showed the characteristic peaks of iron oxide magnetite nanoparticles (Fe~3~O~4~), PEG and of the pure drug peaks are present with slight lesser intensity. The presence of characteristic peaks of Fe~3~O~4~, PEG and GA in the final anticancer nanocomposite confirms the successful formation of the nanocomposite Fe~3~O~4~-PEG-GA. The average particle size of the anticancer nanocomposite was found to be 10 nm with a narrow size distribution of 5--12 nm compared to previously reported designed (P-Fe~3~O~4~-PEG-GA) which had wide distribution 20--50 nm with average size of 31.44 nm \[[@B31-nanomaterials-08-00083]\]. The Percentage GA loading is found to be 35% in this redesigned nanocomposite is much higher compared to our previously reported nanocomposite of (Fe~3~O~4~-PEG-GA) with 7%. In addition to this in vitro release of GA from the nanocomposite was found to be highly sustained which took about 200 h (about 8 days) in human body simulated phosphate buffer saline (PBS) solution of pH 7.4 (blood pH) and pH 4.8 (intracellular lysosomal pH) at human body temperature 37 °C. In this study we tested the free drug GA, empty nanocarrier and the anticancer nanocomposite formulation Fe~3~O~4~-PEG-GA against A549 human lung carcinoma cells, *HT29* human colon adenocarcinoma cell line, repeated on MCF-7 breast cancer cells and with normal 3T3 cells for incubation period of 24, 48 and 72 h. The IC~50~ of the designed nanocomposite Fe~3~O~4~-PEG-GA are found to be much lower compared to free drug GA. The effective IC~50~ which is based on percentage drug (GA) loading in nanocomposite Fe~3~O~4~-PEG-GA which is even further lower. [Table 1](#nanomaterials-08-00083-t001){ref-type="table"} shows the details of the IC~50~ against A549 human lung carcinoma cells, *HT29* human colon adenocarcinoma cell line, and MCF-7 breast cancer cells and with normal 3T3 cells. 4. Materials and Methods {#sec4-nanomaterials-08-00083} ======================== 4.1. Chemicals {#sec4dot1-nanomaterials-08-00083} -------------- Gallic acid of 97% purity, iron oxide coated with polyethylene glycol (PEG) nanocarrier (Fe~3~O~4~-PEG) and gallic acid-iron oxide coated with PEG nanocomposite (Fe~3~O~4~-PEG-GA) were provided by the Material Synthesis and Characterization Laboratory, Institute of Advance Technology University Putra Malaysia (Serdang Selangor Malaysia). All three drugs were used for the preliminary screening of the effectiveness between gallic acid nanocomposite and pure gallic acid against normal cell and three different types of cancer cell lines. To prepare the stock solution, 5 mg of each drug was initially dissolved in 200 µL of dimethyl sulfoxide (DMSO) before the mixture was vortexed and sonicated for at least 30 min to ensure that the drug was completely dissolved. Upon sonication, 800 µL of RPMI 1640 (Nacalai Tesque, Kyoto, Japan) was added and then vortexed for 2 min to make the total volume of 1 mL. The drug sub stock was further diluted to a series of concentrations (0.47--200 µg/mL) and was used on the same day it was prepared. MTT \[3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide\] and Trypsin-EDTA (Ethylene diamine tetraacetate) (0.25%) were purchased from Nacalai Tesque (Kyoto, Japan). 4.2. Cell Lines {#sec4dot2-nanomaterials-08-00083} --------------- Normal human fibroblast cells (3T3), human lung cancer epithelial cells (A549), human breast cancer epithelial cells (MCF-7) and human colon cancer epithelial cells (HT-29) were obtained from the American Tissue Culture Collection (Manassas, VA, USA). The cells were maintained in Roswell Park Memorial Institute (RPMI) 1640 medium (Nacalai Tesque, Kyoto, Japan). The growth medium was supplemented with 10% fetal bovine serum, [l]{.smallcaps}-glutamine 15 mmol/L, penicillin 100 U/mL and streptomycin 100 µg/mL. All cells were incubated at 37 °C in humidified 5% CO~2~/95% air and the media was replaced every 2 to 3 days. 4.3. In Vitro Cell Viability Assay {#sec4dot3-nanomaterials-08-00083} ---------------------------------- The tetrazolium-based Colorimetric Assay (MTT) was used to determine the cytotoxicity of the drugs on the cancer and normal cell lines. In the case of viable cells in the in vitro MTT assay, the (nicotinamide adenine dinucleotide phosphate) NADPH-dependent oxidoreductase enzyme that is produced by the healthy mitochondria of a living cell will be secreted outside the cell membrane and converts the tetrazolium dye into a purple colored compound called formazan. Cells were first maintained in drug free media until the cells reached 80--90% confluency. MTT assay was initiated by the seeding process of plating 1 × 10^5^ cells into each well of a flat bottomed 96-well plate making the total volume of media in each well 100 µL. The cells were then left to attach for 24 h. Cells were then treated with increasing concentrations of pure gallic acid (GA), iron oxide-PEG nanocarrier (Fe~3~O~4~-PEG) and gallic acid-iron oxide coated with PEG nanocomposite (Fe~3~O~4~-PEG-GA) to make the final volume of 200 µL per well and incubated in 5% CO~2~ at 37 °C for 24, 48 and 72 h. At the desired time point, 20 µL of MTT solution (5 mg/mL in PBS) was added to each well and kept in the incubator for 4 h before the cells were centrifuged to discard the supernatant. 100 µL of DMSO was later added and left in the dark for 30 min to dissolve the crystal formazan. Absorbance was measured at 570 and 630 nm was used to measure the background absorbance. Experiments were done in triplicates and cell viability was calculated based on the given formula. Cell Viability ( \% ) = { ( OD \* of Treated ) ÷ ( OD \* of Control ) × 100 } OD \* = Optical density 4.4. Preparation of Magnetite Nanoparticles (Fe~3~O~4~), Polyethylene Glycol (PEG) Coating and GA Loading {#sec4dot4-nanomaterials-08-00083} --------------------------------------------------------------------------------------------------------- Iron oxide magnetite nanoparticles (Fe~3~O~4~) were prepared by previously reported method, in brief in 80 mL deionized water containing 6 mL ammonia hydroxide (25%) a mixture of 2.43 g (FeCl~2~⋅4H~2~O) and 0.99 g (FeCl~3~⋅6H~2~O) was added. After adding the above material sample was subject to ultrasonication for 1 h. Sample was centrifuged and washed thoroughly with water and dried in oven at 80 °C for 24 h and ground to fine powder \[[@B15-nanomaterials-08-00083],[@B26-nanomaterials-08-00083],[@B31-nanomaterials-08-00083]\]. PEG coating was carried out by dissolving the 0.2 g of dried iron oxide magnetite nanoparticles (Fe~3~O~4~) in 1% of PEG solution followed by stirring for one hour after that sample was centrifuged and washed with ethanol thoroughly. GA was loaded on the designed nanocarrier Fe~3~O~4~-PEG by putting this sample to 50 mL (40 mL methanol and 10 mL water) of 1% GA and stirred for 24 h resulting in the formation of Fe~3~O~4~-PEG-GA. Next day sample was washed thoroughly with methanol and water by centrifugation. After that sample was dried in oven and ground to powder and subjected to further characterization. 5. Conclusions {#sec5-nanomaterials-08-00083} ============== The designed magnetite nanocomposite Fe~3~O~4~-PEG-GA was found to have higher drug GA loading 35% compared to previously reported nanocomposite with 7% GA loading. The in vitro release in human physiological pH 7.4 and pH 4.8 at 37 °C was conducted and was found to be highly sustained for up to 8 days compared to previously reported nanocomposite which took 3.5 days for the complete release and that was conducted at room temperature. In this study the magnetite nanocomposite Fe~3~O~4~-PEG-GA was found to be biocompatible with normal 3T3 cells. Both GA and Fe~3~O~4~-PEG-GA nanocomposite displayed time and dose-dependent anticancer activity against A549, MCF-7 and HT-29 cells Most importantly the IC~50~ of the magnetite nanocomposite Fe~3~O~4~-PEG-GA were found against these cell lines were much better than free drug GA. These in vitro study results are highly encouraging to go further animal studies. The study was supported by research grants from the Ministry of Higher Education, namely Fundamental Research Grant Scheme (FRGS), Universiti Putra Malaysia Project (GP/IPS/2015/5524841) and Putra Graduate Initiative (IPS) Putra Grant from the Universiti Putra Malaysia (Vot number: 9510100). We thank our colleagues from the Laboratory for Vaccine and Immunotherapeutics, Institute of Biosciences, Material Synthesis and Characterization Laboratory, Institute of Advanced Technology (ITMA), the Department of Human Anatomy Faculty of Medicine and Health Sciences University Putra Malaysia (UPM) and the Chemistry Department, University of Sheffield. Raihana Rosman, Bullo Saifullah, Sandra Maniam, Dena Dorniani, Sharida Fakurazi and Mohd Zobir Hussein conceived and designed the project. Raihana Rosman, Bullo Saifullah and Dena Dorniani conducted experimental work and wrote the manuscript. Sandra Maniam, Mohd Zobir Hussein and Sharida Fakurazi also contributed in writing up of the manuscript and the improvement of the manuscript. All other authors reviewed and contributed significantly to improve the manuscript for its final submission. The authors declare no conflict of interest. {#nanomaterials-08-00083-f001} ###### (**a**--**c**) shows the cell viability (%) of 3T3 cells estimated by MTT assay after 24, 48 and 72 h incubation respectively; (**d**--**f**) shows the cell viability (%) of A549 cells estimated by MTT assay after 24, 48 and 72 h incubation respectively; (**g**--**i**) shows the cell viability (%) of MCF-7 cells estimated by MTT assay after incubation for 24, 48 and 72 h respectively; (**j**--**l**) shows the cell viability (%) of HT-29 cells estimated by MTT assay after 24, 48 and 72 h incubation respectively.     nanomaterials-08-00083-t001_Table 1 ###### The IC~50~ values of GA, Doxorubicin and Fe~3~O~4~-PEG-GA on cancer cell lines. Cancer Cell Lines IC~50~ (μg/mL) \* Effective IC~50~ (μg/mL) \*\* ------------------- ------------------- ------------------------------- ------------- ------ HT29 14.52 ± 0.94 4.85 ± 0.33 0.33 ± 0.03 1.70 MCF-7 21.35 ± 4.14 7.28 ± 0.64 0.05 ± 0.01 2.55 A549 56.49 ± 4.31 37.49 ± 1.42 0.58 ± 0.01 13.1 **\*** Values are expressed as the mean ± standard deviation of 3 replicates. The IC~50~ value is defined as the concentration of drug needed for 50% cell inhibition; **\***\* Values of actual IC~50~ that were calculated based on 35% GA loading in the nanocomposite.

Background ========== Fusarium wilt of *Arabidopsis thaliana* is an ideal pathosystem for mapping, identifying and characterizing genes responsible for host resistance to vascular wilt fungi. *A. thaliana*, which is the preeminent subject of plant molecular genetic and genomic studies, is susceptible to infection by three phylogenetically-distinct pathogenic forms, or formae speciales, of the soil-borne fungus *Fusarium oxysporum*\[[@B1],[@B2]\]. In the field, *F. oxysporum* forma specialis *conglutinans* (FOC), *F. oxysporum* forma specialis *raphani* and *F. oxysporum* forma specialis *matthioli* (FOM) are isolated from diseased *Brassica* species, radish (*Raphanus sativus*) and garden stock (*Matthioli incana*), respectively \[[@B3]\]. Fusarium wilt of *A. thaliana* recapitulates the development of disease symptoms in field hosts \[[@B1]\]. The response of different accessions of *A. thaliana* to different formae speciales varies from complete resistance to ready susceptibility \[[@B1]\]. For example, the standard laboratory accession Columbia-0 (Col-0) is completely resistant to FOM but expresses only partial resistance to FOC1. Taynuilt-0 (Ty-0), on the other hand, is susceptible to FOM but also expresses partial resistance to FOC1. Two strategies are used to map genes responsible for phenotypic variation in populations \[[@B4]-[@B6]\]. When the population of interest is wild and results from an indeterminate number of undefined crosses, a genome-wide association (GWA) study uses evidence of linkage disequilibrium to associate sequence polymorphisms within or near the genes responsible for the trait. Enabling GWA studies in the plant *A. thaliana* is the primary motivation for the 1001 Genomes Project, which has generated whole genome sequence for hundreds of wild accessions of *A. thaliana*\[[@B7],[@B8]\]. Indeed, the detection of functional sequence diversity in *A. thaliana* using GWA is reported \[[@B9],[@B10]\]. However, GWA studies rarely detect more than a modest fraction of the sequence diversity responsible for variation in existing populations of plant and animal species \[[@B5],[@B9],[@B11]\]. Genetic linkage may be used to map the genes associated with a trait to chromosomal intervals. However, this approach requires that the studied population is derived from controlled crosses between defined parents; and, only the genetic diversity distinguishing the parents of crosses is detected. Nevertheless, linkage analysis has been a powerful and successful approach for detecting and defining the genes responsible for complex traits in *A. thaliana*\[[@B12]\]. With plant species that readily inbreed, such as *A. thaliana*, recombinant inbred (RI) populations are almost exclusively used to map and define genetic loci underlying natural traits \[[@B12],[@B13]\]. RI populations in their simplest form originate from an outcross between parents with dissimilar genotypes. Unique recombinant genotypes of the parents are captured in dozens to hundreds of RI lines that result from propagating individual F~2~ offspring by self-fertilization and single-seed descent. After several filial generations of inbreeding, RI progeny become largely homozygous and thus true-breeding RI lines. However, the effort to propagate and curate an RI population without introducing selection represents a substantial investment in time and effort before QTL analysis begins. The effort to generate an RI population is offset by the fact that RI lines are immortal and can be retested innumerable times and reused in separate studies but need to be genotyped just once. There are now dozens of published RI populations from crosses between wild accessions of *A. thaliana*\[[@B12],[@B14]-[@B16]\]. Recently, a technique for generating haploid *A. thaliana* has made the generation of doubled haploid lines possible \[[@B17]\]. Like RI lines, doubled haploids are homozygous and thus immortal but require fewer generations to create. Other mating strategies generate recombinant mapping populations in less time and with less effort than it takes to generate RI lines. In particular, BC~1~ populations are generated from crosses in two successive generations. An initial outcross between parental genotypes produces the F~1~ hybrid, which is then backcrossed to its recurrent parent. Each resulting BC~1~ hybrid inherits a set of non-recombinant chromosomes from the recurrent parent and a set of recombinant chromosomes from the F~1~ hybrid. Because crossovers resulting from single meioses can be unambiguously assigned to recombinant chromosomes, the BC~1~ mating scheme is often used to generate a model population for the evaluation of novel approaches to QTL analysis \[[@B18]-[@B20]\]. In addition, backcrossing is a common feature in traditional breeding schemes that seek to introgress new traits into elite crop varieties \[[@B21]\]. The appeal of BC~1~ populations is undermined by the need for extensive genotyping, and very few studies of natural traits in *A. thaliana* have used BC~1~ populations for genome-wide mapping \[[@B1],[@B12],[@B22]\]. Because each BC~1~ hybrid possesses a unique recombinant genotype, it is necessary to genotype each tested BC~1~ hybrid genome-wide. Without whole genome sequence information for the parents of a BC~1~ population, the discovery of sequence polymorphism and their development into an appropriate set of DNA markers for genome-wide mapping is a time-consuming and laborious process. Nevertheless, prior genetic analysis of a BC~1~ population shows that the qualitative resistance of Col-0 to FOM is a polygenic trait \[[@B1]\]. Six *RFO* QTLs, accounting for the resistance of Col-0 and susceptibility of Ty-0, segregate in a population generated by crossing Col-0 and Ty-0 and then backcrossing the resistant F~1~ hybrid to its susceptible parent Ty-0. Among *RFO* loci, *RFO1* has the strongest association with resistance to FOM. *RFO1* also appears to interact with three other *RFO* loci, namely *RFO2*, *RFO4* and *RFO6*, because the three interacting loci have significant association only when recombinant BC~1~ hybrids also inherit the Col-0 allele of *RFO1* (*RFO1-C*). *RFO2* is a receptor-like protein (RLP) gene that is homologous to the PSY1 peptide receptor gene, *PSY1R*\[[@B23]\]. The *RFO1*-linked gene At1g79670 is named *RFO1* because the Col-0 sequence of At1g79670, as a transgene, enhances the resistance of Ty-0, and the loss-of-function allele of At1g79670 (*rfo1*) compromises the resistance of Col-0 \[[@B1]\]. At1g79670 is a member of the wall-associated kinase-like kinase subfamily of receptor-like kinase (RLK) genes. Here, I map Fusarium wilt resistance in two new BC~1~ populations (i) to address whether At1g79670 alone is responsible for resistance attributed to the *RFO1* QTL, including interactions with *RFO2*, *RFO4* and *RFO6*, and (ii) to examine whether the same or different *RFO* QTLs mediate resistance to different formae speciales of *F. oxysporum*. In doing so, I present a methodology for genome-wide genotyping that makes the mapping of complex quantitative traits a routine procedure. Importantly, because whole genome sequence is now available for most studied accessions, the same approach could be applied to crosses between any pair of *Arabidopsis* accessions. Results ======= Resistance to FOM in *rfo1* --------------------------- In prior mapping of resistance to FOM, *RFO1* was the most significant of six *RFO* loci in *A. thaliana*, and *RFO1* was epistatic to, or enhanced the resistance of, three other *RFO* loci \[[@B1]\]. In theory, the *RFO1* QTL could represent one gene or multiple genes. To appreciate whether At1g79670 is responsible for all or part of the resistance attributed to the *RFO1* QTL, resistance to FOM was mapped in a new BC~1~ population that included *rfo1*, which is a loss-of-function allele of At1g79670 resulting from a T-DNA insertion in and deletion of coding sequence in the Col-0 genetic background \[[@B1],[@B24]\]. The same crossing scheme that generated the original Col-0 and Ty-0 (C-T) BC~1~ population, was used to generate the new *rfo1* and Ty-0 (*r*-T) BC~1~ population with the exception that *rfo1* replaced wild type as the Col-0 parent: Crossing *rfo1* and Ty-0 produced the F~1~ hybrid that was then backcrossed to Ty-0. Differences in quantitative resistance in the new *r*-T and original C-T populations would include the contribution of At1g79670. As in the C-T population, resistance to FOM segregated in the *r*-T population as a polygenic trait, and most BC~1~ hybrids exhibited resistance that was intermediate to that of either parent \[[@B1]\]. Wilt disease in the F~1~ hybrid, Ty-0 parent and 190 BC~1~ hybrids was evaluated using a health index (HI), an ordinal scale from 0 (dead) to 5 (unaffected), described in Methods. At 18 days post infection (dpi), a broad distribution of HI scores registered the breadth of disease resistance among BC~1~ hybrids and presumably the diversity of resistance genotypes (Figure  [1](#F1){ref-type="fig"}c). In contrast, the parents were consistently either resistant or susceptible. Most F~1~ hybrids (Figure  [1](#F1){ref-type="fig"}a) as well as a minority of BC~1~ hybrids (Figure  [1](#F1){ref-type="fig"}c) exhibited only mild symptoms (with a HI score \> 3); and, at the opposite extreme, most of the Ty-0 parents (60 percent, Figure  [1](#F1){ref-type="fig"}b) as well as 10 percent of BC~1~ hybrids were dead (Figure  [1](#F1){ref-type="fig"}c). Thus, segregation of resistance among BC~1~ hybrids was inconsistent with monogenic inheritance as a single locus would have given a 1:1 segregation ratio in the backcross, *i.e.* one plant as resistant as the F~1~ hybrid to one plant as susceptible as the Ty-0 parent. {#F1} Genome-wide linkage of 40 CHR markers ------------------------------------- To expedite the mapping of resistance, methodology to genotype BC~1~ hybrids was developed with efficiency and economy in mind. Previously, *RFO* QTLs were mapped in the C-T population using the genome-wide genotype of 24 SSLP markers distributed over the five chromosomes of *A. thaliana*\[[@B1]\]. However, genotyping one SSLP in one BC~1~ hybrid from one PCR sample is a prohibitive bottleneck in analysis. For instance, if the same 24 SSLPs were used to genotype the 190 FOM-infected BC~1~ hybrids in the *r*-T population, the effort would entail processing no fewer than 4,560 PCR samples. Instead, as described in Methods, the genome-wide genotype of 40 marker loci in each BC~1~ hybrid was obtained from just three multiplex PCR samples. In comparison to genotyping with SSLPs, the new approach gave genome-wide genotypes of BC~1~ hybrids that were comprised of two-thirds more markers and obtained in one-eighth as many PCR samples. The phenotype of the 40 CHR markers was dominant, and primer pairs for CHR markers directed PCR amplification of marker sequence from Col-0 DNA and not from Ty-0 DNA (Figure  [2](#F2){ref-type="fig"}a). DNA products corresponding to as many as 14 markers were amplified in a single multiplex PCR sample and then separated by size using standard agarose gel electrophoresis, as shown for the three multiplex PCR samples of five representative BC~1~ hybrids in Figure  [2](#F2){ref-type="fig"}a. Because BC~1~ hybrids were either Col-0/Ty-0 (C/T) or Ty-0/Ty-0 (T/T) at any locus, genotypes were scored according to whether PCR-amplified marker DNA was present (C) or absent (T), respectively (Figure  [2](#F2){ref-type="fig"}b). {#F2} Genetic linkage between CHR markers in both *r*-T and C-T populations was consistent with the proximity and order of marker sequences in the *Arabidopsis* reference genome (version TAIR10, <http://www.arabidopsis.org>). Genome-wide genetic maps corresponding to recombination frequencies in *r*-T and C-T populations are shown in Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1 and Additional file [2](#S2){ref-type="supplementary-material"}: Figure S2, respectively. In the *r*-T population, marker intervals had mean, median and total genome distances of 15.8 centiMorgan (cM), 12.9 cM and 551 cM, respectively, while individual marker intervals ranged from 4.8 cM to 27.1 cM. (See Additional file [3](#S3){ref-type="supplementary-material"}: Table S1 for recombination frequencies and genetic distances of all intervals). In the original C-T population, 39 dominant markers and one SSLP marker (in place of the linked CHR2.4 marker) had mean, median and total genome distances of 14.1 cM, 14.5 cM, and 516 cM, respectively (See Additional file [4](#S4){ref-type="supplementary-material"}: Table S2 for recombination frequencies and genetic distances of all intervals). Reliability of CHR markers -------------------------- There was concern that dominant CHR markers would not be as reliable as codominant SSLP markers. The absence of marker DNA, which is the phenotype of genotype T/T, could be the false negative result of insufficient PCR amplification of Col-0 DNA from genotype C/T in which case a genotype of C/T would be miscalled as T/T. The codominant SSLPs, on the other hand, were safeguarded from false negative miscalls because marker primers direct the amplification of Ty-0 DNA in all samples, confirming that PCR was productive. Results with SSLPs and CHR markers were compared in the C-T population. As expected, half of genotypes at SSLP markers (50.4 percent with a standard deviation of 3.4 percent) and half of genotypes at CHR markers (50.9 percent with a standard deviation of 3.0 percent) were T/T, so neither codominant SSLPs nor the Col-0-specific CHR markers were prone to give an excess of T/T. The reliability of dominant markers was further scrutinized by examining recombination in a dataset that combined the genotypes of 39 dominant markers and 24 SSLP markers in the C-T population. Miscalled marker genotypes would exaggerate the number of instances of crossovers in adjacent marker intervals because tightly linked markers usually share the same genotype. The mean recombination frequency in intervals separating the 63 markers was 8.7 percent, so pairs of adjacent intervals were expected to have crossover events only once or twice among 234 BC~1~ hybrids. A miscalled genotype would appear to be flanked, in most cases, by markers with opposite genotype and thus by intervals with spurious crossovers. However, instead of having an excess of adjacent double crossovers, the combined marker dataset had a clear deficit of linked double crossovers (Figure  [3](#F3){ref-type="fig"}a). A total of 80 double crossovers were predicted from the sum of the products of recombination frequencies in adjacent intervals, whereas crossovers in adjacent intervals were observed in just 18 instances. Importantly, double crossovers flanked a similar proportion of dominant markers (10) and SSLPs (8). Thus, dominant markers were no more likely than SSLPs to have genotypes that were different from the genotypes of both flanking markers. In addition, the number of crossovers in two intervals was expected to decline as the number of marker intervals separating crossovers increased, whereas the observed number of double crossovers increased with separation of crossover events (Figure  [3](#F3){ref-type="fig"}a). {#F3} Crossover interference ---------------------- The deficiency of linked crossovers was explained by crossover interference, which is observed in *A. thaliana*\[[@B25]\]. Coincidence, a measure of crossover interference, is defined as the observed frequency of crossovers in two marker intervals divided by the product of recombination frequencies in the same intervals \[[@B26]\]. Positive interference has a coincidence value of less than one and indicates that a crossover in one interval inhibits crossover in the other interval. In Figure  [3](#F3){ref-type="fig"}b, positive interference was observed when crossovers were separated by less than 36 cM. Coincidence values near one or greater than one indicate no interference or negative interference, respectively. A transitional negative interference, which is a common observation when positive interference is present, was apparent when crossovers were separated by roughly 53 cM in Figure  [3](#F3){ref-type="fig"}b \[[@B26]\]. Moreover, clear deficiencies of linked double crossovers were observed in all BC~1~ populations examined here. Only 19 to 33 percent of expected double crossovers in adjacent marker intervals were in fact observed in the five *Arabidopsis* chromosomes. (For chromosomal distribution of expected and observed double crossovers, see Additional file [5](#S5){ref-type="supplementary-material"}: Table S3.) No *RFO1* QTL without At1g79670 ------------------------------- In the *r*-T population, association of resistance at CHR markers was evaluated using the Mann-Whitney rank sum test as previously described in \[[@B1]\]. Briefly, BC~1~ hybrids in the *r*-T population were ranked, from most susceptible to most resistant, according to HI scores. At each marker, a standardized statistic *Z* enumerated the separation of ranks of BC~1~ hybrids that were C/T and T/T, and the sign and magnitude of *Z* indicated the direction and strength of genetic association. Specifically, resistance was found to have significant correlation with genotype C/T when *Z* was greater than 3.3 and with T/T when *Z* was less than -3.3 (when *p* \< 0.05, according to permutation tests). In the *r*-T population, no QTL with major effect was detected on chromosome 1, though both *RFO1* and *RFO2* are located on chromosome 1 and make substantial contribution to resistance in the C-T population (Figure  [4](#F4){ref-type="fig"}) \[[@B1]\]. In the *r*-T population, the correlation of resistance with the Col-0 alleles of *RFO1*- and *RFO2*-linked markers, respectively CHR1.9 and CHR1.3, lacked statistical significance (Figure  [4](#F4){ref-type="fig"}). Thus, *rfo1* abolished the major contributions of *RFO1* and *RFO2*. In the C-T population, *RFO2*'s strong association with resistance among plants that are C/T at *RFO1* is absent among plants that were T/T (Figure  [5](#F5){ref-type="fig"}a) \[[@B1]\]. In the *r*-T population, *RFO2*-linked markers had insignificant association with resistance whether BC~1~ hybrids were C/T or T/T at *RFO1* (Figure  [5](#F5){ref-type="fig"}b). {#F4} {ref-type="fig"} for description of plot details.](1471-2229-13-171-5){#F5} *rfo1* also suppressed the apparent interactions between *RFO1* and either the (CHR5.6-linked) *RFO6* or (CHR4.2-linked) *RFO4*. From prior work, resistance is associated with two loci on chromosome 5: *RFO5* gives resistance that is independent of *RFO1* while *RFO6* is only evident among BC~1~ hybrids that are also C/T at *RFO1* (Figure  [5](#F5){ref-type="fig"}a) \[[@B1]\]. In the *r*-T population, CHR5.6 lacked significant association with resistance among BC~1~ hybrids with or without *RFO1-C* (Figure  [5](#F5){ref-type="fig"}b). Similarly, an apparent interaction between *RFO1* and (CHR4.2-linked) *RFO4* was not evident in the *r*-T population, whereas significant association of resistance at *RFO4* in the C-T population is evident only among plants that also have *RFO1-C* (Figure  [5](#F5){ref-type="fig"}a) \[[@B1]\]. In the *r*-T population, marker CHR4.2 was associated with a major QTL without regard to the genotype of *RFO1* (Figure  [5](#F5){ref-type="fig"}b). As previously observed, *RFO3* and *RFO5* expressed resistance that was independent of *RFO1*\[[@B1]\]. In fact, *RFO3* and *RFO5* had stronger correlation with resistance in the *r*-T population than in the C-T population --compare peak *Z* values at *RFO3-*linked (CHR3.3) and *RFO5*-linked (CHR5.3) in Figure  [4](#F4){ref-type="fig"}. Excluding *RFO3*, *RFO4* and *RFO5*, there was no other significant association with resistance to FOM. *RFO* QTLs are pathogen-specific -------------------------------- To examine whether the same or different *RFO* loci provided resistance to different *F. oxysporum* pathogens, resistance was investigated in a third BC~1~ population that was instead infected with FOC1. The HI scores of 200 FOC1-infected BC~1~ hybrids and the two parental accessions, Col-0 and Ty-0, at 16 dpi are shown in Figure  [1](#F1){ref-type="fig"}d, e and f, respectively. Both parental accessions exhibited partial resistance to FOC1 and had median HI scores of 3.5 (Figure  [1](#F1){ref-type="fig"}d, e). BC~1~ hybrids exhibited a broader range of symptom severity than their parents (Figure  [1](#F1){ref-type="fig"}f): 17 percent of BC~1~ hybrids were unaffected (HI = 5.0) while all parents exhibited at least mild symptoms; and, 15 percent of BC~1~ hybrids exhibit more severe symptoms that either parent (HI \< 2.0). Thus, a third of FOC1-infected BC~1~ hybrids expressed an extreme phenotype that was not seen in either parent. *RFO7* confers resistance to FOC1 --------------------------------- A genome-wide genetic map derived from the recombination frequencies between CHR markers in the FOC1-infected C-T population was consistent with the order of marker sequences in the TAIR10 reference genome (See Additional file [6](#S6){ref-type="supplementary-material"}: Figure S3 for the genome-wide genetic map). Intervals between markers ranged from 4.0 to 24.9 cM, and mean, median and total genome distances were 13.5, 12.8 and 472 cM, respectively (See Additional file [7](#S7){ref-type="supplementary-material"}: Table S4 for recombination frequencies and genetic distances of all marker intervals). Association with resistance to FOC1 was evaluated at 40 CHR markers. For the sake of comparison, *Z* statistics at markers in FOC1-infected and FOM-infected C-T populations are juxtaposed in Figure  [6](#F6){ref-type="fig"}. A single major effect QTL at marker CHR5.7 (*Z* = -8.77) associated genotype T/T with strong resistance to FOC1. Because all previous *RFO* QTLs correlated resistance with genotype C/T and CHR5.7 was not previously associated with resistance, this QTL was new and was named *RFO7*. Among F~2~ offspring of Col-0 and Ty-0, genotype C/C at the *RFO7*-linked SSLP CIW9 was more susceptible to FOC1 than genotype C/T, indicating that Col-0 and Ty-0 alleles of *RFO7* express incomplete dominance (Figure  [7](#F7){ref-type="fig"}). {ref-type="fig"} for description of plot details.](1471-2229-13-171-6){#F6} {#F7} Previously, *RFO1* was shown to confer resistance to FOC1 as well as FOM \[[@B1]\]. In the FOC1-infected C-T population, resistance associated with *RFO1-*linked CHR1.9 had questionable significance (*Z* = 2.54, *p* = 0.28). However, among BC~1~ hybrids that were heterozygotes (C/T) at CHR5.7, which minimized the contribution of *RFO7*, the association of resistance with *RFO1* was significant (Figure  [8](#F8){ref-type="fig"}). {ref-type="fig"} for description of plot details.](1471-2229-13-171-8){#F8} Discussion ========== It was conceivable that more than one gene might be responsible for different aspects of the *RFO1* QTL. However, QTL analysis that included *rfo1* was consistent with the simplest explanation: A single gene was responsible for the major effect of *RFO1* and also for apparent interactions with *RFO2*, *RFO4* and *RFO6*. *RFO2* and *RFO6* were undetected in the *r*-T population while the resistance of *RFO4*, which was dependent on *RFO1-C* in the C-T population, was independent of *RFO1* in the *r*-T population. Why *RFO4*, which only attained significance with *RFO1-C* in the C-T population, was a major QTL in the *r*-T population lacking *RFO1-C* is difficult to explain. Possibly, the expression of *RFO4* was influenced by subtle differences in the progression of wilt disease or environmental factors as the C-T and *r*-T populations were similarly infected on separate occasions. Also, the parents, which were nominally from the same Col-0 and Ty-0 accessions, might have been genetically (or epigenetically) dissimilar as separate crosses generated the two populations. Overall, results obtained from independent FOM-infected populations were consistent, and QTLs in the new *r*-T population were coincident with the previously detected *RFO3*, *RFO4* and *RFO5* in the C-T population. In fact, the association of resistance at the three QTLs appeared stronger in the *r*-T population. Col-0, which was the source of resistance, rarely exhibits wilt symptoms when infected with FOM and presumably expresses more than sufficient resistance. As symptom severity is difficult to discriminate among the more resistant plants, loss of *RFO1-C* in the *r*-T population undoubtedly improved the evaluation of disease in BC~1~ hybrids and thus the detection of *RFO* QTLs. The map position and source of QTLs detected in FOC1- and FOM-infected populations suggest that quantitative resistance to *F. oxysporum* is predominantly specific to the infecting forma specialis. Remarkably, resistance to FOC1 was strongly associated with a single new QTL, *RFO7*, though Col-0 and Ty-0 expressed similar partial resistance to FOC1. In addition to *RFO7*, a region on chromosome 1 had a marginal association with resistance, and *RFO1*-linked markers did attain significant association when BC~1~ hybrids were heterozygous for *RFO7*. However, previous work clearly shows that *rfo1* and transgenic *RFO1* affect resistance to both FOC1 and FOM. Thus, while *RFO1* may have a non-specific role in resistance to the three crucifer-infecting *formae speciales*, it appears that *RFO1* also has a much stronger, specific effect on resistance to FOM. Only dominant traits from the donor parent are expressed in a BC~1~ population, so *RFO* alleles of Col-0 that were recessive to alleles of Ty-0 would not be detected. Nevertheless, the strong resistance of F~1~ hybrids, Col-0/Ty-0 and *rfo1*/Ty-0, suggests that resistance to *F. oxysporum* is in large part a dominant trait. Resistance associated with *RFO7* was confirmed in F~2~ progeny, and positional cloning has identified single genes that are responsible for three *RFO* QTLs, *RFO1*, *RFO2* and *RFO3*\[[@B1],[@B23],[@B27]\]. Of the four confirmed *RFO* loci, three QTLs, *RFO1*, *RFO3* and *RFO7*, express incomplete dominance. Nucleotide sequences of both resistant and susceptible alleles of *RFO1*, *RFO2* and *RFO3* encode apparently functional, full-length membrane-spanning receptor proteins. Thus, competition between or interference by the products of the two alleles, rather than gene dosage of the resistant Col-0 allele, might explain the incomplete dominance of natural *RFO* alleles. Because physical interactions between RLKs and a RLP are critical for signaling in plants \[[@B28]\], genetic interaction between *RFO1* and *RFO2* might be evidence for the direct interaction of the corresponding RFO1 RLK and RFO2 RLP. However, because resistance is a complex phenotype, involving processes that occur at different sites in the host and at different times in the infection cycle, the observed genetic interaction might reveal the priority of *RFO1* before *RFO2* without direct interaction. In addition, when neither resistant nor susceptible allele is the null allele, interpretation of genetic interaction is ambiguous. For example, results with *rfo1* clearly implicate *RFO1-C* in the *RFO1*-*RFO2* interaction, however, it remains unclear whether *RFO1-C* suppresses resistance of *RFO2-T* or enhances resistance of *RFO2-C.* Routine QTL analysis in *A. thaliana* is limited to natural traits that distinguish parents of existing RI populations; otherwise, the generation of new RI lines represents a substantial investment of time and effort \[[@B4],[@B5]\]. In the meantime, there is an increasing availability of whole genome sequence, from the 1001 Genomes Project for example, that makes the sequence diversity in hundreds of *Arabidopsis* accessions accessible. As shown here, the mapping of traits that distinguish any two sequenced accessions, including mutant genotypes, can be conceive and complete in six months using BC~1~ populations. Mapping in BC~1~ populations can be a routine procedure when genotyping is efficient, accessible and economical. In this regard, available whole genome sequence from the 1001 Genomes Project was an invaluable resource for identifying accession-specific polymorphisms \[[@B29]\]. Primer sequences for Col-0-specific dominant markers were readily selected from genome sequence reported to be polymorphic in Col-0 and Ty-0. In the same way, dominant markers could be designed to distinguish any two sequenced accessions. In fact, we have reused most of the Col-0-specific markers for genotyping BC~1~ populations from crosses between Col-0 and accessions Zdr-1 or Kondara (unpublished observation). The methodology for genotyping was designed with efficiency and economy in mind. Starting from crude leaf preparations, multiplex PCR DNA of 40 dominant markers was amplified by just three sets of multiplex PCR primers and visualized using standard agarose-gel electrophoresis. The 200 FOC1-infected BC~1~ hybrids were genotyped genome-wide with little more than six 96-well plates of PCR samples. Markers in a multiplex PCR sample appeared as a ladder of bands in agarose gels when all markers were present. Because annealing of marker primers distinguished the Col-0 and Ty-0 genotypes, markers could be arbitrarily assigned sequence lengths that appeared as regularly spaced bands in agarose gels. Results obtained with dominant markers were as reliable as results from codominant SSLP markers \[[@B1]\]. No unforeseen PCR products were amplified when as many as 14 primer pairs were combined in multiplex PCR, and no primer pairs that were confirmed singly subsequently failed when combined with other primer pairs. In theory, an RI population has roughly twice as many crossovers as a BC~1~ population \[[@B30]\]. However, the additional recombination in RI lines remains largely unappreciated unless a high density of DNA markers are used to genotype RI lines \[[@B31]\]. During the inbreeding cycles that generate RI lines, crossovers tend to accumulate at linked sites, and thus recombination in RI lines has the appearance of negative interference. High-resolution analysis of breakpoints in 98 Col-0/L*er* RI lines found, for example, that 17 percent of intervals between crossovers contained just one gene \[[@B32]\]. For genome-wide linkage analysis in BC~1~ populations, 40 evenly spaced markers should be sufficient to capture most recombination. As already mentioned, just one set of homologous chromosomes in BC~1~ hybrids is recombinant. With an average marker separation of 15 cM, I estimate that just seven percent of crossovers went undetected in the BC~1~ populations because approximately three quarters of expected double crossovers in marker intervals would be suppressed by positive interference. In addition, I took advantage of significantly higher recombination in male meiosis (as compared to female meiosis) in *A. thaliana* when generating the BC~1~ hybrids \[[@B33]\]: Ty-0 was the female parent in the backcross while the F~1~ hybrid, which was the source of recombinant chromosomes, was the male parent. Number of crossovers, or amount of recombination, has little bearing on whether a lone QTL is detected \[[@B4],[@B13]\]. Rather, recombination frequency affects the resolution of map position of a QTL, and less recombination would more poorly resolve multiple QTLs in proximity on a chromosome. The detection of two or more linked loci could be suppressed if the loci that remain unresolved express opposing effects on a trait. Indeed, an example of two opposing QTLs for growth rate within an interval of 210 kbp has been reported in Col-0/L*er* recombinants \[[@B34]\]. QTL mapping is just the first step in the identification and characterization of the genes underlying traits. In this regard, mapping in BC~1~ populations is also advantageous because individual (or specific combinations of) QTLs can immediately be reevaluated and fine-mapped in progeny of selfed BC~1~ hybrids. Even after a potentially lethal test, such as resistance to FOM, I was able to collect seeds from 144 of 236 tested C-T BC~1~ hybrids. Although half of the genome in BC~1~ hybrids was heterozygous, on average, seeds were collected from 16 BC~1~ hybrids that were largely homozygous Ty-0 and heterozygous in just four or fewer chromosomal intervals representing 30 percent or less of the genome. *RFO* QTLs in these heterozygous intervals would again segregate in progeny. Conclusions =========== Genome-wide mapping of quantitative Fusarium wilt resistance was expeditious and reproducible in BC~1~ recombinant populations of *A. thaliana*. In two independent BC~1~ populations, resistance to FOM was associated with QTLs *RFO3*, *RFO4* and *RFO5*. Because the resistance of *RFO1*, *RFO2* and *RFO6* was absent in the BC~1~ population that included *rfo1*, the major effect and epistatic interactions of *RFO1* were solely attributed to At1g79670, the gene disrupted in *rfo1*. In a third BC~1~ population, resistance to a second pathogen FOC1 was instead associated with *RFO7*, a new major effect QTL. Pathogen-specific *RFO* QTLs were largely responsible for resistance to the two pathogens, FOM and FOC1. Methods ======= Growing *A. thaliana* --------------------- Seeds of Ty-0 (CS6768) and *rfo1* (Salk_077975) were obtained from the Arabidopsis Biological Resource Center. Seeds were surface-sterilized in 10% household bleach and 0.1% Triton X-100 for 15 min, rinsed 3 times in sterile water. Seeds were sown on peat pellets (Jiffy-730, Grower's Solution Inc., Cookeville, TN) or first germinated on plant nutrient agar (PNA) before transplanting \[[@B1]\]. Plants were arrayed in flats (1′ × 2′) 5 rows by 10 columns and designated: first by flat (1 through 6), second by row (A to E) and third by column (1 through 10). Plants were grown under medium intensity cool white fluorescent lighting (100 to 140 μmoles m^-2^ sec^-1^) for a 12-hr daylength at 25 to 28°C and irrigated with water or fertilizer (PlugCarePlus, Greencare Fertilizers, Inc., Kankakee, IL). Infection with *F. oxysporum* ----------------------------- *Fusarium oxysporum* forma specialis *conglutinans* race 1 (FOC1, isolate 777) and *Fusarium oxysporum* forma specialis *matthioli* (FOM, isolate 726) are from P.H. Williams by way of H.C. Kistler \[[@B3],[@B35]\]. *F. oxysporum* cultures were stored at -80°C in 50% glycerol, grown on Czapek Dox medium (Oxoid Ltd., Hampshire, England) and harvested as described in \[[@B1]\]. Starting with an excess of 3-week old plants, 200 C-T BC~1~, 25 Col-0 and 25 Ty-0 plants with comparable sizes were infected with FOC1; and, 190 *r*-T BC~1~, 25 *rfo1*, and 15 Ty-0 plants were infected with FOM. Plants were irrigated with an excess of washed conidia (2 × 10^6^ conidia mL^-1^). The FOM-infected *r*-T population was scored 11 days post infection (dpi) for three early symptoms: (i) stunting of leaves, (ii) leaf epinasty and (iii) anthocyanin accumulation, using a graduated scale of 1 (severe) to 4 (unaffected). At 18 and 23 dpi, infected plants were scored using a health index (HI), which is the same as the disease index (DI) in \[[@B1]\], ranging from 0 (dead plants) to 5 (unaffected plants) in intervals of 0.5. The FOC1-infected C-T population was similarly scored on 10, 13 and 16 dpi. At the final time point, plants were rank ordered: For the FOC1-infected C-T population, each flat of 40 plants was ranked separately, from 1 (most susceptible) to 40 (most resistant); and, for the FOM-infected *r*-T BC population, all 190 plants were ranked together, from 1 (most susceptible) to 190 (most resistant). Infection and scoring of the FOM-infected C-T BC population is described in \[[@B1]\]. (See Additional file [8](#S8){ref-type="supplementary-material"}: for spreadsheets with phenotypic data of all three BC~1~ populations). Genotyping with CHR markers --------------------------- In proportion to physical and genetic lengths of chromosomes in The Arabidopsis Information Resource (TAIR, <http://www.arabidopsis.org>), 10, 6, 8, 7 and 9 CHR markers were distributed on chromosomes 1, 2, 3, 4 and 5, respectively. On each chromosome, two markers were placed close to the telomeres, and nucleotide positions for remaining markers were spaced at regular intervals in the reference Col-0 sequence of TAIR10. At the approximate nucleotide positions of markers, marker sequences were reference sequences that were classified as highly diverged, or \"unsequenced\", in whole genome sequencing of Ty-0 ( <http://signal.salk.edu/atg1001>). Appropriate pairs of primer sequences were selected in the highly diverged reference sequence using Primer3Plus software, according to recommendations of the QIAGEN Multiplex PCR Handbook (Qiagen Inc., Valencia, CA) \[[@B36]\]. DNA products of 13, 13 or 14 markers were simultaneously amplified by multiplex PCR using three sets of PCR primers. Each set of multiplex PCR primers were designed to give a logarithmic progression of DNA product sizes, ranging from 200 bp to 650 bp in length, which gave regular spacing of marker bands when products were size-separated by agarose gel electrophoresis. Primer sequences and genomic locations of PCR primers are in Additional file [9](#S9){ref-type="supplementary-material"}: Table S5. Sizes and order of DNA products for each set of primers are in Additional file [10](#S10){ref-type="supplementary-material"}: Table S6. PCR amplification was performed using the QIAGEN Multiplex PCR kit according to the protocol for microsatellite loci. A reaction volume of 5 μL included 1 μL crude leaf DNA preparation, 2.5 μL 2× QIAGEN Master Mix, 1 μL 10× primer mix (containing 2 μM of each oligonucleotide primer), and 1.5 μL water. Amplified PCR products were separated by gel electrophoresis in 2% agarose. Crude leaf DNA preparations were prepared according to \[[@B37]\]. See Additional file [8](#S8){ref-type="supplementary-material"}: for spreadsheets with genotypic data for all markers in all BC~1~ populations. Genotypic data for simple sequence length polymorphism (SSLP) C4H, from a prior study \[[@B1]\], replaced CHR2.4s in the analysis of the FOM-infected C-T population. Genetic distances between markers were calculated using the Kosambi mapping function \[[@B13]\]. Genetic linkage supported the presumed physical linkage of markers in the three BC~1~ populations. Linkage data for markers in the three mapping populations are provided in Additional file [3](#S3){ref-type="supplementary-material"}: Table S1, Additional file [4](#S4){ref-type="supplementary-material"}: Table S2 and Additional file [7](#S7){ref-type="supplementary-material"}: Table S4. Testing association of wilt resistance -------------------------------------- A BC~1~ population of *n* plants was rank ordered according to HI scores, from 1 (most susceptible) to *n* (most resistant). Ranking gave priority to later HI scores over earlier HI scores. Rank distributions of the two possible genotypes C/T and T/T were compared using the Mann-Whitney (M-W) test, the results of which were expressed as a standardized statistic (*Z*), the standard deviation units separating the mean ranks of the two genotypes. (See Additional file [11](#S11){ref-type="supplementary-material"}: Table S7 with values of *Z* at CHR markers in the three BC~1~ populations.) For a major effect QTL, threshold values of *Z* for the three BC~1~ populations were determined by permutation tests with 10,000 trials \[[@B18]\]. From the distribution of highest *Z* values in trials, the threshold value of *Z* at *p* = 0.01 (*Z*~0.01~) was 3.86; and, *Z*~0.05~ was 3.36; and, *Z*~0.20~ was 2.80. In *r-*T population, *Z*~0.01~ = 3.73; and, *Z*~0.05~ = 3.27; and, *Z*~0.20~ = 2.77. In FOC1-infected C-T population, *Z*~0.01~ = 3.63; and, *Z*~0.05~ = 3.16; and, *Z*~0.20~ = 2.68. Probability threshold values of *Z* were also determined for QTLs conditioned by genotype at a major QTL \[[@B20]\]. For FOC1-infected C-T population, the M-W test was performed on subpopulations that were either genotype T/T (*Z*~0.01~ = 3.54; and, *Z*~0.05~ = 3.10; and, *Z*~0.20~ = 2.67) or C/T (*Z*~0.01~ = 3.51; and, *Z*~0.05~ = 3.08; and, *Z*~0.20~ = 2.63) at *RFO7*-linked CHR5.7. For FOM-infected plants, subpopulations were tested that were either T/T (for *r*-T population, *Z*~0.01~ = 3.56; and, *Z*~0.05~ = 3.13; and, *Z*~0.20~ = 2.67; and, for C-T population, *Z*~0.01~ = 3.55; and, *Z*~0.05~ = 3.12; and, *Z*~0.20~ = 2.66) or C/T (for *r*-T population, *Z*~0.01~ = 3.52; and, *Z*~0.05~ = 3.12; and, *Z*~0.20~ = 2.64; and, for C-T population, *Z*~0.01~ = 3.88; and, *Z*~0.05~ = 3.34; and, *Z*~0.20~ = 2.81) at *RFO1*-linked CHR1.9. Competing interests =================== The author declares that he has no competing interests. Supplementary Material ====================== ###### Additional file 1: Figure S1 Genetic map of CHR markers in FOM-infected *r*-T population. ###### Click here for file ###### Additional file 2: Figure S2 Genetic map of SSLP and CHR markers in FOM-infected C-T population. ###### Click here for file ###### Additional file 3: Table S1 Linkage of CHR markers in the FOM-infected *r-*T BC~1~ population. ###### Click here for file ###### Additional file 4: Table S2 Linkage of SSLP and CHR markers in FOM-infected C-T BC~1~ population/. ###### Click here for file ###### Additional file 5: Table S3 Observed and expected double crossovers in BC~1~ populations. ###### Click here for file ###### Additional file 6: Figure S3 Genetic map of CHR markers in FOC1-infected C-T population. ###### Click here for file ###### Additional file 7: Table S4 Linkage of CHR markers in FOC-infected C-T BC~1~ population. ###### Click here for file ###### Additional file 8 **Genotypes and phenotypes of BC**~**1**~**hybrids in three*F. oxysporum*-infected populations.** ###### Click here for file ###### Additional file 9: Table S5 Sequence and location of PCR primers. ###### Click here for file ###### Additional file 10: Table S6 Expected PCR products of three sets of multiplex markers. ###### Click here for file ###### Additional file 11: Table S7 *Z* of CHR markers in the three BC~1~ populations. ###### Click here for file Acknowledgements ================ I thank K. Hirschi, J. Merriam, and F. Laski for reading drafts of the manuscript. I gratefully acknowledge research funding provided by the Dean of Life Sciences at the University of California, Los Angeles.

Description =========== **​**The *daf-2* gene encodes an insulin-like growth factor/IGF-1 receptor that regulates *C. elegans* embryonic and larval development. It has previously been shown that DAF-2 inhibits neurite regeneration of the GABAergic motor neurons and PVD sensory neurons in an age-dependent fashion (Bryne et al., 2014; Kravtsov et al., 2017). Following injury, the posterior lateral microtubule (PLM) neurons are capable of regenerating through axonal fusion, a highly efficient regrowth mechanism in which separated fragments fuse back together (Ghosh-Roy et al., 2010; Neumann et al., 2011; Neumann et al. 2015; Abay et al., 2017). We previously established that a critical event for axonal fusion to occur is the exposure of injury-induced phosphatidylserine (PS) 'save-me' signals (Neumann et al., 2015). The level of PS exposed increases with advancing age (Abay et al., 2017). To determine if *daf-2* is involved in this age-dependent modulation of PS exposure, we visualised and quantified the level of PS exposed after PLM axotomy using a secreted, tagged version of Annexin V (Neumann et al., 2011; Mapes et al., 2012; Neumann et al. 2015). Mutation of *daf-2* had no effect on PS exposure 1 h post-axotomy, with no significant differences observed on either the distal or proximal axon segments (Table 1).​ **Table 1.** Quantification of the relative level of PS exposed 1 h post-axotomy. ---------------- ----------------------------------------------------- ---- ------------------------------------------------------- ---- Genotype PS exposed on distal axon (relative to pre-axotomy) n PS exposed on proximal axon (relative to pre-axotomy) n wild-type 1.53 ± 0.105 28 1.44 ± 0.0855 28 *daf-2(e1370)* 1.51 ± 0.167 26 1.57 ± 0.166 26 ---------------- ----------------------------------------------------- ---- ------------------------------------------------------- ---- Reagents ======== One-day-old adult hermaphrodites were used for all experiments, and were grown under standard conditions at 20°C. The BXN301 \[*daf-2(e1370); smIs95(Phsp-16.2::sAnxV::mRFP); zdIs5(Pmec-4::GFP)*\] strain was used along with the CU4204 \[*smIs95(Phsp-16.2::sAnxV::mRFP); zdIs5(Pmec-4::GFP)*\] control strain. The *daf-2(e1370)* allele has been considered temperature sensitive for the dauer phenotype, but not for the long-lived phenotype. At 20°C, *daf-2(e1370)* animals display a greater than 2-fold increase in lifespan compared to the wild-type (Kenyon et al., 1993). Laser axotomy, microscopy and quantification of data was performed as previously described (Abay et al 2017). Acknowledgments =============== We thank Ding Xue for sharing strains. Funding ======= This work was supported by National Health and Medical Research Council (NHMRC) Project Grant 1101974.

All relevant data are within the paper, and a statement to this effect has been included in the final paragraph of the methods section. Introduction {#sec005} ============ Sexually transmitted infections (STIs) have been a longstanding problem in the United States military. Prior to the discovery of antibiotics, these infections were a significant cause of morbidity and mortality, with a significant toll on operational readiness; in fact, before the invention of modern weaponry, venereal disease was more commonly lethal than combat duty \[[@pone.0167892.ref001]\]. During World War I, STIs were only second to influenza in disabling troops from performing their duties, with 7 million person days lost to STIs in the US army \[[@pone.0167892.ref002]\]. In recent decades, the influence of human immunodeficiency virus (HIV), increased numbers of deployed personnel and enhanced tools for STI screening and diagnosis have necessitated ongoing attention to the burden of STIs. There are many reasons why STI diagnoses are prevalent among military personnel, including the population of healthy, sexually active, risk taking young adults. Demographic characteristics commonly associated with increased risk for STIs, including African-American race, younger age, education at the high school (as opposed to college) level and residence in endemic areas \[[@pone.0167892.ref003],[@pone.0167892.ref004]\], are over-represented in the military compared to the general population. Thus, even at accession, military recruits show a high rate of STIs prior to starting their service \[[@pone.0167892.ref005]\]. In addition, military service may appeal to individuals prone to risk-seeking behavior, and high prevalence of sexual risk behavior is well-described \[[@pone.0167892.ref006],[@pone.0167892.ref007]\]. Last, military service provides more opportunities for health screening, resulting in higher ascertainment of diagnoses; the disparity is particularly noteworthy among women in the military, who receive annual gonorrhea/chlamydia screening, and therefore have higher rates of STI diagnosis than men, who are only screened for HIV infection. Despite frequent screening and surveillance, it is often difficult to determine the longitudinal impact of STIs on the military. Several studies have found high rates of reinfection among individuals with STIs \[[@pone.0167892.ref003],[@pone.0167892.ref008]\], but the cumulative risk over an individual's military career is less well-described. This paper takes a novel approach to this problem by selecting cohorts of service-members based on the year they joined the armed forces and evaluating the occurrence of STI diagnoses over the course of their service based on administrative records in the centralized military healthcare system. Our goal was to gain a better understanding of the prevalence of sexually transmitted infections in the DOD, to identify where more data are needed, and to potentially identify areas for intervention. Materials and Methods {#sec006} ===================== We retrospectively studied data which were collected from the Defense Medical Surveillance System (DMSS) between 1997--2012. The DMSS is a centralized relational database which contains medical encounter data for the US armed forces, including clinical, laboratory and epidemiologic data \[[@pone.0167892.ref009]\]. We sampled 100,005 subjects in cohorts of 6,667 based on year of entry into military service; all cohorts had at least one year of follow-up. We oversampled females by a 2:1 ratio to balance the likely outcome counts: while females represent 14.5% of military personnel, they experience approximately twice the rate of STIs of males. We obtained demographic information, including age at accession, sex, race, home of record (zip code), marital status and military-specific data including date of accession/discharge, service affiliation, job code/rank and deployment status. Medical information included all STI diagnoses (at any anatomic site), hepatitis B virus (HBV) and human papilloma virus (HPV) immunization dates and exemptions, and the date of the first DoD seropositive test among HIV-infected individuals. Both etiologic and syndromic STI diagnoses were considered in our search (see definitions, [Table 1](#pone.0167892.t001){ref-type="table"}). 10.1371/journal.pone.0167892.t001 ###### ICD-9 Diagnosis Codes examined in this study. {#pone.0167892.t001g} Condition Diagnosis Code --------------------------------------------- ---------------------------------------------------------------------- Etiologic *Chlamydia trachomatis* 099.41, 099.5 *Neisseria gonorrhoeae* 098, V02.7 Herpes simplex virus (HSV-genital) 054.1, Human papilloma virus (HPV-genital) 078.1, 079.4, 795.05, 795.09, 795.15, 795.19, 796.75, 796.79, V73.81 Syphilis 090, 091, 092, 093, 094, 095, 096, 097 *Lymphogranuloma venereum* (LGV) 991 Trichomoniasis 131 Hepatitis B virus (HBV) 0703, 0702 Human immunodeficiency virus (HIV) V08, 042 Syndromic Cervicitis (unspecified agent, \<45yo) 616.0 Vaginitis (\<45yo) 616.1 Pelvic Inflammatory Disease (PID) 614 Orchitis/Epididymitis, Prostatitis (\<35yo) 604, 601 Urethritis (\<35yo) 099.4 Data analysis {#sec007} ------------- Descriptive statistics were used to characterize the accession year cohorts. Incident STI diagnoses were counted by each quarter of a calendar year. A bacterial STI was considered incident and only counted once in each quarter. Viral STIs were considered incident only at the first recorded diagnosis. If a syndromic diagnosis was found in the same quarter as an etiologic diagnosis explaining the syndrome (e.g. gonorrhea and urethritis), only the etiologic diagnosis was counted. Incidence rates and unadjusted risk ratios were calculated for each STI diagnosis, and a multivariate Poisson regression was performed to determine the relationship between number of STIs and years in the military. In addition, multivariate logistic regression was used to characterize factors independently associated with a history of diagnosed syndromic or pathogen-specific STIs. All patient demographics and medical records were anonymous and de-identified prior to analysis. All relevant data are included within the manuscript. This study was approved by the Institutional Review Board of the Uniformed Services University of the Health Sciences. Results {#sec008} ======= The total sample size included 70,995 males (71%) and 21,005 females (29%), with baseline demographics shown in [Table 2](#pone.0167892.t002){ref-type="table"}. Owing to the large number of subjects in the sample, all variables of interest demonstrated statistically significant differences in univariate analyses. STI rates were highest among women, African-Americans, enlisted personnel and Air Force/Army compared with other service branches ([Table 3](#pone.0167892.t003){ref-type="table"}). STI rates were higher among individuals without a history of deployment, except for pathogen-specific diagnoses ([Table 3](#pone.0167892.t003){ref-type="table"}), though the end-study prevalence of STIs was higher among individuals who had been deployed than individuals with a history of non-deployment ([Table 2](#pone.0167892.t002){ref-type="table"}). 10.1371/journal.pone.0167892.t002 ###### Baseline characteristics of participants and lifetime prevalence of at least one sexually transmitted infection. {#pone.0167892.t002g} Overall--women N (%) Syndromic---women N (%) Etiologic---women N (%) Overall---men N (%) Syndromic---men N (%) Etiologic---men N (%) --------------------------------------------------------------------- ---------------------- ------------------------- ------------------------- --------------------- ----------------------- ----------------------- **Median Age at entry (IQR)**[^1^](#t002fn001){ref-type="table-fn"} 19 (18--22) 19 (18--21) 19 (18--21) 19(18--21) 19 (18--21) 19 (18--21) **Race-White** 17,876 4,823 (26.9) 2,791 (15.6) 51,994 1,266 (2.4) 1,248 (2.4) **Race-Black** 6,813 3,254 (47.8) 1,844 (27.0) 9,874 247 (2.5) 789 (8.0) **Race-Other** 2,789 780 (28.0) 526 (18.9) 5,567 108 (1.9) 144 (2.6) **Race-Unknown** 1,532 597 (39.0) 324 (21.1) 3,560 130 (3.7) 143 (4.0) **Rank- Enlisted** 11,193 (41.4) 8,526 (31.6) 5,237 (19.4) 3,753 (5.5) 1,654 (2.4) 2,247 (3.3) **Officer** 606 (30.5) 443 (22.3) 248 (12.5) 160 (5.0) 86 (2.7) 77 (2.4) **History of deployment** 5,251 (53.4) 4,026 (41.0) 2,709 (27.6) 2,465 (6.9) 1,091 (3.1) 1,481 (4.1) **No history of deployment** 6,548 (34.1) 4,943 (25.8) 2,776 (14.5) 1,448 (4.1) 660 (1.9) 843 (2.4) **Single** 10,123 (40.8) 8,038 (32.4) 4,838 (19.5) 3,509 (5.5) 1,506 (2.4) 2,150 (3.4) **Married** 1,426 (40.5) 1,205 (34.2) 542 (15.4) 347 (5.5) 210 (3.3) 150 (2.4) **Air Force** 3,233 (41.8) 2,480 (32.0) 1,463 (18.9) 780 (5.9) 407 (3.1) 403 (3.1) **Army** 4,924 (42.3) 3,930 (33.8) 2,138 (18.4) 1,638 (6.0) 666 (2.4) 1,042 (3.8) **Marines** 706 (30.9) 485 (21.3) 364 (16.0) 627 (4.5) 290 (2.1) 359 (2.6) **Navy** 2,936 (39.9) 2,074 (28.2) 1,520 (20.7) 867 (5.2) 376 (2.3) 520 (3.1) **Median years of service**[^1^](#t002fn001){ref-type="table-fn"} 4 (1.8--6.3) 5.3 (3.4,8.5) 5.7 (3.7,8.7) 4(2.4.6.9) 6.3(4,11) 6.2 (4--10) ^1^Interquartile range 10.1371/journal.pone.0167892.t003 ###### Incident rates (per 1,000 person-years) of sexually transmitted infections for women and men. {#pone.0167892.t003g} Syndromic---women (95% CI) Etiologic---women (95% CI) Syndromic---men (95% CI) Etiologic---men (95% CI) --------------------------------------------------------------------- ---------------------------- ---------------------------- -------------------------- -------------------------- **Median Age at entry (IQR)**[^1^](#t003fn001){ref-type="table-fn"} **18** 40.1 (39.2, 41.0) 15.7 (15.1, 16.3) 1.5 (1.4, 1.6) 2.2 (2.1, 2.4) **19** 38.7 (37.3, 40.0) 16.1 (15.3, 17.0) 1.8 (1.6, 1.9) 2.3 (2.1, 2.5) **20--21** 38.4 (37.1, 39.6) 14.1 (13.4, 14.9) 1.8 (1.6, 2.0) 2.2 (2.0, 2.3) **22+** 34.4 (33.4, 35.4) 10.4 (9.9, 11.0) 1.7 (1.5, 1.8) 1.6 (1.4, 1.7) **Race-White** 28.3 (27.7, 29.0) 11.7 (11.3, 12.1) 1.7 (1.6, 1.9) 1.5 (1.4, 1.5) **Race-Black** 64.4 (63.0, 65.9) 20.3 (19.6, 21.1) 1.6 (1.4, 1.8) 5.3 (5.0, 5.6) **Race-Other** 28.4 (26.9, 30.0) 13.7 (12.7, 14.9) 1.3 (1.1, 1.6) 1.7 (1.4, 1.9) **Race-Unknown** 34.1 (32.2, 36.1) 10.9 (9.8, 12.0) 2.2 (1.9, 2.5) 2.0 (1.7, 2.3) **Rank-Officer** 16.7 (15.5, 17.9) 6.1 (5.4, 6.9) 1.2 (1.0, 1.5) 1.0 (0.8, 1.2) **Enlisted** 40.1 (39.6, 40.8) 14.9 (14.5, 15.2) 1.7 (1.6, 1.8) 2.1 (2.1, 2.2) **History of deployment** 35.4 (34.7, 36.1) 13.3 (12.8, 13.7) 1.5 (1.5, 1.6) 2.4 (2.2, 2.5) **No history of deployment** 40.9 (40.1, 41.7) 15.0 (14.5, 15.5) 1.9 (1.8, 2.1) 1.9 (1.8, 2.0) **Single** 37.2 (36.6, 37.8) 14.3 (14.0, 14.7) 1.6 (1.5, 1.7) 2.1 (2.1, 2.2) **Married** 44.1 (42.3, 45.9) 12.8 (11.9, 13.9) 2.4 (2.1, 2.7) 1.0 (0.8, 1.2) **Air Force** 33.8 (32.9, 34.7) 11.3 (10.9, 11.9) 1.9 (1.7, 2.0) 1.6 (1.4, 1.7) **Army** 48.5 (47.4, 49.5) 16.0 (15.3, 16.7) 1.7 (1.5, 1.8) 2.6 (2.4, 2.7) **Marines** 25.2 (23.5, 27.0) 12.7 (11.5, 14.0) 1.7 (1.5, 1.8) 1.7 (1.6, 1.9) **Navy** 31.9 (30.9, 32.9) 15.1 (14.4, 15.8) 1.5 (1.3, 1.6) 2.0 (1.8, 2.1) ^1^Interquartile range The career prevalence of ever being diagnosed with an STI is depicted in [Fig 1](#pone.0167892.g001){ref-type="fig"}. For both men and women, increasing number of years of service was associated with an increased proportion experiencing an STI. After 15 years of service, 41% of women and 6% of men had experienced at least one STI. Among women, 33% were diagnosed with a syndromic STI, and 19% were diagnosed with a pathogen-specific STI. Among men, these rates were 3% and 2%, respectively. Over the course of the study, the proportion of syndromic diagnoses decreased from 1997--2014, while pathogen-specific diagnoses increased (data not shown). {#pone.0167892.g001} The distribution of STIs among males and females is shown in [Table 4](#pone.0167892.t004){ref-type="table"}. With the exception of HIV, pathogen-specific STI rates were higher among women than men. Among women, infection with human papilloma virus (HPV) was the most common pathogen-specific diagnosis, followed by chlamydia and trichomonas while male service members were most likely to have an etiologic diagnosis of chlamydia followed by herpes simplex virus (HSV) and gonorrhea. 10.1371/journal.pone.0167892.t004 ###### Case-rates (per 1,000 person-years) for individual STIs among women and men. {#pone.0167892.t004g} STI Cases---women Rate (95% CI)---women Cases---men Rate (95% CI)---men ------------- --------------- ----------------------- ------------- --------------------- Chlamydia 1,688 3.49 (3.33, 3.66) 907 0.69 (0.65,0.74) Gonorrhea 513 1.06 (0.97, 1.16) 551 0.42 (0.39, 0.46) HIV 19 0.04 (0.02, 0.06) 97 0.07 (0.06, 0.09) HSV 1,127 2.33 (2.20, 2.47) 564 0.43 (0.40, 0.47) HPV 2,547 5.27 (5.07, 5.48) 191 0.15 (0.13, 0.17) Syphilis 69 0.14 (0.11, 0.18) 190 0.15 (0.13, 0.17) Trichomonas 784 1.62 (1.51, 1.74) 93 0.07 (0.06, 0.09) In multivariate analyses, factors associated with both etiologic and syndromic STIs among women included African American race, younger age and fewer years of education ([Table 5](#pone.0167892.t005){ref-type="table"}). A longer period of military service was associated with increased likelihood of syndromic or pathogen-specific STIs among both men and women, while history of deployment was independently associated with increased likelihood of syndromic STI diagnosis in both women (OR 1.33 95% CI \[1.24--1.42\]) and men (OR 1.11 95% CI \[1.00--1.18\]). Men with greater time in the military were more likely to receive an STI diagnosis. Finally, racial patterns were differentially associated with syndromic and pathogen-specific STIs, as African American women were more likely to receive pathogen-specific diagnoses than syndromic diagnoses, in contrast with men, among whom syndromic diagnoses were more common. 10.1371/journal.pone.0167892.t005 ###### Multivariate analysis of factors associated with STI diagnosis during military service. {#pone.0167892.t005g} Syndromic odds---women Etiologic Odds---women Syndromic odds -men Etiologic odds---men ------------------- ------------------------ ------------------------ --------------------- ---------------------- **Age** 0.68 (0.6--0.92) 0.88 (0.8--0.97) 0.82 (0.69--0.98) 1.02 (0.85--1.23) **Race-Black** 1.78 (1.65--1.91) 2.28 (2.14--2.43) 3.36 (3.05--3.69) 0.96 (0.83--1.1) **Race-Other** 1.22 (1.1--1.36) 1.02 (0.93--1.12) 1.1 (0.92--1.31) 0.81 (0.66--0.99) **Rank-Officer** 0.84 (0.67--1.05) 0.75 (0.63--0.9) 0.75 (0.53--1.06) 1.06 (0.74--1.53) **Deployed** 1.33 (1.24--1.42) 1.03 (0.97--1.1) 1.11 (1.0--1.22) 0.96 (0.86--1.08) **Single** 1.05 (0.94--1.17) 0.77 (0.71--0.84) 1.34 (1.12--1.61) 0.71 (0.61--0.84) **Service Years** 1.14 (1.13--1.15) 1.2 (1.19--1.21) 1.13 (1.12--1.14) 1.15 (1.14--1.16) Discussion {#sec009} ========== We found a high career prevalence of STI diagnoses among military men and women, consistent with prior studies of STIs in the military. Interestingly, the overall STI prevalence tended to increase throughout an individual's length of military service. While the majority of STI diagnoses were syndromic, the percentage of women experiencing an STI was especially high, with nearly one-third receiving a pathogen-specific diagnosis at some point in their careers. Lastly, as with prior studies of military populations, we found that increased prevalence of STIs was associated with younger age \[[@pone.0167892.ref003],[@pone.0167892.ref005],[@pone.0167892.ref010]\] and African-American race \[[@pone.0167892.ref003],[@pone.0167892.ref005],[@pone.0167892.ref011]--[@pone.0167892.ref013]\]. The sustained increase over time in individual STI diagnoses was a surprising finding. As expected, STI prevalence increased most dramatically earlier in military service, as the cohort is younger and most likely to receive a new diagnosis. As the number of service-years increases, the proportion of individuals with an STI diagnosis remained constant and decreased after approximately six years of service, likely reflecting the retirement of enlisted personnel who tend to have higher STI prevalence. However, after eight years of service, a significant rise in the proportion of individuals with an STI diagnosis was observed, suggesting ongoing transmission among individuals previously believed to be at lower risk for STI acquisition. Military personnel with longer time in service tend to be older, more likely to be married, and have higher education levels, which are all protective factors in most other studies of STIs in the military. Thus, our findings suggest that there continues to exist an ongoing risk for STI acquisition during military service which may be higher than previously appreciated. The disparity in STI prevalence between women and men likely reflects, at least in part, improved case identification due to screening practices among women; the magnitude of the difference suggests that STIs among military men are not being adequately identified and treated. While prior studies involving random samples of women and men have found higher incidence and prevalence of STIs among women, typically the differences are two- to three-fold, as opposed to the significantly larger differences that we report \[[@pone.0167892.ref003],[@pone.0167892.ref014]--[@pone.0167892.ref018]\]. To reduce the prevalence as well as transmission of STIs in the military, these findings suggest a significant benefit to enhancing male-directed screening and diagnostic interventions. Additional measures, such as expedited partner therapy (EPT), which currently is not routinely offered in the military health system, might also provide benefit as has been shown in other high risk settings. Military STI screening policies may also influence the comparability of data with those from civilian populations. All service women receive well-women exams on an annual basis, and both men and women are screened for HIV infection before deployment. In addition, service members receive HIV screening every other year which is not standard practice for the civilian population as a whole. Finally, military personnel undergo initial hepatitis screening upon entry into service, a practice which is not routinely performed among civilians. Beginning in 2001, the Armed Forces Epidemiological Board (now the Defense Health Board) recommended chlamydia screening for for all female recruits within one year of accession, as well as for all female service members at the time of routine pap smear up to age 25. The Navy and Marine Corps were the first military services to implement universal screening of recruits in the mid-1990s and the Air Force and Coast Guard followed suit in the mid-2000s. By 2008, all females in these services were being screened during recruit training using nucleic acid amplification tests (NAATs) while the Army continued screening only at annual exams. Our data suggests that service members are more likely to be screened for STIs than their civilian counterparts with only 44% of civilian women less than age 25 being screened in accordance to guidelines per Healthcare Effectiveness Data and Information (HEDIS) data from 2008 \[[@pone.0167892.ref019]\]. The disparity between genders is likely explained by screening among women that is not performed among males. Among women, infection with human papilloma virus (HPV) was the most common pathogen-specific diagnosis, which is consistent with previous analysis of the military population during this period \[[@pone.0167892.ref020]\]. While the Food and Drug Administration (FDA) approved Gardasil in 2006 and Cervarix vaccines in 2009, the vaccines are not required on entry into military service, and as a result, uptake remains low. Hala-Maktabi et al found that between 2006--11, only 24% of women eligible for HPV vaccination initiated the series \[[@pone.0167892.ref021]\]. A study performed at Womack Army Medical Center found that only 15% of women received the vaccine between 2006--2009, and of these, only 37% completed the series \[[@pone.0167892.ref020]\]. At the Naval Medical Center, San Diego (NMCSD), Gunther et al found completion rates of only 32% of women who initiated the vaccine series, and completion rates were lowest for active-duty females (16%) and males (3%) \[[@pone.0167892.ref022]\]. Last, LaRocque and Berry-Caban found a similar vaccination rate of 15% among Army Soldiers at Fort Bragg North Carolina, from 2007--2010 \[[@pone.0167892.ref020]\]. The benefits of HPV vaccination are well-known, and even with suboptimal coverage, decreased rates of genital warts have been observed in the military since the introduction of HP vaccine \[[@pone.0167892.ref021]\]. Our paper again demonstrates the impact of HPV on sexual health and highlights the need for vaccination. This study also found high rates of chlamydia among both male and female service members, whereas the overall prevalence for men and female service members was 1% and 5%, respectively. This prevalence is consistent with previous studies of the military population. Based on data from the National Health and Nutrition Examination Survey (NHANES), the US Centers for Disease Control and Prevention (CDC) estimated the prevalence of chlamydia among US males and females aged 14--39 to be 1.4% and 2.0%, respectively \[[@pone.0167892.ref019]\]. Similarly, Hakre et al noted that from 2005 to 2010, the Fort Bragg and surrounding Cumberland County, North Carolina, had STI rates that were twice that of the state as a whole \[[@pone.0167892.ref003]\]. Sena et al examined the Fort Bragg population from 1985 to 1996 and noted that while the adjusted incidence of gonorrhea infections declined to below the adjusted North Carolina rates, the rates of chlamydia infections were 3-to-6 times higher than the state and national levels \[[@pone.0167892.ref018]\]. In these studies, higher chlamydia prevalence may partially reflect increased compliance with US Preventive Services Task Force (USPSTF) guidelines through military screening; however, a higher baseline prevalence does seem likely based on the magnitude of the disparity. Our study presents additional information on the relationship between deployment and STI incidence and prevalence. Interestingly, STI rates were higher among individuals without a history of deployment, though individuals without a history of deployment were less likely to be diagnosed with an STI than those with a deployment history. The reasons for this apparent discrepancy are unclear, and may reflect more frequent screening in non-deployed settings (i.e. higher rates), along with longer time in military for individuals with a deployment history, coinciding with a higher likelihood of STI diagnosis, as we have established. The effect of deployment on STI diagnosis has been seen in other studies; Hakre et al found that deployment was associated with lower incidents of STI diagnosis and recurrence \[[@pone.0167892.ref003]\]. Elsewhere, Aldous et al found that between 2005--09, STI prevalence increased among military personnel in Iraq and Afghanistan, likely reflecting a maturation of theater \[[@pone.0167892.ref014]\]. In our own survey, the combat environment may have affected documentation, screening, and treatment which was not captured in the medical record as well as increasing risk for STI. Further research is necessary to better define the relationship between deployment and STI diagnosis. There are several limitations to this study. This study is retrospective, and the use of clinical data (i.e. diagnosis codes) as opposed to laboratory records for data collection may overestimate the prevalence of STIs, and indeed, we noted a high proportion of syndromic relative to pathogen-specific diagnoses. Additionally, the military's screening policy and compliance were not constant during this study which may have affected the incidence of STI diagnoses in this study. Individuals may also be diagnosed with STIs outside the military system (e.g. public health clinics), which would reduce the observed versus actual prevalence. Still, our reported prevalence was high, indicating that most likely this did not occur to a significant extent. Finally, healthcare encounter data are also subject to errors in documentation and were partly collected during transitions between paper to electronic medical record (EMR); increased utilization of the EMR may have led to increases in case reporting during the later years of this study. The observed trend of increasing STI, however, was seen even in later years, when EMR use was consistent. The strengths of this study include its large sample size, which would overcome many of these limitations and smooth potential inaccuracies. In addition, our analysis by calendar quarter allowed us to examine re-infection rates that may be missed by other research methods. Demographics of such a large sample naturally mirror those of the overall military, mitigating any sampling bias. In fact, the large population allowed for significant analysis to occur on less common STIs which have not been thoroughly analyzed in this population. While a number of cross-sectional studies have shown high prevalence of STIs, our analysis represents the first effort to follow a specific, large cohort to determine STI risk over time. Conclusions {#sec010} =========== We found a high burden of STIs in all branches of the military, and increasing years of service were associated with continued increases in new STI diagnoses. This study reinforces the need for ongoing STI-related screening and educational efforts throughout the military career of all service members. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceived and designed the experiments:** RD OM BA.**Performed the experiments:** RD OM.**Analyzed the data:** RD RB OM BA.**Wrote the paper:** RD RB OM EC JS JM AG GM BA.

1. Introduction {#sec1} =============== High cumulative doses of anthracyclines (300--500 mg/m^2^) are frequently administered to children with cancer. Cardiac toxicity is a serious adverse effect that limits the therapeutic potential of anthracyclines and threatens the cardiac function of pediatric cancer patients leading to debilitating long-term effects resulting in poor quality of life in cancer survivors \[[@B1]--[@B5]\]. This is particularly devastating in children who are cured of their cancer because they have to endure the debilitating cardiac dysfunction for the rest of their lives with limited exercise capacity which may also lead to other chronic illnesses. B-type-natriuretic peptide (BNP) is a polypeptide hormone predominantly released from the cardiac ventricles in response to volume expansion and pressure overload. BNP is found in the circulation as BNP-32 and the NH~2~-terminal portion of ProBNP (Nt-proBNP). BNP levels are elevated in patients with left ventricular systolic dysfunction and correlate with the severity of symptoms and prognosis \[[@B6]--[@B14]\]. Measuring serum Pro-BNP levels is a reliable way to monitor the cardiac function of patients receiving cardiotoxic drugs such as anthracyclines. Selenium (Se) is a trace element distributed in a small amount in the soil and certain foods. It is an important antioxidant, and its absence has been associated with cardiomyopathy in people living in areas with poor levels of soil Se. The concentration of Se in grain varies based on the soil content. Dietary Se is found in meat and seafood. It is a cofactor for glutathione peroxidase which catalyzes the reduction of hydrogen peroxide using glutathione. It is an essential element to remove free radicals from the body and to prevent oxidative tissue damage \[[@B15]--[@B19]\]. Se supplementation could potentially prevent cardiac toxicity of anthracyclines \[[@B16]--[@B20]\]. In this study, we assessed anthracycline-induced cardiotoxicity by measuring Pro-BNP levels and echocardiographic (ECHO) findings, and we investigated the potential protective effect of Se supplementation in a group of children with high Pro-BNP levels and/or cardiac dysfunction. 2. Patients and Methods {#sec2} ======================= Plasma level of Pro-BNP was measured, and echocardiography (ECHO) was performed in 67 pediatric cancer patients (45 boys and 22 girls, ages between 2 and 18 years, median age 12 years) with a variety of tumors (leukemias, lymphomas, solid tumors) after completing anthracycline-containing treatment. Serum Se levels were measured in 37 patients. Sera were stored at −20 degrees centigrade until selenium levels were measured with atomic absorption method. Patients with low level of Se were supplemented with Se (100 mcg/day). 3. Statistical Analysis {#sec3} ======================= Statistical analysis was performed using SPSS (Version 15.0) software package. Comparisons between the groups were done using Mann-Whitney  *U*  test, Wilcoxon sign test, and Fisher\'s exact test. Levels of statistical significance were set at a  *P*  value \< 0.05. The results were expressed as range (minimum and maximum) and median. 4. Results {#sec4} ========== In eleven patients who had high Pro-BNP levels and/or cardiac failure Pro-BNP levels ranged between 10 and 8022 pg/mL with a median of 226.3 pg/mL (normal \< 120 pg/mL). Fifty-six patients had normal Pro-BNP levels (8.2--119.6 pg/mL, median 32.4 pg/mL). As seen in [Table 1](#tab1){ref-type="table"}, the difference in levels of Pro-BNP between these two groups was significant (*P* \< 0.001). Serum Se levels were low in 10 of these 11 patients with high Pro-BNP levels and/or cardiac failure (20--129 mcg/L, median 62 mcg/L). Twenty-six of 56 patients with normal Pro-BNP levels were also investigated for Se levels (51.3--150 mcg/L, median 99.4 mcg/L). There was a significant difference between Se levels of patients in high Pro-BNP and normal Pro-BNP groups (*P* \< 0.001) ([Table 1](#tab1){ref-type="table"}). Abnormal ECHO findings were observed in 7 of 11 (63.6%) patients with high Pro-BNP levels and/or cardiac failure group. Only 1 (3.8%) of 26 patients with normal Pro-BNP levels had abnormal ECHO finding. A patient with normal pro-BNP and low Se level died in 1 month because of progressive disease with respiratory failure and cardiac failure. The probability of having abnormal ECHO findings was significantly higher in patients with high Pro-BNP compared to those with normal Pro-BNP (*P* \< 0.001) ([Table 2](#tab2){ref-type="table"}). Eight of 11 patients with low Se and high Pro-BNP levels were supplemented with Se 100 mcg per day for 4--33 months (median 6 months). Three of 8 patients had cardiac failure according to ECHO and were supplemented with Se in addition to digoxin and ACE inhibitors. All 3 patients were doing well with normal ECHO findings and normal Pro-BNP levels after a follow-up periods of 33, 14, and 5 months. Five patients, 3 with normal ECHO and 2 with diastolic dysfunction (one with low Pro-BNP level, other with high Pro-BNP level) also, were supplemented with selenium (100 mcg per day). One patient who had diastolic dysfunction with normal Pro-BNP did well with Se supplementation with normalization of ECHO findings, but she later died due to progression of her cancer. Another patient with diastolic dysfunction as well as 3 patients with normal ECHO had normal Se and Pro-BNP levels after 4--6 months of Se supplementation. Only 3 patients were not supplemented with Se in the high Pro-BNP and/or cardiac failure group, because one of them had normal Se level, the second one died with progressive disease in a very short period of time, and the third one had Pro-BNP level within normal limits after the removal of intracardiac tumor thrombus with open heart surgery ([Table 3](#tab3){ref-type="table"}). In Se-supplemented group, supplementation period was between 4 and 33 months (median 6 months). Before supplementation, Pro-BNP levels were between 10 and 843 pg/mL (median 175 pg/mL). After supplementation, Pro-BNP levels were 2--536 pg/mL (median 73.5 pg/mL) which were significantly lower than pretreatment levels (*P* = 0.018). Pretreatment Se levels were between 20 and 83 mcg/L (median 57 mcg/L). After supplementation Se levels were 65--109 mcg/L (median 103 mcg/L) which were significantly higher than presupplementation level (*P* = 0.028) ([Table 4](#tab4){ref-type="table"}). After achieving normal Se and Pro-BNP levels, Se supplementation was discontinued. During follow-up period with no Se supplementation, 2--6 months after supplementation repeat measurements of Se levels were 75--106 mcg/L (median 83 mcg/L), and Pro-BNP levels were 10--123.5 pg/mL (median 106.5 pg/mL), which were lower for Se (*P* = 0.068) and higher for Pro-BNP (*P* = 0.109) compared to Se-supplemented period ([Table 4](#tab4){ref-type="table"}). 5. Discussion {#sec5} ============= The main long-term toxicity of anthracyclines is cardiac dysfunction associated with their chronic and/or high-dose administration. Severe cardiomyopathy and congestive heart failure may develop any time after the completion of the treatment. The precise pathogenesis of anthracycline-induced cardiotoxicity is still uncertain, and it is likely to be multifactorial in origin. Nevertheless, pivotal role is attributed to the iron-catalyzed intramyocardial production of reactive oxygen species (ROS), which cause damage of various targets in the myocardial cells \[[@B1]--[@B5]\]. Probrain natriuretic peptide (Pro-BNP) is released by cardiac cells, and serum levels are elevated even before the development of overt cardiac distress symptoms related to impairment of left ventricular systolic or diastolic function leading to increased left ventricular wall stretch. Recent studies have also suggested that ischemia itself may promote release of BNP \[[@B7], [@B20]--[@B25]\]. In the present study, we evaluated cardiotoxicity in 67 pediatric patients with cancer (leukemia, lymphoma, and solid tumor) after they completed treatment with anthracycline-containing regimens. We also evaluated Se levels and the effects of Se supplementation with regard to cardiotoxicity because previous studies with Keshan disease (KD) suggested potential protective role of Se for cardiac dysfunction observed in Se deficiency. KD, a potentially fatal form of cardiomyopathy, first found in Keshan county, northeast China, is one of the most harmful endemic diseases. The disease is characterized by multifocal myocardial necrosis and fibrosis and leads to congestive heart failure and cardiogenic shock. Although the exact etiology of KD is unclear, Se deficiency is a major contributing factor \[[@B19]\]. Investigations into the epidemiology of KD revealed that individuals living in areas with Se-poor soil were under a high risk of development of the disease. Individuals living in those areas had low dietary intakes of Se that were reflected in low serum and hair levels of Se. Populations living in areas of China with Se-rich soil did not develop KD \[[@B26]--[@B29]\]. In this study we have investigated the potential role of Se in anthracycline-induced cardiotoxicity in pediatric cancer patients undergoing chemotherapy. We found an association between low Se levels and anthracycline cardiotoxicity which could be prevented by Se supplementation. These results suggest that Se deficiency may have an effect on anthracycline-induced cardiomyopathy, which may have similarities to KD. The family of selenoproteins includes glutathione peroxidases, the redox enzymes that take advantage of the chemical properties of Se to remove free radicals by reduced glutathione and thus to form oxidized glutathione. Se supplementation had a protective effect on ischemia/reperfusion injury in experimental animals; it improved the recovery of cardiac function, decreased ultrastructural changes, increased the expression of glutathione-related enzymes, and partially affected the antioxidant capacity of the tissues together with an effect on gene transcription level \[[@B30], [@B31]\]. Se supplementation prevented the hypoxia/reoxygenation injury of the isolated neonatal cardiomyocytes and resulted in an NO-related increase of inotropic response of cardiac muscle to the beta-adrenergic stimulation by isoproterenol \[[@B17]\]. Oral Se supplementation has been shown to reverse the biochemical evidence of the Se deficiency \[[@B30]--[@B32]\]. The beneficial effect of treatment with the inorganic form of Se was also demonstrated in experimental models of cardiac injury \[[@B32], [@B33]\]. The mechanism by which Se influences iNOS cardiac expression is unknown. Kim et al. \[[@B34]\] have shown that lipopolysaccharide-activated human T cells with relatively high concentrations of selenite had lower NF-kB-binding and -decreased NO production. Similarly, Turan et al. \[[@B35]\] observed that total NF-kB in the cardiac muscle was reduced by Se. They suggested that Se deficiency or excess affects signal transduction. Se effect can be monitored with Pro-BNP, a good marker of cardiac function \[[@B7], [@B37]\]. Dietary supplementation of 100 *μ*g Se (sodium selenite) in patients receiving total parenteral nutrition has been reported to prevent arrhythmias and cardiomegaly and lead to an increase in left ventricle ejection fraction \[[@B40]\]. In addition, the incidence of Keshan disease, an endemic dilated congestive myocardiopathy in areas of Se deficiency in China and Russia, has been shown to be decreased by oral Se supplementation at a dosage of 150--300 *μ*g/week \[[@B40], [@B39]\]. It should be noted that Se supplementation has also been suggested as a strategy for prevention of myocardial disease in other studies of human cardiac pathology \[[@B40]--[@B38]\]. The results of our study support the hypothesis that Se supplementation could be considered as a strategy for treatment and prevention of anthracycline-induced cardiomyopathy observed in children with cancer. Our results also suggest that Se supplementation should be continued much longer to ameliorate or prevent anthracycline-induced cardiotoxicity. In conclusion, our results suggest that Se supplementation may have a potential role in the protection against anthracycline-induced cardiac toxicity in patients with high pro-BNP level and/or cardiac failure and low Se levels. The authors thank Ankara University Medical School for supporting the project. ###### Selenium and Pro-BNP levels of patients.   Normal Pro-BNP High Pro-BNP and/or abnormal ---------------------------- ---------------- ------------------------------ ---------- ------- \*Pro-BNP (pg/mL) 8.2--119.6 32.4 10--8022 226.3 \*\*Selenium (mcg/L)\*\*\* 51.3--150 99.4 20--129 62 \**P*\<0.001. \*\**P*\<0.001. \*\*\**n* = 26 (twenty-six of fifty-six patients with normal Pro-BNP were measured for Se level) Mann Whitney *U* test was used, and median (range) was given as descriptive statistics. ###### Echo findings of patients with high and normal Pro-BNP levels.   ECHO findings Total number of patients (%) ----------------------- --------------- ------------------------------ ---------- Normal Pro-BNP levels 25 (96.2) 1 (3.8) 26 (100) High Pro-BNP levels 4 (36.4) 7 (63.6) 11 (100) Total 29 (78.4) 8 (21.6) 37 (100) *P* \< 0.001 (Fisher\'s exact test). ###### Se-supplemented patients with low serum Se levels, high Pro-BNP levels, and/or cardiac failure. ------------------------------------------------------------------------------------------------------------------------------------------------------------------- Pt.\ Total anthracy\ Pro-BNP\ Se\ ECHO Cardiac failure Digoxin\ Se\ End results *n* = 11 mg/m^2^ pg/mL mcg/L Enalapril\ Suppl. Furosem ---------- ----------------- ---------- ------- --------------------------------- ----------------- ------------ --------- ------------- ------ ------- ----- ----- ES 550 754 70 Systolic failure \+ \+ 100 mcg Normal 536 123.5 108 86 NBL 180 175 52 Systolic failure \+ \+ 100 mcg Normal 85 95 103 106 ACC 400 10 71 Diastolic failure − − 100 mcg Normal 10 10 65 75 HL 300 843 55 Diastolic failure − − 100 mcg Normal 298 118 72 81 NHL 240 172 49.2 Normal − − 100 mcg Normal 12.6 NA 109 NA HL 400 197.5 20 Normal − − 100 mcg Normal 80 NA 208 NA BL 150 170.4 57 Normal − − 100 mcg Normal 2 NA 75 NA NHL 120 277 83 Systolic failure \+ \+ 100 mcg Normal NA NA NA NA RMS 0 127 62 Intracardiac thrombus − − --- Normal 67 NA NA NA AML 400 8022 65 Normal − − --- Died NA NA NA NA TALL 320 1536 129 Pericardial effusion, tamponade − − --- Died NA NA NA NA ------------------------------------------------------------------------------------------------------------------------------------------------------------------- Pt: patient; NA: not available, ES: Ewing\'s sarcoma, NBL: neuroblastoma, ACC: adrenocortical carcinoma, HL: Hodgkin lymphoma, NHL: non-Hodgkin lymphoma, BL: Burkitt lymphoma, RMS: rhabdomyosarcoma, AML: acute myeloid leukemia; T ALL: T acute lymphoblastic leukemia. ###### Pre- and postsupplementation levels in Se-supplemented patients with low serum Se levels and high Pro-BNP levels and/or cardiac failure. ------------------------------------------------------------------------------------------------------------------------------------------------- *n* = 8 Presupplementation levels,\ 1st postsupplementation levels, range (median) 2nd postsupplementation levels, range (median) range (median) ----------------- ----------------------------- ------------------------------------------------ ------------------------------------------------ Pro-BNP (pg/mL) 10--843 (175) 2--536 (73.5)\* 10--123.5 (106.5)\*\* Se (mcg/L) 20--83 (57) 65--109 (103)\*\*\* 75--106 (83)\*\*\*\* ------------------------------------------------------------------------------------------------------------------------------------------------- \**P* = 0.018. \*\**P* = 0.109. \*\*\**P* = 0.028. \*\*\*\**P* = 0.068. Wilcoxon sign test was used; median and range were given as descriptive statistics. [^1]: Academic Editor: Julian J. Raffoul

Pathogenic response to viral infection varies dramatically between individuals infected with the same viral strain and dose, and much of this variation is heritable. The impact of host genetics is evident both on the primary exposure to a virus during early life ([@bib82]) and upon infection with newly emerging viral strains; the latter, where prior immune exposure to a variant viral strain is not cross-protective, being especially common for quickly evolving RNA viruses such as the influenza A virus (IAV) ([@bib55]). Pathogenesis induced by IAV, whether contracted during early childhood or later in life, is thus likely to have a significant heritable component. A greater understanding of this heritability should improve our ability to not only identify populations at risk of enhanced morbidity and mortality during an emerging pandemic, but also to identify successful options for treatment. The past several years have seen significant progress in identifying and characterizing host genes that modulate susceptibility to IAV infection via knockout mouse studies, *in vitro* screens, and studies of primary immunodeficiencies and allelic variants in humans ([@bib85]). In humans, screening for inborn errors identified a major role for interferon regulatory factor 7 (*Irf7*) in modulating the severity of primary IAV infection ([@bib11]), and allelic variation in *Ifitm3*, which was identified in a high-throughput siRNA screen, was associated with differential severity of IAV-infection outcomes during the 2009 H1N1 pandemic ([@bib18]). Most of our insights into genes modulating host IAV resistance, however, have come from studies on mice. These include studies using knockout mice---which have identified host genetic factors critical to antiviral responses, including *Tlr3* ([@bib31]) and *Isg15* ([@bib45])---and studies that examine differences between laboratory inbred strains. Inbred strain studies were the first to identify the *Myxovirus resistance* (*Mx*) family of proteins as important for host antiviral response ([@bib81]), and inbred studies have continued to demonstrate the relevance of genetic background to multiple aspects of IAV pathogenesis ([@bib80]; [@bib1]; [@bib43]; [@bib74]). Yet, despite the identification of clear phenotypic differences between inbred strains, there have been relatively few attempts to dissect the genetic basis of those differences using traditional quantitative trait locus (QTL) mapping approaches such as the use of F2s or backcrosses (although see [@bib4]). This may be, in part, because traditional QTL mapping approaches tend to rely on outbred animals---and when it comes to studying viral pathogenesis, outbreds are in many respects problematic. One important limitation is phenotyping: studying the response to an infection is equivalent to studying the causal effect of an applied treatment in that its strict definition relies on a comparison between otherwise identical individuals subject to infection *vs.* control. But such like-for-like comparisons are biologically and technically challenging to make in an outbred population, where every individual is genetically distinct, and this has undesirable consequences for downstream interpretation: namely, that when genetic determinants of severe IAV pathogenesis are confounded with those influencing baseline phenotypes, the roles of any detected QTL are ambiguous. A related disadvantage of outbreds from the perspective of genetics is the inability to obtain biological replicates, which makes it harder to distinguish which aspects of pathology are stable consequences of genetics *vs.* products of stochastic variability. This is particularly important, since it also makes it almost impossible to follow-up on genuinely extreme responders for additional mechanistic and genetic analysis. Translating strain differences in IAV pathogenesis to meaningful QTL studies ideally requires an experimental paradigm that combines population-level genetic diversity with individual-level replicability. An exciting opportunity is therefore presented by replicable genetic reference populations, in particular, those based on panels of recombinant inbred (RI) strains. Across a panel of RIs, genetic background varies, providing a basis for QTL mapping; within a RI strain, individuals are genetically identical, providing a basis for replication. The combination allows infection response to be rigorously defined and genomic regions affecting that response to be mapped. It also permits the creation of sophisticated experiments that target a wider range of heritable mechanisms: crossing RIs with each other to form RI intercrosses (RIXs), or crossing them with outside strains, produces replicable systems capable of distinguishing, for example, additive, dominance, and parent-of-origin effects, among others (*e.g.*, [@bib95]; [@bib34], [@bib33]; [@bib42]; [@bib84]; [@bib54]; [@bib26]; [@bib83]; [@bib77]; [@bib47]; [@bib101]; [@bib29]; [@bib93]). RI genetic reference panels range from inbred lines derived from crosses between two mouse strains to more complex multiparental crosses. The BXD RI panel, derived from two founder strains, C57BL/6J and DBA/2J, has been used to study the impact of genetic variation on susceptibility to IAV infection and map QTL associated with these effects. [@bib5] studied H5N1 infection in females from 66 BXD strains, and [@bib58] studied H1N1 infection in 53 BXD strains, with both studies identifying QTL associated with susceptibility to infection. The Collaborative Cross (CC) RI panel is a multiparental population (MPP) descended from eight inbred founder strains ([@bib84]; [@bib9]), with these founders including representatives from the three major domesticated house mouse subspecies ([@bib99]). As such, the CC captures considerably more genetic diversity and, thanks to its breeding structure, this diversity is also more uniformly distributed across the genome, with as many as eight distinct haplotypes segregating at any given locus within the population ([@bib12]; [@bib79]). The eight CC founder strains have distinct pathogenesis profiles in response to influenza virus ([@bib43]), suggesting that the CC RI panel is capable of a broader phenotypic range than would be observed in less complex populations. Indeed, studies using an incompletely inbred, ancestor population of the CC (pre-CC), demonstrated high levels of phenotypic variation across the population and successfully mapped several QTL associated with variation in susceptibility to IAV infection ([@bib19]; [@bib6]). The CC therefore represents a promising resource for studying how genetically diverse populations respond to IAV infection. Determining an optimal strategy for how the CC should be used to study the genetic architecture of IAV pathogenesis is nonetheless complicated because (1) the space of possible experimental designs is vast, and (2) information about what types of heritable effects are likely to be present is extremely limited. Regarding (1), with ∼75 CC strains currently available, including all reciprocal F1 hybrids (so called CC-RIXs), there are \>5600 potential replicable configurations. Since only a subset of these configurations can be explored within any realistic experiment, any chosen experimental design necessarily targets some types of heritable effects to the exclusion of others. Regarding (2), to date, most *in vivo* studies of IAV pathogenesis have been confined to candidate genes or additive interactions at single loci; studies investigating the broader question of what types of heritability are at play during IAV infection are largely absent. To rationally design studies of heritable effects in complex populations such as the CC it is therefore helpful to have advance knowledge of which types of heritable effects might be present. One source of such information is phenotype data collected on the multiparental founders and their F1 hybrid offspring, a combination that can be more formally described as an (inbred) diallel. Diallels have a long history in quantitative genetics, having been used originally in plant breeding studies to judge the relative merits of different strain combinations and subsequently for gaining insight into the heritable architecture of a broad range of phenotypes \[*e.g.*, references in [@bib8], [@bib44], and [@bib62]\], including host--pathogen interactions in, *e.g.*, crickets ([@bib66]) and flies ([@bib92]) (see *Statistical Models and Methods* and *Discussion* for connections to other diallel literature). Here we use a diallel of the CC founders and their reciprocal F1 hybrids (hereafter, a CC founder diallel) to give an overall predictive picture of the range and relative influence of the different types of heritable effects on IAV pathogenesis that are likely to be present in CC founder-derived MPPs, a group that includes not only replicable MPPs such as the CC and the CC-RIX but also irreplicable ones such as the Diversity Outbred (DO) population ([@bib105]). We take advantage of the diallel design's replicability to measure IAV-induced pathogenesis in a precise way, as the response to an applied treatment defined in terms of postinfection (p.i.) weight-loss differences (deltas) between matched sets of mock and infected individuals ([Figure 1](#fig1){ref-type="fig"}). Adapting a recently developed statistical framework for analyzing treatment-response diallels ([@bib14]), we use those deltas to model how pathogenic response to IAV is modulated by parentage, sex, and their interaction, framed in terms of additive genetics, dominance, epistasis, parent-of-origin, and sex-specific versions thereof. After observing that, following IAV infection, diallel individuals show a broad, continuous distribution of day 4 (D4) p.i. weight loss, we find, through statistical modeling, that the IAV-induced weight loss includes substantial contributions of host additive, epistatic, and sex-specific effects, with much of the heritable variation closely tracking the genotype state implied by the three distinct functional alleles of the previously identified resistance locus *Mx1*. Confirming previous findings, the functional CAST/EiJ (CAST) *Mx1* allele, in contrast with functional NZO/HlLtJ (NZO) and PWK/PhJ (PWK) *Mx1* alleles, provides intermediate levels of protection against H1N1 influenza strains. Unexpectedly, and confirmed through additional modeling, we found that different classes of functional *Mx1* alleles exhibit distinct functional patterns, additive or dominant, when combined with null *Mx1* alleles. Further, illustrating our general rationale, we show that the major strain-specific, *Mx1*-effect patterns are consistent across two CC founder-derived MPPs: the pre-CC, as determined from reanalysis of a previously published data set ([@bib19]), and a previously unpublished 105-line CC-RIX, in which we conduct a limited analysis focused on the *Mx1* locus. {ref-type="table"}). Light blue squares represent animals that were mock treated and red squares represent animals infected with IAV, with daily weights for each taken from D1 to D4 p.i. (and these were analyzed with model 2 in [Table 2](#t2){ref-type="table"}). Purple squares represent infection response, the primary quantity of interest, estimated using match quartets of one mock to three infected mice (analyzed with models 3 and 4 in [Table 2](#t2){ref-type="table"}). Labels within each square indicate phenotypes analyzed, where weight = preinfection body weight, pct = p.i. percent change in starting D0 weight (post), and "delta" = infection response, as described in the *Statistical Models and Methods* section. The coloring increases in saturation from D1 to D4 for the influenza and matched quartet groups to indicate an overall increasing amount of p.i. weight loss over time.](427f1){#fig1} Experimental Materials and Methods {#s1} ================================== CC founder diallel mice {#s2} ----------------------- The inbred and F1 mice used within this study were bred in-house at the University of North Carolina at Chapel Hill (UNC-CH). This colony was directly descended from the subset of animals used to generate the initial CC funnels ([@bib12]) and included mice from the following eight strains at The Jackson Laboratory: A/J (AJ; \#000646), C57BL/6J (B6; \#000664), 129S1/SvImJ (129; \#002448), NOD/ShiLtJ (NOD; \#001976), NZO (\#002105), CAST (\#000928), PWK (\#003715), and WSB/EiJ (WSB; \#001145). Mice from the UNC-CH colony were then used to generate all 62 possible inbred and (reciprocal) F1 combinations between these eight strains, excluding NZO × CAST and NZO × PWK matings which are nonproductive ([@bib7]) ([Figure 2A](#fig2){ref-type="fig"}). This yielded a total of 124 distinct combinations of sex and parentage (hereafter, described as "diallel categories"). Lung tissues were collected from a subset of each of the founder inbred strains in this study, at day 2 (D2) and D4 p.i., and were used for a separate comparative RNA-seq analysis by [@bib96]. {#fig2} Mouse infections in the diallel {#s3} ------------------------------- Mice were weaned at ∼21-d old and housed four per cage, within each diallel category, under standard conditions (12 hr light/dark; food and water *ad libitum*). Of the four mice in a cage, one was randomly assigned to mock and three to influenza infection, as there is no evidence that mice can transmit influenza virus. Each cage was then assigned to a harvest time point: D2 p.i. (*n* = 533 mice), or D4 p.i. (*n* = 510 mice). At 8--12 wk of age, based on their assignments, mice were anesthetized with isoflurane and inoculated intranasally with 500 plaque-forming units (PFU) of mouse-adapted IAV (H1N1 A/Puerto Rico/8/1934; short name PR8) or with the diluent, phosphobuffered saline (PBS), alone as a mock control. For each inbred strain and F1 cross, about six mice (range: 5--9) of each sex were infected with IAV PR8, and about two mice (range: 2--3) of each sex were mock infected. This gave a total of 1043 mice across 54 experimental batches. Treatment assignment was random: same-sex siblings from the same cage (and therefore batch) were randomly assigned at weaning to mock or infected groups prior to being moved to new cages. The 1043 mice were housed in ∼260 cages (about four mice per cage), with 775 infected mice and 268 mock-infected mice. Body mass was recorded daily. All animal experiments were carried out in compliance with the Guide for the Care and Use of Laboratory Animals (Institute of Laboratory Animal Resources, National Research Council, 1996, <https://www.ncbi.nlm.nih.gov/books/NBK232589/>). Animal protocols were approved by the Institutional Animal Care and Use Committee of UNC-CH. Mouse infections in the pre-CC and CC-RIX {#s4} ----------------------------------------- To verify that strain-specific haplotype effects measured in the diallel were consistent with those at the host resistance locus *Mx1*, we sought out CC-related IAV infection data sets for which we could isolate *Mx1* locus-specific effects. ### Existing data from pre-CC study: {#s5} In the QTL mapping study of host response to IAV infection of [@bib19], 155 female pre-CC mice from as many pre-CC lines were infected with IAV (PR8) at 8--12 wk of age and assayed for p.i. weight loss via daily weights, with phenotypes collected including starting weight (D0) and weight at D4 p.i. ([Figure 2B](#fig2){ref-type="fig"}). This study did not include mock-infected mice. ### CC-RIX study: {#s6} In total, 1402 female mice were bred from 105 F1 crosses of CC strains (*i.e.*, 105 CC-RIX lines) ([Figure 2C](#fig2){ref-type="fig"} and Supplemental Material Figure S2 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)), as part of an ongoing QTL mapping study. These mice were infected at 8--12 wk of age with 5000 PFU IAV (A/California/04/2009; short name CA04), a human 2009 pandemic H1N1 isolate ([@bib37]), and phenotypes were collected, including starting weight (D0) and weight at D7 p.i. CC-RIX were bred under similar conditions to diallel mice. This experiment, whose broader analysis is still ongoing, included both flu-infected and mock-treated mice. However, since the design did not match these to the same exacting degree as the diallel, with mock controls missing entirely for some batch/line combinations, in the current study we consider data from the infected mice only. CC animals used to generate CC-RIX lines were purchased from the Systems Genetics Core at UNC-CH; information about CC strains available for distribution is found at <http://csbio.unc.edu/CCstatus/index.py?run=AvailableLines> ([@bib56]). Statistical Models and Methods {#s7} ============================== Our statistical analysis of heritable effects in the diallel (hereafter, diallel effects) relies heavily on the BayesDiallel model and approach described by [@bib44] and [@bib14]. BayesDiallel was originally proposed in [@bib44] for diallel analysis of routine, single outcome phenotypes, describing how the mean value of those phenotypes was shifted by changes in parentage and sex. Although in some ways the method was built upon a canon of existing diallel literature (*e.g.*, references in [@bib8]), including more recent work that used random effects ([@bib102]; [@bib88]) and Bayesian hierarchical modeling ([@bib27]), in other ways it represents a new parameterization and a generalization of many earlier methods \[see [@bib44] for explicit connections to those methods\]. In [@bib14], we extended BayesDiallel to treatment-response phenotypes, in particular, to when the modeled outcome is the phenotypic difference between placebo and treated matched pairs; the model in this case describes a causal effect modification, or, in a slight abuse of terminology, a gene-by-treatment (G × T) effect. Herein, that treatment-response approach is extended further to our more complex matching regime of quartets rather than pairs, and with a different imputation procedure to deal with quartets that are incomplete. This section begins by reviewing the BayesDiallel model for single outcome phenotypes. This is used not only to analyze our primary baseline phenotype, body weight at day 0 (D0 weight), but is also foundational for our subsequent analyses. Then we introduce our definition of infection response based on matched quartets, which gives rise to treatment responses defined for each of four time points (D1, D2, D3, and D4 p.i.), and describe how they are modeled using BayesDiallel. The analysis is then modified further to estimate the impact of haplotype state at the resistance locus *Mx1*, and we describe how the interaction of haplotype pairs at this locus is examined by estimating relative degrees of haplotype additivity and dominance. Finally, we describe an illustrative comparative analysis of the effect of the *Mx1* locus on IAV response in pre-CC and CC-RIX mice. Diallel model for single outcome phenotypes {#s8} ------------------------------------------- Diallel effects for single outcome phenotypes, that is, phenotypes measured as a single value per mouse, were modeled using the "fulls" model of BayesDiallel ([@bib44]; [@bib14]). BayesDiallel is a Bayesian linear mixed model that decomposes phenotypic variation into separate heritable components corresponding to additive genetics, dominance/inbred effects, parent-of-origin ("maternal"), epistasis, and all sex-specific versions thereof. It models the phenotype value $y_{i}$ of mouse *i* as$$y_{i} = \mu + \mathbf{c}_{i}^{\text{T}}\mathbf{\alpha} + {\sum\limits_{r = 1}^{R}u_{i}^{(r)}} + \mathbf{d}_{i}^{\text{T}}\mathbf{\beta} + \varepsilon_{i}\,,$$where *μ* is the intercept, and $\varepsilon_{i}$ is the residual error, normally distributed as $\varepsilon_{i} \sim \text{N}\left( {0,\sigma^{2}} \right),$ with variance $\sigma^{2}.$ The $\mathbf{c}_{i}^{\text{T}}\mathbf{\alpha}$ term represents the contribution of an arbitrary set of user-specified fixed effect covariates, with predictors encoded in vector $\mathbf{c}_{i}$ and fixed effects $\mathbf{\alpha};$ the $\sum_{r = 1}^{R}u_{i}^{(r)}$ term represents the contribution of an arbitrary set of *R* user-defined random effect covariates, which for single outcome phenotypes in this study always includes an effect of experimental batch; and the $\mathbf{d}_{i}^{\text{T}}\mathbf{\beta}$ term represents the contribution of heritable components of the diallel, written as a linear combination of the diallel effects vector $\mathbf{\beta}$ and diallel category vector $\mathbf{d}_{i}.$ Here $\mathbf{d}_{i}$ is shorthand for $\mathbf{d}_{{\{{jks}\}}{\lbrack i\rbrack}},$ where $\left\{ {jks} \right\}\left\lbrack i \right\rbrack$ denotes *i*'s diallel category, that is, its unique combination of mother strain *j*, father strain *k*, and sex *s*. The diallel category vector $\mathbf{d}_{\{{jks}\}}$ is defined with the diallel effects $\mathbf{\beta}$ so as to give the linear combination shown in Equation 2, where $a_{j}$ is the additive effect of strain *j* (*e.g.*, the additive effect parameter $a_{\text{AJ}}$ is the expected increase in phenotype on adding one haploid genome of strain AJ); $m_{j}$ is an additional increase in phenotype induced by strain *j* being the mother (parent-of-origin effect); indicator $I_{\{ X\}}$ is 1 if *X* is true and 0 otherwise; $\beta_{\text{inbred}}$ is the overall effect of being inbred; $b_{j}$ is the additional effect of being inbred for strain *j*; $v_{jk}$ is the additional effect of combining strains *j* with *k* regardless of which is the mother (symmetric epistasis); indicator $S_{\{ X\}}$ is $1/2$ if *X* is true and $- {1/2}$ otherwise; $w_{jk}$ is a deviation from $v_{jk}$ induced by parent-of-origin (asymmetric epistasis); *φ* is the effect of being female rather than male; and $\varphi_{j}^{a}$ is the sex-specific deviation from additive effect $a_{j},$ with other superscripted *φ* terms (*e.g.*, $\varphi^{m}$) defined analogously. Each set of related variables, *e.g.*, the additive effects $a_{1},\ldots,a_{J}$ for *J* parents, is modeled as a group via a constrained normal distribution, that is, $a_{1},a_{2},\ldots,a_{J} \sim \text{marginally}\,\text{N}\left( {0,\tau_{a}^{2}} \right),$ but subject to ${\sum_{j}a_{j}} = 0,$ after [@bib14]. The variance of each group, *e.g.*, $\tau_{a}^{2},$ was modeled with a weak inverse gamma prior, $\tau_{a}^{- 2} \sim \chi^{2}\left( {\text{d}.\text{f}. = 0.2,\text{mean} = 0.2} \right),$ with this prior also used for the residual variance $\sigma^{2}.$ The prior for fixed effects, *e.g.*, *μ*, is set to a vague normal distribution, $\mu \sim \text{N}\left( {0,10^{3}} \right).$ A summary of the diallel effects parameters is given in [Table 1](#t1){ref-type="table"}. Model fitting proceeded using Markov chain Monte Carlo (MCMC) via Gibbs sampling (algorithm in [@bib44]), with results based on samples from $12.5 \times 10^{6}$ MCMC iterations (five chains of length 2500, after 500 iterations burn-in). See also the later section *Reporting BayesDiallel results: highest posterior density*, *model inclusion probability*, *variance projection*, *and treatment response variance projections*. d { j k s } T β = a j \+ a k ︸ additive \+ m j − m k ︸ maternal \+ I { j = k } ( β inbred \+ b j ) ︸ inbred penalty \+ I { j ≠ k } ( v j k \+ S { j \< k } ⋅ w j k ) ︸ epistasis \+ S { s = female } ⋅ \[ φ \+ φ j a \+ φ k a ︸ sex × additive \+ φ j m − φ k m ︸ sex × maternal \+ I { j = k } ( φ inbred \+ φ j b ) ︸ sex × inbred penalty \+ I { j ≠ k } ( φ j k v \+ S { j \< k } ⋅ φ j k w ) ︸ sex × epistasis \] , ###### Model parameters, random and fixed (overall), from Equations 1, 2, and 5 Parameter Color Description Type Levels ----------------------------- -------------- --------------------------------------------------------------------- -------- --------------------------------------------- *μ* (or *θ*) Overall mean (or overall infection response) Fixed 1 *α* D0 body weight Fixed 1 $u^{(\text{batch})}$ Experimental batch Random 44 or 52[*^a^*](#t1n1){ref-type="table-fn"} $u^{({Mx1\,\text{diplo}})}$ *Mx1* diplotype Random 6 $a_{j}$ Blue Strain-specific additive Random 8 $m_{j}$ Green Strain-specific maternal (parent-of-origin) Random 8 $\beta_{\text{inbred}}$ Red Overall inbred penalty Fixed 1 $b_{j}$ Orange Strain-specific inbred penalty Random 8 $v_{jk}$ Purple Strain pair-specific symmetric epistasis Random 28 $w_{jk}$ Brown Strain pair-specific asymmetric epistasis (parent-of-origin) Random 28 *φ* Gray Overall female Fixed 1 $\varphi_{j}^{a}$ Light blue Sex-by-strain-specific additive Random 8 $\varphi_{j}^{m}$ Light green Sex-by-strain-specific maternal (parent-of-origin) Random 8 $\varphi_{\text{inbred}}$ Pink Overall female inbred Fixed 1 $\varphi_{j}^{b}$ Light orange Sex-by-strain-specific inbred penalty Random 8 $\varphi_{jk}^{v}$ Lavender Sex-by-strain pair-specific symmetric epistasis Random 28 $\varphi_{jk}^{w}$ Tan Sex-by-strain pair-specific asymmetric epistasis (parent-of-origin) Random 28 Random effect levels for $u^{(\text{batch})}$ differ according to the number of experimental batches within each phenotype being modeled: 52 levels for D0, D1pct, D2pct, D1delta, D2delta; and 44 levels for D3pct, D4pct, D3delta, and D4delta. In the text, *h* is used to indicate the level of batch for a given individual or quartet. Modeling infection response as mock-corrected percent change in body weight p.i. {#s9} -------------------------------------------------------------------------------- A standard measure used to assess pathogenesis in IAV-infected mice is weight loss. Weight loss correlates with several host and viral factors, including viral load, immune response phenotypes, and lung histopathology ([@bib19]; [@bib43]); as such, it provides a simple, noninvasive measure of infection pathology that can be assessed for a large number of mice. We measured the percentage change in body weight relative to D0,$$\text{pct}_{i}^{\text{day}{\lbrack\text{group}\rbrack}} = 100 \times {\text{weight}_{i}^{\text{day}{\lbrack\text{group}\rbrack}}/\text{weight}_{i}^{\text{D}0{\lbrack\text{group}\rbrack}}},$$for mouse *i* on $\text{day} \in \left\{ {\text{D}1,\text{D}2,\text{D}3,\text{D}4} \right\}$ in $\text{group} \in \left\{ {\text{flu},\text{mock}} \right\},$ where, *e.g.*, $\text{weight}_{i}^{\text{D}4{\lbrack\text{flu}\rbrack}}$ and $\text{weight}_{i}^{\text{D}0{\lbrack\text{flu}\rbrack}}$ are the body weights for IAV-infected mouse *i* at D4 and at D0, respectively. These measures, which we describe as single outcome phenotypes, were analyzed using BayesDiallel as above ([Table 2](#t2){ref-type="table"}), but they were not the main focus of our study. Our main focus was a derived measure, IAV-infection response, defined next. ###### Models used for each analysis in this study Model Number Model[*^a^*](#t2n1){ref-type="table-fn"} Phenotype(s) Unit Variance Parameters -------------- ----------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------- ------------------------------------------ ------------- --------------------------------------- 1 $y_{i} = \mu + u_{i}^{(\text{batch})} +$ $\mathbf{d}_{i}^{\text{T}}\mathbf{\beta} + \varepsilon_{i}$ Pre[*^b^*](#t2n2){ref-type="table-fn"} Individuals 12[*^c^*](#t2n3){ref-type="table-fn"} 2 $y_{i} = \mu + u_{i}^{(\text{batch})} + c_{i}^{({\text{D}0})}\alpha +$ $\mathbf{d}_{i}^{\text{T}}\mathbf{\beta} + \varepsilon_{i}$ Post[*^d^*](#t2n4){ref-type="table-fn"} Individuals 12[*^c^*](#t2n3){ref-type="table-fn"} 3 ${delta}_{q} = \theta + u_{q}^{(\text{batch})} + c_{q}^{({\text{D}0})}\alpha +$ $\mathbf{d}_{q}^{\text{T}}\mathbf{\beta} + \varepsilon_{q}$ delta[*^e^*](#t2n5){ref-type="table-fn"} Quartets 12[*^c^*](#t2n3){ref-type="table-fn"} 4 ${delta}_{q} = \theta + u_{q}^{(\text{batch})} + c_{q}^{({\text{D}0})}\alpha + u_{q}^{({Mx1\,\text{diplo}})} +$ $\mathbf{d}_{q}^{\text{T}}\mathbf{\beta} + \varepsilon_{q}$ delta[*^e^*](#t2n5){ref-type="table-fn"} Quartets 13[*^f^*](#t2n6){ref-type="table-fn"} See study design in [Figure 1](#fig1){ref-type="fig"} for overview of analyses. See [Table 1](#t1){ref-type="table"} and *Statistical Models and Methods* for parameter and phenotype definitions. D0 \[all\]. This count includes $\tau_{\text{batch}}^{2},$ {$\tau_{a}^{2},$ $\tau_{m}^{2},$ $\tau_{b}^{2},$ $\tau_{v}^{2},$ $\tau_{w}^{2}$}, {$\tau_{\varphi^{a}}^{2},$ $\tau_{\varphi^{m}}^{2},$ $\tau_{\varphi^{b}}^{2},$ $\tau_{\varphi^{v}}^{2},$ $\tau_{\varphi^{w}}^{2}$}, and $\sigma^{2}.$ D1pct, D2pct, D3pct, D4pct \[mock\] and D1pct, D2pct, D3pct, D4pct \[flu\]. D1delta, D2delta, D3delta, and D4delta. This count includes $\tau_{Mx1\,\text{diplo}}^{2}$ and parameters in *c*. In defining IAV-infection response, we note that from a causal inference perspective (described more fully in Appendix A) weight loss in an IAV-infected mouse (*e.g.*, $\text{pct}_{i}^{\text{D}4{\lbrack\text{flu}\rbrack}}$) reflects two confounded processes: weight loss due to IAV-induced pathogenesis, and weight loss due to other aspects of the experimental procedure. To obtain an unconfounded estimate of weight loss due to IAV-induced pathogenesis alone, we defined IAV-infection response as the difference between weight loss in mice subject to infection by IAV and those subject to mock. Specifically, since in our experimental design we match one mock mouse to three infected---this reflecting our expectation that phenotypes from infected mice will be more variable and will thus need more replicates for comparable precision---infection response was defined in terms of "matched quartets," $q = 1,\ldots,Q,$ where each matched quartet *q* comprised four mice of the same diallel category from the same experimental batch, with the first three mice, $q\left\lbrack 1 \right\rbrack,$ $q\left\lbrack 2 \right\rbrack$ and $q\left\lbrack 3 \right\rbrack,$ being IAV infected and the last mouse, $q\left\lbrack 4 \right\rbrack,$ receiving mock treatment. Infection response at a given day for quartet *q* was thus defined as a delta,$$\text{delta}_{q}^{\text{day}} = \frac{1}{3}{\sum\limits_{f = 1}^{3}{\text{pct}_{q{\lbrack f\rbrack}}^{\text{day}\,{\lbrack\text{flu}\rbrack}} - \text{pct}_{q{\lbrack 4\rbrack}}^{\text{day}\,{\lbrack\text{mock}\rbrack}}}}\,,$$following the more general definition in Equation A2 in Appendix A. Diallel effects on infection response were then modeled using BayesDiallel in manner analogous to the single outcome case in Equation 1, as$$\text{delta}_{q}^{\text{day}} = \theta + \mathbf{c}_{q}^{\text{T}}\mathbf{\alpha} + {\sum\limits_{r = 1}^{R}{u_{q}^{(r)} + \mathbf{d}_{q}^{\text{T}}\mathbf{\beta} + \varepsilon_{q}}}\,,$$where now the unit of observation is the matched quartet *q* rather than the individual *i* and where, for example, $\mathbf{d}_{q}$ is shorthand for $\mathbf{d}_{{\{{jk,s}\}}\,{\lbrack q\rbrack}},$ the diallel category appropriate for *q*. The shift to modeling treatment response does, however, change how the parameters are interpreted. The intercept in the above formula, relabeled as *θ*, now acquires a special meaning, representing an overall causal effect due to infection, and the diallel effects in $\mathbf{\beta}$ now describe how that causal effect is modified by parentage, sex, and their interaction. For example, the additive effect parameter $a_{\text{AJ}}$ is the expected increase in infection response on adding one haploid genome of strain AJ. Regarding covariates, as for the single outcome phenotypes, this model included a random effect of batch and, to reduce potential dependence between the delta's and baseline body weight, we also included a fixed effect covariate for the quartet mean D0 weight (*i.e.*, $\text{D}0_{q} = {{\sum_{f = 1}^{4}\text{weight}_{q{\lbrack f\rbrack}}^{\text{D}0}}/4}$) in $\mathbf{c}_{q}$ ([Table 2](#t2){ref-type="table"}). Although our experimental design stipulated even multiples of four mice per diallel category, practical constraints on animal breeding and availability meant that in some cases this number was three or five, such that some quartets had either missing infecteds or surplus mocks. To ensure the definition of delta in Equation 4 remained consistent, and in particular that delta's from different quartets had comparable precision, the diallel analysis was performed on $M = 1000$ imputed versions of the data, with each imputed data set being comprised of exact quartets in which missing phenotypes had been filled using stochastic regression imputation and surplus mocks had been (randomly) deleted (details in Appendix B). On each imputed data set we collected 125 MCMC samples from 12,500 total time steps (*i.e.*, by recording values at every 100th time step); results were based on the aggregate of these samples from the *M* imputed data sets (*i.e.*, on 125,000 MCMC samples in total). Reporting BayesDiallel results: highest posterior density, model inclusion probability, variance projection, and treatment response variance projections {#s10} -------------------------------------------------------------------------------------------------------------------------------------------------------- Point and interval estimates of individual diallel effects, *e.g.*, additive effect $a_{\text{AJ}},$ are reported as posterior means and 95% highest posterior density (HPD) intervals. The overall contribution of a particular inheritance group is reported in two ways: as a variance projection (VarP), *e.g.*, VarP\[a\] for the contribution of additive effects to a phenotype or treatment response VarPs (TreVarPs), *e.g.*, TreVarP\[a\] for the contribution of additive effects to an infection response; and as a model inclusion probability (MIP), *e.g.*, MIP\[a\] for the probability of additive effects being included in the model. The VarP is a heritability-like measure that predicts how much of the total phenotypic sum of squares would be explained by each component in a new, completely balanced diallel. Unlike traditional heritability, it is calculated based on the effects, $\mathbf{\beta},$ rather than the variance components, $\tau_{a}^{2},\ldots,\tau_{w}^{2},\sigma^{2},$ and as such benefits not only from greater interpretability but also from the stability and accuracy provided by hierarchical shrinkage (as detailed in [@bib14]). Since the VarP is a function of the posterior predictive distribution and calculated at each iteration of the MCMC, it is reported via Bayesian posterior summaries, specifically, the posterior median and the 95% HPD interval. The VarPs for infection response phenotypes are, following [@bib14], given the special name of TreVarPs to acknowledge their more delicate interpretation. The MIP reflects a different type of inference: rather than being a function of the parameters estimated in the full, sexed BayesDiallel model of Equations 1 and 2, it describes the results of model selection, that is, an assessment of which diallel categories could be excluded without a substantial loss in fit. As in [@bib14], we use the exclusionary Gibbs group sampler of [@bib44]. Each diallel category is set to have a prior inclusion probability of 0.5, reflecting a prior opinion that inclusion and exclusion are equally likely. This prior is then updated by the phenotype data and the model selection procedure to give a (posterior) MIP. MIPs are interpreted following the conventions in [@bib14]: MIPs in the range (0.25, 0.75) indicate that the data does not provide sufficient evidence to make an informed decision about exclusion or inclusion; MIPs within (0.05, 0.25\] or \[0.75, 0.95) represent positive evidence for exclusion or inclusion respectively; (0.01, 0.05\] or \[0.95, 0.99) represent strong evidence; and \[0, 0.01\] or \[0.99, 1\] represent strong to decisive evidence. These conventions are based on those proposed by [@bib40] for Bayes factors, which are connected to MIPs by the relation$$\text{Bayes}\,\text{factor} = \frac{\text{MIP}}{1 - \text{MIP}} \times \frac{1 - \text{MIP}_{0}}{\text{MIP}_{0}},$$where $\text{MIP}_{0}$ is the prior inclusion probability, and where the above simplifies to $\text{MIP}/\left( {1 - \text{MIP}} \right)$ in our case of $\text{MIP}_{0} = 0.5.$ Estimating Mx1 effects in the diallel {#s11} ------------------------------------- The critical host resistance factor (*Mx1*) has been shown to drive IAV resistance in the CC founder strains and has been mapped in the pre-CC ([@bib19]). *Mx1* was previously described as having three major, naturally occurring functional classes of resistance to influenza H1N1 arising from the subspecies *Mus musculus domesticus* (hereafter, *dom*; members include AJ, B6, 129, NOD, and WSB), *M. musculus castaneus* (*cast*; CAST), and *M. musculus musculus* (*mus*; PWK and NZO), of which *domesticus* is considered to be null whereas *musculus* and *castaneus* are protective. (Note that *domesticus Mx1* in the CC founder strains is comprised of two unique null alleles, and that the subspecific *Mx1* alleles observed in the CC may not be representative of the those segregating in the wild.) To estimate the contribution of *Mx1* haplotypes as discernible in the diallel, and thereby also estimate the extent of heritable effects that remain after *Mx1* is controlled for, we define the following haplotype combinations (diplotypes) as six levels of the random effect, $u^{({Mx1\,\text{diplo}})}:$ {*dom* × *dom*}, {*dom* × *cast*}, {*cast* × *cast*}, {*cast* × *mus*}, {*mus* × *dom*}, and {*mus* × *mus*}; we then repeat our diallel analysis with this effect included (model 4 in [Table 2](#t2){ref-type="table"}). ### Estimating a dominance index for Mx1 alleles: {#s12} Dominance is typically defined in the context of bialleles, but since in this population *Mx1* has a multiallelic series, we define dominance of *Mx1* between allele pairs. Following [@bib38], which is built on the work of [@bib94], we define the "dominance index" for a wild-type (wt) against a mutant (mut) allele as$$\mathcal{D}^{({\text{wt};\,\text{mut}})} = \frac{u^{(\text{wt~wt})} - u^{(\text{wt~mut})}}{u^{(\text{wt~wt})} - u^{(\text{mut~mut})}},$$where values for $\mathcal{D}$ are close to −0.5 when the effect of the wt is overdominant to the mut (the effect of the mut is underrecessive), 0 when the effect of the wt is completely dominant to the mut (the effect of the mut is recessive), close to 0.5 when the effect of the wt is additive (not dominant, or incompletely dominant) to the mut, close to 1 when the effect of the wt is recessive (the effect of the mut is dominant), and close to 1.5 when the effect is underrecessive (the effect of the mut is overdominant). Overdominance is given by values of $\mathcal{D}$ that are much less than zero and underdominance by values that are much greater than one. This definition is used to define dominance indices $u^{({cast;\, dom})}$ and $u^{({mus;\, dom})},$ describing the degree of dominance of the protective alleles *castaneus* and *musculus*, respectively, against the null allelle *domesticus*. To assess the degree to which *castaneus* and *musculus* differ in their relation to *domesticus*, we further define a "dominance difference index,"$$\mathcal{D}\mathcal{D}^{({mus - cast;\, dom})} = \mathcal{D}^{({mus;\, dom})} - \mathcal{D}^{({cast;\, dom})}\,,$$where negative values indicate that *musculus* has more of a dominance-based relationship to *domesticus* than does *castaneus*, positive values indicate the converse, and zero indicates that the relationships of *castaneus* and *musculus* to *domesticus* show dominance equally. When the BayesDiallel model includes *Mx1* effects, the aforementioned dominance index and dominance difference index are both functionals of the posterior; posterior samples of these indices were therefore obtained by simply applying Equations 6 and 7 to the sampled *Mx1* effects at each time step of the MCMC. The [@bib38] dominance index is a simple reparameterization of the degree of dominance parameter, $a_{\text{CR}},$ defined by [@bib13] and used by [@bib23]. In the Comstock--Robinson model, the mean-centered phenotypes are coded as (translating from our model above): $u^{({\text{wt};\,\text{wt}})} = w,$ $u^{({\text{wt};\,\text{mut}})} = aw,$ and $u^{({\text{mut};\,\text{mut}})} = - w.$ This gives the relation $\mathcal{D}^{({\text{wt};\,\text{mut}})} = {\left( {1 - a_{\text{CR}}} \right)/2}$ or equivalently, $a_{\text{CR}} = 1 - 2\mathcal{D}^{({\text{wt};\,\text{mut}})}.$ This alternate dominance parameterization is explored further using BayesDiallel in [@bib89]. Estimating haplotype effects at the Mx1 locus in the pre-CC and CC-RIX {#s13} ---------------------------------------------------------------------- The additive effect parameters estimated in the diallel do not precisely distinguish the effects at the *Mx1* locus because they are confounded with any potential genome-wide effects that follow the same pattern of strain classification. An unconfounded estimate of haplotype effects at *Mx1* requires a population in which the remainder of the genome is randomized, *e.g.*, by recombination. To this end, we make use of two related data sets on IAV-induced weight loss in two CC-derived MPPs: IAV (PR8) infection in the pre-CC and IAV (CA04) infection in a set of CC-RIX lines. These two studies, described in more detail below, were in other respects less rigorous than our diallel: the experimental measurement of the infection response was based on infected mice only with no mocks in the pre-CC, and although mocks were collected in the CC-RIX, their relative sparsity (200--300 mocks to \>1400 infecteds) complicates analysis based on matching alternate treatment groups; the experimental batching was subject to a less exacting degree of randomization across genetically distinct categories; the available combinations of *Mx1* diplotypes are limited mostly to homozygotes in the pre-CC, and incompletely and unevenly sampled in the CC-RIX; and the *Mx1* diplotype state for each line is known only probabilistically, having been inferred by hidden Markov models (HMMs) applied to genotyping data. Nonetheless, if effects at the *Mx1* locus were largely independent of those elsewhere in the genome, we might expect that *Mx1* effects in the pre-CC and CC-RIX would be broadly consistent with those in the diallel. Estimation of haplotype effects at the *Mx1* locus was performed using the Diploffect model ([@bib100]), a Bayesian hierarchical model that estimates effects of diplotype substitutions at a specified QTL when the diplotype states themselves are known only probabilistically. The effects estimated by Diploffect are analogous to those estimated by BayesDiallel: phenotype $y_{i}$ of mouse *i* is modeled as$$y_{i} = \mu + \mathbf{c}_{i}^{\text{T}}\mathbf{\alpha} + {\sum\limits_{r = 1}^{R}{u_{i}^{(r)} + \text{dip}_{i}^{\text{T}}\mathbf{\beta} + \varepsilon_{i}}}\,,$$where $\text{dip}_{i}$ is a vector representing the diplotype state of mouse *i* at the QTL and is shorthand for $\text{dip}_{{\{{jk}\}}{\lbrack i\rbrack}},$ where $\left\{ {jk} \right\}\left\lbrack i \right\rbrack$ denotes *i*'s diplotype state comprised of haplotypes from CC founder strains *j* and *k*, $\mathbf{\beta}$ are the corresponding effects, and all other variables are as in Equation 1. The diplotype vector $\text{dip}_{\{{jk}\}}$ is defined with $\mathbf{\beta}$ so as to give the linear predictor$$\text{dip}_{\{{jk}\}}^{\text{T}}\mathbf{\beta} = a_{j} + a_{k} + I_{\{{j \neq k}\}}\gamma_{jk}\,,$$where $a_{j}$ and $a_{k}$ are additive (haplotype) effects modeled as $a_{j} \sim \text{N}\left( {0,\tau_{\text{add}}^{2}} \right),$ broadly equivalent to the additive effects in BayesDiallel's Equation 2; and $\gamma_{jk} \sim \text{N}\left( {0,\tau_{\text{dom}}^{2}} \right)$ are dominance deviations, which are the converse to BayesDiallel's inbred parameters. Dominance deviations are expected to be poorly informed when heterozygotes are sparsely represented, as in the CC-RIX and in particular the largely inbred pre-CC, but are nonetheless included to stabilize inference of additive effects. For numerical stability, phenotypes were first centered and scaled to unit variance, and variance parameters ($\sigma^{2}$ or $\tau_{\text{effect}}^{2},$ where effect is add, dom, or $r \in R$) were given mildly informative priors of the form $\tau_{\text{effect}}^{- 2} \sim \text{Ga}\left( {1,1} \right).$ Estimation proceeded by importance sampling (the DF.IS and DF.IS.kinship methods in [@bib100]) using integrated nested Laplace approximations (INLA; [@bib53]), with 100 importance samples taken, and parameter estimates for additive effects are reported as posterior means, posterior medians, and HPD intervals. ### Pre-CC study: {#s14} In the study of [@bib19], IAV-infection response was measured on 155 mice from as many pre-CC lines as weight loss following infection with IAV (PR8 variant, as for the diallel). QTL mapping of D4 p.i. weight loss, equivalent to $\text{pct}_{i}^{\text{D}4}$ in the diallel study, identified a QTL, *HrI1*, containing the *Mx1* gene, with peak marker JAX00072951 (chr16:98,148,641; Mouse Diversity Array of [@bib98]). We estimated haplotype effects at this peak marker using Diploffect ([@bib100]), applied to the phenotype and the original HMM probabilities of [@bib19], with the model including a fixed effect covariate for D0 weight. ### CC-RIX study: {#s15} For the CC-RIX study of infection response to IAV (CA04 strain), we calculated weight loss values for all 1402 infected mice at D7 p.i. (analogous to a $\text{pct}_{i}^{\text{D}7}$ measure), and for all 105 CC-RIX lines obtained diplotype probabilities at marker UNC27478095 (16:97,591,482; MegaMUGA array, described in [@bib57]) from the Inbred Strain Variant database (ISVdb; [@bib63]). Haplotype effects were then estimated by Diploffect applied to debatched CC-RIX line means as follows. First, we fit a linear mixed model (by REML using the R package lme4 of [@bib2]) to the individual-level phenotypes (*n* = 1402) with fixed effects of D0 weight and laboratory (two levels), and random effects of mating (107 levels: 105 RIXs + 2 additional levels distinguishing minor breeding differences, when CC010 and CC042 strains were rederived from breeder females into a new facility) and infection date (59 levels). The residuals of this model were then averaged over the $n_{i}$ mice of each CC-RIX line *i* and used as the response $y_{i}$ in Equation 8 with precision weighting $\varepsilon_{i} \sim \text{N}\left( {{0,\sigma^{2}}/n_{i}} \right)$ and a between-line polygenic random effect $\mathbf{u} \sim \text{N}\left( {0,\,\mathbf{G}\tau_{G}^{2}} \right),$ where the $105 \times 105$ genetic relationship matrix $\mathbf{G}$ was calculated between all CC-RIX pairs based on the founder haplotype probabilities (dosages) at each locus, according to the method described in [@bib24]. Data availability {#s16} ----------------- Analyses were conducted in the statistical programming language R ([@bib67]). In addition to R packages cited above, we used the packages BayesDiallel ([@bib44]) and Diploffect.INLA ([@bib100]). The data, analysis software, and scripts are available on the flu-diallel repository on GitHub at <https://github.com/mauriziopaul/flu-diallel>. A static version is posted as a public, open-access Zenodo repository at <http://dx.doi.org/10.5281/zenodo.293015>. Phenotype data from the diallel and CC-RIX animals used in this study will be available on the Mouse Phenome Database ([@bib28]) at <https://phenome.jax.org> with persistent identifier RRID:SCR_003212. [File S1](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS1.pdf) contains an account of the supplemental files which can be used to reproduce our analysis. [File S2](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS2.gz) contains the software packages used for this analysis. [File S3](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS3.csv) contains the diallel data file, and [File S4](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS4.gz), [File S5](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS5.zip), and [File S6](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS6.gz) contain the data analysis files required for analyzing the diallel, pre-CC, and CC-RIX, respectively. [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip) contains supplemental figures, tables, and an algorithm. After unzipping, the files FluDiData.csv, Flu-pre-CC-data.csv, and Flu-CC-RIX-data.csv contain raw phenotypes, cross (or line, strain), and mouse ID information from the three mouse populations used in this study. The script files MIMQ\*.sh are used in bash to call R scripts to run the BayesDiallel analysis on diallel phenotypes. The script files main_analysis\*.R are used with Diploffect to run Diploffect analysis on the pre-CC and CC-RIX phenotypes. Additional \*.RData, \*.pl, \*.alleles, and \*.csv files are uploaded which contain settings, genotypes, and founder haplotype probabilities used by the scripts. Results {#s17} ======= Mice from the eight inbred founder strains of the CC were used to generate a near-complete $8 \times 8$ diallel. This study used offspring (*n* = 1043) of both sexes (519 females and 524 males) representing 62 of the 64 crosses (Figure S1 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)), including all inbred combinations (*n* = 129) and all F1 hybrids (*n* = 914) except NZO × CAST and NZO × PWK. Within each diallel category---defined as the combination of sex and (reciprocal) parentage---and in each experimental batch, mice were randomly assigned at weaning to infection or mock groups in a ratio of 3:1; complete sets of three infected with one mock were described as matched quartets. Mice in the infected group were inoculated with IAV PR8, and in the mock group with PBS. For each mouse, body weight was measured prior to infection (D0 or baseline weight), and at days 1--4 p.i. (D1, D2, D3, D4). D0 weight is reported in grams whereas p.i. weight is hereafter reported as a percentage of D0 weight, *e.g.*, D4pct. Not all mice survived the protocol: one infected mouse died after D3 weights were taken and one mouse died from anesthesia on D0. F1 hybrids of the CC founders show a wide range of phenotypic outcomes {#s18} ---------------------------------------------------------------------- The CC founders include five strains we have previously characterized as susceptible to IAV-induced pathology (AJ, B6, 129, NOD, and WSB), two strains as resistant (NZO and PWK), and one (CAST) that exhibits a distinct intermediate weight loss phenotype ([@bib19]). Results for the inbred founders measured in our diallel replicate those earlier findings, and the p.i. weight loss among the infected F1 hybrids spanned the range of phenotypes observed in the founders (Figure S3 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)), consistent with the notion of IAV-induced weight loss being a complex trait with contributions from multiple loci. Diallel effects on baseline mouse weight strongly replicate previous CC founder diallel studies {#s19} ----------------------------------------------------------------------------------------------- The effects of parentage and sex on D0 weight were estimated using BayesDiallel. Described further in *Statistical Models and Methods*, BayesDiallel decomposes the heritable effects observable in the diallel into 160 parameters (diallel effects) grouped into 13 distinct heritability classes. In sketch form, it models the average phenotype of mice of sex *s* bred from mother of strain *j* and father of strain *k* as$$\text{ave}.\text{phenotype}_{jks} = \begin{matrix} {\text{overall}\,\text{mean}} \\ {\&\,{covariates}} \\ \end{matrix} + \underset{\text{diallel}\,\text{effects}}{\underset{︸}{a_{j} + a_{k} + \text{inbred}_{j} + \text{other}_{jks}}},$$where covariates always includes experimental batch, $a_{j}$ and $a_{k}$ are the additive effects of the two parents, $\text{inbred}_{j}$ is an additional effect included only when $j = k$, and $\text{other}_{jks}$ models the effects of further nuances of sex and parentage as deviations from this base model (listed in *Statistical Models and Methods* and [Table 1](#t1){ref-type="table"}). Diallel effects estimated for D0 weight are reported in Figure S6A in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip) as 95% HPD intervals for each parameter, and two summary measures, VarPs and MIPs, for each of the 13 heritability classes are given in Figure S6, B and C, in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip). Briefly, VarPs (Figure S6C in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)) report the contribution of the effect group as the proportion of the total phenotypic variance, whereas MIPs (Figure S6B in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)) assess the strength of support for whether an effect group should be included at all, with probabilities near 1 providing stronger support for inclusion, probabilities near 0 supporting exclusion, and probabilities near 0.5 reflecting a lack of information either way. The pattern of effects for D0 weight was strikingly similar to that seen for baseline body weight in two previous diallels of the CC founders ([@bib44]; [@bib14]), despite those earlier studies being independent experiments with no particular attempt made to align experimental protocols, and included substantial additive effects, strain-specific parent-of-origin effects, signals of epistasis, and sex-specific versions thereof. For example, we largely replicated the pattern of inbred, additive, and maternal effects observed in both [@bib44] and [@bib14], and also found a higher-order, sex-specific PWK × CAST symmetric epistatic effect in [@bib44]. We also observed some new epistatic and sex-specific epistatic effects largely due to increased power from a larger sample size. Diallel effects on IAV-infection response {#s20} ----------------------------------------- Infection response was defined as the percentage change in body weight induced by IAV infection, with more negative values indicating more severe pathology. This was calculated at each time point---D1, D2, D3, and D4 p.i.---as the difference between matched infected and mock mice, yielding a single infection response number (a delta, *e.g.*, D4delta) for each matched quartet (three infected mice and one mock). The effects of parentage and sex on infection response were then analyzed for each time point separately using BayesDiallel as above, with an additional covariate of D0 weight (see *Statistical Models and Methods* for details). Although results are provided in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip) for all time points, we will focus on those for D4 p.i. since this showed the greatest difference between infected and mock. IAV infection causes weight loss through D4 p.i., with greater susceptibility in females {#s21} ---------------------------------------------------------------------------------------- IAV infection in the diallel induced an overall mean change in body weight (*i.e.*, overall infection response *θ* in Equation 5 and [Table 1](#t1){ref-type="table"}) of −0.13% (95% HPD interval: −0.48, 0.22; MIP = 1) on D1 p.i., −0.83% (−1.33, −0.32%; MIP = 1) on D2 p.i., −5.60% (−6.47, −4.73%; MIP = 1) on D3, and −8.85% (−9.92, −7.78%; MIP = 1) on D4 (Table S3 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip); see also progression in [Figure 3](#fig3){ref-type="fig"}). Consistent with previous mouse studies of sex effects on infection ([@bib49]; [@bib71]), females given the same dose of virus as male mice had increased weight loss: a negative effect of female sex was estimated at all four time points p.i., gradually increasing in magnitude from −0.89% (−1.45, −0.36%) at D1 p.i. to −2.11% (−3.87, −0.30%) at D4 p.i. ([Figure 4](#fig4){ref-type="fig"}), suggesting that enhanced susceptibility in females may occur at least as early as D1 p.i. Although all mice received the same dose of virus regardless of starting body weight, heavier mice experienced a transient increase in percent weight loss at D2 p.i. compared with lighter mice: the D0 weight effect (*α* in Equation 5) on the infection response at D2 p.i. was −0.31% (−0.52, −0.09%), such that for every 10 g of starting weight beyond 0 g, an additional ∼3.1% weight was lost on D2; however, this effect disappeared by D3 p.i. No other significant effects of starting weight on IAV-induced weight loss were detected at other time points, indicating that heavier mice were infected at least as effectively as lighter mice, and that starting body weight does not in general confound our exploration of strain- and cross-specific effects. {#fig3} {#fig4} ### Diallel effects on infection response reflect mostly additive genetics, consistent with differences in Mx1 haplotype: {#s22} Infection response in our diallel was strongly driven by additive effects. On D3 p.i., enhanced susceptibility to weight loss in infected animals was affected the most by contributions from strain AJ, −2.17% (−3.72, −0.61%), and enhanced resistance from contributions of NZO, 2.54% (0.72, 4.27%), and PWK, 1.70% (0.12, 3.23%), strains. On D4 p.i., enhanced susceptibility was greatest from AJ, −2.77% (−4.66, −0.86%), and WSB, −3.09% (−5.01, −1.18%), with enhanced resistance greatest from NZO, 4.07% (1.95, 6.12%), and PWK, 4.06% (1.97, 6.08%) ([Figure 4A](#fig4){ref-type="fig"}). In terms of its additive effect, CAST was more resistant than the *Mx1*-null strains (AJ, B6, 129, NOD, and WSB) but about half as resistant as the *Mx1*-functional strains (NZO and PWK), consistent with it conferring intermediate protection in the heterozygote state. To summarize these effects: for each dose of AJ or WSB genomes inherited from a parent, ∼2--3% of additional starting body weight is lost p.i., indicating enhanced susceptibility compared with the overall mean weight loss; for each NZO and PWK genome inherited, ∼4% more of starting body weight is retained p.i., compared with the mean treatment effect, indicating enhanced resistance. Diallel effects explained over half of the total variance of infection response at D4, with a treatment-response VarP for all effect groups collectively of 57% (TReVarP\[all\] = 0.571; 0.418, 0.721). The variance explained by additive effects only, which is related to the narrow-sense heritability, was estimated as 34.8% (TReVarP\[a\] = 0.348; 0.190, 0.491), and also detected were potential additional contributions of epistasis (TReVarP\[v\] = 0.069; −0.001, 0.212) and maternal effects (TReVarP\[m\] = 0.020; 0.000, 0.059) ([Figure 4, B and D](#fig4){ref-type="fig"} and Table S1 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). ### Evidence for additive, inbred, epistatic, and parent-of-origin effects mounts as disease progresses: {#s23} The relevance of diallel effects to infection response became more marked with time (Figures S7--S10 and Table S3 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). At D1 and D2 p.i., model inclusion probabilities gave strong support only to an overall infection response, with no evidence of this effect being modified by sex or parentage (Figures S7 and S8 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). At D3 p.i., however, we found positive to strong evidence of additive (MIP\[a\] = 0.978), inbred (MIP\[b\] = 0.958), and asymmetric epistatic (MIP\[w\] = 0.820; *i.e.*, parent-of-origin epistatic) effects (Figure S9 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). By D4 p.i., support for additive (MIP\[a\] = 0.998) and inbred (MIP\[b\] = 0.999) effects had become decisive (see *Statistical Models and Methods* for MIP interpretation) and there was strong support for both symmetric epistatic (MIP\[v\] = 0.960) and asymmetric epistatic (MIP\[w\] = 0.966) effects (Figure S10 and Table S3 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). Modeling effects consistent with Mx1 haplotype {#s24} ---------------------------------------------- To help distinguish diallel effects that are consistent with the subspecies haplotype of the resistance factor *Mx1* (hereafter, *Mx1* effects), we incorporated the *Mx1* subtype explicitly into the model as a genotype covariate with three alleles, one for each subspecies branch: *domesticus* (AJ, B6, 129, NOD, and WSB), *castaneus* (CAST), and *musculus* (NZO and PWK). ### Mx1 effects are increasingly evident with disease progression and explain ∼40% of the diallel effects at D4 p.i.: {#s25} In keeping with the increased support seen for diallel effects over time, evidence for a nonzero *Mx1* effect increases from positive evidence of exclusion on D1 (MIP = 0.035), to no evidence for inclusion or exclusion on D2 (MIP = 0.552), to decisive evidence for inclusion on D3 (MIP = 1.000) and D4 (MIP = 1.000) (Figures S11--S14 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)); a comparable level of support for inclusion in the model was seen only for effects of overall treatment and batch. After controlling for *Mx1*, the variance explained by diallel effects at D4 was substantially reduced, from 57 to 33.8% (TReVarP\[all$\left| \text{Mx}1 \right.$\] = 0.338; 0.174, 0.537) ([Figure 4, C and E](#fig4){ref-type="fig"} and Table S2 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). This was consistent with *Mx1* accounting for ∼40% of the variance explained by the diallel, including most of the additive effects (mathematically the *Mx1* term models effects that compete with a subset of the additive and dominance diallel effects). ### Evidence for distinct additive and nonadditive effects of Mx1 functional groups: {#s26} After controlling for other diallel effects, the predicted weight loss over the course of 4 d varies in a manner consistent with *Mx1* allele combination ([Figure 5A](#fig5){ref-type="fig"}). We observed that, as expected, *domesticus* × *domesticus* crosses were predicted to have much more overall post-IAV-infection weight loss at D3 and D4 compared with all other crosses. Notably, the most protected group appeared to be the *domesticus* × *musculus* haplotype, at both D3 and D4 p.i., although the HPD intervals overlap with other *Mx1*-functional groups. The rank order of effects changes from D3 to D4 due to the dramatic slowing of weight loss in the *musculus* × *musculus* crosses from D3 to D4 compared with D2 to D3. {#fig5} Although we did not observe any strain- or pairwise-specific nonadditive effects in the diallel prior to inclusion of the *Mx1* random effect, we did observe a pattern of dominance in crosses between *musculus* and *domesticus*, even as there was a pattern of additivity in the crosses between *castaneus* and *domesticus* ([Figure 5B](#fig5){ref-type="fig"}). Whereas it might be expected that host alleles from *Mx1*-null strains should act in a recessive manner, this appears not to be the case for this phenotype and time point in crosses of *castaneus* with *domesticus*, such that the functional *Mx1* allele from CAST appears to operate in an additive manner. This further supports the previous observation that the CAST *Mx1* alleles differ from the *musculus Mx1* alleles in their protective host response to IAV ([@bib19]). ### Dominance and additivity of Mx1 alleles against the functional null: musculus is dominant, castaneus acts additively: {#s27} To better characterize how the *Mx1* effects on infection response exhibit aspects of genetic dominance *vs.* genetic additivity, we estimated for each functional *Mx1* allele a dominance index, after [@bib38]. This measures the distance between the expected phenotype of a homozygous functional allele, in our case *musculus* or *castaneus*, and the heterozygote formed with a null allele, in our case *domesticus*. On this scale, 0 denotes the functional allele being dominant to the null, 1 denotes it being recessive, and 0.5 indicates pure additivity (see *x*-axis scale in [Figure 6, A and B](#fig6){ref-type="fig"}, and more details in *Statistical Models and Methods*). {#fig6} The dominance indices of the two functional *Mx1* alleles, *musculus* and *castaneus*, were sharply different ([Figure 6, A and B](#fig6){ref-type="fig"}, and Table S8 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). We found that *musculus* against *domesticus* was −0.278 \[= posterior mode of $\mathcal{D}^{({mus;\, dom})};$ 80% HPD interval −2.547, 0.329\] at D3 and 0.068 at D4 (−0.568, 0.380), a clear signal of *musculus* exerting classical dominance over the functional null. In contrast, the dominance index of *castaneus* against *domesticus* was 0.421 (−0.534, 0.907) and 0.491 (−0.028, 0.836) for D3 and D4, consistent with *castaneus* and the functional null being codominant (*i.e.*, having an additive relationship). The difference of the two dominance indices, whose posterior distribution is shown in [Figure 6](#fig6){ref-type="fig"} for each time point, quantifies the distinction between *musculus* and *castaneus* more directly, putting the probability that *musculus* is more dominant than *castaneus* (*i.e.*, $P\left\lbrack {\mathcal{D}^{({mus;\, dom})} > \mathcal{D}^{({cast;\, dom})}} \right\rbrack$) at 83.6% for D3 and 86.6% for D4. Mx1 effects show consistent pattern in related MPPs for pre-CC and CC-RIX {#s28} ------------------------------------------------------------------------- We examined effects associated with the *Mx1* locus in two related recombinant CC populations, the pre-CC of [@bib19] and a set of CC-RIX lines first described here, and observed that the pattern of locus-specific strain haplotype effects was strikingly similar to that observed in our diallel ([Figure 7](#fig7){ref-type="fig"}). This suggests that the pattern of genome-wide additive effects in the diallel is largely driven by the effect of *Mx1* haplotypes in the founder strains. This similarity in pattern is consistent, even though the virus isolate and the peak weight loss time point differed in the CC-RIX population (CA04 human pandemic strain, D7 p.i.) compared with the diallel and pre-CC (PR8 mouse-adapted strain, D4 p.i.) (Table S4 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). In all three populations, NZO and PWK alleles provide the most resistance to IAV-induced weight loss, and CAST alleles are slightly less protective. In the pre-CC, effects of AJ, B6, 129, NOD, and WSB haplotypes are all approximately the same, and clearly separated from the additive effects of strains with functional *Mx1*. In the diallel and in the CC-RIX (at *Mx1*), however, AJ and WSB haplotypes are on average more susceptible than the B6 haplotype, and there is less separation between additive effects of CAST and those from *Mx1*-null strains. The proportion of variance in weight loss explained by *Mx1* was estimated as 0.5 (95% HPD interval: 0.43, 0.54) and 0.54 (0.42, 0.63) for pre-CC and CC-RIX mice, respectively (Figures S16 and S17 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). Note that an in-depth analysis of dominance indices for the *Mx1* locus was not possible in these populations because of the relatively sparse coverage of heterozygote diplotype states in the pre-CC and homozygous functional diplotype states in the CC. {ref-type="fig"}). (B) Additive strain haplotype effects at the *Mx1* locus for female pre-CC mice (*n* = 155) infected with IAV (PR8) at D4 p.i. (C) Additive strain haplotype effects at *Mx1* for female CC-RIX mice (*n* = 1402) infected with IAV (CA04) at D7 p.i. Estimates are shown as HPD intervals as described in [Figure 1](#fig1){ref-type="fig"}, with blue lines connecting posterior means. Parameter scales are given as additional IAV-induced weight loss per dose of strain in % of (A) D0, and (B and C) normalized effect size.](427f7){#fig7} Discussion {#s29} ========== We describe a general approach for investigating heritable effects on host susceptibility to virus-induced disease, in our case pathogenesis induced by IAV, using a diallel cross of the eight CC founder strains. The results from this diallel are informative not only in more clearly defining the genetic architecture of the host influenza response, but also prospectively: they anticipate sources of heritable variation likely to be present in the CC, the DO, and other derived experimental populations, and therefore provide a ready basis for the rational design of future studies. As an illustration of this, we demonstrate concordant effects of viral resistance locus *Mx1* across the CC founder diallel, pre-CC, and a set of CC-RIX lines. With regard specifically to IAV pathogenesis, our study sought to better understand host genetic effects on this outcome in terms of their (1) time-dependence, (2) consistency across related populations, and (3) conditionality---for example, dependence on interactions between alleles at the same locus (dominance, at *Mx1*) or at different loci (epistasis). Regarding time-dependence (1), we found that whereas the effect of being female rather than male is evident from D1, the effects of genetics appear later, becoming evident only on D3 and then increasing through D4 p.i. Regarding consistency (2), we found that the effects of the *Mx1* alleles seen previously in the CC founders remain stable across inbred, F1, and recombinant populations. Regarding conditionality (3), we found something unexpected: evidence that the two *Mx1* functional classes, *castaneus* (CAST) and *musculus* (NZO and PWK), which were previously characterized as being functional alleles, in fact behave differently when present in the heterozygous state with susceptible *Mx1* alleles from *domesticus* (AJ, B6, 129, NOD, and WSB). Specifically, the protection conferred by the presence of a *musculus Mx1* allele is the same regardless of whether it is in the homozygote state or paired as a heterozygote with the null *domesticus* allele; the *musculus* allele is therefore dominant to *domesticus*. But for the CAST allele, when paired in the same way with *domesticus*, its protection is weakened to an extent consistent with CAST and *domesticus* being codominant, that is, having an additive relationship. Level of resistance to IAV among different inbred mice is conditional on IAV subtype and strain {#s30} ----------------------------------------------------------------------------------------------- Differences in *Mx1* function have been identified between a variety of inbred mouse strains, including the CC founders ([@bib19]; [@bib96]; [@bib43]). Our results were largely consistent with those studies. Notably, in their examination of the CC founders with H3N2 infection, [@bib43] identified AJ and WSB strains as being most susceptible, and NZO and PWK as being most resistant, which agrees with our diallel additive effects. However, in contrast with our results showing partial protection against H1N1 IAV with CAST *Mx1*, which is consistent with our prior findings in the pre-CC ([@bib19]); they found CAST mice, grouping with AJ and WSB, to be highly susceptible. This difference could arise for at least two reasons. First, across the influenza field, even in identical RI panels ([@bib5]; [@bib58]), host genetic effects appear to be IAV subtype specific. Second, the effectiveness of *Mx1*'s antiviral activities can vary depending on IAV subtypes ([@bib70]; [@bib15]; [@bib103]; [@bib51]; [@bib90]). Differentiating these two possibilities, however, is beyond the scope of this work. Although the molecular differences in CAST *Mx1* that produce a deficient response in comparison with *musculus Mx1* have not been defined, some work has been done in inbred mice to better understand CAST (strain)-specific antiviral responses. To interpret what they saw as a unique antiviral deficiency of CAST mice, transcriptomic experiments by [@bib43] suggested enhanced susceptibility is due to leukocyte recruitment deficiency (relative to NZO and PWK) in the lung. In the CC founder study of [@bib96], several transcriptomic differences separated the CAST response to PR8 from the that of the other strains, including differential splicing of *Irak1* and lack of *Ifng* expression at D4 p.i., which was consistent with *Ifng* deficiency observed by [@bib16] leading to lethal monkeypox infection of CAST mice. Because these studies were completed in inbred CAST mice, the role of CAST *Mx1* is confounded with the genome-wide differences between CAST and the other CC founders. Thus, there are several challenges to understanding the unique IAV-resistance profile of CAST *Mx1* based on existing studies: (1) studies in inbred lines are unable to probe the overall or *Mx1*-specific dominance architecture due to a lack of heterozygosity, and (2) studies in nonrecombinant lines that identify a unique phenotype in CAST compared with other founders are unable to separate the effect of CAST *Mx1* from effects arising from the rest of the CAST genome. Our study in part circumvents these shortcomings by (1) additionally examining F1 hybrids; and (2) exploring the emerging phenotypes from an ongoing IAV-infection screen using CC-RIX, themselves F1s of RI strains. Complex additive effects patterns mask strong signals of dominance {#s31} ------------------------------------------------------------------ In our initial analysis, we found that most of the phenotypic variation explained in infection response is driven by additive genetics with no particular signal of dominance. However, when we explicitly modeled *Mx1* status, using a term that competes with a subset of the additive and dominance diallel effects, we found that the *Mx1* functional classes act in a manner consistent with a strong dominance pattern for *musculus Mx1* ([Figure 5](#fig5){ref-type="fig"}). It seems striking that such a pattern of dominance could be underlying an apparently heavily additive effect signal. Identifying dominance requires a good basis for comparing inbreds with hybrids. However, since the diallel is mainly composed of F1 hybrids with relatively few (8 *vs.* 54) inbreds, this basis for comparison is often weak. The BayesDiallel model handles this by considering the hybrid state as the baseline and treats the inbred state as the exception (a deviation) relevant to a minority of categories, as discussed further in [@bib44]. Inferred dominance effects are therefore vague because the data that informs them is sparse, and low estimates of dominance variance comes from absence of information rather than from information about the absence of an effect. Nonetheless, greater precision was available when considering dominance of substrain-specific *Mx1* because dominance information was pooled across multiple strains and strain pairs. The fact that the proportion of estimated additive *vs.* nonadditive variance is influenced by model parameterization motivates careful consideration of both study design and analysis. As [@bib35] have recently described, model parameterizations can have critical effects on the detection of nonadditivity, with the same data strongly supporting evidence for mostly additive or mostly nonadditive effects, depending on the model. Related issues have been described at the locus level by [@bib73], who showed that when applying penalized regression to multi-SNP fine-mapping in GWAS, genotype parameterization interacts with how priors/penalties are assigned and can make biallelic dominance hard to identify in some cases. Yet even when dominance is not of interest *per se*, failure to accommodate it can disrupt estimation of additivity: in the pre-CC QTL mapping study of [@bib65], dominance signals arising from residual heterozygosity disrupted detection of an additive QTL for basal levels of CD23 (encoded by *Fcer*2); this was resolved by treating heterozygote diplotypes, whose occurrence was too sparse to be modeled, as inherently noisier via downweighting. Antiviral genes are expected to be dominant, but CAST Mx1 exhibits additivity {#s32} ----------------------------------------------------------------------------- The degree of genetic dominance of host resistance factors to viral infection in humans and mice has not been thoroughly explored. In general, in the context of biochemical and immunological studies one might expect, just as with *musculus Mx1* combined with *domesticus Mx1*, that genes encoding strong-acting antivirals when combined with a null mut would be mostly dominant. In quantitative genetics, however, it is more often expected that genetic contributions will be mostly additive. In this study, at the *Mx1* locus, we observe both. In genetic crosses of functional and null mice, major host determinants of pathogenesis are normally expected to be classified as either recessive or dominant: recessive when null results in loss of function for a host factor required for disease susceptibility, dominant when null results in loss of function for a host gene required for virus resistance. The recessive case is especially true of passive immunity gained by knockout of host genes critical to viral entry and life cycle, and has been demonstrated in a variety of studies on crop resistance ([@bib20]; [@bib39]; [@bib87]; [@bib30]) and explored in studies of the effects of CCR5 deficiency (CCR5-Δ32 deletion) in resistance to HIV infection and pathogenesis in humans ([@bib46]; [@bib75]; [@bib36]), however the degree of protection in the CCR5-Δ32 heterozygous individuals is not fully understood ([@bib52]; [@bib86]). The dominant case could be considered for a viral sensor, where a single inherited functional copy still provides sufficient sensitivity for viral detection and control, resembling that of an individual inheriting two copies, one from each parent. This type of dominance is best explained by the model proposed by [@bib38], a metabolic--enzymatic model for the architecture of dominance at specific loci, and has been explored further in studies of viral resistance in plants, such as in [@bib22] and [@bib21]. The Kacser--Burns model also provides a mechanism that could in some cases give rise to additivity. [@bib38] predicted that, biochemically, for most enzymes, if there is a 50% reduction in enzyme activity in the heterozygote of a null × functional cross, then in most cases the resulting phenotype will resemble that in the homozygous functional individual and the null allele would likely be characterized as operating in a "recessive" manner. According to their model, the phenotype (or "flux") resulting from a given enzymatic pathway with multiple enzymes joined by "kinetic linking" is a summation of the change in flux due to each specific enzyme activity ("selectivity coefficient"). This means that even a dramatic change in activity for any one enzyme in a physiological system results in barely discernible changes in the system overall, as long as some functional enzyme from the locus of interest is produced. However, the authors also describe two cases where systemic flux can be partially reduced in the heterozygote: (1) in pathways where there are exceptionally few enzymes involved in the system (this case is unlikely for an IFN-responsive antiviral pathway such as *Mx1*); and (2) in pathways where the selectivity coefficient (functional activity) of the enzyme is very low, a case termed heterozygote "indeterminacy," which we henceforth equate to additivity. As further explored by [@bib41], dominance may be incomplete when less active allelic members of a series are involved in a cross with null muts, resulting in a more additive relationship; this seems most likely to explain our observation of CAST *Mx1* effects, and the lower antiviral activity of CAST *Mx1* observed in [@bib60], discussed below, appears to support this. Recent work exploring CAST Mx1 antiviral deficiency {#s33} --------------------------------------------------- Important insights into why CAST *Mx1* might be additive come from recent functional studies. [@bib60] engineered B6 mice expressing either the CAST-derived or A2G-derived MX1 proteins. A2G encodes an MX1 protein sequence similar to the NZO and PWK *musculus* class described in this study. CAST MX1 differs from A2G and *musculus*, with corresponding amino acid changes G83R and A222V in the G domain, which is important for enzymatic and antiviral function. [@bib60] clearly show that CAST provides intermediate protection from IAV in their case using H7N7 (SC35M) and H5N1 (R65) viruses, and suggest that sequence changes in the CAST *Mx1* allele result in reduced enzyme stability, metabolic instability, and possibly in altered dimerization of MX1 monomers and/or changes in MX1 GTPase antiviral activity. It is unknown whether the differences they observed would lead to changes in the dominance of CAST and A2G *Mx1*, although we might expect this to be the case given our mouse infection results. We have verified that the same variants, G83R and A222V, differentiate CAST coding sequence from NZO and PWK, as in [@bib80] and using <http://isvdb.unc.edu> ([@bib63]), and that these are the only nonsynonymous variants on coding transcripts of *Mx1* that differentiate CAST from NZO and PWK. Although we see substantial protection from weight loss in CAST mice, we see a deficiency in the antiviral effects (as measured by RNA-seq viral reads in infected lungs) of CAST *Mx1* on D2 and D4 p.i. \[data not shown, via RNA-seq reads from [@bib96] and transcript analysis in [@bib19]\]. Our work motivates further functional studies of the MX1 protein using *Mx1* transgenic mice. Mx1-independent effects and their follow-up: new studies should leverage CAST Mx1 additivity {#s34} -------------------------------------------------------------------------------------------- A substantial proportion of heritable variance in the diallel was *Mx1* independent (VarP\[all$\left| \text{Mx}1 \right.$\] = 33.81, Table S2 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip)). This was broadly driven by additive genetics and both symmetric and asymmetric epistasis (*i.e.*, differing by parent-of-origin) ([Figure 4, C and E](#fig4){ref-type="fig"}). Relatedly, in our analysis of the *Mx1* locus in the CC-RIX, we estimated *Mx1*-independent effects attributable to overall genome similarity to account for 21% of phenotypic variance. Both observations suggest the presence of additional QTL that could be drawn out given a suitable follow-up design. Consider the design of a second CC-RIX. Here our knowledge of differences in *Mx1* dominance becomes a valuable guide: prioritizing CC F1s with one copy of *musculus Mx1* would reduce power because it would cause *Mx1*-independent drivers to be masked; however, prioritizing CC F1s with one or fewer copies of *castaneus Mx1* would leave the *Mx1*-independent effects exposed and QTL underlying them more easily detected. The inclusion of mice with a single functional *Mx1* in a mapping population provides a basis for mapping loci that modify the effect of *Mx1*, as well as mapping *Mx1*-independent loci controlling disease. [@bib78] showed that even the protectiveness of *Mx1* from the A2G inbred strain is conditional and depends on host genetic background. Thus, CC-RIX designs that incorporate heterozygous classes of *domesticus Mx1* crossed with either CAST *Mx1* or *musculus Mx1* can be of substantial benefit for mapping novel loci affecting infection outcomes, and at least 40% of the F1 crosses in our CC-RIX study incorporate lines which have one single copy (CAST or *musculus*) of *Mx1*. Practical use of the diallel in quantitative genetics {#s35} ----------------------------------------------------- Diallels have a long history in quantitative genetics ([@bib76]; and references in, *e.g.*, [@bib8]; [@bib91]; [@bib44]). They have most commonly been used as a way to assess the relative potency of different genomes with respect to a studied trait, yielding, for example, estimates of generalized combining ability for each strain and estimates of specific combining ability for each F1. More ambitiously, they have been used to obtain an overall picture of a trait's genetic architecture. In many respects, this picture is clearly incomplete: even within the limited genetic space spanned by the founders, the diallel shows only the effects of swapping intact haploid genomes, with no ability to see the effects of recombination. But in other respects it is comprehensive: in considering every F1 combination, one can observe evidence for types of effects---dominance, epistasis, parent-of-origin, epistasis by parent-of-origin, and all sex-specific versions thereof---that would be hard or impossible to identify in other settings, *e.g.*, outbreeding populations derived from the same set of founders. A number of studies have sought to combine the features of a diallel with those of such derived outbred crosses to obtain a picture of genetic architecture that is in some way informed by both. These include studies that map QTL across multiple biparental (*e.g.*, F2) crosses derived from a diallel or diallel-like population (*e.g.*, [@bib68]; [@bib97]; [@bib48]; [@bib69]; [@bib61]) and at least one theoretical study, that of [@bib91], examining the extent to which such information can be analyzed jointly and reconciled with data from the original diallel itself. The goals of our study were more prospective. We use the diallel to prioritize follow-up designs in target populations that segregate genetic material from the same set of founders: the diallel provides evidence of heritable features that would be expected to exist in the CC, and that could be examined in more detail in a suitably designed CC-based experiment. Of course, a comprehensive view of IAV-resistance architecture, even within the genetic space of the CC founder genomes, would be achievable only asymptotically through countless, diverse studies; but in this, the diallel can be seen as a compass, identifying promising initial directions. Summary {#s36} ------- Our study demonstrates the use of diallel crosses for identifying different types of heritable effects that can affect host responses to IAV infection. As such, we find reproducible effects of *Mx1* alleles across first order crosses and recombined populations (despite coordination between protocols being inexact), confirming our previous findings that the CAST *Mx1* allele exhibits an intermediate resistance phenotype against H1N1 strains of influenza virus ([@bib19]), and also identifying novel attributes of the CAST and *musculus Mx1* alleles with respect to additivity and dominance. Despite a body of literature on the effects of null mutations in *Mx1*, the importance of allelic variation at this antiviral gene is just beginning to be understood. A GWAS study published in 2011 found that *Mx1* allelic variation likely plays a role in viral disease manifestation in humans, specifically with regards to West Nile virus infection ([@bib3]), highlighting a need for further study of the role of natural allelic variation in *Mx1* on virus infections in future research. Supplementary Material {#s37} ====================== Supplemental material is available online at [www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1). ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. We thank Peter Palese, at the Icahn School of Medicine at Mount Sinai, New York City, NY, for provision of mouse-adapted IAV H1N1 PR8 viral stocks (A/Puerto Rico/8/1934) for the diallel and pre-CC infections. We thank Yoshihiro Kawaoka, at the University of Wisconsin Department of Pathobiological Sciences, for providing the IAV H1N1 CA04 (A/California/04/2009) infectious clone plasmids that were used to generate viral stocks for the CC-RIX infections. We thank Timothy A. Bell for breeding the diallel and CC-RIX mice. We thank Alan B. Lenarcic for assistance with statistical analysis software and Sarah D. Turner for assistance with data visualization. We are grateful to the referees and editors who contributed to this article by providing the authors with detailed questions and comments. We acknowledge support from the National Institutes of Health T32 Virology Training Grant (5T32AI007419-23) to P.L.M.; National Institutes of Health U19 AI100625 to M.T.H., R.S.B., M.T.F., and F.P.-M.d.V.; National Institutes of Health U54 AI081680 to M.T.H., R.S.B., and F.P.-M.d.V.; and National Institutes of Health R01 GM104125 to W.V. and G.R.K. The organizations that funded this study did not have any role in the design, data collection/interpretation, nor the decision to submit the article for publication. Author contributions: P.L.M., M.T.F., G.R.K., M.T.H., and W.V. wrote the manuscript. P.L.M., M.T.F., D.W.T., F.P.-M.d.V., M.T.H., R.S.B., and W.V. designed experiments. P.L.M., M.T.F., D.R.M., A.C.W., A.W., C.R.M., K.E.N., K.S.P., A.S.C., and F.P.-M.d.V. performed experiments. G.D.S. bred the mice. P.L.M., G.R.K., and W.V. performed the statistical analysis. The authors declare no conflicts of interest. *Note added in proof:* See Turner *et al.* 2018 (pp. [411--426](http://www.g3journal.org/content/8/2/411.full)) for a related work. Communicating editor: L. McIntyre In the potential outcomes framework of [@bib59] and [@bib72], the causal effect of an applied treatment on a measured outcome in an individual *i* is defined as the difference between the outcome under treatment and the outcome that would have been observed if *i* were instead to have received the control. In our case, for some outcome measure *y*, we defined the causal effect as the infection response$$\Delta_{i} = y_{i}^{\text{flu}} - y_{i}^{\text{mock}}\,,$$where $y_{i}^{\text{flu}}$ and $y_{i}^{\text{mock}}$ are "potential outcomes," one of which is observed (the factual) and other of which is unobserved (the counterfactual). Since it is impossible to observe both simultaneously, the causal effect $\Delta_{i}$ can never be measured directly ([@bib32]). It can however be estimated as$${\hat{\Delta}}_{i} = y_{i}^{\text{flu}} - y_{i^{\prime}}^{\text{mock}}\,,$$with the accuracy of this estimate depending on how closely $i^{\prime}$ matches *i*. Our desire for lack of bias in this measure motivates our treatment assignment being randomly assigned within a group of matched individuals. In the treatment-response diallel, we are primarily interested not in infection response for a particular mouse but rather the expectation of this quantity for mice within a given diallel category, or more generally within a group of matched individuals *q*, {#s38} Δ q = E ( y q flu − y q mock ) = E ( y q flu ) − E ( y q mock ) , where in our case *q* is defined as mice specific to a given diallel category and experimental batch. In practice it is natural to estimate this quantity as$${\hat{\Delta}}_{q} = \frac{1}{n_{\text{flu}}}{\sum\limits_{i \in q{\lbrack\text{flu}\rbrack}}y_{i}^{\text{flu}}} - \frac{1}{n_{\text{mock}}}{\sum\limits_{i^{\prime} \in q{\lbrack\text{mock}\rbrack}}y_{i^{\prime}}^{\text{mock}}}\,,$$where $q\left\lbrack \text{flu} \right\rbrack$ and $q\left\lbrack \text{mock} \right\rbrack$ are, respectively, the set of mice in group *q* assigned to flu and mock treatment groups. The variance of this estimate is$${Var}\left( {\hat{\Delta}}_{q} \right) = \frac{{Var}\left( y_{q}^{\text{flu}} \right)}{n_{\text{flu}}} + \frac{{Var}\left( y_{q}^{\text{mock}} \right)}{n_{\text{mock}}},$$and if it is considered likely, as in this study, that the infected phenotypes will be more variable than the mock, ${Var}\left( y^{\text{flu}} \right) > {Var}\left( y^{\text{mock}} \right),$ then it is most efficient experimentally to devote more individuals to the infected arm than the mock arm, *i.e.*, $n_{\text{flu}} > n_{\text{mock}}.$ In our experimental design, we have flu:mock in the ratio 3:1 for each group *q*. It is therefore natural to define the unit of observation *q* as a quartet, that is, Equation A2 with $n_{\text{flu}} = 3$ and $n_{\text{mock}} = 1.$ This means that each diallel category can be represented by multiple quartets, corresponding to multiple observations of ${\hat{\Delta}}_{q},$ denoted in the *Statistical Models and Methods* as ${delta}_{q}.$ We now note several assumptions and connections. Equation A1 equates unit-level with marginal causal effects and thereby assumes no interference between units, specifically, that mice in the same quartet do not affect each other's outcomes; this is approximately true based on the well-established evidence that mice do not transmit H1N1 influenza virus ([@bib50]; [@bib17]), a finding we have also verified by weight-loss profiles and RNA-seq of CC founder strains cohoused with H1N1(PR8)-infected mice ([@bib96]). Last, we note that the definition of ${\hat{\Delta}}_{q}$ in Equation A2 is analogous to an inverse probability-weighted causal effect estimate. The equation for the variance of ${\hat{\Delta}}_{q},$ namely Equation A3, implies that modeling the residual in Equation 5 as homoskedastic would require $n_{\text{flu}}$ and $n_{\text{mock}}$ to be constant throughout; in other words, to ensure comparable precision of infection responses, all quartets should be complete. However, in the diallel experiment, some combinations of batch and diallel category had one or more flu mice missing. In these cases, quartets were defined to have missing values that would be filled in by imputation. The imputation scheme used here corresponds to stochastic regression imputation (*e.g.*, [@bib25]), whereby the incomplete data set is repeatedly augmented to a completed data set using sampled variates from a prediction model, each completed data set is subject to the BayesDiallel analysis described in *Statistical Models and Methods*, and then results across the completed data sets are aggregated. Each imputation required two steps: since the target phenotype of a missing mouse, namely its p.i. weight loss, was considered potentially dependent on its D0 weight, we first imputed missing values for D0 and then imputed missing p.i. weight loss conditional on D0. In addition, at each day p.i., there was one batch/diallel category combination with one mock and four infecteds; for this case only, in each round of imputation, we created a completed quartet by randomly deleting one of the four infecteds. The two-step stochastic regression imputation was performed as follows. Define the observed diallel data for D0 and $\text{pct}^{\text{day}{\lbrack\text{flu}\rbrack}}$ as $\text{D}0_{\text{obs}}$ and $\text{pct}_{\text{obs}}^{\text{day}{\lbrack\text{flu}\rbrack}},$ respectively, and let $\overset{\text{sim}}{\sim}$ represent regression with BayesDiallel followed by stochastic regression imputation, that is, sampling from the posterior predictive. For each $t = 1,\ldots,1000$ round of imputation, we first impute missing D0 values as {#s39} D 0 mis ( t ) ∼ sim BayesDiallel ( D 0 obs ) , where BayesDiallel is fitted as model 1 in [Table 2](#t2){ref-type="table"}. These imputed values are then combined with observed D0 values to give the completed set,$$\text{D}0_{\text{complete}}^{(t)} = \left\{ {\text{D}0_{\text{mis}}^{(t)},\text{D}0_{\text{obs}}} \right\},$$which are then used to impute p.i. weight loss at all time points (D1--D4) for the missing mice as$$\text{pct}_{\text{mis}}^{\text{day}{\lbrack\text{flu}\rbrack},{(t)}}\overset{\text{sim}}{\sim}\text{BayesDiallel}\left( {\text{pct}_{\text{obs}}^{\text{day}{\lbrack\text{flu}\rbrack}},\text{D}0_{\text{complete}}^{(t)}} \right);$$leading in each case to the completed flu p.i. weight loss data,$$\text{pct}_{\text{complete}}^{\text{day}{\lbrack\text{flu}\rbrack},{(t)}} = \left\{ {\text{pct}_{\text{mis}}^{\text{day}{\lbrack\text{flu}\rbrack},{(t)}},\text{pct}_{\text{obs}}^{\text{day}{\lbrack\text{flu}\rbrack}}} \right\},$$and subsequent calculation of quartet-based infection response values as$$\text{delta}^{\text{day},{(t)}} = \text{Quartets}\left\{ {\text{pct}_{\text{complete}}^{\text{day}{\lbrack\text{flu}\rbrack},{(t)}},\text{pct}^{\text{day}{\lbrack\text{mock}\rbrack}}} \right\}.$$Each of the 1000 infection response data sets is analyzed separately using BayesDiallel (model 3 in [Table 2](#t2){ref-type="table"}). The MCMCs from each replicate are aggregated and thinned (sampled across even intervals) and the aggregate results are reported according to the procedure outlined in Algorithm 1 in [File S7](http://www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300438/-/DC1/FileS7.zip). The number of animals imputed were 33 at D1, 33 at D2, 15 at D3, and 16 at D4, in each case corresponding to a small proportion (2.8--3.1%) of the total data set. Phenotypes were not imputed for nonproductive diallel genotypes (which contain no mock or infected mice at all), *i.e.*, $jk \in${NZO × CAST, NZO × PWK} ([@bib7]).

Bacterial Resistance to Aminoglycoside Antibiotics ================================================== Aminoglycoside (AG) antibiotics (Box [1](#BX1){ref-type="boxed-text"}) have not been an exception to the fact that after their introduction in clinical practice, resistance has been recorded ([@B85]). In fact, the three classes of aminoglycoside-modifying enzymes, aminoglycoside acetyltransferases (AACs), aminoglycoside phosphotransferases (APHs), and aminoglycoside nucleotidyltransferases (ANTs), have been widely detected in most pathogenic bacteria as a major determinant of resistance; in these bacteria the presence of aminoglycoside-modifying enzymes correlated with patterns of AG susceptibility ([@B76]). The modified AG (either by acetylation, phosphorylation or nucleotidylation) fails to inhibit their bacterial target, the 30S ribosomal subunit ([@B76]). Most genes encoding aminoglycoside-modifying enzymes are plasmid-located (indicative of a potential acquisition by horizontal gene transfer processes) and confer the bacterial hosts with high levels of AG resistance ([@B19]). **Box 1.** Aminoglycosides: origins, structure, mode of action, resistance and clinical use. Most aminoglycoside antibiotics are produced by bacterial species of the genus *Streptomyces*, such as the antitubercular streptomycin that is produced by *S. griseus* being the first antibiotic identified from bacteria. Other genera producing aminoglycosides are *Micromonospora* and *Bacillus*. Many semi-synthetic aminoglycosides, such as amikacin, have also been produced ([@B55]; [@B51]; [@B74]). Structurally, aminoglycosides are formed by an aminocyclitol (commonly, 2-deoxystreptamine) with additional amino sugars bound by glycosidic bonds. There are two large families of 2-deoxystreptamine aminoglycosides, those carrying substitutions at positions 4 and 5 of the 2-deoxystreptamine ring (including neomycin, paromomycin, lividomycin, ribostamycin and butirosin) and those being substituted at positions 4 and 6 of the 2-deoxystreptamine (including kanamycin, amikacin, tobramycin, dibekacin, arbekacin, gentamicin, isepamicin, sisomicin, and netilmicin). The carbon atoms in the sugar bound to position 4 of the 2-deoxystreptamine ring are named with primed numbers ('), and those in the sugar bound to positions 5 or 6 of the 2-deoxystreptamine ring are named with double-primed numbers ("). Other aminoglycosides contain aminocyclitols distinct to 2-deoxystreptamine (this is the case of streptomycin or apramycin) or are formed by fused amino sugar rings (i.e., spectinomycin) ([@B51]; [@B74]). The following figure shows the structure of kanamycin A, an example of 4,6 di-substituted 2-deoxystreptamine aminoglycoside antibiotic. {#d35e349} The bacterial target of aminoglycosides is the 30S small ribosomal subunit, and their global effect is the interference with protein synthesis. In bacterial cells, translation is initiated when the 30S ribosomal subunit binds the Shine-Dalgarno sequence (because of sequence complementarity between the Shine-Dalgarno sequence and the 16S rRNA molecule of the 30S ribosomal subunit), which is normally present in the 5′ untranslated region of mRNAs. Next, initiation factors and fMet-tRNA will join the complex that finally will recruit the 50S ribosomal subunit in order to start translation. Aminoglycoside binding to the 30S subunit does not affect the binding of mRNA and the large 50S subunit, so translation can proceed. However, aminoglycosides differ in their binding site at the 30S subunit, hence affecting the protein production at different levels. Whereas spectinomycin blocks translocation (hence being bacteriostatic), streptomycin and most 2-deoxystreptamine aminoglycosides lock the ribosome in a conformation that is prone to introducing erroneous aminoacyl-tRNAs. The accumulation of aberrant proteins in the bacteria results in cell death ([@B19]; [@B51]; [@B73]). Resistance to aminoglycosides may be due to several mechanisms ([@B19]; [@B51]). Reduced uptake (which can be a consequence of alterations in the composition of bacterial membrane or to metabolic conditions like anaerobiosis) or the action of efflux pumps can lead to limited intracellular concentration of aminoglycosides, hence causing resistance. Mutations in 16S ribosomal RNA or certain ribosomal proteins such as S12 (encoded by *rpsL* gene) lead to aminoglycoside resistance through target modification; this is also achieved after the action of methyltransferases, which introduce methyl groups in guanine or adenine nucleotides of 16S ribosomal RNA. The presence of aminoglycoside-modifying enzymes is, however, the most prevalent mechanism of aminoglycoside resistance; there are three types of aminoglycoside-modifying enzymes: aminoglycoside *N*-acetyltransferases (AAC), *O*-phosphotransferases (APH), and *O*-nucleotidyltransferases (ANT). Aminoglycoside-modifying enzymes are named by using the aforementioned abbreviations, followed by a number in brackets indicating the site of modification in the aminoglycoside molecule (as explained above), a Roman numeral related with substrate profile, and a lower-case letter for differentiating isoenzymes, i.e., AAC(2′)-Ib. Clinically, due to their oto- and nephrotoxicity and the rising prevalence of resistance, aminoglycosides are commonly reserved as a second line of treatment of serious infections. Due to their low absorption when given orally, aminoglycosides need to be administered through injections. Streptomycin was a first-line drug in the treatment of tuberculosis, and in our days, kanamycin and amikacin are listed as second line drugs against this disease. Spectinomycin is used against *Neisseria gonorrhoeae* infections ([@B78]). *Pseudomonas aeruginosa* infections in cystic fibrosis patients, septicemia, endocarditis and several other infections caused by non-tuberculous mycobacteria, Gram-positive or Gram-negative bacteria can be treated efficiently by using aminoglycosides, either alone or in combinations with other antibacterials such as the beta-lactam antibiotics ([@B55]; [@B51]; [@B9]). In mycobacteria, however, resistance to AGs resulted mainly from mutations of the ribosome components that prevent the drugs from inhibiting its function ([@B43]; [@B93]; [@B72]). This is due to the fact that most mycobacterial species have either one (like *Mycobacterium. tuberculosis*) or two (like *Mycobacterium fortuitum*) ribosomal operons, hence making dominant those mutations acquired in their components ([@B51]; [@B74]). The presence of aminoglycoside-modifying enzymes, mostly AAC, in mycobacterial species has been reported over the years (as detailed in the following sections), and the role of such AACs has been explored, originally for their contribution to AG resistance, and more recently for their role in other bacterial processes, which has resulted in the interest of developing inhibitors of these enzymes ([@B40]; [@B46]). In this review, we will summarize major findings on two mycobacterial AACs, the AAC(2′)-I and Eis enzymes, that have resulted in a Copernican turn for AACs in mycobacteria, from being putative drug resistance mechanisms, to reach the status of novel drug targets. Aacs in Non-Tuberculous Mycobacteria ==================================== The first detection of AACs in mycobacteria ([@B39]) was reported in a group of *M. fortuitum* isolates, an opportunistic fast-growing mycobacteria. Biochemical assays of crude extracts from *M. fortuitum* strains revealed the presence of AAC activity, strongly acetylating gentamicin and kanamycin A, along with other AGs. This substrate profile was consistent with that of AAC(3) enzymes that had been previously described in *Pseudomonas* and *Enterobacteriaceae* ([@B7]), although confirmation at the genetic or molecular levels were not done at that time. Surprisingly, the AG susceptibility profile of *M. fortuitum* could not be correlated with the activity of AACs, indicating that in this species AACs were not the major responsible for AG resistance; it was neither correlated with the presence of plasmids, hence suggesting a chromosomal location ([@B39]). In fact, the frequency of resistant mutants to kanamycin and amikacin in *M. fortuitum* and the related species *Mycobacterium chelonae* ranged between 10^-4^ and 10^-7^ ([@B86]). This relatively high frequency of mutations, along with the fact that AAC activity was detected at similar levels between susceptible and resistant strains, led the authors to suggest that ribosome alterations were the main factor responsible of AG resistance in these species ([@B86]). In another study ([@B82]), altered transport or permeability of AGs was identified as a contributor to AG resistance in *M. fortuitum*, since ribosomes from a clinical isolate were inhibited by one tenth of the MIC of AGs: for example, the MIC of kanamycin for *M. fortuitum* was 50 μg/ml, and in cell-free systems, 5 μg/ml of kanamycin reduced the activity of ribosomes to 13% in comparison with drug-free controls; similar results were obtained when using gentamicin or paromomycin ([@B82]). The biochemical analysis of crude extracts from other non-tuberculous mycobacteria such as *Mycobacterium smegmatis, Mycobacterium phlei, Mycobacterium vaccae*, and *Mycobacterium kansasii*, from both clinical and environmental origins, revealed similar characteristics to those found in *M. fortuitum*: crude extracts from all strains contained AAC enzymatic activity, but no correlation with AG susceptibility profile could be established ([@B80]; [@B35]). In other mycobacterial species such as *Mycobacterium avium* and *Mycobacterium intracellulare*, however, AAC activity could not be detected ([@B35]). In a recent study, using cell-free translation assays, ribosomes of *Mycobacterium abscessus* and *M. smegmatis* were inhibited by both AGs having a 2′-amino group (tobramycin, dibekacin, and kanamycin B) and by those having a 2′-hydroxyl group (amikacin, and kanamycin A). However, in *M. abscessus*, those AGs having a 2′-amino group (tobramycin, dibekacin, and kanamycin B) were less efficient in killing the cells and inhibiting its ribosomes, which is consistent with the presence of a highly active AAC(2′) activity in this species. These findings suggested that in *M. abscessus*, AACs could somewhat mitigate the bactericidal effect of its AGs substrates ([@B52]). Interestingly, when characterizing the enzymatic activity of crude extracts, early reports detected that substrates such as amino sugars, malonyl-CoA, propionyl-CoA or butyryl-CoA inhibited the enzymatic activity of AACs from mycobacterial species. Such effect of non-acetyl CoA donors had never been described for the AAC(3) enzymes from other bacteria ([@B80]). Altogether, these findings strongly suggested for the very first time that, in mycobacteria, AACs could have important metabolic functions, and their contribution to AG resistance could be marginal ([@B81]). Aac(2′), a New Enzyme Comes Into Scene ====================================== In *Providencia stuartii*, a Gram-negative species, phylogenetically distant from mycobacteria, a novel class of AAC was identified as AAC(2′), because of its acetylating activity against gentamicin and lack of enzymatic activity against kanamycin A. The gene encoding AAC(2′)-Ia was cloned from *P. stuartii* ([@B66]) and found to be present in the chromosome of all isolates of this bacteria. In *P. stuartii*, the expression of the *aac(2′)-Ia* gene was controlled by several transcriptional regulators ([@B50]), suggesting that this enzyme could play an important role, beyond its contribution to drug resistance ([@B22]). In fact, AAC(2′)-Ia contributes to *O*-acetylation of peptidoglycan affecting cell morphology and the expression of autolysins, and can use acetylated peptidoglycan precursors as donors of acetyl groups, another indication about the role of this enzyme beyond resistance ([@B61], [@B62]; [@B15]; [@B60]). AAC(2′) in Mycobacteria ----------------------- Given the high AG acetylating activity against gentamicin, tobramycin, netilmicin and its derivatives 2′-*N*-ethyl- and 6′-*N*-ethyl-netilmicin, kanamycin A and kanamycin B found in *M. fortuitum* ([@B39]; [@B3]), we launched a molecular approach aimed at characterizing the determinant of AG resistance in this species. A genomic library of *M. fortuitum* was transformed in *M. smegmatis*, and characterization of AG resistant clones allowed the identification of a *M. fortuitum* gene showing sequence similarity to *aac(2′)-Ia* of *P. stuartii*. The enzyme of *M. fortuitum* was named AAC(2′)-Ib ([@B3]) and was capable of acetylating gentamicin, but not kanamycin A, hence indicating that another AAC enzyme should be present in *M. fortuitum*. Similarly to *P. stuartii*, the *aac(2′)-Ib* gene was found in all strains of *M. fortuitum* regardless of the phenotype of AG resistance, suggesting other roles for the AAC(2′)-Ib in this species. Further studies (done by database searching or by southern blot analysis using the probe of *aac(2′)-Ib* gene) demonstrated the presence of *aac(2′)-I* genes in other mycobacterial species, including *M. tuberculosis* (the major pathogenic species in this genus), *Mycobacterium leprae*, and *M. smegmatis*, indicating thus that the presence of *aac(2′)-I* genes in mycobacteria could be universal ([@B4]). Interestingly, the expression of the *aac(2′)-Id* gene in *M. smegmatis* was driven from two promoters, and the strongest one produced a leaderless transcript having a GTG translation start codon at its 5′ end ([@B54]). Leaderless transcripts are those in which the transcription start site coincides with the translation start codon; although representing a rather unusual feature in the model organism *E. coli*, they are quite common in mycobacteria, where 25% of all transcripts are predicted to be leaderless ([@B73]); in fact, *eis* gene of *M. tuberculosis* (see the section "The Eis Protein Becomes a Novel Aminoglycoside Acetyltransferase," below) is transcribed also as a leaderless mRNA ([@B92]). In leaderless mRNAs, there is no Shine-Dalgarno sequence, and 70S ribosomes bind directly to the 5′-end of the mRNA in order to initiate translation ([@B73]). New Roles for AAC(2′) in Mycobacteria ------------------------------------- Two major evidences supported that in mycobacteria, AAC(2′)-I enzymes could have additional roles other than acetylation of AGs containing 2′-amino group: first, the capability of amino sugars and diverse acyl-CoA molecules to inhibit the acetylating activity of mycobacterial crude extracts ([@B81]); and second, the implication of AAC(2′)-Ia (an enzyme of the same class) in acetylation of cell wall substrates in *P. stuartii*. In a series of laboratory mutants of *P. stuartii* expressing *aac(2′)-Ia* gene at different levels, it was found that the extent of peptidoglycan acetylation correlated with the activity of AAC(2′)-Ia, hence suggesting a partial contribution of this enzyme (along with other enzymes) to peptidoglycan acetylation. Many bacterial pathogens acetylate their peptidoglycan as a way to resist the action of muramidase enzymes. Under *in vitro* conditions, AAC(2′)-Ia was able to acetylate tobramycin having acetylated peptidoglycan as donor of acetyl groups ([@B61], [@B62]; [@B60]). We investigated the hypothesis of mycobacterial AAC(2′)-I enzymes having also a role in cell wall metabolism, and a gene knock-out mutant of *M. smegmatis* deleted in *aac(2′)-Id* gene was constructed. This mutant (named EP-10) defective in AAC(2′)-I activity was more susceptible to gentamicin, tobramycin, dibekacin, and netilmicin than the parental wild-type strain, and crude extracts of *M. smegmatis* EP-10 failed to acetylate 2′-amino group containing AGs, whereas wild-type strains of *M. smegmatis* readily acetylated such AGs ([@B4]; [@B52]). *M. smegmatis* EP-10 was also twofold more susceptible to lysozyme, a feature that has been associated with the extent of peptidoglycan acetylation ([@B4]). Hence, we concluded that in *M. smegmatis*, AAC(2′)-Id enzyme could also contribute to acetylation of peptidoglycan, since a less extensively acetylated peptidoglycan in the knock-out strain would be more susceptible to lysozyme degradation affecting cell viability. Biochemical Analysis of Mycobacterial AAC(2′)-I Enzymes ------------------------------------------------------- The presence of an AG (2′)-*N*-acetyltransferase gene in the genome of *M. tuberculosis* was intriguing. In this species, AAC activity had never been reported ([@B56]) and the expression of the *aac(2′)-Ic* gene \[annotated as Rv1258c gene in the *M. tuberculosis* H37Rv genome ([@B16])\] in the surrogate host *M. smegmatis* could not be associated with any change in the levels of susceptibility to AGs ([@B4]). We constructed a knock-out mutant of *M. tuberculosis* H37Rv deleted in the *aac(2′)-Ic* gene and observed that the mutant (named *M. tuberculosis* B1) was twofold more susceptible than the original wild-type strain to AGs containing a 2′-amino group such as gentamicin, tobramycin and dibekacin, and fourfold more susceptible to 6′-*N*-ethyl-netilmicin. This indicated that the *aac(2′)-Ic* gene was being expressed in *M. tuberculosis* although at a very low level, and that the AAC(2′)-Ic enzyme in *M. tuberculosis* would acetylate all these four AGs in the wild type strain; hence, acetylated AGs would bind less efficiently to the ribosome, and the AAC(2′)-Ic enzyme would contribute to basal AG resistance in this species. In order to find the physiological role of this enzyme in *M. tuberculosis*, recombinant *E. coli*-produced AAC(2′)-Ic enzyme from *M. tuberculosis* was studied and found to efficiently acetylate AG antibiotics containing an amino group at the 2′ position (as expected; for example, *K*~m~ for kanamycin B was 1.4 μM), and surprisingly, it was also capable of acetylating (although to a much lesser extent) other AGs such as kanamycin A (*K*~m~ 320 μM) and amikacin (*K*~m~ 968 μM) that have a hydroxyl group in the 2′ position, hence suggesting this enzyme to be capable of both *N*- and *O*-acetylation ([@B34]; [@B20]). This residual activity of AAC(2′)-Ic against kanamycin A and amikacin does not affect their bactericidal activity, and these two antibiotics are used as second line drugs against drug resistant *M. tuberculosis* ([@B89]). The activity of *M. tuberculosis* AAC(2′)-Ic is dependent on metal ions, being inhibited by Cu^2+^ and Au^3+^ ([@B48]). Another study consisted in binding covalently the AGs kanamycin A, tobramycin, neamine and neomycin B to an agarose matrix in order to quantify the extent of AG acetylation by AAC enzymes and their subsequent ability to bind an artificial probe mimicking the A-site of the ribosome ([@B8]). In such experimental model system, the AGs acetylated by *M. tuberculosis* AAC(2′)-Ic enzyme (only tobramycin, neamine and neomycin B) maintained their binding affinity with the probe mimicking A-site of the ribosome at detectable levels, maybe because the percent of AG acetylation by AAC(2′)-Ic was low. In contrast, these AGs were very efficiently acetylated by *E. coli* AAC(3) (at a different amino group) and had readily lost their affinity for binding this artificial probe. These experiments suggested that subtle differences in the structure of modified AGs (i.e., acetylation at the amino group in 2′ or 3 position) are sufficient to drastically affect their capability of binding to the A-site probe, and this would be expected to correlate with their activity as ribosome inhibitors. In consequence, in mycobacteria, AAC(2′)-I enzymes would not play a major role in resistance to these drugs ([@B8]). High resolution crystal structures of AAC(2′)-Ic complexed with AGs demonstrated that this enzyme is a member of the GNC5 acetyltransferase superfamily and suggested a role in the synthesis of mycothiol, a metabolite that has a key role in regulating redox potential in mycobacteria ([@B84], [@B83]). In agreement with the preferential *N*-acetylating activity over the *O*-acetylating activity that was found in the *M. tuberculosis* enzyme ([@B34]; [@B20]), *M. abscessus* clinical isolates were found to be more susceptible to AGs containing a hydroxyl group at the 2′ position (such as amikacin and kanamycin A) than to AGs with a 2′-amino group (such as tobramycin, dibekacin and kanamycin B), as the latter group would be substrates of *M. abscessus* AAC(2′)-I enzyme ([@B52]). In fact, crude extracts of *M. abscessus* efficiently acetylated kanamycin B, whereas kanamycin A was not acetylated at detectable levels. In this species, deletion of the *aac(2′)-I* gene resulted in increased susceptibility to kanamycin B, tobramycin, dibekacin and gentamicin C (all of them containing a 2′-amino group) ([@B70]). These two reports demonstrate that in *M. abscessus*, the presence of an AAC(2′)-I enzyme contributes to decreased innate susceptibility to AGs containing a 2′-amino group ([@B49]). Becoming a Drug Target: Developing Inhibitors of AAC(2′)-I ---------------------------------------------------------- The interest in developing inhibitors against AAC(2′)-I enzymes came from a study in the non-tuberculous species *M. abscessus* ([@B53]). It was found that AGs such as amikacin, gentamicin or tobramycin, which are normally bactericidal against *E. coli*, do not have such activity against *M. abscessus* or *M. smegmatis*. However, disruption of the chromosomally encoded *aac(2′)-I* gene in these species restored the bactericidal activity of these AGs ([@B53]). Given that AGs are used as second-line drugs in treatment of multidrug resistant (MDR) tuberculosis infections, and also in the treatment of other infections caused by non-tuberculous mycobacteria, the possibility of developing compounds that could enhance bactericidal activity of current antimycobacterial treatments became an interesting approach. To date, the only putative inhibitor of *M. tuberculosis* AAC(2′)-Ic is andrographolide, a natural product that was identified in methanolic extracts of a plant that were capable of inhibiting growth of *M. tuberculosis* strains ([@B63]). *In silico* analysis predicted that this compound could potentially bind with a high affinity the AAC(2′)-Ic enzyme, as well as isocitrate lyase (a metabolic enzyme of the glyoxylate shunt, involved in persistence and virulence of *M. tuberculosis*) and other *M. tuberculosis* proteins ([@B63]). However, the specificity of this binding and the ability to really inhibit such putative target proteins were not tested. Given that the gene encoding AAC(2′)-Ic is not essential in *M. tuberculosis*, a direct link between AAC(2′)-Ic inhibition and bacterial growth inhibition could be discarded. Hence, the ability of plant extracts containing andrographolide to inhibit growth of *M. tuberculosis* could be due to the presence of additional compounds in the extract, or to multiple effects on *M. tuberculosis* cells. Other *in silico* analysis revealed that AAC(2′)-Ic enzyme from *M. tuberculosis* could interact with ten other proteins (including a protein of a putative RND-like efflux pump), suggesting that inhibition of AAC(2′)-Ic could also impact many other metabolic processes, hence conferring this enzyme with a relevant role in drug discovery of antituberculosis agents ([@B42]). The Eis Protein Becomes a Novel Aminoglycoside Acetyltransferase ================================================================ Investigation of *M. tuberculosis* virulence factors lead to the identification of a protein, that was required for infecting and survival in human macrophages; this protein was named Eis for [e]{.ul}nhanced [i]{.ul}ntracellular [s]{.ul}urvival (Box [2](#BX2){ref-type="boxed-text"}) ([@B87]). Bioinformatic analysis revealed that Eis protein of *M. tuberculosis* was an acetyltransferase of the GCN-5 family ([@B71]). **Box 2.** Discovery and characterization of enhanced intracellular survival (Eis) protein of *M. tuberculosis.* A genome wide investigation of the ability of *M. tuberculosis* for infecting macrophages resulted in the identification of a coding sequence that, once cloned in the non-pathogenic *M. smegmatis* species, conferred capacity for infecting the human macrophage-like cell line U937. This gene was named *eis* (Rv2416c in the *M. tuberculosis* genome; [@B16]) for enhanced intracellular survival ([@B87]), and was detected only in pathogenic species of mycobacteria. The promoter of the *eis* gene is similar to consensus *E. coli* sigma-70 dependent promoters ([@B69]) and it is recognized by *M. tuberculosis* SigA sigma factor ([@B90]). The Eis protein was found to be mostly hydrophilic, but having a hydrophobic N-terminal end, so that it could be found in the cytosolic but also in other cell fractions such as the membrane, cell wall or among the secreted proteins of *M. tuberculosis* ([@B17]). In fact, antibodies against Eis could detect this protein in the sera of tuberculosis patients ([@B17]) and in the cytoplasm of infected macrophages ([@B71]). Also, it was found that Eis protein, directly added to cultures of human monocytes, modulated the secretion of pro-inflammatory cytokines in a similar way to that found in *M. tuberculosis* infected cells ([@B71]), hence suggesting a role of Eis as an effector protein. Further studies demonstrated that Eis inhibits the extra-cellular signal-regulated kinase 1/2 (ERK1/2) and JAK pathways, and in consequence it inhibited the production of TNF-alpha and IL-4, and stimulated the production of IFN-gamma and IL-10 ([@B47]). The effect of Eis in increasing production of IL-10 was found to be related to Eis-mediated acetylation of histone H3, which binds the promoter of the human IL-10 gene ([@B21]). Hence, by disturbing cross-regulation of T-cells and impairing TH1 and TH2 response, Eis could mediate *M. tuberculosis* pathogenicity ([@B47]). In fact, an isolate of the Beijing family (which are more transmissible and virulent than other *M. tuberculosis* genetic lineages ([@B33]) was found to contain elevated levels of Eis protein, mediated by increased expression of SigA ([@B90]). Other key factors in host immune response to tuberculosis are also mediated by Eis, which increased production of ROS and consequently modulated processes such as autophagy, inflammation, and cell death ([@B75]). These processes are started by Eis-dependent acetylation of dual-specificity phosphatase-16 (DUSP16)-mitogen-activated protein kinase phosphatase-7 (MKP-7), which dephosphorylates the JNK protein leading to its inactivation ([@B44]; [@B91]). Other studies have revealed the activity of Eis for acetylating arylalkylamines such as histamine, octopamine, or tyramine, suggesting novel roles for this protein in *M. tuberculosis* pathogenicity ([@B59]). Later on, the analysis of kanamycin resistant *M. tuberculosis* laboratory and clinical strains revealed mutations in the -10 and -35 regions of the *eis* gene promoter, which resulted in increased levels of *eis* mRNA and Eis protein. These mutations were related to low-level resistance to kanamycin (MIC 25 μg/ml), but not to amikacin (MIC \< 4 μg/ml), whereas 16S rRNA mutations confer higher levels of resistance to kanamycin (MIC \> 80 μg/ml) and frequently cross-resistance to amikacin. These *eis* mutants also displayed increased levels of AAC activity, hence demonstrating that Eis was a novel class of AAC, highly divergent from all other previously known AACs. Eis is capable of acetylating kanamycin more efficiently than amikacin, and streptomycin was not found to be a substrate of Eis ([@B92]). From then on, detection of mutations in *eis* promoter has become a relevant assay in clinical microbiology laboratories for determining susceptibility to kanamycin ([@B27]); the diagnostic and clinical implications of these tests are beyond the scope of this review. Formation of stable hexamers by Eis is required for its AAC activity ([@B23]; [@B6]), which can acetylate multiple amine groups of different AG antibiotics, including netilmicin, sisomicin, neamine, ribostamycin, paromomycin, neomycin B, kanamycin, amikacin, tobramycin and hygromycin, resulting in mono-, di-, tri, and tetraacetylated products ([@B12]; [@B37]), being able to use not only acetyl-CoA but also other acyl-CoA derivatives ([@B13]). The Eis protein works by a random-sequential bisubstrate mechanism of acetylation ([@B79]). Interestingly, Eis is also able to acetylate capreomycin, a polypeptide second-line antituberculosis drug commonly used in the treatment of MDR tuberculosis infections ([@B67]). Several metal ions such as Au^3+^, Cd^2+^ and Zn^2+^ inhibited Eis activity *in vitro* ([@B48]). Other mycobacterial species such as *M. smegmatis* and *M. abscessus* have ortholog (and even paralog) *eis* genes, although the Eis proteins have distinct biochemical features and impact on AG susceptibility in comparison with Eis of *M. tuberculosis* ([@B14]; [@B70]). For example, *M. abscessus* has two *eis* genes, and the deletion of one (but not the other one) resulted in altered AG susceptibility ([@B70]; [@B49]). Consistently with these findings, mutational changes in the amino acid residues lining the substrate binding site of *M. tuberculosis* Eis altered its substrate specificity ([@B41]). It is important to note that in *M. tuberculosis*, transcription of the *eis* gene is activated by the regulator WhiB7 ([@B67]). Mutations in the promoter of *whiB7* gene that led to increase in the mRNA of this gene resulted in increased expression of *eis* gene, along with other genes such as *rv1258c* (encoding the Tap efflux pump; [@B2]), hence resulting in cross resistance to several drugs including kanamycin (mediated by Eis protein) or streptomycin (mediated by Tap efflux pump). Similarly, in *M. abscessus*, WhiB7 controlled the expression of one of the two *eis* genes in this species and also that of the *erm(41)* gene, which encodes a ribosomal methyltransferase that by altering target structure is associated with resistance to macrolide antibiotics ([@B64]; [@B49]). Subinhibitory concentrations of clarithromycin induced the *whiB7* gene and consequently decreased Eis-mediated susceptibility to AGs, such as amikacin that is currently used in the treatment of *M. abscessus* infections ([@B64]). Aminoglycosides and Beyond... ----------------------------- The unusual properties of Eis acetyltransferase include its capability for acetylating peptides and proteins ([@B44]; [@B38]; [@B91]), in contrast with other AACs. The *M. tuberculosis* nucleoid-associated protein HU (encoded by Rv2986c gene) can be acetylated by Eis on multiple lysine residues, hence decreasing its ability to interact with DNA, and altering its DNA compactation activity ([@B28]; [@B31]). Overexpression of Eis led to a hyperacetylation of HU protein, and consequently, to a decompactation of the genome ([@B28]). The reverse effect (condensation of relaxed DNA) could be reached through the deacetylation of HU protein, which is mediated by a Sir2 family protein from *M. tuberculosis* encoded by the Rv1155c gene ([@B5]; [@B31]). Controlling the architecture of DNA is a key process in any bacteria, and so, the HU protein is essential for *M. tuberculosis*. In fact, inhibitors of HU have been discovered ([@B10]), which could act in synergy with potential inhibitors of Eis, as described in the next section. Finding Inhibitors of Eis Protein --------------------------------- The crystal structure of *M. tuberculosis* Eis protein was determined by several groups ([@B12]; [@B44]), which has been useful for determining docking properties of potential inhibitory compounds; also, its comparison with the crystal structure of *M. smegmatis* Eis protein revealed several distinct structural features that may account for the biochemical and substrate differences between the two proteins ([@B45]). A first screening of potential inhibitors of Eis from *M. tuberculosis* resulted in the identification of 25 molecules (including the antiseptic chlorhexidine) that inhibited Eis with IC~50~ values in the low micromolar range. In addition, this inhibition was specific to the *M. tuberculosis* Eis protein, since these molecules could not inhibit significantly AACs from AAC(2′), AAC(3), and AAC(6′) families ([@B32]), nor Eis protein from *Bacillus anthracis* ([@B29]). Later studies revealed the presence of *eis*-like genes in many pathogenic and non-pathogenic bacteria (remarkably, many mycobacterial species have two or even three paralogs of the *eis* gene), and chlorhexidine was capable of inhibiting (to different levels) all Eis proteins that were capable of acetylating AGs ([@B30]). A second screening of a larger collection of small-molecule compounds resulted in the identification of several families of compounds capable of inhibiting Eis activity. These contained diverse chemical scaffolds ([@B24],[@B25], [@B26]; [@B88]; [@B57]). Besides, these inhibitors were highly selective for Eis, and did not inhibited AACs from other families ([@B25]). More importantly, as these inhibitors bound in the AG pocket of the Eis protein, they were able to reverse kanamycin resistance of a *M. tuberculosis* isolate ([@B25]; [@B88]; [@B57]). Some of these inhibitors, such as those based on a pyrrolo\[1,5-a\]pyrazine scaffold, also lacked any toxicity on mammalian cell lines ([@B26]). Other Aminoglycoside-Modifying Enzymes in Mycobacteria ====================================================== Genome-wide analysis of *M. tuberculosis* genome identified only one other potential AAC (encoded by the Rv1347c gene), although such enzymatic activity could not be detected on the recombinant protein ([@B20]). Later studies related the product of the Rv1347c gene with a role in the synthesis of mycobactin, the mycobacterial siderophore ([@B11]). Leaving apart AACs, only a few reports of other classes of aminoglycoside-modifying enzymes have been done in mycobacteria. An APH enzyme of the APH(3″) family, conferring resistance to the AG streptomycin only, has been characterized in *M. fortuitum* ([@B65]) and *M. abscessus* ([@B18]; [@B49]); the latter species encodes up to 11 additional putative APH enzymes ([@B68]). Furthermore, a putative APH of the APH(3′) class, encoded by the Rv3168 gene of *M. tuberculosis*, was identified and expressed as a recombinant enzyme in *E. coli*, being related to kanamycin phosphotransferase activity ([@B1]). Back to the Start Point: Eis in *M. fortuitum* ============================================== We started this review referring to previous work that has shown that crude extracts of *M. fortuitum* harbored AAC activity having gentamicin, tobramycin, netilmicin and its derivatives 2′-*N*-ethyl- and 6′-*N*-ethyl-netilmicin, kanamycin A and kanamycin B as substrates. So far, the only AG acetyltransferase identified in this species has been AAC(2′)-Ib ([@B3]), which cannot explain the acetyltransferase activity against kanamycin A and 2′-*N*-ethyl-netilmicin detected in *M. fortuitum* crude extracts. Later studies characterized Eis AAC in diverse mycobacterial species, but no report were been done on a putative Eis protein in *M. fortuitum*. In view of the universal presence of Eis proteins in mycobacteria, and its activity as AAC, we hypothesized that *M. fortuitum* could also have a putative Eis protein that would be responsible for the acetyltransferase activity against kanamycin A and 2′-*N*-ethyl-netilmicin detected in crude extracts of this species, as it was reported earlier ([@B39]; [@B3]). Thus, we first ascertained the existence of an *eis* gene in the genome of *M. fortuitum* ([@B36]), which presented a 94% of identity with respect to the one reported in *M. tuberculosis;* the sequence of both Eis proteins from *M. tuberculosis* and *M. fortuitum* also presented high levels of identity ([Figure 1](#F1){ref-type="fig"}). We designed two oligonucleotides for amplifying specifically a DNA fragment from *M. fortuitum* genome containing *eis* gene. This DNA fragment was subsequently cloned in pMV261 vector ([@B77]), which expresses genes constitutively under the control of the *hsp60* gene promoter, resulting in plasmid pFS2. Given that this plasmid still contains the Tn*903*-derived aminoglycoside-3′-phosphotransferase (*aph*) gene (present in the original pMV261 cloning vector) conferring kanamycin resistance as selection marker, we anticipated that *M. smegmatis* strains harboring pMV261 vector or its derivatives would be intrinsically resistant to kanamycin A; this would prevent from determining whether this antibiotic is a substrate of the *M. fortuitum* Eis protein. Additionally, the selection marker could have cross-effect with other AGs and thus interfering with the resistance phenotype conferred by *eis* gene. To circumvent these problems, we generated a derivative of pFS2 through the disruption of the *aph* gene with the ampicillin resistance cassette (*bla* gene) from pGEM^®^-T easy (Promega). The resistance to 2′-*N*-ethylnetilmicin conferred by *eis* gene, as demonstrated for the parental plasmid pFS2 (see [Table 1](#T1){ref-type="table"}) was used as resistance marker for transformant selection and plasmid maintenance. This process resulted in plasmid pEAC which contains the *M. fortuitum eis* gene and no other determinant of AG resistance. {#F1} ###### MICs (μg/ml) of antibiotics of different structural families in the *M. smegmatis* mc^2^ 155 strains that overexpress *eis* gene from *M. fortuitum*. Strain *M. smegmatis* mc^2^ 155 ------------------------- -------------------------- ------------ ------------ ------- Plasmid None pMV261 pFS2 pEAC Marker gene on plasmid None *aph* *aph* *bla* *M. fortuitum eis* gene -- -- \+ \+ Gentamicin 0.78-1.56 0.78 3.12 3.12 2′-*N*-ethyl netilmicin 6.25 3.12--6.25 50--100 50 6′-*N*-ethyl netilmicin 3.12 3.12 3.12--6.25 25 Kanamycin A 1.56 \>100 \>100 7.8 Kanamycin B 6.25 \>100 \>100 12.5 Hygromycin 15.6 15.6 31.2--62.5 62.5 Amikacin 0.39 0.39 0.39 0.39 Capreomycin 1.95 1.95 3.9 3.9 Streptomycin 0.25 0.25 0.25 0.25 Spectinomycin 62.5 31.2--62.5 62.5 62.5 The values separated by a dash indicate that growth was detected in both concentrations. The range of antibiotic concentrations spanned from 0.78 μ g/ml to 100 μ g/ml, and from 0.25 μ g/ml to 125 μ g/ml; a twofold difference in the MIC was not considered as significant. aph: aminoglycoside 3′-phosphotransferase from Tn903; bla: beta-lactamase. The three plasmids \[original empty vector pMV261 as a control, and the two plasmids (pFS2 and pEAC) containing *M. fortuitum eis* gene\] were introduced in *M. smegmatis* mc^2^ 155 in order to over-express the ortholog *eis* gene from *M. fortuitum* and to elucidate its hypothetical implication in AG susceptibility. The antibiotic susceptibility assay was made, based on a double dilution protocol with the addition of resazurin dye ([@B58]). We observed that plasmids pFS2 and pEAC produced detectable changes in AG susceptibility of *M. smegmatis* mc^2^ 155, which can be attributed to the expression of the plasmid-borne *M. fortuitum eis* gene. The major shift in the MICs was detected for 2′-*N*-ethylnetilmicin ([Table 1](#T1){ref-type="table"}), since the MIC increased from 3.12 to 6.25 μg/ml in the control strains to 50--100 μg/ml in the strains containing *M. fortuitum* Eis, accounting for a 8- to 32-fold increase; similar changes were observed for 6′-*N*-ethylnetilmicin (8-fold increase in the MIC). A moderate decrease in the susceptibility to kanamycin A (fivefold), hygromycin (twofold to fourfold) and gentamicin (twofold to fourfold) was also observed. Finally, slight changes (twofold) in the MICs were detected for kanamycin B and capreomycin; this finding is consistent with previous reports on the activity of *M. tuberculosis* Eis against capreomycin ([@B67]). The levels of susceptibility to the AGs amikacin, streptomycin and spectinomycin, and to other non-AG compounds tested (isoniazid, rifampicin, ethambutol, ciprofloxacin, tetracycline, chloramphenicol) were not altered significantly by the presence of the plasmid-encoded *eis* gene ([Table 1](#T1){ref-type="table"} and data not shown), suggesting that either these antimicrobials are not substrates of the Eis enzyme or the corresponding acetylations (if any) might just not affect antibacterial activity. Concluding Remarks ================== The presence and activity of AAC(2′)-I and Eis AACs in mycobacteria have clearly demonstrated that their primary role is little related with susceptibility to AGs. In the case of AAC(2′)-I enzymes, its presence in phylogenetically distant genera as *Providencia* and *Mycobacterium* remains to be an evolutionary mystery. In contrast with this restricted distribution of AAC(2′)-I enzymes among bacteria, Eis enzymes seem to be more widely distributed, being present even in non-pathogenic and environmental species, which suggest a general function of Eis-like enzymes in bacterial metabolism, and virtually excludes any potential selection of *eis* genes due to the use of AGs, or its horizontal transfer from other species. It is clear that in mycobacteria ribosomal modifications constitute the major mechanism of AG resistance, given that only one or two copies of ribosomal RNA operons are present in these species, hence making likely the acquisition of mutations conferring high levels of AG resistance. Then, AAC(2′)-I enzymes only contribute modestly to innate low level susceptibility to AGs, and despite other roles have been suggested in the literature for mycobacterial AAC(2′)-I enzymes, their relevance as potential drug targets is still modest, especially in comparison with Eis acetyltransferase. The contribution of Eis acetyltransferase to virulence of *M. tuberculosis*, and the finding that Eis is related with resistance to kanamycin (a second line drug for the treatment of tuberculosis) in clinical isolates has greatly attracted the attention and promoted the interest in developing Eis inhibitors. In some way, Eis inhibitors would fall into the class of anti-virulence and anti-resistance mechanisms compounds, which is a trending topic in the age of antimicrobial resistance. Globally, antimicrobial resistance is a major public health threat, and multi and extensively drug resistant (MDR, XDR) tuberculosis is a case of particular concern, so progress and major advances that can be expected from the coming years will be greatly welcomed. Author Contributions ==================== JA, LR, and CM contributed to the conception and design of the study. FS-G, EA-C, AL, LR, and EP-H carried out the experimental work. FS-G, EA-C, and JA wrote the manuscript. All authors contributed to manuscript revision, read and approved the submitted version. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** FS-G and EA-C are recipients of FPU fellowships from the Spanish Ministry of Economy and Competitivity. JA acknowledges funding from the European Comission \[More Medicines for Tuberculosis (MM4TB) grant 260872\] and the Spanish Ministry of Economy and Competitivity (grants SAF-2013-48971-C2-2-R and SAF2017-84839-C2-1-R). Dessi Marinova was acknowledged for critical reading of the manuscript. [^1]: Edited by: Silvia Buroni, University of Pavia, Italy [^2]: Reviewed by: Peter Sander, University of Zurich, Switzerland; Roland Brosch, Institut Pasteur, France [^3]: ^†^Present address: Fernando Sanz-García, Departamento de Biotecnología Microbiana, Centro Nacional de Biotecnología, Consejo Superior de Investigaciones Científicas, Madrid, Spain; Esther Pérez-Herrán, GlaxoSmithKline, Tres Cantos Medicines Development Campus, Tres Cantos, Spain; Liliana Rodrigues, Unit of Medical Microbiology, Global Health and Tropical Medicine, Instituto de Higiene e Medicina Tropical, Universidade Nova de Lisboa, Lisbon, Portugal [^4]: ^‡^These authors have contributed equally to this work [^5]: This article was submitted to Evolutionary and Genomic Microbiology, a section of the journal Frontiers in Microbiology

Introduction {#s1} ============ It is commonly accepted that in order for a neuronal population to encode the value of a quantity x, it must contain cells tuned to a range of values of x. Thus for example the retina can encode information about the wavelength of light because it contains three different types of cones with different tuning to wavelength, and the primary visual cortex can encode feature orientation because it contains neurons tuned to a range of orientations. This is unproblematic because natural images contain a wide range of light wavelengths and object orientations. However, the same argument applied to stereo vision produces some more challenging conclusions. The expected vertical disparity in natural viewing depends on position in the retina, with opposite signs in opposite quadrants of the visual field. The range in vertical disparities encountered at a given position depends on a number of assumptions about eye movement and scene statistics, but all attempts to estimate it agree that it is extremely narrowly distributed compared to horizontal disparity [@pcbi.1000754-Liu1], [@pcbi.1000754-Hibbard1], [@pcbi.1000754-Read1]. Thus, if disparity sensors in the brain were to reflect disparity in the natural world, we would expect the distribution of two-dimensional disparity tuning at a given retinotopic location to be highly elongated, virtually one-dimensional, with a wide range of horizontal disparity and a narrow range of vertical disparity, centered on the value expected for that retinotopic location. Yet, vertical disparities which hardly ever occur in normal visual experience can still have demonstrable effects on perception in the lab [@pcbi.1000754-Helmholtz1], [@pcbi.1000754-Ogle1], and there is evidence that stereo matching occurs in all 2D directions, vertical as well as horizontal [@pcbi.1000754-Farell1]. Thus, the brain clearly can extract unusual vertical disparities, on relatively local scales [@pcbi.1000754-SerranoPedraza1], [@pcbi.1000754-Rogers1], [@pcbi.1000754-Kaneko1]. This has led to the conclusion that the brain must contain neurons tuned to a range of vertical disparities, including highly unusual ones, on the assumption that otherwise, these disparities could not be perceived [@pcbi.1000754-Durand1], [@pcbi.1000754-Durand2], [@pcbi.1000754-Gonzalez1]. Motivated by this, a number of physiological studies have examined two-dimensional disparity tuning in cortical neurons in monkey primary visual cortex (V1). Near the fovea, most disparity-tuned neurons are tuned to vertical disparities which are not significantly different from zero, given the confidence interval on the measurement [@pcbi.1000754-Cumming1]. In the visual periphery, neurons tuned to non-zero vertical disparities have been reported [@pcbi.1000754-Durand1], [@pcbi.1000754-Durand2], [@pcbi.1000754-Gonzalez1]. Unfortunately, these studies only reported disparity in head-centric coordinates, which can differ substantially from retino-centric disparity [@pcbi.1000754-Read2]. For example, it is perfectly possible for a neuron tuned to a substantial head-centric vertical disparity, say 0.3°, to be tuned to a vertical disparity of 0° on the retina [@pcbi.1000754-Read1]. Thus, the published data do not enable us to draw any conclusions about 2D disparity tuning on the retina. Furthermore, these studies did not report the retinal location of individual neurons, making it impossible to assess whether a range of vertical disparity tuning is found at a single retinotopic location. Given this lack of data from physiology, theoretical considerations become important. A clear understanding of how, in principle, neurons could represent two-dimensional disparity is essential for guiding future physiology experiments. We recently argued [@pcbi.1000754-Read3] that a population of model binocular neurons like that shown in [Figure 1](#pcbi-1000754-g001){ref-type="fig"}, tuned to a range of horizontal disparities and orientations but all tuned to zero vertical disparity on the retina, nevertheless encodes information about the vertical disparity of the stimulus. This original model only extracted the magnitude, not the sign, of the local vertical disparity, and we later demonstrated that this was inconsistent with human psychophysics [@pcbi.1000754-SerranoPedraza2]. However, this model did not make optimal use of the information available in the population. In the present paper, I show that this population of disparity sensors does contain information about both the magnitude and the sign of the vertical disparity at that point in the retina, even if all neurons in the population are tuned to the same vertical disparity. With an appropriate decoding technique, information about the two-dimensional disparity can be deduced from activity in this one-dimensional population. This result is of interest in its own right as a theoretical demonstration that it is possible to extract the value of a quantity from a neuronal population, all of whose members respond optimally to the same value of that quantity. From the point of view of understanding stereo vision, it means that two-dimensional disparity may be represented far more efficiently than previously appreciated. ![A neuronal population which explicitly encodes horizontal, but not vertical, disparity.\ The shaded region represents the space of two-dimensional disparity on the retina [@pcbi.1000754-Read2]. The purple disks represent the preferred 2D disparity of an idealized population of disparity sensors. Although these sensors form a one-dimensional population, all tuned to zero vertical disparity, they can nevertheless encode two-dimensional stimulus disparity, e.g. the stimulus disparity represented by the green dot, which has both a horizontal and a vertical component. (Cf [figure 1](#pcbi-1000754-g001){ref-type="fig"} of Serrano-Pedraza & Read [@pcbi.1000754-SerranoPedraza2].)](pcbi.1000754.g001){#pcbi-1000754-g001} Methods {#s2} ======= Overview {#s2a} -------- The essential insight guiding this paper is relatively trivial. According to the stereo energy model of disparity-selective neurons [@pcbi.1000754-Ohzawa1], [@pcbi.1000754-Ohzawa2], cells with obliquely-oriented receptive fields will also have obliquely-oriented disparity tuning surfaces, like the one illustrated in [Figure 2A](#pcbi-1000754-g002){ref-type="fig"}. This cell\'s optimal disparity is marked with a red circle. It has zero vertical component, i.e. the cell responds best to zero vertical disparity. [Figure 2B](#pcbi-1000754-g002){ref-type="fig"} shows two cross-sections through this surface, corresponding to vertical disparity tuning curves for two different horizontal disparities, as indicated by the vertical lines in [Figure 2A](#pcbi-1000754-g002){ref-type="fig"}. At the optimal horizontal disparity (red curve), the cell responds best to zero vertical disparity. But at horizontal disparities away from the optimum (e.g. purple curve), the cell\'s response is reduced, but is now tuned to a non-zero vertical disparity. Thus, while the cell in [Figure 2](#pcbi-1000754-g002){ref-type="fig"} is "tuned to zero vertical disparity" in that its optimum 2D disparity has zero vertical component, when it is probed at horizontal disparities on either side of the optimum, it responds best to vertical disparities on either side of zero. This suggests that, given cells tuned to a range of orientations and horizontal disparities, one could potentially extract the stimulus orientation, horizontal disparity *and* vertical disparity. Of course, it may not be quite that simple. In order to use the cells\' tuning to vertical disparity away from the optimal horizontal disparity, one has to know what the horizontal disparity is. Extracting this may be hard in the presence of vertical disparity, since then none of the cells in the population is tuned to the correct stimulus disparity. Also, because the tuning to vertical disparity occurs only at sub-optimal horizontal disparities, the neuron\'s activity is weaker, so more subject to noise. Thus, this intuitive idea has to be rigorously tested by simulation. This is what is achieved in this paper. {#pcbi-1000754-g002} The simulations consist of two neuronal populations: one encoding population, which takes left and right retinal images and performs the initial encoding of binocular disparity, and one decoding population, which estimates the disparity of the stimulus. The encoding population is like that in [Figure 1](#pcbi-1000754-g001){ref-type="fig"}: it consists of a set of neurons tuned to a range of horizontal disparities, orientations and spatial frequencies, but all tuned to the same vertical disparity. For simplicity, I shall set this vertical disparity to be zero, which is appropriate for the parafoveal region. The encoding neurons are based on the stereo energy model [@pcbi.1000754-Ohzawa1], normalized so as to report the effective local binocular correlation [@pcbi.1000754-Read3], [@pcbi.1000754-Banks1], [@pcbi.1000754-Filippini1]. The activity of this population is then decoded by a separate, higher-level population, using a template-matching approach like that of Tsai & Victor [@pcbi.1000754-Tsai1]. The synaptic weights from the encoding to the decoding population store the mean response of the population to stimuli with a range of different two-dimensional disparities. To estimate the two-dimensional disparity of a test image, I simply calculate the correlation between the population response to the test image, and the stored average population response for each known 2D disparity. The stimulus disparity is taken to be that giving the highest correlation, i.e. the best match to the mean response. Disparity encoding {#s2b} ------------------ ### Receptive fields {#s2b1} The monocular receptive fields were Gabor functions varying in their preferred orientation θ, spatial frequency f, receptive field size σ, receptive field phase φ, and position on the retina ([Figure 3](#pcbi-1000754-g003){ref-type="fig"}). The two receptive fields of a given binocular neurons always had the same orientation, frequency and size, but could differ in their phase and position, reflecting the properties of real neurons in primary visual cortex [@pcbi.1000754-DeAngelis1], [@pcbi.1000754-Anzai1], [@pcbi.1000754-Prince1], [@pcbi.1000754-Bridge1], [@pcbi.1000754-Read4]. Thus, the model binocular simple cells in general had both position and phase disparity [@pcbi.1000754-DeAngelis1]. All model binocular simple cells were tuned to the same cyclopean position, which was the origin. That is, the mean of the receptive field centers in the left and right eyes was (0,0) for all cells. {ref-type="supplementary-material"}.](pcbi.1000754.g003){#pcbi-1000754-g003} The aim of this study is to demonstrate that vertical disparity can be implicitly encoded by a population of neurons that are all tuned to a single vertical disparity. Here, I choose this single vertical disparity tuning to be zero, reflecting the vertical disparity expected at the fovea, (0,0). At other retinotopic locations, a different value would be appropriate, reflecting the expected vertical disparity at that location [@pcbi.1000754-Read2]. The particular value chosen is not important to the demonstration, only the fact that it is the same for all neurons in the population. Including phase disparity in the model makes this slightly more complicated, since for neurons tuned to non-vertical orientations, phase disparity adds both a horizontal and a vertical component to the preferred disparity. To deal with this, each neuron is given a position disparity chosen to cancel out the component introduced by the phase disparity. Thus, even in considering a single neuron, there are several different meanings of disparity to distinguish. In this paper, Δ*x* ~enc~ will indicate the preferred horizontal disparity of an encoding neuron, i.e. the horizontal disparity which elicits its maximum firing rate (the preferred vertical disparity of all encoding neurons is Δ*y* ~enc~ = 0). Δφ indicates the phase disparity of an encoding neuron. Finally (Δ*x* ~pos~,Δ*y* ~pos~) indicates the two-dimensional position disparity, chosen to beFor sufficiently narrow-band cells, this ensures that the neuron is tuned to the desired horizontal disparity of Δ*x* ~enc~, and to zero vertical disparity. The left and right eye receptive fields of the binocular simple cell tuned to orientation θ, frequency f, receptive field size σ, phase φ and horizontal disparity Δx are thenwhere x′ and y′ are retinal coordinates offset to the centre of the receptive field, and rotated to line up with the cell\'s preferred orientation: taking the + signs for *x*′~L~, *y*′~L~, and the − minus signs for *x*′~R~, *y*′~R~, and where the position disparity (Δ*x* ~pos~,Δ*y* ~pos~) is as specified in Equation 1. The population included a range of values for preferred orientation θ, spatial frequency *f*, receptive field size σ, phase φ, phase disparity Δφ and horizontal disparity Δ*x* ~enc~ , as follows: 1. Orientation θ: 6 values, −60°, −30°, 0°, 30°, 60° and 90°. 90° is horizontal, 0° is vertical. 2. Phase φ: 2 values, 0 or π/2 (this is all that is needed to achieve a phase-invariant complex cell) 3. Horizontal position disparity Δ*x* ~enc~: 21 values, −10 to 10 pixels in steps of 1 pixel. 4. Spatial frequency: 5 values, 0.200, 0.112, 0.0707, 0.0420, 0.0250 cycles per pixel, corresponding to spatial periods λ of 5.00, 8.41, 14.14, 23.81, 40.00 pixels. Receptive field size σ was set equal to 0.35λ. 5. Phase disparity Δφ: 5 values, 0, ±π/4 and ±π/2. Thus, there were 6×2×21×5×5 = 6300 binocular simple cells. These values were chosen to maximize physiological plausibility while giving reasonable simulation run-times. The different parameters have different effects on the model\'s performance. Self-evidently, sensitivity to a range of horizontal disparities is essential. The model\'s ability to extract the sign of vertical disparity depends on neurons tuned to oblique orientations ([Figure 2](#pcbi-1000754-g002){ref-type="fig"}). A range of spatial frequencies is not required for the model to extract vertical disparity in principle, but does improve the range of vertical disparity magnitudes over which the model performs well. For small vertical disparities, neurons tuned to high spatial frequencies are most sensitive to the disparity. For large vertical disparities, it is neurons tuned to low spatial frequencies which are most informative, since only these have receptive fields large enough to detect the disparity. A range of phase and phase disparity is not necessary for the model to work in principle, but helps to improve the model\'s accuracy [@pcbi.1000754-Read5]. ### Stereo energy model {#s2b2} The output from each receptive field was taken to be the inner product of each eye\'s image I(x,y) with the corresponding receptive field:and similarly for *v* ~R~. *I*(*x*,*y*) represents the contrast of the image at the point (*x*,*y*) relative to the mean luminance: positive values represent bright pixels, and negative values dark ones. In the standard energy model [@pcbi.1000754-Ohzawa1], [@pcbi.1000754-Ohzawa2], [@pcbi.1000754-Qian1], [@pcbi.1000754-Qian2], the response of binocular simple cells would beIt will be convenient to split this into monocular and binocular terms: Energy-model complex cells, which are invariant to stimulus phase, are built by summing the response of binocular simple cells tuned to different phases:As noted in the previous section, my population of simple cells includes only two values of phase, 90° apart. This produces the same results as summing over large number of simple cells with randomly-scattered phase, and is thus a widely-used short-cut in simulating complex-cell responses [@pcbi.1000754-Qian1], [@pcbi.1000754-Adelson1], [@pcbi.1000754-Fleet1]. The stereo energy, E, represents something close to the cross-correlation function between the filtered, windowed images. The problem with using this to extract stimulus disparity is that it reflects not only the degree of similarity between the shifted left- and right-eye images, but also their monocular contrast energy. Thus an energy-model unit may respond strongly either because it is genuinely tuned to the stimulus disparity, or because both its monocular receptive fields happen to contain features which drive them well -- whether or not those features match between the eyes. This makes it difficult to extract stimulus disparity from the stereo energy computed in Equation 3. ### Effective binocular correlation {#s2b3} To overcome this, I based my template-matching on the response of normalized correlation detectors [@pcbi.1000754-Read3], [@pcbi.1000754-Banks1], [@pcbi.1000754-Filippini1]. These are based on the stereo energy model, but are normalized so that their response ranges between +1 (when the left and right images are identical), and −1 (when the left image is an inverted version of the right). This is achieved by dividing the binocular terms of the energy-model complex cell by the monocular terms:Physiologically, this could be computed by combining the outputs of energy-model neurons with phase-disparities π apart. If two neurons are identical except that their phase-disparities are π apart, then if the first neuron computes E = (M+B), the second will compute (M−B). M and B are then available from the sum and difference of this pair of neurons. Thus the simulations implicitly use the full range of phase disparity, even though only phase disparity in the range \[−π/2,+π/2\] is explicitly simulated. The quantity *C* computes the correlation coefficient between filtered, local regions of the left and right eye\'s images [@pcbi.1000754-Read5]. It can be thought of as the effective binocular correlation experienced by that cell, and takes values in the range \[−1,1\]. To avoid any later confusion, note that this correlation is quite distinct from the Pearson product-moment correlation coefficient used below to assess how well population activity elicited by a test stimulus matches a template. I view the population of binocular correlation detectors, *C*(*θ*, *f*,Δφ,Δ*x* ~enc~), as performing the initial encoding of disparity within my model. Recall that there are 6 different orientations, 5 different frequencies, 5 different phase disparities and 21 different horizontal disparities, so the population *C*(*θ*, *f*,Δφ,Δ*x* ~enc~) consists of 3150 different correlation-detectors. Normalizing the stereo energy *E* so as to obtain the effective binocular correlation *C* removes the confounding effect of monocular contrast, making it much easier to extract the stimulus disparity from peaks in the population activity. *C* has the useful property that it is exactly equal to 1 when the stimulus disparity matches the cell\'s preferred disparity. This is true for *any* pair of stereo images, irrespective of spectral content etc, provided only that the left eye\'s image is related to the right eye\'s image by exactly the same offset relating left and right receptive fields. Under these circumstances, *v* ~L~(*θ, f,φ,Δφ,Δx* ~enc~) = *v* ~R~(*θ, f,φ,Δφ,Δx* ~enc~) for all *θ, f,φ,Δφ,Δx* ~enc~; 2*v* ~L~ *v* ~R~ is then the same as *v* ~L~ ^2^+*v* ~R~ ^2^, and it follows immediately that *C* = 1. ### Noise {#s2b4} As [Figure 2](#pcbi-1000754-g002){ref-type="fig"} makes clear, these neurons become effectively tuned to non-zero vertical disparities only when stimulated at their non-optimal horizontal disparity. Thus, in this model, vertical disparity is encoded only by neurons firing at below their optimal rate. Given this, it becomes important to be sure that this signal would not be lost in noise in a real neuronal population. To incorporate realistic neuronal noise, I convert the correlation *C*, which can take values \[−1,1\], into an observed spike count, which is necessarily positive or zero. First, I define the mean spike count, *R* ~m~, as *R* ~m~ = *U*(1+*C*), where *U* is the mean number of spikes elicited by a binocularly uncorrelated stimulus. R~m~ is in the range \[0,2*U*\], where 2*U* is the mean number of spikes a perfectly binocularly correlated stimulus elicits from neurons tuned to its disparity. I model neuronal noise as a Poisson process [@pcbi.1000754-Dean1], [@pcbi.1000754-Bair1]. Thus, the actual number of spikes elicited by the stimulus on any given presentation is *R*, where *R* is a random variable drawn from a Poisson distribution with mean *R* ~m~. The effective level of neuronal noise then depends on the value chosen for *U*. This will depend on the neurons\' maximal firing rate and the length of time assumed to be available for the judgment. If we assume that the firing rate for the optimal disparity is 100Hz [@pcbi.1000754-Prince2] and that the neuronal response is averaged over a 160ms window (since humans can discriminate temporal changes in disparity up to ∼6Hz, [@pcbi.1000754-Norcia1]), this suggests that the most active neurons might fire 16 spikes in the time available for a disparity judgment, yielding an estimate of around 8 spikes for *U*. Since the variance of Poisson noise is equal to its mean, larger values of *U* produce lower noise, and smaller values would mean greater neuronal noise. In fact, as I discuss below, the model is extremely resilient to neuronal noise. To demonstrate this, the results presented here use *U* = 1. This means that the average neuron fires only 1 spike in the time available for a perceptual judgment, resulting in a very large amount of neuronal noise (coefficient of variance 70% for even optimally-tuned neurons). Variation in the stimuli also contributes an additional effective source of noise. In this model, a stereo stimulus where left and right images are related simply by a shift will always produce an effective binocular correlation of *C* = 1 in neurons tuned to the disparity of the stimulus. However, neurons which are not tuned to the stimulus will produce a correlation which is on average less than 1, but whose precise value depends on the particular properties of the image, e.g. where the regions of high and low contrast happen to fall in relation to the receptive fields. When it comes to estimating the disparity of a single image, this stimulus-driven variation in response has the same deleterious effect as neuronal noise. If the stimulus disparity has a vertical component, it will stimulate none of the neurons optimally, meaning that *C* will be less than 1 (thus variable) for all neurons, and the neurons will be firing at a lower rate (thus subject to more Poisson noise). Thus, both sources of noise are larger for stimuli with vertical disparity. Disparity decoding {#s2c} ------------------ ### Storing templates {#s2c1} The first step was to generate many examples of the population\'s response to stimuli of known disparity. These "template" stimuli were uniform-disparity random noise patterns. Each pixel in the left eye\'s image, *I* ~L~, was given a random value drawn from a Gaussian with zero mean and unit standard deviation. The right eye\'s image, *I* ~R~, was offset horizontally and/or vertically from the first eye\'s image, and new random pixels were generated to fill the gap ([Figure 4](#pcbi-1000754-g004){ref-type="fig"}). {ref-type="supplementary-material"}.](pcbi.1000754.g004){#pcbi-1000754-g004} I produced random noise images with different horizontal and vertical disparities Δ*x* ~stim~ and Δ*y* ~stim~. Δ*x* ~stim~ and Δ*y* ~stim~ both ranged from −10 to 10 pixel in steps of 1 pixel, making a total of 441 different two-dimensional stimulus disparities. At each of these 441 stimulus disparities, I generated 500 random image-pairs, each generated with a different random seed *j*, making a total of 220,500 test stereograms. For each image-pair (Δ*x* ~stim~,Δ*y* ~stim~, *j*), I calculated the effective binocular correlation as described in Equation 4. I converted this to a mean spike count, and averaged this over 500 different random images, to obtain *W* is the mean number of spikes produced by sensors tuned to orientation θ, frequency *f*, phase disparity Δφ and horizontal disparity tuning Δ*x* ~enc~, when averaged over many different presentations of many different noise images with the same 2D stimulus disparity (Δ*x* ~stim~,Δ*y* ~stim~). The averaging over different presentations of the same image removes the neuronal noise, while the averaging over different images removes stimulus-dependent noise. I envisage this as representing the information stored in the system as a result of visual experience. ### Template matching {#s2c2} The disparity of an unknown test stimulus can then be estimated by comparing the response of the population to that particular test image with the stored, average response elicited by stimuli with known two-dimensional disparity. The stimulus is taken to have the 2D disparity whose stored activity profile best matches the current activity [@pcbi.1000754-Tsai1]. Let *R* ~test~(*θ*, *f*, Δ*φ*, Δ*x~enc~*) be the number of spikes fired by the encoding population to the particular test image under consideration. Remember that this neuronal population includes cells tuned to 6 different orientations θ, 5 different frequencies *f*, 5 different phase disparities and 21 different horizontal disparities Δ*x* ~enc~, so *R* ~test~(*θ*, *f*, Δ*φ*, Δ*x* ~enc~) is a set of 3150 individual spike-counts. To estimate the disparity of the test stimulus, I compare the population\'s response to the test image, *R* ~test~(θ, *f*, Δ*φ*, Δ*x~enc~*), with the stored mean spike-counts, W, for each of the 441 template stimulus disparities. That is, for each possible two-dimensional disparity (Δ*x* ~dec~, Δ*y* ~dec~) (subscript "dec" for decoding), I calculate the Pearson correlation coefficient, *r*(Δ*x* ~dec~, Δ*y* ~dec~), between the set of 3150 spike-counts obtained for this particular test image, *R* ~test~(θ, *f*, Δφ, Δ*x~enc~*), and the set of 3150 values stored in *W*(*θ*, *f*, Δ*φ*,Δ*x* ~enc~;Δ*x* ~dec~, Δ*y* ~dec~):where Corr(a,b) represents the usual Pearson product-moment correlation coefficient between a and b:where the sum Σ, averages \<\> and standard deviations std are all taken over *θ*, *f*, Δ*φ*, Δ*x* ~enc~, while holding Δ*x* ~dec~ and Δ*y* ~dec~ constant. I shall always use the word Pearson when referring to this correlation, in order to avoid possible confusion with the effective binocular correlation computed by the encoding neurons, Equation 4. In the figures, I shall use a "jet" colormap (running from blue-green-red) to represent spike-counts based on effective binocular correlation, and a "hot" colormap (black-red-yellow-white) to represent Pearson correlation. To model the lack of sensitivity to disparity in anti-correlated stereograms [@pcbi.1000754-Cogan1], [@pcbi.1000754-Read6], [@pcbi.1000754-Tanabe1], [@pcbi.1000754-Janssen1], [@pcbi.1000754-Cumming2], I finally set any negative correlations to zero, computingwhere ⌊⌋ indicates halfwave rectification: ⌊*x*⌋ = *x* for *x*\>0, and zero otherwise. The two-dimensional disparity of the test stimulus is then taken to be the values (Δ*x* ~dec~, Δ*y* ~dec~) which maximizes the halfwave-rectified Pearson correlation *P*(Δ*x* ~dec~,Δ*y* ~dec~). Matlab code (The Mathworks, Natick, MA; [www.mathworks.com](http://www.mathworks.com)) to run the simulations and generate most of the figures is available as Supplementary Material (although due to the size of the neuronal populations, running all the simulations presented in this paper takes weeks). Details of which functions to use are given in each figure legend. Other functions called by this code are grouped together in the file [Protocol S11](#pcbi.1000754.s011){ref-type="supplementary-material"}. Results {#s3} ======= All members of the neuronal population are tuned to zero vertical disparity {#s3a} --------------------------------------------------------------------------- First, it is important to establish that -- despite their wide range in phase disparity, position disparity and orientation -- all the units in our encoding population genuinely are tuned to zero vertical disparity. To this end, [Figure 5](#pcbi-1000754-g005){ref-type="fig"} shows two-dimensional disparity tuning surfaces for 15 example members of the model population of 3150 neurons. Disparity tuning surfaces like this have been measured for real neurons by Cumming [@pcbi.1000754-Cumming1], Durand et al, [@pcbi.1000754-Durand1], [@pcbi.1000754-Durand2] and Gonzalez et al [@pcbi.1000754-Gonzalez1]. Each panel in [Figure 5](#pcbi-1000754-g005){ref-type="fig"} shows the disparity tuning surface for a different model neuron in the encoding population. The pseudocolor represents the mean number of spikes fired by that neuron to stimuli with a given disparity, averaged over many different random noise images. All the neurons shown have the same spatial frequency, f = 0.071cyc/pix, and preferred horizontal disparity, Δ*x* ~enc~ = 6pix. The three rows show neurons tuned to different orientations: vertical, oblique and horizontal, as specified to the left of each row. The five columns show neurons with different phase-disparities Δ*φ*, as labelled at the top of each column. The phase disparity controls the symmetry of the disparity tuning surface: odd-symmetric for Δ*φ* = ±*π*/2, even-symmetric for Δ*φ* = 0, intermediate for Δ*φ* = ±*π*/4. As described in the [Methods](#s2){ref-type="sec"}, phase disparity shifts the preferred disparity in a direction orthogonal to the neuron\'s orientation. Model neurons in the encoding population were given just the right amount of position disparity (Equation 1) to cancel this out and place their peak sensitivity in the region expected for normal vision. This 2D position disparity (Δ*x* ~pos~,Δ*y* ~pos~) is indicated above each panel. When there is no phase disparity (Δ*φ* = 0, middle column), the position disparity is simply equal to the desired disparity tuning, here (6,0). Elsewhere, the model neurons have to be given additional amounts of vertical and/or horizontal position disparity in order to bring the preferred 2D disparity back to the desired value. The white cross in each panel marks the stimulus disparity which elicited the highest response from that neuron, averaged over the 500 images. In every case this is very close to (6,0), indicating that the position disparity specified in Equation 1 has had the desired effect. This was true for all 1350 neurons in our population, as well as the 15 examples shown in [Figure 5](#pcbi-1000754-g005){ref-type="fig"}, demonstrating that Equation 1 achieves its aim of making all neurons in the encoding population respond best to zero vertical disparity. {ref-type="supplementary-material"}, averaging over 500 stimuli, and the figure was generated with [Protocol S4](#pcbi.1000754.s004){ref-type="supplementary-material"}.](pcbi.1000754.g005){#pcbi-1000754-g005} Vertical disparity is implicitly encoded in the pattern of activity across the population {#s3b} ----------------------------------------------------------------------------------------- We now move to considering how stimulus vertical disparity is encoded within this population. To do this, instead of plotting the mean response of individual neurons to stimuli with different disparities, as was done in [Figure 5](#pcbi-1000754-g005){ref-type="fig"}, we now plot the mean response of many neurons to stimuli with a given disparity. This is what is shown in [Figure 6](#pcbi-1000754-g006){ref-type="fig"}. {ref-type="fig"}, where the neuron\'s preferred horizontal disparity was fixed in each panel and the horizontal axis was the horizontal disparity of the stimulus). Each panel shows the mean number of spikes which stimuli with this disparity elicit from 126 neurons, tuned to 21 different horizontal disparities Δx~enc~ and 6 orientations θ, plotted on the horizontal and vertical axes respectively. The 5 panels in each row show sets of 126 neurons tuned to 5 different preferred spatial frequencies. Thus together each row shows the mean response of the zero-phase-disparity sub-population, 630 neurons, averaged over 500 random stimuli with the same stimulus disparity. The stimulus horizontal disparity, Δx~stim~, was set equal to −2 pixels throughout (marked with the arrow in each panel); the stimulus vertical disparity, Δy~stim~, was set to a different value in each row, as indicated to the left of each row. The colorscale is the same as in [Figure 5](#pcbi-1000754-g005){ref-type="fig"}, indicated on the right. Matlab code: The mean responses were obtained with [Protocol S3](#pcbi.1000754.s003){ref-type="supplementary-material"}, and the figure was generated with [Protocol S5](#pcbi.1000754.s005){ref-type="supplementary-material"}.](pcbi.1000754.g006){#pcbi-1000754-g006} Each row of [Figure 6](#pcbi-1000754-g006){ref-type="fig"} shows the average spike count, *W*(*θ*, *f*,Δ*φ*,Δ*x* ~enc~;Δ*x* ~stim~,Δ*y* ~stim~), for all zero-phase-disparity neurons in the population, elicited by one particular stimulus disparity (Δ*x* ~stim~,Δ*y* ~stim~). (The choice to display the 630 neurons with Δφ = 0 is arbitrary; qualitatively similar plots are obtained for the other phase disparities.) The 6 rows show the response of this population to 6 different stimulus vertical disparities Δy~stim~, as indicated to the left of each row. In each case the stimulus horizontal disparity is Δ*x* ~stim~ = −2 pixels, marked with the arrow in each panel. Each panel shows *W*(*θ*, *f*,Δ*φ*,Δ*x* ~enc~;Δ*x* ~stim~,Δ*y* ~stim~) as a function of Δ*x* ~enc~ (horizontal axis) and θ (vertical axis), for the spatial frequency *f* indicated at the top of the column. Thus, the 6 rows of [Figure 6](#pcbi-1000754-g006){ref-type="fig"} correspond to 6 of the 441 stored responses of this population, which will be used in our template-matching algorithm to extract an estimate of stimulus disparity. The neurons above the arrow in each panel are those tuned to the horizontal disparity of the stimulus under consideration, Δ*x* ~enc~ = Δ*x* ~stim~. As one would expect, the effective correlation is generally high in this region (dark red colors). The stimulus vertical disparity Δ*y* ~stim~ is 4 pixels in row A, 2 pixels in row B, 0 pixels in row C, and so on as indicated to the left of each row. Although the cells in the population are tuned to many different horizontal disparities, Δ*x* ~enc~, they are all tuned to zero vertical disparity. Thus the middle row, [Figure 6C](#pcbi-1000754-g006){ref-type="fig"}, is the only case where any neurons are tuned to the exact two-dimensional disparity of the stimulus. Here, neurons with Δ*x* ~enc~ = Δ*x* ~stim~ = −2 have receptive fields which exactly match the binocular disparity of the stimulus. Their correlation is therefore *C* = 1 for every noise image with this disparity, and so the mean spike-count *W* = (1+*C*) is exactly 2. The mean spike-count falls below 2 to either side of the arrow, as the difference between the horizontal disparity of the stimulus and that preferred by the neurons increases. The rate of decrease depends on the spatial frequency channel, since this controls the size of the receptive fields. For the left-most column, *f* = 0.2 cycles/pixel, the standard deviation of the receptive field envelope, σ, is just 1.25 pixels. For the right-most column, *f* = 0.025 cycles/pixel and σ = 10 pixels, meaning that the effective correlation experienced by these neurons is still high even for neurons tuned to disparities several pixels away from the stimulus. The rate of decrease also depends on the orientation. In our model population, the receptive field envelopes are isotropic, but the rate of change of the receptive field function is still fastest orthogonal to the cell\'s preferred orientation *θ* (see [Figure 3](#pcbi-1000754-g003){ref-type="fig"}). Thus, for each spatial frequency channel, the rate of change along the horizontal direction is fastest for the vertically-oriented cells (*θ* = 0°), and slowest for the horizontally-oriented ones (*θ* = ±90°). This effect can be seen in [Figure 6C](#pcbi-1000754-g006){ref-type="fig"}: the red region of high correlation extends further to either side of the optimal disparity for the horizontally-oriented cells at the top and bottom of each panel. The same effect of receptive-field size can be seen as we look at rows other than row C, thus increasing the distance between the neurons\' preferred vertical disparity (0) and that of the stimulus. The peak response anywhere in the population declines as we move along a column away from Δ*y* ~stim~ = 0, as described by Read & Cumming [@pcbi.1000754-Read3]. Again, this decrease is most apparent for the higher-frequency channels, where receptive fields are smaller. For the highest-frequency channel (0.2 cyc/pix), where σ is just 1.25 pixels, a vertical disparity of −8 pixels (row F) is enough to make the portions of the images falling within the left and right-eye receptive fields completely uncorrelated. This means that the average binocular correlation is zero, and so with the spiking model I have adopted, the mean spike count is just 1, everywhere in the panel. The most interesting, and informative, panels of [Figure 6](#pcbi-1000754-g006){ref-type="fig"} are those where the stimulus has a non-zero, but relatively small, vertical disparity (rows A,B,D,E). Here, the effective binocular correlation C has fallen below 1, but is still above zero. In this case, the red region of high spike-counts takes on a distinctive diagonal slant, whose direction depends on the sign of stimulus vertical disparity. Where stimulus vertical disparity is positive (rows A, B), spike-counts are highest for receptive fields tilted counter-clockwise from vertical (positive *θ*) when horizontal disparity is positive, and for receptive fields tilted clockwise from vertical (negative *θ*) when horizontal disparity is negative. When stimulus vertical disparity is negative (rows D, E, F), the situation is reversed. The reason is exactly the geometry sketched in [Figure 2](#pcbi-1000754-g002){ref-type="fig"}. This slant is the "signature" of vertical disparity, and will enable us to decode vertical disparity from this population. 2D stimulus disparity can be extracted from the response of this population {#s3c} --------------------------------------------------------------------------- [Figure 6](#pcbi-1000754-g006){ref-type="fig"} showed the average response of a neuronal population, averaged across thousands of stimuli with the same disparity. As we have seen, this average response possesses a structure which reflects the vertical disparity of the stimulus. However, this averaging process conceals important features of the response to single images. Most importantly, the response of the neuronal population to single images is affected not only by the disparity, but also by the luminance features of the particular image. These features cancel out to nothing when averaged over many random images, but the brain cannot take advantage of this when estimating the disparity of a single image. The stereo correspondence problem is complicated by these "false matches" due to particular features of the image [@pcbi.1000754-Fleet1]. Normalizing stereo energy so as to calculate the effective binocular correlation *C* is enough to solve the problem in the absence of vertical disparity. Then, as explained in the [Methods](#s2){ref-type="sec"}, the stimulus horizontal disparity can be identified from the horizontal disparity tuning of the cell with *C* = 1 (mean spike count = 2*U*). However, when there is a mismatch between the cell\'s preferred vertical disparity and the vertical disparity of the stimulus, the correlation will not usually reach 1 even for cells tuned to the horizontal disparity of the stimulus, so the false-match problem creeps in again. Secondly, neuronal populations are subject to noise. In principle, this may be reduced by averaging either over a long time period, or over a large pool of neurons with similar tuning and independent noise. Here, I have made the conservative assumption that neither of these options is available, so the neuronal population is subject to very large amounts of trial-to-trial noise, with the coefficient of variation at least 70%. To bring home just how much variation these two sources of noise contribute, [Figure 7](#pcbi-1000754-g007){ref-type="fig"} shows the spikes elicited in response to a single example test image, with stimulus disparity Δ*x* ~stim~ = −2 and Δ*y* ~stim~ = +2 pixels. For comparison, [Figure 6B](#pcbi-1000754-g006){ref-type="fig"} showed the average response of the same population to stimuli with this disparity, with both neuronal and stimulus-driven noise averaged away. The 5 panels of [Figure 6B](#pcbi-1000754-g006){ref-type="fig"} are thus the "template" which [Figure 7](#pcbi-1000754-g007){ref-type="fig"} is meant to match (though note that because up to 6 spikes were produced by the single presentation in [Figure 7](#pcbi-1000754-g007){ref-type="fig"}, while the mean number of spikes never rises above 2, different colorscales were used in the two plots). At first glance, the task might appear to be impossible, given the very high levels of noise. However, certain features of similarity are indeed detectable between [Figure 7](#pcbi-1000754-g007){ref-type="fig"} and [Figure 6B](#pcbi-1000754-g006){ref-type="fig"}. At the lower spatial frequencies (right-hand panels), where the stimulus vertical disparity is not so large as a fraction of receptive field size, there is a slight tendency for neurons tuned to the horizontal disparity of the stimulus, marked with the arrows, to fire more spikes. Similarly, the slanted structure of the most responsive region is already hinted at. Furthermore, recall that for reasons of space, [Figures 6](#pcbi-1000754-g006){ref-type="fig"} and [7](#pcbi-1000754-g007){ref-type="fig"} show only the 630 neurons with zero phase-disparity; once we include the other phase disparities, there are a further 2520 neurons whose instantaneous response can be matched to the corresponding template. As I show below, despite the major differences between the single-image response shown in [Figure 7](#pcbi-1000754-g007){ref-type="fig"} and its template shown in [Figure 6B](#pcbi-1000754-g006){ref-type="fig"}, the population provides enough information for the correct template to be reliably identified. {ref-type="fig"}, only neurons with zero phase disparity are shown, Δφ = 0. The different panels each show 126 neurons tuned to different spatial frequencies *f*, while 21 preferred horizontal disparity tunings Δx~enc~ and 6 orientations θ are shown by the horizontal and vertical axes, respectively. In each panel, an arrow marks the neurons tuned to the horizontal disparity of the stimulus. The colorscale is the same in all panels. The average response of the population to all Gaussian-noise stimuli with this disparity was shown in [Figure 6B](#pcbi-1000754-g006){ref-type="fig"} (note different colorscale). This mean response differs from the single-stimulus response shown here because the latter is affected by stimulus-dependent variation, reflecting the properties of this particular image, and Poissonian noise on neuronal spiking. Matlab code: This figure was generated by [Protocol S6](#pcbi.1000754.s006){ref-type="supplementary-material"}.](pcbi.1000754.g007){#pcbi-1000754-g007} As described in the [Methods](#s2){ref-type="sec"}, I assess the quality of the match between the population response to a single image and the stored average population response by calculating the Pearson correlation coefficient between the two. [Figure 8](#pcbi-1000754-g008){ref-type="fig"} uses pseudocolor to show the Pearson correlation coefficients *r*(Δ*x* ~dec~,Δ*y* ~dec~) for all 441 disparities. The black cross marks the disparity of the stimulus. In this example, the highest Pearson correlation is obtained from the decoder tuned to this disparity, so for this single test image, the stimulus disparity is correctly extracted. {ref-type="supplementary-material"}.](pcbi.1000754.g008){#pcbi-1000754-g008} [Figure 9](#pcbi-1000754-g009){ref-type="fig"} quantifies the accuracy with which this algorithm performs across many test images. The plots show frequency histograms for the estimated disparity (red for horizontal disparity, blue for vertical) for 1000 different random test images with a fixed disparity. None of the 1000 test images was in the set of 500 images used to obtain the template responses, although they were all Gaussian noise images like those in [Figure 4](#pcbi-1000754-g004){ref-type="fig"}. Each column in [Figure 9](#pcbi-1000754-g009){ref-type="fig"} shows results for a different test disparity (Δ*x* ~test~,Δ*y* ~test~). The root-mean-squared error between the disparity estimated for each test image and its actual value is given above each panel. The algorithm\'s performance does not depend on the horizontal disparity of the test image (provided, of course, that it falls within the range to which the encoding population is tuned), so the three particular horizontal disparities chosen are immaterial. In contrast, performance does depend strongly on the vertical disparity tested. The three rows of [Figure 9](#pcbi-1000754-g009){ref-type="fig"} show results for increasing vertical disparity magnitudes: A: Δ*y* ~test~ = 0, B: Δ*y* ~test~ = 2, C: Δ*y* ~test~ = −4 pixels. {ref-type="fig"}. The root-mean-squared error between the estimated disparity and the correct value is indicated at the top of each panel. Matlab code: The disparity estimates were obtained with [Protocol S7](#pcbi.1000754.s007){ref-type="supplementary-material"}, and the figure was generated with [Protocol S8](#pcbi.1000754.s008){ref-type="supplementary-material"}.](pcbi.1000754.g009){#pcbi-1000754-g009} In [Figure 9A](#pcbi-1000754-g009){ref-type="fig"}, the test images had zero vertical disparity. Thus, the encoding population contains sensors tuned to the exact 2D disparity of the test images. Under these circumstances, unsurprisingly, both horizontal and vertical disparity are reconstructed with great accuracy. In [Figure 9B](#pcbi-1000754-g009){ref-type="fig"}, the test images had a vertical disparity of 2 pixels. An example population response to a single test image with this disparity was shown in [Figure 7](#pcbi-1000754-g007){ref-type="fig"}, while the template response (averaged over many training images with this disparity) was shown in [Figure 6B](#pcbi-1000754-g006){ref-type="fig"}. Here, no sensors in the encoding population are tuned to the 2D disparity of the stimulus. This naturally reduces the accuracy, but the RMS error is still only half a pixel. Critically, both the magnitude and sign of the vertical disparity can still be estimated from the reduction in the peak spike count [@pcbi.1000754-Read3] and the slant in the region of high spike count. [Figure 9C](#pcbi-1000754-g009){ref-type="fig"} shows results when the test images had a vertical disparity of −8 pixels. This is large compared to the receptive field size of most channels, so the RMS error increases further, but the sign of the vertical disparity is still reliably detected. Horizontal disparity is also extracted, but with a larger error which would correspond to a reduced stereoacuity. This is qualitatively consistent with human performance: human stereo perception becomes worse as vertical disparity increases, and is destroyed by relatively small amounts [@pcbi.1000754-Stevenson1], [@pcbi.1000754-Prazdny1]. Here, almost all the "work" is being done by the low spatial-frequency channels, but these are still enough to extract 2D disparity, without being excessively degraded by the higher-frequency channels for which the stimulus is effectively uncorrelated. Ultimately, of course, as vertical disparity moves beyond the range spanned by the largest receptive fields, performance will fall to chance, again as human performance does. Response to anti-correlated stereograms {#s3d} --------------------------------------- Disparity is encoded within this model by the population of binocular correlation detectors *C*(*θ*, *f*,Δ*x*). This population, which is all tuned to zero vertical disparity on the retina, performs the initial encoding of disparity. It was chosen to resemble primary visual cortex, V1. For example, these initial disparity encoders are tuned to a particular spatial frequency and orientation, and they continue to respond to disparity in anti-correlated stimuli. Anti-correlated stereograms are those in which one eye\'s image has been contrast-inverted, so that black pixels are replaced with white. Since I use zero to represent the mean luminance, this corresponds to inverting the sign of one eye\'s image. Thus, the product *v* ~L~ *v* ~R~ changes sign when the stimulus is made anti-correlated. This means that the disparity tuning of binocular correlation-detectors inverts for anti-correlated stimuli. A similar inversion is found in V1 [@pcbi.1000754-Cumming3], [@pcbi.1000754-Ohzawa3], although with a slight reduction in amplitude. Disparity is extracted from the activity of these V1 correlation-detectors by a higher-level brain area. The properties of this decoding area should ideally match those of human perception. For example, neurons in this region should not respond to disparity in anti-correlated stereograms, since these produce no perception of depth in humans or monkeys [@pcbi.1000754-Cogan1], [@pcbi.1000754-Read6], [@pcbi.1000754-Cumming2], and neurons in higher visual areas such as IT and V4 do not respond to disparity in anti-correlated stimuli [@pcbi.1000754-Tanabe1], [@pcbi.1000754-Janssen1]. In this paper, I have used the Pearson correlation coefficient, *r*, to quantify how well the population response to a test image matches the mean population response to template images. To match the lack of response to disparity in anti-correlated stereograms, I set the response of the decoding population equal to the half-wave-rectified Pearson correlation, replacing negative *r* with 0. This has no effect on correlated stereograms, where the maximum *r* is positive, but it prevents the decoder responding systematically to disparity in anti-correlated stereograms. [Figure 8](#pcbi-1000754-g008){ref-type="fig"} illustrated the response of the population of disparity decoders (prior to the half-wave rectification) to one example test stimulus, showing that the maximally-responding decoders were those tuned to disparities close to that of the stimulus. [Figure 10](#pcbi-1000754-g010){ref-type="fig"} plots the disparity tuning surface of a single disparity decoder, the one tuned to (Δ*x* ~stim~,Δ*y* ~stim~) = (−6,−3), for both correlated and anti-correlated stereograms. The pseudocolor of each pixel shows the mean \<*P*(Δ*x* ~stim~,Δ*y* ~stim~)\> averaged across 40 different random images with the same disparity (Δ*x* ~test~,Δ*y* ~test~), specified by the pixel\'s position on the axes. [Figure 10A](#pcbi-1000754-g010){ref-type="fig"} shows the disparity tuning surface for normal, correlated stereograms. Unsurprisingly, the response is largest when the two-dimensional disparity of the test stimulus matches the preferred disparity of the decoder, indicated with the cross. Similar disparity tuning surfaces were plotted in [Figure 5](#pcbi-1000754-g005){ref-type="fig"} for the encoding neurons. The disparity tuning surfaces for the decoding neurons differ in two respects. First, they are isotropic rather than elongated, because the decoding neurons receive inputs from cells tuned to all orientations ([Figure 11](#pcbi-1000754-g011){ref-type="fig"}). Second, the peak response is obtained for a non-zero vertical disparity, whereas the encoding neurons were all tuned to zero vertical disparity. {ref-type="supplementary-material"} and the figure was plotted by [Protocol S10](#pcbi.1000754.s010){ref-type="supplementary-material"}.](pcbi.1000754.g010){#pcbi-1000754-g010} {#pcbi-1000754-g011} [Figure 10B](#pcbi-1000754-g010){ref-type="fig"} shows the disparity tuning surface for the same decoder as in [Figure 10A](#pcbi-1000754-g010){ref-type="fig"}, but this time obtained with anti-correlated stereograms. As noted, anti-correlated stimuli elicit no perception of depth, and neurons in brain areas which are believed to have solved the correspondence problem do not discriminate disparity in anti-correlated stereograms. The Pearson correlation coefficient *r* between the response to an anti-correlated stereogram and the stored average responses for correlated stereograms is almost always negative, meaning that half-wave rectification ensures the decoder response *P*(Δ*x* ~stim~,Δ*y* ~stim~) is zero. Accordingly, the disparity tuning surface in [Figure 10B](#pcbi-1000754-g010){ref-type="fig"} is almost completely flat, in agreement with the physiological data for areas IT and V4 [@pcbi.1000754-Tanabe1], [@pcbi.1000754-Janssen1]. Thus, both encoding and decoding neurons in this simulation have properties consistent with those of the corresponding neuronal populations, as far as these are known. Discussion {#s4} ========== This paper has implemented a simple physiologically-inspired two-dimensional stereo correspondence algorithm. It consists of two model "brain areas": one which performs the initial encoding of binocular disparity between left and right images, and one which decodes this activity so as to arrive at an estimate of the two-dimensional disparity in the images. The unusual feature of this model is that the encoding neurons are all tuned to the same vertical disparity (zero). Despite this, the decoding neurons are able to successfully recover 2D stimulus disparity. This is possible because vertical disparity causes distinctive patterns of activity across the encoding population. The model uses its stored knowledge about these patterns, in the form of templates of expected activity, to deduce the stimulus disparity. Neuronal correlates {#s4a} ------------------- The model has a simple physiological interpretation. The population of disparity encoders, C(θ,f,Δx~enc~), was designed to represent primary visual cortex, V1. Neurons in this area are tuned to different orientations *θ*, spatial frequencies *f* and horizontal disparities Δ*x* ~enc~, and respond to disparity in anti-correlated stereograms. This encoding area projects to a higher brain area which extracts stimulus disparity. Neurons in this decoding area are tuned to both horizontal and vertical disparity, but are not sensitive to orientation or spatial frequency. They do not respond to disparity in anti-correlated stereograms. The perceived disparity corresponds to the preferred disparity of the most active neuron in the decoding area. The stored templates of the population activity expected for different stimulus disparities, *W*, can be viewed as the synaptic weights in the projection from the early encoding area to the decoding area ([Figure 11](#pcbi-1000754-g011){ref-type="fig"}). That is, *W*(*θ*, *f*,Δ*φ*,Δ*x* ~enc~;Δ*x* ~dec~,Δ*y* ~dec~) describes the strength of the synaptic connection from the encoding neuron tuned to orientation *θ*, frequency *f*, phase disparity Δ*φ* and horizontal disparity Δ*x* ~enc~, onto the decoding neuron tuned to horizontal disparity Δ*x* ~dec~ and vertical disparity Δ*y* ~dec~. The firing rate of the decoding neuron depends on the total activity of its input neurons weighted by the strength of each synapse (the term *Σ R* ~test~ *W* in Equation 6), after undergoing a subtractive and a divisive normalization, and finally a threshold non-linearity (Equation 7). The threshold non-linearity is a universal feature of neuronal circuits, since firing rates cannot go negative. Both subtractive and divisive normalization have been extensively discussed in the literature, and plausible neuronal mechanisms have been proposed to implement them [@pcbi.1000754-Carandini1], [@pcbi.1000754-Carandini2], [@pcbi.1000754-Heeger1], [@pcbi.1000754-Simoncelli1], [@pcbi.1000754-Tolhurst1], [@pcbi.1000754-Ayaz1]. Robustness to noise {#s4b} ------------------- This model is able to successfully decode two-dimensional disparity, including both the magnitude and sign of vertical disparity, from the activity of the encoding population. This demonstrates that information regarding vertical disparity is implicitly encoded within this population. The accuracy of this information, unsurprisingly, declines as the vertical disparity of the stimulus increases ([Figure 9](#pcbi-1000754-g009){ref-type="fig"}), consistent with psychophysical data. In the model, this decline occurs because information about the stimulus disparity is being carried by neurons which are not optimally tuned to it. Partly, this is because of neuronal noise: the effective signal-to-noise level declines as we move away from the peak of the neuron\'s tuning surface. I modelled neuronal spiking as a Poisson process, and deliberately chose a low spike count so that the Poisson noise would be large. In these simulations, neurons optimally tuned to the stimulus disparity have a coefficient of variation (CV, the ratio of standard deviation to mean) of 70%, while neurons which are tuned so far from the stimulus disparity that it appears effectively uncorrelated to them have a CV of 100%. However, the main reason for the decline in decoding accuracy is not neuronal noise, but fluctuations in the stimulus. For the uniform-disparity stimuli examined here, receptive fields tuned to the 2D stimulus disparity always experience an effective binocular correlation of exactly 1 (CV = 0%), whereas away from the 2D stimulus disparity the effective binocular correlation is, on average, smaller, and also much more variable. This means that as vertical disparity moves away from the value to which the neurons are tuned (here, zero), the stimulus-dependent fluctuations contribute much more variability to the neuronal spiking. Nevertheless, despite these two potent sources of noise in the model, the simulations reveal that it performs extremely well. This is because the decoding process uses the responses of thousands of encoding neurons. Although every neuron is tuned to different parameters, and so their responses cannot be directly pooled, the decoding process effectively averages out noise when it correlates the responses of thousands of neurons with the stored templates. For this reason, the model is extremely robust to neuronal noise. If the reader runs the code in the Supplementary Material, reducing the Poisson noise by setting `Neurons.MeanSpikeUncorr` to a value greater than its current value of 1, s/he will be able to verify that the results show only a slight improvement in accuracy. Relationship to previous models of vertical disparity encoding {#s4c} -------------------------------------------------------------- The model of Read & Cumming [@pcbi.1000754-Read3] was discussed in the [Introduction](#s1){ref-type="sec"}. That model worked by detecting changes in vertical disparity magnitude across the visual field. In contrast, the present model is purely local; all neurons simulated were tuned to the same cyclopean position in the visual field. This model would therefore work even with the induced-effect stimulus of Serrano-Pedraza & Read [@pcbi.1000754-SerranoPedraza2]. Serrano-Pedraza & Read [@pcbi.1000754-SerranoPedraza2] were correct to reject the particular decoding model proposed by Read & Cumming [@pcbi.1000754-Read3], but wrong to conclude that vertical disparity must be explicitly encoded. A more sophisticated decoding of the same encoding population is consistent with their psychophysical results. Matthews et al. [@pcbi.1000754-Matthews1] also modelled the perceptual effects of vertical disparity using energy-model neurons with different orientation tuning. The present algorithm differs substantially from theirs. Most importantly, their model does not ever estimate stimulus vertical disparity. Their decoding algorithm extracts a one-dimensional estimate of horizontal disparity, assuming that vertical disparity is zero. This means that when vertical disparity actually is present, it causes horizontal disparity to be mis-estimated: a vertical disparity V is misinterpreted as a horizontal disparity of Vcotθ, where θ is the cell\'s preferred orientation relative to horizontal (eq. 6 of Matthews et al.). They postulate that the perceptual effects of vertical disparity are a direct consequence of this confusion between horizontal and vertical disparity components. In contrast, the present model explicitly decodes both horizontal and vertical disparity. Vertical disparity does not cause horizontal disparity to be systematically mis-estimated (although it does increase the random error, [Figure 9](#pcbi-1000754-g009){ref-type="fig"}). Thus, the present model is agnostic on the question of how vertical disparity causes its perceptual effects: the two-dimensional disparity decoded by the present algorithm would have to be fed into one of the many models of that process (e.g. [@pcbi.1000754-Read2], [@pcbi.1000754-Garding1],[@pcbi.1000754-Backus1],[@pcbi.1000754-Gillam1],[@pcbi.1000754-Mayhew1]. Second, in order to explain how the "mistaken" disparity *V*cot*θ* produces a perceptual effect when averaged over neurons tuned to all possible orientations *θ*, Matthews et al. [@pcbi.1000754-Matthews1] invoke a radial bias for *θ* [@pcbi.1000754-Leventhal1], [@pcbi.1000754-Vidyasagar1], [@pcbi.1000754-Bauer1]. The present algorithm does not depend on any such anisotropy. In the simulations presented here, *θ* was assumed to be isotropic; any anisotropy would not affect the performance of the algorithm. This means that the present model is almost the opposite of that in Matthews et al. Their neuronal population explicitly encodes both horizontal and vertical disparity, but their decoding algorithm deliberately extracts only horizontal disparity. My population explicitly encodes only horizontal disparity, but my decoding algorithm extracts both horizontal and vertical disparity. Consistency with known physiology {#s4d} --------------------------------- As sketched in [Figure 2](#pcbi-1000754-g002){ref-type="fig"}, the present algorithm depends critically on the obliquely-oriented disparity-tuning surfaces predicted by the stereo energy model. It is therefore important to know whether real neurons display such oriented disparity-tuning surfaces. In monkey V1, Cumming [@pcbi.1000754-Cumming1] examined two-dimensional disparity-tuning surfaces for random-dot patterns, and compared their orientation to the cell\'s orientation tuning for grating stimuli. He found many cells with the obliquely-oriented disparity tuning used here. However, most cells had disparity-tuning surfaces elongated along the horizontal axis, independent of the cell\'s orientation tuning for gratings. Cumming argued that this represented a specialization for horizontal disparity not predicted by the energy model. This non-energy-model population can be modeled by combining several energy-model units with different horizontal disparity tuning [@pcbi.1000754-Read1]. The oblique disparity tuning predicted by the energy model is also found in cat visual cortex [@pcbi.1000754-Sasaki1], and in peripheral monkey V1 [@pcbi.1000754-Durand2]. Thus, the existing physiological evidence suggests that neurons with the obliquely-oriented disparity-tuning surfaces used by this model do exist, and may form the inputs for a second stage of disparity encoding consisting of neurons with horizontally-oriented disparity-tuning surfaces. Neurons in V1 contain both position and phase disparity [@pcbi.1000754-DeAngelis1], [@pcbi.1000754-Anzai1], [@pcbi.1000754-Prince1], [@pcbi.1000754-Anzai2]. The model presented here works equally well whether position disparity alone, or both position and phase disparity, are included. In this paper, I specified a relationship between position disparity, phase disparity, frequency and orientation (Equation 1) which ensured that all neurons in the population were tuned to zero vertical disparity. (If this relationship did not hold, the model would contain neurons tuned to a range of vertical disparities, so its success would be trivial.) No physiological study has yet quantified both phase disparity and vertical disparity tuning, yet the results of [@pcbi.1000754-Cumming1] imply that something like Equation 1 may hold in reality, at least in the central 10° or so of the visual field. In the visual periphery, very little is currently known about the distribution of 2D retinal disparity, despite the fact that this is where the range of naturally-occurring vertical disparities is largest [@pcbi.1000754-Read2], [@pcbi.1000754-Rogers2]. The existing physiological studies have reported their results only in head-centric Helmholtz coordinates, and have not examined tuning as a function of position on the retina. The encoding population described here, where all neurons at a given retinotopic location are tuned to the same vertical disparity on the retina ([Figure 1](#pcbi-1000754-g001){ref-type="fig"}), is consistent with the very limited existing physiological data available [@pcbi.1000754-Read3]. Only future physiological studies can resolve the issue. These should obtain a full 2D disparity tuning surface for every neuron; as [Figure 2](#pcbi-1000754-g002){ref-type="fig"} shows, 1D cross-sections can give misleading results. They should be clear about the definition of vertical disparity they are using, reporting data in retinal, as well as head-centric, coordinates. Finally, they need to examine disparity tuning as a function of position on the retina (not just eccentricity), in order to test whether the mean and variation in preferred vertical disparity varies across the retina as predicted from natural image statistics [@pcbi.1000754-Read2]. These studies should be carried out in both early visual cortex and in higher areas such as IT believed to underlie perception. The present model predicts that the range of preferred vertical disparities will be larger in the higher cortical areas. Significance {#s4e} ------------ This paper demonstrates a highly efficient strategy for representing 2D stimulus disparity. 2D disparity is represented explicitly only at the decoding level, with the initial encoding being one-dimensional. Because the disparity decoding area does not represent other stimulus properties such as orientation, spatial frequency and phase, this results in a huge reduction in the number of neurons required. Irrespective of whether the model here is ultimately validated physiologically, it nevertheless provides a vivid demonstration that populations of disparity-tuned neurons contain a much richer array of information than previously appreciated. It places a caveat on the common wisdom that in order to encode a quantity *X*, a neuronal population needs to be tuned to a range of values of *X*. In this example, horizontal and vertical disparity are completely independent quantities in the external world, but they are bound together with orientation at the initial encoding stage in the brain. Subsequently, vertical disparity can be extracted from neurons via their tuning to horizontal disparity and orientation alone. Under these very special circumstances, the common wisdom ceases to hold. Supporting Information {#s5} ====================== ###### Matlab code for running the simulations presented in this paper (Fig_ExampleRFs.m) (1.00 KB TXT) ###### Click here for additional data file. ###### Matlab code for generating [Fig 4](#pcbi-1000754-g004){ref-type="fig"} (Fig_ExampleImages.m) (1.00 KB TXT) ###### Click here for additional data file. ###### Matlab code ([www.mathworks.com](http://www.mathworks.com)) for running the simulations presented in this paper (gets templates). GetTemplates.m (1.00 KB TXT) ###### Click here for additional data file. ###### Matlab file, for generating [Fig 5](#pcbi-1000754-g005){ref-type="fig"} (Fig_DispTunSurfEncoders.m) (3.00 KB TXT) ###### Click here for additional data file. ###### Matlab code ([www.mathworks.com](http://www.mathworks.com)) for running the simulations presented in this paper (Fig_MeanResponses.m) (2.00 KB TXT) ###### Click here for additional data file. ###### Matlab code ([www.mathworks.com](http://www.mathworks.com)) for running the simulations presented in this paper (Fig_FitDisparity.m) (3.00 KB TXT) ###### Click here for additional data file. ###### Matlab code ([www.mathworks.com](http://www.mathworks.com)) for running the simulations presented in this paper (FitDisparity.m) (2.00 KB TXT) ###### Click here for additional data file. ###### Matlab code ([www.mathworks.com](http://www.mathworks.com)) for running the simulations presented in this paper (Fig_FreqHists.m) (3.00 KB TXT) ###### Click here for additional data file. ###### Matlab code for generating [Fig 10](#pcbi-1000754-g010){ref-type="fig"} (DispTunSurfDecoders.m) (2.00 KB TXT) ###### Click here for additional data file. ###### Matlab code for generating [Fig 10](#pcbi-1000754-g010){ref-type="fig"} (Fig_DispTunSurfDecoders.m) (1.00 KB TXT) ###### Click here for additional data file. ###### Zip archive containing 7 files with Matlab functions necessary to run the simulations and generate the figures in the paper (Protocol_S11.zip) (0.14 MB ZIP) ###### Click here for additional data file. Thanks to Bruce Cumming for helpful discussions, and to Ignacio Serrano-Pedraza for helpful comments on the manuscript. The author has declared that no competing interests exist. This research was supported by Royal Society University Research Fellowship UF041260 ([www.royalsociety.org](http://www.royalsociety.org)) and Medical Research Council New Investigator Award 80154 ([www.mrc.ac.uk](http://www.mrc.ac.uk)). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: JCAR. Performed the experiments: JCAR. Analyzed the data: JCAR. Wrote the paper: JCAR.

1. Introduction =============== Tumor angiogenesis, that is, the development of new blood vessels from pre-existing ones, plays an essential role in tumor growth, invasiveness, and metastasis \[[@B1-molecules-17-06854],[@B2-molecules-17-06854]\]. It is estimated that angiogenesis in tumors contributes to more than 90% of all cancer deaths. Tumor angiogenesis is a promising therapeutic target for treatment of cancer. Angiogenesis is initiated by cell proliferation and migration in response to chemotactic agents, such as VEGF, which is widely expressed in the majority of cancers and is a critical component of tumor angiogenesis \[[@B3-molecules-17-06854]\]. VEGF mediates its signals through interactions with the receptor tyrosine kinases expressed on endothelial cells \[[@B4-molecules-17-06854]\]. VEGFRs activation leads to the activation of diverse intracellular signaling molecules, including extracellular signal-regulated kinases (ERKs), phosphoinositide 3-kinase/AKT kinase, and mammalian target of rapamycin (mTOR)/ribosomal protein S6 kinase (p70S6K) \[[@B5-molecules-17-06854],[@B6-molecules-17-06854]\] that promote the proliferation, migration, differentiation, and survival of endothelial cells in the pre-existing vasculature. Thus, VEGFR2 and those intracellular signaling molecules appear to be critical targets for the suppression of tumor angiogenesis. Multiple angiogenesis inhibitors have been therapeutically validated in treatments for many cancers \[[@B7-molecules-17-06854]\]. Pristimerin ([Figure 1](#molecules-17-06854-f001){ref-type="fig"}) is a natural triterpenoid isolated from Celastrus and Maytenus spp. \[[@B8-molecules-17-06854]\] that has been shown to possess a variety of biological activities. Antitumor properties have also been reported for pristimerin. Previous studies have shown that pristimerin can inhibit tumor cell proliferation by inhibiting the NF-κB pathway and cell cycling \[[@B8-molecules-17-06854],[@B9-molecules-17-06854],[@B10-molecules-17-06854],[@B11-molecules-17-06854],[@B12-molecules-17-06854]\]. In addition, pristimerin has been reported to induce caspase-dependent apoptosis of breast cancer cells \[[@B13-molecules-17-06854]\] and prostate cancer cells *in vitro* \[[@B11-molecules-17-06854]\]. Recently, pristimerin is reported to inhibit xenografted plasmacytoma tumors in mice through the suppression of 20s proteasome chymotrypsin-like activity \[[@B9-molecules-17-06854]\]. Despite reports of the antitumor properties of pristimerin, whether this compound directly affects tumor angiogenesis or has potential value for breast cancer prevention *in vivo* remains unknown. {#molecules-17-06854-f001} In this study, we investigated how pristimerin inhibits tumor angiogenesis by targeting key signaling pathways. Our results provide the first evidence that pristimerin significantly inhibits VEGF-stimulated endothelial cell proliferation, migration, tube formation, and tumor angiogenesis by targeting VEGFR2 activation, leading to the suppression of tumor growth. 2. Results and Discussion ========================= 2.1. Pristimerin Inhibits Tumor Growth and Tumor Angiogenesis in a Xenograft Mouse Model ---------------------------------------------------------------------------------------- To investigate the effects of pristimerin on tumor growth and tumor angiogenesis *in vivo*, we used a xenograft human breast cancer model and found that 3 mg/kg of pristimerin applied every other day significantly reduced both tumor volume ([Figure 2](#molecules-17-06854-f002){ref-type="fig"}A) and tumor weight ([Figure 2](#molecules-17-06854-f002){ref-type="fig"}B). As shown in [Figure 2](#molecules-17-06854-f002){ref-type="fig"}A, the tumors in the control group increased in size from 114.95 ± 27.50 to 501.06 ± 135.10 mm^3^, whereas the tumors in the pristimerin-treated group shrank from 115.76 ± 29.80 to 109.32 ± 54.40 mm^3^. Additionally, the average final tumor weight in the pristimerin-treated group was dramatically reduced compaired with that in the control group ([Figure 2](#molecules-17-06854-f002){ref-type="fig"}B), suggesting that pristimerin strongly inhibited tumor growth in this xenograft mouse breast tumor model. To investigate further whether pristimerin inhibited tumor angiogenesis in the xenograft mouse model, we stained solid tumor sections with a blood vessel staining kit. The mean number of blood vessels in the pristimerin-treated group was 37.06 ± 19.09/HPF compared with 100 ± 24.87/HPF in the control group ([Figure 2](#molecules-17-06854-f002){ref-type="fig"}D), indicating that pristimerin significantly inhibited tumor angiogenesis. {#molecules-17-06854-f002} 2.2. Pristimerin Inhibits Angiogenesis in Vivo and VEGF-induced Vessel Sprouting *ex Vivo* ------------------------------------------------------------------------------------------ We performed a chick embryo chorioallantoic membrane (CAM) assay to determine the antiangiogenic effects of pristimerin *in vivo*. As illustrated in [Figure 3](#molecules-17-06854-f003){ref-type="fig"}A, new blood vessels formed well on CAMs in the control group after a 2-day incubation, whereas incubation with pristimerin at 40 nmol/egg resulted in a notable inhibition, and pristimerin at 80 nmol/egg drastically impaired CAM neovascularization accompanied by a lack of prominent vessel networks. These results demonstrate that pristimerin suppresses angiogenesis in a CAM model. The aortic ring assay mimics several stages in angiogenesis, including endothelial cell proliferation, migration, and tube formation and is widely used to evaluate the antiangiogenic effects of putative therapeutic compounds \[[@B14-molecules-17-06854]\]. To further investigate whether pristimerin inhibited VEGF-induced angiogenesis *ex vivo*, we examined the sprouting of vessels from aortic rings in the absence or presence of pristimerin. VEGF (20 ng/mL) significantly stimulated microvessel sprouting around the aortic rings ([Figure 3](#molecules-17-06854-f003){ref-type="fig"}B). Treatment with pristimerin antagonized the VEGF-induced sprouting in a dose-dependent manner. Treatment with pristimerin at 1 μM yielded a notable suppression of microvessel formation *versus* the control ([Figure 3](#molecules-17-06854-f003){ref-type="fig"}A), and 2 μM of pristimerin completely blocked the VEGF-induced microvessel sprouting of rat aortic rings ([Figure 3](#molecules-17-06854-f003){ref-type="fig"}B). The presence of pristimerin resulted in decreasing capillary sprouting from the rat aorta rings. {#molecules-17-06854-f003} 2.3. Pristimerin Inhibits the VEGF-Induced Chemotactic Motility, Capillary Structure Formation and Proliferation of HUVECs *in Vitro* ------------------------------------------------------------------------------------------------------------------------------------- The migration of endothelial cells is essential for blood vessel formation in angiogenesis and thus for tumor growth. The effects of pristimerin on the chemotactic motility of HUVECs were assessed in a wound-healing migration assay and in a Transwell cell migration assay. As shown in [Figure 4](#molecules-17-06854-f004){ref-type="fig"}A, pristimerin significantly inhibited HUVEC migration in the wound-healing migration assay and dramatically reduced the VEGF-induced migration at 1 μM in the Transwell assay ([Figure 4](#molecules-17-06854-f004){ref-type="fig"}B). Pristimerin inhibits VEGF-induced HUVEC chemotactic motility in a concentration-dependent manner. Tube formation by endothelial cells is one of the key steps of angiogenesis. We next evaluated the effects of pristimerin on tube formation by HUVECs in Matrigel assays. In this assay, HUVECs become elongated and form capillary-like structures on Matrigel \[[@B15-molecules-17-06854]\]. Stimulation with 10 ng/mL VEGF promoted the differentiation of HUVECs to form robust tubular-like structures ([Figure 4](#molecules-17-06854-f004){ref-type="fig"}C). Pristimerin significantly inhibited the VEGF-induced tube formation by HUVECs on Matrigel at 0.5 μM, suggesting the potential effect of pristimerin on angiogenesis Angiogenesis requires the proliferation of endothelial cells. We next examined the inhibitory effects of pristimerin on VEGF-induced HUVEC proliferation using a cell proliferation assay involving bromodeoxyuridine (BrdU) incorporation. As shown in [Figure 3](#molecules-17-06854-f003){ref-type="fig"}C, 0.5 μM pristimerin significantly decreased the VEGF-induced proliferation of HUVECs ([Figure 4](#molecules-17-06854-f004){ref-type="fig"}D). To determine if the effect of pristimerin on the tumor vasculature, we evaluated the MDA-MB-231 cell viability after 24 h of pristimerin treatment. Pristimerin at the same dose had less inhibitory effect on MDA-MB-231 cell viability (\<5%). Pristimerin had higher inhibitory effect on endothelial cells than that on cancer cells. These results demonstrate that pristimerin blocks VEGF-induced *in vitro* angiogenesis by inhibiting cell proliferation, chemotactic motility and tube formation. {#molecules-17-06854-f004} 2.4. Influence of Pristimerin on VEGFR2 and Related Signaling Pathways ---------------------------------------------------------------------- To determine the molecular basis of pristimerin-mediated antiangiogenesis, we examined the signaling molecules and pathways activated by the interaction of VEGFR2 with VEGF using Western blotting assays. [Figure 5](#molecules-17-06854-f005){ref-type="fig"}A illustrates how pristimerin inhibited VEGF-activated VEGFR2 phosphorylation in a dose-dependent manner. VEGFR2 activation leads to the activation of diverse intracellular signaling molecules that are responsible for endothelial cell migration, proliferation, and survival. To further investigate the intracellular signaling pathway affected by pristimerin, we screened some key kinases involved in angiogenesis signaling. We found that 0.5 and 1 μM of pristimerin significantly suppressed the activation of AKT and ERK1/2 ([Figure 5](#molecules-17-06854-f005){ref-type="fig"}A), respectively, and that 2 μM of pristimerin significantly inhibited the activation of mTOR and p70S6K ([Figure 5](#molecules-17-06854-f005){ref-type="fig"}B), which suggested that pristimerin exerted its antiangiogenic effects through the inhibition of VEGFR2 activation on the surfaces of endothelial cells and suppression of the AKT/mTOR/S6K kinase signaling pathway. {#molecules-17-06854-f005} 2.5. Discussion --------------- In recent years, increasing effort has been focused on the identification of new anti-cancer compounds. Phytochemicals play an essential role in the search for new anticancer molecules; triterpenoids, possessing a wide range of unique biological activities, have received the most attention in this area \[[@B16-molecules-17-06854],[@B17-molecules-17-06854],[@B18-molecules-17-06854]\]. Pristimerin is a triterpenoid isolated from Celastrus and Maytenus spp., and antitumor properties have also been reported for pristimerin \[[@B9-molecules-17-06854],[@B10-molecules-17-06854],[@B11-molecules-17-06854],[@B19-molecules-17-06854]\]. In this study, we found that pristimerin displays potent antiangiogenic activities both *in vitro* and *in vivo*. Pristimerin inhibits tumor angiogenesis in a xenograft mouse breast tumor model, reducing the neovascularization of chicken chorioallantoic membrane *in vivo* and abrogating VEGF-induced microvessel sprouting in an *ex vivo* rat aortic ring assay. Pristimerin blocks VEGF-induced *in vitro* angiogenesis by inhibiting cell proliferation, chemotactic motility and tube formation. It is noteworthy that pristimerin at the same concentration exhibits a higher potency to suppress proliferation of activated endothelial cells (at angiogenic state) compared with that of cancer cells, suggesting pristimerin is a relatively affordable drug that specifically targets activated endothelial cells. Pristimerin exhibits a potent inhibition of angiogenesis. When angiogenesis was suppressed by pristimerin, tumor growth was substantially inhibited. VEGF is the primary and the most potent inducer of angiogenesis. VEGF signaling events related to tumor growth and angiogenesis are mainly mediated by VEGFR2 \[[@B4-molecules-17-06854]\]. VEGF-A and VEGFR2 play crucial roles in vessel sprouting and new vessel initiation through the induction of the proliferation, migration, and survival of endothelial cells \[[@B20-molecules-17-06854]\]. We found that pristimerin significantly downregulated the phosphorylation of VEGFR2 induced by VEGF in a dose-dependent manner. VEGFR2 receptor tyrosine phosphorylation induced by binding with VEGF can stimulate intracellular signaling pathways, such as the MAPK/ERK pathway and the phosphatidylinositol 3-kinase/AKT pathway \[[@B21-molecules-17-06854]\]. We found that pristimerin caused a downregulation of the VEGF-induced phosphorylation of both ERK1/2 and Akt, which are the key signaling molecules of these signaling pathways. ERK1/2 activation exerts its regulatory effects on the proliferation, differentiation and survival of endothelial cells \[[@B22-molecules-17-06854]\], and the protein AKT is importantly involved in cell proliferation, survival and migration in endothelial cells \[[@B23-molecules-17-06854]\]. Our results in the VEGF stimulation models *in vitro* showed that pristimerin inhibits multiple steps of angiogenesis including VEGF-induced cell proliferation, motility, and tube formation. The inhibitory effect of pristimerin on the cell proliferation, motility, and tube formation of HUVECs appeared to be associated with its ability to suppress the phosphorylation of VEGFR2 and ERK1/2 and Akt. Our results demonstrate that pristimerin potently inhibits angiogenesis by significantly inhibiting VEGFR2 activation. Notably, pristimerin has a more potent inhibitory effect on phosphorylation of Akt than on the phosphorylation of VEGFR2. This highly inhibitory effect suggests that pristimerin may also affect other pathway involved in regulating angiogenesis. Recently, the Akt/mTOR/P70S6K signaling pathway has been identified as a novel functional target in angiogenesis \[[@B6-molecules-17-06854],[@B18-molecules-17-06854],[@B24-molecules-17-06854]\]. mTOR is critical for cellular proliferation and growth in endothelial cells \[[@B25-molecules-17-06854]\]. The S6K1 (ribosomal p70S6 kinase) protein is mainly a downstream effector of mTORC1, leading to the initiation of translation and thereby increasing protein synthesis \[[@B26-molecules-17-06854]\], additionally regulating cell migration by inducing actin filament remodeling to form filopodia and lamellipodia structures \[[@B27-molecules-17-06854]\]. Our present data demonstrate that pristimerin significantly inhibits the activation of AKT/mTOR pathway in endothelial cells, including AKT, mTOR kinase and its downstream ribosomal S6 kinase, in a concentration-dependent manner. This inhibitory effect of pristimerin on the AKT/mTOR/P70S6K signaling pathway may, at least in part, contribute to the inhibition of endothelial cell proliferation, migration, and capillary structure formation. Thus, the antiangiogenic properties of pristimerin may be partially due to its inhibition of the AKT/mTOR/P70S6K signaling pathway. Importantly, our study revealed that pristimerin decreased the VEGF secretion of MDA-MB-231 cells and had no effect on MDA-MB-231 cell viability at the same concentration (data not shown). Pristimerin exerts inhibitory effects on NFκB activation \[[@B9-molecules-17-06854],[@B28-molecules-17-06854]\], which is known to regulate the expression of angiogenic gene products; Additionally, the mTOR/p70S6K pathway has been found to regulate the expression of hypoxia-inducible factor-1 α to influence the secretion of proangiogenic factors in various human carcinomas \[[@B29-molecules-17-06854],[@B30-molecules-17-06854],[@B31-molecules-17-06854]\]. Previous studies have examined the effects of mTORC1/mTORC2 inhibition on VEGF secretion *in vitro* and *in vivo* \[[@B6-molecules-17-06854],[@B31-molecules-17-06854]\]. All the above mechanisms might contribute to this compound's antiangiogenic properties in the xenograft mouse tumor model. We are interested in investigating the mechanism of the decreased secretion of proangiogenic molecules (VEGF) by pristimerin. 3. Experimental =============== 3.1. Cell lines and Cell Culture -------------------------------- Primary human umbilical vascular endothelial cells (HUVECs) were isolated from human umbilical cord veins by collagenase treatment as described previously \[[@B14-molecules-17-06854]\]. The HUVECs were cultured in endothelial cell growth medium (ECGM): M199 medium (Gibco-BRL, Grand Island, NY, USA) supplemented with 20 μg/mL bovine endothelial cell growth factor (Roche Molecular Biochemicals, Laval, Quebec, Canada), penicillin (50 IU/L), streptomycin (50 mg/L), 50 μg/mL amphotericin B, 0.1 mg/mL heparin (Sigma, St. Louis, MO, USA), 15 mmol/L HEPES buffer, NaHCO~3~ (44 mmol/L) and 20% fetal bovine serum (FBS, Gibco-BRL) at 37 °C under a humidified 95%:5% (v/v) mixture of air and CO~2~, as described previously \[[@B32-molecules-17-06854]\]. Human breast cancer (MDA-MB-231) cells were purchased from the American Type Culture Collection and cultured in Leibovitz\'s L-15 medium (Gibco-BRL). supplemented with 10% FBS (Gibco-BRL) at 37 °C under a humidified 100% air atmosphere. 3.2. Reagents and Antibodies ---------------------------- A 10 mM stock solution of pristimerin (purchased from Enzo Life Sciences, Lausen, Switzerland, purity ≥98%) was stored at −20 °C in small aliquots protected from light and was diluted to the required concentrations in cell culture medium. Recombinant human VEGF 165 was purchased from R&D Systems (Minneapolis, MN, USA), and growth factor-reduced Matrigel was purchased from BD Biosciences (Bedford, MA, USA). The antibodies against β-actin were obtained from Santa Cruz Biotechnology (Santa Cruz, CA, USA). The anti-AKT, anti-ERK1/2, anti-VEGFR-2, anti-mTOR, anti-p70S6K1, and phosphorylated-specific anti-AKT (Ser473), anti-ERK1/2 (Thr202/Tyr204), anti-VEGFR2 (Tyr1175), anti-mTOR (Ser2448), and anti-p70S6K1 (Thr389) antibodies were purchased from Cell Signaling Technology (Danvers, MA, USA). 3.3. Human Breast Tumor Xenograft Mouse Model --------------------------------------------- The experimental protocol was established according to the NIH guidelines for Animal Research and Care and was approved by the Institutional Animal Care and Use Committee. A xenograft human breast cancer mouse model assay was performed as previously described \[[@B33-molecules-17-06854]\]. Six-week-old female BALB/c nude mice (National Rodent Laboratory Animal Resources, Shanghai, China) weighing approximately 22 g were randomly divided into two groups of six each. MDA-MB-231 cells were s.c. injected into the mice (8 × 10^6^ per mouse). After the tumors grew to approximately 100 mm^3^, the mice were injected subcutaneously with either vehicle or pristimerin (3 mg/kg) every other day for 16 days. The tumor size was measured with a caliper (calculated volume = shortest diameter^2^ × longest diameter/2) at two day intervals. 3.4. Histology and Immunohistochemistry --------------------------------------- Solid tumors were removed and fixed with 10% formaldehyde and embedded in paraffin. Specific blood vessel staining was performed on 5-μm deparaffinized sections with a specific blood vessel staining kit (von Willebrand Factor; Millipore, Billerica, MA, USA). Section images were obtained with using a Leica DM 4000B photo microscope (Solms, Germany) at a magnification of 200×. Blood vessels were counted (n = 5). 3.5. Chicken Chorioallantoic Membrane (CAM) Assay ------------------------------------------------- The antiangiogenic activity of pristimerin on CAM was assayed as described previously \[[@B17-molecules-17-06854]\]. Briefly, fertilized chicken eggs were incubated at 37 °C and 60%--70% relative humidity for 6 days. After this incubation, a small hole was punched into the broad side of the egg, and a window was carefully created through the egg shell. Sterilized filter paper disks (5 × 5 mm) saturated with either vehicle or pristimerin (20, 40, or 80 nmol/egg) were placed on the CAMs. The window was sealed with cellophane tape, and the eggs were replaced in the incubator and maintained at 37 °C for another 2 days. Finally, the sterilized filter paper disks were removed, after which the CAMs were photographed. Ten eggs were used per group, and the number of newly formed vessels was counted. The angiogenic index was defined as the mean number of visible blood vessel branches, the percentage of inhibition was expressed relative to untreated wells at 100%. Three independent experiments were performed in each trial. 3.5. Rat Aortic Ring Assay -------------------------- The rat aortic ring assay was performed as described previously \[[@B14-molecules-17-06854]\]. Thoracic aortas isolated from 6-week-old Sprague-Dawley rats were cleaned of periadventitial fat and connective tissues and cut into 1- to 1.5-mm-thick rings. After being rinsed five times with M199, these rings were placed in 24-well plates. The clotting medium contained M199+ (M199 with 100 U/mL penicillin and 100 μg/mL streptomycin) plus 0.3% fibrinogen and 0.5% ε-amino-*n*-caproic acid (ACA, Sigma). The growth medium consisted of M199+ with 0.5% FBS and 0.5% ACA. Next, 20 ng/mL VEGF in 1 mL of M199+ with or without various concentrations of pristimerin were added to the wells. As a control, medium alone was assayed. After 7 days, microvessel growth was quantified by manually counting the number of microvessels sprouting from the rat aortic rings, with six rings being used as a group. The percentage of inhibition was expressed relative to untreated wells at 100%. The experimental protocol was established according to the NIH guidelines for Animal Research and Care and was approved by the Institutional Animal Care and Use Committee. Three independent experiments were performed in each trial. 3.7. Cell Proliferation Assay by Bromodeoxyuridine (BrdU) Incorporation ----------------------------------------------------------------------- Cell proliferation was measured using a BrdU ELISA kit (Roche Applied Science, Mannheim, Germany) according to the manufacturer's instructions \[[@B34-molecules-17-06854]\]. HUVECs (5 × 10^3^ cells/well) were plated in 96-well plates. After 24 h, the cells were first synchronized in serum-free ECGM containing 0.5% FBS for 6 h and cultured in 100 μL of serum-free ECGM containing various concentrations of pristimerin with or without VEGF for 24 h at 37 °C. The cells were incubated with 20 μL of a BrdU-labeling solution per well for 4 h and were dried, fixed, and detected using anti-BrdU mAb. Finally, the BrdU incorporation was determined by measuring the optical density at a wavelength of 450 nm using a reference wavelength of 690 nm. Triplicate wells were analyzed for each concentration, and the BrdU assays were repeated in triplicate. 3.8. Wound-Healing Migration Assay ---------------------------------- HUVECs were allowed to grow to full confluence in 6-well plates precoated with 0.1% gelatin (Sigma) and were subsequently starved with serum-free ECGM containing 0.5% FBS for 6 h to inactivate cell proliferation. The cells were wounded with pipette tips and washed with phosphate-buffered saline. Serum-free ECGM containing 0.5% FBS was added to the wells with or without 10 ng/mL VEGF and various concentrations of pristimerin. Images of the cells were recorded after 8 h of incubation at 37 °C in a 95%:5% (v/v) mixture of air and CO2. Three independent experiments were performed. 3.9. Transwell Migration Assay ------------------------------ The chemotactic motility of the HUVECs was determined using a Transwell migration assay (BD Biosciences) as described previously \[[@B18-molecules-17-06854]\]. Briefly, the bottom chambers were filled with 600 μL of serum-free ECGM containing 0.5% FBS supplemented with 10 ng/mL VEGF. The top chambers were seeded with inactivated HUVECs (1.8 × 10^4^ cells) suspended in 100 μL of serum-free ECGM containing 0.5% FBS plus various concentrations of pristimerin. After 8 h of migration, the non-migrated cells were removed with cotton swabs, and the migrated cells were fixed with cold 4% paraformaldehyde and stained with 1% crystal violet. Images were recorded (Olympus; magnification, ×100). The migrated cells were counted manually, and the percentage of inhibition was expressed in comparison with the untreated control wells at 100%. Three independent experiments were performed. 3.10. Capillary-Like Tube Formation Assay ----------------------------------------- Tube formation was assessed as previously described \[[@B35-molecules-17-06854]\]. Growth factor--reduced Matrigel was pipetted into pre-chilled 96-well plates (60 μL Matrigel/well) and polymerized for 60 min at 37 °C. HUVECs were first incubated in serum-free ECGM containing 0.5% FBS for 6 h, subsequently pretreated with various dilutions of pristimerin for 30 min, and finally seeded onto the Matrigel layer in 96-well plates at a density of 2.0 × 10^4^ cells/well. Serum-free ECGM containing 0.5% FBS was added with or without 10 ng/mL VEGF. After 6 h of incubation at 37 °C in a 95%:5% (v/v) mixture of air and CO2, the tubular structures of the endothelial cells were photographed using an inverted microscope (Olympus; magnification, ×100). The tube length of the capillary-like structure was calculated randomly from five fields. Inhibition percentage was expressed using untreated wells as 100%. Three independent experiments were performed. 3.11. Western Immunoblot Analysis --------------------------------- To determine the signaling mechanism of pristimerin involved in angiogenesis, HUVECs were first starved in serum-free ECGM for 6 h and were pretreated with or without pristimerin for 60 min followed by stimulation with 50 ng/mL of VEGF for 2 min (for VEGFR2 activation) or 20 min (to activate signaling downstream of VEGFR2). After stimulation, the cells were harvested and lysed followed by centrifugation. The protein concentrations of the supernatants were measured with the BCA (bicinchoninic acid) assay using a Varioskan multimode microplate spectrophotometer (Thermo, Waltham, MA, USA). About 30--40 μg of cellular protein from each sample was applied to 8--12% SDS-polyacrylamide gels and probed with specific antibodies followed by exposure to horseradish peroxidase-conjugated goat anti-mouse or goat anti-rabbit antibodies. 3.12. Statistical Analysis -------------------------- All of the data in the different experimental groups were expressed as means ± SD. The data reported herein were obtained in at least three independent experiments. Statistical comparisons between the treated groups and the untreated group were performed by one-way analysis of variance (ANOVA) followed by Dunnet's test, the difference between two groups was analyzed by Student's *t*-test. A P value of \<0.05 was considered statistically significant. 4. Conclusions ============== In conclusion, our results demonstrate that pristimerin potently suppresses angiogenesis *in vitro* and *in vivo* by targeting VEGFR2 activation, leading to the suppression of tumor growth. Our discovery suggests that pristimerin has great potential as an anti-cancer therapeutic agent. This study was partially supported by grants from the Specific Fund for Public Interest Research of Traditional Chinese Medicine, Ministry of Finance (No. 200707008), and Mega-Projects of Science Research for the 11th Five-Year Plan: Standardized platform construction and scientific application in new technologies for new drug screening No. 2009ZX09302-002, and the 111 Project (111-2-07), and the International Scientific and Technological Cooperation Projects (No 2010DFB33710). *Samples Availability:* Samples of the compounds are available from the authors.

1. Introduction {#sec1} =============== Cocaine abuse is relatively common in our society. Recently, the Drug Enforcement Agency (DEA) has reported that seventy percent (70%) of cocaine seized at United States (US) borders has been adulterated with levamisole \[[@B1]\]. This agent, which is an antihelminthic and chemotherapeutic drug, indirectly increases the number of D1 dopamine receptors in the brain and has a cholinergic effect that seems to potentiate the effect of cocaine. It also has immune stimulating effect producing antineutrophils cytoplasmic antibodies (ANCAs). Several side effects have been associated with its use, including severe neutropenia and necrotizing vasculitis. Levamisole is a challenging drug to test; it has a half-life about 5.6 hours and specific testing is not routinely available \[[@B2]\]. Recently, several cases with this manifestation have presented in our institution. Although cases have been previously reported in the USA, none had been reported in Puerto Rico or the Caribbean. 2. Case Report {#sec2} ============== A 45-year-old male cocaine user without any previous systemic illness complains of progressive painful erythematous lesions in both auricles and extremities, constant burning, 10/10 pain in upper and lower extremities, although more prominent in upper extremities. Patient also refers suffering of unquantified fever, congested nose, and fatigue; denies nausea, vomiting, and shortness of breath or changes in mental status. Four days prior to admission, patient was evaluated in a health clinic for the above-mentioned lesions and was referred for a follow-up visit to surgery for a skin biopsy. One day prior to admission, patient was taken to the same clinic for the worsening of his skin lesions; he was transferred to our institution for further evaluation and management. History was remarkable for smoking (use of) cocaine; last reported use was 10 days prior to evaluation. Physical examination demonstrated several purpuric retiform patches with erythematous borders in the helix of both ears and both upper and lower extremities ([Figure 1](#fig1){ref-type="fig"}). A working diagnosis of vasculitis was made with suspicion of cryoglobulinemia; high-dose steroids were initiated. Toxicology test in urine was negative for cocaine; hepatitis profile and HIV test were nonreactive. C-ANCA (1.7 IU/mL), P-ANCA (1.7 IU/mL), ANA, and cryoglobulins were positive. Complements levels were decreased C3 = 92.50 mg/dL and C4 = 12.30 mg/dL. WBC = 4.7 × 10^3^/uL, Hb = 10.2 g/dL, Hct = 29.8%, and platelets = 390 × 10^3^ uL. Skin biopsy (punch) demonstrated intravascular thrombosis and mild perivascular lymphocytic infiltrates ([Figure 2](#fig2){ref-type="fig"}). High-dose steroids were continued and plasmapheresis was initiated. Complete clinical resolution of skin lesions was noted after treatment and thirty (30) days of no cocaine use ([Figure 3](#fig3){ref-type="fig"}). The second patient is a 52-year-old male with medical history of constant crack cocaine and marihuana abuse, complaining of bilateral upper and lower extremities necrotic skin lesions, which were also found on ears, nose, and genital area associated with a burning, 10/10 pain of two (2) weeks of evolution. Chills, nauseas, and vomits were referred as associated symptoms; patient denied fever, shortness of breath, or changes in mental status. History remarkable for smoking cocaine, last reported use was the night before initial evaluation at the emergency room. Physical examination showed necrotic lesions with an erythematous base, tender to palpation, on both ears, nose, upper and lower extremities, and genital areas ([Figure 4](#fig4){ref-type="fig"}). Laboratory evaluation was remarkable for neutropenia: WBC = 3.5 × 10^3^/uL, microcytic anemia, Hb = 9.92 g/dL, Hct = 28%, and platelet count of 290 × 10^3^ uL. Due to the appearance of necrotic tissue, a diagnosis of vasculitis with infected necrotic ulcers was made. High-dose steroids and broad-spectrum antibiotics were initiated. Toxicology test was positive for cocaine and cannabinoids, while the hepatitis profile and HIV test were nonreactive. C-ANCA (2.55 IU/ml), P-ANCA (1.69 IU/ML), and cryoglobulins were positive. ANA was negative in this occasion. Complements levels were decreased, C3 = 55.30 mg/dL, and C4 = 8.86 mg/dL. As with the previous patient, skin biopsy revealed homogenous eosinophilic material with blood vessels and lymphocytic infiltrate in dermis; however, no evidence of intravascular thrombosis was found. As recommended by hematology service, plasmapheresis was started, and antibiotics and high-dose steroids were continued. Clinical resolution was observed, but patient left against medical advice before the treatment was completed. The third patient is a 60-year-old homeless female who complained of suffering 8/10, burning pain in both ears for about six months prior to the initial evaluation, and that exacerbated since the day before arrival to our institution. Patient denied any associated symptoms. History of smoking (use of) cocaine: last reported use was the day of admission. Physical examination remarkable for purpuric retiform patches with erythematous borders and necrotic tissue on both ears ([Figure 5](#fig5){ref-type="fig"}). Purpuric retiform patches also observed on the tip of the patient\'s nose and inferior aspect of second to fifth right toe. Patient also presented with foul smelling necrotic ulcer in the distal right, lower extremity. Patient was admitted with broad-spectrum antibiotics for her ulcer but high-dose steroids were not initiated. Laboratory data showed neutropenia, WBC = 3.6 × 10^3^/uL, and microcytic anemia, Hb = 11.3 g/dL, Hct = 34.2%, and Plts = 368 × 10^3^ uL. Toxicology was positive for cocaine: yet, hepatitis profile and HIV test were nonreactive. Complements levels C3 (131 mg/dL) and C4 (24 mg/dL) were within normal values. A skin biopsy revealed vascular thrombosis in the small-sized vascular channels in the dermis. P-ANCA and C-ANCA levels could not be obtained. Although clinical resolution was observed during hospitalization, patient decided to left against medical advice. 3. Discussion {#sec3} ============= Cocaine abuse affects more than five (5) million people in the USA with side effects that are not limited to the parent drug. To enhance profitability and acceptability of the product, it is not uncommon for illicit drugs to undergo several processes. Among these, adulteration (intentional addition of a substance with similar pharmacologic properties which attenuates the effects of the parent drug) \[[@B6]\] presents as a new challenge to the medical community. Levamisole is an anthelmintic agent used in veterinary medicine that was formerly used as an adjuvant with fluorouracil in the treatment of colorectal cancer; as immunomodulator in rheumatoid arthritis; as treatment in nephrotic syndrome. It is known for augmenting dopamine and endogenous opiates levels in the brain, increasing D1 dopamine receptors, and producing cholinergic effects \[[@B3]\]. Levamisole may inhibit MAO and catechol-O-methyltransferase activity, prolonging the presence of catecholamine neurotransmitters in the synapse, adding to the reuptake-inhibition effect of cocaine, and augmenting its effects (some cases of mood elevation have been reported in humans as a side effect of adjuvant therapy with levamisole for colon carcinoma) \[[@B4]\]. Levamisole\'s immune stimulating effect leads to the production of autoantibodies (antinuclear and antineutrophils antibodies), which leads to ANCA positivity (presented in the first two patients), severe neutropenia, and focal necrotizing features of vascular damage \[[@B5]\]. Levamisole-induced, occlusive, necrotizing vasculitis is an uncommon side effect of this agent \[[@B6]\]. It presents as purpuric retiform lesions with necrotic patches typically manifesting on the ears and extremities, as previously reported in patients who received treatment with levamisole for nephrotic syndrome. Also, it differs from lesions directly related to pure cocaine described as palpable purpura or Wegener\'s granulomatosis like \[[@B7]\]. Microscopically, levamisole can produce a mixed pattern of leukocytoclastic vasculitis, thrombotic vasculitis, and/or vascular occlusion as seen in skin biopsies from children using levamisole for the treatment of nephrotic syndrome \[[@B8]\]. Although levamisole levels were not measured because of its short half-life of 5.6 hours and because routine tests for its detection are not commonly available, we believe that the adulteration of cocaine with levamisole was the cause of the clinical presentation of the three evaluated patients. Retiform purpuric lesions that later become necrotic, an histopathologic report of intravascular thrombosis with a mild perivascular lymphocytic infiltrates, and ANCA positivity in a patient with recent history of smoking (use of) cocaine should raise the suspicion of a levamisole-induced vasculitis. Treatment is primarily supportive, however steroids have been used in some cases with success \[[@B6]\]. Since their initial evaluation and treatment, the first two patients have been admitted in several occasions for the same lesions; the use (smoking) of cocaine was reported prior to every admission, thus linking cocaine use to the appearance of these lesions. Clinicians should recognize this presentation and be aware that the discontinuation of cocaine use, more than the use of immunosuppresors, leads to the resolution of these lesions. The authors declare that they have no conflict of interests. The authors would like to thank the Department of Dermatology of the University of Puerto Rico School of Medicine, for their contribution in providing the images of biopsies used in this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} [^1]: Academic Editors: G. S. Alarcon, A. Giusti, and M. Soy

ANNOUNCEMENT {#s1} ============ The ascomycete fungus Fusarium graminearum is the main pathogen causing *Fusarium* head blight (FHB), an important cereal disease worldwide ([@B1]). The F. graminearum genome from a South American strain was not previously available. Here, we report an annotated assembly for strain CML3066 (DON/15-ADON), isolated in 2009 in Rio Grande do Sul state, Brazil (latitude, −28.327, longitude, −51.271). This strain was isolated from a symptomatic wheat spike with 21.5% FHB incidence ([@B2]). Genomic DNA of F. graminearum CML3066 was extracted from mycelia grown for 3 days in potato dextrose broth (PDB) medium using a cetyltrimethylammonium bromide (CTAB) protocol ([@B3]) and quantified using a Qubit 2.0 fluorometer (Life Technologies). Libraries were prepared with TruSeq DNA high-throughput (HT) (Illumina) and SMRTbell (PacBio) kits. Genome sequencing was done both on an Illumina HiSeq 2000 platform, producing 100-bp paired-end reads (91× coverage) with no quality control required, and on a PacBio RS II platform with a postquality filter of minimum polymerase read quality of 0.80 and minimum subread length of 500 bp, resulting in 160× coverage with a subread total of 6,392,721,477 bp, 1,358,615 reads, and an *N*~50~ value of 5,621 bp. Default parameters were used for all software unless otherwise noted. The *de novo* assembly was carried out using SOAPdenovo2 v2.0.4 using the Illumina data with a range of k-mer values (61 to 99) and the SMRT analysis portal using the PacBio data. The PacBio assembly was manually gap filled and further scaffolded using Lastz v1.04.03 alignments with the complementary Illumina assembly, resulting in four complete chromosomes from telomere to telomere with no gaps or N bases. Reference sequence statistics were extracted from Geneious v8.1. The genome annotation of CML3066 was done using the MAKER v2.30 ([@B4]) annotation pipeline with RepeatMasker v4.50 ([@B5]). Gene calls were generated using both AUGUSTUS v2.7 ([@B6]) using the F. graminearum species model and GeneMark ([@B7]), which was trained using strain PH-1 ([@B8], [@B9]). The CML3066 assembly is 36,908,675 bp long with a GC content of 47.9%. The CML3066 genome is predicted to contain 14,188 genes, 286 of which are not present in the PH-1 genome. Using the PH-1 reference, a minimum of 80% of the length of the gene was required to have mapped reads to be considered present. Single nucleotide polymorphism (SNP) calling was performed with SAMtools using default settings. SNP effects were predicted using SnpEff 4.2. Comparison of SNP frequencies along all four chromosomes of both CML3066 and PH-1 ([@B8], [@B10]) revealed that all telomere proximal regions displayed the highest SNP density windows. In addition, chromosomes 1, 2, and 4 were found to have one or two large interstitial regions with a high SNP density. To predict secreted proteins, Blast2GO v3.2 was used to identify signal peptide and transmembrane domains. Prediction of glycosylphosphatidylinositol (GPI)-anchored membrane proteins, cellular protein localization, and effectors was performed using Big-Pi, WoLF PSort and ProtComp v9.0 (Softberry), and EffectorP v1.0 ([@B11][@B12][@B14]), respectively. The secretome was predicted to contain 874 genes. A genome comparison between CML3066 and the reference strain PH-1 is summarized in [Table 1](#tab1){ref-type="table"}. ###### Genome sequence assemblies for F. graminearum strains PH-1 and CML3066 Characteristic Value for strain: --------------------------------------------------------- ------------------- ------------ Genome size (bp)[^*b*^](#ngtab1.2){ref-type="table-fn"} 36,663,736 36,908,675 No. of chromosomes 4 4 GC content (%)[^*c*^](#ngtab1.3){ref-type="table-fn"} 48.2 47.9 No. of spanned gaps 0 0 No. of predicted genes 14,145 14,188  Reannotated genome ([@B10]). Including all scaffolds and the mitochondrial genome but excluding the large repetitive sequence at the carboxyl end of chromosome 4. Excluding the mitochondrial genome. Data availability. {#s1.1} ------------------ The raw data and assembled/annotated sequences have been deposited in the European Nucleotide Archive (ENA). The study accession number is [PRJEB12819](https://www.ebi.ac.uk/ena/data/view/PRJEB12819). The accession numbers for the assembled chromosomes and mitochondrial genome are [LT222053](https://www.ebi.ac.uk/ena/data/view/LT222053) to [LT222057](https://www.ebi.ac.uk/ena/data/view/LT222057). The secretome and effector predictions can be found at <https://github.com/ana321wood/Secretome_CML3066_Feb2020/blob/master/Secretome_CML3066_Feb2020_AMW.txt>. This research was supported by the Biotechnology and Biological Sciences Research Council of the United Kingdom (BBSRC) through the Institute Strategic Programmes 20:20 Wheat (BB/J/00426X/1) and Designing Future Wheat (BB/P016855/1), the BBSRC Embrapa joint wheat projects (BB/N004493/1 and BB/N018095/1), the CAPES Foundation of Brazil Ph.D. scholarship (BEX 1266-13-6), and the CNPq research fellow grant 303216/2012-3. We thank Ludwig Pfenning for depositing the strain in the Coleção Micológica de Lavras (CML) in Brazil. We also thank the staff at the European Nucleotide Archive (ENA), Cambridge, United Kingdom. [^1]: **Citation** Machado Wood AK, King R, Urban M, Nicolli CP, Del Ponte EM, Hammond-Kosack KE. 2020. Genome sequence of *Fusarium graminearum* strain CML3066, isolated from a wheat spike in southern Brazil. Microbiol Resour Announc 9:e00157-20. <https://doi.org/10.1128/MRA.00157-20>.

Background {#Sec1} ========== Diabetes is a chronic disease that is responsible for high rates of morbidity and mortality which can be attributed to atherosclerosis and cardiovascular disease \[[@CR1]\]. It is estimated that type II diabetes doubles the risk of cardiovascular disease even after adjustment of other cardiovascular risk factors \[[@CR2]\]. Despite the increase in the rate of treatment of diabetic patients with statins and glucose lowering drugs achieving target glycated hemoglobin (HBA1C) levels and low-density lipoprotein (LDL) levels \[[@CR3]\], another strategy of effective management of diabetes lies in management of the disease process at earlier stage \[[@CR1]\]. Prediabetes is a collective term that encloses individuals with glucose levels lower than cutoff levels for diabetes but too high to be considered normal. It is the term used for individuals with impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT) and/or HbA1C levels ranging from 5.7 to 6.4% \[[@CR3]\]. Prediabetes is not an uncommon condition with an estimated worldwide prevalence of 343 million individuals expected to rise to 471 million by 2035 \[[@CR4]\]. Prediabetes is a serious clinical condition that not only increases the risk of developing diabetes but also increases the burden of cardiovascular disease risk. Compared to normoglycemic individuals, patients with prediabetes show a 20% higher risk of developing cardiovascular disease (CVD) \[[@CR5]\]. Prediabetes is a toxic state in which both micro- and macrovascular complications of diabetes can manifest \[[@CR6]\]. The prompt diagnosis and proper management of prediabetes are necessary to prevent progression to diabetes mellitus and to prevent microvascular and macrovascular complications that manifest early in the prediabetic state \[[@CR7]\]***.*** Aim of the work {#Sec2} =============== Observes the effect of prediabetes on the severity of coronary artery disease in patients undergoing elective coronary angiography. Methods {#Sec3} ======= The current study was carried out at the cardiology department at a university hospital. Inclusion criteria {#Sec4} ------------------ Patients who were admitted for elective coronary angiography and/or PCI starting from September 2017 to August 2018. Exclusion criteria {#Sec5} ------------------ No exclusion criteria were applied. After an informed written consent, all patients involved in the study were subjected to: **A- History taking and examination** with special emphasis on age, sex, risk factors for coronary artery disease (smoking, HTN, DM, dyslipidemia, positive family history for premature CVDs), history of CKD detected either by reduction in GFR or high serum creatinine, history of prior percutaneous coronary intervention (PCI) or coronary arteries bypass grafting (CABG), or acute coronary syndrome (ACS). **B- Laboratory tests**: Level of HBA1C and serum creatinine on admission. **C- Estimation of renal function**: eGFR was estimated using MDRD formula: eGFR = 186 × (serum creatinine)--1.154 × age--0.0203 (× 1.210 if black) (× 0.742 if female) \[[@CR8]\] **D- Interventional data**: Number of vessels affected and atherosclerotic burden of CAD assessed by Gensini score \[[@CR9]\]***.*** For patients undergoing PCI, additional data was collected regarding number of stents used, type of stents used, and total length of stents used. The studied patients were divided according to HbA1C level to 3 groups: 1- Group A: Normoglycemic patients (HBA1C \< 5.7%) 2- Group B: Prediabetic patients (HBA1C 5.7--6.4%) 3- Group C: Diabetic patients (HBA1C \> 6.4%) \[[@CR10]\] Statistical analysis {#Sec6} -------------------- Data were collected and revised on PC. Data were tabulated and statistically analyzed using SPSS 17 software, mean and standard deviation (± SD), and range for parametric numerical data, while the median was used for nonparametric numerical data. Student t-test was used to assess the statistical significance of the difference between two study group means. Mann--Whitney test (U test) was used to assess the statistical significance of the difference of a nonparametric variable between two study groups. Chi-squared test was used to examine the relationship between two qualitative variables. Fisher's exact test was used to examine the relationship between two qualitative variables when the expected count is less than 5 in more than 20% of cells. Results {#Sec7} ======= Patients were divided to group A (normoglycemic group, *N* = 228), group B (prediabetes group, *N* = 177), and group C (diabetic group, *N* = 326). Prediabetics represented 24% of the study population (Table [1](#Tab1){ref-type="table"}). Table 1Group classificationHBA1CGroup A(*n* = 228)\ (31.2%)Group B(*n* = 177)\ (24.2%)Group C(*n* = 326)\ (44.6%)Mean ± SD5.25 ± 0.246.00 ± 0.228.92 ± 1.60Range4.5--5.65.7--6.46.5--13 Among patients with HBA1C in the prediabetic range, there were only 8 patients who were known prediabetic and on medical treatment. Among the diabetic group, 7% of patients were newly diagnosed, denoting that newly diagnosed prediabetics and diabetics represent 26% of the study population. Demographic and clinical characteristics {#Sec8} ---------------------------------------- There was no significant difference regarding age among the three groups, yet group C showed higher prevalence of male gender and a lower prevalence of smoking. Both DM and prediabetes group showed significantly higher prevalence of HTN. The normoglycemic group showed a stronger family history of CAD (Table [2](#Tab2){ref-type="table"}). Table 2Demographic and clinical characteristics of the groupsGroup AGroup BGroup CTest\ value*P* valueSig.Post hoc\ analysisNo. = 228No. = 177No. = 326P1P2P3Age (years)56.68 ± 9.2157.10 ± 9.8458.26 ± 8.872.1660.115NS------Sex(male)171 (75.0%)132 (74.6%)213 (65.3%)7.8230.020S0.9240.0150.033Smoking114 (50.0%)99 (55.9%)111 (34.0%)26.5880.000S0.2360.0000.000HTN84 (36.8%)90 (50.8%)198 (60.7%)30.6490.000S0.0050.0000.032Dyslipidemia120 (52.6%)86 (49.0%)165 (50.6%)0.660.7NS------Known CKD12 (5.3%)9 (5.1%)10 (3.1%)2.0020.367NS------Family history of CAD97 (42.5%)36 (20.3%)82 (25.2%)28.8040.000S0.0000.0000.224*P* value \> 0.05, nonsignificant; *P* value \< 0.05, significant; *P* value \< 0.01, highly significant\*: Chi-squared test; •: one-way ANOVA testP1: *P* value group A vs group BP2: *P* value group A vs group CP3: *P* value group B vs group C Assessment of renal function {#Sec9} ---------------------------- On comparing the three groups, there was no significant difference regarding the mean eGFR or prevalence of CKD (Table [3](#Tab3){ref-type="table"}). Table 3Assessment of renal functionGroup AGroup BGroup CTest\ value*P* valueSig.No. = 228No. = 177No. = 326CreatinineMean ± SD1.03 ± 0.261.05 ± 0.251.02 ± 0.290.980•0.376NSRange0.6--1.60.6--1.90.5--2.9eGFRMean ± SD78.86 ± 23.3176.92 ± 23.2878.41 ± 24.120.361•0.697NSRange35--14237--14225--149CKD54 (23.7%)33 (18.6%)75 (23.0%)1.711\*0.425NS*P* value \> 0.05, nonsignificant; *P* value \< 0.05, significant; *P* value \< 0.01, highly significant\*, Chi-squared test; •, one-way ANOVA testP1: Group A vs group BP2: Group A vs group CP3: Group B vs group C Prior history of ischemia {#Sec10} ------------------------- There was no significant difference in history of PCI or CABG prior to the current procedure between the different groups with significantly higher prevalence of prior ACS in patients with prediabetes (Table [4](#Tab4){ref-type="table"}). Table 4History of CAD among the different groupsGroup AGroup BGroup CTest\ value\**P* valueSig.Post hoc analysisNo. (%)No. (%)No. (%)P1P2P3Prior PCI42 (18.4%)39 (22.0%)75 (23.0%)1.7470.417NS------Prior CABG9 (3.9%)3 (1.7%)15 (4.6%)2.7840.249NS------Prior ACS81 (36.0%)93 (52.5%)114 (35.0%)16.5430.000S0.0010.8030.000*P* value \> 0.05, nonsignificant; *P* value \< 0.05, significant; *P* value \< 0.01, highly significant\*: Chi-squared testP1: Group A vs group BP2: Group A vs group CP3: Group B vs group C Interventional data {#Sec11} ------------------- Regarding the type of procedure performed, group A showed a lower rate of PCI compared to group C. Both group B and group C showed a larger number of vessels with significant disease when compared to group A. LM disease was significantly higher in groups B and C when compared to group A. Group B showed a more complex coronary anatomy with a higher Gensini score than group A and comparable to group C. The type of stent used was similar among the different groups. Length of stents used was higher in prediabetic when compared to normoglycemic group denoting a longer length of lesions (Table [5](#Tab5){ref-type="table"}, Figs. [1](#Fig1){ref-type="fig"} and [2](#Fig2){ref-type="fig"}). Table 5Interventional data among the different groupsGroup AGroup BGroup CTest\ value*P* valueSig.Post hoc analysisNo. = 228No. = 177No. = 326P1P2P3ProcedureCA99 (43.4%)66 (37.3%)116 (35.6%)14.507\*0.024S0.2150.0090.662CA + PCI69 (30.3%)57 (32.2%)93 (28.5%)CA + PTCA3 (1.3%)0 (0.0%)0 (0.0%)PCI57 (25.0%)54 (30.5%)117 (35.9%)No. of vessels042 (18.4%)18 (10.2%)38 (11.7%)41.574\*0.000S0.0000.0000.4351102 (44.7%)54 (30.5%)87 (26.7%)248 (21.1%)66 (37.3%)111 (34.0%)336 (15.8%)39 (22.0%)87 (26.7%)40 (0.0%)0 (0.0%)3 (0.9%)LM disease11(4.8%)21(11.8)%34(10.4%))7.4180.0245S0.0090.010.6Gensini scoreMedian (IQR)35.75 (24--64.5)66 (49--94)65 (36--96)72.404≠0.000S0.0000.0000.967Range0--1520--1350--156Type of stentNo0 (0.0%)3 (2.7%)3 (1.4%)6.393\*0.172NS------BMS0 (0.0%)3 (2.7%)3 (1.4%)DES120 (100.0%)105 (94.6%)204 (97.1%)No. of stents03 (2.4%)3 (2.7%)3 (1.4%)22.963\*0.003S0.1030.0000.331178 (63.4%)54 (48.6%)90 (42.9%)239 (31.7%)48 (43.2%)99 (47.1%)30 (0.0%)3 (2.7%)15 (7.1%)43 (2.4%)3 (2.7%)3 (1.4%)LengthMedian (IQR)33 (21.5--50)42 (32.5--59)48 (28--66)16.055≠0.000S0.0040.0000.500Range12--11012--9610--147*P* value \> 0.05, nonsignificant; *P* value \< 0.05, significant; *P* value \< 0.01,highly significant\*, Chi-square test; •, one-way ANOVA testP1: Group A vs group BP2: Group A vs group CP3: Group B vs group C Fig. 1CAD severity among different groups represented by Gensini score (median and IQR) Fig. 2Length of stents used among different groups Discussion {#Sec12} ========== Our study included 731 patients who presented to our university hospital to undergo elective coronary angiography for the diagnosis and treatment of CAD starting from September 2017 to August 2018. We aimed to evaluate the effect of prediabetes on angiographic outcomes in those patients. One hundred and seventy-seven patients were prediabetics constituting 24% of the study population. Similar prevalence of prediabetes was demonstrated among elective PCI patients and ACS patients in various registries \[[@CR11], [@CR12]\]. Patients with prediabetes had the same age range as diabetics and normoglycemic subjects, yet female gender was more prevalent among the diabetic group. This can be explained by the findings of Kodama et al. \[[@CR13]\] suggesting that cardiovascular risk in the diabetic population is higher among women than in men. Although Kataoka et al. \[[@CR14]\] and Choi et al. \[[@CR11]\] found no significant difference in age between normoglycemic and prediabetic groups, the results of both showed male preponderance across the different groups. There were more smokers in the prediabetes group compared to diabetics (Choi et al.) \[[@CR11]\]**.** However, we found that smoking was not significantly different between normoglycemic patients and prediabetics. There was a parallel increase in the prevalence of hypertension with increase in HBA1C. This can be attributed to insulin resistance promoting both hypertension and diabetes (Sowers) \[[@CR15]\] or a myriad of genetic and environmental factors contributing to the development of both diabetes and hypertension \[[@CR16]\]*.* Choi et al. \[[@CR11]\] also demonstrated a higher prevalence of hypertension among prediabetic patients than normoglycemic patients undergoing elective PCI. Similarly, Zhang et al. \[[@CR17]\] demonstrated that hypertension was more common in prediabetics than normoglycemic subjects and in diabetic group more than prediabetic group. Patients with prediabetes had a prevalence of dyslipidemia which was comparable to diabetics and normoglycemic subjects. Nakamura et al. \[[@CR18]\] demonstrated that among CAD patients, prediabetics and diabetics showed a higher prevalence of dyslipidemia, yet this was evident in postprandial lipid levels and not the fasting lipid levels which are used as the standard screening test. Similarly, Açar et al. \[[@CR12]\] found no difference in prevalence of dyslipidemia between prediabetic, normoglycemic, and diabetic subjects. The prevalence of CKD was not significantly different among the three groups, although diabetes is known as a common comorbid risk factor for CKD \[[@CR19]\] as well as CKD pathophysiology starting in prediabetic subjects \[[@CR20]\]. Those results are similar to Zhang et al. \[[@CR17]\] and Choi et al. \[[@CR11]\] who found no significant difference in prevalence of CKD among CAD patients. This can be attributed to hindering of both pharmacological and interventional treatment of cardiovascular disease by the presence of CKD in addition to increments in the risk of contrast-induced nephropathy with worsening of renal function; the management plan of CAD in CKD patients is directed towards more conservative management \[[@CR21], [@CR22]\]. Prediabetic subjects showed involvement of coronary arteries with a more aggressive atherosclerotic process resulting in CAD severity that was significantly higher than normoglycemic subjects and comparable to diabetic subjects. The number of coronary arteries with significant disease was higher in the prediabetic group than the normoglycemic group, yet there was no significant difference when compared with the diabetic group. This is similar to the findings of Santos et al. \[[@CR23]\] who demonstrated that among patients with CAD confirmed by angiography, prediabetes was more commonly associated with multivessel disease. In addition, Açar et al. \[[@CR12]\] found that among patients presenting with acute coronary syndrome, diabetic and prediabetic patients showed significantly higher prevalence of three vessel diseases when compared to normoglycemic patients. The complexity of CAD assessed by Gensini score was higher in the prediabetic than in normoglycemic subjects and comparable with diabetics. This is similar to the results of Açar et al. \[[@CR12]\] where patients with prediabetes and diabetes showed a more complex coronary anatomy than normoglycemic subjects with a higher proportion of patients with three vessel diseases and higher CAD severity assessed by both SYNTAX and Gensini scores. This is in accordance with the results of Kataoka et al. \[[@CR14]\]; both prediabetes group and diabetes group showed a higher Gensini score when compared to those without diabetes. The glycemic state didn't affect the type of stent used, with drug-eluting stents (DESs) used in most of patients across the three groups. This goes hand in hand with Choi et al. \[[@CR11]\] as all patients of the different groups received DESs. When comparing the length of stent used among the different groups, both prediabetics and diabetics required significantly longer stents than normoglycemic patients. This can be attributed to the findings of De Rosa et al. \[[@CR24]\] who assessed plaque characteristics in stable CAD patients and demonstrated that both prediabetes and diabetes were associated with a higher and longer plaque burden. Zhang et al. \[[@CR17]\] assessed OCT data regarding non-infarct-related plaques in patents presenting with ACS and found that raised HBA1C in prediabetic subjects was associated with more complex and active plaque structure with longer lipid length, higher lipid content, thinner fibrous cap, higher macrophage infiltration, wider lipid arc, and more calcification than normal subjects but was comparable to diabetic subjects. HBA1C was independently associated with significantly higher lipid length. Those results agree with the findings of Kataoka et al. \[[@CR14]\] which demonstrated that both prediabetes and diabetes were associated with high average lesion length in patients with CAD assessed by quantitative coronary angiography. Similarly, Choi et al. \[[@CR11]\] found significantly longer lesions in prediabetics when compared to normoglycemic subjects. Conclusion {#Sec13} ========== Prediabetes is not merely a step closer to diabetes, it is a stage of diabetes which shows a similar atherosclerotic disease progression causing more complex coronary anatomy and requiring a higher number of longer stents. Yet, such a stage is always overlooked. Prediabetes confers high yet modifiable cardiovascular risk. Rigorous lifestyle interventions and medical treatment can help flatten the risk of conversion to diabetes, regression to normoglycemia, and reduction of the cardiovascular disease burden in this population. ACS : Acute coronary syndrome ANOVA : Analysis of variance BMS : Bare metal stent CA : Coronary angiography CABG : Coronary artery bypass grafting CAD : Coronary artery disease CKD : Chronic kidney disease CVD : Cardiovascular disease DES : Drug eluting stent DM : Diabetes mellitus eGFR : Estimated glomerular filtration rate GFR : Glomerular filtration rate HTN : Hypertension IFG : Impaired fasting glucose IGT : Impaired glucose tolerance IQR : Interquartile range LDL : Low-density lipoprotein LM : Left main MDRD : Modification of diet in renal disease No. : Number OCT : Optical coherence tomography PCI : Percutaneous coronary intervention SD : Standard deviation **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Not applicable. AM, SA, and AS performed the angiographies and angioplasties and analyzed and interpreted the patient data. MTZ, AM, and AS were major contributors in writing the manuscript. All authors read and approved the final manuscript. No financial support or scholarship All data and equipment were available at Ain Shams University Ethics approval and consent to participate This study was approved by the Ethical Committee of Ain Shams university. All the procedures in the study were in accordance with the 1975 Helsinki Declaration, updated in 2013. Informed consent was obtained from all of the participants included in the study. Not applicable. The authors declare that they have no competing interests.

INTRODUCTION ============ Sudden infant death syndrome (SIDS) is death of an infant that is neither attributable to medical history nor explained after autopsy or by death scene investigation. SIDS is the leading cause of death in the first year of life after the neonatal period and is currently responsible for 0.53 deaths per 1000 infants. High incidence, catastrophic impact on affected families and absence of mechanistic insight means that SIDS represents a major medical challenge. Since the 'back to sleep' campaign in 1994, there have been no further reductions in SIDS incidence. A number of causative mechanisms have been proposed to lead to SIDS, but without any unifying theory or correlation with pathological findings ([@b7-0060503]). A study of 33,034 infants found that 50% of infants who died of SIDS had a prolonged QTc interval in the first week of life ([@b13-0060503]). Approximately 10% of SIDS cases carry functionally significant genetic variants in sodium and potassium channels causing long QT ([@b2-0060503]), or variants in the gap junction protein Connexin43 (Cx43) ([@b16-0060503]). This circumstantial evidence suggests a role for abnormal electrical conduction in SIDS, but the underlying cause(s) in the vast majority of cases remains unexplained. Most risk factors for SIDS, including prone sleeping position, respiratory disorders and high altitude, are associated with a reduced oxygen environment. Furthermore, hypoxia is associated with a prolonged QT interval in the adult ([@b11-0060503]; [@b15-0060503]). We therefore hypothesised that neonatal hypoxia leading to abnormal electrical conduction is a potential cause of sudden death. We used non-invasive electrocardiography to characterize the postnatal maturation of the cardiac electrical conduction system in neonatal mice ([@b6-0060503]). We investigated whether reduced ambient oxygen environment or genetically manipulated hypoxic signalling affected maturation of the cardiac electrical conduction system and the subsequent risk of sudden death. RESULTS ======= Maturation of ECG morphology in wild-type mice ---------------------------------------------- To assess electrocardiographic changes immediately following birth, unborn pups were removed from pregnant females at embryonic day (E)18.5 and placed with a foster mother. We performed electrocardiography in the same pups sequentially at 0, 1, 3, 6, 12 and 24 hours after birth. ECG morphology changes were detectable at 1 hour after birth, becoming significant at 3 hours. Heart rate increased whereas QRS, QTc and QTc dispersion (where 'c' denotes correction for heart rate) declined rapidly and then plateaued over the 24 hours ([Fig. 1](#f1-0060503){ref-type="fig"}). We recorded resting ECGs from postnatal day (P) 0.5 to P10 ([Fig. 1](#f1-0060503){ref-type="fig"}). The trends of increased heart rate and declining QRS, QTc and QTc dispersion continued over this timescale ([Fig. 1](#f1-0060503){ref-type="fig"}). {#f1-0060503} Hypoxia prevents maturation of the ECG -------------------------------------- Neonates reared in 10% oxygen for 24 hours showed reduced heart rate and increased QTc and QTc dispersion compared with normoxic controls ([Fig. 2](#f2-0060503){ref-type="fig"}). These parameters were similar to those in newborn neonates. {#f2-0060503} *αMHC-Cre::VHL^fl/fl^* mice exhibit immature ECG morphology and sudden death ---------------------------------------------------------------------------- αMHC*-Cre::VHL^fl/fl^* mice have cardiac-specific deletion of von Hippel-Lindau protein (VHL), causing constitutive upregulation of cardiac hypoxia inducible factor (HIF) signalling. Their neonates showed decreased heart rate with increased QRS, QTc and QTc dispersion compared with control αMHC-*Cre::VHL^+/+^*or *αMHC-Cre::VHL^fl/+^*littermates at 10 days after birth ([Fig. 2](#f2-0060503){ref-type="fig"}). αMHC-*Cre::VHL^fl/fl^* mice died between P16 and P18. Before death, there were no observable differences between mutant and control littermates in behaviour or weight. *αMHC-Cre::VHL^fl/fl^* mice exhibited frequent cardiac arrhythmia, consistent with sudden cardiac death, as did hypoxic wild-type mice ([Fig. 3](#f3-0060503){ref-type="fig"}). {#f3-0060503} ###### TRANSLATIONAL IMPACT **Clinical issue** Sudden infant death syndrome (SIDS) remains one of the major enigmas in modern medicine. The 'back to sleep' campaign in 1994, which encouraged parents to place infants on their backs to sleep, promoted a reduction in SIDS incidence from 2 to 0.53 infants per 1000 births. Since then, there has been no reduction in this figure. Thus, advice for parents remains limited, and tragedies that might be preventable continue to occur. It has been previously documented that ∼50% of infants that die from SIDS display a prolonged QTc interval in the first few weeks of life, but the mechanisms underlying this observation have been elusive (except in cases where rare channelopathies and other genetic abnormalities are present). **Results** In this study, the authors addressed this issue using a recent innovation in electrocardiography that allows non-invasive recording of the electrocardiogram (ECG) in mice. This enabled the first reported catalogue of ECG changes from birth to 10 days postnatally, measuring changes in heart rate, QTc interval and QRS duration. By altering ambient oxygen concentration or genetically manipulating cellular hypoxic signalling in neonatal mice, the authors show that an increase in ambient oxygen concentration after birth is important for driving maturation of cardiac electrical conduction. Reduced oxygen predisposed mice to arrhythmia and sudden death, which was associated with ECG abnormalities. At the cellular level, reduced oxygen caused aberrant gap junction phosphorylation and distribution, and misexpression of ion channels, in the heart. These findings are consistent with known risk factors of SIDS -- such as head covering, high altitude, respiratory infections, central nervous system abnormalities and the prone sleeping position -- all of which are directly or indirectly associated with a hypoxic environment. **Implications and future directions** This study provides a link between neonatal hypoxia, ECG abnormalities and sudden death, which might provide an explanation for many SIDS cases. The results support the use of regular ECG screening of infants, and subsequent close monitoring of infants displaying long QTc interval, as well as ensuring a well-ventilated environment in cots and the use of other hypoxia-prevention strategies. The mouse models used in this study will facilitate further investigation into SIDS, and the non-invasive ECG approach used here can be applied by other researchers investigating cardiac conduction defects in mice. Risk of sudden death on exposure to hypoxia decreases with age in neonates -------------------------------------------------------------------------- We hypothesised that postnatal electrocardiac maturation is oxygen dependent and that exposure of neonatal mice to hypoxia at later points after birth would result in lower rates of sudden death. When neonates were raised from birth in a hypoxic environment for 24 hours, mortality was 58%. When neonates were raised from birth in normoxia for 1, 6 and 12 hours, then placed into 10% hypoxia for 24 hours, mortality was reduced to 33, 25 and 17%, respectively ([Fig. 3](#f3-0060503){ref-type="fig"}). Connexin43 distribution and quantification ------------------------------------------ Cx43 is essential for normal electrical conduction in the heart; cardiac-restricted inactivation of Cx43 leads to slower ventricular conduction and lethal arrhythmias in mice ([@b8-0060503]). We therefore performed immunohistochemistry to investigate left ventricular distribution of Cx43 in neonatal mice reared in 10% oxygen, in *αMHC-Cre::VHL^fl/fl^* mice and in normoxic controls. We found no difference between hypoxic mice and controls in Cx43 distribution and quantification (data not shown). In *αMHC-Cre::VHL^fl/fl^* mice, Cx43 was observed in intracellular aggregates rather than at the cell membrane ([Fig. 4](#f4-0060503){ref-type="fig"}). Western blots using antibodies to phosphorylated and non-phosphorylated Cx43 revealed that total Cx43 was unaltered, whereas the presence of phosphorylated Cx43, thought to be targeted to the plasma membrane ([@b14-0060503]), was nearly undetectable ([Fig. 4](#f4-0060503){ref-type="fig"}). {#f4-0060503} Ion channel expression ---------------------- Microarray gene expression analysis showed reduced expression of several cardiac ion channels (potassium channels, potassium inwardly rectifying channels and sodium channels) in neonates reared in 10% oxygen for 24 hours compared with normoxic controls ([Fig. 4](#f4-0060503){ref-type="fig"}). DISCUSSION ========== We describe, for the first time, oxygen-dependent maturation of cardiac conduction in the mouse over the hours and days following birth. Increased postnatal heart rate and decreased QRS duration, QTc interval and QT dispersion during the first postnatal week are dependent on downregulation of hypoxia signalling in the heart. Elevation of neonatal cardiac hypoxia signalling leads to arrhythmia and sudden death. This in turn suggests a previously unknown mechanism for SIDS pathogenesis. Our results link hypoxia, a major risk factor for SIDS, with several genetic mutations found in SIDS victims ([@b2-0060503]; [@b16-0060503]). The link between prolonged QT interval and risk of SIDS has been firmly established ([@b13-0060503]); however, with the exception of genetic variations in ion channels ([@b2-0060503]), the root of prolonged QT remains unknown in most SIDS cases. In some studies of human SIDS, no correlation has been made with QT prolongation. This might reflect the relatively poor sensitivity of surface ECGs to detect changes in QT interval, the age of testing or the use of small cohorts. Our mouse models suggest that hypoxia could be an important precipitant of prolonged QT, and thus sudden death, by hypoxia-induced downregulation of ion channels and Cx43 dephosphorylation. Indeed, hypoxia, prolonged QT interval and risk of lethal cardiac arrhythmias are causally linked in adults ([@b11-0060503]; [@b15-0060503]). It is unclear whether hypoxia alone can be causal in human SIDS cases, as in our models, or whether it exacerbates underlying genetic variations and is additive with other SIDS risk factors. We found overall levels of Cx43 to be unaltered in mice with constitutively elevated cardiac HIF signalling, but a significant reduction in membrane targeting consistent with Cx43 dephosphorylation ([Fig. 4](#f4-0060503){ref-type="fig"}), which has been reported in adult hypoxic myocardium ([@b3-0060503]). We also found significant downregulation of potassium, sodium and calcium channels when neonates were raised in hypoxia ([Fig. 4](#f4-0060503){ref-type="fig"}). In our current study, dead pups were rapidly eaten by the dam, so the quality of tissue available for autopsy was poor. It will be important to analyse the cause of death by ECG telemetry and immediate autopsy, to compare with human pathological findings in SIDS ([@b7-0060503]). Sensitivity to myocardial hypoxia decreases with time after birth, with risk of death declining with age of exposure to hypoxia ([Fig. 3](#f3-0060503){ref-type="fig"}). It is not definitively known how the timescale of postnatal development in mice relates to that of humans, but it is well documented that sensitivity to SIDS in humans decreases 4 months after birth. Interestingly, this is when the QTc interval is known to peak in humans ([@b12-0060503]), whereas mice display unidirectional change ([Fig. 1](#f1-0060503){ref-type="fig"}). We propose a 'ratchet' effect whereby oxygen causes maturation of the electrical conduction system, with declining ability to revert to immature phenotype with increasing age. These pathological cardiac changes could represent a predisposition to cardiac death and might themselves be serious enough to lead to death (as in our model), or be lethal in combination with other risk factors such as brain-stem malfunction. The discrepancies between hypoxia-reared neonates and *αMHC-Cre::VHL^fl/+^* mice might be due to a dosage effect of HIF signalling. It could be that, at 10% FiO~2~, neonatal cardiac HIF signalling is not maximally upregulated, whereas it is in VHL deletion. Systemic effects of generalised hypoxia, such as increased sympathetic activation, might also contribute to the slightly differing phenotypes. In summary, we propose a model that links neonatal hypoxia with sudden death by cardiac arrhythmia by misregulation of cardiac Cx43 and ion channels. Our model is consistent with existing theories of SIDS pathogenesis and links hypoxia, the major known risk factor for SIDS, with many of the candidate genes for pathogenesis. Our electrocardiographic characterisation in the developing neonatal mouse serves as a benchmark for future studies and we believe that the neonatal hypoxic model and the *αMHC-Cre::VHL^fl/fl^*mouse will facilitate further investigation into SIDS. The lack of validated animal models of SIDS is puzzling given that this is a large clinical problem with little mechanistic insight at the moment. We feel that our study adds further evidence to prompt the use of regular ECG screening of infants and subsequent close-monitoring of those infants displaying long QTc interval to ensure a well-ventilated environment in the infant\'s cot and the use of other strategies to prevent hypoxia. MATERIALS AND METHODS ===================== Animal husbandry ---------------- All studies were performed in accordance with the Home Office Animal Procedures Act (1986) and guidelines established by the European Convention for the Protection of Laboratory Animals. In studies characterising ECG maturation in the hours after birth and in hypoxic studies, embryonic F1(CBA/Ca × C57BL/10) mice were removed and fostered at E18.5 onto a Parkes mouse who had littered the previous day. We were able to distinguish the fostered mice by the black colouration of the eyes in the F1(CBA/Ca × C57BL/10) pups compared with unpigmented eyes of the Parkes strain pups. The transgenic mouse strain, *αMHC Cre^+^::VHL^fl/fl^,*was created by crossing transgenic mice with a floxed VHL allele ([@b9-0060503]) with mice containing Cre driven by the α-myosin heavy chain promoter (αMHC Cre), resulting in cardiac specificity ([@b1-0060503]). PCR amplification was performed on tail-derived genomic DNA to determine genotype. Electrocardiography ------------------- ECGs were recorded non-invasively in conscious mice using the ECGenie system (Mouse Specifics). Data acquisition was carried out using the program LabChart 6 (ADInstruments). Analysis of individual ECG signals was then performed using e-MOUSE physiologic waveform analysis software (Mouse Specifics) as described ([@b6-0060503]). In this system, ECG recordings are assessed by the user before being analysed by automated algorithms; signals which contain too much noise or incorrectly called waveforms are removed. All data were obtained during daylight hours, when the mouse heart rate is more stable than during the more active nocturnal hours. In evaluating waveforms and intervals, the end of the T wave was determined as the return of the signal to the isoelectric line as previously described ([@b6-0060503]). QTc was calculated according to Bazett\'s formula modified specifically for mice ([@b10-0060503]), i.e. QTc=QT~0~/\[(RR~0~/100)^1/2^\]. Connexin43 studies ------------------ Hearts from embryos and neonatal mice were dissected in cold PBS and immediately snap-frozen in liquid nitrogen. Immunohistochemistry and western blotting for Cx43 was performed as described ([@b5-0060503]). Anti-Connexin43 (Zymed) and anti-phosphorylated-Connexin43 (Invitrogen) were diluted 1:200 for histology. An immunoblot for actin protein (anti-actin antibody; Sigma-Aldrich) was used as a control for equal protein loading in western blotting. All antibodies including secondary antibodies (Sigma-Aldrich) were diluted 1:5000. Full details of immunochemistry and western blotting procedures are available on request. Ion channel gene expression --------------------------- RNA extraction and microarray analysis was performed on hearts as previously described ([@b4-0060503]). Briefly, the specimen was placed in 1 ml of TRIzol reagent and homogenised using glass homogenisers and plungers (Uniform, Jencoms, England). Chloroform (200 μl*)*was added, samples mixed by vortex and left at room temperature for 5 minutes. The tubes were then centrifuged at 13,000 r.p.m. for 15 minutes at 4°C. The aqueous phase was then transferred to a new centrifuge tube and 500 μl of chilled isopropanol added and mixed by vortex. After incubation at room temperature for 20 minutes, tubes were centrifuged at 13,000 r.p.m. for 30 minutes at 4°C. The supernatant was discarded and 500 μl of ice cold 70% ethanol (v/v) added to the residual pellet. Tubes were vortexed before centrifuging at 8000 r.p.m. for 5 minutes at 4°C. The supernatant was discarded and the pellets air-dried. The resulting RNA was resuspended in 50 μl of nuclease-free water (Ambion, Huntingdon, UK) and frozen at −80°C until use. RNA cleanup was carried out on samples prior to microarray analysis, followed by assessment of RNA yield and purity (full details available on request). Labelled RNA was hybridised to the mouse 430Plus 2.0 chip (Affymetrix) (full details available on request) and the raw data analysed using GeneSpring software version 11.0 (Silicon Genetics/Agilent Technologies). Gene lists were quality filtered to remove genes with expression levels below background and limited to report genes that changed by 1.5-fold or greater with a significance of *P*\<0.05 according to an unpaired *t*-test. **FUNDING**: This work was funded by the Medical Research Council to T.M. \[grant number U117562103\]. **COMPETING INTERESTS:**The authors declare that they do not have any competing or financial interests. **AUTHOR CONTRIBUTIONS:**M.T.N., R.A.B. and T.J.M. conceived and designed the experiments. M.T.N. performed the experiments. M.T.N. analysed the data. M.T.N. and R.A.B. wrote the paper. T.J.M. edited the paper.

Introduction {#Sec1} ============ One of the remarkable features of the Integer Quantum Hall Effect (QHE) is the impressive precision of the quantization of the plateaus observed in the experiments. While the experimental samples have a very complex microscopic structure, depending on a huge number of non-universal details related to molecular forces and the atomic structure, the conductance appears to be quantized at a very high precision, and the result only depends on fundamental constants. The understanding of this phenomenon, via a connection between the Hall conductivity and a topological invariant \[[@CR4], [@CR40]\] was a major success of theoretical condensed matter in the 80s. The argument was later generalized to non-interacting disordered systems \[[@CR1], [@CR5], [@CR9], [@CR10]\] and to clean multi-particle systems \[[@CR3], [@CR37]\]: however, the definition of conductivity in the interacting case required the presence of an unphysical averaging over fluxes, expected to be unimportant in the thermodynamic limit, but a proof remained elusive for many years. Arguments based on Ward Identities for Quantum ElectroDynamics in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(2+1)$$\end{document}$-dimensions \[[@CR12], [@CR24], [@CR30]\], or on the properties of anomalies \[[@CR15]\], offered an alternative view on the QHE: they indicated that quantization should persists in the presence of many body interaction, but such conclusions were based on manipulations of divergent series, or of effective actions arising in a formal scaling limit. The problem of a mathematical proof of the quantization of the Hall conductivity in the presence of many-body interactions remained open for several years. After the works \[[@CR1], [@CR3], [@CR5], [@CR9], [@CR10], [@CR37]\], it was dormant for more than a decade, and then, in recent years, it was actively reconsidered again, starting from \[[@CR27]\], which proved the quantization of the Hall conductance of an interacting electron system using quasi-adiabatic evolution of the groundstate around a flux-torus, under the *assumption* of a volume-independent spectral gap. In \[[@CR22]\] we followed a different approach, and proved the quantization of the interacting Hall conductivity by writing it as a convergent series, and by showing that all the interaction corrections cancel exactly, thanks to Ward Identities. Our result holds for interacting fermionic Hamiltonians of the form $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0+U {\mathcal {V}}$$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$ is quadratic and gapped, with the chemical potential in the middle of a spectral gap of width $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta _0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {V}}$$\end{document}$ is a many body interaction, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\ll \Delta _0$$\end{document}$; this smallness condition is assumed to ensure the convergence of the power series expansion in *U* of the Euclidean correlations. The same result also follows from \[[@CR27]\], in combination with a proof of the stability of the spectral gap for such fermionic Hamiltonians \[[@CR13], [@CR26]\]. See also \[[@CR6], [@CR7]\] for alternative proofs of the main theorem in \[[@CR27]\]. Recently, the bulk-edge correspondence for a class of weakly interacting fermionic systems displaying single-mode chiral edge currents was also proved \[[@CR2]\]. Given these results on the quantization of the Hall conductivity in weakly interacting systems (i.e., with interaction strength smaller than the gap), one naturally wonders what happens for stronger interactions. We focus on the interacting extension of the spinful Haldane model \[[@CR23]\], which has been recently realized in cold atoms experiments \[[@CR31]\] and can be used as the building block of more general topological insulators \[[@CR25]\]. Extensions to related systems is straightforward, in particular to the interacting, spin-conserving, Kane--Mele model, for which the quantization of the edge conductivity has been recently established \[[@CR34]\]. We recall that, in the absence of interactions, the phase diagram of the spinful Haldane model consists of two 'trivial' insulating phases, with vanishing transverse conductivity, and two quantum Hall phases, with transverse conductivity $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{12} =\pm \,2 e^2/h$$\end{document}$, separated by two critical curves. By \[[@CR22]\], we know that, away from the critical lines, for interactions *U* smaller than the spectral gap, the Hall conductivity is quantized and independent of *U*. However, what happens close to the critical lines, in cases where the interaction is much larger than the gap? This question, and in particular the possible emergence of new quantum phases, has been extensively investigated in the literature, mainly via mean-field, variational, and numerical studies, see \[[@CR28], [@CR29], [@CR38], [@CR41], [@CR42]\] and references therein. These works show evidence for the appearance of a new interaction-induced phase with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{12} =\pm \,e^2/h$$\end{document}$, but the numerics is inconclusive on whether this phase, in the thermodynamic limit, emerges at arbitrarily small, positive, interactions or, rather, above a finite threshold. The main result of this work excludes the first possibility: no new phases appear close to the transition lines, as long as the interaction strength is sufficiently small, compared with the bandwidth $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_0$$\end{document}$. More precisely, we compute the Hall conductivity for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\ll t_0$$\end{document}$ and all the values of the parameters outside two critical curves, across which the model undergoes a 'topological' phase transition: the Hall coefficient remains integer and constant as long as we continuously deform the parameters without crossing the curves; when this happens, the Hall coefficient jumps abruptly to a different integer. The main difficulties in proving such results are related to the fact that the critical lines are non-universal (i.e., interaction-dependent), thus making a naive perturbative approach ineffective. The 'dressing' of the critical lines is analogous to what happens in the theory of second order phase transitions, where the critical temperature is modified by the interaction, and one needs to appropriately tune the temperature as the interaction is switched on, in order to stay at criticality. Technically, we proceed in a similar way: we do not expand around the non-interacting Hamiltonian but, rather, around a reference quadratic Hamiltonian, characterized by the same gap as the interacting system, whose value is fixed self-consistently. Note that our problem $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0+U{\mathcal {V}}$$\end{document}$ naturally comes with three energy scales: the spectral gap $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta _0$$\end{document}$ of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal H_0$$\end{document}$; the bandwidth $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_0$$\end{document}$ of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$; and the interaction strength *U*. Our methods are not applicable in the regime of non-perturbatively strong interactions, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U| > rsim t_0$$\end{document}$: they are limited to the case where *U* is smaller than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_0$$\end{document}$ but, as remarked above, they are allowed to be much larger than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta _0$$\end{document}$. Even in this regime, the interaction can induce drastic changes of the physical properties of the system, as well known in the context of interacting, gapless, 2D electron gases, where weak interactions can in general produce quantum (e.g., superconducting) instabilities. The reason why this does not happen in our case is due to a key feature of the model under investigation, namely that the critical, gapless, Hamiltonian has energy bands with conical intersections: this ensures that the interaction is irrelevant in a Renormalization Group sense, uniformly in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta _0$$\end{document}$. In more general cases, the interaction may be marginal, as in the case of the anisotropic Hofstadter model, recently considered in \[[@CR33]\]: in this case, the gaps with integer label are stable, but new gaps corresponding to fractional labels are expected to open. It would be, of course, very interesting to further investigate such cases, where fractional Hall conductances may potentially appear, as well as to include disorder effects, which are essential for the very existence of Hall plateaus. Our results extend and complement those of \[[@CR19]\], where we considered the same model (in the special case of ultra-local interactions) and we proved: (i) existence of the critical curves, but without an explicit control on their regularity properties, and (ii) universality of the jump in the Hall coefficient across the critical curves, but without a proof that the coefficient remains constant in each connected component of the complement of the critical curves. Combining the results of \[[@CR19]\] with those presented here, we have a complete construction of the topological phase diagram of the interacting Haldane model. Our presentation is organized as follows: in Sect. [2](#Sec2){ref-type="sec"} we define the class of interacting Haldane models that we are going to consider, and we state our main result. In Sect. [3](#Sec7){ref-type="sec"} we prove the quantization of the Hall coefficient, under suitable regularity assumptions on the Euclidean correlation functions of the interacting model; we stress that this part of the proof holds in great generality, for a class of interacting fermionic systems much larger than the interacting Haldane model. In Sect. [4](#Sec12){ref-type="sec"} we prove the regularity assumptions on the correlations for the model at hand, via rigorous renormalization group methods. In Sect. [5](#Sec13){ref-type="sec"} we put things together and complete the proof of our main result. Main Result {#Sec2} =========== The Model {#Sec3} --------- The Haldane model describes spinless fermions on the honeycomb lattice hopping on nearest and next-to-nearest neighbours, in the presence of a transverse magnetic field, with *zero net flux* through the hexagonal cell, and of a staggered potential. In this section we introduce an interacting, spinful, version of the Haldane model. Note that, in the presence of interactions, the spin could induce a qualitatively different behaviour, as compared with the spinless case (this is a well known fact in the context of one-dimensional fermions \[[@CR14]\], including the edge theory of 2D topological insulators \[[@CR2], [@CR34]\]). Note also that the experimental realization of the interacting Haldane model involves, indeed, spin-1 / 2 particles, see \[[@CR31]\]. Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda =\big \{{\vec {x}} \mid {\vec {x}} = n_{1} {\vec {\ell }}_{1} + n_{2} {\vec {\ell }}_{2},\; n_{i} \in {\mathbb {Z}}\}\subset {\mathbb {R}}^{2}$$\end{document}$ be the infinite triangular lattice generated by the two basis vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_{1} = \frac{1}{2}(3,\, -\sqrt{3}), {\vec {\ell }}_{2} = \frac{1}{2}(3, \sqrt{3})$$\end{document}$. Given $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L\in {\mathbb {N}}$$\end{document}$, we also let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{L}=\Lambda / L\Lambda $$\end{document}$ be the corresponding finite periodic triangular lattice of side *L*, which will be identified with the set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _L= \big \{ {\vec {x}} \mid {\vec {x}} = n_{1} {\vec {\ell }}_{1} + n_{2} {\vec {\ell }}_{2},\; n_{i} \in {\mathbb {Z}}\cap [0,L) \big \}$$\end{document}$ with periodic boundary conditions. The lattice is endowed with the Euclidean distance on the torus, denoted by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {\vec {x}} - {\vec {y}}|_L=\min _{m\in {\mathbb {Z}}^2}| {\vec {x}} - {\vec {y}}+m_1 {\vec {\ell }}_1 L+m_2 {\vec {\ell }}_2 L|$$\end{document}$. The number of sites of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{L}$$\end{document}$ is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\Lambda _{L}| = L^{2}$$\end{document}$. The periodic honeycomb lattice can be realized as the superposition of two periodic triangular sublattices $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda ^{\text {A}}_{L} \equiv \Lambda _{L}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda ^{\text {B}}_{L}\equiv \Lambda _{L} + {\vec {e}}_1$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {e}}_1=(1,0)$$\end{document}$ the first Euclidean basis vector. Equivalently, we can think the honeycomb lattice as a triangular lattice, with two internal degrees of freedom corresponding to the *A*, *B* sublattices. It is convenient to define the model in second quantization. The one-particle Hilbert space is the set of functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {h}}_{L} = \{ f: \Lambda _{L}\times \{ \uparrow , \downarrow \}\times \{A, B\}\rightarrow {\mathbb {C}} \} \simeq {\mathbb {C}}^{L^{2}}\otimes {\mathbb {C}}^{4}$$\end{document}$. We let the fermionic Fock space $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {F}}_{L}$$\end{document}$ be the exterior algebra of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {h}}_{L}$$\end{document}$. Notice that for fixed *L*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {F}}_{L}$$\end{document}$ is a finite-dimensional space. For a given site $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {x}}\in \Lambda _{L}$$\end{document}$, we introduce fermionic annihilation operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi _{{\vec {x}}, \rho , s}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in \{A, B\}$$\end{document}$ the sublattice label and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s \in \{\uparrow , \downarrow \}$$\end{document}$ the spin label, and we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi ^\dagger _{{\vec {x}},\rho ,s}$$\end{document}$ their adjoint, the creation operators. They satisfy the standard canonical anticommutation relations $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \psi ^{\dagger }_{{\vec {x}}, \rho , s}, \psi _{{\vec {y}}, \rho ', s'}\} = \delta _{\rho ,\rho '} \delta _{s,s'} \delta _{{\vec {x}},{\vec {y}}}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \psi ^{\dagger }_{{\vec {x}}, \rho , s}, \psi ^{\dagger }_{{\vec {y}}, \rho ', s'}\} = \{ \psi _{{\vec {x}}, \rho , s}, \psi _{{\vec {y}}, \rho ', s'}\} = 0$$\end{document}$. The operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi _{{\vec {x}},\rho ,s}$$\end{document}$ are consistent with the periodic boundary conditions on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{L}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi _{{\vec {x}} + n_{1} L + n_{2} L,\rho ,s} = \psi _{{\vec {x}},\rho ,s}$$\end{document}$. The reciprocal lattice $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{L}^{*}$$\end{document}$ of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{L}$$\end{document}$ is the triangular lattice generated by the basis vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_{1}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_{2}$$\end{document}$, such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_i\cdot {\vec {\ell }}_j=2\pi \delta _{i,j}$$\end{document}$. Explicitely, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_{1} = \frac{2\pi }{3}(1,\, -\sqrt{3})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_{2} = \frac{2\pi }{3}(1,\, \sqrt{3})$$\end{document}$. We define the finite-volume Brillouin zone as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {B}}_{L} := \Big \{ {\vec {k}} \in {\mathbb {R}}^{2} \mid {\vec {k}} = \frac{n_1}{L} {\vec {G}}_{1} + \frac{n_2}{L} {\vec {G}}_{2},\; n_{i}\in {\mathbb {Z}}\cap [0,L) \Big \}$$\end{document}$. We define the Fourier transforms of the fermionic operators as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \psi _{{\vec {x}},\rho ,s} = \frac{1}{L^{2}} \sum _{{\vec {k}} \in {\mathcal {B}}_{L}} e^{- i {\vec {k}}\cdot {\vec {x}}} {{\hat{\psi }}}_{{\vec {k}}, \rho ,s} \quad \forall {\vec {x}}\in \Lambda _{L} \Longleftrightarrow {{\hat{\psi }}}_{{\vec {k}}, \rho , s} = \sum _{{\vec {x}}\in \Lambda _{L}} e^{+ i{\vec {k}}\cdot {\vec {x}}} \psi _{{\vec {x}}, \rho , s}\quad \forall {\vec {k}}\in {\mathcal {B}}_{L}\;.\nonumber \\ \end{aligned}$$\end{document}$$With this definition, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\psi }}}_{{\vec {k}},\rho ,s}$$\end{document}$ is periodic over the Brillouin zone, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\psi }}}_{{\vec {k}} + m_{1} {\vec {G}}_{1} + m_{2} {\vec {G}}_{2},\rho ,s} = {{\hat{\psi }}}_{{\vec {k}},\rho ,s}$$\end{document}$. Moreover, the Fourier transforms of the fermionic operators satisfy the anticommutation relations: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ {{\hat{\psi }}}^{\dagger }_{{\vec {k}}, \rho , s}, {{\hat{\psi }}}_{{\vec {k}}', \rho ', s'} \} = L^{2} \delta _{{\vec {k}}, {\vec {k}}'}\delta _{\rho ,\rho '}\delta _{s,s'}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ {{\hat{\psi }}}^{\dagger }_{{\vec {k}}, \rho , s}, {{\hat{\psi }}}^{\dagger }_{{\vec {k}}', \rho ', s'} \} = \{ {{\hat{\psi }}}_{{\vec {k}}, \rho , s}, {{\hat{\psi }}}_{{\vec {k}}', \rho ', s'} \} = 0$$\end{document}$. The Hamiltonian of the model is: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}} = {\mathcal {H}}_0 + U{\mathcal {V}}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$ the noninteracting Hamiltonian and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {V}}$$\end{document}$ the many-body interaction of strength *U*. We have:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}_0= & {} -t_{1} \sum _{{\vec {x}} \in \Lambda _{L}}\sum _{s = \uparrow , \downarrow } [ \psi ^{\dagger }_{{\vec {x}}, A, \sigma } \psi _{{\vec {x}}, B, s} + \psi ^{\dagger }_{{\vec {x}}, A, s} \psi _{{\vec {x}} -{\vec {\ell }}_{1}, B, s} + \psi ^{\dagger }_{{\vec {x}}, A, s} \psi _{{\vec {x}} - {\vec {\ell }}_{2}, B, s} + \text {h.c.} ] \nonumber \\&- t_{2} \sum _{{\vec {x}} \in \Lambda _{L}}\sum _{\begin{array}{c} \alpha = \pm \\ j=1,2,3 \end{array}} \sum _{s=\uparrow \downarrow } [ e^{i\alpha \phi } \psi ^{\dagger }_{{\vec {x}},A,s}\psi _{{\vec {x}} + \alpha {\vec {\gamma }}_{j}, A, s} + e^{-i\alpha \phi }\psi ^{\dagger }_{{\vec {x}},B,s}\psi _{{\vec {x}} + \alpha {\vec {\gamma }}_{j}, B, s} ]\nonumber \\&+ W \sum _{{\vec {x}}\in \Lambda _{L}} [n_{{\vec {x}}, A} - n_{{\vec {x}}, B}] - \mu \sum _{{\vec {x}}\in \Lambda _{L}} [ n_{{\vec {x}}, A}+n_{{\vec {x}},B}]\;, \end{aligned}$$\end{document}$$with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\gamma }}_{1} = {\vec {\ell }}_{1} - {\vec {\ell }}_{2}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\gamma }}_{2} = {\vec {\ell }}_{2}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\gamma }}_{3} = -{\vec {\ell }}_{1}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\vec {x}},\rho } = \sum _{s=\uparrow ,\downarrow }\psi ^{\dagger }_{{\vec {x}},\rho ,s}\psi _{{\vec {x}},\rho ,s}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in \{A,B\}$$\end{document}$. For definiteness, we assume that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{1} > 0$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{2} > 0$$\end{document}$. The term proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_1$$\end{document}$ describes nearest neighbor hopping on the hexagonal lattice. The term proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_2$$\end{document}$ describes next-to-nearest neighbor hopping, with the complex phases $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{\pm \, i\phi }$$\end{document}$ modeling the effect of an external, transverse, magnetic field. The term proportional to *W* describes a staggered potential, favoring the occupancy of the *A* or *B* sublattice, depending on whether *W* is negative or positive. Finally, the term proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu $$\end{document}$ is the chemical potential, which controls the average particle density in the Gibbs state. See Fig. [1](#Fig1){ref-type="fig"}. Concerning the many-body interaction, we assume it to be a density--density interaction of the form:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {V}} = \sum _{{\vec {x}},{\vec {y}}\in \Lambda _{L}}\sum _{\rho = A, B} (n_{{\vec {x}}, \rho }-1) v_{\rho \rho '}({\vec {x}}-{\vec {y}}) (n_{{\vec {y}},\rho '}-1)\;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{AA}({\vec {x}})=v_{BB}({\vec {x}})=v({\vec {x}})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{AB}({\vec {x}})=v({\vec {x}}-{\vec {e}}_1)$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{BA}({\vec {x}})=v({\vec {x}}+{\vec {e}}_1)$$\end{document}$, with *v* a finite range, rotationally invariant, potential (we recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {e}}_1$$\end{document}$ is the first Euclidean basis vector).Fig. 1The honeycomb lattice of the Haldane model. The empty dots belong to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda ^{\text {A}}_{L}$$\end{document}$, while the black dots belong to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda ^{\text {B}}_{L}$$\end{document}$. The oval encircles the two sites of the fundamental cell, labeled by the position of the empty dot, i.e., of the site of the *A* sublattice. The nearest neighbor vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\delta }}_i$$\end{document}$, are shown explicitly, together with the next-to-nearest neighbor vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\gamma }}_i$$\end{document}$, and the two basis vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_{1,2}$$\end{document}$ of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _L$$\end{document}$ The noninteracting Hamiltonian can be rewritten as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}_0= \sum _{{\vec {x}},{\vec {y}}} \sum _{\rho , \rho ', s} \psi ^{\dagger }_{{\vec {x}}, \rho , s} H_{\rho \rho '}({\vec {x}}- {\vec {y}}) \psi _{{\vec {y}}, \rho ', s}\;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{\rho \rho '}({\vec {x}}- {\vec {y}})$$\end{document}$ are the matrix elements of the Haldane model; we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H({\vec {x}}- {\vec {y}})$$\end{document}$ the corresponding $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\times 2$$\end{document}$ block. We introduce the Bloch Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{H}}}(k) = \sum _{{\vec {z}}\in \Lambda _{L}} e^{-i{\vec {k}}\cdot {\vec {z}}} H({\vec {z}})$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {k}} \in {\mathcal {B}}_{L}$$\end{document}$. An explicit computation gives:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{H}}}( {\vec {k}}) = \begin{pmatrix} -2t_2\alpha _1({\vec {k}})\cos \phi + m({\vec {k}})-\mu &{} -t_{1} \Omega ^*( {\vec {k}}) \\ - t_{1}\Omega ( {\vec {k}}) &{} -2t_2\alpha _1({\vec {k}})\cos \phi - m({\vec {k}})-\mu \end{pmatrix} \end{aligned}$$\end{document}$$where:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&\alpha _{1}({\vec {k}}) = \sum _{j=1}^3\cos ({\vec {k}}\cdot {\vec {\gamma }}_j)\;,\qquad m({\vec {k}}) = W - 2t_{2}\sin \phi \, \alpha _{2}({\vec {k}})\;, \\&\alpha _{2}({\vec {k}}) = \sum _{j=1}^3\sin ({\vec {k}}\cdot {\vec {\gamma }}_j) \;,\qquad \Omega ({\vec {k}}) = 1 + e^{-i{\vec {k}}\cdot {\vec {\ell }}_1} + e^{-i{\vec {k}}\cdot {\vec {\ell }}_2}\;. \end{aligned} \end{aligned}$$\end{document}$$The corresponding energy bands are$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \varepsilon _{\pm }({\vec {k}}) =-2t_2\alpha _1({\vec {k}})\cos \phi \pm \sqrt{m({\vec {k}})^{2} + t_1^{2}|\Omega ({\vec {k}})|^{2}}\;. \end{aligned}$$\end{document}$$The size of the bands can be bounded by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max _{{\vec {k}}}\varepsilon _+({\vec {k}})- \min _{{\vec {k}}}\varepsilon _-({\vec {k}})$$\end{document}$, which we call the *bandwidth*. To make sure that the energy bands do not overlap, we assume that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_2/t_1<1/3$$\end{document}$. For $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L\rightarrow \infty $$\end{document}$, the two bands can touch only at the *Fermi points*$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {k}}_{F}^{\pm } = \big ( \frac{2\pi }{3}, \pm \frac{2\pi }{3\sqrt{3}} \big )$$\end{document}$, which are the two zeros of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega ({\vec {k}})$$\end{document}$, around which $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega ({\vec {k}}_F^\pm +{\vec {k}}')\simeq \frac{3}{2}(ik_1'\pm k_2')$$\end{document}$. The condition that the two bands touch at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {k}}_F^\omega $$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \in \{+,-\}$$\end{document}$, is that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_\omega =0$$\end{document}$, with$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} m_{\omega } \equiv m({\vec {k}}_{F}^{\omega }) = W +\omega 3\sqrt{3}\,t_{2}\sin \phi \;. \end{aligned}$$\end{document}$$If, instead, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_+$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_-$$\end{document}$ are both different from zero, then the spectrum of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{H}}}({\vec {k}})$$\end{document}$ is gapped for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {k}}$$\end{document}$, corresponding to an insulating phase. Lattice Currents and Linear Reponse Theory {#Sec4} ------------------------------------------ Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\vec {x}}} = \sum _{\rho = A,B} n_{{\vec {x}}, \rho }$$\end{document}$ be the total density operator at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {x}}$$\end{document}$. Its time-evolution is given by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\vec {x}}}(t) = e^{i{\mathcal {H}} t} n_{{\vec {x}}} e^{-i{\mathcal {H}} t}$$\end{document}$, which satisfies the following *lattice continuity equation*:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \partial _{t} n_{{\vec {x}}}(t) = i[ {\mathcal {H}}, n_{{\vec {x}}}(t) ] \equiv \sum _{{\vec {y}}}j_{{\vec {x}},{\vec {y}}}(t)\;, \end{aligned}$$\end{document}$$with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{{\vec {x}},{\vec {y}}}$$\end{document}$ the *bond current*:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} j_{{\vec {x}},{\vec {y}}} = \sum _{\rho ,\rho '=A,B}\ \sum _{s=\uparrow ,\downarrow } (i\psi ^{\dagger }_{{\vec {y}}, \rho ',s} H_{\rho '\rho }({\vec {y}}-{\vec {x}}) \psi _{{\vec {x}},\rho ,s} + \text {h.c.})\;. \end{aligned}$$\end{document}$$Notice that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{{\vec {x}},{\vec {y}}} = -j_{{\vec {y}},{\vec {x}}}$$\end{document}$. Thus, using that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H({\vec {x}}) \ne 0$$\end{document}$ if and only if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {x}} = {\vec {0}}, \pm {\vec {\ell }}_{1},\pm {\vec {\ell }}_2, \pm ({\vec {\ell }}_{1} - {\vec {\ell }}_{2})$$\end{document}$, Eq. ([2.9](#Equ9){ref-type=""}) implies:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \partial _{t} n_{{\vec {x}}}(t)= & {} \sum _{{\vec {y}}} j_{{\vec {x}},{\vec {y}}}(t) = \sum _{i=1,2} [j_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{i}} + j_{{\vec {x}}, {\vec {x}} - {\vec {\ell }}_{i}}] + j_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{1} - {\vec {\ell }}_{2}} + j_{{\vec {x}}, {\vec {x}} - {\vec {\ell }}_{1} + {\vec {\ell }}_{2}}\nonumber \\\equiv & {} -\text {d}_{1} \tilde{\text {J}}_{1, {\vec {x}}} - \text {d}_{2} \tilde{\text {J}}_{2, {\vec {x}}}\;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {d}_{i} f({\vec {x}}) = f({\vec {x}}) - f({\vec {x}} - {\vec {\ell }}_{i})$$\end{document}$ is the lattice derivative along the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_{i}$$\end{document}$ direction, and:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tilde{\text {J}}_{1,{\vec {x}}} = -j_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{1}} - j_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{1} - {\vec {\ell }}_{2}}\;,\qquad \tilde{\text {J}}_{2,{\vec {x}}} = -j_{{\vec {x}}, {\vec {x}} + {\vec {\ell }}_{2}} - j_{{\vec {x}}, {\vec {x}} - {\vec {\ell }}_{1} + {\vec {\ell }}_{2}}\;. \end{aligned}$$\end{document}$$The operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\text {J}}_{i, {\vec {x}}}$$\end{document}$ are the components along the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_{i}$$\end{document}$ directions of the total vectorial current, defined as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\vec {\jmath }}_{{\vec {x}}} = \tilde{\text {J}}_{1,{\vec {x}}} {\vec {\ell }}_{1}+ \tilde{\text {J}}_{2,{\vec {x}}}{\vec {\ell }}_{2}\;. \end{aligned}$$\end{document}$$Note that, given the definitions of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_{1,2}$$\end{document}$, the components of the lattice current along the two reference, orthogonal, coordinate directions are:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} j_{1,{\vec {x}}}=\frac{3}{2}(\tilde{\text {J}}_{1,{\vec {x}}}+\tilde{\text {J}}_{2,{\vec {x}}}),\qquad j_{2,{\vec {x}}}=\frac{\sqrt{3}}{2}(-\tilde{\text {J}}_{1,{\vec {x}}}+\tilde{\text {J}}_{2,{\vec {x}}}). \end{aligned}$$\end{document}$$We are interested in the transport properties of the Haldane--Hubbard model, in the linear response regime. The *Gibbs state* of the interacting model is defined as: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \cdot \rangle _{\beta , L} = \mathrm {Tr}_{{\mathcal {F}}_{L}} \cdot e^{-\beta {\mathcal {H}}} / {\mathcal {Z}}_{\beta , L}$$\end{document}$ with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {Z}}_{\beta , L} = \mathrm {Tr}_{{\mathcal {F}}_{L}} e^{-\beta {\mathcal {H}}}$$\end{document}$ the partition function. We define the conductivity matrix via the *Kubo formula*, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i, j =1,2$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{ij}:= & {} \frac{1}{|{\vec {\ell }}_1\wedge {\vec {\ell }}_2|}\lim _{p_0 \rightarrow 0^{+}} \frac{1}{p_0}\Big [-i \int _{-\infty }^{0} dt\, e^{p_0 t} \lim _{\beta , L\rightarrow \infty } \frac{1}{L^{2}} \langle [ {\mathcal {J}}_{i}\,, {\mathcal {J}}_{j}(t) ] \rangle _{\beta , L} \nonumber \\&+ i \lim _{\beta , L\rightarrow \infty } \frac{1}{L^{2}} \langle [{\mathcal {J}}_i,{\mathcal {X}}_j] \rangle _{\beta , L}\Big ]\;, \end{aligned}$$\end{document}$$with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec {{\mathcal {J}}} = \sum _{{\vec {x}}\in \Lambda _L} {\vec {\jmath }}_{{\vec {x}}}$$\end{document}$ the total current operator, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec {{\mathcal {X}}}$$\end{document}$ the second quantization of the position operator[1](#Fn1){ref-type="fn"}, and where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\beta ,L\rightarrow \infty }$$\end{document}$ must be understood as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\beta \rightarrow \infty }\lim _{L\rightarrow \infty }$$\end{document}$, i.e., thermodynamic limit first, and then temperature to zero. Note that formally, in the thermodynamic limit, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec {{\mathcal {J}}} =i[\mathcal H,\vec {{\mathcal {X}}}]$$\end{document}$, as it should. Equation ([2.15](#Equ15){ref-type=""}) describes the linear response of the average current at the time $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t= 0$$\end{document}$ to an adiabatic external potential of the form $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{\eta t} {\vec {E}}\cdot \vec {{\mathcal {X}}}$$\end{document}$, see e.g. \[[@CR17]\] for a formal derivation, and \[[@CR8], [@CR36], [@CR39]\] for a rigorous derivation in a slightly different setting. **Remark.** The indices *i*, *j* labelling the elements of the conductivity matrix ([2.15](#Equ15){ref-type=""}) refer to the two reference, orthogonal, coordinate directions. Sometimes, a similar definition of the Kubo matrix is given, where, instead, the indices *i*, *j* label the two lattice coordinate directions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_1,{\vec {\ell }}_2$$\end{document}$ ('adapted basis'). The two definitions are, of course, related in a simple way, via the transformation induced by the change of basis. In particular, the transverse conductivities defined in the orthogonal and in the adapted basis are the same, up to an overall multiplicative factor, equal to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|{\vec {\ell }}_1\wedge {\vec {\ell }}_2|$$\end{document}$. The longitudinal conductivities are, instead, related via a matrix relation that mixes up the diagonal and non-diagonal components of the conductivity matrix. For ease of comparison with experimental papers on graphene, or graphene-like materials, we prefer to use the definition involving the orthogonal reference directions, which we find more natural. In the absence of interactions, the Kubo conductivity matrix of the Haldane model can be computed explicitly. Suppose that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\omega } \ne 0$$\end{document}$, both for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$ and for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =-$$\end{document}$, and let us choose the chemical potential in the spectral gap. For instance, let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu = -2t_{2} \cos \phi \alpha _{1}(k_{F}^{\omega })$$\end{document}$, which corresponds to choosing the chemical potential in the 'middle of the gap'. Then, it turns out that \[[@CR23]\]:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{11} = 0\;,\qquad \sigma _{12} = -\sigma _{21} = \frac{\nu }{2\pi }\;,\qquad \nu = \mathrm{sign}(m_{+}) - \mathrm{sign}(m_{-})\;. \end{aligned}$$\end{document}$$The integer $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu $$\end{document}$ is the Chern number of the Bloch bundle associated to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{H}}}({\vec {k}})$$\end{document}$. The zeros of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\omega }=W+\omega 3\sqrt{3} t_2\sin \phi $$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \in \{+,-\}$$\end{document}$, define the *critical curves* of the Haldane model, which separate the different topological phases, corresponding to different values of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu $$\end{document}$. *On* the curves, the spectrum is gapless: the energy bands intersect with conical intersection, and the system displays a quantization phenomenon of the *longitudinal* conductivity:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{11} = \sigma _{22} = \frac{1}{8}\;, \end{aligned}$$\end{document}$$while $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{11} =\sigma _{22}= \frac{1}{4}$$\end{document}$ at the 'graphene points' $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{+} = m_{-} = 0$$\end{document}$. Main Result: Interacting Topological Phases and Phase Transitions {#Sec5} ----------------------------------------------------------------- Let us now turn on the many-body interaction, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U\ne 0$$\end{document}$. In previous works, it was proved that the quantization of the conductivity persists, but only for interactions of strength *much smaller than the gap of*$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$. Our main result, summarized in the next theorem, overcomes this limitation. ### Theorem 2.1 {#FPar1} There exists $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{0} >0$$\end{document}$, independent of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W,\phi $$\end{document}$, such that for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U| < U_0$$\end{document}$ the following is true. There exist two functions, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {d}}(U,W, \phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {z}}(U,W, \phi )$$\end{document}$, analytic in *U* and continuously differentiable in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W,\phi $$\end{document}$, such that, if the chemical potential is fixed at the value $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu = -2t_{2} \cos \phi \alpha _{1}(k_{F}^{\omega }) {-} {\mathfrak {z}}(U,W, \phi )$$\end{document}$, then, for all the values of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W,\phi $$\end{document}$ such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${m_{\mathrm {R},\omega }}(W,\phi ):=W+\omega 3\sqrt{3} t_2\sin \phi +{\omega {\mathfrak {d}}(U,-\omega W,\phi )}$$\end{document}$ is different from zero, both for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$ and for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =-$$\end{document}$, the interacting Hall conductivity is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{12}(U)= \frac{1}{2\pi }\big [ \mathrm{sign}(m_{\mathrm {R},+}) - \mathrm{sign}(m_{\mathrm {R},-}) \big ]\;. \end{aligned}$$\end{document}$$Moreover, the conditions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_\omega ^R(W,\phi )=0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \in \{\pm \}$$\end{document}$, define two $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^1$$\end{document}$ curves $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=W^R_\omega (\phi )$$\end{document}$, called 'critical curves', which are $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^1$$\end{document}$ close to the unperturbed curves $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=-\omega 3\sqrt{3} t_2\sin \phi $$\end{document}$. The two critical curves have the same qualitative properties as the unperturbed ones, in the sense that: (i) they intersect at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(W,\phi )=(0,0), (0,\pi )$$\end{document}$; (ii) they are one the image of the other, under the reflection $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W\rightarrow -W$$\end{document}$; (iii) they are monotone for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \in [-\frac{\pi }{2},\frac{\pi }{2}]$$\end{document}$; (iv) they are odd in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$, and their periodic extension to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb {R}}$$\end{document}$ is even under the reflection $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \rightarrow \pi -\phi $$\end{document}$. An illustration of how the interaction deforms the critical lines is shown in Fig.[2](#Fig2){ref-type="fig"}.Fig. 2Illustration of the deformation of the critical lines induced by the electron--electron interaction. The solid red curve corresponds to the non-interacting Haldane model with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_1=1$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_2=0.1$$\end{document}$. The dotted and dashed-dotted lines correspond to the interacting case, with ultra-local (on-site) interaction and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U=\pm \,0.5$$\end{document}$; the lines are computed by truncating perturbation theory to first order, see \[[@CR19], Sect. III.E\] for details. The connected regions of the complement of the critical lines are labelled NI (resp. TI), if they correspond to the 'normal' (resp. 'topological') insulating phase. Notice that, in the considered example, repulsive interactions have the effect of enhancing the topological insulating phase. It would be interesting to have a conceptual understanding of this phenomenon, that is, of why repulsive interactions favor the non-trivial topological phase The main improvement of the result stated in Theorem [2.1](#FPar1){ref-type="sec"} compared to previous works is that it establishes the quantization of the Hall conductivity for values of the coupling constant *U* that are *much larger* than the gap of the bare Hamiltonian: it states that the interaction does not change the value of the interacting Hall conductivity, provided we do not cross the interacting critical curves, which we construct explicitly; this universality of the Hall coefficient holds, in particular, arbitrarily close to the critical curves. On the critical curves the system is massless, i.e., correlations decay algebraically at large distances, and we do not have informations on the transverse conductivity coefficient. However, the critical longitudinal conductivity displays the same quantization phenomenon as the non-interacting one: namely, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=W^{R}_\omega (\phi )$$\end{document}$, for either $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$ or $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =-$$\end{document}$, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \ne 0,\pi $$\end{document}$, then$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{11} = \sigma _{22} = \frac{1}{8}\;, \end{aligned}$$\end{document}$$while $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{11} =\sigma _{22}= \frac{1}{4}$$\end{document}$ for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(W,\phi )=(0,0),(0,\pi )$$\end{document}$; see \[[@CR19]\] for the proof. We remark that the proof of Theorem [2.1](#FPar1){ref-type="sec"} is constructive: therefore, a patient reader can extract from it an explicit bound on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_0$$\end{document}$. Such a bound would certainly be far from optimal; optimizing it would be a non-trivial, interesting, exercise, requiring a computer-assisted proof (at least if one is interested in getting a physically significant bound). In any case, conceptually, the only important requirement should be that *U* is sufficiently small, compared to the bandwidth of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$, see the definition after ([2.7](#Equ7){ref-type=""}). Finally, concerning the model: we expect that the specific choice of the interacting Haldane model is not crucial for the validity of the result. The proof extends straightforwardly to strictly related models, such as the spin-conserving Kane--Mele model. An appropriate adaptation should apply, more generally, to any interacting Hamiltonian of the form $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}} = {\mathcal {H}}_0 + U{\mathcal {V}}$$\end{document}$, with: (i) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {V}}$$\end{document}$ a short-range, spin-independent, interaction, (ii) \|*U*\| small compared to the bandwidth, and (iii) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$ a quadratic Hamiltonian that can become gapless as a parameter is varied: in the gapless case, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0$$\end{document}$ has a degenerate, point-like, Fermi surface, around which the dispersion relation has a linear, 'graphene-like', behavior. Note that, as discussed in the introduction, the latter condition is needed to guarantee the irrelevance of the interaction. Even if conceptually non problematic, the extension to such a general class of many-body Hamiltonians would require a discussion of their symmetry properties, in connection with the classification of the possible relevant and marginal effective coupling that can be generated under the multiscale Renormalization Group construction of the Euclidean correlations, cf. with Sect. [4](#Sec12){ref-type="sec"} below. This goes beyond the scopes of this article: for this reason, we restrict to the specific example of the interacting Haldane model, which is physically the most relevant for applications to 2D topological insulators. ### Strategy of the Proof {#Sec6} Let us give an informal summary of the main steps of the proof. For simplicity, we limit ourselves to the generic case $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W\ne 0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \ne 0$$\end{document}$, the special, symmetric, complementary case ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=0$$\end{document}$ and/or $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi =0$$\end{document}$) being treatable analogously. Thanks to the symmetries of the model, see Eqs.([4.7](#Equ62){ref-type=""})--([4.13](#Equ68){ref-type=""}) below, we further restrict ourselves, without loss of generality, to the range of parameters$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W>0,\qquad 0<\phi \leqslant \frac{\pi }{2}, \end{aligned}$$\end{document}$$which corresponds to the case $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{+}>|m_{-}|$$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_\pm $$\end{document}$ are defined in ([2.8](#Equ8){ref-type=""}). Note that, under these conditions, the amplitude of the bare gap is given by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|m_-|$$\end{document}$. We expect the interaction to modify ('renormalize') in a non trivial way both the chemical potential and the width of the gap[2](#Fn2){ref-type="fn"}. In order to compute the interacting gap, we proceed as follows. For the purpose of this discussion, let us denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0(W,\phi ,\mu )$$\end{document}$ the non-interacting Hamiltonian ([2.2](#Equ2){ref-type=""}), thought of as a function of the parameters $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(W,\phi ,\mu )$$\end{document}$, at fixed $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_1,t_2$$\end{document}$. We rewrite $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu $$\end{document}$ in the form $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu =-2t_{2}\cos \phi \, \alpha _{1}(k_{F}^{\omega })-{{\mathfrak {z}}}$$\end{document}$, and, recalling that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=m_-+3\sqrt{3} t_2\sin \phi $$\end{document}$, we rewrite $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=(m_--{{\mathfrak {d}}})+3\sqrt{3}t_2\sin \phi +{{\mathfrak {d}}} \equiv m_{\text {R},-}+3\sqrt{3}t_2\sin \phi +{{\mathfrak {d}}}$$\end{document}$, where the parameter $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathfrak {d}}}$$\end{document}$ will be chosen in such a way that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}=m_--{{\mathfrak {d}}}$$\end{document}$ has the interpretation of *renormalized gap*. By using these rewritings, we find:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}={\mathcal {H}}_0(W,\phi ,\mu )+U{\mathcal {V}}=\mathcal H_0^{\text {R}}(m_{\text {R},-},\phi )+ U{\mathcal {V}}+ {{\mathfrak {d}}}\sum _{{\vec {x}}\in \Lambda _{L}}[n_{{\vec {x}}, A} - n_{{\vec {x}}, B}] +{{\mathfrak {z}}} \sum _{{\vec {x}}\in \Lambda _L}n_{{\vec {x}}},\nonumber \\ \end{aligned}$$\end{document}$$where$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}_0^{\text {R}}(m_{\text {R},-},\phi ):=\mathcal H_0(m_{\text {R},-}+3\sqrt{3} t_2\sin \phi , \phi ,-2t_{2}\cos \phi \, \alpha _{1}(k_{F}^{\omega })). \end{aligned}$$\end{document}$$Let us now introduce the reference Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal H^{\text {R}}$$\end{document}$, thought of as a function of the parameters $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U,m_{\text {R},-},\phi $$\end{document}$, defined by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}^\text {R}: ={\mathcal {H}}_0^{\text {R}}(m_{\text {R},-},\phi ) + U{\mathcal {V}}+ \delta (U,m_{\text {R},-},\phi )\sum _{{\vec {x}}\in \Lambda _{L}}[n_{{\vec {x}}, A} - n_{{\vec {x}}, B}] +\xi (U,m_{\text {R},-},\phi ) \sum _{{\vec {x}}\in \Lambda _L}n_{{\vec {x}}}.\nonumber \\ \end{aligned}$$\end{document}$$Note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$ in ([2.21](#Equ21){ref-type=""}) has the same form as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}$$\end{document}$ in ([2.22](#Equ22){ref-type=""}), with the important difference that in passing from $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$ to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal H^{\text {R}}$$\end{document}$, the parameters $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {d}}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathfrak {z}}$$\end{document}$ have been replaced by the two functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U,m_{\text {R},-},\phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi (U,m_{\text {R},-},\phi )$$\end{document}$; for the moment, these two functions should be thought of as being arbitrary: they will be conveniently fixed below. Therefore, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^\text {R}$$\end{document}$ is in general different from the original Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$. However, by construction, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}={\mathcal {H}}^{\text {R}}$$\end{document}$, provided that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu =-2t_{2}\cos \phi \, \alpha _{1}(k_{F}^{\omega })-\xi (U,m_{\text {R},-},\phi )$$\end{document}$, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$ is a solution of the fixed point equation$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} m_{\text {R}, -} =W-3\sqrt{3} t_2\sin \phi -\delta (U,m_{\text {R},-},\phi )\;. \end{aligned}$$\end{document}$$Our construction, described below, will allow us to fix the counterterms $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi (U,m_{\text {R},-},\phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U,m_{\text {R},-},\phi )$$\end{document}$ in such a way that they are small, of order *O*(*U*), and that, as anticipated above, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$ has the interpretation of renormalized gap: in particular, the condition $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$ implies that the system is massive, that is, correlations decay exponentially at large distances, with decay rate $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$. Given these definitions, the main steps of the proof are the following.(i)We introduce the Euclidean correlations and the Euclidean Hall conductivity, which are formally obtained from the corresponding real-time formulas via a 'Wick rotation' of the time variable. In Lemma [3.4](#FPar7){ref-type="sec"}, by differentiating the Ward Identities associated with the continuity equation, and by combining the result with the Schwinger--Dyson equation, we show that the Euclidean Hall conductivity of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text{ R }}$$\end{document}$ is constant in *U*, provided that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi (U,m_{\text {R},-},\phi ),\delta (U,m_{\text {R},-},\phi )$$\end{document}$ are differentiable in *U* and that the Fourier transform of the Euclidean correlation functions is smooth (i.e., at least of class $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{3}$$\end{document}$) in the momenta, for any fixed $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$.(ii)As a second step, we prove the assumptions of Lemma [3.4](#FPar7){ref-type="sec"}. More precisely, we prove that there exist two functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi (U, m_{\text {R}, -}, \phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U, m_{\text {R}, -}, \phi )$$\end{document}$, analytic in *U*, such that the Euclidean correlations of the model ([2.22](#Equ22){ref-type=""}) are analytic in *U* and, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$, they are exponentially decaying at large space-time distances, with decay rate $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$; in particular, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-} \ne 0$$\end{document}$, their Fourier transform is smooth in the momenta.(iii)Next, we prove the equivalence between the original model and the model with Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}$$\end{document}$, anticipated above. In particular, we prove that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta $$\end{document}$ is differentiable in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R}, -}$$\end{document}$, with small (i.e., *O*(*U*)) derivative; therefore, eq. ([2.23](#Equ23){ref-type=""}) can be solved via the implicit function theorem, thus giving $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} m_{\text {R},-} =W-3\sqrt{3} t_2\sin \phi {-{\mathfrak {d}}(U,W,\phi )}, \end{aligned}$$\end{document}$$ and we show that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|{\mathfrak {d}}(U,W,\phi )|\leqslant C|U| (W+\sin \phi )$$\end{document}$. The equation for the interacting critical curve has the form: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=3\sqrt{3}\,t_{2}\sin \phi +\delta (U,0,\phi )=(1+O(U))\,3\sqrt{3} t_2\sin \phi $$\end{document}$.(iv)Finally, once we derived explicit estimates on the decay properties of the Euclidean correlations, we infer the identity between the Euclidean and the real-time Kubo conductivity, via \[[@CR2], Lemma B.1\].The key technical difference with respect to the strategy in \[[@CR22]\] is the rewriting of the model in terms of the renormalized reference Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}_0$$\end{document}$: this allows us to take into account the renormalization of the gap and of the chemical potential, which characterizes the interacting critical point of the theory. Lattice Conservation Laws and Universality {#Sec7} ========================================== In this section, we show how lattice conservation laws can be used to prove the universality of the Euclidean Kubo conductivity, see step (i) above. The main result of this section is summarized in Lemma [3.4](#FPar7){ref-type="sec"}. Before getting to this lemma, in Sect. [3.1](#Sec8){ref-type="sec"} we introduce the Euclidean formalism and derive the *Ward identities*, associated with the lattice continuity equation ([2.9](#Equ9){ref-type=""}), for the Euclidean correlations. In Sects. [3.1.1](#Sec9){ref-type="sec"} and [3.1.2](#Sec10){ref-type="sec"} we differentiate and manipulate the Ward identities, under the assumption that the current--current correlations are sufficiently smooth in momentum space, thus getting some important identities, summarized in Lemma [3.1](#FPar2){ref-type="sec"} and [3.2](#FPar4){ref-type="sec"}. Finally, in Sect. [3.2](#Sec11){ref-type="sec"}, we prove Lemma [3.4](#FPar7){ref-type="sec"}, by combining these identities with the Schwinger--Dyson equation. Euclidean Formalism and Ward Identities {#Sec8} --------------------------------------- Given an operator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}$$\end{document}$ on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {F}}_{L}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\in [0, \beta )$$\end{document}$, we define the imaginary-time evolution generated by the Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^\text {R}$$\end{document}$, Eq. ([2.22](#Equ22){ref-type=""}), as: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}_{t} := e^{t {\mathcal {H}}^{\text {R}}} {\mathcal {O}} e^{-t{\mathcal {H}}^{\text {R}}}$$\end{document}$. Notice that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}_{t} \equiv {\mathcal {O}}(-it)$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}(t)$$\end{document}$ the real-time evolution generated by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}$$\end{document}$. Given *n* operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}_{t_{1}}^{(1)},\ldots , {\mathcal {O}}_{t_{n}}^{(n)}$$\end{document}$ on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {F}}_{L}$$\end{document}$, each of which (i) can be written as a polynomial in the time-evolved creation and annihilation operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi ^\pm _{(t,{\vec {x}}),\rho } = e^{t {\mathcal {H}}^\text {R}} \psi ^\pm _{ {\vec {x}},\rho }e^{-t{\mathcal {H}}^\text {R}}$$\end{document}$, (ii) is normal-ordered, and (iii) is either even or odd in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi ^\pm _{(t,{\vec {x}}),\rho }$$\end{document}$, we define their time-ordered average, or *Euclidean correlation function*, as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \langle \mathbf{T}\, {\mathcal {O}}^{(1)}_{t_1}\cdots {\mathcal {O}}^{(n)}_{t_{n}} \rangle _{\beta ,L}^{\text {R}} := \frac{\mathrm {Tr}_{{\mathcal {F}}_{L}} e^{-\beta {\mathcal {H}}^\text {R}} {\mathbf {T}} \big \{ {\mathcal {O}}_{t_{1}}^{(1)}\cdots {\mathcal {O}}_{t_{n}}^{(n)} \big \} }{\mathrm {Tr}_{{\mathcal {F}}_{L}} e^{-\beta {\mathcal {H}}^{\text {R}}}} \;, \end{aligned}$$\end{document}$$where the (linear) operator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf {T}}$$\end{document}$ is the fermionic time-ordering, acting on a product of fermionic operators as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbf {T}} \big \{ \psi ^{\varepsilon _{1}}_{(t_1,{\vec {x}}_1),s_1}\cdots \psi ^{\varepsilon _{n}}_{(t_n,{\vec {x}}_n),s_n} \big \} = \text {sgn}(\pi ) \psi ^{\varepsilon _{\pi (1)}}_{(t_{\pi (1)},{\vec {x}}_{\pi (1)}),s_{\pi (1)}}\cdots \psi ^{\varepsilon _{\pi (n)}}_{ (t_{\pi (n)},{\vec {x}}_{\pi (n)}),s_{\pi (n)}} \;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon _i\in \{\pm \}$$\end{document}$ (with the understanding $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi ^-_{(t,{\vec {x}}),\rho ,s}\equiv \psi _{(t,{\vec {x}}),\rho ,s}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\psi ^+_{(t,{\vec {x}}),\rho ,s}\equiv \psi ^\dagger _{(t,{\vec {x}}),\rho ,s}$$\end{document}$), and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi $$\end{document}$ is a permutation of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{1,\ldots , n\}$$\end{document}$ with signature $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {sgn}(\pi )$$\end{document}$ such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{\pi (1)}\geqslant \ldots \geqslant t_{\pi (n)}$$\end{document}$. If some operators are evaluated at the same time, the ambiguity is solved by normal ordering. We also denote the *connected* Euclidean correlation function, or cumulant, by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \mathbf{T}\, {\mathcal {O}}^{(1)}_{t_1}\,; {\mathcal {O}}^{(2)}_{t_2}\,; \cdots \,; {\mathcal {O}}^{(n)}_{t_{n}} \rangle _{\beta ,L}^{\text {R}}$$\end{document}$. Let *O* be a self-adjoint operator on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {F}}_{L}$$\end{document}$. We define its time Fourier transform as: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{{\mathcal {O}}}_{p_0} = \int _{0}^{\beta } dt\, e^{-ip_0 t} {\mathcal {O}}_{t}$$\end{document}$ with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0 \in \frac{2\pi }{\beta }{\mathbb {Z}}$$\end{document}$ the *Matsubara frequencies*. Also, we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{{\mathcal {O}}}_{\mathbf{p}}$$\end{document}$, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}= (p_{0}, p_{1}, p_{2})$$\end{document}$, the joint space-time Fourier transform of the operator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {O}}_{(t,{\vec {x}})}$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\widehat{{\mathcal {O}}}_{\mathbf{p}} = \int _{0}^{\beta } dt\, \sum _{{\vec {x}}\in \Lambda _{L}} e^{-i\mathbf{p}\cdot \mathbf{x}} {\mathcal {O}}_{\mathbf{x}},\end{aligned}$$\end{document}$$with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{x}= (t, x_{1}, x_{2})\equiv (x_{0}, x_{1}, x_{2})$$\end{document}$. Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{\mu , {\vec {x}}}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu \in \{0,1,2\}$$\end{document}$, be the three-component operator such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{0,{\vec {x}}}:=n_{{\vec {x}}}$$\end{document}$, while $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{i,{\vec {x}}}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i\in \{1,2\}$$\end{document}$, are the components of the total current along the reference, orthogonal, coordinate directions, see ([2.14](#Equ14){ref-type=""}). Note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{\mu ,{\vec {x}}}$$\end{document}$ is the natural current operator, associated both with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$ and with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal H^{\text {R}}$$\end{document}$, because $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i[{\mathcal {H}},n_{{\vec {x}}}]=i[\mathcal H^{\text {R}},n_{{\vec {x}}}]$$\end{document}$. Therefore, its imaginary-time evolution with respect to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}$$\end{document}$ satisfies the analogue of the continuity equation ([2.11](#Equ11){ref-type=""}), cf. with ([3.5](#Equ29){ref-type=""}) below. We define the normalized current--current correlation functions as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{K}}_{\mu _{1}, \ldots , \mu _{n}}^{\beta , L; \text {R}}(\mathbf{p}_{1}, \ldots , \mathbf{p}_{n-1}) := \frac{1}{\beta L^2} \langle \mathbf{T}\, {\hat{\jmath }}_{\mu _1, \mathbf{p}_{1}}\,; {\hat{\jmath }}_{\mu _{2}, \mathbf{p}_{2}}\,; \cdots \,; {\hat{\jmath }}_{\mu _{n}, -\mathbf{p}_{1}-\ldots - \mathbf{p}_{n-1}} \rangle _{\beta , L}^{\text {R}} \end{aligned}$$\end{document}$$for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _i\in \{0,1,2\}$$\end{document}$. We also denote the infinite volume, zero temperature limit of the Euclidean correlations by: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{\mu _{1}, \ldots , \mu _{n}}(\mathbf{p}_{1}, \ldots , \mathbf{p}_{n-1}) := \lim _{\beta \rightarrow \infty }\lim _{L\rightarrow \infty } {\widehat{K}}_{\mu _{1}, \ldots , \mu _{n}}^{\beta , L; \text {R}}(\mathbf{p}_{1}, \ldots , \mathbf{p}_{n-1})$$\end{document}$. The *Euclidean conductivity matrix* for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text {R}}$$\end{document}$ is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma _{ij}^{\text {E}, \text {R}} :=\frac{1}{|{\vec {\ell }}_1\wedge {\vec {\ell }}_2|} \lim _{p_{0}\rightarrow 0^{+}}\frac{1}{p_{0}} \Big (- {\widehat{K}}^{\text {R}}_{i,j}\big ((-p_0,{\vec {0}})\big ) +i \pmb {\langle } [ {\mathcal {J}}_{i}, {\mathcal {X}}_{j} ] \pmb {\rangle }_{\infty }^{\text {R}}\Big )\;, \end{aligned}$$\end{document}$$where, in the second term, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pmb {\langle }\cdot \pmb {\rangle }_{\infty }^R:=\lim _{\beta \rightarrow \infty }\lim _{L\rightarrow \infty }\frac{1}{L^{2}} \langle \cdot \rangle _{\beta , L}^{\text {R}}$$\end{document}$, and the expression $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[ {\mathcal {J}}_{j}, {\mathcal {X}}_{i} ]$$\end{document}$ must be understood as explained in the footnote 1 above. This definition can be obtained via a formal 'Wick rotation' of the time variable, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow -it$$\end{document}$, starting from the original definition of the Kubo conductivity, ([2.15](#Equ15){ref-type=""}), see, e.g., \[[@CR17]\]. A posteriori, we will see that in our context the two definitions coincide, see Sect. [5](#Sec13){ref-type="sec"} below. The structure correlation functions, and hence the conductivity, is severely constrained by *lattice Ward identities*. These are nonperturbative implications of lattice continuity equation, which we rewrite here in imaginary time (cf. with ([2.11](#Equ11){ref-type=""})):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} i\partial _{x_{0}} j_{0,\mathbf{x}} + \text {div}_{{\vec {x}}}{\vec {\jmath }}_{\mathbf{x}}=0\;, \end{aligned}$$\end{document}$$where we used the notation $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {div}_{{\vec {x}}}{\vec {\jmath }}_{\mathbf{x}}:=\sum _{i=1,2}\text {d}_{i}\tilde{\text {J}}_{i,\mathbf{x}}$$\end{document}$. For instance, consider the current--current correlation function[3](#Fn3){ref-type="fn"},$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \langle \mathbf{T}\, j_{0, \mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L} = \theta (x_{0} - y_{0}) \langle j_{0, \mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L} + \theta (y_{0} - x_{0}) \langle j_{\nu , \mathbf{y}}\,; j_{0, \mathbf{x}} \rangle ^{\text {R}}_{\beta ,L}\;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta (t)$$\end{document}$ is the Heaviside step function and the correlations in the right side are the time-unordered ones (i.e., they are defined without the action of the time-ordering operator). Using the continuity equation Eq. ([3.5](#Equ29){ref-type=""}):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} i\partial _{x_{0}}\langle \mathbf{T}\, j_{0, \mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L}= & {} \langle \mathbf{T}\, i\partial _{x_{0}}j_{0, \mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L} + i\langle [ j_{0, {\vec {x}}}\, , j_{\nu , {\vec {y}}} ] \rangle ^{\text {R}}_{\beta ,L} \delta (x_{0} - y_{0})\nonumber \\= & {} -\langle \mathbf{T}\, \text {div}_{{\vec {x}}} {\vec {\jmath }}_{\mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L} + i\langle [ j_{0, {\vec {x}}}\, , j_{\nu , {\vec {y}}} ] \rangle ^{\text {R}}_{\beta ,L} \delta (x_{0} - y_{0})\;. \quad \end{aligned}$$\end{document}$$Let us now take the Fourier transform of both sides: integrating by parts w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x_0$$\end{document}$ and using ([3.7](#Equ31){ref-type=""}), we find$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_{0} {\widehat{K}}^{\beta , L; \text {R}}_{0,\nu }(\mathbf{p})= & {} -\frac{1}{\beta L^2} \int _{0}^{\beta } dx_{0} \int _{0}^{\beta } dy_{0}\, \sum _{{\vec {x}}, {\vec {y}}\in \Lambda _L} e^{-ip_{0}(x_{0} - y_{0})}e^{-i{\vec {p}} \cdot ({\vec {x}}-{\vec {y}})} i\partial _{x_{0}}\langle \mathbf{T}\, j_{0, \mathbf{x}}\,; j_{\nu , \mathbf{y}} \rangle ^{\text {R}}_{\beta ,L}\nonumber \\= & {} \sum _{i=1,2} (1 - e^{-i{\vec {p}}\cdot {\vec {\ell }}_{i}}) \frac{1}{\beta L^{2}} \langle \mathbf{T}\, \hat{{\vec {\jmath }}}_{\mathbf{p}}\cdot \frac{{{\vec {G}}_i}}{2\pi }\,; {{\hat{\jmath }}}_{\nu , -\mathbf{p}} \rangle ^{\text {R}}_{\beta ,L} - i\sum _{{\vec {x}}} e^{-i{\vec {p}}\cdot {\vec {x}}} \langle [ j_{0, {\vec {x}}}\, , j_{\nu , {\vec {0}}} ] \rangle ^{\text {R}}_{\beta ,L}\nonumber \\\equiv & {} \sum _{i,i'=1,2} (1 - e^{-i{\vec {p}}\cdot {\vec {\ell }}_i})\frac{({\vec {G}}_i)_{i'}}{2\pi } {\widehat{K}}^{\beta , L; \text {R}}_{i',\nu }(\mathbf{p}) + {{\widehat{S}}}^{\beta , L; \text {R}}_{\nu }(\mathbf{p})\;, \end{aligned}$$\end{document}$$where we used that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\text {J}}_{i,\mathbf{x}}={\vec {\jmath }}_{\mathbf{x}}\cdot \frac{{\vec {G}}_i}{2\pi }$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {G}}_i$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i=1,2$$\end{document}$, the vectors of the dual basis, see definition in Sect. [2.1](#Sec3){ref-type="sec"}. More generally, denoting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(0,\nu _2,\ldots ,\nu _{n})$$\end{document}$ by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(0,{{\underline{\nu }}})$$\end{document}$, one has:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&p_{1,0} {\widehat{K}}^{\beta , L; \text {R}}_{0, {\underline{\nu }}}(\{\mathbf{p}_{i}\}_{i=1}^{n-1}) = \sum _{i,i'=1,2} (1 - e^{-i{\vec {p}}_{1}\cdot {\vec {\ell }}_i}) \frac{({\vec {G}}_i)_{i'}}{2\pi } {\widehat{K}}^{\beta , L; \text {R}}_{i', {\underline{\nu }}}(\{\mathbf{p}_{i}\}_{i=1}^{n-1}) + {{\widehat{S}}}^{\beta , L; \text {R}}_{{{\underline{\nu }}}}(\{\mathbf{p}_{i}\}_{i=1}^{n-1})\;,\qquad \nonumber \\&{{\widehat{S}}}_{{\underline{\nu }}}^{\beta , L; \text {R}}(\cdots ) := -\frac{i}{\beta L^2}\sum _{j=2}^{n} \langle \mathbf{T}\,C_{\nu _{j}}(\mathbf{p}_{1}, \mathbf{p}_{j})\,; {\hat{\jmath }}_{\nu _{2},\mathbf{p}_{2}}\,;\ldots \,; {\hat{\jmath }}_{\nu _{j-1},\mathbf{p}_{j-1}}\,; {\hat{\jmath }}_{\nu _{j+1},\mathbf{p}_{j+1}}\,; \cdots \,; {{\hat{\jmath }}}_{\nu _{n},\mathbf{p}_{n}} \rangle _{\beta , L}^{\text {R}},\nonumber \\ \end{aligned}$$\end{document}$$with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{\nu }(\mathbf{p}_{1}, \mathbf{p}_{2}) = \int _{0}^{\beta } dt\, e^{-it (\omega _{1} + \omega _{2})} [ {{\hat{\jmath }}}_{0,(t, {\vec {p}}_{1})}\,, {\hat{\jmath }}_{\nu , (t, {\vec {p}}_{2})}]$$\end{document}$ (here, with some abuse of notation, we let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\jmath }}}_{\mu ,(t,{\vec {p}})}$$\end{document}$ be the imaginary-time evolution at time *t* of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\jmath }}}_{\mu ,{\vec {p}}}$$\end{document}$), and with the understanding that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}_{n} = -\mathbf{p}_{1} - \ldots - \mathbf{p}_{n-1}$$\end{document}$. Even more generally, the identity remains valid if some of the current operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{\nu _i,\mathbf{p}_i}$$\end{document}$ are replaced by other local operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{{\mathcal {O}}}_{i,\mathbf{p}_i}$$\end{document}$: in this case, of course, the operators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{\nu _i}$$\end{document}$ must be modified accordingly. In the following, we will be interested in replacing one of the current operators either by the staggered density$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\jmath }}}_{3,\mathbf{p}}:=n_{\mathbf{p},A}-n_{\mathbf{p},B}\,, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{\mathbf{p},\rho }$$\end{document}$ is the Fourier transform of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{(t,{\vec {x}}),\rho }:=\sum _\sigma \psi ^+_{(t,{\vec {x}}),\rho ,\sigma }\psi ^-_{(t,{\vec {x}}),\rho ,\sigma }$$\end{document}$, or by the quartic interaction potential$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{{\mathcal {V}}}}_{\mathbf{p}} := \int _{0}^{\beta } dx_{0}\, e^{-ip_{0} x_{0}} \sum _{{\vec {x}}} e^{-i{\vec {p}}\cdot {\vec {x}}} \sum _{{\vec {y}}, \rho , \rho '} v_{\rho ,\rho '}({\vec {x}} - {\vec {y}}) \big ( (n_{{\vec {x}}, \rho }-1) (n_{{\vec {y}}, \rho '}-1)\big )_{x_{0}}\;. \end{aligned}$$\end{document}$$As we shall see below, the combination of the identity ([3.9](#Equ33){ref-type=""}) together with the regularity of the correlation functions has remarkable implications on the structure of the correlations. ### Consequences of the Ward Identities for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^1$$\end{document}$ Correlations {#Sec9} Here we start by discussing the consequences of the Ward identities for continuously differentiable correlations. #### Lemma 3.1 {#FPar2} Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}_{\beta , L} \in \frac{2\pi }{\beta }{\mathbb {Z}} \times \frac{2\pi }{L} {\mathbb {Z}}^{2}$$\end{document}$, such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\beta , L\rightarrow \infty }\mathbf{p}_{\beta , L} = \mathbf{p}\in B_{\varepsilon }(\mathbf{0}) := \{\mathbf{q}\in {\mathbb {R}}^{2} \mid |\mathbf{q}|<\varepsilon \}$$\end{document}$, for some $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon >0$$\end{document}$. Suppose that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{K}}}^{\text {R}}_{\mu ,\nu }(\mathbf{p}) = \lim _{\beta ,L\rightarrow \infty } {{\widehat{K}}}^{\beta , L; \text {R}}_{\mu ,\nu }(\mathbf{p}_{\beta , L})$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{S}}}^{\text {R}}_{j}(\mathbf{p}) = \lim _{\beta ,L\rightarrow \infty } {{\widehat{S}}}^{\beta , L; \text {R}}_{j}(\mathbf{p}_{\beta , L})$$\end{document}$ exist and that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{K}}}^{\text {R}}_{\mu ,\nu }(\mathbf{p}), {\widehat{S}}^{\text {R}}_{j}(\mathbf{p}) \in C^{1}(B_{\varepsilon }(\mathbf{0}))$$\end{document}$. Then,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma ^{\text {E}, \text {R}}_{ij}= \frac{1}{|{\vec {\ell }}_1\wedge {\vec {\ell }}_2|} \frac{\partial }{\partial p_{0}} {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{0})\;. \end{aligned}$$\end{document}$$ #### Proof {#FPar3} Consider Eq. ([3.8](#Equ32){ref-type=""}) with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu = j$$\end{document}$, in the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta , L\rightarrow \infty $$\end{document}$ limit. We differentiate both sides w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{i}$$\end{document}$, and take the limit $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}\rightarrow \mathbf{0}$$\end{document}$, thus getting (recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\ell }}_i\cdot {\vec {G}}_j=2\pi \delta _{i,j}$$\end{document}$):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} 0 = i {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{0}) + \frac{\partial }{\partial p_{i}} {{\widehat{S}}}^{\text {R}}_{j}(\mathbf{0}). \end{aligned}$$\end{document}$$Now, recall the definition of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\beta ,L;\text {R}}_{j}(\mathbf{p})$$\end{document}$ from Eq. ([3.8](#Equ32){ref-type=""}): $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\beta ,L;\text {R}}_{j}(\mathbf{0})=- i\sum _{{\vec {x}}} e^{-i{\vec {p}}\cdot {\vec {x}}} \langle [n_{{\vec {x}}}\, , j_{j, {\vec {0}}} ] \rangle ^{\text {R}}_{\beta ,L}$$\end{document}$, where we also used that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j_{0,{\vec {x}}}=n_{{\vec {x}}}$$\end{document}$. Taking the limit $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ,L\rightarrow \infty $$\end{document}$ and the derivative with respect to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_i$$\end{document}$, we get $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{\partial }{\partial p_{i}} {{\widehat{S}}}^{\text {R}}_{j}(\mathbf{0})=-\pmb {\langle } [\mathcal X_{i},{\mathcal {J}}_j ] \pmb {\rangle }^{\text {R}}_{\infty }$$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pmb {\langle }\cdot \pmb {\rangle }_{\infty }^R$$\end{document}$ was defined in ([3.4](#Equ28){ref-type=""}), and the expression $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[ {\mathcal {J}}_{j}, {\mathcal {X}}_{i} ]$$\end{document}$ must be understood as explained in the footnote 1 above. In conclusion,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{0}) =i\pmb {\langle } [{\mathcal {J}}_j, {\mathcal {X}}_{i} ] \pmb {\rangle }^{\text {R}}_{\infty }\;, \end{aligned}$$\end{document}$$and, if we plug this identity in ([3.4](#Equ28){ref-type=""}), noting that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pmb {\langle } [{\mathcal {J}}_j, {\mathcal {X}}_{i} ] \pmb {\rangle }^{\text {R}}_{\infty }$$\end{document}$ is even under the exchange , we obtain the desired identity. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ ### Consequences of the Ward Identities for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{3}$$\end{document}$ Correlations {#Sec10} Next, we discuss some other implications of the Ward identities for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^3$$\end{document}$ three-point correlations of the current operator (twice) with either the staggered density $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{j}}}_{3,\mathbf{p}}$$\end{document}$ (see ([3.10](#Equ34){ref-type=""})), or the interaction potential (see ([3.11](#Equ35){ref-type=""})), defined as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{K}}^{\beta , L; \text {R}}_{\mu , \nu , 3}(\mathbf{p}, \mathbf{q}):= & {} \frac{1}{\beta L^2}\langle \mathbf{T}\,{{\hat{\jmath }}}_{\mu , \mathbf{p}}\,; {\hat{\jmath }}_{\nu ,\mathbf{q}}\,; {{\hat{\jmath }}}_{3,-\mathbf{p}-\mathbf{q}} {\rangle }_{\beta , L} \nonumber \\ {\widehat{K}}^{\beta , L; \text {R}}_{\mu , \nu , V}(\mathbf{p}, \mathbf{q}):= & {} \frac{1}{\beta L^2}\langle \mathbf{T}\,{{\hat{\jmath }}}_{\mu , \mathbf{p}}\,; {\hat{\jmath }}_{\nu ,\mathbf{q}}\,; \hat{{\mathcal {V}}}_{-\mathbf{p}-\mathbf{q}} {\rangle }_{\beta , L} \;. \end{aligned}$$\end{document}$$We also let$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{S}}^{\beta , L; \text {R}}_{j, 3}(\mathbf{p}, \mathbf{q}):= & {} -\frac{i}{\beta L^2}\langle C_{j}(\mathbf{p}, \mathbf{q})\,; {\hat{\jmath }}_{3,-\mathbf{p}-\mathbf{q}}\rangle _{\beta , L}^{\text {R}}\;, \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{S}}^{\beta , L; \text {R}}_{j, V}(\mathbf{p}, \mathbf{q}):= & {} -\frac{i}{\beta L^2}\langle C_{j}(\mathbf{p}, \mathbf{q})\,; \hat{{\mathcal {V}}}_{-\mathbf{p}-\mathbf{q}} {\rangle }_{\beta , L}^{\text {R}} \end{aligned}$$\end{document}$$be the new Schwinger terms (recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{C}}}_j$$\end{document}$ was defined right after ([3.9](#Equ33){ref-type=""})). As usual, we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{\mu , \nu , \sharp }$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j, \sharp }$$\end{document}$ the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta , L\rightarrow \infty $$\end{document}$ limits of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\beta , L; \text {R}}_{\mu , \nu , \sharp }(\cdots )$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\beta , L; \text {R}}_{j, \sharp }(\cdots )$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sharp \in \{3,V\}$$\end{document}$. #### Lemma 3.2 {#FPar4} Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sharp \in \{0,3,V\}$$\end{document}$. Suppose that the limiting functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{\mu , \nu , \sharp }(\mathbf{p},\mathbf{q})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j, \sharp }(\mathbf{p},\mathbf{q})$$\end{document}$ exist in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_{\varepsilon }(\mathbf{0})\times B_\varepsilon (\mathbf{0})$$\end{document}$, and that they are of class $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{3}$$\end{document}$ in this domain. Then:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{\partial }{\partial p_{0}} {\widehat{K}}^{\text {R}}_{i, j, \sharp }((p_{0}, {\vec {0}}), (-p_{0}, {\vec {0}})) = \frac{\partial }{\partial p_{0}}\Big [ p_{0}^2 \frac{\partial ^{2}}{\partial p_{i} \partial q_{j}} {\widehat{K}}^{\text {R}}_{0,0, \sharp }((p_{0}, {\vec {0}}), (-p_{0}, {\vec {0}}))\Big ]\;. \end{aligned}$$\end{document}$$In particular, the left side of Eq. ([3.18](#Equ42){ref-type=""}) vanishes as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{0}\rightarrow 0$$\end{document}$. #### Proof {#FPar5} Taking the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ,L\rightarrow \infty $$\end{document}$ limit of the Ward Identity ([3.9](#Equ33){ref-type=""}) with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\nu }}}=(0,\sharp )$$\end{document}$, we find$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_{0} {\widehat{K}}^{\text {R}}_{0,0, \sharp }(\mathbf{p}, \mathbf{q}) = \sum _{i,i'=1,2} (1 - e^{-i{\vec {p}}\cdot {\vec {\ell }}_{i}})\frac{({\vec {G}}_i)_{i'}}{2\pi } {\widehat{K}}^{\text {R}}_{i', 0, \sharp }(\mathbf{p}, \mathbf{q})\;. \end{aligned}$$\end{document}$$Similarly, choosing $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\nu }}}=(j,\sharp )$$\end{document}$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_{0} {\widehat{K}}^{\text {R}}_{0,j, \sharp }(\mathbf{p}, \mathbf{q}) = \sum _{i,i'=1,2} (1 - e^{-i{\vec {p}}\cdot {\vec {\ell }}_{i}})\frac{({\vec {G}}_i)_{i'}}{2\pi } {\widehat{K}}^{\text {R}}_{i', j, \sharp }(\mathbf{p}, \mathbf{q}) + {{\widehat{S}}}^{\text {R}}_{j, \sharp }(\mathbf{p},\mathbf{q})\;, \end{aligned}$$\end{document}$$and, exchanging the roles of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{q}$$\end{document}$, we also get$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} q_{0} {\widehat{K}}^{\text {R}}_{i,0, \sharp }(\mathbf{p}, \mathbf{q}) = \sum _{j,j'=1,2} (1 - e^{-i{\vec {q}}\cdot {\vec {\ell }}_{j}})\frac{({\vec {G}}_j)_{j'}}{2\pi } {\widehat{K}}^{\text {R}}_{i, j', \sharp }(\mathbf{p}, \mathbf{q}) + {{\widehat{S}}}^{\text {R}}_{i, \sharp }({\mathbf{q},\mathbf{p}})\;. \end{aligned}$$\end{document}$$Combining ([3.19](#Equ43){ref-type=""}) with ([3.21](#Equ45){ref-type=""}), we find$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} q_{0} p_{0} {\widehat{K}}^{\text {R}}_{0,0, \sharp }(\mathbf{p}, \mathbf{q})= & {} \sum _{i,i' = 1,2}\Big [ \sum _{j,j' = 1,2}(1 - e^{-i{\vec {p}}\cdot {\vec {\ell }}_{i}}) (1 - e^{-i{\vec {q}}\cdot {\vec {\ell }}_{j}}) \frac{({\vec {G}}_i)_{i'}}{2\pi }\frac{({\vec {G}}_j)_{j'}}{2\pi } {\widehat{K}}^{\text {R}}_{i', j', \sharp }(\mathbf{p}, \mathbf{q}) \nonumber \\&+ (1-e^{-i{\vec {p}}\cdot {\vec {\ell }}_{i}})\frac{({\vec {G}}_i)_{i'}}{2\pi }{{\widehat{S}}}^{\text {R}}_{i', \sharp }({\mathbf{q},\mathbf{p}})\ \ \Big ]\;. \end{aligned}$$\end{document}$$We now derive w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{i}, q_{j}$$\end{document}$, and then set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}= -\mathbf{q}= (p_{0}, {\vec {0}})$$\end{document}$, thus finding[4](#Fn4){ref-type="fn"}:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_{0}^{2} \frac{\partial ^{2}}{\partial p_{1,i} \partial p_{2,j}} {\widehat{K}}^{\text {R}}_{0,0, \sharp }\big ((p_0,{\vec {0}}),(-p_0,{\vec {0}})\big )= & {} {\widehat{K}}^{\text {R}}_{i, j, \sharp }\big ((p_{0}, {\vec {0}}), (-p_{0}, {\vec {0}})\big ) \nonumber \\&-i \frac{\partial }{\partial p_{1,j}}{\widehat{S}}^{\text {R}}_{i, \sharp }\big ((-p_0,{\vec {0}}),(p_0,{\vec {0}})\big )\;. \end{aligned}$$\end{document}$$Finally, notice that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\partial _{p_{1,j}} S^{\text {R}}_{i, \sharp }\big ((-p_0,{\vec {0}}),(p_0,{\vec {0}})\big )$$\end{document}$ is constant in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{0}$$\end{document}$ (recall the definition of Schwinger term, Eq. ([3.16](#Equ40){ref-type=""}), and of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{j}$$\end{document}$, Eq. ([3.9](#Equ33){ref-type=""})). Therefore, after differentiation in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{0}$$\end{document}$, the final claim follows. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ Universality of the Euclidean Conductivity Matrix {#Sec11} ------------------------------------------------- Here we prove the universality of the Euclidean conductivity matrix, defined in Eq. ([3.4](#Equ28){ref-type=""}). We restrict to the range of parameters ([2.20](#Equ20){ref-type=""}), as discussed at the beginning of Sect. [2.3.1](#Sec6){ref-type="sec"}. In terms of the renormalized parameters, we restate ([2.20](#Equ20){ref-type=""}) as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} 0<\phi \leqslant \frac{\pi }{2}\;,\qquad m_{\text {R},+}>|m_{\text {R},-}|\;, \end{aligned}$$\end{document}$$where$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} m_{R,+}:=m_{R,-}+6\sqrt{3}\,t_{2}\sin \phi \;. \end{aligned}$$\end{document}$$A key ingredient in the proof is the following regularity result for the correlation functions. ### Proposition 3.3 {#FPar6} There exists $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{0}>0$$\end{document}$ such that, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|<U_{0}$$\end{document}$ and for parameters $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\phi ,m_{\text {R},-})$$\end{document}$ in the range ([3.24](#Equ48){ref-type=""}), the following is true. There exist functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi (U, m_{\text {R}, -}, \phi )$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U, m_{\text {R}, -}, \phi )$$\end{document}$, analytic in *U* and vanishing at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U=0$$\end{document}$, such that the Euclidean correlation functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{K}}}^{\text {R}}_{\mu ,\nu }(\mathbf{p})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{\mu , \nu , \sharp }(\mathbf{p}, \mathbf{q})$$\end{document}$, as well as the Schwinger terms $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j}(\mathbf{p})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j, \sharp }(\mathbf{p},\mathbf{q})$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sharp \in \{0,3,V\}$$\end{document}$, are analytic in *U*; moreover, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$, they are $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{3}$$\end{document}$ in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}, \mathbf{q}\in B_{\varepsilon }(\mathbf{0})$$\end{document}$, uniformly in *U* and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$. The proof of this proposition is postponed to the next section. Its content, combined with the (consequences of the) Ward identities discussed above, immediately implies the universality of the Euclidean conductivity matrix. ### Lemma 3.4 {#FPar7} Under the same assumptions as Proposition [3.3](#FPar6){ref-type="sec"}, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{R,-}\ne 0$$\end{document}$, then$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma ^{\text {E},\text {R}}_{12} = \frac{1}{2\pi } \big [ \mathrm{sign}(m_{\text {R},+}) - \mathrm{sign}(m_{\text {R},-})\big ]\;. \end{aligned}$$\end{document}$$ ### Proof {#FPar8} (*Assuming the validity of Proposition* [3.3](#FPar6){ref-type="sec"}). Thanks to Proposition [3.3](#FPar6){ref-type="sec"}, we know that the correlation functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{K}}}^{\text {R}}_{\mu ,\nu }(\mathbf{p})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{\mu , \nu , \sharp }(\mathbf{p}, \mathbf{q})$$\end{document}$, and the Schwinger terms $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j}(\mathbf{p})$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{j, \sharp }(\mathbf{p},\mathbf{q})$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sharp \in \{0,3,V\}$$\end{document}$, are $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{2}$$\end{document}$ in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}, \mathbf{q}\in B_{\varepsilon }(\mathbf{0})$$\end{document}$, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|< U_0$$\end{document}$. Therefore, we can apply Lemma [3.1](#FPar2){ref-type="sec"} and Lemma [3.2](#FPar4){ref-type="sec"}. Using Lemma [3.1](#FPar2){ref-type="sec"}, we rewrite the Euclidean conductivity matrix as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sigma ^{\text {E}, \text {R}}_{ij} =\frac{1}{|{\vec {\ell }}_1\wedge {\vec {\ell }}_2|} \frac{\partial }{\partial p_{0}} {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{0})\;. \end{aligned}$$\end{document}$$Then, we rewrite $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{i,j}$$\end{document}$ in terms of the non-interacting current--current correlation associated with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}_0^{\text {R}}$$\end{document}$, via the following *interpolation formula*:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{p}) = {\widehat{K}}^{\text {R},0}_{i,j}(\mathbf{p}) + \int _{0}^{U} dU'\, \frac{d}{dU'} {\widehat{K}}^{\text {R},U'}_{i,j}(\mathbf{p})\;. \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R},U'}_{i,j}(\mathbf{p})$$\end{document}$ is the correlation associated with the ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ,L\rightarrow \infty $$\end{document}$ limit of the) Gibbs measure with Hamiltonian$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathcal {H}}^\text {R}_{U'}: ={\mathcal {H}}_0^{\text {R}} + U'{\mathcal {V}}+ \delta (U',m_{\text {R},-},\phi )\sum _{{\vec {x}}\in \Lambda _{L}}[n_{{\vec {x}}, A} - n_{{\vec {x}}, B}] + \xi (U',m_{\text {R},-},\phi )\sum _{{\vec {x}}\in \Lambda _L}n_{{\vec {x}}},\nonumber \\ \end{aligned}$$\end{document}$$cf. with Eq. ([2.22](#Equ22){ref-type=""}). Computing the derivative in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U'$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\widehat{K}}^{\text {R}}_{i,j}(\mathbf{p})= & {} {\widehat{K}}^{\text {R},0}_{i,j}(\mathbf{p})- \int _{0}^{U} dU'\, \Big [\frac{\partial \delta }{\partial U'}(U',m_{\text {R},-},\phi )\, {\widehat{K}}^{\text {R},U'}_{i,j,3}(\mathbf{p},-\mathbf{p})\nonumber \\&+ \frac{\partial \xi }{\partial {U'}}(U',m_{\text {R},-},\phi )\, {\widehat{K}}^{\text {R},U'}_{i,j,0}(\mathbf{p},-\mathbf{p}) + {{\widehat{K}}}^{\text {R}, U'}_{i,j,V}(\mathbf{p}, -\mathbf{p})\Big ]\;. \end{aligned}$$\end{document}$$We now take the derivative w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{0}$$\end{document}$ and take $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0\rightarrow 0$$\end{document}$. Using Lemma [3.2](#FPar4){ref-type="sec"}, we immediately get:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{\partial }{\partial p_{0}}{\widehat{K}}^{\text {R}}_{i,j}(\mathbf{0}) =\frac{\partial }{\partial p_{0}}{\widehat{K}}^{\text {R},0}_{i,j}(\mathbf{0})\;, \end{aligned}$$\end{document}$$that is, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma ^{\text {E},\text {R}}_{ij}= \sigma ^{\text {E}, \text {R}}_{ij}\Big |_{U=0}$$\end{document}$ (we recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma ^{\text {E}, \text {R}}_{ij}\Big |_{U=0}$$\end{document}$ is the non-interacting Euclidean conductivity associated with the quadratic Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal H_0^{\text {R}}$$\end{document}$ at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$, which is assumed to be different from zero). The final claim, Eq. ([3.26](#Equ50){ref-type=""}), follows from a direct computation of the non-interacting conductivity, cf. with \[[@CR22], Appendix B, Eq. (B.8)\]. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ Proof of Proposition [3.3](#FPar6){ref-type="sec"} {#Sec12} ================================================== The proof of Proposition [3.3](#FPar6){ref-type="sec"} is a rather standard application of RG methods for fermions (see, e.g., \[[@CR11], [@CR16], [@CR18], [@CR32]\] for reviews). A similar analysis for interacting graphene, which corresponds to the case $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{2} = W = 0$$\end{document}$, has been discussed in \[[@CR20], [@CR21]\], which we refer to for further details. See also \[[@CR19]\], where an application to the Haldane--Hubbard model was discussed. The RG construction of the ground-state correlation functions, uniformly in the gap, is ultimately made possible by the fact that the many-body interaction, in the critical, massless, case, is *irrelevant* in the RG sense. The only qualitative effect of the interaction, with respect to the non-interacting theory, is a finite renormalization of the gap, of the chemical potential, of the Fermi velocity and of the wave function renormalization. We recall once more that we restrict the discussion to the range of parameters ([3.24](#Equ48){ref-type=""}). Moreover, we assume that *W* is not too large, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W\leqslant M_0$$\end{document}$, for a pre-fixed constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_0$$\end{document}$, the case of large *W* being substantially simpler, and left to the reader (for large *W*, the system is massive and is in a trivial, non-topological, insulating phase, as it follows from the proof of \[[@CR22]\]). Finally, for simplicity, we set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_1=1$$\end{document}$, that is, we set the scale of the bandwidth equal to one. Proof {#FPar9} ----- The starting point is the well-known representation of the Euclidean correlation in terms of Grassmann integrals (see, for instance, \[[@CR20], [@CR22]\]). The generating functional of the correlations is denoted by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {W}}(f,A)$$\end{document}$, with *f* an external Grassmann field coupled to the fermionic fields, and *A* a (five-component) external complex field conjugated to the lattice currents and the quartic interaction. We have:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} e^{{{\mathcal {W}}}(f,A)}=\frac{\int P(d\Psi )e^{-V(\Psi )+(\Psi ,f)+(J,A)}}{\int P(d\Psi )e^{-V(\Psi )}}, \end{aligned}$$\end{document}$$where: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^\pm _{\mathbf{x},s}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{x}=(x_0,{\vec {x}})\in \mathbb [0,\beta )\times \Lambda _L$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s\in \{\uparrow ,\downarrow \}$$\end{document}$, is a two-component Grassmann spinor, whose components will be denoted by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^\pm _{\mathbf{x},\rho ,s}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho =A,B$$\end{document}$; $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(d\Psi )$$\end{document}$ is the fermionic Gaussian integration with propagator$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} g_{s,s'}(\mathbf{x},\mathbf{y})=\frac{\delta _{s,s'}}{\beta L^2}\sum _{k_0\in \frac{2\pi }{\beta }({\mathbb {Z}}+\frac{1}{2})}\ \sum _{{\vec {k}}\in \frac{2\pi }{L}{\mathbb {Z}}_L^2}e^{-i\mathbf{k}(\mathbf{x}-\mathbf{y})} {{\hat{g}}}(\mathbf{k}), \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb {Z}}_L={\mathbb {Z}}/L{\mathbb {Z}}$$\end{document}$ and, letting$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R({\vec {k}})= -2t_2\cos \phi \big (\alpha _1({\vec {k}})-\alpha _1({\vec {k}}_F^{\pm })\big ),\qquad m_{R}({\vec {k}})=m_{R,-}+2t_2(\alpha _2(k)-\alpha _2(k_F^-))\sin \phi , \end{aligned}$$\end{document}$$and recalling that we set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_1=1$$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{g}}}(\mathbf{k})=\begin{pmatrix} &{}-ik_0+R({\vec {k}})+ m_R({\vec {k}}) &{} - \Omega ^*({\vec {k}})\\ &{}- \Omega ({\vec {k}}) &{} -i k_0+R({\vec {k}})- m_R({\vec {k}}) \end{pmatrix}^{\!\!\!-1}, \end{aligned}$$\end{document}$$with the understanding that, at contact, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g(\mathbf{x},\mathbf{x})$$\end{document}$ should be interpreted as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\varepsilon \rightarrow 0^+}[g(\mathbf{x}+(\varepsilon ,{\vec {0}}),\mathbf{x}) +g(\mathbf{x}-(\varepsilon ,{\vec {0}}),\mathbf{x})]$$\end{document}$;$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} V(\Psi )= & {} \int _{0}^\beta dx_0\sum _{{\vec {x}}\in \Lambda _L} \Big [U \sum _{{\vec {y}}\in \Lambda _L}\sum _{\rho , \rho '=A,B} n_{\mathbf{x}, \rho } v_{\rho ,\rho '}({\vec {x}} - {\vec {y}}) n_{(x_0,{\vec {y}}), \rho '}\nonumber \\&+\delta (U,m_{\text {R},-},\phi )(n_{\mathbf{x},A}-n_{\mathbf{x},B}) +\xi (U,m_{\text {R},-},\phi ) n_{\mathbf{x}} \Big ], \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{\mathbf{x},\rho }=\sum _{s=\uparrow ,\downarrow }\Psi ^+_{\mathbf{x},\rho ,s}\Psi ^-_{\mathbf{x},\rho ,s}$$\end{document}$ is the Grassmann counterpart of the density operator, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{\mathbf{x}}=\sum _{\rho =A,B}n_{\mathbf{x},\rho }$$\end{document}$; finally,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&(\Psi ,f)=\int _0^\beta dx_0\sum _{{\vec {x}}\in \Lambda _L}\sum _{s=\uparrow \downarrow }(\Psi ^+_{\mathbf{x},s}f^-_{\mathbf{x},s}+f^+_{\mathbf{x},s}\Psi ^-_{\mathbf{x},s}), \\&(J,A)=\frac{1}{\beta L^2}\sum _{p_0\in \frac{2\pi }{\beta }{\mathbb {Z}}}\ \sum _{{\vec {p}}\in \frac{2\pi }{L}{\mathbb {Z}}^2}\ \sum _{\mu =0}^4{{\hat{A}}}_{\mathbf{p},\mu }{{\hat{J}}}_{\mathbf{p},\mu }, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{J}}}_{\mathbf{p},\mu }=\int _0^\beta dx_0 \sum _{{\vec {x}}\in \Lambda _L}e^{-i\mathbf{p}\cdot \mathbf{x}} J_{\mathbf{x},\mu }$$\end{document}$ and: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{\mathbf{x},0}=n_\mathbf{x}$$\end{document}$ is the Grassmann counterpart of the density; $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{\mathbf{x},1},J_{\mathbf{x},2}$$\end{document}$ are the Grassmann counterparts of the two components of the lattice current,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} J_{\mathbf{x},1}=\frac{3}{2}({{\tilde{J}}}_{\mathbf{x},1}+{{\tilde{J}}}_{\mathbf{x},2}),\qquad J_{\mathbf{x},2}=\frac{\sqrt{3}}{2}(-{{\tilde{J}}}_{\mathbf{x},1}+\tilde{J}_{\mathbf{x},2}), \end{aligned}$$\end{document}$$with$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{J}}}_{\mathbf{x},1}= -J_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{1}} - J_{{\vec {x}}, {\vec {x}}+ {\vec {\ell }}_{1} - {\vec {\ell }}_{2}},\qquad {{\tilde{J}}}_{\mathbf{x},2} = -J_{{\vec {x}}, {\vec {x}} + {\vec {\ell }}_{2}} - J_{{\vec {x}}, {\vec {x}} - {\vec {\ell }}_{1} + {\vec {\ell }}_{2}}, \end{aligned}$$\end{document}$$and$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} J_{{\vec {x}},{\vec {y}}} = \sum _{s=\uparrow ,\downarrow } \big [i\Psi ^{+}_{{\vec {y}}, s} H({\vec {y}}-{\vec {x}}) \Psi ^{-}_{{\vec {x}},s} -i \Psi ^{+}_{{\vec {x}}, s} H({\vec {x}}-{\vec {y}}) \Psi ^{-}_{{\vec {y}},s}\big ]\;; \end{aligned}$$\end{document}$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{\mathbf{x},3}=n_{\mathbf{x},A}-n_{\mathbf{x},B}$$\end{document}$ is the Grassmann counterpart of the staggered density; $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{\mathbf{x},4}$$\end{document}$ is the Grassmann counterpart of the quartic interaction,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} J_{\mathbf{x},4}= \sum _{{\vec {y}}, \rho , \rho '} n_{\mathbf{x}, \rho } v_{\rho ,\rho '}({\vec {x}} - {\vec {y}})n_{(x_0,{\vec {y}}), \rho '}\;. \end{aligned}$$\end{document}$$The derivatives of the generating functional computed at zero external fields equal the Euclidean correlation functions, cf. with, e.g., \[[@CR19], Eq. (27), (28)\]. Needless to say, the Euclidean correlations satisfy non trivial Ward Identities, following from the lattice continuity equation. For an example, cf. with \[[@CR19], Eq. (19),(20)\]. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ In order to compute the generating functional $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {W}}(f,A)$$\end{document}$ in Eq. ([4.1](#Equ56){ref-type=""}), we use an expansion in *U*, which is convergent uniformly in the volume and temperature, and uniformly close to (and even on) the critical lines $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R}, \pm } = 0$$\end{document}$. Note that, in the parameter range ([3.24](#Equ48){ref-type=""}) the propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{g}}}(\mathbf{k})$$\end{document}$ is singular only when $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}=0$$\end{document}$, in which case the singularity is located at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}_{F}^{-} := (0, {\vec {k}}_{F}^{-})$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {k}}_{F}^{\pm } = \big ( \frac{2\pi }{3}, \pm \frac{2\pi }{3\sqrt{3}} \big )$$\end{document}$. Due to this singularity, the Grassmann integral has, a priori, an infrared problem, which we resolve by a multi-scale re-summation of the corresponding singularities. The multi-scale computation of the generating function proceeds as follows. First of all, we distinguish the ultraviolet modes, corresponding to large values of the Matsubara frequency, from the infrared ones, by introducing two compactly supported cut-off functions, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _\pm (\mathbf{k})$$\end{document}$, supported in the vicinity of the Fermi points $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}_F^\pm =(0,{\vec {k}}_F^\pm )$$\end{document}$; more precisely, we let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _\pm (\mathbf{k})=\chi _0(\mathbf{k}- \mathbf{k}_F^\pm )$$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _0$$\end{document}$ is a smooth characteristic function of the ball of radius $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0$$\end{document}$ equal to, say, 1 / 3) and by letting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _{\text {uv}}(\mathbf{k})=1-\sum _{\omega =\pm }\chi _\omega (\mathbf{k})$$\end{document}$. We correspondingly split the propagator in its ultraviolet and infrared components:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} g(\mathbf{x},\mathbf{y})=g^{(1)}(\mathbf{x},\mathbf{y})+\sum _{\omega =\pm } e^{-i{\vec {k}}_F^\omega ({\vec {x}}-{\vec {y}})}g_{\omega }^{(\leqslant 0)}(\mathbf{x},\mathbf{y}) \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(1)}(\mathbf{x},\mathbf{y})$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g_{\omega }^{(\leqslant 0)}(\mathbf{x},\mathbf{y})$$\end{document}$ are defined in a way similar to Eq.([4.2](#Equ57){ref-type=""}), with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{g}}}(\mathbf{k})$$\end{document}$ replaced by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _{\text {uv}}(\mathbf{k}) {{\hat{g}}}(\mathbf{k})$$\end{document}$ and by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _{0}(\mathbf{k}) {{\hat{g}}}(\mathbf{k}+\mathbf{k}_F^\omega )$$\end{document}$, respectively. We then split the Grassmann field as a sum of two independent fields, with propagators $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(1)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(\leqslant 0)}$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi _{\mathbf{x},s}^\pm =\Psi ^{\pm (1)}_{\mathbf{x},s}+\sum _{\omega =\pm }e^{\pm i{\vec {k}}_F^\omega {\vec {x}}} \Psi _{\mathbf{x},s,\omega }^{\pm (\leqslant 0)} \end{aligned}$$\end{document}$$and we rewrite the Grassmann Gaussian integration as the product of two independent Gaussians: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(d\Psi )=P(d\Psi ^{(\leqslant 0)})P(d\Psi ^{(1)})$$\end{document}$. By construction, the integration of the 'ultraviolet' field $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(1)}$$\end{document}$ does not have any infrared singularity and, therefore, can be performed in a straightforward manner, thus allowing us to rewrite the generating function $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {W}}(f,A)$$\end{document}$ as the logarithm of$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{e^{{{\mathcal {W}}}^{(0)}(f,A)}}{{\mathcal {N}}_0}\int P(d\Psi ^{(\leqslant 0)})e^{-V^{(0)}(\Psi ^{(\leqslant 0)})+B^{(0)}(\Psi ^{(\leqslant 0)}, f, A)}, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(0)}$$\end{document}$ are, respectively, the effective potential and the effective source: they are defined by the conditions that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}(0)=0$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(0)}(0,f,A)=B^{(0)}(\Psi ,0,0)=0$$\end{document}$. The normalization constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {N}}_0$$\end{document}$ is fixed in such a way that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {N}}_0=\int P(d\Psi ^{(\leqslant 0)})e^{-V^{(0)}(\Psi ^{(\leqslant 0)})}$$\end{document}$. All $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal W^{(0)}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(0)}$$\end{document}$ are expressed as series of monomials in the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ,f,A$$\end{document}$ fields, whose kernels (given by the sum of all possible Feynman diagrams with fixed number and fixed space-time location of the external legs) are *analytic functions* of the interaction strength, for *U* sufficiently small. The precise statement and the proof of these claims are essentially identical to those of \[[@CR20], Lemma 2\], see also \[[@CR22], Lemma 5.2\] or \[[@CR17], Section 6\]; details will not belabored here and are left to the reader. In order to integrate the infrared scales, one has to exploit certain lattice symmetries of the model (which replace those of \[[@CR20], Lemma 1\]), which allow us to reduce the number of independent *relevant* and *marginal* terms generated by the multi-scale integration. In particular, the symmetries under which the effective potential $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}(\Psi )$$\end{document}$ is invariant are the following \[[@CR19], Sect. III.B\]. \(1\) Discrete rotation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}_{\mathbf{k}',s,\omega }^-\rightarrow e^{i\omega \frac{2\pi }{3}n_-}e^{-i{\vec {k}}'\cdot {\vec {\ell }}_2\,n_-}{{\hat{\Psi }}}_{T\mathbf{k}',s,\omega }^-\;,\quad {{\hat{\Psi }}}_{\mathbf{k}',s,\omega }^+\rightarrow {{\hat{\Psi }}}_{T\mathbf{k}',s,\omega }^+e^{i{\vec {k}}'\cdot {\vec {\ell }}_2\,n_-}e^{-i\omega \frac{2\pi }{3}n_-} \end{aligned}$$\end{document}$$where, denoting the Pauli matrices by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _1,\sigma _2,\sigma _3$$\end{document}$, we defined$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} n_-=(1-\sigma _3)/2\;,\qquad T\mathbf{k}'=(k_0',e^{-i\frac{2\pi }{3}\sigma _2}{\vec {k}}')\;; \end{aligned}$$\end{document}$$that is, *T* is the spatial rotation by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\pi /3$$\end{document}$ in the counter-clockwise direction. \(2\) Complex conjugation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}^{\pm }_{\mathbf{k}',s,\omega }\rightarrow {{\hat{\Psi }}}^{\pm }_{-\mathbf{k}',s,-\omega },\quad c\rightarrow c^{*}\;,\quad \phi \rightarrow -\phi \;, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c\in {\mathbb {C}}$$\end{document}$ is a generic constant appearing in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(d\Psi )$$\end{document}$ or in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V(\psi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c^*$$\end{document}$ is its complex conjugate. \(3\) Horizontal reflections:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}^{-}_{\mathbf{k}',s,\omega }\rightarrow \sigma _1{{\hat{\Psi }}}^-_{R_h\mathbf{k}',s,\omega }\;,\quad {{\hat{\Psi }}}^{+}_{\mathbf{k}',s,\omega }\rightarrow {{\hat{\Psi }}}^+_{R_h\mathbf{k}',s,\omega }\sigma _1 \,,\quad (W,\phi )\rightarrow (-W,-\phi ) \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_h\mathbf{k}'=(k_0',-k_1',k_2')$$\end{document}$. \(4\) Vertical reflections:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}^{\pm }_{\mathbf{k}',s,\omega }\rightarrow {{\hat{\Psi }}}^{\pm }_{R_v\mathbf{k}',s,-\omega }\;,\quad \phi \rightarrow -\phi . \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_v\mathbf{k}'=(k_0',k_1',-k_2')$$\end{document}$. \(5\) Particle-hole:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}^{-}_{\mathbf{k}',s,\omega }\rightarrow i{{\hat{\Psi }}}^{+,T}_{P\mathbf{k}',s,-\omega }\;,\quad {{\hat{\Psi }}}^{+}_{\mathbf{k}',s,\omega }\rightarrow i{{\hat{\Psi }}}^{-,T}_{P\mathbf{k}',s,-\omega }\;,\quad \phi \rightarrow -\phi \;. \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\mathbf{k}'=(k_0',-k_1',-k_2')$$\end{document}$. \(6\) Magnetic reflections:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{\Psi }}}^{-}_{\mathbf{k}',s,\omega }\rightarrow -i \sigma _1\sigma _3{{\hat{\Psi }}}^-_{-R_v\mathbf{k}',s,\omega }\;,\quad {{\hat{\Psi }}}^{+}_{\mathbf{k}',s,\omega }\rightarrow -i{{\hat{\Psi }}}^+_{-R_v\mathbf{k}',s,\omega }\sigma _3\sigma _1\;,\quad \phi \rightarrow \pi -\phi . \end{aligned}$$\end{document}$$These symmetries have nonperturbative consequences on the structure of the effective interaction action $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}$$\end{document}$. At fixed $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W,\phi $$\end{document}$, the theory is invariant under the transformations (1), (2)+(4), and (2)+(5). In particular, these transformations leave the quadratic part$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Q^{(0)}(\Psi )=\sum _{s,\omega }\int \frac{d\mathbf{k}'}{2\pi |\mathcal B|}\,{{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega }{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{k}'){\hat{\Psi }}^-_{\mathbf{k}',s,\omega } \end{aligned}$$\end{document}$$of the effective potential $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(0)}(\Psi )$$\end{document}$ invariant (in ([4.14](#Equ69){ref-type=""}), $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int \frac{d\mathbf{k}'}{2\pi |{\mathcal {B}}|}$$\end{document}$ is a shorthand for the Riemann sum $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\beta L^2)^{-1} \sum _{k_0\in \frac{2\pi }{\beta }{\mathbb {Z}}}\sum _{{\vec {k}}\in {\mathcal {B}}_L}$$\end{document}$). This means that:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{k}')= & {} e^{-i(\omega \frac{2\pi }{3}+{\vec {k}}'\cdot {\vec {\ell }}_1)n_-} {{\hat{W}}}_{2;\omega }^{(0)}(T^{-1}\mathbf{k}')e^{i(\omega \frac{2\pi }{3}+{\vec {k}}'\cdot {\vec {\ell }}_1)n_-}\nonumber \\= & {} \big [{{\hat{W}}}_{2;\omega }^{(0)}(-k_0',-k_1',k_2')\big ]^*=\big [{{\hat{W}}}_{2;\omega }^{(0)}(-k_0',k_1',k_2')\big ]^\dagger . \end{aligned}$$\end{document}$$The values of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{k}')$$\end{document}$ and of its derivatives at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'=\mathbf{0}$$\end{document}$ define the *effective coupling constants*. By computing Eq. ([4.15](#Equ70){ref-type=""}) at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'=\mathbf{0}$$\end{document}$, we find, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =\pm $$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})=e^{-i\frac{2\pi }{3}\omega n_-}{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})e^{i\frac{2\pi }{3}\omega n_-}= \big [{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})\big ]^{*}=\big [{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})\big ]^\dagger \;. \end{aligned}$$\end{document}$$This implies:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})=\xi _{\omega ,0}+\delta _{\omega ,0}\sigma _3, \end{aligned}$$\end{document}$$for two *real* constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,0}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{\omega ,0}$$\end{document}$. Let us now discuss the structure of the derivative of the kernel of the quadratic terms. By taking the derivative of Eq. ([4.15](#Equ70){ref-type=""}) w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'$$\end{document}$ and then setting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'=\mathbf{0}$$\end{document}$, we get:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \partial _{\mathbf{k}'}{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})= & {} e^{-i\frac{2\pi }{3}\omega n_-}T\partial _{\mathbf{k}'}{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0})e^{i\frac{2\pi }{3}\omega n_-}=(-R_v)\partial _{\mathbf{k}'} {{\hat{W}}}_{2;\omega }^{(0)*}(\mathbf{0})\nonumber \\= & {} (-P)\partial _{\mathbf{k}'}{{\hat{W}}}_{2;\omega }^{(0)\dagger }(\mathbf{0}), \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_v$$\end{document}$ (resp. *P*) is the diagonal matrix with diagonal elements $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1,1,-1)$$\end{document}$ (resp. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1,-1,-1)$$\end{document}$). [4.18](#Equ73){ref-type=""}) implies that:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathbf{k}'\partial _{\mathbf{k}'}{{\hat{W}}}_{2;\omega }^{(0)}(\mathbf{0}) = \begin{pmatrix} -i z_{1,\omega } k_{0}' &{} -u_{\omega }(-i k_{1}' +\omega k_{2}') \\ -u_{\omega }(i k_{1}' + \omega k_{2}') &{} -i z_{2,\omega } k_{0}' \end{pmatrix}, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u_\omega , z_{1,\omega },z_{2,\omega }$$\end{document}$ are *real* constants. The integration of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(\leqslant 0)}_\omega $$\end{document}$ is performed iteratively. One rewrites $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(\leqslant 0)}_{\omega } = \sum _{h\leqslant 0} \Psi ^{(h)}_{\omega }$$\end{document}$, for suitable single-scale fields $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(h)}_{\omega }$$\end{document}$. The covariance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{g}}}^{(h)}_{\omega }$$\end{document}$ of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(h)}_{\omega }$$\end{document}$, supported for quasi-momenta $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'$$\end{document}$ such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_{0}2^{h-1}\leqslant |\mathbf{k}'|\leqslant a_{0}2^{h+1}$$\end{document}$, will be defined inductively. We consider two different regimes. The first corresponds to scales $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\geqslant h^*_1$$\end{document}$, with$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} h^*_1:=\min \{0,\lfloor \log _2 m_{R,+}\rfloor \}, \end{aligned}$$\end{document}$$and the rest to scales $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1\geqslant h\geqslant h^*_2$$\end{document}$ with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_2:=\min \{0,\lfloor \log _2 |m_{R,-}|\rfloor \}$$\end{document}$ (recall that we are focusing on the case that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{R,+}>|m_{R,-}|$$\end{document}$.). We describe the iteration in an inductive way. Assume that the fields $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(0)}, \Psi ^{(-1)},\ldots ,\Psi ^{(h+1)}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\geqslant h^*_1$$\end{document}$, have been integrated out and that after their integration the generating function has the following structure, analogous to the one at scale 0:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} e^{ {{\mathcal {W}}}(f,A) }= \frac{e^{{{\mathcal {W}}}^{(h)}(f,A)}}{{\mathcal {N}}_h}\int P(d\Psi ^{(\leqslant h)})e^{-V^{(h)}(\Psi ^{(\leqslant h)})+B^{(h)}(\Psi ^{(\leqslant h)}, f, A)}, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(h)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(h)}$$\end{document}$ are, respectively, the effective potential and source terms, satisfying the conditions that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(h)}(0)=0$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(h)}(0,f,A)=B^{(h)}(\Psi ,0,0)=0$$\end{document}$. The normalization constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {N}}_h$$\end{document}$ is fixed in such a way that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {N}}_h=\int P(d\Psi ^{(\leqslant h)})e^{-V^{(h)}(\Psi ^{(\leqslant h)})}$$\end{document}$. Here, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(d\Psi ^{(\leqslant h)})$$\end{document}$ is the Grassmann Gaussian integration with propagator (diagonal in the *s* and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega $$\end{document}$ indices)$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} g^{(\leqslant h)}_\omega (\mathbf{x},\mathbf{y})=\int P(d\Psi ^{(\leqslant h)})\Psi ^{-(\leqslant h)}_{\mathbf{x},s,\omega }\Psi ^{+(\leqslant h)}_{\mathbf{y},s,\omega } =\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, e^{-i\mathbf{k}'(\mathbf{x}-\mathbf{y})} {{\hat{g}}}_\omega ^{(\leqslant h)}(\mathbf{k}'), \end{aligned}$$\end{document}$$where, letting$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&r_\omega ({\vec {k}}')= R({\vec {k}}'+{\vec {k}}_F^\omega ),\quad s_\omega ({\vec {k}}')=-[\Omega ({\vec {k}}'+{\vec {k}}_F^{\,\omega })-\frac{3}{2}(ik_1'+\omega k_2')], \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&m_{-}({\vec {k}}')=m_{\text {R},-}+2t_2\big (\alpha _2({\vec {k}}'+{\vec {k}}_F^-)-\alpha _2({\vec {k}}_F^-)\big )\sin \phi , \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&m_{+}({\vec {k}}')=m_{\text {R},-}+6\sqrt{3}t_2\sin \phi +2t_2\big (\alpha _2({\vec {k}}'+{\vec {k}}_F^+)-\alpha _2({\vec {k}}_F^+)\big )\sin \phi , \end{aligned}$$\end{document}$$and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _h(\mathbf{k}')=\chi _0(2^{-h}\mathbf{k}')$$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\hat{g}}}_\omega ^{(\leqslant h)}(\mathbf{k}')=\chi _h(\mathbf{k}') \begin{pmatrix} a_{1,\omega ,h}(\mathbf{k}') &{} b^*_{\omega ,h}(\mathbf{k}')\\ b_{\omega ,h}(\mathbf{k}') &{} a_{2,\omega ,h}(\mathbf{k}')\end{pmatrix}^{\!\!\!-1}, \end{aligned}$$\end{document}$$with$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&a_{\rho ,\omega ,h}(\mathbf{k})=-ik_0Z_{\rho ,\omega ,h}+r_\omega ({\vec {k}}')+(-1)^{\rho -1} m_{\omega }({\vec {k}}'),\nonumber \\&b_{\omega ,h}(\mathbf{k}')=-v_{\omega ,h} (ik_1'+\omega k_2')+s_\omega ({\vec {k}}')\;, \end{aligned}$$\end{document}$$and the understanding that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(-1)^{\rho -1}$$\end{document}$ is equal to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+1$$\end{document}$, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho =A$$\end{document}$, and equal to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1$$\end{document}$, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho =B$$\end{document}$. The quantities $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,h}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{\omega ,h}$$\end{document}$ are *real*, and they have, respectively, the meaning of wave function renormalizations and of effective velocities. Note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_\omega ({\vec {k}}')$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_\omega ({\vec {k}}')$$\end{document}$ are both of order $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O(|{\vec {k}}'|^2)$$\end{document}$, while the mass satisfies (again, recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},+}=m_{\text {R},-}+6\sqrt{3} t_2\sin \phi $$\end{document}$):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} m_{\omega }({\vec {k}}')=m_{\text {R},\omega }+t_\omega ({\vec {k}}'), \quad \text {with}\quad t_\omega ({\vec {k}}')=O(|{\vec {k}}'|^2). \end{aligned}$$\end{document}$$By definition, the representation above is valid at the initial step, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=0$$\end{document}$. In order to inductively prove its validity at the generic step, let us discuss how to pass from scale *h* to scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h-1$$\end{document}$, that is, how to integrate out the field $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(h)}$$\end{document}$, and how to re-express the resulting effective theory in the form ([4.21](#Equ76){ref-type=""}), with *h* replaced by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h-1$$\end{document}$. Before integrating the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(h)}$$\end{document}$ field out, we split $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(h)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(h)}$$\end{document}$ into their *local* and *irrelevant* parts (here, for simplicity, we spell out the definitions only in the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f=0$$\end{document}$ case): $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(h)}={{\mathcal {L}}}V^{(h)}+{{\mathcal {R}}}V^{(h)}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(h)}={{\mathcal {L}}}B^{(h)}+{{\mathcal {R}}}B^{(h)}$$\end{document}$, where, denoting the quadratic part of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{(h)}$$\end{document}$ by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}Q^{(h)}(\Psi ) = \sum _{\omega ,s}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, {\hat{\Psi }}^+_{\mathbf{k}',s,\omega } {{\hat{W}}}^{(h)}_{2;\omega }(\mathbf{k}') {\hat{\Psi }}^-_{\mathbf{k}',s,\omega },\end{aligned}$$\end{document}$$and the part of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^{(h)}$$\end{document}$ of order (2, 0, 1) in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\psi ,f,A)$$\end{document}$ by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Q^{(h)}(\Psi , A) = \sum _{\omega ,s,\mu }\int \frac{d\mathbf{p}}{(2\pi )^3}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, {{\hat{A}}}_{\mathbf{p},\mu }{\hat{\Psi }}^+_{\mathbf{k}'+\mathbf{p},s,\omega } {{\hat{W}}}^{(h)}_{2,1;\mu ,\omega }(\mathbf{k}',\mathbf{p}){{\hat{\Psi }}}^-_{\mathbf{k}',s,\omega } \end{aligned}$$\end{document}$$we let:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\mathcal {L}}}V^{(h)}(\Psi )=\sum _{\omega ,s}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, {{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega } [{{\hat{W}}}^{(h)}_{2;\omega }(\mathbf{0})+\mathbf{k}'\partial _{\mathbf{k}'}{{\hat{W}}}^{(h)}_{2;\omega }(\mathbf{0})\big ]{\hat{\Psi }}^-_{\mathbf{k}',s,\omega }, \end{aligned}$$\end{document}$$and$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\mathcal {L}}}B^{(h)}(\Psi ,0,A)=\sum _{\omega , s, \mu }\int \frac{d\mathbf{p}}{(2\pi )^3}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, {{\hat{A}}}_{\mathbf{p},\mu }{{\hat{\Psi }}}^+_{\mathbf{k}'+\mathbf{p},s,\omega } {{\hat{W}}}^{(h)}_{2,1;\mu ,\omega }(\mathbf{0},\mathbf{0}){\hat{\Psi }}^-_{\mathbf{k}',s,\omega }. \end{aligned}$$\end{document}$$By the symmetries of the model,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\mathcal {L}}}V^{(h)}(\Psi )= & {} \sum _{\omega , s}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}\, \Big [2^h\xi _{\omega ,h}{\hat{\Psi }}^+_{\mathbf{k}',s,\omega } {{\hat{\Psi }}}^-_{\mathbf{k}',s,\omega }+ 2^h\delta _{\omega ,h}{{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega } \sigma _3 {\hat{\Psi }}^-_{\mathbf{k}',s,\omega } \nonumber \\&+{{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega } \begin{pmatrix} -i z_{1,\omega ,h} k_{0}&{} -u_{\omega ,h}(-i k_{1}' +\omega k_{2}') \\ -u_{\omega ,h}(i k_{1}' + \omega k_{2}') &{} -i z_{2,\omega ,h} k_{0} \end{pmatrix} {\hat{\Psi }}^-_{\mathbf{k}',s,\omega }\Big ], \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,h}, \delta _{\omega ,h},z_{\rho ,\omega ,h}, u_{\omega ,h}$$\end{document}$ are real constants and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _3$$\end{document}$ is the third Pauli matrix. We also denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _{\mu ,\omega ,h}:={{\hat{W}}}^{(h)}_{2,1;\mu ,\omega }(\mathbf{0},\mathbf{0})$$\end{document}$ the *vertex functions*, entering the definition of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal {L}}}B^{(h)}(\Psi ,0,A)$$\end{document}$. Notice that their structure is constrained by the Ward Identities. E.g., using \[[@CR19], Eq. (20)\], one finds that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _{0,\omega ,h}=-\sum _{\rho =1}^2(Z_{\rho ,\omega ,h}+z_{\rho ,\omega ,h})n_\rho $$\end{document}$ (where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_\rho =(1+(-1)^{\rho -1}\sigma _3)/2$$\end{document}$), $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _{1,\omega ,h}=-(v_{\omega ,h}+u_{\omega ,h})\sigma _2$$\end{document}$, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _{2,\omega ,h}=-\omega (v_{\omega ,h}+u_{\omega ,h})\sigma _1$$\end{document}$. However, in the following, we will neither need these identities, nor to identify any special structure of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _{\mu ,\omega ,h}$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu =3,4$$\end{document}$. Once the effective potential and source have been split into local and irrelevant parts, we combine the part of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal {L}}}V^{(h)}$$\end{document}$ in the second line of ([4.27](#Equ82){ref-type=""}) with the Gaussian integration $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(d\Psi ^{(\leqslant h)})$$\end{document}$, thus defining a dressed measure $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{P}(d\Psi ^{(\leqslant h)})$$\end{document}$ whose propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{g}}}^{(\leqslant h)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$ is analogous to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(\leqslant h)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$, with the only difference that the functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_{\rho ,\omega ,h}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{\omega ,h}$$\end{document}$ in ([4.25](#Equ80){ref-type=""})-([4.26](#Equ81){ref-type=""}) are replaced by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{a}}}_{\rho ,\omega ,h-1}(\mathbf{k})= & {} -ik_0\tilde{Z}_{\rho ,\omega ,h-1}(\mathbf{k}')+r_\omega ({\vec {k}}')+(-1)^{\rho -1} m_{\omega }({\vec {k}}'), \\ {{\tilde{b}}}_{\omega ,h-1}(\mathbf{k}')= & {} -{{\tilde{v}}}_{\omega ,h-1}(\mathbf{k}') (ik_1'+\omega k_2')+s_\omega ({\vec {k}}'), \end{aligned}$$\end{document}$$with$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{Z}}}_{\rho ,\omega ,h-1}(\mathbf{k}')= & {} Z_{\rho ,\omega ,h}+z_{\rho ,\omega ,h}\,\chi _h(\mathbf{k}'), \\ {{\tilde{v}}}_{\omega ,h-1}(\mathbf{k}')= & {} v_{\omega ,h}+u_{\omega ,h}\,\chi _h(\mathbf{k}'). \end{aligned}$$\end{document}$$Now, by rewriting the support function $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _h(\mathbf{k}')$$\end{document}$ in the definition of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{g}}}^{(\leqslant h)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$ as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _h(\mathbf{k}')=f_h(\mathbf{k}')+\chi _{h-1}(\mathbf{k}')$$\end{document}$, we correspondingly rewrite: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{g}}}^{(\leqslant h)}_\omega (\mathbf{x},\mathbf{y})=\tilde{g}^{(h)}_\omega (\mathbf{x},\mathbf{y})+g^{(\leqslant h-1)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(\leqslant h-1)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$ is defined exactly as in ([4.25](#Equ80){ref-type=""}), ([4.26](#Equ81){ref-type=""}), with *h* replaced by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h-1$$\end{document}$, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,h-1}, v_{\omega ,h-1}$$\end{document}$ defined by the flow equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Z_{\rho ,\omega ,h-1}=Z_{\rho ,\omega ,h}+z_{\rho ,\omega ,h},\qquad v_{\omega ,h-1}=v_{\omega ,h}+u_{\omega ,h}. \end{aligned}$$\end{document}$$We are now ready to integrate the fields on scale *h*. We define:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&e^{-V^{(h-1)}(\Psi )+B^{(h-1)}(\Psi ,f,A)+w^{(h)}(f,A)}\nonumber \\&\quad =C_h\int {{\tilde{P}}}(d\Psi ^{(h)}) e^{-F_\xi ^{(h)}(\Psi ^{(h)}+\Psi )-F_\delta ^{(h)}(\Psi ^{(h)}+\Psi )}\times \nonumber \\&\qquad \times e^{-{{\mathcal {R}}}V^{(h)}(\Psi ^{(h)}+\Psi )+{\mathcal {L}} B^{(h)}(\Psi ^{(h)}+\Psi , f, A)+ {\mathcal {R}} B^{(h)}(\Psi ^{(h)}+\Psi , f, A)}, \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{P}}}(d\Psi ^{(h)})$$\end{document}$ is the Gaussian integration with propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{g}}}^{(h)}_\omega $$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} F_\xi ^{(h)}(\Psi )= & {} \sum _{\omega ,s}2^h\xi _{\omega ,h}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}{{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega } {\hat{\Psi }}^-_{\mathbf{k}',s,\omega }, \qquad \\ F_\delta ^{(h)}(\Psi )= & {} \sum _{\omega ,s}2^h\delta _{\omega ,h}\int \frac{d\mathbf{k}'}{(2\pi )^{3}}{{\hat{\Psi }}}^+_{\mathbf{k}',s,\omega } \sigma _3{\hat{\Psi }}^-_{\mathbf{k}',s,\omega }, \end{aligned}$$\end{document}$$and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_h^{-1}= \int \tilde{P}(d\Psi ^{(h)})e^{-F_\xi ^{(h)}(\Psi ^{(h)})+{{\mathcal {R}}}V^{(h)}(\Psi ^{(h)})}$$\end{document}$. Finally, letting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {W}}^{(h-1)}={\mathcal {W}}^{(h)}+w^{(h)}$$\end{document}$, we obtain the same expression as ([4.21](#Equ76){ref-type=""}), with *h* replaced by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h-1$$\end{document}$. This concludes the proof of the inductive step, corresponding to the integration of the fields on scale *h*, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\geqslant h^*_1$$\end{document}$. By construction, the running coupling constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {\tau }}_h=(\xi _{\omega ,h},\delta _{\omega ,h}, Z_{A,\omega ,h}, Z_{B,\omega ,h},v_{\omega ,h})_{\omega \in \{\pm \}}$$\end{document}$ verify the following recursive equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{\omega ,h-1}= & {} 2\xi _{\omega ,h}+\beta ^\xi _{\omega ,h}(U, {\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0),\nonumber \\ \delta _{\omega ,h-1}= & {} 2\delta _{\omega ,h}+\beta ^\delta _{\omega ,h}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0),\nonumber \\ Z_{\rho ,\omega ,h-1}= & {} Z_{\rho ,\omega ,h}+\beta ^{Z,\rho }_{\omega ,h}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0),\nonumber \\ v_{\omega ,h-1}= & {} v_{\omega ,h}+\beta ^v_{\omega ,h}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0), \end{aligned}$$\end{document}$$for suitable functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ^\sharp _{\omega ,h}$$\end{document}$, known as the (components of the) *beta function*. Note that the initial data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,0},\delta _{\omega ,0},Z_{\rho ,\omega ,0},v_{\omega ,0}$$\end{document}$ are analytically close to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi ,\delta ,1,\frac{3}{2} $$\end{document}$, respectively; they are not exactly independent of the indices $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho ,\omega $$\end{document}$, due to the effect of the ultraviolet integration. However, for small values of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},+}$$\end{document}$, the difference between the initial data, for different values of the indices, differ at most by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O(U m_{\text {R},+})$$\end{document}$ (note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},+}=O(|m_{\text {R},+}|+\sin \phi )$$\end{document}$). As we shall see below, the running coupling constants remain analytically close to their initial data, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\leqslant 0$$\end{document}$. Similarly, the vertex functions satisfy recursive equations driven by the running coupling constants themselves:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \gamma _{\mu ,\omega ,h-1}=\gamma _{\mu ,\omega ,h}+\sum _{h'=h}^0\gamma _{\mu ,\omega ,h'}\,{{\tilde{\beta }}}^\gamma _{\mu ,\omega ,h'}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0)\;, \end{aligned}$$\end{document}$$whose solution remains analytically close to the corresponding initial data, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\leqslant 0$$\end{document}$. From the structure and properties of the effective propagator on scale *h*, see ([4.25](#Equ80){ref-type=""}) and following lines, one recognizes that the effective theory at scale *h* is a lattice regularization of a theory of relativistic fermions with masses $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R}, \pm }$$\end{document}$. As anticipated above, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,h}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{\omega ,h}$$\end{document}$ remain analytically close to their initial data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1,\frac{3}{2}$$\end{document}$, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\leqslant 0$$\end{document}$: therefore, it is straightforward to check that the single scale propagator satisfies$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |g^{(h)}_\omega (\mathbf{x},\mathbf{y})|\leqslant C_N \frac{2^{2 h}}{1+(2^h|\mathbf{x}-\mathbf{y}|)^N }\;,\qquad \forall N\geqslant 1\;. \end{aligned}$$\end{document}$$Moreover, the single-scale propagator admits the decomposition:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} g^{(h)}_\omega (\mathbf{x},\mathbf{y}) = G^{(h)}_{\omega }(\mathbf{x},\mathbf{y}) + g^{(h)}_{\omega ,r}(\mathbf{x},\mathbf{y}) \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G^{(h)}_{\omega }(\mathbf{x},\mathbf{y})$$\end{document}$ is obtained from $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(h)}_\omega (\mathbf{x},\mathbf{y})$$\end{document}$ by setting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},\omega }=0$$\end{document}$, and where the remainder term $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(h)}_{\omega ,r}$$\end{document}$ satisfies the same bound as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(h)}_\omega $$\end{document}$ times an extra factor $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R}, \omega } 2^{-h}$$\end{document}$, which is small, for all scales larger than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1$$\end{document}$. Due to the fact that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R}, +} \geqslant |m_{R, -}|$$\end{document}$, once we reach the scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h = h^{*}_{1}$$\end{document}$, the infrared propagator of the field corresponding to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega = +$$\end{document}$ satisfies the following bound:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |g^{(\leqslant h_1^*)}_+(\mathbf{x},\mathbf{y})|\leqslant C_N \frac{2^{2 h^*_1}}{1+(2^{ h^*_1}|\mathbf{x}-\mathbf{y}|)^N }\;; \end{aligned}$$\end{document}$$that is, it admits the same qualitative bound as the corresponding single scale propagator on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h = h^{*}_{1}$$\end{document}$. For this reason, it can be integrated in a single step, without any further need for a multiscale analysis. We do so and, after its integration, we are left with an effective theory on scales $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\leqslant h^*_1$$\end{document}$, depending only on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi ^{(\leqslant h^{*}_{1})}_{-}$$\end{document}$, which we integrate in a multiscale fashion, similar to the one described above, until the scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h = h^{*}_{2}$$\end{document}$ is reached. At that point, the infrared propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(\leqslant h_2^*)}_-$$\end{document}$ satisfies a bound similar to ([4.32](#Equ87){ref-type=""}), with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^{*}_{1}$$\end{document}$ replaced by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^{*}_{2}$$\end{document}$, and the corresponding field can be integrated in a single step. The outcome of the final integration is the desired generating function. The iterative integration procedure described above provides an explicit algorithm for computing the kernels of the effective potential and sources. In particular, they can be represented as sums of *Gallavotti--Nicolò trees*, identical to those of \[[@CR20], Section 3\], modulo the following minor differences. The endpoints *v* on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_{v} = +1$$\end{document}$ are associated either with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F^{(0)}_\xi (\Psi ^{(\leqslant 0)})$$\end{document}$, or with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F^{(0)}_\delta (\Psi ^{(\leqslant 0)})$$\end{document}$, or with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {L}}B^{(0)}(\Psi ^{(\leqslant 0)},f,A)$$\end{document}$, or with one of the terms in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal {R}}}{{\mathcal {V}}}^{(0)}(\Psi ^{(\leqslant 0)})$$\end{document}$ or in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal {R}}}B^{(0)}(\Psi ^{(\leqslant 0)},f,A)$$\end{document}$; the endpoints on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_{v}\leqslant 0$$\end{document}$ are, instead, associated either with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F^{(h_v-1)}_\xi (\Psi ^{(\leqslant h_v-1)})$$\end{document}$, or with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F^{(h_v-1)}_\delta (\Psi ^{(\leqslant h_v-1)})$$\end{document}$, or with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal {L}}}B^{(h_v-1)}(\Psi ^{(\leqslant h_v-1)},f,A)$$\end{document}$. The most important novelty of the present construction, as compared with \[[@CR20]\], is the presence of the relevant couplings $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,h},\delta _{\omega ,h}$$\end{document}$, whose flow must be controlled by properly choosing the counterterms $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi $$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta $$\end{document}$, see discussion below. Recall that the flows of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,h}$$\end{document}$ stop at scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1$$\end{document}$; for smaller scales, we let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h'}=\delta _{+,h'}=0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\forall h'<h^*_1$$\end{document}$. Similarly, we let the other running coupling constants with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$, that is, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,+,h}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{+,h}$$\end{document}$, be zero for scales smaller than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1$$\end{document}$. It turns out that the tree expansion is *absolutely convergent*, provided that *U* is small enough and the relevant couplings remain small, uniformly in the scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\leqslant 0$$\end{document}$. More precisely, the kernels of the effective potential satisfy the following bound (a similar statement is valid, of course, for the kernels of the effective source). Notation-wise, we let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_n^{(h)}(\mathbf{x}_1,\ldots ,\mathbf{x}_n)$$\end{document}$ be the kernel of the effective potential $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {V}}^{(h)}(\Psi )$$\end{document}$ associated with the monomial in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi $$\end{document}$ of order *n*; of course, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^{(h)}_n$$\end{document}$ is non zero only if *n* is even. The arguments $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{x}_1,\ldots ,\mathbf{x}_n$$\end{document}$ are the space-time coordinates of the Grassmann fields; the kernel implicitly depends also on the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho ,\omega $$\end{document}$ indices of the external fields, but we do not spell out their dependence explicitly. We also let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert W^{(h)}_n\Vert _1:=\int d\mathbf{x}_2\cdots d\mathbf{x}_n |W^{(h)}_n(\mathbf{x}_1,\ldots ,\mathbf{x}_n)|$$\end{document}$ (here $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\mathbf{x}$$\end{document}$ is a shorthand for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int _0^\beta dx_0\sum _{{\vec {x}}\in \Lambda _L}$$\end{document}$), which is independent of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{x}_1$$\end{document}$, due to translational invariance. Lemma 4.1 {#FPar10} --------- There exist positive constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_0, \theta , C_0$$\end{document}$, such that the following is true. Suppose that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max _{\rho ,\omega ,k\geqslant h}\{|Z_{\rho ,\omega ,k} - 1|,|v_{\omega , k} - \frac{3}{2}|,|\xi _{\omega , k}|,|\delta _{\omega , k}|\}\leqslant C|U|$$\end{document}$. Then, the kernels of the effective potential on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h-1$$\end{document}$ are analytic in *U* for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\leqslant U_0/(C+1)$$\end{document}$, and satisfy the bound$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\Vert {W}^{(h-1)}_{2} \Vert _{1}\leqslant C|U|2^{h}+C_0|U|2^{h(1+\theta )}\;, \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\Vert {W}^{(h-1)}_{n} \Vert _{1}\leqslant C_0^n |U|^{\frac{n}{2}-1} 2^{h(3 - n+\theta )}\,, \qquad \forall n\geqslant 4\;. \end{aligned}$$\end{document}$$The components of the beta function are analytic in *U* in the same domain, and satisfy:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \big |\beta ^\#_{\omega ,h}(U, {\vec {\tau }}_h,\ldots , {\vec {\tau }}_0)\big |\leqslant C_0 |U| 2^{\theta h}\,. \end{aligned}$$\end{document}$$ The proof of the lemma goes along the same lines as the proof of \[[@CR20], Theorem 2\], see also the review \[[@CR18]\], and will not be repeated here. Two key ingredients in the proof are: the representation of the iterated truncated expectations in terms of the Brydges--Battle--Federbush determinant formula, and the Gram-Hadamard bound. The factors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{\theta h}$$\end{document}$ appearing in the right sides of ([4.33](#Equ88){ref-type=""}), ([4.34](#Equ89){ref-type=""}) and ([4.35](#Equ90){ref-type=""}), represent a 'dimensional gain', as compared to a more basic, naive, dimensional bound, proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{(3-n)h}$$\end{document}$, which is suggested by the fact that the scaling dimension of the contributions to the effective potential with *n* external fermionic is equal to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3-n$$\end{document}$, in the RG jargon (we use the convention that positive/negative scaling dimensions correspond to relevant/irrelevant operators). Such a dimensional gain is due to the *RG irrelevance* of the quartic interaction (note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3-n=-1$$\end{document}$ for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=4$$\end{document}$) and to the so-called short-memory property of the Gallavotti-Nicolò trees ("long trees are exponentially suppressed"): all the contributions to the effective potential associated with trees that have at least one endpoint on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+1$$\end{document}$ have this additional exponentially decaying factor. The only contributions not having such a gain are those associated with trees without endpoints on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+1$$\end{document}$. The key remark is that, since the running coupling constants are all associated with quadratic contributions in the fermionic fields, such contributions are very simple and explicit: they can all be represented as sums of linear Feynman diagrams with two external legs ('chain diagrams'), obtained by contracting in all possible ways the two-legged vertices corresponding to the running coupling constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,k},\delta _{\omega ,k}$$\end{document}$. Therefore, they only contribute to the quadratic part of the effective potential, and they lead to the first term in the right side of ([4.33](#Equ88){ref-type=""}). Note also that such diagrams do not contribute to the beta function: in fact, the beta function at scale *h* is obtained by taking the 'local part' of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{2}^{(h)}$$\end{document}$, which is equal to the value of the Fourier transform $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widehat{W}}}_2^{(2)}$$\end{document}$ at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'=\mathbf{0}$$\end{document}$. If we compute the chain diagrams at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{k}'=\mathbf{0}$$\end{document}$, we see that the quasi-momenta of all the propagators of the chain diagram are equal to zero; therefore, the value of the diagram is zero, too, due to the compact support properties of the single-scale propagator. The idea, now, is to use the bound on the beta function to inductively prove the assumption on the running coupling constants, or, more precisely, the following improved version of the inductive assumption:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&|Z_{\rho ,\omega ,h} - 1|\leqslant C|U|,\quad |v_{\omega , h} - \frac{3}{2}|\leqslant C|U|,&\forall h^*_2\leqslant h\leqslant 0\;,\nonumber \\&|\xi _{-, h}|\leqslant C|U|2^{\theta h},\quad |\delta _{-, h}|\leqslant C|U|2^{\theta h},&\forall h^*_2\leqslant h\leqslant 0\;, \nonumber \\&|\xi _{+, h}-\xi _{-,h}|\leqslant C|U|2^{h^*_1-h},\quad |\delta _{+, h}-\delta _{-,h}|\leqslant C|U|2^{h^*_1-h},&\forall h^*_1\leqslant h\leqslant 0\;, \end{aligned}$$\end{document}$$for a suitable $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C>0$$\end{document}$ (recall that, by definition, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}=\delta _{+,h}=Z_{\rho ,+,h}=v_{+,h}=0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\forall h<h^*_1$$\end{document}$). Note that the bound on the beta function is already enough to prove the assumption for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,h}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{\omega ,h}$$\end{document}$. The subtle point is to control the flow of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega , h}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{\omega , h}$$\end{document}$, provided the initial data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi ,\delta $$\end{document}$ are properly chosen. This is the content of the next lemma. Lemma 4.2 {#FPar11} --------- There exist positive constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_0$$\end{document}$, *C*, and functions $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta =\delta (U,m_{\text {R},-},\phi )$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi =\xi (U,m_{\text {R},-},\phi )$$\end{document}$, analytic in *U* for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\leqslant U_0/(C+1)$$\end{document}$ and vanishing at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U=0$$\end{document}$, such that ([4.36](#Equ91){ref-type=""}) are verified. Proof {#FPar12} ----- We solve the beta function by looking at it as a fixed point equation on a suitable space of sequences. The fixed point equation arises by iterating the beta function equation and then imposing that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,h^*_2}=\delta _{-,h^*_2}=0$$\end{document}$. By iterating the first two equations of ([4.29](#Equ84){ref-type=""}), we get, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_2\leqslant h\leqslant 0$$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{\omega ,h}= & {} 2^{-h}\big (\xi _{\omega ,0}+\sum _{k=h+1}^{0} 2^{k-1} \beta ^\xi _{\omega ,k}(U, {\vec {\tau }}_k,\ldots ,{\vec {\tau }}_0)\big )\nonumber \\ \delta _{\omega ,h}= & {} 2^{-h}\big (\delta _{\omega ,0}+\sum _{k=h+1}^{0} 2^{k-1}\beta ^\delta _{\omega ,k}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0)\big )\;, \end{aligned}$$\end{document}$$with the understanding that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}=\delta _{+,h}=0$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\forall h<h^*_1$$\end{document}$. Consider first the case $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =-$$\end{document}$. By imposing the condition that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,h^*_2}=\delta _{-,h^*_2}=0$$\end{document}$, we find that$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{-,0}=-\sum _{k=h+1}^{0} 2^{k-1} \beta ^\xi _{-,k}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0),\qquad \delta _{-,0}=-\sum _{k=h+1}^{0} 2^{k-1}\beta ^\delta _{-,k}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0).\nonumber \\ \end{aligned}$$\end{document}$$Plugging these identities back in ([4.37](#Equ92){ref-type=""}) with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =-$$\end{document}$ gives$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{-,h}= & {} -\sum _{h^*_2<k\leqslant h} 2^{k-h-1}\beta ^\xi _{-,k}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0),\qquad \nonumber \\ \delta _{-,h}= & {} -\sum _{h^*_2<k\leqslant h} 2^{k-h-1}\beta ^\delta _{-,k}(U,{\vec {\tau }}_h,\ldots ,{\vec {\tau }}_0), \end{aligned}$$\end{document}$$which is the desired equation for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,h},\delta _{-,h}$$\end{document}$. Consider next the case $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$. The initial data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,0}, \delta _{+,0}$$\end{document}$ in the right side of ([4.37](#Equ92){ref-type=""}) are regarded as given functions of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi $$\end{document}$, whose explicit form follows from the ultraviolet integration, such that both $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,0}-\xi _{-,0}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,0}-\delta _{-,0}$$\end{document}$ are of the order $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O(U \min \{m_{\text {R},+},1\})$$\end{document}$. More explicitly, we write,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{+,0}= & {} \xi _{-,0}+\bar{x}_+(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi ), \qquad \nonumber \\ \delta _{+,0}= & {} \delta _{-,0}+{\bar{d}}_+(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi ), \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{x}}}_+$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{d}}}_+$$\end{document}$ are analytic in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U,\xi _{-,0},\delta _{-,0}$$\end{document}$ for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\xi _{-,0}|,|\delta _{-,0}|\leqslant C|U|$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\leqslant U_0/(C+1)$$\end{document}$, and satisfy:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&|{{\bar{x}}}_+(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )|\leqslant C_1|U|\min \{m_{\text {R},+},1\},\nonumber \\&|{{\bar{x}}}_+(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )-{{\bar{x}}}_+(U,\xi _{-,0}',\delta _{-,0}',m_{\text {R},-},\phi )| \nonumber \\&\quad \leqslant C_1|U|\min \{m_{\text {R},+},1\}(|\xi _{-,0}-\xi _{-,0}'|+|\delta _{-,0}-\delta _{-,0}'|)\;, \end{aligned}$$\end{document}$$for some $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1>0$$\end{document}$, and analogously for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{d}}}_+$$\end{document}$. Plugging ([4.40](#Equ95){ref-type=""}), with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,0},\delta _{-,0}$$\end{document}$ written as in ([4.38](#Equ93){ref-type=""}), back in ([4.37](#Equ92){ref-type=""}) with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$, we get the desired equation for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h},\delta _{+,h}$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{+,h}= & {} 2^{-h}\big ({{\bar{x}}}_++\sum _{k=h+1}^{0} 2^{k-1} (\beta ^\xi _{+,k}-\beta ^\xi _{-,k}) -\sum _{k=h^*_2+1}^h2^{k-1}\beta ^\xi _{-,k}\big )\;,\nonumber \\ \delta _{+,h}= & {} 2^{-h}\big ({{\bar{d}}}_++\sum _{k=h+1}^{0} 2^{k-1} (\beta ^\delta _{+,k}-\beta ^\delta _{-,k}) -\sum _{k=h^*_2+1}^h2^{k-1}\beta ^\delta _{-,k}\big )\;, \end{aligned}$$\end{document}$$for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1\leqslant h\leqslant 0$$\end{document}$. The equations ([4.39](#Equ94){ref-type=""}) and ([4.42](#Equ97){ref-type=""}), together with the analogues of ([4.37](#Equ92){ref-type=""}) for the running coupling constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,h},v_{\omega ,h}$$\end{document}$, are looked at as a fixed point equation on the space $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {M}}$$\end{document}$ of sequences of running coupling constants $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\underline{\tau }}:=\{{\vec {\tau }}_{h^*_2},\ldots ,{\vec {\tau }}_0\}$$\end{document}$, endowed with the norm$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert {{\underline{\tau }}}\Vert _\theta= & {} \max \Big \{\max _{\begin{array}{c} h\leqslant 0\\ \omega ,\rho \end{array}}\{|Z_{\rho ,\omega ,h}-1|,|v_{\omega ,h}-\frac{3}{2} |,2^{-\theta h}|\xi _{-,h}|,2^{-\theta h}|\delta _{-,h}|\},\nonumber \\&\quad \max _{h^*_1\leqslant h\leqslant 0}\{|\xi _{+,h}-\xi _{-,h}|2^{h-h^*_1},|\delta _{+,h}-\delta _{-,h}|2^{h-h^*_1}\}\Big \}. \end{aligned}$$\end{document}$$More precisely, the sequence of running coupling constants, solution of the flow equation with boundary data such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,h^*_2}=\delta _{-,h^*_2}=0$$\end{document}$, is the fixed point of the map $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\tau }}}\rightarrow {{\underline{\tau }}}'=\mathbf{T}({{\underline{\tau }}})$$\end{document}$ that, in components, reads (we write the argument of the beta function as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(U,{{\underline{\tau }}})$$\end{document}$, and we do not indicate the argument of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{x}}}_+$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{d}}}_+$$\end{document}$, for short):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{-,h}'= & {} -\sum _{k=h^*_2+1}^h 2^{k-h-1}\beta ^\xi _{-,k}(U, {{\underline{\tau }}}),\quad \forall h^*_2\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \delta _{-,h}'= & {} -\sum _{k=h^*_2+1}^h 2^{k-h-1}\beta ^\delta _{-,k}(U, {{\underline{\tau }}}),\quad \forall h^*_2\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \xi _{+,h}'= & {} 2^{-h}{{\bar{x}}}_+ +\sum _{k=h+1}^{0} 2^{k-h-1} (\beta ^\xi _{+,k}(U, {{\underline{\tau }}})-\beta ^\xi _{-,k}(U, {{\underline{\tau }}}))\nonumber \\&- \sum _{k=h^*_2+1}^h2^{k-h-1}\beta ^\xi _{-,k}(U, {{\underline{\tau }}}),\quad \forall h^*_1\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \delta _{+,h}'= & {} 2^{-h}{{\bar{d}}}_+ +\sum _{k=h+1}^{0} 2^{k-h-1} (\beta ^\delta _{+,k}(U, {{\underline{\tau }}})-\beta ^\delta _{-,k}(U, {{\underline{\tau }}}))\nonumber \\&-\sum _{k=h^*_2+1}^h2^{k-h-1}\beta ^\delta _{-,k}(U, {{\underline{\tau }}}), \quad \forall h^*_1\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Z_{\rho ,\omega ,h}'= & {} 1+{{\bar{z}}}_{\rho ,\omega } +\sum _{k=h+1}^{0}\beta ^{Z,\rho }_{\omega ,k}(U, {{\underline{\tau }}}),\quad \forall h^*_2\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} v_{\omega ,h}'= & {} \frac{3}{2}t_1+{{\bar{v}}}_{\omega } +\sum _{k=h+1}^{0} \beta ^v_{\omega ,k}(U, {{\underline{\tau }}})\;,\quad \forall h^*_2\leqslant h\leqslant 0 \end{aligned}$$\end{document}$$with the understanding that the running coupling constants with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =+$$\end{document}$ are zero for all scales smaller than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1$$\end{document}$: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}=\delta _{+,h}=Z_{\rho ,+,h}=v_{+,h}=0$$\end{document}$, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h<h^*_1$$\end{document}$. Moreover, in the last two lines, we rewrote $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_{\rho ,\omega ,0}=1+{{\bar{z}}}_{\rho ,\omega }$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{\omega ,0}=\frac{3}{2}+{{\bar{v}}}_\omega $$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{z}}}_{\rho ,\omega }={\bar{z}}_{\rho ,\omega }(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{v}}}_{\omega }={\bar{v}}_\omega (U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )$$\end{document}$ are analytic in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U,\xi _{-,0},\delta _{-,0}$$\end{document}$ for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\xi _{-,0}|,|\delta _{-,0}|\leqslant C|U|$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|U|\leqslant U_0/(C+1)$$\end{document}$, and satisfy:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&|{{\bar{z}}}_{\rho ,\omega }(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )|\leqslant C_1|U|,\nonumber \\&|{{\bar{z}}}_{\rho ,\omega }(U,\xi _{-,0},\delta _{-,0},m_{\text {R},-},\phi )-{{\bar{z}}}_{\rho ,\omega }(U,\xi _{-,0}',\delta _{-,0}',m_{\text {R},-},\phi )|\nonumber \\&\quad \leqslant C_1|U|(|\xi _{-,0}-\xi _{-,0}'|+|\delta _{-,0}-\delta _{-,0}'|)\;, \end{aligned}$$\end{document}$$and analogously for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{v}}}_\omega $$\end{document}$. In addition, the differences $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{z}}}_{\rho ,+}-{{\bar{z}}}_{\rho ,-}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{v}}}_+-{{\bar{v}}}_-$$\end{document}$ satisfy the same bound as ([4.41](#Equ96){ref-type=""}). We want to show that the map $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\tau }}} \rightarrow \mathbf{T}({{\underline{\tau }}})$$\end{document}$ admits a unique fixed point in the ball $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_0=\{{{\underline{\tau }}}\in {\mathcal {M}}: \Vert {{\underline{\tau }}}\Vert _\theta \leqslant C|U|\}$$\end{document}$, for a suitable $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C>0$$\end{document}$. In order to prove this, we show that, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\tau }}},{{\underline{\tau }}}'\in B_0$$\end{document}$,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert \mathbf{T}({{\underline{\tau }}})\Vert _\theta \leqslant C|U|,\qquad \Vert \mathbf{T}({{\underline{\tau }}})-\mathbf{T}({{\underline{\tau }}}')\Vert _\theta \leqslant C|U|\, \Vert {{\underline{\tau }}}-{{\underline{\tau }}}'\Vert _\theta \,, \end{aligned}$$\end{document}$$for a suitable *C*. Once ([4.51](#Equ106){ref-type=""}) is proved, the existence of a unique fixed point in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_0$$\end{document}$ follows via the Banach fixed point theorem, and we are done: such a fixed point defines the initial data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,0},\delta _{-,0}$$\end{document}$ generating a solution to the flow equation satisfying ([4.36](#Equ91){ref-type=""}), as desired. Of course, fixing $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{-,0},\delta _{-,0}$$\end{document}$ is equivalent (thanks to the analytic implicit function theorem) to fixing $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi ,\delta $$\end{document}$: therefore, the existence of such a fixed point proves the statement of the lemma. We are left with proving ([4.51](#Equ106){ref-type=""}). If $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\underline{\tau }\in B_0$$\end{document}$, by using the bound ([4.35](#Equ90){ref-type=""}) on the beta function, as well as the assumptions ([4.41](#Equ96){ref-type=""}), ([4.50](#Equ105){ref-type=""}) on the initial data (together with their analogues for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{d}}}_+,{{\bar{v}}}_\omega $$\end{document}$), it is immediate to check that$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |Z_{\rho ,\omega ,h}'-1|\leqslant C|U|, \quad |v_{\omega ,h}'-\frac{3}{2}|\leqslant C|U|,\quad |\xi _{-,h}'|\leqslant C|U|2^{\theta h},\quad |\delta _{-,h}'|\leqslant C|U|2^{\theta h},\nonumber \\ \end{aligned}$$\end{document}$$for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_2\leqslant h\leqslant 0$$\end{document}$ and a suitable constant *C*. Therefore, in order to check that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert \mathbf{T}({{\underline{\tau }}})\Vert _\theta \leqslant C|U|$$\end{document}$, we are left with proving that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max \{ |\xi _{+,h}'-\xi _{-,h}'|,|\delta _{+,h}'-\delta _{-,h}'|\}\leqslant C|U|2^{h^*_1-h}$$\end{document}$, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1\leqslant h\leqslant 0$$\end{document}$. We spell out the argument for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}'-\xi _{-,h}'$$\end{document}$, the proof for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,h}'-\delta _{-,h}'$$\end{document}$ being exactly the same. By using ([4.44](#Equ99){ref-type=""})--([4.46](#Equ101){ref-type=""}), we have: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,h}'-\xi _{-,h}'=2^{-h}{{\bar{x}}}_++\sum _{k=h+1}^{0} 2^{k-h-1} (\beta ^\xi _{+,k}(U, {{\underline{\tau }}})-\beta ^\xi _{-,k}(U, {{\underline{\tau }}}))$$\end{document}$. Now, the first term in the right side is bounded by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{-h}|{{\bar{x}}}_+|\leqslant 2C_1|U|$$\end{document}$, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h\geqslant h^*_1$$\end{document}$, by ([4.41](#Equ96){ref-type=""}) and the very definition of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1$$\end{document}$, ([4.20](#Equ75){ref-type=""}). In order to bound the sum $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sum _{k=h+1}^{0} 2^{k-h-1} (\beta ^\xi _{+,k}(U, {{\underline{\tau }}})-\beta ^\xi _{-,k}(U, {{\underline{\tau }}}))$$\end{document}$, we note that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ^\xi _{+,k}-\beta ^\xi _{-,k}$$\end{document}$ can be expressed as a sum over trees with root on scale *k*, at least an endpoint on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+1$$\end{document}$ (recall the discussion after the statement of Lemma ([4.1](#FPar10){ref-type="sec"})) and: either an endpoint corresponding to a difference $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,k'}-\xi _{-,k'}$$\end{document}$, or an endpoint corresponding to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,k'}-\delta _{-,k'}$$\end{document}$, or a propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_{+}-g^{(k')}_-$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k'\geqslant k$$\end{document}$. The propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_{+}-g^{(k')}_-$$\end{document}$ admits a dimensional bound that is the same as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_{\omega }$$\end{document}$ times a gain factor $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{h^*_1-k'}$$\end{document}$; the differences $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{+,k'}-\xi _{-,k'}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,k'}-\delta _{-,k'}$$\end{document}$ are proportional to the same gain factor, due to the assumption that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\underline{\tau }}}\in B_0$$\end{document}$. All in all, recalling the basic bound on the beta function, ([4.35](#Equ90){ref-type=""}), we find a similar bound, improved by the gain factor $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{h^*_1-k}$$\end{document}$:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \big |\beta ^\xi _{+,k}(U,{{\underline{\tau }}})-\beta ^\xi _{-,k}(U,{{\underline{\tau }}})\big |\leqslant 2C_0|U| 2^{h^*_1-k}2^{\theta k}. \end{aligned}$$\end{document}$$This, together with the bound on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{-h}{{\bar{x}}}_+$$\end{document}$, implies the desired bound, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\xi _{+,h}'-\xi _{-,h}'|\leqslant C|U|2^{h^*_1-h}$$\end{document}$, for all $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_1\leqslant h\leqslant 0$$\end{document}$ and *C* sufficiently large. Exactly the same argument implies the desired bound for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{+,h}'-\delta _{-,h}'$$\end{document}$. The proof of the second of ([4.51](#Equ106){ref-type=""}) goes along the same lines, and we only sketch it here. A similar argument, discussed in all details, can be found in \[[@CR11], Section 4\]. Let us focus, for simplicity, on the first component of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{T}({{\underline{\tau }}})-\mathbf{T}({{\underline{\tau }}}')$$\end{document}$, which reads:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} -\sum _{k=h^*_2+1}^h 2^{k-h-1}\big (\beta ^\xi _{-,k}(U, {{\underline{\tau }}})-\beta ^\xi _{-,k}(U, {{\underline{\tau }}}')\big ). \end{aligned}$$\end{document}$$The difference $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ^\xi _{-,k}(U,{{\underline{\tau }}})-\beta ^\xi _{-,k}(U,{{\underline{\tau }}}')$$\end{document}$ can be represented as a sum over trees with root on scale *k*, at least an endpoint on scale $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+1$$\end{document}$, and: either an endpoint corresponding to a difference $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _{\omega ,k'}-\xi _{\omega ,k'}'$$\end{document}$, or an endpoint corresponding to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{\omega ,k'}-\delta _{\omega ,k'}'$$\end{document}$, or a propagator corresponding to the difference between $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_{\omega }$$\end{document}$ computed at the values $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(Z_{\rho ,\omega ,k'},v_{\omega ,k'})$$\end{document}$ of the effective parameters and the same propagator computed at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(Z_{\rho ,\omega ,k'}',v_{\omega ,k'}')$$\end{document}$, for some $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k'\geqslant k$$\end{document}$. The difference between the propagators computed at different values of the effective parameters can be bounded dimensionally in the same way as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_{\omega }$$\end{document}$, times an additional factor $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max _{\rho ,\omega }\{|Z_{\rho ,\omega ,k'}-Z_{\rho ,\omega ,k'}'|, |v_{\omega ,k'}-v_{\omega ,k'}'|\}$$\end{document}$. Therefore, recalling the basic bound on the beta function, ([4.35](#Equ90){ref-type=""}), we find a similar bound, multiplied by the norm of the difference between the running coupling constants:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Big |\beta ^\xi _{-,k}(U,{{\underline{\tau }}})-\beta ^\xi _{-,k}(U,{{\underline{\tau }}}')\Big |\leqslant 2C_0|U| 2^{\theta k} \Vert {{\underline{\tau }}}-{{\underline{\tau }}}'\Vert _\theta , \end{aligned}$$\end{document}$$which implies the desired estimate on the first component of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{T}({{\underline{\tau }}})-\mathbf{T}({{\underline{\tau }}}')$$\end{document}$. A similar argument is valid for the other components, but we will not belabor the details here. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ We now have all the ingredients to prove Proposition [3.3](#FPar6){ref-type="sec"}. In fact, in view of Lemma [4.1](#FPar10){ref-type="sec"} and Lemma [4.2](#FPar11){ref-type="sec"}, we can fix the counterterms $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi ,\delta $$\end{document}$ in such a way that the kernels of the effective potential on all scales are analytic in *U*, uniformly in the scale, and satisfy ([4.33](#Equ88){ref-type=""}). A simple by-product of the proof shows that the kernel $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^{(h)}_{n}(\mathbf{x}_1,\ldots ,\mathbf{x}_n)$$\end{document}$ decays faster than any power in the tree distance among the space-time points $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{x}_1,\ldots ,\mathbf{x}_n$$\end{document}$, with a decay length proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{-h}$$\end{document}$. Analogous claims are valid for the kernels of the effective source term and of the generating function. In particular, recalling that the scale *h* is always larger or equal than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h^*_2$$\end{document}$, we have that the kernels of the effective potential, which are nothing else but the multi-point correlation functions, are analytic in *U* and decay faster than any power in the tree distance among their arguments, with a typical decay length of the order $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{h^*_2}\sim |m_{\text {R},-}|$$\end{document}$. Therefore, for any $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$, the Fourier transform of any multi-point correlation of local operators is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^\infty $$\end{document}$ in the momenta. In the massless case, the correlations are dimensionally bounded like in the graphene case \[[@CR20], [@CR21]\]: in particular, the two-point density--density, or current--current correlations decay like $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\mathbf{x}-\mathbf{y}|^{-4}$$\end{document}$ at large Euclidean space-time separation. For further details about the construction and estimate of the correlation functions, the reader is referred to, e.g., \[[@CR16], [@CR21]\]. This concludes the proof of Proposition [3.3](#FPar6){ref-type="sec"}. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ Proof of Theorem [2.1](#FPar1){ref-type="sec"} {#Sec13} ============================================== In order to conclude the proof of Theorem [2.1](#FPar1){ref-type="sec"}, we need to prove that: there exists a choice of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R,-}}$$\end{document}$ for which the Euclidean correlations of the reference model with Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text{ R }}$$\end{document}$, see ([2.22](#Equ22){ref-type=""}), coincide with those of the original Hamiltonian $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$; the Euclidean Kubo conductivity coincides with the real-time one. Cf. with the last two items, (iii) and (iv), of the list after ([2.23](#Equ23){ref-type=""}). We also need to prove the regularity and symmetry properties of the critical curves, stated in Theorem [2.1](#FPar1){ref-type="sec"}. Let us start with discussing item (iii), as well as the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^1$$\end{document}$ regularity of the critical curves. In order to prove the equivalence of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {H}}^{\text{ R }}$$\end{document}$, it is enough to fix the counterterms as discussed in the previous section, and choose $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$ to be the solution of ([2.23](#Equ23){ref-type=""}). Let us then show that ([2.23](#Equ23){ref-type=""}) can be inverted in the form $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{R,-}= m_{R,-}(U, W, \phi )$$\end{document}$, with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{R,-}(U, W, \phi )$$\end{document}$ analytic in *U* and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{1}$$\end{document}$ in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W, \phi $$\end{document}$. We want to appeal to the analytic implicit function theorem. For this purpose, we need to estimate the derivative of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U,m_{\text {R,-}},\phi )$$\end{document}$ w.r.t. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}$$\end{document}$. Recall that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{-,0}=\delta _{-,0}(U,m_{\text {R},-},\phi )$$\end{document}$ satisfies the second of ([4.38](#Equ93){ref-type=""}), and that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U,m_{\text {R},-},\phi )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{-,0}(U,m_{\text {R},-},\phi )$$\end{document}$ are analytically close (they differ only because of the effect of the ultraviolet integration). Therefore,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \delta (U,m_{R,-},\phi )=-\sum _{k=h^*_2+1}^1 2^{k-1}\beta ^\delta _{-,k}(U,{{\underline{\tau }}}), \end{aligned}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta ^\delta _{-,k}(U,{{\underline{\tau }}})$$\end{document}$ accounts for the difference between $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta $$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta _{-,0}$$\end{document}$ due to the ultraviolet integration. Differentiating both sides with respect to the mass, we find:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{\partial \delta (U,m_{R,-},\phi )}{\partial m_{R,-}}=-\sum _{k=h^*_2+1}^1 2^{k-1}\frac{\partial \beta ^\delta _{-,k}}{\partial m_{R,-}}(U,{{\underline{\tau }}}), \end{aligned}$$\end{document}$$which should be looked at as (a component of) a fixed point equation for the derivatives of the running coupling constants, analogous to the ones solved in the proof of Lemma ([4.2](#FPar11){ref-type="sec"}). When acting on the beta function, the derivative with respect to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{R,-}$$\end{document}$ can either act on a propagator $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_\omega $$\end{document}$, or on a running coupling constant. When acting on a propagator, it replaces $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_\omega $$\end{document}$ by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial g^{(k')}_\omega }{\partial m_{\text {R},-}}$$\end{document}$, which is bounded dimensionally in the same way as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g^{(k')}_\omega $$\end{document}$, times an extra factor proportional to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{-k'}$$\end{document}$. On the other hand, the action of the derivative on a running coupling constant should be bounded inductively, in the same spirit as the proof of Lemma [4.2](#FPar11){ref-type="sec"}. All in all, recalling also the basic bound on the beta function, ([4.35](#Equ90){ref-type=""}), we get$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Big |\frac{\partial \delta (U,m_{R,-},\phi )}{\partial m_{R,-}}\Big |\leqslant \sum _{k=h^*_2+1}^1 2^{k}C_0|U|2^{\theta k}2^{-k}\leqslant C_2|U|, \end{aligned}$$\end{document}$$for a suitable constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_2$$\end{document}$. Exactly the same argument and estimates are valid for the derivative with respect to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$, so that$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Big |\frac{\partial \delta (U,m_{R,-},\phi )}{\partial \phi }\Big |\leqslant C_2|U|\;. \end{aligned}$$\end{document}$$The last estimate is optimal for small $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$. For larger values of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$, one can also take advantage of the symmetry under exchange $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \rightarrow \pi -\phi $$\end{document}$ (the 'magnetic reflections', see ([4.13](#Equ68){ref-type=""})) to conclude that the derivative of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta $$\end{document}$ with respect to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$ vanishes continuously as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \rightarrow (\pi /2)^-$$\end{document}$. Moreover, by the symmetry properties of the model, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta (U,0,0)=0$$\end{document}$. Therefore, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\delta (U, m_{\text {R}, -}, \phi )|\leqslant 2C_2|U| (|m_{\text {R},-}|+ \sin \phi )$$\end{document}$. Using these bounds and the implicit function theorem, we see that ([2.23](#Equ23){ref-type=""}) can be inverted in the form ([2.24](#Equ24){ref-type=""}), with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|{{\mathfrak {d}}}(U,W,\phi )|\leqslant C|U| (W+\sin \phi )$$\end{document}$ for some constant *C*. The equation for the critical curve in the parameter range we are considering is simply $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R,-}}=0$$\end{document}$, that is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=3\sqrt{3}t_2\sin \phi +\delta (U,0,\phi )$$\end{document}$, which is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^1$$\end{document}$ in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi $$\end{document}$ and, thanks to the symmetries of the problem, it satisfies the properties stated in Theorem [2.1](#FPar1){ref-type="sec"}. We are left with discussing item (iv), that is, the equivalence between the Euclidean and real-time Kubo conductivities. Given our bounds on the Euclidean correlations, the equivalence follows from result discussed in previous papers. In fact, our bounds imply that the current--current correlations, at large space-time separations, decay either faster-than-any-power decay, if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\text {R},-}\ne 0$$\end{document}$, or like $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\mathbf{x}-\mathbf{y}|^{-4}$$\end{document}$, otherwise: therefore, we can repeat step by step the proof of \[[@CR22], Theorem 3.1\], as the reader can easily check. For a slightly modified and simplified proof, see also \[[@CR2], Appendix B\] and \[[@CR35], Section 5\]. This concludes the proof of Theorem [2.1](#FPar1){ref-type="sec"}. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square $$\end{document}$ Concluding Remarks {#Sec14} ------------------ In conclusion, the universality of the Hall conductivity (i.e., its independence from the interaction strength) can be seen as a consequence of lattice conservation laws, combined with the regularity properties of the correlation functions. The quantization of the interacting Hall conductivity then follows from its quantization in the non-interacting case: however, an important point in the proof is to compare the interacting system and its conductivity with the right reference non-interacting system, that is, the one with the right value of the mass; this is the reason why we introduce a reference non-interacting system with mass equal to the renormalized mass of the interacting system; in order to fix the correct value of the renormalized mass, we need to solve a fixed point equation for it. The same strategy we proposed in the present context can be easily extended to prove that the Hall conductivity is constant against *any* deformation of the Hamiltonian, even non-translationally invariant ones, provided that the off-diagonal decay of the Euclidean correlations in space and imaginary time is sufficiently fast, in the sense specified by[5](#Fn5){ref-type="fn"} Proposition [3.3](#FPar6){ref-type="sec"}. Note that our universality result is valid as soon as the Fourier transform of the current--current-interaction correlations are $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^3$$\end{document}$ in momentum space, which corresponds to a space-time decay faster than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(dist.)^{-6}$$\end{document}$ (a critical analysis of the proof shows that we need even less: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^{2+\varepsilon }$$\end{document}$ with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon >0$$\end{document}$ is a sufficient condition for our construction to work; this translates into a space-time decay faster than $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(dist.)^{-5}$$\end{document}$). This means that we do not require the existence of a spectral gap, in the strong sense of exponential decay of correlations: sufficiently fast polynomial decay is actually enough. It would be nice to provide a realistic example of a gapless model with fast polynomial decay of correlations, exhibiting a non-trivial, universal behavior of the transverse conductivity; or, in alternative, to exclude the possibility that such a model exists. A problem connected with the one discussed in this paper, but much more challenging, is to prove universality of the conductivity for clean massless models with slow polynomial decay of correlations: by 'slow', here, we mean that Proposition [3.3](#FPar6){ref-type="sec"} cannot be applied. A first example is the Haldane model, considered in this paper, for values of the parameters *on* the critical line. In this case, as already recalled after the statement of Theorem ([2.1](#FPar1){ref-type="sec"}), one can prove the universality of the *longitudinal conductivity* \[[@CR19]\]: the proof, which generalizes the one in \[[@CR21]\], uses lattice Ward Identities, combined with the symmetry properties of the current--current correlation functions. It would be very interesting to establish the universality, or the violation thereof, of the transverse conductivity on the critical line. Another context, where the issue of the universality of the conductivity naturally arises, is the case of bulk massive systems in non-trivial domains with, say, Dirichlet conditions imposed at the boundary. In such a setting, usually, massless edge states appear, and the edge system is characterized by correlations with slow polynomial decay. Nevertheless, universality holds as a consequence of a more subtle mechanism, which relies on the non-renormalization of the edge chiral anomaly. Using these ideas, two of us proved the validity of the bulk-edge correspondence in lattice Hall systems with single-mode chiral edge currents \[[@CR2]\], and in the spin-conserving Kane--Mele model \[[@CR34]\]. It would be very interesting to generalize these findings to lattice systems with several edge modes, as well as to continuum systems. Finally, it would be extremely interesting to include disorder effect, even in the regime where the interaction is smaller than the non-interacting gap. Understanding the combined effects of disorder and interactions in the vicinity of the critical lines is a major open problem, even from a theoretical physics perspectives. We do not expect that the phase diagram will remain qualitatively unchanged in their presence: new quantum phases may in general arise in the vicinity of unperturbed critical lines. In this sense, we expect that the stability of the phase diagram, if valid at all, will depend on the specific features of the model under investigation. However, as far as we know, not even the effects of disorder alone are well understood in the vicinity of the critical lines. There is an issue in defining the position operator on the torus. In order to avoid the problem, we interpret the second term in ([2.15](#Equ15){ref-type=""}) as being equal to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i\frac{\partial }{\partial q_j}\pmb {\langle } [ {\mathcal {J}}_{i}(\vec {q}), N(-\vec {q})]\pmb {\rangle }_\infty \big |_{\vec {q}=\vec {0}}$$\end{document}$, where: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec {{\mathcal {J}}}({\vec {q}})= \sum _{{\vec {x}}\in \Lambda _L} {\vec {j}}_{{\vec {x}}}e^{i{\vec {q}}{\vec {x}}}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N({\vec {q}})=\sum _{{\vec {x}}\in \Lambda _L} n_{{\vec {x}}}e^{i{\vec {q}}{\vec {x}}}$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pmb {\langle } [ {\mathcal {J}}_{i}({\vec {q}}), N(-{\vec {q}})]\pmb {\rangle }_\infty := \lim _{\beta ,L\rightarrow \infty }\frac{1}{L^2}\langle [ {\mathcal {J}}_{i}({\vec {q}}^{(L)}), N(-{\vec {q}}^{(L)})]\rangle _{\beta ,L}$$\end{document}$, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\vec {q}}^{(L)}$$\end{document}$ a sequence of vectors in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal {B}}_L$$\end{document}$ such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{L\rightarrow \infty }{\vec {q}}^{(L)}={\vec {q}}$$\end{document}$. Here, by 'gap' we mean the rate of the exponential decay of the Euclidean correlations. The definition in ([3.6](#Equ30){ref-type=""}) is only valid for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x_0\ne y_0$$\end{document}$. However, the value at $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x_0=y_0$$\end{document}$ has no influence on the following formulas, in particular on ([3.8](#Equ32){ref-type=""}), which is the main goal of the following manipulations. In ([3.23](#Equ47){ref-type=""}), we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}_1$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{p}_2$$\end{document}$ the first and second arguments of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{K}}^{\text {R}}_{0,0, \sharp }$$\end{document}$, as well as of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{S}}^{\text {R}}_{i, \sharp }$$\end{document}$; correspondingly, we denote by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial }{\partial p_{1,i}}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial }{\partial p_{2,i}}$$\end{document}$ the derivatives with respect to the *i*-th components of the first and second arguments thereof. Proposition [3.3](#FPar6){ref-type="sec"} is formulated in terms of the regularity of the Fourier transform, but, by anti-trasforming and going back to real space, the stated properties of the correlation functions can be straightforwardly translated into a condition of sufficiently fast polynomial decay in space and imaginary time. Such a formulation would be the right one in order to deal with additional, possibly non-translationally invariant, perturbations, including weak random potentials (chosen in such a way that the non-interacting spectral gap is *not* closed by the randomness). **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The work of A. G. and of V. M. has been supported by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (ERC CoG UniCoSM, grant agreement n.724939). V. M. acknowledges support also from the Gruppo Nazionale di Fisica Matematica (GNFM). The work of M. P. has been supported by the Swiss National Science Foundation, via the grant "Mathematical Aspects of Many-Body Quantum Systems". [^1]: Communicated by Michael Aizenman.

Introduction {#S1} ============ Understanding the endogenous pathways that regulate inflammatory responses is of critical importance for the development of novel therapies for chronic inflammatory disease. Recent approaches for the treatment of inflammatory diseases have focussed around blocking specific inflammatory mediators such as the cytokines TNF-α, IL-1β, and IL-6 ([@B1]--[@B3]). Whilst such approaches have been very effective for certain diseases \[e.g., anti-TNF-α monoclonal antibodies for the treatment of rheumatoid arthritis (RA)\] a large percentage of patients either do not respond to treatment or become refractory to therapeutic antibody treatment ([@B4], [@B5]). It is known that cytokines play a central role in shaping the immune response to invading pathogens and in the context of chronic inflammation. It is now appreciated that there exists a plethora of lipid mediators as well as other immune modulating proteins and peptides that play important roles during both the onset as well as the resolution of inflammation ([@B6]--[@B8]). Annexin-A1 and its N-terminal peptide ac2-26 have been demonstrated to exert potent pro-resolution properties in multiple inflammatory disease models ([@B8], [@B9]). Similarly, a number of lipid mediators including the resolvins, lipoxins, maresins, and protectins have been more recently come to the fore ([@B7], [@B8]). These novel immune modulatory mediators and others may represent new avenues to explore for the treatment of chronic inflammatory disease. Murine chemerin is a 16 kDa protein produced and secreted as an inactive precursor, pro-chemerin, predominantly by hepatocytes and adipocytes ([@B10]). Pro-chemerin is present in the liver, spleen, skin, and plasma ([@B11], [@B12]). During an inflammatory response, inflammatory proteases produced locally by granulocytes and the coagulation cascade, cleave the carboxyl terminus of pro-chemerin to generate active chemerin isoforms at sites of inflammation ([@B13], [@B14]). Chemerin has been implicated in the pathology of a range of inflammatory diseases including RA, inflammatory bowel disease, psoriasis, diabetes, and cardiovascular disease ([@B15]--[@B19]). However, the exact role played by chemerin and its receptors in inflammatory disease remains unclear. Our group and others have demonstrated that active chemerin, once generated, is a potent chemoattractant for macrophages, immature dendritic cells (DCs), plasmacytoid dendritic cells (pDCs), and NK cells ([@B20]--[@B23]). Chemerin binds to three G protein-coupled receptors (GPCRs) with high affinity; CMKLR1, GPR1, and CCRL2 ([@B24]). Murine GPR1 is expressed in white adipose tissue, skin, muscle, and in the central nervous system ([@B25]). Although chemerin binding to GPR1 has been reported to induce downstream signalling, chemerin is thought to act as a partial agonist at GPR1 whilst chemerin acts as a full agonist at CMKLR1 ([@B26], [@B27]). CMKLR1 is expressed on various immune cells including macrophages, DCs, NK cells, and pDCs ([@B12]). It is also expressed on endothelial cells and adipocytes ([@B28], [@B29]). *Cmklr1* mediates the chemotactic effects of chemerin, and its activation has been reported to lead to rapid downstream signalling cascades, which are G~i/0~ coupled ([@B23], [@B26]). The Chemerin/*Cmklr1* axis has been implicated in driving the recruitment of immature DCs, pDCs, and NK cells to local sites of inflammation in a number of inflammatory diseases ([@B22], [@B30]--[@B32]). Interestingly, *Cmklr1* has also been reported to play an anti-inflammatory role in a number of inflammatory disease models, although these have predominantly been allergic inflammatory models ([@B12], [@B33]). In addition, our group and others have reported anti-inflammatory effects of synthetic chemerin-derived peptides in a number of inflammation models and these effects seem to be dependent on CMKLR1 ([@B34]--[@B36]). CCRL2 is a seven transmembrane receptor that lacks the DRYLAIV intracellular motif required for classical downstream signalling by GPCRs ([@B37]). It binds chemerin but does not induce classical downstream signalling nor does it internalise chemerin ([@B20], [@B37], [@B38]). CCRL2 is expressed on a range of cell types including macrophages, DCs, endothelial cells, and epithelial cells amongst others ([@B38], [@B39]). Expression of CCRL2 is upregulated in response to inflammatory stimuli but the function of CCRL2 during inflammation remains incompletely understood ([@B39], [@B40]). Zabel et al. have proposed a model in which CCRL2 binds to the non-signalling N-terminus of chemerin and then presents it to other cells expressing CMKLR1. In this way, CCRL2 could function to concentrate chemerin at local sites to augment chemerin signalling during inflammation ([@B38]). The aim of this study was to further explore the role of the non-signalling chemerin receptor CCRL2 during a self-resolving model of acute inflammation. We report, for the first time, that animals lacking the chemerin receptor CCRL2 displayed exaggerated neutrophil and inflammatory monocyte recruitment in models of acute inflammation. These effects were due in part to increased levels of chemerin, which augmented production of the neutrophil chemoattractant CXCL1, resulting in increased neutrophil recruitment. Materials and Methods {#S2} ===================== Animals {#S2-1} ------- B6.129-*Ccrl2^tm1Dgen^*/J mice were obtained from Jackson Laboratories (Bar Harbour, ME, USA). These mice, originally produced by Deltagen (San Mateo, CA, USA), were backcrossed for 12 generations onto the C57BL/6J background. All animal studies were conducted with ethical approval from the Dunn School of Pathology Local Ethical Review Committee and in accordance with the UK Home Office regulations (Guidance on the Operation of Animals, Scientific Procedures Act, 1986). Reagents {#S2-2} -------- Recombinant murine chemerin (aa17--156) was reconstituted in PBS supplemented with 0.1% BSA. Neutralising anti-chemerin antibody and goat IgG control were reconstituted in PBS. Both chemerin and antibodies were purchased from R&D Systems (Abingdon, UK). Zymosan was purchased from Sigma-Aldrich (Dorset, UK). Recombinant CCL5 was purchased from Peprotech (London, UK). Bio-gel P100 (45--90 µm) fine polyacrylamide beads were obtained from BIO-RAD Laboratories, Hemel Hempstead, Hertfordshire, UK. Murine Bone Marrow-Derived Macrophages (BMDMs) {#S2-3} ---------------------------------------------- Bone marrow-derived macrophages were generated as previously described ([@B41]). Briefly, fresh bone marrow cells from tibiae and femurs of C57BL/6J mice (8--12 weeks) were isolated and cultured in DMEM media supplemented with 10% heat inactivated fetal bovine serum, 10% L929 cell-conditioned media as a source of macrophage colony-stimulating factor, 100 U/ml penicillin, and 100 µg/ml streptomycin for 7 days. A total of 4 × 10^6^ bone marrow cells were seeded into 10 ml of medium in 100 mm Petri dishes (Sterilin, Abergoed, UK.) and on day 3 an additional 5 ml of medium was added. Cells were harvested by gentle agitation to lift cells off surface. Human Umbilical Vein Endothelial Cells (HUVEC) {#S2-4} ---------------------------------------------- Human umbilical vein endothelial cells were isolated as previously described ([@B42]) and frozen for storage in liquid N~2~. HUVEC were thawed and resuspended in endothelial cell growth medium with supplement mix C (PromoCell, Germany), 100 U/ml penicillin, and 100 µg/ml streptomycin and cultured in pre-coated 0.5% gelatin (Sigma) flask/dishes in a 37°C 5% CO~2~ incubator. Cell Activation Assays {#S2-5} ---------------------- Cells were seeded into 6-well plates at a concentration of 1 × 10^6^/ml. They were then exposed to TLR ligands and cytokines for 16 h in a 37°C 5% CO~2~ incubator as described previously. The TLR ligands and cytokines were added to cells at a final concentration of: LPS 100 µg/ml, IFNγ 20 ng/ml, Poly I:C 10 µg/ml, zymosan 10 µg/ml, flagellin 500 ng/ml, IL-4 20 ng/ml, and IL-13 20 ng/ml. RNA Isolation and Reverse Transcription and RT-PCR {#S2-6} -------------------------------------------------- Total RNA was extracted using the QIAGEN RNeasy Mini kit as instructed by the manufacturer as described previously ([@B43]). RNA concentration and purity was determined using ND-1000 spectrophotometer (Nanodrop, Thermo Scientific) at 260/280 and 260/230 nm. cDNA was synthesised from 500 to 800 ng of total RNA using QuantiTect Reverse Transcription kit following the manufacturer's instructions. cDNA was amplified for 15 min at 42°C and then 3 min at 95°C. Real-time quantitative PCR was performed using Sybr Select Master Mix (Applied Biosystems, Life Technologies) in the Step One Plus Real-time PCR System (Applied Biosystems). All primers were from QuantiTect Primer Assay (Qiagen). The thermal profile included an initiation step for 2 min at 50 and 95°C followed by 40× cycles of 15 s at 95°C and 1 min at 60°C. Cycle threshold values were determined by the StepOne software v2.3 and the mRNA content of samples was inferred by normalising to the housekeeping β-actin gene. Relative expression results were plotted as mRNA expression over actin, normalised to basal samples ([@B44]). Zymosan or Thioglycollate Induced Peritonitis {#S2-7} --------------------------------------------- Male *Ccrl2*^−^*^/^*^−^ or littermate controls were injected i.p. (0.5 ml) with 100 µg zymosan resuspended in PBS, or 1 ml of 4% thioglycollate (Thioglycollate brewers yeast; Sigma-Aldrich, Dorset, UK) prepared as described previously ([@B45]). Mice were sacrificed at specified time points and peritoneal cavities were lavaged with 5 ml ice-cold PBS supplemented with 2 mM EDTA. Blood was collected from the hepatic portal vein into EDTA-coated vacutainers and centrifuged at 2,000 × *g* for 20 min at 4°C to obtain plasma. Modulation of Chemerin Levels *In Vivo* {#S2-8} --------------------------------------- C57BL/6J mice were treated with either recombinant murine chemerin (4 µg) or PBS i.p. for 1 h before zymosan challenge for 4 h. Animals were sacrificed and peritoneal lavage was collected as before. *Ccrl2*^−^*^/^*^−^ mice were pretreated for 24 h with either 100 ng of isotype control IgG antibody or 100 ng of anti-chemerin polyclonal antibody i.p. This was followed by a second injection of isotype control IgG or anti-chemerin antibody 1 h before challenge with zymosan. Mice were sacrificed 4 h later and peritoneal lavage fluid was collected as before. Flow Cytometry {#S2-9} -------------- Peritoneal exudate cells (PECs) were resuspended in fluorescence-activated cell sorting (FACS) buffer (PBS; 2% FCS, 25 mM HEPES, 5 mM EDTA) containing CD16/CD32 FCγIIR blocking antibody (eBioscience). Specific staining with the following antibodies was performed with appropriate isotype controls; F4/80 (AbD Serotec, clone CI:A3-1), Ly6B.2 (AbDSerotec, Clone 7/4), Ly6 G (BioLegend, clone 1A8), CD11b (BioLegend, clone M1/70), CD115 (BioLegend, clone AFS98), CD4 (BioLegend, clone RM4-5), CD3 (BioLegend, clone 17A2), B220 (eBioscience, clone RA3-6B2), CD8 (BD Pharminogen, clone 53-6.7), and analysed by flow cytometry. Cells were counted using CountBright Absolute Counting Beads (Life Technologies). Cells were analysed with a Dako Cyan ADP flow cytometer (Beckman Coulter Ltd., High Wycombe, UK) and FlowJo software V10 (Tree Star Incorporation, Ashland, OR, USA). Monocytes and neutrophils in the peritoneum were defined as CD45^+^, 7/4^+^, Ly6G^−^ and CD45^+^, 7/4^+^, Ly6G^+^, respectively ([@B46], [@B47]). Blood was collected *via* hepatic portal vein into EDTA-coated vacutainers. Blood was treated in the same manner as the PECs, but red blood cells were lysed after antibody staining using BD FACS Lysing Solution (Buffered solution with \<15% formaldehyde and \<50% diethylene glycol) before fixation. Ly6C^hi^ blood monocytes were defined as CD45^+^, CD11b^+^, CD115^+^, Ly6C^hi^. Ly6C^lo^ monocytes were defined as CD45^+^, CD11b^+^, CD115^+^, Ly6C^lo^ ([@B48], [@B49]). Fluorescence-Activated Cell Sorting {#S2-10} ----------------------------------- Male C57BL/6J mice were injected i.p. (0.5 ml) with 100 µg zymosan resuspended in PBS. Steady state and zymosan challenged mice were sacrificed 4 h later, and peritoneal cavities were lavaged with 5 ml ice-cold PBS supplemented with 2 mM EDTA. PECs were stained for flow cytometry as described previously. Peritoneal macrophages, monocytes, and neutrophils were FACS sorted using a Beckman Astrios cell sorter directly into RLT buffer for RNA isolation using the QIAGEN RNeasy Mini kit. Detection of Secreted Protein by ELISA and Luminex {#S2-11} -------------------------------------------------- CXCL1, CCL2, IL-6, and chemerin in peritoneal exudate fluid and plasma were detected using ELISA (R&D Systems, Abingdon, UK). Sandwich ELISAs for chemerin, IL-6, CCL2, and CXCL1 were performed according to manufacturer's instructions. Custom multiplex polyacrylamide bead assays were purchased from R&D Systems to determine levels of CCL3, CCL4, IL-10, CXCL10, and MMP9 in peritoneal exudate fluid. Briefly, colour-coded beads were pre-coated with antibodies against the targets of interest. Biotinylated detection antibodies specific for each analyte were added, followed by phycoerythrin (PE)-conjugated streptavidin. Samples were read using a laser detection system, which quantifies the amount of PE present for each analyte. The 96-well plates were read on a Bio-Rad Bioanalyser with Bio-Plex Manager software (Hemel Hempstead, Hertfordshire, UK). ACEA xCELLigence Chemotaxis Assay {#S2-12} --------------------------------- 8- to 10-week-old male *Ccrl2*^−^*^/^*^−^ or littermate controls were injected i.p. with 1 ml 2% Bio-gel (P100 Fine, 45--90 µm). PECs (mixture of inflammatory macrophages and neutrophils) were isolated by peritoneal lavage with ice-cold PBS supplemented with 2 mM EDTA 4 days later. Real-time chemotaxis assays were performed using the ACEA RTCA-DP instrument as described previously ([@B50], [@B51]). Briefly, vehicle, chemerin, or recombinant murine CCL5 (160 µl) at 5 nM final concentration was added to the lower chamber of a CIM-16 plate. The upper chamber was attached, and 50 µl of prewarmed chemotaxis buffer added to each of the upper chambers. Following equilibration for 30 min, the plate was transferred into the RTCA-DP system. Bio-gel elicited PECs (50 µl---4 × 10^5^ cells/well), were then added to all upper wells. Cell index (CI) measurements were then taken every 5 s over the 3--4 h assay period. Chemotactic responses were assessed by quantifying the slope of the curve in the first 40 min and by the maximum CI minus minimum CI (Max--Min) values of the curve over the entire experiment. CMKLR1 β-Arrestin Recruitment Assays {#S2-13} ------------------------------------ Recruitment of β-Arrestin to Cmklr1 was measured using the Discoverx PathHunter^®^ eXpress β-Arrestin GPCR Assay following the manufacturer's protocol. Briefly, CHO-K1 cells stably expressing murine Cmklr1 were seeded into 1/2 area 96-well plates (15,000 cells/well) and incubated at 37°C, 5% CO~2~ for 48 h. Cells were then treated with either vehicle (Cell assay reagent) or indicated concentrations of anti-chemerin antibody for 45 min at 37°C, 5% CO~2~ before stimulation with vehicle or recombinant chemerin (20 nM) for 90 min at 37°C, 5% CO~2~. Cells were lysed and detection of total recruited β-arrestin was determined following the manufacturers protocol and as previously described ([@B51]). Luminescence measurements were taken using a PHERAstar microplate reader (BMG Labtech). Statistical Analysis {#S2-14} -------------------- All quantitative data are reported as mean ± SEM of *n* independent biological replicates. Statistical significance was assessed using a Student's unpaired *t*-test, one-way or two-way analysis of variance (ANOVA) with Dunnett's multiple comparison *post hoc* test (Prism 6 GraphPad Software, San Diego, CA, USA), *P* \< 0.05 was taken to be statistically significant. Results {#S3} ======= Chemerin Levels and *Ccrl2* Expression Are Increased during Acute Inflammation {#S3-1} ------------------------------------------------------------------------------ We and others have previously demonstrated that i.p. challenge with zymosan induces robust inflammatory mediator production in the peritoneum as well as inflammatory cell recruitment ([@B47]). Injection of 100 µg zymosan resulted in robust inflammatory cell recruitment with neutrophils peaking at 4 h post zymosan challenge and monocytes peaking at 8 h (Figure [1](#F1){ref-type="fig"}A). The role played by chemerin during inflammation remains incompletely understood. To interrogate this in our model of acute inflammation, we quantified total chemerin levels in the peritoneum during this acute inflammatory response and found that levels were significantly increased compared with naïve mice 4 h following zymosan challenge (2.5 ± 0.4 ng/ml at 4 h compared with 1.2 ± 0.3 ng/ml in naïve mice) (Figure [1](#F1){ref-type="fig"}B). One caveat using an ELISA to measure chemerin levels is that it is a quantification of total chemerin levels, including inactive pro-chemerin and potentially shorter bioactive chemerin isoforms generated during inflammation. To investigate chemerin bioactivity during this acute inflammatory response, we compared the ability of the peritoneal exudate fluid from animals injected with zymosan at different time points to activate the CMKLR1 chemerin receptor on CMKLR1 transfected CHO-K1 cells (Figure [1](#F1){ref-type="fig"}C). Using this bioassay, chemerin bioactivity progressively increased after zymosan injection peaking at 8 h and decreasing to baseline levels by 96 h (Figure [1](#F1){ref-type="fig"}C). These data suggest that although total chemerin levels as measured by ELISA peaked at 4 h and decreased rapidly by 8 h, chemerin bioactivity continued to increase up to 8 h and remained increased up to 72 h post zymosan injection. {#F1} The function of CCRL2 during inflammation remains to be fully elucidated, but the expression of CCRL2 has been reported to be increased during inflammation ([@B39]). To confirm and extend this observation, we FACS sorted resident peritoneal macrophages as well as recruited peritoneal neutrophils and monocytes from wild-type (WT) C57BL/6J mice following a 4 h zymosan challenge. We found that *Ccrl2* was expressed on three cell types but expression was highest on recruited neutrophils following zymosan challenge (Figure [1](#F1){ref-type="fig"}D). In addition, we cultured BMDMs and HUVECs and challenged them with a range of inflammatory stimuli (Figures [1](#F1){ref-type="fig"}E,F). Addition of zymosan, LPS, interferon-γ, poly I:C all increased BMDM expression of *Ccrl2* with the combination of LPS and interferon-γ resulting in the largest increase in expression (\~235-fold increase compared with vehicle) (Figure [1](#F1){ref-type="fig"}E). Zymosan and LPS also increased *Ccrl2* expression in HUVECs (Figure [1](#F1){ref-type="fig"}F). These results are in agreement with the published literature, indicating that *Ccrl2* expression is indeed rapidly upregulated on a number of cells during acute inflammation ([@B39], [@B40]). *Ccrl2*^−*/*−^ Mice Displayed Increased Monocyte and Neutrophil Recruitment to the Inflamed Peritoneum {#S3-2} ------------------------------------------------------------------------------------------------------ To investigate the potential functional consequence of upregulation of Ccrl2 during acute inflammation *in vivo*, we challenged WT mice or *Ccrl2*^−/−^ mice with 100 µg zymosan for 4 h and quantified inflammatory cell recruitment and mobilisation. Zymosan challenge resulted in significantly more total cells recruited to the peritoneum of *Ccrl2*^−^*^/^*^−^ mice compared with WT after 4 h (Figure [2](#F2){ref-type="fig"}B). There was also a striking increase in monocyte recruitment (0.4 ± 0.05 × 10^6^ in *Ccrl2*^−/−^ mice compared with 0.2 ± 0.03 × 10^6^ in WT mice) and neutrophil recruitment (6.2 ± 0.52 × 10^6^ in *Ccrl2*^−/−^ mice compared with 3.1 ± 0.35 × 10^6^ in WT mice) to the peritoneum of *Ccrl2*^−^*^/^*^−^ mice (Figures [2](#F2){ref-type="fig"}A--D). Other cell populations including CD4 T cells, CD8 T cells and B cells were also quantified but no significant differences were observed between the two genotypes (data not shown). {#F2} We next examined the blood from these animals 4 h post zymosan challenge to investigate whether the increased monocyte and neutrophil recruitment we observed at the site of inflammation was mirrored in the blood, indicating increased mobilisation of innate immune cells from bone marrow and spleen. Similarly to the peritoneum, there was a twofold increase in blood neutrophils in *Ccrl2*^−^*^/^*^−^ mice challenged with zymosan compared with WT (Figure [2](#F2){ref-type="fig"}E). There was also an increase in Ly6C^hi^ monocytes, although this was not significant (Figure [2](#F2){ref-type="fig"}F). In agreement with the results from the peritoneum, there were no differences in circulating B cells, CD4 T cells or CD8 T cells between these genotypes (data not shown). Since we observed increased neutrophil and monocyte numbers in the peritoneum and blood of *Ccrl2*^−^*^/^*^−^ mice following zymosan challenge, we investigated if the increased numbers of systemic leucocytes was associated with differences in levels in the bone marrow (Figures [2](#F2){ref-type="fig"}H--J). *Ccrl2*^−^*^/^*^−^ did not display any significant differences in total bone marrow neutrophils (Figure [2](#F2){ref-type="fig"}H) monocytes (Figure [2](#F2){ref-type="fig"}I) or B cells (Figure [2](#F2){ref-type="fig"}J). Collectively, our results suggest that the elevated neutrophil levels observed in the peritoneum are the result of increased initial systemic inflammatory responses of *Ccrl2*^−^*^/^*^−^ mice. Intriguingly, Ccrl2 appears to be primarily important in the initial stages of the acute inflammatory response. When we investigated later time points following zymosan challenge (48 h), we found no significant differences in neutrophil or monocyte numbers in the peritoneum or blood between *Ccrl2*^−/−^ mice and WT mice (Figures S2B,C,E,F in Supplementary Material). In addition, there were no significant differences in local or systemic chemerin levels at this late time point (Figures S2D,G in Supplementary Material). Steady State *Ccrl2*^−*/*−^ Mice Displayed No Obvious Alterations in Resident Leucocyte Populations in Any Tissues Examined {#S3-3} --------------------------------------------------------------------------------------------------------------------------- To establish whether the increased inflammatory cell recruitment observed in *Ccrl2*^−^*^/^*^−^ mice during zymosan challenge could be explained by increased circulating myeloid cells under resting conditions, we undertook a phenotypic analysis of the peritoneum, blood, bone marrow and spleen of unchallenged WT and *Ccrl2*^−^*^/^*^−^ mice (Figure [2](#F2){ref-type="fig"}). We found no differences in neutrophils or monocytes between the two groups (Figures [2](#F2){ref-type="fig"}C,D). There were no differences in Ly6C^hi^ monocytes, Ly6C^lo^ monocytes or neutrophils in the blood between the two groups (Figures [2](#F2){ref-type="fig"}E--G). Similarly, there were no differences in monocytes, neutrophils or B cells in the bone marrow (Figures [2](#F2){ref-type="fig"}H--J). Nor were there obvious differences in the spleens under homeostatic conditions (data not shown). Finally, we endeavoured to assay basal levels of CXCL1 and IL-6 in the peritoneal lavage fluid of these mice, but all samples tested were below the limit of detection (set at 15 pg/ml) of the assays (data not shown). These results indicate that the increased neutrophil and monocyte recruitment to the peritoneum seen following zymosan challenge were due to differences in initial inflammatory responses rather than a constitutive increase in circulating neutrophil and monocyte numbers. *Ccrl2*^−*/*−^ Mice Displayed Increased CXCL1 and Chemerin Levels at Early Time Points following Zymosan Challenge {#S3-4} ------------------------------------------------------------------------------------------------------------------ Having shown that mice lacking the CCRL2 chemerin receptor displayed exaggerated acute inflammatory responses following zymosan challenge, we next investigated if the increased monocyte and neutrophil recruitment was due to elevated chemokine or inflammatory mediator levels. There were no significant differences in mediator levels between the two genotypes at the 4 h time point (Table [1](#T1){ref-type="table"}). However, a number of mediators including CXCL1 (an important neutrophil chemoattractant) and IL-6 displayed rapid induction following zymosan challenge peaking at 2 h (Figures [3](#F3){ref-type="fig"}A,B). ###### Local mediators (pg/ml) produced in peritoneal exudate cell fluid following 4 h challenge with 100 µg zymosan i.p. as measured by Luminex. IL-6 CCL2 CCL3 CCL4 CXCL1 CXCL2 CXCL10 -------------- ----------- ------------ ------------- ------------ ------------ ----------- ---------- Wild type 759 ± 123 1,218 ± 81 34.50 ± 5.3 744.3 ± 41 78.6 ± 6.4 506 ± 166 620 ± 50 *Ccrl2*^−/−^ 705 ± 83 1,013 ± 21 26.40 ± 2.1 736.8 ± 32 127.8 ± 28 400 ± 148 525 ± 24 *Data are presented as mean ± SEM of n = 8 animals*. *Statistical significance was assessed by a Student's unpaired t-test, ns = P \> 0.05*. {#F3} Since CXCL1 and IL-6 levels peaked rapidly following zymosan insult, we challenged both *Ccrl2*^−/−^ mice and littermate controls with zymosan for 2 h and assessed the resulting inflammatory responses. Similar to the 4 h time point, we observed significantly more neutrophils recruited to the peritoneum in the *Ccrl2*^−^*^/^*^−^ mice compared with WT (Figure [3](#F3){ref-type="fig"}E). There were very few if any monocytes recruited to the peritoneum at this early time point (Figure [1](#F1){ref-type="fig"}B). We observed a twofold increase in CXCL1 levels in the peritoneum of these animals as well as significantly elevated chemerin levels (Figures [3](#F3){ref-type="fig"}F,G). Whilst we observed no obvious differences in monocyte or neutrophil numbers in the blood (Figures [2](#F2){ref-type="fig"}H,I), there was a threefold increase in CXCL1 plasma levels (Figure [3](#F3){ref-type="fig"}K). Chemerin levels were also significantly elevated in the blood of the *Ccrl2*^−/−^ mice compared with WT at the 2 h time point (Figure [3](#F3){ref-type="fig"}L). *Ccrl2*^−/−^ Mice Displayed Increased Neutrophil Numbers in the Peritoneum after Thioglycollate Challenge {#S3-5} --------------------------------------------------------------------------------------------------------- To explore if the increased neutrophil recruitment observed in the *Ccrl2*^−^*^/^*^−^ mice was also evident with other inflammatory stimuli, we challenged these mice i.p. with thioglycollate. Thioglycollate is a well-established inflammatory stimulus used to elicit inflammatory macrophages (after 4--5 days) ([@B45], [@B52]). However, it can also be used to interrogate neutrophil recruitment in the initial stages of inflammation. Following a 1 h challenge with 4% thioglycollate, we observed a small but distinct population of neutrophils recruited to the peritoneum (Figure [4](#F4){ref-type="fig"}A). We chose this time point because it represents an early stage of neutrophil recruitment in response to thioglycollate. *Ccrl2*^−^*^/^*^−^ mice displayed a twofold increase in neutrophil recruitment to the peritoneum compared with WT controls (0.07 ± 0.01 × 10^6^ neutrophils in WT compared with 0.13 ± 0.01 × 10^6^ neutrophils in *Ccrl2*^−/−^ mice) (Figure [4](#F4){ref-type="fig"}C). There were also significantly higher CXCL1 and chemerin levels in the peritoneum of the *Ccrl2*^−^*^/^*^−^ mice compared with WT mice at this early time point (Figures [4](#F4){ref-type="fig"}D,E). {#F4} When we examined the blood of these mice, there was no difference in neutrophil numbers between *Ccrl2*^−^*^/^*^−^ mice and WT controls (Figures [4](#F4){ref-type="fig"}F,G). This is perhaps not surprising given the early time point tested. There were, however, elevated chemerin levels in the blood of the *Ccrl2*^−^*^/^*^−^ mice compared with WT controls (Figure [4](#F4){ref-type="fig"}I). As we did not observe any differences in neutrophil numbers in the blood of these animals, it seems unlikely there would be changes in the bone marrow at this early time point but this was not assessed ([@B53]). Collectively, these observations demonstrate that mice lacking *Ccrl2* display exaggerated neutrophil recruitment to local sites of inflammation as well as elevated chemerin and CXCL1 levels irrespective of the stimulus used to elicit the response. Neutralisation of Endogenous Chemerin in *Ccrl2*^−*/*−^ Mice Abrogated the Exaggerated Inflammatory Phenotype {#S3-6} ------------------------------------------------------------------------------------------------------------- From our previous experiments, *Ccrl2*^−^*^/^*^−^ mice displayed increased myeloid cell recruitment in short-term models of acute inflammation. This was associated with elevated CXCL1 and chemerin levels. Given our group and others have identified the CCRL2 ligand chemerin, as an important modulator of inflammation and chemotaxis, it seemed plausible that the elevated levels of chemerin during acute inflammation in *Ccrl2*^−^*^/^*^−^ mice was responsible for the increased myeloid cell recruitment ([@B20], [@B21], [@B23]). To test this hypothesis, we used a blocking anti-chemerin antibody to neutralise endogenous chemerin levels in the *Ccrl2*^−^*^/^*^−^ mice before zymosan challenge (Figure [5](#F5){ref-type="fig"}). We first confirmed that this antibody could indeed block signalling at the Cmklr1 receptor. Using Cmklr1 transfected CHO-K1 cells, we demonstrated that preincubation with the antibody efficiently blocked chemerin induced β-arrestin recruitment to Cmklr1 (Figure [5](#F5){ref-type="fig"}A). We also tested the antibody in primary cells and confirmed that it effectively blocked chemerin-induced chemotaxis of biogel-elicited macrophages in a real-time chemotaxis system (Figure [5](#F5){ref-type="fig"}B). Following a 24 h pretreatment with isotype control or anti-chemerin antibody, *Ccrl2*^−/−^ mice were challenged for 4 h with zymosan as before (Figure [5](#F5){ref-type="fig"}C). Animals that received the anti-chemerin antibody displayed significantly less total leucocyte and neutrophil recruitment to the peritoneum compared with mice that received the isotype control (Figures [5](#F5){ref-type="fig"}D,E; Figure S1 in Supplementary Material). There was no significant effect on monocyte recruitment (Figure [5](#F5){ref-type="fig"}F). When we measured mediator levels in the peritoneum of these mice, we observed a twofold reduction in CXCL1 levels in the anti-chemerin treated mice compared with isotype control treated mice and no differences in IL-6 levels between the groups (Figures [5](#F5){ref-type="fig"}G,H). These results indicate that the elevated chemerin levels observed in *Ccrl2*^−^*^/^*^−^ mice contribute to the exaggerated leucocyte recruitment as well as the elevated CXCL1 levels observed in these mice during acute inflammation. {#F5} Chemerin, the Ligand for CCRL2, Increased Neutrophil Recruitment in WT Mice {#S3-7} --------------------------------------------------------------------------- Chemerin has previously been reported to induce the expression of various inflammatory cytokines and chemokines (including CXCL1 and CCL2) in epithelial and endothelial cells ([@B54], [@B55]). This provides a possible mechanism by which the higher chemerin levels in *Ccrl2*^−/−^ mice could cause increased CXCL1 levels during an inflammatory response. To further investigate the role of chemerin in exacerbating inflammation in our model, we pretreated WT mice with recombinant murine chemerin (4 µg/mouse) for 1 h before zymosan challenge (Figures [6](#F6){ref-type="fig"}A,B). Mice that received chemerin pretreatment displayed increased total cell recruitment to the peritoneum compared with mice that received PBS (4.5 × 10^6^ leucocytes in PBS treated animals compared with 7.5 × 10^6^ leucocytes in chemerin pretreated) (Figure [6](#F6){ref-type="fig"}C). Chemerin pretreated mice also displayed increased neutrophil recruitment to the peritoneum compared with PBS pretreated mice (2.5 ± 0.5 × 10^6^ neutrophils in zymosan alone compared with 4.5 ± 0.7 × 10^6^ in chemerin pretreated) (Figure [6](#F6){ref-type="fig"}D). We did not observe any differences in monocyte recruitment between the groups (Figure [6](#F6){ref-type="fig"}E). Importantly, injection of chemerin alone did not result in any neutrophil or monocyte recruitment (Figures [6](#F6){ref-type="fig"}B--E). When we analysed the inflammatory exudate from these mice, we found that chemerin pretreated mice had significantly higher levels of CXCL1 and IL-6 (Figures [6](#F6){ref-type="fig"}F,G). Mice that were treated with chemerin displayed a 2-fold and 2.6-fold increase in CXCL1 and IL-6 levels, respectively. We also observed similar effects using a lower dose of 0.5 µg chemerin per mouse as a pretreatment before zymosan challenge (data not shown). {#F6} Increased Recruitment of Neutrophils and Monocytes in *Ccrl2*^−*/*−^ Mice Was Not due to a Direct Effect of CCRL2 on Chemerin Induced Chemotaxis {#S3-8} ------------------------------------------------------------------------------------------------------------------------------------------------ One possible explanation for the observed differences in inflammatory cell recruitment was that lack of Ccrl2 may in someway alter the migratory behaviour of inflammatory cells in response to chemerin or other chemotactic ligands. To test this, we used Bio-gel elicited PECs, which we recently demonstrated to be a mixture of inflammatory macrophages and neutrophils ([@B50], [@B51]). We first demonstrated that Bio-gel elicited neutrophils do not express the chemerin Cmklr1 receptor (Figure [7](#F7){ref-type="fig"}A). Using the real-time chemotaxis platform, we demonstrated that deletion of CCRL2 had no appreciable effect on PEC migration towards chemerin, CCL5, CXCL1, or C5a, ruling out any direct effect of CCRL2 on cell migration (Figures [7](#F7){ref-type="fig"}B--E). {#F7} Discussion {#S4} ========== In this study, we have demonstrated for the first time that Ccrl2 plays a non-redundant role in dampening the recruitment of myeloid cells to local sites of acute inflammation. We report that Ccrl2 is expressed on resident peritoneal macrophages as well as recruited monocytes and neutrophils following zymosan challenge (Figure [1](#F1){ref-type="fig"}D). Interestingly, the highest levels of expression appear to be on recruited neutrophils and this is in agreement with a recent report by Del Prete et al. ([@B56]). In addition, *Ccrl2* expression on macrophages and endothelial cells is increased after exposure to inflammatory stimuli. This is in agreement with published reports (Figures [1](#F1){ref-type="fig"}E,F) ([@B39], [@B40]). After 4 h of zymosan challenge we observed that mice lacking expression of the Ccrl2 chemerin receptor displayed a twofold increase in monocyte and neutrophil recruitment to the peritoneum (Figures [2](#F2){ref-type="fig"}B,C). These effects of Ccrl2 were not localised only to the site of inflammation, as we also observed increased neutrophil numbers in the blood, suggesting more systemic effect (Figure [2](#F2){ref-type="fig"}E). These systemic effects were predominantly associated with neutrophil recruitment as we did not observe significant differences in monocyte numbers in the blood or bone marrow (Figures [2](#F2){ref-type="fig"}F,I). The fact that there were significant differences only in local monocyte numbers but not in systemic numbers may be a feature of the early time point evaluated here, as we know from our kinetic studies that monocyte numbers peak later (Figure [1](#F1){ref-type="fig"}B). Importantly, at an earlier time point (2 h) in which we could more easily evaluate cytokine and chemokine levels, we observed that mice lacking Ccrl2 displayed higher levels of CXCL1 in the peritoneum (Figure [3](#F3){ref-type="fig"}F). The elevated levels of CXCL1 both locally and systemically would explain why the majority of differences we observe in the *Ccrl2*^−/−^ mice were in the context of neutrophil migration. This phenotype was consistently associated with higher endogenous levels of chemerin in *Ccrl2*^−/−^ mice and was abrogated after blocking chemerin activity with an anti-chemerin antibody (Figure [5](#F5){ref-type="fig"}). Furthermore, this phenotype was recapitulated in WT mice by injection of recombinant murine chemerin, further indicating a role for chemerin in driving this exaggerated neutrophil recruitment (Figure [6](#F6){ref-type="fig"}). Finally, this exaggerated inflammatory phenotype of *Ccrl2*^−/−^ mice appears to be predominantly associated with the initial stages of acute inflammatory responses. Absence of Ccrl2 did not appear to affect neutrophil or monocyte numbers locally or systemically at later time points (Figure S2 in Supplementary Material). Ccrl2 is one of the less studied chemerin receptors and arguably performs the least obvious function. It binds chemerin but there is no reported downstream signalling upon receptor ligation in primary cells ([@B38], [@B39]). Whilst it was initially thought to serve a similar function to the decoy receptors ACKR1 (or DARC) and ACKR2 (or D6), which dampen inflammation by binding and internalising inflammatory chemokines, Ccrl2 does not internalise chemerin or any other chemokines ([@B38], [@B57]). Whilst the exact role of Ccrl2 during inflammation has yet to be fully elucidated, its expression has been documented by a number of groups including in this report (Figures [1](#F1){ref-type="fig"}E,F) to be rapidly upregulated during inflammation, suggesting a conserved function required during inflammatory responses ([@B39], [@B40]). Several studies have endeavoured to clarify the role played by Ccrl2 *in vivo* but there have been conflicting results in different experimental models. Zabel et al. demonstrated that mice lacking Ccrl2 displayed reduced ear swelling and leucocyte influx in a mast cell dependent model of atopic allergy. The exact mechanism by which Ccrl2 modulated these responses was not clear. However, cells expressing Ccrl2 were shown to bind and concentrate chemerin locally *in vitro*. When these cells were incubated with *Cmklr1* transfected cells, they induced calcium flux, indicating a possible role for Ccrl2 in chemerin presentation ([@B38]). In the *in vitro* model, Ccrl2 expressing cells were capable of binding chemerin and presenting it to Cmklr1 expressing cells when in close proximity. It was postulated this mode of action played a role in mast cell responses to low dose IgE challenge. However, this seems unlikely to be the case in our model of acute inflammation. Administration of recombinant bioactive chemerin to WT mice exacerbated the acute inflammatory response (Figure [6](#F6){ref-type="fig"}). If Ccrl2 were functioning to increase chemerin signalling, which appears to be pro-inflammatory here, one would expect to observe reduced inflammation following zymosan challenge in *Ccrl2*^−/−^ mice. Yet in Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}, we observed the converse. An alternative hypothesis, therefore, is that *in vivo*, increased expression of Ccrl2 on other cell types during inflammation such as vascular endothelial cells may enable binding of free chemerin, which in turn reduces systemic levels of chemerin ([@B39]). This would thereby decrease the chemerin available to interact with Cmklr1 expressing cells and therefore dampen the ensuing pro-inflammatory responses elicited by chemerin as described by Neves et al. ([@B55]). Mazzon et al. reported that mice lacking Ccrl2 displayed exacerbated disease in a model of experimental autoimmune encephalitis (EAE) as well as increased *Ccrl2* expression on mononuclear cells in WT mice during EAE ([@B58]). The authors reported that in mice lacking Ccrl2, T cells, and macrophages were further polarised towards an inflammatory phenotype during disease, but the mechanism by which deletion of the *Ccrl2* gene enhanced disease remained unexplained ([@B58]). At the time of writing, the same group more recently reported a new study in which mice lacking Ccrl2 were protected in two murine models of arthritis in contrast to their previous study using the EAE model, highlighting the complexity of the chemerin/Ccrl2 axis ([@B56], [@B58]). Importantly, the authors report that the protective effect of deletion of *Ccrl2* in these models was due to defects in neutrophil recruitment and trafficking due to CXCR2 heterodimersastion with Ccrl2. These results seem to be at odds with our current findings but possibly reflect differences in the disease models used to interrogate this biology. The kinetics and disease pathology differ significantly between the acute models of inflammation used in our study and the more chronic disease models used by Del Prete et al. in their most recent report ([@B56]). Clearly, further studies will be necessary to clarify the exact role played by Ccrl2 during inflammation but it seems likely to be disease and even tissue specific ([@B59]). In agreement with an earlier study by Monnier et al. who reported higher chemerin levels in the blood of *Ccrl2*^−/−^ mice following intranasal administration of LPS, we observed elevated chemerin levels in the blood of *Ccrl2*^−/−^ animals following zymosan and thioglycollate challenge (Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}) ([@B39]). In addition, using BMDMs from WT animals and HUVECs, we observed increased expression of *Ccrl2* mRNA after stimulation with inflammatory stimuli (up to \~230-fold increase on BMDMs) (Figure [1](#F1){ref-type="fig"}E). We also report increased neutrophil recruitment and CXCL1 levels both locally and systemically (Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}) in *Ccrl2*^−^*^/^*^−^ mice challenged with 100 µg zymosan. To the best of our knowledge, this is the first report of such an observation. The increased neutrophil recruitment during acute inflammation in these mice can be at least partly explained by increased CXCL1 and chemerin levels. Previous studies from our lab and others have not detected the chemotactic chemerin receptor Cmklr1 on murine neutrophils ([@B23], [@B56], [@B60]). There has since been one report of murine neutrophils expressing Cmklr1; however, we failed to detect any Cmklr1 expression on murine neutrophils (Figure [7](#F7){ref-type="fig"}A) ([@B35]). It seems unlikely, therefore, that the increased neutrophil recruitment we observed in our model is due to any direct chemotactic effects of chemerin on neutrophils. We did not observe any differences in migration between WT and *Ccrl2*^−/−^ cells in response any mediators tested in agreement with Del Prete et al. (Figures [7](#F7){ref-type="fig"}C,D) ([@B56]). Importantly, the phenotype observed in *Ccrl2*^−^*^/^*^−^ mice is not related to the receptors that detect zymosan (dectin-1, TLR2/6) as we observed a similar phenotype with thioglycollate challenge, which is a more severe inflammatory insult ([@B61]). Our results support the hypothesis that Ccrl2 is important for regulating both local and systemic chemerin levels during an acute inflammatory response. This exacerbated inflammatory response was not simply a feature of increased basal myeloid cell numbers, as we observed no differences in leucocyte populations between the two groups in any tissue tested under steady state conditions (Figure [2](#F2){ref-type="fig"}). Rather, the increased myeloid cell recruitment in *Ccrl2*^−^*^/^*^−^ mice was associated with increased chemerin and CXCL1 levels both locally and systemically. Chemerin has previously been reported to induce pro-inflammatory signalling in microvascular endothelial and smooth muscle cells ([@B55]). Chemerin treatment increased mRNA expression of a number of pro-inflammatory mediators including CCL2, TNF-α, and VCAM-1 ([@B55]). Another study by Lin et al. demonstrated that chemerin administration to WT mice induced more severe inflammation in a model of DSS-induced colitis, which was characterised by increased inflammatory cytokines and an inhibition of M2 polarisation of resident macrophages ([@B62]). Taken together, these reports support our hypothesis that increased chemerin levels observed in the *Ccrl2*^−^*^/^*^−^ mice were responsible for increased myeloid cell recruitment *via* the increased production of inflammatory mediators such as CXCL1 during sterile peritonitis. We confirmed the importance of chemerin in the phenotype of *Ccrl2*^−^*^/^*^−^ mice using an anti-chemerin polyclonal blocking antibody to inhibit chemerin signalling in these animals before zymosan challenge (Figure [5](#F5){ref-type="fig"}). Neutralisation of endogenous chemerin in the *Ccrl2*^−^*^/^*^−^ mice decreased the total leucocytes and neutrophils recruited to the peritoneum compared with isotype controls, indicating that elevated chemerin levels did indeed play a role in the exaggerated neutrophil recruitment observed in these animals (Figures [5](#F5){ref-type="fig"}C,D). We previously observed that blockade of endogenous chemerin in WT mice resulted in increased neutrophil and monocyte recruitment to the peritoneum of WT mice following challenge with 10 µg zymosan ([@B34]). In our current study, we did not interrogate chemerin blockade in WT mice but when chemerin activity was blocked in *Ccrl2*^−/−^ mice during 100 µg zymosan challenge, we observed decreased neutrophil recruitment compared with isotype control treated *Ccrl2*^−/−^ mice. These data highlight further differences in chemerin behaviour depending on the intensity of inflammatory insult and genetic backgrounds used. From the results in our current study, it is plausible that chemerin may be more important for driving neutrophil rather than monocyte recruitment in this acute model of inflammation, as chemerin blockade had no significant effect on monocyte numbers. However, it is known that neutrophils are important for the recruitment of monocytes during an inflammatory response and are capable of secreting a number of monocyte chemoattractants ([@B63], [@B64]). Chemerin bioactivity may not have been completely blocked in these animals, hence although we observed a significant decrease in neutrophil recruitment, this may not been sufficiently blunted to in turn appreciably decrease monocyte recruitment at this time point. When we interrogated mediator levels, we observed a significant decrease in local levels of CXCL1 but no change in IL-6 following blockade of endogenous chemerin (Figures [5](#F5){ref-type="fig"}F,G). CXCL1 and IL-6 were evaluated as CXCL1 is a key driver of neutrophil recruitment and IL-6 is a systemic marker of inflammation ([@B65]--[@B68]). As predicted from our results using chemerin blocking antibodies, chemerin pretreatment of WT mice before zymosan challenge increased total cell and neutrophil recruitment to the peritoneum as well as elevated levels of inflammatory chemokines and cytokines, similar to what we observed in the *Ccrl2*^−^*^/^*^−^ mice (Figure [6](#F6){ref-type="fig"}). Importantly, injection of chemerin alone did not induce any neutrophil recruitment after 4 h at the dose used (Figures [6](#F6){ref-type="fig"}B--E). We demonstrate here that full-length bioactive chemerin can increase neutrophil recruitment when administered before an inflammatory stimulus but previous studies from our laboratory have reported that a synthetic chemerin-derived peptide called C15 has anti-inflammatory effects in a similar model (albeit using 10-fold lower dose of zymosan as an inflammatory stimulus) ([@B34]). The most likely explanation for this apparent discrepancy is differences in the pharmacology and downstream signalling between the full-length chemerin protein and the 15 amino acid chemerin peptide. The signalling cascades that are activated at the murine Cmklr1 receptor upon chemerin binding for example, are relatively well characterised (involving the MAP kinase, RhoA, ROCK, MEK1/2, and P38 proteins amongst others) ([@B23], [@B26]). However, very little if anything is known about the signalling pathways induced by the C15 peptide. Whilst chemerin is known to act as a potent chemoattactant for macrophages, DCs, and NK cells, C15 does not exhibit similar effects on any cell types investigated thus far ([@B20], [@B22], [@B23], [@B30], [@B50]). Clearly, the full-length protein and the smaller peptide elicit quite different intracellular signalling pathways downstream of the Cmklr1 chemerin receptor and this would suggest they play quite different roles *in vivo*. We have endeavoured to confirm the requirement for CXCL1 in the exaggerated neutrophil recruitment in *Ccrl2*^−/−^ mice by pretreating WT and *Ccrl2*^−/−^ mice with a blocking anti-CXCL1 antibody before zymosan challenge. However, we did not observe significant decreases in neutrophil recruitment in either group despite previous reports (data not shown) ([@B65]). However, these results are perhaps not surprising given the high affinity of these chemokines for their receptors as well as the functional redundancy in the system ([@B69]). Hence, we have demonstrated that chemerin is elevated in *Ccrl2*^−/−^ mice and that elevated chemerin is capable of driving the exaggerated neutrophil recruitment and increased levels of CXCL1 during acute inflammation. However, we cannot exclude the possibility that other mediators or pathways are also involved. Further studies will be required to fully clarify this. In summary, we have demonstrated that the non-signalling chemerin receptor Ccrl2 serves a non-redundant role in dampening acute inflammatory responses *in vivo*. The absence of Ccrl2 resulted in exaggerated acute inflammatory responses as well as elevated CXCL1 and chemerin levels. Blockade of endogenous chemerin in *Ccrl2*^−^*^/^*^−^ mice abrogated the elevated neutrophil recruitment. We observed that administration of recombinant murine chemerin to WT mice induced similar exaggerated inflammatory responses to those seen in *Ccrl2*^−^*^/^*^−^ mice. Importantly, deletion of the *Ccrl2* gene did not have any direct effect on myeloid cell chemotaxis towards chemerin or other chemokines. Our data are consistent with a model in which Ccrl2 serves to bind and maintain chemerin levels below a pathological threshold during acute inflammation and the more severe inflammatory responses observed in *Ccrl2*^−^*^/^*^−^ mice are due to significantly elevated free chemerin levels. The increased chemerin signalling in turn induces higher production of inflammatory chemokines such as CXCL1, which results in elevated myeloid cell recruitment and more severe inflammation. Our experiments suggest that chemerin could be a therapeutic target in the treatment of inflammatory diseases, particularly RA, in which chemerin has consistently been implicated in the pathology of this disease ([@B16], [@B19], [@B54], [@B70]). Ethics Statement {#S5} ================ All animal studies were conducted with ethical approval from the Dunn School of Pathology Local Ethical Review Committee and in accordance with the UK Home Office regulations (Guidance on the Operation of Animals, Scientific Procedures Act, 1986). Author Contributions {#S6} ==================== DR-K, SV, CR, LT, TK, DG, and AI performed experiments; DR-K, SV, and AI analysed results and made the figures; DR-K, SV, AI, and DG designed the research and wrote the paper. All the authors provided critical revision of the manuscript. Conflict of Interest Statement {#S7} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors thank Linda Randall and Nicky Hamp for excellent technical assistance. **Funding.** This work was supported by British Heart Foundation grants (FS/11/82/29332, PG/10/6028496, RE/13/1/30181) and MRC Industrial CASE Studentship (MR/K017160/1). Supplementary Material {#S8} ====================== The Supplementary Material for this article can be found online at <http://www.frontiersin.org/article/10.3389/fimmu.2017.01621/full#supplementary-material>. ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Fulvio D'Acquisto, Queen Mary University of London, United Kingdom [^2]: Reviewed by: Francesco Maione, University of Naples Federico II, Italy; Vily Panoutsakopoulou, Biomedical Research Foundation of the Academy of Athens, Greece [^3]: ^†^These authors share first authorship. [^4]: ^‡^These authors share senior authorship. [^5]: Specialty section: This article was submitted to Inflammation, a section of the journal Frontiers in Immunology

INTRODUCTION {#sec1-1} ============ *Gynura* genus belongs to the family Asteraceae, consisting of 12 species in China.\[[@CIT1]\] Many species are edible medicinal plants and the leaves are also used as a vegetable by the locals in Southwestern China.\[[@CIT2]\] *G. divaricata* is a traditional Chinese medicinal plant, which is called "Bai Bei San Qi" in Chinese. It has a long history of use for treatment of diabetes in the folk medicine. The ethanol extract of aerial parts of *G. divaricata* was reported to demonstrate hypoglycemic activity in vivo, the flavonoid compounds were the active constituents.\[[@CIT3][@CIT4]\] It also has been reported that many constituents with antiproliferation activity exist in *G. divaricata*.\[[@CIT5][@CIT6]\] The chemical constituents of *G. divaricata* include flavonols, phenolic acids, cerebrosides, polysaccharides, alkaloids, terpenoids, and sterols.\[[@CIT5]--[@CIT10]\] Flavonols were the principal constituents of the plant, 4 flavonol compounds, including quercetin, isoquercitrin, rutin, and kaempferol-3-O-rutinoside, have been isolated and identified from the aerial parts of the plant.\[[@CIT9]\] This article herein describes the isolation and structure elucidation of the flavonol and phenolic acid compounds from the ethanol extract of *G. divaricata* DC. leaves by NMR and high-performance liquid chromatography-diode array detector-electrospray ionization-mass spectrometry (HPLC-DAD-ESI-MS) methods. MATERIALS AND METHODS {#sec1-2} ===================== General {#sec2-1} ------- The^1^H-NMR and^13^C-NMR spectra were measured with a Bruker Avance-600 FT-NMR spectrometer (Bruker, Coventry, Germany), with TMS internal standard. HPLC-DAD-ESI-MS were recorded on Waters 2995 Series LC and ZQ-4000 Mass spectrometer (Waters Corporation, Milford, MA, USA). Column chromatography was carried out with Silica gel (Qingdao Marine Chemistry Co. Ltd., 200-300 mesh, Qingdao, China), Sephadex LH-20, and Reverse phase octadecylsilyl (RP-ODS) (Pharmacia Co. Ltd., Minnesota, USA). Thin layer chromatography (TLC) was carried out with Silica gel GF~254~ (Qingdao Marine Chemistry Co. Ltd., Qingdao, China), and the compounds were prepared either by spraying with 10% sulfuric acid ethanol or under UV lamp at 254 nm. HPLC-grade acetonitrile was purchased from Merck Company (Merck, Darmstadt, Germany), other solvents were analytical grade from Sinopharm Chemical Reagent Co. Ltd. (Shanghai, China). Plant material {#sec2-2} -------------- The *Gynura divaricata* plant was obtained in 2009 from Guangdong province, China. A voucher specimen (201001) was deposited at the Department of Chemistry, Nanchang University. The leaves of *G. divaricata* were dried at 40°C in an air oven and finely powdered. Extraction and isolation {#sec2-3} ------------------------ The weighed portion of the crude drug 5 kg was extracted twice with 60% ethanol (v/v) under reflux at 90°C. The extract was evaporated to dryness *in vacuo*. Extract yield with respect to the dried herb was 25%. The dry extract was suspended in water and subjected to sequential liquid-liquid extraction with chloroform, ethyl acetate (EA), and *n*-butanol, the yield of those 3 extracts were 31.2, 56.5, and 89.5 g, respectively. The EA fraction was chromatographed using flash column on a Silica gel eluted with chloroform-methanol step-gradient (starting with 100:0 to 4:1), eluted fractions were combined on their TLC pattern to yield 8 fractions. The chloroform-methanol fraction (10:1) was chromatographed on a Sephadex LH-20 column eluted with chloroform-methanol (1:1) to yield compounds 1 and 2. The chloroform-methanol fraction (6:1) chromatographed on a Sephadex LH-20 column eluted with methanol and further chromatographed on an RP-ODS column gradient eluted with methanol-water (40%-60%, v/v) gave compounds 3 and 7. The chloroform-methanol fraction (4:1) chromatographed on a Sephadex LH-20 column eluted with methanol yields compound 4 \[[Figure 1](#F0001){ref-type="fig"}\]. {#F0001} The *n*-butanol fraction was chromatographed using flash RP-ODS column gradient eluted with methanol-water (10%-50%, v/v), and the eluted fractions were combined on their HPLC pattern to yield 4 fractions. The methanol-water fraction (25%, v/v) was further chromatographed using flash RP-ODS column and isocratic eluted with methanol-water (18%, v/v) gave compounds 5 and 6. The other minor constituents of *n*-butanol extracts were separated and identified by HPLC-DAD-ESI-MS method. HPLC-MS instrument and conditions {#sec2-4} --------------------------------- The HPLC-DAD-ESI-MS system consists of a Waters 2995 Series LC and ZQ-4000 Mass spectrometer (Waters, USA), equipped with a vacuum degasser, a quaternary pump, an autosampler, a thermostatted column compartment, a diode array detector (DAD), and an ion-trap mass spectrometer with electrospray ionization interface, controlled by Waters 2995 Series LC/ZQ-4000 Trap Software. Shimadzu shimpack VP-ODS (150 mm × 4.6 mm i.d., 5 μm particle size) was used for separation. Solvents for the mobile phase were water-0.1% acetic acid (A) and acetonitrile (B). The gradient elution was 0-30 min, linear gradient 10%-30% B; 30-40 min, linear gradient 30%-100% B. The flow rate was 0.8 mL/min and the column was operated at 30°C. Peaks were detected with the DAD at 254 nm. The ESI negative and positive ionization (NI and PI) total ion current (TIC) modes were used for MS detection. The *m/z* values of the monitored ions were from 100 to 800. The other parameters were as follows: capillary voltage, 3.5 kV; cone voltage, 30 V; extraction voltage, 5 V; RF voltage, 0.5 V; source temperature, 90°C; nitrogen gas flow for desolvation, 300 L/h; and temperature of the nitrogen gas for desolvation, 350°C. Samples for assay were dissolved in 45% MeOH as 3 mg/mL solutions and centrifuged at 12,000 rpm (Beckman, USA) for 15 min to remove particles before injection. RESULTS AND DISCUSSION {#sec1-3} ====================== The compounds were identified using UV, ESI-MS, and NMR spectral data, and determined as quercetin,\[[@CIT11][@CIT12]\] kaempferol,\[[@CIT13][@CIT14]\] kaempferol-3-O-*β*-D-glucopyranoside,\[[@CIT15]\] quercetin-3-O-rutinoside,\[[@CIT15]\] kaempferol-3-O-rutinoside-7-O-*β*-D-glucopyranoside,\[[@CIT16][@CIT17]\] kaempferol-3,7-di-O-*β*-D-glucopyranoside,\[[@CIT16]\] and 3,5-Dicaffeoylquinic acid.\[[@CIT18]\] Compound 1 was obtained as a yellow powder, the ESI-MS yielded a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 301 and \[M+H\]^+^ at *m/z* 303. The UV spectrum showed λ~max~ at 256 and 370 nm. The^1^H-NMR spectrum showed 2 peaks at δ 6.18 (1H, d, *J* = 2.0 Hz) and 6.40 ppm (1H, d, *J* = 2.0 Hz) consistent with the meta protons H-6 and H-8 on A-ring and an ABX system at 7.68 (1H, d, *J* = 2.2 Hz, H-2'), 7.54 (1H, dd, *J* = 2.0 Hz, 8.4 Hz, H-6'), and 6.88 (1H, d, *J* = 8.4 Hz, H-5') corresponding to the catechol protons on B-ring. The^13^C-NMR spectrum indicated the presence of 15 carbon atoms, the signal at δ 177.9 was attributed to a carbonyl carbon placed at C-4, and the other signals were compatible with those literatures\[[@CIT11][@CIT12]\] on quercetin. Compound 2 was obtained as a yellow powder, the ESI-MS yielded a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 285 and \[M+H\]^+^ at *m/z* 287. The UV spectrum showed λ~max~ at 265 and 366 nm. The^1^H-NMR spectrum showed 2 peaks at δ 6.17 (1H, d, *J* = 1.8 Hz) and 6.42 ppm (1H, d, *J* = 1.8 Hz) consistent with the meta protons H-6 and H-8 on A-ring and an AA'BB' system at 8.04 (2H, d, *J* = 8.9 Hz, H-2', 6') and 6.93 (2H, d, *J* = 8.9 Hz, H-3', 5') corresponding to the protons on B-ring. The MS and^1^ H-NMR data were compatible with the literatures\[[@CIT13][@CIT14]\] of kaempferol. Compound 3 was obtained as a faint yellow powder, the ESI-MS yielded a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 447 and \[M+H\]^+^ at *m/z* 449. The UV spectrum showed λ~max~ at 265 and 346 nm. The^1^ H-NMR spectrum showed 2 peaks at δ 6.21 (1H, d, *J* =1.8 Hz) and 6.44 ppm (1H, d, *J* =1.8 Hz) consistent with the meta protons H-6 and H-8 on A-ring and an AA'BB' system at 8.04 (2H, d, *J* =8.9 Hz, H-2', 6') and 6.89 (2H, d, *J* =8.9 Hz, H-3', 5') corresponding to the protons on B-ring. Compound 3 presented the same aglycone signal patterns of compound 2, but the signal at 5.47 (1H, d, *J* =7.2 Hz) followed by other characteristic additional signals indicates the presence of a sugar moiety in compound 3. The hexose was determined to be a glucopyranosyl unit bound to the C-3 position of the aglycone by comparison of proton and carbon upfield shift values with the literature data.\[[@CIT15]\] Therefore, compound 3 was identified as kaempferol-3-O-*β*-D-glucopyranoside. Compound 4 was obtained as a faint yellow powder, the ESI-MS yielded a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 609 and \[M+H\]^+^ at *m/z* 611. The UV spectrum showed λ~max~ at 258 and 356 nm. The^1^H-NMR spectrum showed 2 peaks at δ 6.20 (1H, d, *J* = 2.0 Hz) and 6.40 ppm (1H, d, *J* = 2.0 Hz) consistent with the meta protons H-6 and H-8 on A-ring and an ABX system at 7.54 (1H, d, *J* = 2.2 Hz, H-2'), 7.59 (1H, dd, *J* = 2.0 Hz, 9.0 Hz, H-6') and 6.85 (1H, d, *J* = 9.0 Hz, H-5') corresponding to the catechol protons on B-ring. Compound 4 presented the same aglycone signal patterns of compound 1, two anomeric proton signals at 5.32 (1H, d, *J* =7.2 Hz) and 4.39 (1H, d, *J* = 1.6 Hz) were assignable to H-1 of a *β*-glucosyl proton and to the H-1 of an *α*-rhamnosyl proton, respectively. A methyl signal 0.99 (3H, d, *J* =6.2 Hz) in the high-field region was assigned to rhamnose. In the^13^C-NMR of compound 4, the C-6 signal (68.5) of glucose showed a downfield shift of 7.3 ppm in comparison with the corresponding C-6 signal (61.2) of quercetin-3-O-*β*-D-glucopyranoside,\[[@CIT15]\] indicating a 1-6 linkage between the glucose and the rhamnose. Therefore, compound 4 was identified as rutin. Compound 5 was obtained as a faint yellow powder, the molecular formula C~27~H~30~O~16~ was suggested by a mass spectrum with a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 609, further confirmed by the positive mode mass spectral ions: 611 \[M+H\]^+^, 449 \[M+H-162\]^+^, 287 \[M+H-162-162\]^+^. The UV spectrum showed λ~max~ at 264 and 347 nm typical of a kaempferol glycoside derivative.\[[@CIT15][@CIT16][@CIT19]\] In the aromatic region of the^1^ H-NMR spectrum an AA'BB' system, appearing as two doublets at δ 8.06 (2H, d, *J* = 8.9 Hz, H-2', 6') and 6.90 (2H, d, *J* = 8.9 Hz, H-3', 5'), and two meta coupled doublet protons at δ 6.78 and 6.44 were evident. In the saccharide region of the spectrum two anomeric proton signals were present as large doublets at δ 5.48 and 5.08. The coupling constant (*J* = 7.2 Hz) of the two anomeric protons characteristic for *β*-configuration. The downfield shift of the H-6 and H-8 proton, as well as downfield shift of the corresponding carbons at δ 99.8 and δ 94.9, with respect to the corresponding signals of aglycone, suggested the linkage with the sugar moiety across the oxygen of the C(7)-OH group.\[[@CIT17]\] The chemical shift (δ 5.48) suggested that the other sugar moiety is directly attached to the C(3)-OH group, further confirmed by the upfield shift of the signal assigned to C-3 (133.9).\[[@CIT15][@CIT16]\] Acid hydrolysis of compound 5 afforded kaempferol and glucose comparison with the authentic samples on TLC. From the above data, compound 5 was identified as kaempferol-3,7-di-O-*β*-D-glucopyranoside. Compound 6 was obtained as a faint yellow powder, the molecular formula C~33~H~40~O~20~ was suggested by a mass spectrum with a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 755. The UV and^1^ H-NMR spectrum of compound 6 was similar to that of 5, suggesting that compound 6 also was a kaempferol glycoside derivative, the only difference being the presence of a methyl signal (δ 0.99) in the high-field region, which was assigned to rhamnose, further confirmed by the doublet proton at δ 4.44, was assigned to the anomeric proton of rhamnose with a coupling constant (*J* = 1.6 Hz) characteristic for *α*-linked rhamnose. The^13^C-NMR spectrum of 6 confirms that compound 6 is a triglycoside of kaempferol \[[Table 1](#T0001){ref-type="table"}\]. Careful examination of the^13^C-NMR spectrum of 6 showed that the signal assigned to the glucose C-6 \[[Table 1](#T0001){ref-type="table"}\] was shifted downfield by appropriately 6 ppm (from 61.3 to 67.3) confirming that the rhamnose moiety linkage to the glucose C-6.\[[@CIT17]\] From the above data, compound 6 was identified as kaempferol-3-O-rutinoside-7-O-*β*-D-glucopyranoside. ###### The 1H-NMR and 13C-NMR spectrum data of kaempferol-3,7-di-O-*β*-d-glucopyranoside and kaempferol-3-O-rutinoside-7-O-*β*-d-glucopyranoside (DMSO-d6) Atom Kaempferol-3,7-di-O-*β*-d-glucoside Kaempferol-3-O-rutinoside-7-O-*β*-d-glucoside ---------------- ------------------------------------- ----------------------------------------------- ------- -------- ----- ------- 2 156.5 156.5 3 133.9 140.0 4 178.1 178.1 4a 105.9 161.4 5 161.3 106.1 6 6.44 d 2.0 99.8 6.45 d 2.0 99.8 7 163.3 163.4 8 6.78 d 2.0 94.9 6.76 d 2.0 95.1 8a 157.3 157.8 1' 121.1 121.2 2' 8.06 d 8.6 131.4 8.01 d 8.8 131.5 3' 6.90 d 8.6 115.6 6.90 d 8.8 115.6 4' 160.6 160.6 5' 115.6 115.6 6' 131.4 131.5 3-O-Rutinoside G1 5.48 d 7.2 101.2 5.35 d 7.2 101.7 G2 74.6 74.7 G3 76.9 76.9 G4 70.4 70.4 G5 78 77.7 G6 61.3 67.3 R1 4.44 d 1.6 101.2 R2 70.8 R3 71.1 R4 72.3 R5 68.7 R6 0.99 d 6.2 18.2 7-O-Glucoside G'1 5.08 d 7.2 100.2 5.08 d 7.2 100.3 G'2 73.5 73.6 G'3 76.9 76.3 G'4 70.0 70.1 G'5 77.6 76.9 G'6 61.1 61.1 Compound 7 was obtained as amorphous powder, the ESI-MS yielded a quasi-molecular ion peak \[M-H\]^-^ at *m/z* 515 and \[M+H\]^+^ at *m/z* 517. The UV spectrum showed λ~max~ 327, 294 (sh), and 248 nm (sh), which were characteristic of caffeic acid derivatives. In the^1^ H-NMR spectrum, two caffeoyl groups were presented at δ 7.50 (1H, d, *J*=16.0 Hz, H-7'), 7.43 (1H, d, *J*=16.0 Hz, H-7"), 7.05 (2H, brs, H-2', 2"), 7.01 (2H, brd, *J*=2.0 Hz, H-6, 6"), 6.78 (1H, d, *J*=8.0 Hz, H-5'), 6.76 (1H, d, *J*=8.0 Hz, H-5"), 6.26 (1H, dd, *J*=16.0 Hz, H-8'), 6.14 (1H, dd, *J*=16.0 Hz, H-8"). A quinic acid moiety was presented at 5.42 (1H, brs, H-3), 5.18 (1H, m, H-5), 3.86 (1H, brs, H-4), 2.20 (2H, m, H-6), 2.01(2H, m, H-2). The^1^H-NMR data were in agreement with the literature\[[@CIT18]\] and compound 7 was identified as 3,5-Dicaffeoylquinic acid. An HPLC-DAD-ESI-MS method was developed to identify the minor phytochemical constituents of *n*-butanol fraction of *G. divaricata* extract. The chromatogram of MS TIC in negative mode is shown in [Figure 2a](#F0002){ref-type="fig"}. As shown in [Figure 2b](#F0002){ref-type="fig"}, 13 major peaks were detected under the HPLC conditions with DAD detection at 254 nm. Peaks of 2, 3, and 11, 12 were co-eluted in the present conditions and unequivocally determined to be kaempferol-3,7-di-O-*β*-D-glucopyranoside, kaempferol-3-O-rutinoside-7-O-D-glucopyranoside, 3,5-Dicaffeoylquinic acid, and kaempferol-3-O-*β*-D-glucopyranoside, respectively. And peak 5 was identified as quercetin-3-O-rutinoside. All of those 5 peaks were identified by comparing the retention time (RT), UV \[[Figure 3](#F0003){ref-type="fig"}\], and ESI-MS values with isolation compounds. The other compounds were tentatively identified based on the UV adsorption value, *m/z* value, and elution order compared with the published data. {#F0002} {#F0003} Peak 1 was believed to be an unidentified minor flavonol glycoside due to its low concentration in the extract, peak 1 and 3 are a pair of isomers, the UV (λ~max~) and *m/z* values \[[Table 2](#T0002){ref-type="table"}\] were similar to peak 3 (identified as kaempferol-3-O-rutinoside-7-O-*β*-D-glucopyranoside). The elution order of peak 1 being prior to peak 3 \[[Table 2](#T0002){ref-type="table"}\] suggested that rutinose of peak 3 was substituted by a robinobiose, and the structure of peak 1 was proposed to be kaempferol-3-O-robinobioside-7-O-*β*-D-glucopyranoside.\[[@CIT20]\] ###### HPLC-DAD-ESI-MS (positive and negative ionization TIC modes) fingerprint of *n*-butanol fraction of *G. divaricata* extracts Peak No. *t*~R~ (min) *λ*~max~(nm) Product ions (ESI-, *m/z*) Product ions (ESI+, *m/z*) Identification of compounds ---------- -------------- -------------- ----------------------------------- ------------------------------------------------------------------------------------- -------------------------------------------------- 1 15.22 265, 346 755 \[M-H\]^-^ Kaempferol-3-O-robinobioside-7-O-*β*-D-glucoside 2 16.43 264, 347 609 \[M-H\]^-^ 611 \[M+H\]^+^ 449 \[M+H-162\]^+^ 287 \[M+H-162-162\]^+^ Kaempferol-3,7-di-O-*β*-D-glucoside 3 16.51 264, 347 755 \[M-H\]^-^ 757 \[M+H\]^+^ 611 \[M+H-146\]^+^ 449 \[M+H-146-162\]^+^ 287 \[M+H-146-162-162\]^+^ Kaempferol-3-O-rutinoside-7-O-*β*-D-glucoside 4 22.68 247, 307 337 \[M-H\]^-^ 191 \[M-H-146\]^-^ 339 \[M+H\]^+^ 147 \[M+H-192\]^+^ *p*-Coumaoylquinic acid 5 25.07 254, 356 609 \[M-H\]^-^ 611 \[M+H\]^+^ 465 \[M+H-146\]^+^ 303 \[M+H-146-162\]^+^ Quercetin-3-O-rutinoside 6 27.08 256, 354 463 \[M-H\]^-^ 465 \[M+H\]^+^ 303 \[M+H-162\]^+^ Quercetin-3-O-*β*-D-glucoside 7 27.26 265, 346 593 \[M-H\]^-^ 595 \[M+H\]^+^ 449 \[M+H-146\]^+^ 287 \[M+H-146-162\]^+^ Kaempferol-3-O-robinobioside 8 28.05 265, 347 593 \[M-H\]^-^ 595 \[M+H\]^+^ 449 \[M+H-146\]^+^ 287 \[M+H-146-162\]^+^ Kaempferol-3-O-rutinoside 9 29.12 265, 346 447 \[M-H\]^-^ 449 \[M+H\]^+^ 287 \[M+H-162\]^+^ Kaempferol-3-O-*β*-D-galacoside 10 29.30 248, 327 515 \[M-H\]^-^ 353 \[M-H-162\]. 499 \[M+H-18\]^+^ 163 \[M+H-162-192\]^+^ 3,4-Dicaffeoylquinic acid 11 30.37 248, 325 515 \[M-H\]^-^ 353 \[M-H-162\]. 499 \[M+H-18\]^+^ 163 \[M+H-162-192\]^+^ 3,5-Dicaffeoylquinic acid 12 30.37 265, 347 447 \[M-H\]^-^ 449 \[M+H\]^+^ 287 \[M+H-162\]^+^ Kaempferol-3-O-*β*-D-glucoside 13 31.28 248, 325 515 \[M-H\]^-^ 353 \[M-H-162\]. 499 \[M+H-18\]^+^ 163 \[M+H-162-192\]^+^ 4,5-Dicaffeoylquinic acid HPLC-DAD-ESI-MS: High-performance liquid chromatography-diode array detector-electrospray ionization-mass spectrometry, TIC: Total ion current, Identification was supported by comparison with reference standards where available and by correlation with previous literature reports. Peaks 2, 3 and 11, 12 were co-eluted. Peak numbers and retention times (TR) refer to HPLC chromatograms in [Figure 2b](#F0002){ref-type="fig"} Peak 4 yielded a \[M-H\]^-^ ion at *m/z* 337, and \[M+H\]^+^ ion at *m/z* 339, \[M+H-192\]^+^ ion at *m/z* 147. The UV spectrum showed λ~max~\> at 307, 293 (sh), and 247 nm (sh), which is characteristic of a Cinnamic acid derivative\[[@CIT19][@CIT21]\]; hence, the structure of peak 4 was proposed to be p-coumaroylquinic acid.\[[@CIT19][@CIT21]\] Peak 6 yielded a \[M-H\]^-^ ion at *m/z* 463, and \[M+H\]^+^ ion at *m/z* 465, \[M+H-162\]^+^ ion at *m/z* 303. The UV spectrum showed λ~max~ at 255 and 356 nm, suggesting that this as a quercetin glycoside.\[[@CIT21]\] By examining the known flavonol glycoside in the genus *Gynura*, isoquercitrin was consistent with the above data. And the elution order of isoquercitrin was in agreement with the compound prior toKaempferol-3-O-robinobioside (peak 7) and afterward with rutin (peak 5).\[[@CIT22]--[@CIT24]\] Thus, peak 6 was tentatively identified as isoquercitrin. Peak 7 and 8 were a pair of isomers. Both of them gave a \[M-H\]^-^ ion at *m/z* 593, and \[M+H\]^+^ ion at *m/z* 595, \[M+H-146\]^+^ ion at *m/z* 449, \[M+H-146-162\]^+^ ion at *m/z* 287. The UV spectrum showed λ~max~ at 265 and 347 nm, which suggested peak 7 and 8 were kaempferol glycoside derivatives.\[[@CIT15]--[@CIT17][@CIT19]\] By examining the known kaempferol glycoside in the genus *Gynura*, Kaempferol-3-O-robinobioside and kaempferol-3-O-rutinoside were consistent with the above data.\[[@CIT25]\] The elution order in HPLC of Kaempferol-3-O-robinobioside being prior to kaempferol-3-O-rutinoside has been reported by many in the literature.\[[@CIT26][@CIT27]\] Thus, peak 7 and 8 were identified as Kaempferol-3-O-robinobioside and kaempferol-3-O-rutinoside, respectively. Peak 9 yielded a \[M-H\]^-^ ion at *m/z* 447, and \[M+H\]^+^ ion at *m/z* 449, \[M+H-162\]^+^ ion at *m/z* 287. The UV spectrum showed λ~max~ at 265 and 346 nm, suggesting this as a kaempferol glycoside. So peak 9 is an isomer of kaempferol-3-O-*β*-D-glucopyranoside (peak 12). Thus, peak 9 was tentatively identified as kaempferol-3-O-*β*-D-galacopyranoside. Peak 10, 11, and 13 are isomers. Both of them gave a \[M-H\]^-^ ion at *m/z* 515, \[M-H-162\]^-^ ion at *m/z* 353, and \[M+H\]^+^ ion at *m/z* 517, \[M+H-18\]^+^ ion at *m/z* 499, \[M+H-162-192\]^+^ ion at *m/z* 163. The 3 compounds also had similar UV absorptions with maxima at 327, 294 (sh), and 248 nm (sh), which is characteristic of caffeic acid derivatives.\[[@CIT28]--[@CIT31]\] Peak 11 was isolated by the chromatography column and identified as 3,5-Dicaffeoylquinic acid by the NMR and ESI-MS spectrum data. According to the elution order in HPLC of Dicaffeoylquinic acid reported in the literature,\[[@CIT31]--[@CIT33]\] 3,4-Dicaffeoylquinic acid is prior to 3,5-Dicaffeoylquinic acid, which is prior to 4,5-Dicaffeoylquinic acid, in a sequence. Thus, peak 10 and 13 were tentatively identified as 3,4-Dicaffeoylquinic acid and 4,5-Dicaffeoylquinic acid, respectively. The flavonoid and phenolic acid compounds were affected by the concentration of extraction ethanol. The single-factor experiment showed that 60% ethanol was suitable to extract the phenolic constituents from the plant. The levels of phenolic contents were decreased as the concentration of ethanol increased. Chloroform was used to remove the nonpolar constituents, while little extracts were obtained using diethyl ether and petroleum ether. The ethyl acetate extracts showed powerful antioxidant activity and highest total phenolic content. HPLC analysis showed that ethyl acetate extracts only shared 3 principal peaks, and the kaempferol-3-O-*β*-D-glucopyranoside was the major constituent. However, *n*-butanol extract shared numerous flavonoid compounds, while the total phenolic was lower. In order to fully elaborate the phenolic compounds of the extract from *G. divaricata*, the extracts of ethyl acetate and *n*-butanol were isolated using chromatograph column and HPLC-DAD-ESI-MS method. To our best knowledge, the present study is the first report of the isolation and identification of triglycoside of kaempferol and Dicaffeoylquinic acid from the leaves of *G. divaricata*. And we also developed a HPLC-DAD-ESI-MS method to separate and identify the minor constituents of the *n*-butanol extracts. The bioactive evaluation of the isolated compounds and the crude drug deserved further research. CONCLUSION {#sec1-4} ========== Seven phenolic compounds were isolated and identified from the leaves of *G. divaricata*, and the structures were fully elucidated by the spectrum methods. HPLC-DAD-ESI-MS method was used to identify the other 8 minor phenolic constituents of the *n*-butanol extracts. This was the first time to use the HPLC-DAD-ESI-MS method to identify the phytochemicals of the genera *Gynura*, and kaempferol-3-O-rutinoside-7-O-*β*-D-glucopyranoside and 3,5-Dicaffeoylquinic acid were identified for the first time from the genus *Gynura*. This project was supported by the National Natural Sciences Foundation of China (No.20662008). **Source of Support:** National Natural Sciences Foundation of China (No.20662008) **Conflict of Interest:** None declared

1. Introduction {#sec1} =============== *Toxoplasma gondii* can infect all warm-blooded vertebrates, including mammals and birds \[[@B1]--[@B3]\]. Genetic diversity of*T. gondii* is widespread due to the biological and epidemiological diversity of this parasite.*T. gondii* isolates can be clustered into six major clades \[[@B4]\], and genetic diversity of*T. gondii* is especially common in South America \[[@B4]\]. Utilizing 11 genetic markers,*T. gondii* isolates in North America and Europe are grouped into four major clonal lineage types (I, II, III, and 12) \[[@B5], [@B6]\] using PCR-RFLP. Rhoptry kinases are involved in mediating pathogenesis of*T. gondii*\[[@B7]\], and they are also master regulators that manipulate the host inflammatory responses \[[@B8], [@B9]\].*T. gondii*rhoptry protein 17 (ROP17), a member of the ROP2 subfamily \[[@B10]\], was predicted to have a cellular localization on the parasitophorous vacuole membrane (PVM), which may participate in the manipulation of the host signalling pathways \[[@B9]\]. Previous studies have shown the existence of sequence variation in some ROP genes, such as*rop7*,*rop9*,*rop13*, and*rop38* \[[@B11]--[@B14]\]. However, it is yet to be known whether sequence diversity exists in*rop17*gene of*T. gondii*. The objective of the present study was to examine sequence variation in*rop17*gene among*T. gondii*strains representing different genotypes and host and origins. 2. Materials and Methods {#sec2} ======================== 2.1. *T. gondii* Isolates {#sec2.1} ------------------------- Ten*T. gondii* strains collected from different hosts and locations were used for analysis in this study ([Table 1](#tab1){ref-type="table"}). These strains have been genotyped and their genomic DNA has been prepared as described previously \[[@B15]--[@B17]\]. 2.2. Amplification of*rop17* Genes and Sequencing {#sec2.2} ------------------------------------------------- The*rop17* gene was amplified by PCR. Two primers were designed based on the*rop17* sequence of*T. gondii* RH strain available in GenBank (accession number: KC997178): ROP17F, 5′-AGGACAACACTAGGTAGCGAGAACC-3′, and ROP17R, 5′-TGGCGAAGTCAAGAGACGACGCAG-3′. Each reaction was performed in a total volume of 25 *μ*L containing 12.5 *μ*L*Premix Taq*(TaKaRa, Dalian, China), ROP17F (20 pmol) 0.25 *μ*L, ROP17R (20 pmol) 0.25 *μ*L, template DNA (200 ng) 2 *μ*L, and ddH~2~O 8 *μ*L, and the reaction conditions were 94°C for 5 min, then 35 cycles of 30 sec at 94°C, 30 sec at 55°C, and 1 min 20 s at 72°C, and a final extension at 72°C for 10 min. All the PCR products were then cloned into pMD18-T vector (TaKaRa, China) after purification using the DNA purification kit (TIANGEN, China) and then sequenced by Songon Biotech Co., Ltd. (Shanghai, China). 2.3. Sequence Analysis and Reconstruction of Phylogenetic Relationships {#sec2.3} ----------------------------------------------------------------------- The*rop17* gene sequences of different*T. gondii* strains were aligned using Multiple Sequence Alignment Program, Clustal X 1.83 \[[@B18]\], and the sequence differences were determined according to Chilton et al. \[[@B19]\] and Zhao et al. \[[@B20]\]. Phylogenetic reconstruction was based on the*rop17* gene sequences determined in the present study plus the corresponding sequences of strains TgC7, PRU, and RH available in GenBank (accession numbers: KC997176, KC997177, and KC997178) using three inference methods, namely, neighbor-joining (NJ), maximum likelihood (ML), and maximum parsimony (MP), using the sequence of*Neospora caninum* (NCLIV_027930) as the outgroup. All the analyses were conducted following previous studies \[[@B20], [@B21]\]. Phylograms were drawn using the Tree View program version 1.65 \[[@B22]\]. 3. Results and Discussion {#sec3} ========================= The length of the*rop17* genes from all examined*T. gondii* isolates was 1375 bp and A+T contents varied from 49.45% to 50.11%. The alignment of 10*rop17* sequences plus the corresponding sequences of the RH, PRU, and TgC7 strains available in GenBank revealed nucleotide polymorphisms at 33 positions, with an intraspecific variation of 0--2.1%. The genetic diversity in*rop17* gene was higher than our previous studies for PLP1 \[[@B23]\], ROP7 \[[@B11]\], eIF4A \[[@B24]\], and MIC13 \[[@B25]\] genes and the whole genome, secretome, and kinome of*T. gondii*\[[@B8]\]. 16 variable positions were identified as transitions and the rest variable nucleotides were classified as transversions, and no deletions were detected in the 13*rop17* gene sequences. Phylogeny reconstruction using MP, NJ, and ML analyses revealed two major clusters ([Figure 1(a)](#fig1){ref-type="fig"}). Topologies of all trees based on nucleotide sequences inferred by three different methods were similar, with only the small difference of bootstrap values. The classical genotypes II and III and atypical Type 12 strain were clustered in one clade. The subtree of NJ analysis further showed that genotype III (strain CTG) was separated from other strains which were supported by bootstrap analysis, and the atypical Type 12 (TgWtSc40 strain) was closely related to classical genotype II (strain PRU) ([Figure 1(b)](#fig1){ref-type="fig"}).*T. gondii* genotype II is one of the parental lineage of Type 12 based on the analysis of the inheritance of multilocus genotypes \[[@B6], [@B26]\]. The somewhat close relationship between Type II and Type 12 strains coincided with analyses of*UPRT* and*SAG1* loci \[[@B6]\]. All the strains belonging to genotype I in this study were clustered together, including strain TgPLh and typical strains GT1 and RH. Atypical strains TgCat1, TgToucan, TgCatBr64, and TgCatBr5 were phylogenetically clustered more closely with Type I strains. Of these, TgCatBr64 and TgCatBr5 strains which originated from cats in Brazil were grouped together. Based on the limited*T. gondii* strains examined in the present study, the*rop17* gene sequences could distinguish the major clonal lineages, but not all, showing the weak differentiation of*T. gondii* genotypes compared to analyses using GRA5, GRA6, and AK gene sequences as genetic markers \[[@B27]--[@B29]\]. Further validation of the*rop17* gene sequences as genetic marker is warranted by sampling more*T. gondii*strains from wider geographical locations and more hosts. The analyses of sequence variations in nucleotides and amino acids among different strains showed high ratio of nonsynonymous to synonymous polymorphisms (\>1), suggesting that*T. gondii rop17* shows signs of positive selection, although more isolates will be required to determine whether*rop17* gene is under selection at the population level. Under the immunized stresses of host cells, the positive selection occurring in*rop17* gene may increase stress resistance. Ongoing positive selection is also found in several polymorphic dense granule (GRA) antigens \[[@B30], [@B31]\] and some other ROPs \[[@B8]\]. 4. Conclusion {#sec4} ============= In summary, the present study demonstrated the existence of slightly high sequence variability in the*rop17* gene sequences among*T. gondii*strains from different hosts and regions, which may be explored as a new genetic marker for population genetic studies of*T. gondii* isolates, and contributed to discovery of the new strategies for vaccination, treatment, or diagnosis. Project support was provided by National Natural Science Foundation of China (Grant nos. 31228022, 31172316, and 31230073) and the Science Fund for Creative Research Groups of Gansu Province (Grant no. 1210RJIA006). Associate Professor Chunlei Su at the Department of Microbiology, the University of Tennessee, Knoxville, USA, is gratefully thanked for providing reference*T. gondii*strains. Conflict of Interests ===================== The authors declare that there is no conflict of interests in this paper. {#fig1} ###### Details of *Toxoplasma gondii*strains used in the present study. Strain Host Geographical origin Genotype∗ -------------------- ------------------- --------------------- -------------------------------------- GT1 Goat United States Reference, Type I PTG Sheep United States Reference, Type II, ToxoDB number 1 CTG Cat United States Reference, Type III, ToxoDB number 2 TgCatBr5 Cat Brazil Reference, ToxoDB number 19 TgCatBr64 Cat Brazil Reference, ToxoDB number 111 TgCgCa1 Cougar Canada Reference, ToxoDB number 66 TgToucan (TgRsCr1) Toucan Costa Rica Reference, ToxoDB number 52 TgPLh Pig China Type I, ToxoDB number 10 QHO Sheep China Type II, ToxoDB number 1 TgWtdSc40 White-tailed deer United States Type 12, ToxoDB number 5 \*Based on genotyping results of Zhou et al. \[[@B15], [@B16]\] and Su et al. \[[@B17]\]. [^1]: Academic Editor: Rekha PD

Amandita FY, Rembold K, Vornam B, et al. DNA barcoding of flowering plants in Sumatra, Indonesia. Ecol Evol. 2019;9:1858--1868. 10.1002/ece3.4875 1. INTRODUCTION {#ece34875-sec-0001} =============== DNA barcoding is a species identification method, using a short, standardized DNA region, so‐called DNA barcode (Hebert, Cywinska, Ball, & de Waard, [2003a](#ece34875-bib-0040){ref-type="ref"}). In principle, DNA barcodes contain variation that can be posed as a character to differentiate species. Although the utility of DNA barcoding for species identification has raised debates over its feasibility (Collins & Cruickshank, [2013](#ece34875-bib-0016){ref-type="ref"}; Krisnamurthy & Francis, [2012](#ece34875-bib-0056){ref-type="ref"}), the method has been increasingly applied during the last decade, especially to facilitate biodiversity studies of very diverse but taxonomically poorly known regions (Blaxter, [2004](#ece34875-bib-0007){ref-type="ref"}; Hajibabaei et al., [2005](#ece34875-bib-0038){ref-type="ref"}), such as Sumatran tropical rainforests. Sumatran tropical rainforests are very rich in flora and fauna (Davis, Heywood, & Hamilton, [1995](#ece34875-bib-0021){ref-type="ref"}; Laumonier, [1997](#ece34875-bib-0059){ref-type="ref"}; Whitten, Damanik, Anwar, & Hisyam, [2000](#ece34875-bib-0089){ref-type="ref"}); nonetheless, they are only sparsely studied compared to other islands in the Malayan Archipelago (Laumonier, [1997](#ece34875-bib-0059){ref-type="ref"}). In terms of plant diversity, the Sumatran forests are comparable to the forests of Borneo and are richer than those found in Java and Sulawesi (Meijer, [1981](#ece34875-bib-0064){ref-type="ref"}). Sumatra is reported as one of the global centers of vascular plant diversity with a species density of 3,000 to 5,000 species per 10,000 km^2^ (Barthlott, Mutke, Rafiqpoor, Kier, & Kreft, [2005](#ece34875-bib-0006){ref-type="ref"}). Roos, Keßler, Gradstein, and Baas ([2004](#ece34875-bib-0074){ref-type="ref"}) estimated a total number of 10,600 plant species in Sumatra with more than 300 endemic species. Laumonier ([1997](#ece34875-bib-0059){ref-type="ref"}) argued that many scientists mistakenly consider that the flora of Sumatra is sufficiently well known since it is similar to that of the Malaysian peninsula, but many parts, especially the center of the island, are floristically unexplored territories. Despite the importance of conserving the ecosystem, the total forest area in Sumatra has decreased from over 23 million hectares to probably less than 16 million hectares between 1985 and 1997 (World Bank, [2001](#ece34875-bib-0090){ref-type="ref"}). The southern provinces of Sumatra have lost most of their lowland forests, including those in protected areas (Lambert & Collar, [2002](#ece34875-bib-0058){ref-type="ref"}). Approximately 7.5 million hectares of primary forest loss were recorded in Sumatra during 1990--2010 and an additional 2.3 million hectares of primary forest were degraded (Margono et al., [2012](#ece34875-bib-0062){ref-type="ref"}). Between 2000 and 2010, the deforestation rate was estimated to be above 5% per year in the eastern lowlands of Sumatra (Miettinen, Shi, & Liew, [2011](#ece34875-bib-0066){ref-type="ref"}). The total deforested areas in Sumatra within 2011 alone were recorded to be approximately 2,200 hectares or as much as 3,520 soccer fields (BP‐REDD+, [2015](#ece34875-bib-0067){ref-type="ref"}). The causes of these massive deforestation and forest degradation are a large‐scale conversion into timber or estate crop plantations, illegal logging, and forest fires. By 2010, 3.9 million hectares of Sumatran lowland forests had been converted into oil palm (*Elaeis guineensis*) plantations (Koh, Miettinen, Liew, & Ghazoula, [2011](#ece34875-bib-0050){ref-type="ref"}). The extensive loss of natural habitat puts a great number of species at risk and may lead to the loss of tropical fauna including forest‐dwelling birds (Koh et al., [2011](#ece34875-bib-0050){ref-type="ref"}), mammals (Maddox, Priatna, Gemita, & Salampessy, [2007](#ece34875-bib-0061){ref-type="ref"}), and orangutan (Gaveau et al., [2009](#ece34875-bib-0032){ref-type="ref"}). Undoubtedly, the destruction also affects the plant diversity (Brook, Sodhi, & Ng, [2003](#ece34875-bib-0008){ref-type="ref"}; Corlett, [1992](#ece34875-bib-0017){ref-type="ref"}; Rembold, Mangopo, Tjitrosoedirdjo, & Kreft, [2017](#ece34875-bib-0073){ref-type="ref"}; Turner et al., [1994](#ece34875-bib-0084){ref-type="ref"}). The rate of species loss in tropical forests seems to be higher than the species exploration due to lack of resources and sound species conservation management such as limited number of taxonomists working in this region, inadequate herbarium collections, and inaccessible taxonomic literature (Kiew, [2002](#ece34875-bib-0047){ref-type="ref"}; Meyer & Paulay, [2005](#ece34875-bib-0065){ref-type="ref"}; Tautz, Arctander, Minelli, Thomas, & Vogler, [2003](#ece34875-bib-0082){ref-type="ref"}). Species explorations become more challenging when the species cannot be identified morphologically. Identification keys based upon morphological characteristics can be difficult to use if features are not present (e.g., in sterile or juvenile specimens) or not well developed. The use of DNA barcoding might help to overcome the limitations of morphological characters and might help to speed up species identification. This has been made possible because DNA barcoding can identify organisms at any stage of development (e.g., Barber & Boyce, [2006](#ece34875-bib-0005){ref-type="ref"}; Hausmann et al., [2011](#ece34875-bib-0039){ref-type="ref"}; Heimeier, Lavery, & Sewell, [2010](#ece34875-bib-0043){ref-type="ref"}; Ko et al., [2013](#ece34875-bib-0049){ref-type="ref"}), or at particular gender (e.g., Elsasser, Floyd, Herbert, & Schulte‐Hostedde, [2009](#ece34875-bib-0027){ref-type="ref"}), or specimens isolated from small and incomplete tissue, whether it is fresh, broken, or old (e.g., Hajibabaei et al., [2006](#ece34875-bib-0037){ref-type="ref"}; Valentini, Pompanon, & Taberlet, [2008](#ece34875-bib-0086){ref-type="ref"}). DNA barcoding may also help to discover new species and to identify cryptic species (e.g., Hebert, Penton, Burns, Janzen, & Hallwachs, [2004](#ece34875-bib-0041){ref-type="ref"}; Pauls, Blahnik, Zhou, Wardwell, & Holzenthal, [2010](#ece34875-bib-0072){ref-type="ref"}; Ward, Costa, Holmes, & Steinke, [2008](#ece34875-bib-0087){ref-type="ref"}). DNA barcoding is now well established for animals (Crawford et al., [2013](#ece34875-bib-0019){ref-type="ref"}; Hebert, Cywinska, Ball, & deWaard, [2003a](#ece34875-bib-0040){ref-type="ref"}; Hebert, Ratnasingham, & de Waard, [2003b](#ece34875-bib-0042){ref-type="ref"}; Hebert et al., [2004](#ece34875-bib-0041){ref-type="ref"}; Lim, [2012](#ece34875-bib-0060){ref-type="ref"}; Nagy, Sonet, Glaw, & Vences, [2012](#ece34875-bib-0068){ref-type="ref"}; Ward, Zemlak, Innes, Lasr, & Hebert, [2005](#ece34875-bib-0088){ref-type="ref"}) by using the mitochondrial DNA *CO1* (cytochrome c oxidase subunit 1) as a standard region. However, this region is ineffective for plant identification due to generally low nucleotide substitution rates in plant mitochondria (Chase et al., [2005](#ece34875-bib-0013){ref-type="ref"}; Fazekas, Kesanakurti, & Burgess, [2009](#ece34875-bib-0029){ref-type="ref"}). A number of candidate gene regions were suggested as potential barcodes for plants including coding genes and noncoding genes in the nuclear and plastid genomes (e.g., Chase, Cowan, & Hollingsworth, [2007](#ece34875-bib-0014){ref-type="ref"}; Kress & Erickson, [2007](#ece34875-bib-0051){ref-type="ref"}; Kress, Wurdack, Zimmer, Weigt, & Janzen, [2005](#ece34875-bib-0055){ref-type="ref"}; Taberlet et al., [2007](#ece34875-bib-0079){ref-type="ref"}). Some studies suggested DNA barcoding based on a single chloroplast region (e.g., Lahaye et al., [2008](#ece34875-bib-0057){ref-type="ref"}) or a combination of different regions (e.g., Chase et al., [2007](#ece34875-bib-0014){ref-type="ref"}; Hollingsworth et al., [2009a](#ece34875-bib-0044){ref-type="ref"}; Kress & Erickson, [2007](#ece34875-bib-0051){ref-type="ref"}). A study by Kress and Erickson ([2007](#ece34875-bib-0051){ref-type="ref"}) showed that the various combinations of two loci were all more powerful at differentiating between species than either locus individually. In 2009, the Plant Working Group under The Consortium for Barcode of Life ([CBOL](#ece34875-bib-0012){ref-type="ref"}) suggested that there were no other two‐loci or multi‐loci barcode provided appreciably greater species resolution than the *matK+rbcL* combination. However, in some complex groups, such as in the genus *Berberis* (Roy et al., [2010](#ece34875-bib-0075){ref-type="ref"}), the combination of *matK* with *rbcL* is not sufficient to distinguish all species. The investigation of these markers will contribute to the development of useful barcode information for plant identification and to document plant species globally. This study aims to generate DNA barcodes of flowering plant species in four land‐use systems in Jambi Province (Sumatra) using two DNA chloroplast markers (*matK* and *rbcL*) and to evaluate the effectiveness of these two markers as DNA barcodes for flowering plants. Crucial characteristics for evaluating the performance of DNA barcodes include universal applicability, ease of data retrieval, and sufficient variability of the used marker (Fazekas et al., [2008](#ece34875-bib-0028){ref-type="ref"}; Kress & Erickson, [2007](#ece34875-bib-0051){ref-type="ref"}). 2. METHODS {#ece34875-sec-0002} ========== 2.1. Study sites {#ece34875-sec-0003} ---------------- This study was carried out in the EFForTS project sites (<https://www.uni-goettingen.de/efforts>) in Jambi Province (Sumatra, Indonesia) comprises of 32 core plots sized 50 m × 50 m. Details about the EFForTS project sites and plot design are described in Drescher et al. ([2016](#ece34875-bib-0026){ref-type="ref"}). 2.2. Specimen collection and identification {#ece34875-sec-0004} ------------------------------------------- Herbarium specimens were collected from three individuals of as many as possible vascular plant species within the 32 core plots. The plant survey included all trees with a diameter at breast height (DBH) ≥10 cm within the entire plot and all vascular plants within five 5 m × 5 m subplots nested within each core plot. Leaf tissue (approximately 2 cm^2^) was collected from each fresh herbarium specimen and dried in silica gel for DNA barcoding analysis. Herbarium vouchers were prepared, morphologically identified, and deposited at the herbarium of the Southeast Asian Regional Centre for Tropical Biology (SEAMEO‐BIOTROP), the Herbarium Bogoriense---Research Center for Biology, LIPI, and herbarium of the University of Jambi. The results of the morphological identification were then compared to the molecular identification results. Molecular identification was conducted for all samples that were successfully barcoded, but only samples that have been morphologically identified were included in the further analysis. 2.3. DNA analysis {#ece34875-sec-0005} ----------------- Based on the result of morphological species identification, two specimens per species were selected for genetic analysis. DNA extractions were performed on healthy dried leaf tissues from all selected samples using the DNeasy 96 Plant Kit (Qiagen, Hilden, Germany) following the manufacturer\'s protocols. The concentration and quality of the extracted DNA were checked by 0.8%--1% agarose gel electrophoresis with Lambda DNA as standard (Roche), visualized by UV illumination and saved using a polaroid camera. Each extracted DNA was amplified by performing polymerase chain reaction (PCR) using universal primers listed in Table [1](#ece34875-tbl-0001){ref-type="table"}. For rbcL, the amplification was straightforward, while for matK, two different amplification reactions were performed. First, the DNA of all investigated samples were amplified using the universal primer pair 1RKIM_f and 3FKIM_r (Table [1](#ece34875-tbl-0001){ref-type="table"}). The second amplification reaction, using the primer pair 390f and 990r (Table [1](#ece34875-tbl-0001){ref-type="table"}), included only those samples which showed no amplification product or produced multiple PCR products in the first amplification reaction. ###### Universal primers of *matK* and *rbcL* used in DNA amplification and sequencing No. Region Name of primer Primer sequence (5′ → 3′) References ------------ ----------------------------- ------------------------------------------------------------------------------------- ---------------------------- ----------------------------------------------------------------- 1 *matK* *3F_KIM_f* CGTACAGTACTTTTGTGTTTACGAG Ki‐Joong Kim (unpublished) *1R_KIM_r* ACCCAGTCCATCTGGAAATCTTGGTTC Ki‐Joong Kim (unpublished) *390f* CGATCTATTCATTCAATATTTC Cuenoud et al. ([2002](#ece34875-bib-0020){ref-type="ref"}) *990r* GGACAATGATCCAATCAAGGC Dayananda, Ashton, Williams, & Primack ([1999](#ece34875-bib-0022){ref-type="ref"}) 2 *rbcL* *rbcLa_f* ATGTCACCACAAACAGAGACTAAAGC Krees and Erickson ([2007](#ece34875-bib-0051){ref-type="ref"}) *rbcLa_r* GAAACGGTCTCTCCAACGCAT Fazekas et al. ([2008](#ece34875-bib-0028){ref-type="ref"}) John Wiley & Sons, Ltd The sequencing reactions were performed using the ABI PrismTM Big DyeTM Terminator Cycle Sequencing Ready Reaction Kit v1.1 (Applied Biosystems), based on the principles described by Sanger, Nicklen, and Coulson ([1977](#ece34875-bib-0076){ref-type="ref"}). Data were collected from capillary electrophoresis on an ABI Prism 3100® Genetic Analyzer with the Sequence Analysis Software v3.1 (Applied Biosystems). The sequencing was performed with the same primers used for amplification in both directions. The amplification and sequencing reaction mixtures are shown in Supporting Information Appendix 1, while the temperature profiles of the PCR for amplification and sequencing are shown in Supporting Information Appendix 2. 2.4. Sequence analysis {#ece34875-sec-0006} ---------------------- To ensure the generated DNA barcodes were as accurate as possible, sequence editing was performed using CodonCode Aligner software (CodonCode Corporation, Dedham, USA). Furthermore, each of these edited barcodes was assigned to a particular taxon by comparing it with the nucleotide sequences in GenBank database and Barcode of Life Database (BOLD). Moreover, the results of sequence identification were cross‐checked with the morphological identification results. The match between morphological and molecular identification results was counted into three levels: species, genus, and family. The following decisions were made for correct identification assignments, namely: (a) when the species name from the molecular identification matched the species name from the morphological identification, then it was counted as a correct species identification, (b) when the identification result only matched the genus or family, then it was counted as correct genus or family identification, and (c) when the result between morphological and molecular identification did not match, it was counted as incorrect identification if *matK* and *rbcL* both showed similar results at least at family level, or it was counted as mislabeling/contamination if the results of *matK* and *rbcL* were different. Herbarium specimens were double‐checked in cases of incorrect identification. Sequence alignment was carried out independently for each marker in two stages. First, multiple sequences were aligned according to their families using the ClustalW program (Thompson, Higgins, & Gibson, [1994](#ece34875-bib-0083){ref-type="ref"}) embedded in MEGA6 (Tamura, Stecher, Peterson, Filipski, & Kumar, [2013](#ece34875-bib-0081){ref-type="ref"}). Reference sequences were downloaded from GenBank/BOLD and included in the alignment for those species represented with only one sample. The alignment results were subsequently checked for the occurrence of ambiguities caused by the presence of indels and/or substitutions and edited if necessary. In the second stage, all aligned sequences from each family were manually aligned with sequences from other families. Gaps were added if necessary, and the final alignment was trimmed at both ends. The aligned sequences of *rbcL* and *matK*were combined to obtain two‐loci DNA barcodes using SequenceMatrix software (Vaidya, Lohman, & Meier, [2011](#ece34875-bib-0085){ref-type="ref"}). Identification success was also calculated with best‐close match analysis as implemented in TaxonDNA (Meier, Kwong, Vaidya, & Ng, [2006](#ece34875-bib-0063){ref-type="ref"}). This analysis only included the species with at least two representatives. A threshold value T was determined for each dataset as a divergence percentage in which 95% of all intraspecific distances were found. In this method, all recovered barcodes were formatted as both database and query. A query can only be identified if the corresponding sequence has a match in the dataset that falls into the 0% to T% interval. If the species name was identical, the query was considered to be successfully identified. A query was considered ambiguously identified when it matched more than one sequence of different species besides the correct species. On the other hand, a query was considered incorrectly identified when it matched to sequences belonging to other species. All queries without such a match would remain unidentified. Pairwise distance matrices were created to calculate the genetic distance using MEGA6 (Tamura et al., [2013](#ece34875-bib-0081){ref-type="ref"}) based on the Tamura‐Nei model ([1993](#ece34875-bib-0080){ref-type="ref"}) assuming the differences in substitution rate between nucleotides and the inequality of nucleotide frequencies with gamma‐distributed rates between sites and the pattern between lineages were assumed to be heterogeneous. The calculation results of intra‐ and interspecific divergences in these matrices were separated using ExcaliBAR (Aliabadian et al., [2014](#ece34875-bib-0001){ref-type="ref"}) to facilitate the measures of distance range and distance mean of each type of divergence. Frequency (%) distribution of intra‐ and interspecific divergences of each marker was calculated and depicted in graphics using Excel to find possible "gap" between these two divergences. This so‐called barcoding gap illustrates the effectiveness of DNA barcodes in discriminating query species from one to another. An ideal barcode can be determined by the presence of a barcoding gap, which occurs when the minimum value of the interspecific divergence is higher than the maximum level of intraspecific divergence (Meyer & Paulay, [2005](#ece34875-bib-0065){ref-type="ref"}). Based on the aligned sequences, phylogenetic trees were reconstructed using MEGA6 (Tamura et al., [2013](#ece34875-bib-0081){ref-type="ref"}) with three different algorithms: maximum parsimony (MP), maximum likelihood (ML), and neighbor joining (NJ). Percentages of species, genus, and family monophyletic clades were calculated from each reconstructed tree. Furthermore, ordinal‐level phylogenies were reconstructed based on maximum likelihood trees of each used marker and were compared to APG III (APG III [2009](#ece34875-bib-0002){ref-type="ref"}) phylogenies to see if there were inconsistencies between these two topologies. 3. RESULTS {#ece34875-sec-0007} ========== From all 5,328 samples collected from the field, only 2,590 samples were included in the study due to time restriction. The selection of studied samples was based on the consideration to involve as much species as possible, and each of these species should be represented at least by two samples. Species with only one sample were still included, but the barcodes generated from single‐sampled species were excluded from the pairwise analysis. We extracted DNA from dried leaf specimens without noticeable difficulties. The amplification and sequencing, however, turned out to be more problematic especially when using *matK*primers. Recoverability of DNA sequences for *rbcL* was overall high (amplification and sequencing success were 96.9% and 94.7%, respectively). The amplification and sequencing results using the primer of *matK* were only moderately successful (79.1% and 65.8%, respectively). A total of 1,207 *matK* barcodes representing 441 species of 97 families of 40 orders, and 2,376 *rbcL* barcodes representing 750 species of 126 families of 44 orders, were generated in this study. For both markers, the highest match between morphological and molecular identification was at genus level (46.6% with *matK* and 51.3% with *rbcL*). The matched identification at species level was higher with *matK* than with *rbcL* (30.2% and 22.4%, respectively). Meanwhile, incorrect identification was relatively low for both regions (3.5%). To maintain the accuracy of the analysis, we excluded all misidentified or presumably mislabeled barcodes from the dataset. Since the study aims at comparing the performance of *matK* and *rbcL* and to generate two‐loci barcodes, only samples from which both *matK* and *rbcL* barcodes were successfully recovered were included in the further analysis. Consequently, only 322 samples from 161 species (two samples per species) were included in best‐close match and barcode‐gap analysis and 334 samples from 334 species (one sample per species) were included in phylogenetic analysis. According to the best‐close match analysis, *matK* has higher overall species identification success compared to *rbcL* (78.3% and 71.4%, respectively), and the highest correct species identification was obtained by the combination of both markers (81.1%). There were 22 species which remained unidentified by each marker and the two‐loci marker. Furthermore, this study showed that the mean value of intraspecific divergences (0.0008--0.0014) was very low and the mean value of the interspecific divergences (0.1--0.3) was significantly higher (unpaired *t*‐test, *p* \< 0.01). The frequency (%) distribution of intraspecific and interspecific divergence using three markers (Figure [1](#ece34875-fig-0001){ref-type="fig"}) showed that no barcode gaps existed as the intraspecific divergences overlapped with interspecific divergences. {#ece34875-fig-0001} As expected, *matK* had a higher discrimination level than *rbcL* (80% and 73%, respectively) but the difference was not significant (one‐way ANOVA, *p* \> 0.05). The combination of *matK* and *rbcL* improved the discrimination up to 89%. Forty‐four out of 161 species could not be discriminated by *rbcL* and eleven of them were not discriminated by any of the markers including the two‐loci barcode. These species were mostly from species‐rich genera, such as *Ficus*(Moraceae), *Santiria*(Burseraceae), and *Litsea*(Lauraceae). Nine phylogenetic trees (Supporting information Appendix 3--11) were constructed based on multiple sequence alignments of *matK*, *rbcL*, and *matK+rbcL* using three different methods: maximum parsimony (MP), neighbor joining (NJ), and maximum likelihood (MP). Each tree was observed and similar topologies were found amongst these trees (Table [2](#ece34875-tbl-0002){ref-type="table"}). ###### Percentage of monophyletic clades recovered in nine reconstructed phylogenetic trees Barcode Monophyletic with support value \>70% ----------------- --------------------------------------- ------ ------ ------- ------ ------ ------- ------ ------ ***matK*** 95.9 68.4 73.9 93.9 66.7 69.6 98.0 64.9 68.9 ***rbcL*** 95.9 63.2 60.3 93.9 63.2 64.0 89.9 63.2 55.9 ***matK+rbcL*** 100.0 71.9 73.3 100.0 64.9 73.9 100.0 70.2 75.2 John Wiley & Sons, Ltd Seventeen families were not included in the calculation of family‐level monophyletic percentage as these families were presented with only one taxon. The two‐loci marker provided 100% taxonomic resolution at family level with all three different methods. Twenty‐two species were nonmonophyletic in all phylogenetic trees (Supporting information Appendix 12). The nonmonophyletic species mostly originated from species‐rich families, such as Burseraceae, Myristicaceae, Moraceae, Phyllanthaceae, Lauraceae, Sapindaceae, and Annonaceae. The ordinal‐level phylogeny of flowering plants shows the relationship between orders of flowering plants and the grouping of these orders (Figure [2](#ece34875-fig-0002){ref-type="fig"}). The matK marker misplaced Myrtales and failed to separate Laurales from Magnoliales. Meanwhile, the rbcL marker misplaced Aquifoliales and grouped Malpighiales and Brassicales into one monophyletic clade. This marker also failed to make Santalales a monophyletic clade. However, this marker successfully separated Laurales from Magnoliales. Finally, the combination of matK and rbcL improved the topologies of the tree and put nearly all orders into the right position compared to APG III phylogeny. {ref-type="ref"}). The dash lines indicate that the two orders are not clearly separated. ^\*^Santalales in rbcL phylogeny tree is a nonmonophyletic clade](ECE3-9-1858-g002){#ece34875-fig-0002} 4. DISCUSSION {#ece34875-sec-0008} ============= 4.1. Recoverability and quality of *matK* and *rbcL* barcodes {#ece34875-sec-0009} ------------------------------------------------------------- The *rbcL* universality as DNA barcode observed in this study confirms that DNA sequences could be easily obtained with *rbcL* primers from a wide range of tropical plant species (e.g., Gonzales et al., [2009](#ece34875-bib-0035){ref-type="ref"}; Lahaye et al., [2008](#ece34875-bib-0057){ref-type="ref"}; Parmentier et al., [2013](#ece34875-bib-0071){ref-type="ref"}). In contrast to *rbcL*, *matK* seems to be less suitable for tropical floras compared to temperate one (e.g., Bruni et al., [2012](#ece34875-bib-0010){ref-type="ref"}; de Vere et al., [2012](#ece34875-bib-0024){ref-type="ref"}; Gonzales et al., [2009](#ece34875-bib-0035){ref-type="ref"}). This might be due to higher evolutionary rates in tropical compared to temperate plants (Gillman, Keeling, Gardner, & Wright, [2010](#ece34875-bib-0034){ref-type="ref"}). The PCR of *matK* performed in this study was using two pairs of primers which were found to be effective to generate DNA barcodes from specific taxa, such as *Tetrastigma* (Fu, Jiang, & Fu, [2011](#ece34875-bib-0030){ref-type="ref"}), *Hedyotis*(Guo, Simmons, But, Shaw, & Wang, [2011](#ece34875-bib-0036){ref-type="ref"}), or Asteraceae (Gao et al., [2010](#ece34875-bib-0031){ref-type="ref"}). These primers, however, became less effective when they were used for a wide range of species (Gonzales et al., [2009](#ece34875-bib-0035){ref-type="ref"}; Kress et al., [2010](#ece34875-bib-0053){ref-type="ref"}). A certain primer pair did not always yield a PCR product in all members of a group of seemingly closely related taxa, indicating that the primers themselves are not conserved. The use of *matK* as a barcode has been criticized mainly because universal primers are not available (e.g., Bafeel et al., [2011](#ece34875-bib-0004){ref-type="ref"}; Dong et al., [2015](#ece34875-bib-0025){ref-type="ref"}). A study by Fazekas et al. ([2008](#ece34875-bib-0028){ref-type="ref"}) showed a relatively high rate of sequencing success for this marker after using up to 10 primer pairs. The usefulness of *matK* primers is proven when they are used in specific species or taxa, such as *Camellia sinensis* (Stoeckle et al., [2011](#ece34875-bib-0078){ref-type="ref"}), Lamiaceae (De Mattia et al., [2011](#ece34875-bib-0023){ref-type="ref"}), or palms (Jeanson, Labat, & Little, [2011](#ece34875-bib-0046){ref-type="ref"}). In a review of the best barcode for plants, Hollingsworth, Graham, and Little ([2011](#ece34875-bib-0045){ref-type="ref"}) indicated that *matK* still needs optimization in regard to primer combinations and needs to be adapted to specific taxonomic groups. 4.2. Plant species identification success using *matK* and *rbcL* {#ece34875-sec-0010} ----------------------------------------------------------------- As one way to evaluate the success rate of species identification, we compared the results from morphological identification with the results from molecular identification. Some authors suggested a superiority of molecular identification in comparison with morphological identification (Newmaster, Ragupathy, & Janovec, [2009](#ece34875-bib-0069){ref-type="ref"}; Stace, [2005](#ece34875-bib-0077){ref-type="ref"}). However, this study showed that DNA barcoding alone is not sufficient to assign all DNA sequences to a correct species name. Only 22%--30% of the samples were correctly assigned to the correct species, while the majority of correct identifications was limited to genus level (46%--51%). Approximately three percent of mismatch between morphological identification results and DNA identification results were found in this study that could be due to several reasons. A specimen could be misidentified when it was found to have the highest similarity to a reference sequence that was falsely identified. The mismatch between morphological and molecular identification could also happen when the taxonomist misidentified the voucher. Morphological identification is difficult in the absence of certain features, such as flowers or fruits, especially when dealing with species‐rich groups. A high percentage of nonfertile material is particularly common in ecological projects such as ours. In the case of incorrect morphological identification, the herbarium vouchers of corresponding samples should be verified morphologically once again. The success of species identification using DNA barcoding depends very much on the taxa in question, as much as the utilized marker. For example, in this study, the family Piperaceae resulted in high species‐matched identification when using *matK* (60%) but no success at all when using *rbcL*. Meanwhile, for the family Asteraceae, the species‐matched identification was higher with *rbcL* (50%) than with *matK* (30%). Another factor affecting the success of species identification using DNA barcoding is the availability of nucleotide data of the corresponding taxa in the DNA sequences database such as GenBank and BOLD. Through this study, 303 newly barcoded tropical plant species have been uploaded to BOLD. Forty‐one percent of the 772 species investigated in this study still had no nucleotide data in BOLD and Genbank. Thus, a significant proportion of samples belonging to species which were not yet recorded in the reference databases lead to increased rates of unassigned samples. Incorrect specimen assignment is more often due to the incompleteness of molecular datasets rather than the data analysis (Bruni et al., [2010](#ece34875-bib-0009){ref-type="ref"}; Burgess et al., [2011](#ece34875-bib-0011){ref-type="ref"}; Cowan & Fay, [2012](#ece34875-bib-0018){ref-type="ref"}). An accurate and complete molecular database, especially for plant species, is still far from being achieved in the present state. Such a database will hopefully be developed in the future as many studies and projects of plant DNA barcoding are going on (e.g., <http://botany.si.edu/projects/DNAbarcode/intro.htm>; <http://xmalesia.info/index.html>). 4.3. Discriminatory power of *matK* and *rbcL* {#ece34875-sec-0011} ---------------------------------------------- None of the markers used in this study successfully obtained a DNA barcoding gap. All of the minimum values of interspecific divergence obtained from three different markers were lower than the maximum values of intraspecific divergence. In studies of DNA barcoding of specific plant taxa, for example, *Ludwigia*(Ghahramanzadeh et al., [2013](#ece34875-bib-0033){ref-type="ref"}), *Abies,* *Cupressus* (Armenise, Simeone, Piredda, & Schirone, [2012](#ece34875-bib-0003){ref-type="ref"}), and *Tetrastigma* (Fu, Jiang, & Fu, [2011](#ece34875-bib-0030){ref-type="ref"}), the distribution of intra‐ versus interspecific distances was relatively well separated. Meanwhile, large‐scale plant diversity inventories (Lahaye et al., [2008](#ece34875-bib-0057){ref-type="ref"}; Parmentier et al., [2013](#ece34875-bib-0071){ref-type="ref"}) reported the absence of barcoding gaps by using a combination of potential markers. The richness of the dataset might have contributed to the wider distribution of the intra‐ and interspecific divergences which then increase the possibility of them to overlap. This implies that the sampling intensity and variety would influence the distribution of the intra‐ and interspecific variation within the dataset. Despite the absence of barcoding gaps, the barcodes generated in this study have relatively high discriminatory power. According to Hollingsworth et al. ([2011](#ece34875-bib-0045){ref-type="ref"}), most of the plant barcodes would have discriminatory power of more than 70%. Studies by Kress et al. ([2009](#ece34875-bib-0052){ref-type="ref"}) and Burgess et al. ([2011](#ece34875-bib-0011){ref-type="ref"}) showed that barcoding of distantly related taxa typically results in high levels of discriminatory power. The *matK*+rbcL marker has the highest number of discriminated species compared to *matK* or *rbcL* alone. This is because the use of two‐loci barcodes maximized the genetic variation, thus minimizing the number of identical barcodes between different species. All species that could not be discriminated have barcodes identical to other species from the same family. Identical barcodes across different genera of the same family were uncommon with *matK* but more common with *rbcL*. However, *matK* and *rbcL* mostly failed to discriminate different species from the same genus. These two plastid markers are therefore not variable enough to be effective barcodes for closely related species in certain taxa. To improve the analysis of closely related taxa, noncoding plastid genes, such as *trnH‐psbA*, could be used as an additional marker (Hollingsworth et al., [2011](#ece34875-bib-0045){ref-type="ref"}). A study by Kress and Erickson ([2007](#ece34875-bib-0051){ref-type="ref"}) showed that *trnH‐psbA* has dramatically higher sequence variability than the coding genes because it has a higher number of single‐nucleotide polymorphisms (SNPs). Hence, *trnH‐psbA* can be a suitable marker to discriminate among closely related species. Moreover, nuclear genomic regions, such as the internal transcribed spacer (ITS) region, were suggested as potential DNA barcodes by Kress et al. ([2005](#ece34875-bib-0055){ref-type="ref"}). ITS sequences generally show high levels of interspecific sequence variability (Cowan & Fay, [2012](#ece34875-bib-0018){ref-type="ref"}) and has been used successfully to classify angiosperms (Li et al., [2011](#ece34875-bib-0015){ref-type="ref"}). 4.4. The phylogeny of flowering plants of Jambi based on *matK* and *rbcL* {#ece34875-sec-0012} -------------------------------------------------------------------------- Both *matK* and *rbcL* showed high family‐level resolution, and the combination of *matK* and *rbcL* succeeded to resolve all of the families into monophyletic clades with high bootstrap value. Furthermore, the taxonomic resolution at the genus level was much lower compared to the family level which was expected. Surprisingly, the genus‐level monophyletic percentages were found slightly lower compared to the species level in all trees, except for MP and ML trees using *rbcL*. A similar study by Gonzalez et al. ([2009](#ece34875-bib-0035){ref-type="ref"}) reported larger numbers of monophyletic genera compared to monophyletic species. This difference can be explained by the fact that the proportion of distantly related species included in the dataset in this study was higher than the proportion of closely related species. Thus, the probability of resolving monophyletic‐species clades was higher than to resolve the monophyletic‐genus clade. Finally, the species‐level resolution in this study is comparable to similar studies (Gonzalez et al., [2009](#ece34875-bib-0035){ref-type="ref"}; de Vere et al., [2012](#ece34875-bib-0024){ref-type="ref"}). However, the two‐loci barcode did not improve the species‐level resolution significantly. Combining these two chloroplast markers was not sufficient to provide 100% of species monophyly. Of 76 families included in the phylogenetic tree reconstruction, Burseraceae and Phyllanthaceae were the families with the highest number of unresolved genera. Most of the species in these genera were found to have identical sequences, so they could not be separated from each other. Identical sequences between species of different genera could be common if the marker was not variable enough, such as *matK* and *rbcL*. In this study, it was revealed that *matK* and *rbcL* were not sufficiently variable for species‐rich groups. The phylogenetic trees based on the *rbcL* marker resulted in larger numbers of unresolved species than *matK*. At least eighteen species were nonmonophyletic according to *rbcL* but monophyletic according to *matK*. The unresolved species found in this study could be explained by two reasons. First, these species might have identical genetic information with other species belonging to the same genera/family. Second, these species might have higher intraspecific than interspecific divergence; thus, they were grouped with the allospecies but not with the conspecies. A number of constraints are limiting DNA barcoding of plant species including slow evolution rates (Palmer et al., [2000](#ece34875-bib-0070){ref-type="ref"}) and high incidence of hybridization (Knobloch, [1972](#ece34875-bib-0048){ref-type="ref"}). The genetic variation caused by hybridization cannot be simply detected by plastid markers (Fazekas et al., [2008](#ece34875-bib-0028){ref-type="ref"}, [2009](#ece34875-bib-0029){ref-type="ref"}). Nevertheless, none of the plant DNA markers are perfect in every case (Hollingsworth et al., [2011](#ece34875-bib-0045){ref-type="ref"}). Indeed, one of the future challenges for plant DNA barcoding is to find the most suitable marker to tackle these problems. As the DNA sequencing technology and bioinformatic tools are progressively advancing, the development of new primers will be much easier and at the end will increase the success of DNA barcoding. The application of next‐generation sequencing (NGS) technology will enhance the capability of DNA barcoding as a powerful tool in the studies of ecology, evolution, and conservation biology (Kress, Garcia‐Robledo, Uriarte, & Erickson, [2014](#ece34875-bib-0054){ref-type="ref"}). 5. CONCLUSION {#ece34875-sec-0013} ============= We conclude that the two plastid markers *matK* and *rbcL* as plant barcodes work reasonably well in identifying flowering plant species in Sumatran lowland rainforest and surrounding agricultural systems, at least up to genus level. However, there are taxa that are difficult to be distinguished using *matK* and *rbcL*. These taxa mostly belong to species‐rich clades with low interspecific divergences. DNA barcoding of closely related species results in low success, especially when using coding plastid markers, such as *matK* and *rbcL*. The success of species identification strongly depends on the availability of an accurate and complete molecular database. Such database should include sufficient barcodes for each species distributed over its entire distribution range to cover the full range of its intraspecific variability. Thus, future studies ideally include all congeneric species from a geographic region and maximize the geographic diversity of samples for each species. Moreover, utilization of supplement markers, such as *psbA‐trnH* or ITS, is highly recommended in combination with *matK* and *rbcL*. All of DNA barcodes generated in this study, comprises more than 500 species of flowering plants, are uploaded to BOLD. This, coupled with the collection of herbarium vouchers, will improve the usability of DNA barcodes for plant identification. AUTHOR CONTRIBUTIONS {#ece34875-sec-0015} ==================== F.Y.A. performed specimen collection, laboratory work, sequence analyses and wrote the manuscript. K.R. performed specimen collection, morphology identification and provided critical review of the manuscript. B.V. supported part of the laboratory work and data analysis, and revised the manuscript. S.R. provided author citation for each botanical name of species barcoded in this study and revised the manuscript. I.Z.S. provided the sample collection permit and mutual transfer agreement (MTA) documents, and revised the manuscript. H.K. and R.F. supervised the research and revised the manuscript. Supporting information ====================== ######   ###### Click here for additional data file. We would like to thank the German Research Foundation (DFG) for funding this research as part of the EFForTS project in the framework of the Collaborative Research Centre 990 (<http://www.uni-goettingen.de/crc990>). Sincere thanks to Lembaga Ilmu Pengetahuan Indonesia (LIPI), and Ministry of Environment and Forestry of Republic of Indonesia (KLHK). Special thanks to Hardianto Mangopo, Fajar Adityarama, Melanie Schmitt and Alexandra Dolynska for the technical assistances, and to the staff of Harapan Rainforest (PT. Restorasi Ekosistem Indonesia) and Bukit Duabelas National Park. We would also like to thank the Ministry of Research, Technology and Higher Education (RISTEKDIKTI) for research permission in Indonesia. DATA ACCESSIBILITY {#ece34875-sec-0014} ================== All data for the project were managed in the BOLD database in a project called "DNA Barcoding of Vascular Plants in Jambi, Indonesia" (project code CRCZ). A list of all species barcoded in this study is available as Supporting Information (Table [S1](#ece34875-sup-0001){ref-type="supplementary-material"}).

In many physiological processes, cells migrate by moving through narrow channels defined by the surrounding environment. One example is cancer metastasis, where a cancer cell squeezes through the endothelium to reach the blood stream and eventually forms a secondary tumor elsewhere in the body[@b1][@b2][@b3][@b4]. Over recent years, the study of cancer from a physical sciences point of view has drawn much attention[@b3][@b5][@b6][@b7][@b8][@b9][@b10]: Physical principles are believed to offer an alternative perspective of the disease and may help to optimize treatments[@b11] and detection[@b12]. The model we present in this paper emphasizes the role of the elastic properties of cancer cells and surrounding normal cells on the metastatic potential of the former. Our simulations show that elasticity mismatch *alone* can reproduce features of cancer cell migration observed in experiments. More precisely, we propose a multiple scale model to study the motility of individual cells in a larger cells-on-substrate assembly that comprises normal and cancer cells. We will focus on the nearly confluent scenario which describes monolayers. Understanding the behavior of cell monolayers is an important biological question that goes beyond the physics of cancer since epithelial tissues, which support the structure of embryos and organs, often have a monolayer structure[@b13]. Examples of cells-on-substrate experiments that are not directly related to cancer include studies of collective behavior[@b14][@b15], wound healing[@b9][@b16][@b17] and colony growth[@b18]. Our work is motivated by recent experiments performed by Lee *et al*.[@b9] which showed that the Young modulus of metastatic human breast carcinoma cells (MDA-MB-231) is about *three times smaller* than the one of human breast epithelial cells (MCF10A). In the same study, the authors showed that the motility of a cancer cell embedded in a confluent monolayer of mostly normal cells was much larger than in the case where the layer is made of cancer cells only. This observation was partly attributed to the fact that short speed "bursts" decorate the trajectory of the cancer cell. These bursts typically occur when a cancer cell, highly deformed due to temporary crowding by the neighboring normal cells, rapidly relaxes to a less deformed shape as the cell escapes the crowded configuration. Hence, it was proposed that the elasticity mismatch between cancer cells and normal cells significantly contributes to the observed "bursty" migration behavior and the concomitantly larger motilities of the cancer cells. In the experiments, the increased motility of the metastatic cancer cells is probably due to many factors where one is the cell mechanical properties. Additional differences between cancer and normal cells include inter cellular adhesions[@b9] and protrusion activity[@b19]. Here, the model parameters will be chosen so that all cells in the monolayer have identical properties except for their elasticity: Cancer cell are softer, normal cells are stiffer. The main results of our simulation studies demonstrate that elasticity mismatch alone is sufficient to trigger bursty migration behavior and significantly increase the motility of the soft cell. Moreover, the simulated migratory behavior of cancer cells in a layer of mostly normal cells is in qualitative agreement with the experiments[@b9]. The model that we use permits the description of very large cell shape deformations. We will show that this point is crucial to accurately describe bursty migration. The effect of deformability of cells and vesicles has recently been studied in other contexts. Many of these studies were based on a beads-and-springs model for the cell shape and focused on red blood cells in capillaries[@b20][@b21], bacteria in biofilms[@b22][@b23] and tissue growth[@b24]. Such models complement recent Potts model studies of cell sorting[@b25] and vertex model dynamical studies[@b26][@b27] of soft tissues. The phase-field model that we propose is more general than these other approaches. First, it can be easily extended to include more complexity (i.e., cell internal degree of freedom). Second, the inactive part of the dynamics is self-consistently derived from non-equilibrium thermodynamics principles. In that sense, our approach more closely resembles that used in Refs [@b6],[@b7], which focused on tumor growth, and that used in Refs [@b28], [@b29], [@b30], [@b31], [@b32], [@b33], which focused on single cell behavior. Our phase-field model approach is applied to an *assembly* of cells and it retains shape and motion details at the single cell level. Modelling the system behavior down to the single cell level is important to describe the cooperation between normal and cancer cells that leads to the bursty migration behavior and the increased motility of the latter. The paper is organized as follows. The next section contains the Results. The first subsection gives a brief summary of our cell monolayer model. Simulations results are given in the next two subsections. The second one reports the cell arrangement predicted by our model for monolayers of inactive (non-migrating) cells. These determine the initial conditions for the simulations of motile cells presented in the third subsection. There, the migratory behavior of a tagged cell in monolayers of varying cell mechanical properties is analyzed. Following is the Discussion. It contains a summary of our findings and discusses avenues to be explored. Finally, the Methods section gives extra details on the statics and dynamics aspects of the model and it gives a brief overview of the numerical procedure. Results ======= We model monolayers that comprise normal and cancer cells using a phase-field approach similar to the one recently employed by Najem *et al*.[@b30], who studied chemotropism in neural cells, by Löber *et al*.[@b31], who studied cell crawling on soft substrate, and by Larazo *et al*.[@b32][@b33], who studied the shape of red blood cells under flow in small channels. Here, we treat the monolayer as a 2D system. This is a good level of description since the cells dynamics are constrained in a region close to the plane defined by the substrate. In our model, it is assumed that cells do not grow nor divide. The idea is that each cell is described by a "field" which rapidly varies in the region of the cell boundary. Hence, we denote by *ϕ*~*n*~(*x*,*y*; *t*) the field associated with cell *n* where *x*,*y* are the two spatial coordinates and *t* is the time. An example of a cell field is shown in [Fig. 1A](#f1){ref-type="fig"}. The interior of each cell is defined by regions where *ϕ*~*n*~ = 1 while the exterior is defined by regions where *ϕ*~*n*~ = 0. At the cell boundary, *ϕ*~*n*~ interpolates rapidly between 0 and 1. From now on, all monolayer images that we will present only show the boundary of each cell except for a tagged cell that will be highlighted with a color as in [Fig. 1B](#f1){ref-type="fig"}. Details of the model are presented in the Methods section and in the [Supplementary Information](#S1){ref-type="supplementary-material"}. However, a brief overview is given in the next subsection. Model outline ------------- The most important advantages of our phase-field monolayer model are: 1) The cell boundary does not have to be tracked explicitly. 2) Extremely large deformations can be described. 3) The mechanical properties and the velocities of each cell can be modeled individually. The latter point is particularly important since one of our goals is to make connections with the experiments of Lee *et al*.[@b9]. Our approach differs from vertex models[@b26][@b27] by allowing all types of cell deformations and any degree of cell coverage. The difference between our phase field model and an equivalent Cells Potts model[@b25] is at the level of the dynamics since the former is the continuum limit of the latter. The time dependent behavior of the monolayer is described by dynamical equations for the cell fields, where is the monolayer free energy, **v**~*n*~ is the translational velocity of cell number *n* and *δ* denotes a functional derivative. Note that this equation is written in terms of dimensionless units, which will be used throughout the paper. The relationships between dimensionless and real units are detailed in the [Supplementary Information](#S1){ref-type="supplementary-material"}, briefly reviewed at the end of the Methods section. The right-hand side of [Eq. (1)](#eq1){ref-type="disp-formula"} describes cell shape dynamics, which are determined by free energy changes. Details of the model free energy are given in the Methods section and in particular, by [Eqs (7](#eq14){ref-type="disp-formula"}) and ([10](#eq19){ref-type="disp-formula"}). The free energy contains several parameters and its minimum determines the prefered state of the system. Briefly, one parameter controls the elastic response of each cell to shape deformation (*γ*~*n*~ in [Eq. (7)](#eq14){ref-type="disp-formula"}). Another controls the preferred radius of the cells (*R* in see [Eq. (7)](#eq14){ref-type="disp-formula"}), which tend to be circular when they are not perturbed by other cells. Also, there is a parameter that controls the energy penalty for overlapping cells (*κ* in [Eq. (10)](#eq19){ref-type="disp-formula"}, which is chosen to be large). Note that the interactions between cells are strictly repulsive. The velocity of each cell is the sum of two distinct contributions as described in [Eq. (11)](#eq20){ref-type="disp-formula"}: 1) The inactive part, **v**~*I*~, is due to the interaction force exerted by the other cells. Like the cell shape dynamics, the inactive part of the velocity is determined by free energy changes. 2) The active, or self-propulsion, part of the velocity, **v**~*A*~, is due to the cell engine. The relative strength of the two contributions to the velocity is determined by another parameter (*ξ* in [Eq. (12)](#eq21){ref-type="disp-formula"}) which also controls the maximum cell shape deformation. The active part of the velocity is chosen so that the motion of isolated cells on the substrate is described by a persistent random walk[@b34][@b35] where the characteristic cell speed and the reorientation statistics match the experimental observations[@b9]. Further, we assume that there is a large separation of time scales between the cell shape relaxation (fast) and the cell translation dynamics (slow). This approximation is physical since 1 *μm* deep indentations on cells typically relax within seconds[@b36] and the motile cells we model translate by 1 *μm* within minutes[@b9]. To highlight the role of cell elasticity mismatch, we considered 2 types of cell monolayers assemblies: a single cancer cell in a layer of normal cells (hereafter referred as the "soft-in-normal" case) and normal cells only (hereafter referred as the "all-normal" case). We performed simulations at two packing fractions, *ρ* = 0.85 and 0.9, describing nearly confluent monolayers, where, where *A*~*B*~ is the area of the simulation box. Overall, we will present a total of 4 simulations, each of which contains *N*~*cell*~ = 72 cells. All model parameters are listed in [Table 1](#t1){ref-type="table"} and explained in details in the Methods section and in the [Supplementary Information](#S1){ref-type="supplementary-material"}. Our aim is to isolate the effect of the mechanical properties of motile cells on their migratory behavior. Hence, *all parameters will be identical for both types of cells with the exception of the parameter which controls the cell stiffness* (we set *γ*~cancer~/*γ*~normal~ = 0.35, as observed experimentally). That includes the sequence of random numbers which determines the initial positions of the cells and the reorientation events. Hence, in the absence of elasticity mismatch and at the same packing fraction, the soft-in-normal and all-normal simulations would give identical results. Aging ----- In the first stage of all simulation, the cells are randomly placed in the simulation box at the appropriate packing fraction. All cells are initially circular with radius *R* given in [Table 1](#t1){ref-type="table"}. Any cell whose center is within a distance smaller than *R* to the center of a previously placed cell is randomly re-assigned to a new position. At a given packing fraction, the initial position of the cells for the soft-in-normal and all-normal cases are enforced to be the same. In the initial configuration, the monolayer is far from equilibrium since many cells overlap. Hence, we allow it to relax by numerically propagating [Eq. (1)](#eq1){ref-type="disp-formula"} *without* the term that gives rise to self-propulsion, **v**~*n*,*A*~ in [Eq. (11)](#eq20){ref-type="disp-formula"}. During this "aging" period, the system rearranges to minimize the free energy and the cells act as mutually immiscible "dead" droplets. [Figure 2](#f2){ref-type="fig"} shows the final configuration of all 4 monolayers after aging and the net displacement of each cells. A tagged cell is highlighted with a black boundary and a colored filling. This cell always starts in the middle of the simulation box. This is the cancer cell in the soft-in-normal simulations. The soft cancer cell is colored in Green and normal cells in Blue. The arrows that report the displacement of the cells during aging are colored in the same way. The length of the arrows is equal to *twice* the cell displacements magnitude. Movies of all 4 aging simulations are given in the [Supplementary Information](#S1){ref-type="supplementary-material"}. Note that the cell displacements do not seem to correlate with the elasticity, but rather on the initial nearest-neighbor configuration of each cell. This is most easily seen in the [Supplementary Movies](#S1){ref-type="supplementary-material"}. All [supplementary movies](#S1){ref-type="supplementary-material"} are available at *http://dx.doi.org/10.6084/m9.figshare.1439474*. On the other hand, the deformation of the tagged soft cell in the soft-in-normal case is clearly larger than that of the other cells and that of the tagged normal cell in the all-normal case. The distortion of each cell can be characterized by the length of their interface, or cell perimeter. With the cell field *ϕ*~*n*~ given by Eq. (S2), it is simple to show that the perimeter is proportional to the following measure, which is equal to 1 for perfectly circular cells. Because the cell area is nearly conserved, the cells can only reduce their interactions energy by deforming their boundaries. The perimeter increases with increasing cell deformation. [Figure 2](#f2){ref-type="fig"} also shows the distribution of *L*~*n*~ values at the end of the 4 aging simulations for all 72 cells. Note that the value of the tagged cell has been singled out and it is indicated by an arrow. Comparing the two rows that correspond to the soft-in-normal and the all-normal cases clearly shows that the soft cell undergoes deformations that are significantly larger than the deformation of the surrounding normal cells. At the end of the all-normal aging simulations, the system is probably close to a metastable state which differs from true equilibrium. In that latter case, the distribution of *L*~*n*~ values in the bottom row of [Fig. 2](#f2){ref-type="fig"} (all-normal simulations) would show a single peak since the minimum free energy state of the system has a hexagonal symmetry where all cells have the same perimeter. This equilibrium state may not be attainable, even with long simulation times. Motile cells ------------ We now present the results obtained for motile cells simulations that correspond to the case where the velocity of each cell is given by [Eq. (11)](#eq20){ref-type="disp-formula"} *including* the self-propulsion term, **v**~*n*,*A*~. The final configurations after aging shown in [Fig. 2](#f2){ref-type="fig"} are used as initial conditions for the motile cell simulations. [Figure 3](#f3){ref-type="fig"} summarizes the motile cell simulations and is the main result of our work. It clearly shows that cell elasticity mismatch plays a significant role on the cells dynamics. Again, we will focus on the tagged cell. The color scheme is the same as before; the Green curves correspond to the soft cancer cell (soft-in-normal case) and the Blue curves correspond to the tagged normal cell (all-normal case). The path traced by the soft cancer cell embedded in the soft-in-normal simulations largely differs from the path traced by its normal cell counterpart in the all-normal simulations. This is shown in the square boxes in [Fig. 3A](#f3){ref-type="fig"}. In particular, the space explored by the soft cancer cell exceeds that of its normal counterparts. Second, the soft cancer cell undergoes very large deformations (see the time evolution of *L* in [Fig. 3A](#f3){ref-type="fig"}) compared to the tagged normal cell at the same packing fraction. More precisely, in the soft-in-normal case, the perimeter of the soft cell reaches *L* = 1.5 which is 150% that of an undeformed cell. This contrasts the all-normal case where the same cell is normally stiff and its perimeter does not exceed *L* = 1.1. Third, the instantaneous velocity of the soft cancer cell contains several spikes that correspond to short speed bursts. These are shown in the time evolution of in [Fig. 3A](#f3){ref-type="fig"}. A comparison with the evolution of *L* shows that these bursts occur simultaneously with a rapid decrease of the cell perimeter; as the cell rapidly relaxes to a lower energy, more circular, configuration. Bursts are clearly observed in the bottom part of [Fig. 3A](#f3){ref-type="fig"} (and the corresponding [Supplementary Movie)](#S1){ref-type="supplementary-material"} for the soft-in-normal case at the largest packing fraction. Note that, at the two packing fractions considered, the tagged cell does not display any speed bursts in the all-normal case. In this case, increasing the packing fraction simply reduces the mean instantaneous velocity of the normal cell. These observations qualitatively reproduce many of the experimental features reported by Lee *et al*.[@b9] for confluent monolayers of live cells. Their experimental data is reported in [Fig. 3B](#f3){ref-type="fig"}, with permission. In the experiment, 2 bursts where observed in a 16 hours time window. Converted back to real units (see Sec. 0.3), the simulation results shown in [Fig. 3A](#f3){ref-type="fig"} corresponds to a ≈40 hours time window during which 6--8 instantaneous velocity spikes (bursts) are observed. Of course, a statistically meaningful comparison between the simulation and experimental results cannot be done due to the small number if bursts observed experimentally and theoretically. However, [Fig. 3A,B](#f3){ref-type="fig"} show that the simulations at *ρ* = 0.90 and the experiments are strikingly similar. One crucial point of our work is that the *only* difference between the cases considered, at fixed packing fraction, is in the cell elastic properties. Hence, our results support the hypothesis of Lee *et al*.[@b9] that cell elasticity mismatch *alone* can enhance the motility of the softer cells. [Fig. 3C](#f3){ref-type="fig"} shows snapshots of the soft-in-normal simulation for *ρ* = 0.85 (top panels) and *ρ* = 0.90 (bottom panels). The snapshots are reported at times labeled by the marks i, ii and iii in [Fig. 3A](#f3){ref-type="fig"}, which respectively corresponds to a time before the soft cancer cell undergoes a speed burst, after the speed burst and at a later time when the soft cell is largely deformed for the *ρ* = 0.90 case. Note that, at any time shown for the lowest packing fraction snapshots, the soft cell does is not largely deformed and many "empty spaces" are seen in the simulation box. To further investigate the mechanism that leads to bursty migration, we used the soft-in-normal simulation at the largest packing fraction to generate [supplementary movies](#S1){ref-type="supplementary-material"} that show the soft cancer cell, its nearest neighbors and their respective instantaneous velocities. [Figure 3D](#f3){ref-type="fig"} reports three frames; before, during and after the speed burst that occurs right after the time marked by iii in [Fig. 3A](#f3){ref-type="fig"}. The Black arrow in [Fig. 3D](#f3){ref-type="fig"} shows the instantaneous velocity of the cells. It turns Red when the velocity magnitude of the soft cancer cell is larger than the magnitude of the active part alone, which is the speed the soft cell would have if it was isolated. Further, the figure shows that the increase in velocity is in part due to the fact that two neighboring cells effectively pinch the rear part of the cell, which is opposite to its velocity, and thereby increase its forward motion. In fact, the two rear-end cells undergo a T1 topological swap; a processes which has been observed in cell rearrangement in tissues[@b37] and in the relaxation of topological defects in 2D[@b38]. We next analyze in more details the behavior of the tagged cell for the simulations where the effect of the elasticity mismatch is more apparent (for *ρ* = 0.90). Simulations at that packing fraction were run for a much longer time (*t* = 2.0 × 10^6^ which corresponds to ≈200 hours in real time). These longer runs are used to perform a meaningful statistical analysis of the instantaneous velocity of the tagged cell. [Figure 3E](#f3){ref-type="fig"} illustrates the correlation between cell perimeter change and instantaneous velocity. The long simulations are used to bin the perimeter change Δ*L* (between succesive times) according to the instantaneous velocity of the cell, . The figure reports , the average perimeter change in each velocity bin. Hence, negative Δ*L* corresponds to a cell that relaxed by decreasing its perimeter. The tagged normal cell (Blue circles) does not achieve the largest velocities of the tagged cancer cell (Green triangles). More importantly, the largest velocities of the cancer cell correlate with a large decrease in cell perimeter, particularly at , which is the magnitude of the self-propulsion speed. Further, the long runs were used to calculate probability distributions for the *x* and *y*-components of the instantaneous velocity of the tagged cell in the soft-in-normal and all-normal cases. The results are shown in [Fig. 4](#f4){ref-type="fig"}. The first panel of [Fig. 4](#f4){ref-type="fig"} shows the probability distribution (scaled such that a Gaussian distribution is a straight line) of the active part of the cell velocity which is the velocity the cell would have if it was alone on the substrate. The curve is the exact result, where *i* = *x* or *y* and the marks are the numerical results. Note that the distribution of the active part of the velocity is extremely non-Gaussian. However, the second panel in [Fig. 4](#f4){ref-type="fig"} shows that, for the all-normal case, the self-averaging induced by the interactions with the other cells transform the instantaneous velocity distribution into a Gaussian one, *P*~*G*~(*v*~*i*~) where the standard deviation of the distribution, *σ*~*G*~, is a fitting parameter. On the other hand, the last panel in [Fig. 4](#f4){ref-type="fig"} shows that a Gaussian fit (Black line) cannot describe the instantaneous velocity distribution of the soft cancer cell in the soft-in-normal case (see the last panel in [Fig. 4](#f4){ref-type="fig"}). Note that the simulation data has long tails (i.e., higher probability for large velocities) that are not accounted for by the Gaussian fit. We propose that the soft cancer cell is in one of two regimes. In the first and most probable one, it behaves like a normal cell and its velocity is described by a Gaussian. In the other, the cell undergoes bursty migration and the self-propulsion adds to the Gaussian contribution of the velocity. Mathematically, this is described by where *ζ* represents the fraction of time that the soft cancer cell is bursty. The right panel of [Fig. 4](#f4){ref-type="fig"} shows the fit of the data with this 2-regimes model (Red curve) where the standard deviation of *P*~*G*~ was obtained from the all-normal simulations and hence, *ζ* is the only fitting parameter which we found to be *ζ* = 0.038. This implies that the soft cell is in the bursty regime ≈3--4% of the time, in qualitative agreement with [Fig. 3A](#f3){ref-type="fig"}. Note the point near v~*i*~ ≈ 0.015 in [Fig. 3B](#f3){ref-type="fig"} is the only one that is significantly outside of the Gaussian distribution. In fact, this point is due to a single speed burst observed for the tagged normal cell at *ρ* = 0.9. In comparison, the soft tagged cell at the same packing fraction shows about 40 speed bursts. Of course, other distributions with a longer than Gaussian tail can fit the soft-in-normal data in [Fig. 4](#f4){ref-type="fig"}. In particular, the Student-t distribution[@b39] with *β* = 7 degrees of freedom gives an equally good fit. The Student-t distribution arises when *β* + 1 samples are drawn from a Gaussian distribution of unknown variance. We find it intriguing that the number of degrees of freedom, *β* = 7, that fits our data is very close to the mean coordination number of each cell. We think that this may be due to the fact that the soft cell in the soft-in-normal case only "feels" its immediate nearest neighbors whereas in the all-normal case, all cells feel each other. The motilities of the tagged cells can be extracted from their trajectories. More precisely, the mean-squared displacement (see [Eq. (14)](#eq28){ref-type="disp-formula"}) of any tagged cell can be used to calculate an effective diffusion coefficient. [Figure 5](#f5){ref-type="fig"} shows the mean-squared displacement of the tagged cell for the long simulations at *ρ* = 0.90 which was computed using a time averaging procedure, where must be smaller than the total simulation time (). The curves in [Fig. 5](#f5){ref-type="fig"} are smooth, which is due to the fact that the average is performed over a long time interval. The effective diffusion coefficients that are reported in the figure are obtained from the long-time part of the curves in [Fig. 5](#f5){ref-type="fig"}. The tagged soft cancer cell in the soft-in-normal simulation has the larger diffusion coefficient, which is about a factor of 1.5 larger than the one of the tagged normal cell in the all-normal simulation. Also note that the inset of [Fig. 5](#f5){ref-type="fig"} compares the analytical prediction of the mean squared-displacement given by [Eq. (14)](#eq28){ref-type="disp-formula"} with the one calculated from the trajectory that the tagged cell would have if it was alone on the substrate for the same simulation time. The analytical result gives *D*~*eff*~ (isolated) = 5.6 × 10^−14^ *m*^2^/*s* which is about one order of magnitude larger than the effective diffusion coefficient for the tagged soft cancer cell or normal cell in a dense monolayer. The comparison with the numerical calculation shows that the latter gives an effective diffusion coefficient which is about 10% smaller than the analytical prediction. This discrepancy is due to the finite sampling of reorientation events. Note that it is particularly difficult to sample the tail of the exponential distribution that governs the reorientation statistics (see. [Eq. (13)](#eq27){ref-type="disp-formula"}). This issue probably also affects the calculated diffusion coefficients for the tagged cells in monolayers reported in [Fig. 5](#f5){ref-type="fig"} which may in fact have slightly larger values if the simulations were run much longer. However, the important point here is the *relative* values of *D*~*eff*~ for the soft-in-normal and all-normal cases. Discussion ========== In this paper, we proposed a multiple scale model to describe cell dynamics in monolayers with any degree of confluence. Results obtained with the model focused on a nearly confluent monolayer comprising of cancer cells and normal cells. Our main goal was to assess the role of cell elasticity mismatch on the migration potential of the cells (i.e., the metastatic potential of the cancer cells), in light of the recent experimental studies of Lee *et al*.[@b9] who observed that human breast carcinoma cells (MDA-MB-231) embedded in a monolayer of mostly normal human breast epithelial cells (MCF10A) displayed an increased motility in comparison to the case where they are alone on the substrate. This larger migration potential was partly attributed to the fact that the cancer cells, whose Young modulus is about 3 times smaller compared to the normal cells, can deform their shape easily and squeeze between neighboring normal cells leading to bursty migration. The simulation results obtained with the model treated cancer and normal cells identically, except for their elastic constant. This allowed us to quantify the role of elasticity mismatch *alone*. The most important results that we obtained are: 1) Our minimum energy model shows that the motility (quantified by an effective diffusion coefficient) of a soft cancer cell in a monolayer of normal cells can be 50% larger (see [Fig. 5](#f5){ref-type="fig"}) as the one of a normal cell in the monolayer. 2) The trajectory of the soft cell in a layer of normal cells is decorated by speed bursts, where the velocity significantly deviates from its average value, in qualitative agreement with the experimental observations. 3) The velocity distribution of the soft cell in a layer of normal cells shows longer tails that are inconsistent with a Gaussian distribution. Of course, in our simulations, elasticity mismatch is solely responsible for these effects. Several speed bursts are shown in the bottom two panels of [Fig. 3A](#f3){ref-type="fig"} (Green curve). We have further observed that most bursts are induced by a T1 topological swap where two normal cells, initially separated by the cancer cell, move toward each other and connect by pinching the rear-end of the cancer cell. This process gives a net push to the cancer cell (see [Fig. 3D](#f3){ref-type="fig"} and the corresponding [Supplementary Movie](#S1){ref-type="supplementary-material"}) which rapidly relaxes to a more circular shape. We observed 6--8 bursts in the soft-in-normal simulation at the highest packing fraction (*ρ* = 0.90) over a time scale that correspond to ≈40 hours in real time. Lee *et al*.[@b9] reported 2 bursts over an observation period of 16 hours (see [Fig. 3B](#f3){ref-type="fig"}). Of course, the small number of bursts prevents us to compare the frequency/amplitude of our bursts with the experimental ones in a statistically meaningful way. Nevertheless, the qualitative agreement between the experiments and our simulations at *ρ* = 0.90 is striking. Moreover, our simulations allows us to quantify the correlation between the instantaneous velocity of the tagged cell and the instantaneous change in cell perimeter (see [Fig. 3E](#f3){ref-type="fig"}). A calculation shown in Sec. S6 of the [Supplementary Information](#S1){ref-type="supplementary-material"} further shows that most bursts can be described as short events were the tagged cell moves with the velocity it would have if it was alone on the substrate *plus* an extra contribution that arises from the cell displacement due to the rapid cell shape change (i.e., when a highly deformed tagged cell rapidly relaxes to a more circular shape). To some degree, our results are sensitive on the choices of the model parameters. In particular, for the lowest packing fraction considered, the soft cancer cell does not show clean speed bursts. This observation seems to suggest that confluence is required for bursty migration. There is much more empty space between cells in the *ρ* = 0.85 simulations compared to the *ρ* = 0.90 (compare the top and bottom rows of [Fig. 3C](#f3){ref-type="fig"} and the [Supplementary Movies](#S1){ref-type="supplementary-material"}). Further note that *ρ* does not necessarily correspond to the cell packing fraction in the real system. The difference arises from the fact that in the model, the cell is an elastomer defined by a single elastic constant. On the other hand, the elastic restoring force that prevents real cells from migrating through narrow pathways primarily comes from the nucleus since the cytoplasm is much softer[@b40]. Future studies will be performed to quantify the role of packing fraction more precisely. If our results are sensitive to the choice of *model parameters*, they should not be sensitive to the choice of the model. In fact, we believe that our results could have been obtained using a Cell Potts description[@b25]. One important difference between our work and the one of Lee *et al*.[@b9] is the following. Recall that our goal was to isolate the role of cell elasticity. Hence, we set the active, self-propulsion, part of the velocity of all cells to be the same. However, in the experiment, that does not seem to be the case. In fact, the cancer cell appears to move much less by itself than the normal cells[@b9]. Hence, when it is embedded in a layer of normal cells, its motility may primarily be increased because it gets pushed around by the other, more motile cells. In our simulations, the strength of the self-propulsion of the cancer and normal cells is identical. Hence, when any type of cells are brought together in a dense monolayer, to first order, their motility decreases since the cells act as obstacles to each other. In the future, we want to study the effect of active velocity mismatch to understand how a soft cell that is embedded in a layer of normal cells (with a different active velocity magnitude) could cross over from the case where it is more (less) motile in the monolayer in comparison to the case where it is alone on the substrate. Another difference between our simulation studies and the experiments of Lee *et al*. is that cell-cell adhesion and more importantly, how it varies with cell type, was neglected. In fact, the experiment showed that the magnitude of the bursts is largest when normal cells adhere with each other, but not with the soft cancer cell[@b9]. Cell-cell adhesion can be included in our phase field monolayer model, but this goes beyond the objective of the current work. We are planning to include it in future extensions of the model. Note that in real monolayers, the integrity of a cluster of cells is maintained through cell adhesion. In our model, the cells remain clustered due to the boundary conditions. Hence, our model includes adhesion in its simplest form. One important advantage of the phase-field description for monolayers of motile cells that we propose is that it scales favorably with the number of fields one may want to add to each cells to include details that we left over for simplicity. A natural example are fields that describe the density of actin and myosin in the cytoplasm or focal adhesions on the cell surface which can generate protrusive forces, contractile forces and strain in the substrate[@b41][@b42]. Such cell internal degrees of freedom have been included in recent 2D models of a single migrating cell[@b28][@b29][@b31]. Including new fields that describe such cell internal processes does not increase the computational effort significantly, as long as they do not require using a finer mesh. Generalizing our model to 3D may also help to understand why motile cells appear to migrate using drastically different mechanisms on 2D substrates compared to the case when they are embedded inside a 3D collagen matrix[@b35][@b44][@b45]. Phase-field models can be extended from 2D to 3D, but the concomitant increase in numerical cost usually requires the use of more sophisticated numerical methods. Methods ======= Model free energy ----------------- We treat each cell as a 2D soft body for which the equilibrium shape minimizes the following free-energy, where it is understood that, by *ϕ*~*n*~, we mean *ϕ*~*n*~(*x*,*y*; *t*). The form of [Eq. (7)](#eq14){ref-type="disp-formula"} guarantees that the preferred values for the cell fields are *ϕ*~*n*~ = 0 and 1. The length over which *ϕ*~*n*~ varies from 0 to 1, *λ*, corresponds to the width of the boundary of any cell. In the model, non-interacting cells tend to be circular with a radius *R*. Energetic costs associated to changes in cell area are determined by *μ*~*n*~. We next show that *γ*~*n*~ is the parameter that controls the elasticity of the cells. We calculated the energy cost predicted by the model and that results from a sinusoidal deformation of the cell boundary (see Sec. S1 of the [Supplementary Information](#S1){ref-type="supplementary-material"}). For a deformation wavelength equal to 2*πR*/*k*, where *k* is a wavenumber, and an amplitude equal to *ε*, the energetic cost is where periodicity imposes that *k* is an integer larger than 1 and where we assume that *ε*/*R* ≪ 1. For identical cell sizes, interface widths and deformations, depends on cell type only through *γ*~*n*~. Hence, *γ*~*n*~ is the parameter that controls the cell stiffness. Note that the energy cost due to shape deformation can scale with the wavenumber, *k*, with a different power law if the curvature energy is taken into account[@b46] (in which case it scales like *k*^4^) or if the cell interior is treated as an elastic medium[@b47] (in which case it scales like ). Here, these contributions are neglected since our focus is not on the details of the underlying restoring force. We now comment on the last term in [Eq. (7)](#eq14){ref-type="disp-formula"} which constrains the area of the cell. In the simulations, *μ*~*n*~ will be chosen to be large enough so that the cells will prefer to deform when they are brought together rather than shrink. It is not crucial that the cell area be exactly conserved in our description of cell monolayers. In reality, it is the cell volume which should be conserved for cells that neither grow nor divide. In our 2D model, the cell area can be interpreted as the area of a 3D cell "projected" onto the substrate. This projected area can deviate from its preferred value as long as the cell thickness above the substrate is simultaneously adjusted such that the overall cell volume remains constant[@b25][@b26]. The free energy given by [Eq. (7)](#eq14){ref-type="disp-formula"} describes each cells individually. The total free energy of the monolayer is, where describes the interaction between cells that prevents them from overlapping. Both *ϕ*~*n*~ and *ϕ*~*m*~ are non-zero in a region where cells *n* and *m* overlap and the resulting energy cost is controlled by *κ* (like *μ*~*n*~, *κ* will be chosen sufficiently large so that cells will deform when they are brought together rather than overlap). In terms of *ϕ*, [Eq. (10)](#eq19){ref-type="disp-formula"} is the lowest order term that gives the desired repulsion between cells. More details on the model parameters and their meaning are given in the [Supplementary Information](#S1){ref-type="supplementary-material"}. Dynamics -------- The motion of each cell within the monolayer is described by [Eq. (1)](#eq1){ref-type="disp-formula"}. The time-dependent cell velocity, **v**~*n*~, is chosen to be spatially uniform for simplicity. Our model fall into the category of "model A" in the classification scheme of Hohenberg and Halperin[@b48] for dynamical critical phenomena, but with an extra convective term. The right-hand-side of [Eq. (1)](#eq1){ref-type="disp-formula"} determines how rapidly a deformed cell returns to its circular, equilibrium shape. Such a relaxation time scale can be determined from cell viscosity measurements (for a recent example, see Ref. [@b36]) or can be estimated by equating the shape relaxation rate with the viscous dissipation of the water inside the cell (see Ref. [@b49]). The velocity of each cell is divided in two parts, where **v**~*n*,*I*~ and **v**~*n*,*A*~ are the inactive and active (self-propulsion) parts of the velocity of cell *n*, respectively. The inactive part is due to forces exerted on cell *n* by the other cells while the self-propulsion part is due to internal processes that require energy consumption. The constitutive equations for **v**~*n*,*I*~ can be determined from thermodynamic principles. With **v**~*n*,*A*~ = 0, the free energy of the system, which is an assembly of dead deformable droplets, should be a strictly decreasing function of time. In Sec. S2 of the [Supplementary Information](#S1){ref-type="supplementary-material"}, we use this condition to show that, where *ξ* is interpreted as the friction between the cells, the substrate and the surrounding water. Together with [Eq. (1)](#eq1){ref-type="disp-formula"}, the last equation guarantees that the monolayer will tend toward thermodynamic equilibrium in the absence of active forces. At this point the cells do not flow, **v**~*n*,*I*~ = 0, and their boundaries do not move, . Of course, live cells never reach thermodynamic equilibrium. This is taken into account by which is chosen such that the instantaneous velocity of isolated cells has a constant magnitude; . On the other hand, the time dependence of the cell motion is due to the orientation of which we describe as a random process where the time interval between reorientation events, , follows an exponential distribution given by, where *τ* is the average time between two reorientations. Hence, in our model, an isolated cell moves in a straight line with constant speed between two reorientation events. [Eq. (13)](#eq27){ref-type="disp-formula"} can also be used to calculate the mean-squared displacement of cell *n* when it is isolated on the substrate, where x~*n*~(*t*) is the center of cell *n* and implies an average over an ensemble of isolated cell trajectories or equivalently, a longer time average over a single isolated cell trajectory as given by [Eq. (6)](#eq11){ref-type="disp-formula"}. The long time behavior of the mean-squared displacement is linear in time and hence, isolated cells can be characterized by an "effective" diffusion coefficient in two dimensions, . Rather than depending on temperature, the effective diffusion coefficient depends on internal processes that require energy consumption and that determine *v*~*A*~ and *τ*. C. Model Parameters and Numerical Considerations ------------------------------------------------ All model parameters are chosen to be identical for both types of cells with the exception of *γ*~*n*~. The simulation will be performed by propagating [Eq. (1)](#eq1){ref-type="disp-formula"} numerically on a uniform mesh. [Table 1](#t1){ref-type="table"} summarizes the model parameters used in the simulations. As stated in the Results section, the parameters are obtained from the experiments or by invoking physical approximations. Additional details are given in Sec. S4. The determination of *ξ* requires further comments. Consider two cells that move toward each other in an "head on collision" manner so that they have a constant and opposite active force parallel to their separation vector, see [Fig. 6](#f6){ref-type="fig"}. As the two cells start interacting/touching, a) they deform and b) they decelerate. Hence, the cells will reach a conformation where the force term due to interactions with the other cells exactly cancels the active force so that **v**~*I*~ + **v**~*A*~ = 0. For increasing *ξ*, cells need to be increasingly deformed for the interaction forces to decrease the velocities. [Figure 6](#f6){ref-type="fig"} reports the total velocity of one cell for the head-on collision just described with *ξ* *=* 1.5 × 10^3^ (as listed in [Table 1](#t1){ref-type="table"}). The long-time maximum deformation of the cells is in qualitative agreement with the largest cell deformation observed in monolayers. Note that the two cells end up in a metastable configuration from which they can escape at long time due to numerical error build-up. Fig. 6 has an accompanying [supplementary movie](#S1){ref-type="supplementary-material"}. Simulations of the monolayer model are performed in a square simulation box of area *A*~*B*~ = *N*~*mesh*~ × *N*~*mesh*~ with *N*~*mesh*~ the number of mesh points along one axis of the box. [Eq. (1)](#eq1){ref-type="disp-formula"} is integrated on a mesh, periodic boundary conditions are used and additional details of the numerical procedure are given in the [Supplementary Information](#S1){ref-type="supplementary-material"}. Importantly, our simulation results can be converted back to real units using the following simple arguments. The cells we are modeling have a radius of the order of 10 *μm* and our cells have a radius of 49 mesh points. Hence, the distance between mesh points in our simulations is ≈0.2 *μm*. In the experiment, the average instantaneous velocity of normal cells in a confluent monolayer of mostly normal cells is ≈10 *μm*/*hour* (see [Fig. 3B](#f3){ref-type="fig"}). Alone on the substrate, it should be larger since the motion of any given cell is not blocked by the others. Hence, we assume that *v*~*A*~ = 20 *μm*/*hour* which means that *t* = 1 in our simulations is equivalent to 0.36 *s* in real time. Additional Information ====================== **How to cite this article**: Palmieri, B. *et al*. Multiple scale model for cell migration in monolayers: Elastic mismatch between cells enhances motility. *Sci. Rep*. **5**, 11745; doi: 10.1038/srep11745 (2015). Supplementary Material {#S1} ====================== ###### Supplementary Information ###### Supplementary Movie S1 ###### Supplementary Movie S2 ###### Supplementary Movie S3 ###### Supplementary Movie S4 ###### Supplementary Movie S5 ###### Supplementary Movie S6 ###### Supplementary Movie S7 ###### Supplementary Movie S8 ###### Supplementary Movie S9 ###### Supplementary Movie S10 We would like to thank Meng-Horng Lee for providing the experimental data shown in [Fig. 3](#f3){ref-type="fig"} and are grateful for discussions with Nir Gov, Samuel Safran, Paul François, Paul Wiseman, Cristiano Diaz, Sabrina Leslie, Bojing Jia, Marcia Simons, Miriam Rafailovich, Nigel Goldenfeld, Nikolas Provatas, Ken Elder, Niloufar Faghihi, Anna Mkrtchyan, Yani Yuval and James Sethna. The Natural Sciences and Engineering Research Council of Canada and the Fonds québécois de la recherche sur la nature et les technologies are gratefully acknowledged for funding this research as well as Compute Canada for providing access to the CLUMEQ Supercomputing facility. B.P. furthers thanks Dr. Juan Gallego for helping with numerical issues. The authors declare no competing financial interests. **Author Contributions** B.P. developed and implemented the model, wrote the main text and the [Supplementary Information](#S1){ref-type="supplementary-material"} and prepared all figures and [Supplementary movies](#S1){ref-type="supplementary-material"}. D.W. and M.G. assisted with the model development. Y.B. and M.G. assisted in the analysis of the simulation data, in writing the main text and the [Supplementary Information](#S1){ref-type="supplementary-material"} and in the preparation of all figures. {#f1} {ref-type="disp-formula"}), after aging. The arrow points toward the tagged cell. The left and right groups of three panels respectively correspond to the packing fractions *ρ* = 0.85 & 0.9 and the top and bottom groups respectively correspond to the soft-in-normal and all-normal cases.](srep11745-f2){#f2} ![Summary of motile cells simulations.\ **(A)** The perimeter and instantaneous velocity, *L* and , of the tagged cell as a function of dimensionless time. Green corresponds to the soft cancer cell (soft-in-normal simulations) and Blue corresponds to its normal counterpart (all-normal simulations). The top (bottom) two panels report the simulation results obtained with *ρ* = 0.85(0.90). The total simulation time corresponds to ≈40 hours when converted back in real time (see the Methods section). The two boxes show the trajectories of the tagged cells at the two packing fractions. **(B)** Experimental results reproduced with permission from Ref. [@b9] for a monolayer that comprises very few cancer cells and mostly normal cells. The value of *L* is calculated from the deformation of the cell nucleus and the total observation time is 16 hours. **(C)** Snapshots of the motile soft-in-normal monolayer simulations at *ρ* = 0.85 (top) and *ρ* = 0.90 (bottom) that correspond to the times indicated by the labels i, ii and iii in part A. **(D)** Three snapshots of the soft-in-normal simulation at *ρ* = 0.90 are shown at times just before, during, and right after the speed burst right after the label iii in part A. Only the soft cancer cell (filled in Green) and its neighbors (Blue boundaries) are shown. The length of the Black arrow on top of each cell is proportional to its instantaneous velocity. When collective effects between cells induce a speed burst to the soft cancer cell (i.e., when its instantaneous velocity is larger than the active part), the arrow is shown in Red. **(E)** Average change in cell perimeter, Δ*L*, between succesive time steps binned according to the cell instantaneous velocity, . Note that part A and D have accompanying [supplementary movies](#S1){ref-type="supplementary-material"}.](srep11745-f3){#f3} {ref-type="disp-formula"}). In the all-normal case, the velocity distribution of the tagged cell is well described by a Gaussian fit (Black line) with standard deviation *σ*~*G*~ = 0.0029. In the soft-in-normal case, the velocity distribution of the soft cancer cell is not well described by a Gaussian fit (Black line) with *σ*~*G*~ = 0.0028. It is better described by the distribution proposed in the main text, [Eq. (5)](#eq10){ref-type="disp-formula"}, which is shown here as the Red curve.](srep11745-f4){#f4} {ref-type="disp-formula"}.](srep11745-f5){#f5} ![Simulation results for two normal cells that move toward each other with opposite velocities.\ The parameter *ξ* that controls the relative strength of the active force, **v**~*A*~ and the interaction force, **v**~*I*~, is chosen to be *ξ* = 1.5 × 10[@b3]. **(A)** The *x*-component of the dimensionless *total* velocity of the left cell is reported as a function of time. Snapshots of the cells configuration are given at four times. **(B)** At long times, the *y*-component of the inactive part of the velocity deviates away from zero due to small numerical errors. This causes the cells to deviate from their original direction and pass by one another, deforming their shapes in the process.](srep11745-f6){#f6} ###### Numerical values used for the dimensionless model parameters. Normal and Cancer cells only differ in the parameter that controls their stiffness, *γ*~*n*~. Dimensionless parameter *γ*~*n*~= *λ*= *R*= *κ*= *μ*= *τ*= *v*~*A=*~ *ξ*= ------------------------- ----------- -------- ------ ------ ------ ------ ----------- -------- ------------- numerical value 0.35 1 7 49 10 1 10^4^ 10^−2^ 1.5 × 10^3^ Cell type cancer normal all

1. Introduction {#sec0005} =============== The prevalence of endometriosis in women during reproductive life is about 10%--15% \[[@bib0005]\]. It can affect not only peritoneum and ovary but also bowel, urinary tract, pericardium and lungs. Gastrointestinal localizations most commonly occur in the rectosigmoid. Colonic endometriosis can lead to a complete bowel obstruction \[[@bib0010], [@bib0015]\]. In emergency settings it is most frequently treated with stoma placement. This approach brings about all the risks related to emergency surgery and might have important psychological and biological side effects. We herein present a case of sigmoid endometriosis with complete bowel obstruction treated with endoscopic stenting and delayed one step laparoscopic procedure. We only found another similar case reported in literature \[[@bib0020]\]. This work has been reported in line with the SCARE criteria \[[@bib0025]\]. 2. Case report {#sec0010} ============== A 38 years old woman presented at emergency care with a history of abdominal pain started two days earlier and constipation started nine days earlier, she reported nausea but no vomit. The patient had personal history of endometriosis and laparoscopic right ovarectomy was carried out a few years before; no similar episodes of abdominal pain were reported. She had no family history of intestinal diseases. The abdomen was meteoric and tender; vital signs were normal with the exception of tachycardia (105bpm). On laboratory exams the WBC was 14090/mm3 and CRP was 3,1 mg/L. A plain abdominal X-ray was performed with evidence of small and large bowel distension and an Abdomen CT detected an irregular mass (diameter 2 cm) at the proximal sigmoid colon determining stenosis. In consideration of the occlusive state, of the radiologic findings and of the likelihood of endometriosis, emergency recto-sigmoidoscopy was performed. The procedure revealed only lumen narrowing without mucosal alterations. A metallic auto-expansible stent was placed to treat bowel obstruction and to delay surgery. Fasting, parenteral rehydration, a double intravenous antibiotic therapy and analgesic drugs were started. Over the next 48 h the bowel obstruction was resolved. The patient underwent a transvaginal utltrasonography (TVUS) with evidence of peritoneal endometriosis in the Douglas pouch and suspected sigmoid deep endometrioid localization. CA-125 levels were increased (114,8 U/L). After 5 days from endoscopy a laparoscopic sigmoidectomy was performed without stoma placement. Histological investigation revealed the presence of endometrioid foci with inflammation and fibrosis affecting the entire sigmoid wall \[[Fig 1](#fig0005){ref-type="fig"}\].Fig. 1Section of the sick sigma with the endoscopic metallic stent inside.Fig. 1 The patient was discharged at fifth postoperative day in good conditions and was referred to Gynecologists. At one month surgical follow-up she had no more abdominal pain and constipation. 3. Discussion {#sec0015} ============= Endometriosis is the growth of ectopic endometrium, most commonly on ovary and pelvic peritoneum \[[@bib0030]\]. It usually leads to pelvic pain, deep dyspareunia, dysmenorrhea and infertility \[[@bib0035], [@bib0040]\]. It can also affects other organs determining different clinical pictures. Even though intestinal localizations occur in about 5--15% of patients, only in about 1% bowel resection is required \[[@bib0010], [@bib0015]\]. Laparoscopy should be considered the diagnostic gold standard for Endometriosis. At present clinical evaluation, imaging and serologic markers can lead to a correct diagnosis leaving surgery to selected patients with a "see and treat" rationale \[[@bib0045]\]. This is also true for deep infiltrating endometriosis; in fact TVUS has a reported sensitivity of 91% and specificity of 98% in detecting bowel localizations \[[@bib0050]\]. Furthermore elevated serum levels of CA-125 can be considered for diagnosis \[[@bib0055]\]. The low incidence of bowel obstruction due to Endometriosis makes the diagnosis unlikely. Contrast abdominal CT has a low specificity and clinical presentation (constipation, nausea, vomit, abdominal pain, rectal bleeding) is unspecific. Other much more common conditions such as Cancer, Inflammatory Bowel Disease and obstruction due to bowel adhesions have a similar onset \[[@bib0010]\]. This is why the diagnosis is usually made by gross histology once the therapeutic decision has already been taken. In the case described patient's age, personal history and the endoscopic findings guided the diagnostic and therapeutic flow-chart. A very important aspect of the disease consists of the psycho-physical implication related to therapies that can drastically alter patient\'s quality of life \[[@bib0060]\]. For this reason the best management of endometriosis is by integrate approach of both medical and surgical treatment \[[@bib0045], [@bib0065], [@bib0070]\]. In the literature some cases of acute colonic obstruction due to endometriosis are described. Hartmann's procedure or direct anastomosis with defunctioning stoma were performed, either open or laparoscopic \[[@bib0075], [@bib0080], [@bib0085], [@bib0090]\]. Our patient was treated with endoscopic stenting as a bridge to elective laparoscopic surgery. We consider that this approach should be taken into account when colonic obstruction due to endometriosis is suspected, especially in young women with positive personal history. Endoscopic stenting is a relatively safe procedure, potentially avoids the costs of two steps surgical intervention and the psychological drawbacks related to stoma placement. Laparoscopic procedure also allows a higher pregnancy rate after surgery \[[@bib0005]\]. In the literature we only found another similar case reported to have good outcomes \[[@bib0020]\]. Conflict of interest {#sec0020} ==================== All the Authors declare that there is no potential personal conflict of interest or financial disclosures or acknowledgements. Funding {#sec0025} ======= This research do not receive any specific grant from funding agencies in the public, commercial or not-for-profit sector. Ethical approval {#sec0030} ================ Ethical approval has been exempted by our Institution, because our paper is not a research but a case report. Consent {#sec0035} ======= Written informed consent has been obtained. A copy of the written consent is available for review by the Editor-in-Chief of this journal on request. Author contribution {#sec0040} =================== Pietro Calcagno: corresponding author who wrote the paper. Matteo Viti: contribute by giving the paper concept. Alessandro Cornelli: the consultant surgeon who managed the patient and run the operation. Davide Galli: the assistance surgeon in patient's operation. Corrado D'Urbano: head physician who receive the article and gave final approval. Guarantor {#sec0045} ========= Corrado D'Urbano.

Introduction {#Sec1} ============ The relationships between different anatomical measurements are a fundamental aspect of human physiology, as has been elegantly depicted by Leonardo da Vinci in his seminal work 'The Vitruvian Man' (da Vinci [@CR12]; Vitruvius [@CR46]). This work shows these relationships are highly conserved, even for men of variable length. This ancient idea of describing *relationships* between different properties has reached many other fields of research, for example quantitative genetics (Steppan et al. [@CR36]) and individual differences psychology (Goldberg [@CR15]). Also the levels of many metabolites in a biological system may be highly interrelated through the biochemical pathways. Perturbations of these biological systems (e.g. diet or disease) may alter enzyme activity and therefore the link between different metabolites. However, the extent of such alterations may also differ between individuals. Thereby also in- or decreases of the inter-individual metabolite level differences may indicate system change. Then not only absolute metabolite level differences between experimental groups, but also the relationships between the metabolites may indicate change. Such Between Metabolite Relationships (BMRs) therefore describe an aspect of metabolism that is complementary to the changes that are common to all individuals (Weckwerth et al. [@CR43]). Recent advances in 'omics'-research brought the study of BMRs closer, because metabolomics emerges more and more as a system-wide approach to observe metabolism (Bino et al. [@CR3]; Fiehn [@CR13]; Hall [@CR16]). The data of a metabolomics study usually consists of a list of numerous metabolites, of which the levels are given for every measured sample (e.g. individual and/or time-point). Of prime interest to metabolomics studies may be to find the in- or decrease of specific metabolite levels between different groups of individuals (e.g. before and after an experimental perturbation) (Fig. [1](#Fig1){ref-type="fig"}a). However, this paradigm holds a major shortcoming for the system-wide view provided by metabolomics analyses, because it may disregard metabolite combinations that show interesting variation where the individual metabolites do not.Fig. 1Three paradigms to observe metabolic differences between two groups: **a** Level difference of an individual metabolite (e.g. ANOVA), **b** Level difference in a combination of, i.e. a component of more metabolites (e.g. PLS), **c** Changes in the combined relationship between metabolites (INDSCAL) Univariate methods that quantify level changes of individual metabolites (e.g., ANalysis Of VAriance, ANOVA (Sokal and Rohlf [@CR35])) disregard the interrelations between levels of different metabolites and thereby the system-wide aspect of metabolism. Therefore in general multivariate methods are used to analyse data generated in metabolomics studies, mostly those from the 'Component Analysis' family such as PCA and PLS-DA (Barker and Rayens [@CR2], Jolliffe [@CR24]). These summarize data into a small number of 'components'---latent variables that gather information about the importance of all measured metabolites. These profiles are constructed based on the levels of all metabolites and express the relative importance of every metabolite in combination with all other metabolites. PCA or PLS-DA models in e.g. case--control studies enable to describe differences in metabolite combinations between groups, even if the levels of single metabolites are not significantly different (Fig. [1](#Fig1){ref-type="fig"}b). However, neither ANOVA nor PLS-DA explicitly reveals the changes in the relationships between metabolites in different experimental groups. Also unsupervised methods like PCA (Jolliffe [@CR24]) may be insufficient to describe BMRs, because these methods cover all metabolic variation simultaneously. The BMRs---schematically depicted in Fig. [1](#Fig1){ref-type="fig"}c---usually remain entangled with other sources of metabolic change and remain beyond reach of any method in these two metabolomic paradigms. Several studies focus on relations between metabolites (Steuer [@CR37]), enzymes and genes (van Erk et al. [@CR41]; Zhai et al. [@CR47]). These studies visualise such relations by Correlation Networks that show the relationships between all metabolite/enzymes/genes pairs (Steuer et al. [@CR38]). However, as already mentioned 'Due to the sheer number of pairwise metabolic correlations, large overview network graphs easily get incomprehensible' (Weckwerth et al. [@CR43]) which is specifically relevant in metabolomics. Therefore a method that both specifically focuses on BMRs and is based on interpretable components that describe the behaviour of the entire system (i.e. all pairwise metabolite relations together) is required. It will provide a novel and complementary view on metabolism. In the field of individual differences psychology, a component method appropriate for the analysis of BMRs called Individual Differences Scaling (INDSCAL) is already available (Carroll [@CR7]). This method translates the changes in covariance or correlations between metabolites upon experimental manipulation into a series of scores and loadings, analogous to those from PCA or PLS-DA. A voluminous yet well-readable publication reveals that INDSCAL is a special version of Parallel Factor Analysis (PARAFAC) (Harshman and Lundy [@CR18]).[1](#Fn1){ref-type="fn"} The PARAFAC model has been used earlier to solve a range of questions in metabolomics studies that focused on changes in metabolite profiles, see e.g. (Montoliu et al. [@CR30]; Jansen et al. [@CR21]; Forshed et al. [@CR14]; Sinha et al. [@CR32]; Verouden et al. [@CR42]) and will therefore provide a view on BMRs intuitive to metabolomics researchers. First BMRs and the INDSCAL model are presented. Then two metabolomics data sets are analysed with INDSCAL, one with a very prominent response of plant chemistry to herbivory and another with a much more subtle response of obese humans to catechin-enriched green tea extract (GTE). The results of standard data analysis methods used in metabolomics, such as ANOVA, PCA, and PLS-DA are compared to that of INDSCAL. Theory {#Sec2} ====== In metabolomics experiments, one or more experimental factors can be manipulated (e.g. doses of a toxicant, different populations) to observe their effect on the metabolites present in an organism, often on different time-points after the manipulation. Metabolomic data consists of comprehensive biochemical descriptions of each sample as a list of metabolites with their corresponding levels. An 'experimental group' of multiple individuals---called biological replicates---undergo a combination of experimental factors. Technical and financial limitations usually lead to considerably more measured metabolites than the number of biological replicates. The 'conceptual model' underlying most metabolomics experiments states that an experimental manipulation may change the levels of several metabolites. When this manipulation is performed on several biological replicates, their response should be similar to the other replicates, up to a certain deviation caused by natural and technical variation. When quantified in a linear model for one factor with groups 1...*k*...*K*, this leads to Eq. [1](#Equ1){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\mathbf{X}}_{k} = 1_{{I_{k} }} {\varvec{\upmu}}^{\text{T}} + 1_{{I_{k} }} {\varvec{\upmu}}_{k}^{\text{T}} + {\mathbf{S}}_{k} \quad {\text{for}}\,k = 1 \ldots K $$\end{document}$$where **X**~*k*~ is the (*I*~*k*~ × *J*) matrix containing the levels of each metabolite, indicated by 1...*j*...*J* in the biological replicates 1~*k*~...*i*~*k*~...*I*~*k*~ of experimental group *k*, **μ** is the length *J* 'centroid' vector of all samples, vector **μ**~*k*~ the centroid vector for group *k* expressed as a deviation from **μ**; matrix **S**~*k*~ contains the deviation of each individual biological replicate from vector **μ**~*k*~; see Supplementary Table 1 for a list of symbols used throughout the paper. Equation [1](#Equ1){ref-type=""} is generally used to quantify the significance of this experimental manipulation on levels of a small subset of single metabolites. This can be done by ANOVA (Sokal and Rohlf [@CR35]) that estimates the treatment effects expressed in a series of vectors $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\varvec{\upmu}}_{k} \left( {k = 1 \ldots K} \right) $$\end{document}$: interesting putative biomarkers are then identified as variables *j* for which variation across the *j*-th elements of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\varvec{\upmu}}_{k} \left( {k = 1 \ldots K} \right) $$\end{document}$ is high relative to the natural and technical variation of the biological replicates derived from **S**~*k*~. The model in Eq. [1](#Equ1){ref-type=""} does not make any assumptions about the relationships between metabolites, which falls in the realm of the component analysis paradigm. Multivariate components {#Sec3} ----------------------- A major objective in metabolomics is to understand the underlying biochemical system, which makes observation of the variations in each individual metabolite insufficient. The relations between different metabolites may both lead to a more parsimonious model---the biochemical system will constrain the complexity of the metabolic changes resulting from the experiment---and may lead to hitherto unknown relations between the metabolites that will provide a better insight into the observed system (Jansen et al. [@CR22]). To model these system-wide relationships metabolomics embraced the multivariate 'component' paradigm that models the relationships between all *J* metabolite descriptors (Fig. [1](#Fig1){ref-type="fig"}b). The 'standard' methods in this field may also be expressed using the partitioning of the variation in Eq. [1](#Equ1){ref-type=""}. Principal Component Analysis simultaneously describes **μ**~*k*~ and **S**~*k*~, so that this model will give a convoluted description of the paradigms in Fig. [1](#Fig1){ref-type="fig"}b, c. The often-used method Partial Least Squares-Discriminant Analysis (PLS-DA) aims---like ANOVA---to describe **μ**~*k*~ at the expense of the 'biological variation' (inter-individual variation, natural variation) in matrix **S**~*k*~. Clearly, thereby PLS-DA does exactly the opposite of what is of interest to BMRs. The analysis of BMRs requires separation of the variation in **μ**~*k*~ from that in **S**~*k*~, because the BMR-related information (between the individual biological replicates) is contained only in the latter matrix. Therefore a component analysis method needs to be developed that focuses on the relations between the metabolites within this contribution. Between Metabolite Relationships {#Sec4} -------------------------------- In characterising BMRs, the strength of relationship between metabolites is of high interest. However, also how much variation in each experimental group is associated with this relationship is important. Although Pearson correlations are widely used in metabolomics, they overlook this aspect, because in Pearson correlations the variation in the levels of both metabolites is scaled by their standard deviations. Therefore covariances are the preferred measure for BMRs. A BMR-describing component model should focus upon the differences between groups in the systematic part of the biological variation. This information is hidden in **S**~*k*~, specifically in the relationships between the metabolites. A view on BMRs therefore necessarily revolves around quantifying relations between the columns of **S**~*k*~. This can be done by covariances, like in Eq. [2](#Equ2){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\mathbf{R}}_{k} = I_{k}^{ - 1} {\mathbf{S}}_{k}^{\text{\,T}} {\mathbf{S}}_{k} $$\end{document}$$where **R**~*k*~ is the covariance matrix of experimental group *k* with dimensions (*J* × *J*). Because interpreting **R**~*k*~ may be tedious for many metabolite covariances, the holistic and simple view of a component model of BMRs may be highly desirable. Individual differences scaling {#Sec5} ------------------------------ A component model for BMRs needs to describe the relations between metabolites, rather than the levels themselves as well as possible. An existing component model that does just this is INdividual Differences SCALing (INDSCAL) model (Kruskal and Wish [@CR27]; Harshman and Lundy [@CR18]; Carroll [@CR7]; Carroll and Chang, [@CR8]), which is given in Eq. [3](#Equ3){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \begin{array}{lll} {\text{Model}}&{\mathbf{R}}_{k} = {\mathbf{AG}}_{k} {\mathbf{A}}^{\text{T}} + {\mathbf{E}}_{k} \hfill \\ {\text{Minimization}}&f\left( {{\mathbf{A}},{\mathbf{G}}_{k} } \right) = \sum\limits_{k = 1}^{K} {\left\|{{\mathbf{R}}_{k} - {\mathbf{AG}}_{k} {\mathbf{A}}^{\text{T}} }\right\|^{2} } \hfill \\ {\text{Constraints}}&diag\left({{\mathbf{A}}^{\text{T}} {\mathbf{A}}} \right) = 1_{R},\,\,\,{\mathbf{G}}_{k} \,\,\,{\text{is diagonal with nonnegative elements}} \hfill \\ \end{array} $$\end{document}$$where **G**~*k*~ is an (*R* × *R*) score matrix of group *k*; matrix **A** of size (*J* × *R*) contains the chemical loadings; **E**~*k*~ contains the residuals of which the sum-of-squares is minimized. The constraints are imposed to arrive at identified and meaningful solutions. The INDSCAL model loadings **A** describe the important relations between metabolites and the scores **G**~*k*~ describe the magnitude of the variation of these relations within each experimental group, such that both important aspects of the BMRs are described. The INDSCAL model is strongly related to Parallel Factor Analysis (PARAFAC) (Bro [@CR6]; Harshman [@CR17]; Smilde et al. [@CR33]), an often-used component model in metabolomics. The INDSCAL model can be fitted by modelling the covariance matrices **R**~*k*~ (arranged in a (*R* × *J* × *J*) three-way array) by PARAFAC (ten Berge and Kiers [@CR39]). The additional nonnegativity constraint on **G**~*k*~ can be straightforwardly imposed by publicly available software (Andersson and Bro [@CR1]). Like PARAFAC, the components of an INDSCAL model are unique. Model visualization and interpretation {#Sec6} -------------------------------------- Conventionally, the INDSCAL loadings **A** are shown in such a way that high loading values relate to the relevance of those metabolites in the BMRs important on each component. However, it may be better interpretable to rearrange the loadings following the structure of the covariance matrices, i.e. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\mathbf{R}}_{k} = \sum\nolimits_{r = 1}^{R} {g_{kr} } {\mathbf{a}}_{r} {\mathbf{a}}_{r}^{\text{T}} + {\mathbf{E}}_{k} = \sum\nolimits_{r = 1}^{R} {g_{kr} } {\mathbf{A}}_{r} + {\mathbf{E}}_{k} $$\end{document}$, where the matrices **A**~*r*~ of dimensions *J* × *J* are symmetric. Then high values in **A**~*r*~ directly indicate important relations between metabolites. It may therefore be easier to interpret a heat map of **A**~*r*~ than a conventional loading plot of **A** to identify relevant metabolites. However, since such heat maps do not allow comparison between components in one figure, both may be of value to gain insight in the BMRs. The scores **G**~*k*~ (or rather the diagonal elements g~*kr*~) show for which group *k* the relations in **A**~*r*~ are important. A score of zero implies that the corresponding relations are absent in group *k*. Just like in PARAFAC, the components fitted for INDSCAL are not orthogonal. The amount of information explained by the model can therefore only be calculated for the entire model. Furthermore, adding INDSCAL components modifies all other components (Smilde et al. [@CR33]), which means a proper number of components has to be chosen before interpreting the model. Number of components, stability and validation {#Sec7} ---------------------------------------------- The amount of information each component adds to the model may be used to determine the appropriate number of INDSCAL components, by comparing the information explained in a model to those with fewer components. Whether the fitted model is prone to local optima can be tested by using multiple random starting values: the models need to be comparable, otherwise the model may contain too many components, thereby covering technical or other non-systematic variation. Because the INDSCAL model describes entire experimental groups rather than individual biological replicates, the significance of observed effects is not expressed in the scores **G**~*k*~. An earlier-proposed jack-knife approach relies heavily on distributional assumptions (Weinberg et al. [@CR44]), not likely fulfilled by metabolomics data. Therefore we quantify this significance by resampling: the results (i.e. scores and loadings) of models where individual biological replicates are left out are compared to the original model, which shows how individual replicates influence the model. This resampling strategy is fully explained in the supplementary material. Also a schematic pipeline to describe BMRs by INDSCAL is given there (Supplementary Fig. 1). Materials and methods {#Sec8} ===================== Induced plant response study data set {#Sec9} ------------------------------------- This experiment studied the 'induced plant response' of cabbage plants to simulated herbivory to the plant shoot (*SJA*) or root (*RJA*), by the plant hormone jasmonic acid. These plants were compared to control (*CON*) plants, not treated with the hormone. The defense was characterized by the glucosinolate compound class: 11 compounds were profiled in plants harvested at 1, 7 and 14 days after the simulated attacks. The dataset contains 6--10 replicate plants *per* herbivory type/harvest time. This study was described in much more detail in an earlier paper (Jansen et al. [@CR23]). Human nutritional intervention study data set {#Sec10} --------------------------------------------- In a double-blinded, placebo-controlled nutritional intervention study with a parallel design, 186 human subjects with abdominal obesity (BMI 25--35 kg/m^2^ and a waist circumference of over 80 cm for women or 95 cm for men) consumed either catechin-enriched green tea extract drink (GTE; 600 mg catechins/day, 87 subjects) or a placebo drink (placebo, green tea-flavoured drink without any active ingredients, 99 subjects) over a period of 12 weeks. The experiment was conducted at University of Nottingham and approved by the University of Nottingham Medical School Ethics Committee. Fasted serum samples were collected at the baseline (start of the experiment; T0) and after 4, 8 and 12 weeks of intervention (T4, T8 and T12). For each serum sample a metabolic profile was obtained, composed of 136 lipid metabolites expressed as ratios between the peak areas of the metabolite and internal standard. The supplementary material contains a description of the analytical method and Supplementary Table 2 lists the measured metabolites. Software {#Sec11} -------- All statistical analyses were carried out in MATLAB 2009a (The Mathworks Inc., Natick, Massachusetts, USA), using in-house routines, partly based on the N-way Toolbox (Andersson and Bro [@CR1]). They have been made available on [www.bdagroup.nl](http://www.bdagroup.nl). Results and discussion {#Sec12} ====================== Induced plant response study---comparison of PCA and INDSCAL results {#Sec13} -------------------------------------------------------------------- In the "induced plant response" study, the metabolic effect of shoot herbivory (SJA) or root herbivory (RJA) are of interest. Initially a PCA model was fitted, slightly modified to exclude the average time-profile of all plants related to uninteresting chemical variation (see earlier paper: Jansen et al. [@CR23]). The results of this analysis are given (again) in Supplementary Fig. 2. The PCA results have been described in great detail in the earlier paper and are only briefly repeated here. The induced response to jasmonic acid consists of an increase in Glucobrassicin (GBC) and Neoglucobrassicin (NEO) for both treatments, but considerably larger for SJA. The RJA plants have higher levels of Progoitrin (PRO) and Glucobrassicanapin (GBN) after 7 and 14 days (PC 3). These changes are consistent for all the plants in the relevant treatment-time combinations and will therefore end up in **μ**~*k*~ in Eq. [1](#Equ1){ref-type=""}. The increase in NEO and GBC will differ between plants, but the model also revealed that SJA plants harvested after 7 and 14 days with more NEO, contain less GBC, this will typically end up in matrix **S**~*k*~ of Eq. [1](#Equ1){ref-type=""} and therefore be the target of INDSCAL analysis. The earlier study also showed the increase in PRO and GBN levels in RJA plants after 7 and 14 days is preceded by an increase in the natural variation of these levels, which should also be revealed by INDSCAL. A 4-component INDSCAL model explains a highly unstable amount of information (see Table [1](#Tab1){ref-type="table"}), leading to three-components. This INDSCAL model (Fig. [2](#Fig2){ref-type="fig"}) corresponds very well to the PCA results. The first component explains a BMR in the *SJA* plants that increases from absence to an enormous contribution 14 days after harvest and explains the high NEO with low GBC levels (see Fig. [2](#Fig2){ref-type="fig"}c). The positive relation between PRO and GBN, expected to be high specifically high 1 day after *RJA* is indeed present in the second INDSCAL component (Fig. [2](#Fig2){ref-type="fig"}d). The component is also important 14 days after *SJA*, which after further inspection of the large confidence interval on the PCA scores. The third INDSCAL component describes the consistently larger variation in NEO and in GBC related to the natural variation between the different *SJA* (1--14 days) and *RJA* plants (1--7 days) described by the first PCA component (Fig. [2](#Fig2){ref-type="fig"}e). In this data set, the qualitatively observed BMRs in the earlier PCA model could be quantified in the INDSCAL model.Table 1Number of components for INDSCAL model of plant data set\# components% convergence% variance explained11009221009731009948099.555599.7The second column shows how many models converged to a stable solution and the third how much information is described by the modelFig. 2INDSCAL model of plant data set. **a** Group scores for component 1 vs. component 2, **b** Group scores for component 2 vs. component 3, *circles* refer to control group (CON), *squares* to root herbivory (RJA) and crosses to shoot herbivory (SJA). Loading are presented as heatplots separately for each component: **c** loadings for component 1, **d** loadings for component 2 and **e** loadings for component 3 Example of human nutritional metabolomics study and BMRs {#Sec14} -------------------------------------------------------- In this section, BMR analysis is applied on lipid profiles from obese human subjects consuming either green tea extract (GTE) or placebo during 12 weeks. Changes in the lipidome were expected, because dietary supplementation of GTE has been proposed as a strategy for weight loss (Maki et al. [@CR29]; Kovacs and Mela [@CR26]). It has been hypothesized to promote lipolysis as a mechanism by which GTE stimulates fat oxidation (Westerterp-Plantenga [@CR45]) and to affect lipid metabolism by inhibiting lipid absorption and digestion (Koo and Noh [@CR25]). The effect of GTE on the lipid profiles was analysed by investigating changes in individual metabolite levels, in multi-metabolite profiles and in BMR components fitted by INDSCAL. Variables were not scaled and INDSCAL was performed with covariances to compare all results to each other. ### Changes in individual metabolite levels {#Sec15} Univariate, nonparametric U-Mann--Whitney statistic tests (Sokal and Rohlf [@CR35]) did not reveal any statistically significant changes in individual metabolite levels between T12 (end of intervention period) and T0 (baseline) (see Supplementary Table 3), for either GTE or placebo. This indicates GTE did not induce an effect stronger than the inter-individual variation. This is frequently observed in dietary intervention studies within healthy human subjects, where effects are typically subtle and obscured by large inter-individual variations. ### Profiles of multiple metabolites {#Sec16} An unsupervised PCA model did not reveal any relevant difference between GTE and placebo groups (see Supplementary Fig. 3). A supervised PLS-DA model also did not provide statistically significant differences between the GTE and placebo group metabolite profiles between T0 and T12 (see Supplementary Table 4). A multiway-PLS-DA (N-PLS-DA) was employed to simultaneously evaluate time-related metabolic changes induced by GTE at all four time points (Bro [@CR5]; Castro and Manetti [@CR9]), while retaining the structure of the repeated measurements on the same individuals (Smilde et al. [@CR34]). The diagnostic outcome of this model was very weak: e.g. 47.1% of the samples were misclassified (see Supplementary Table 4). This shows GTE intervention did not change the serum lipid profiles significantly as observed by the 'standard' multivariate data analysis methods most widely used in metabolomics (Jansen et al. [@CR22]; Trygg et al. [@CR40]; Lindon et al. [@CR28]; Holmes et al. [@CR19]). ### Between Metabolite Relationships {#Sec17} To include all available a priori knowledge about the experimental design into the INDSCAL model, a 'baseline' group (BL) was constructed of all individuals measured at the start of the experiment (T0), assuming all subjects to belong to a homogeneous population before the nutritional intervention. For the remaining samples, covariance matrices were calculated for each experimental groups, i.e. treatment and measurement time-point combination: GTE-T4, GTE-T8, GTE-T12, placebo-T4, placebo-T8 and placebo-T12. To determine the number of INDSCAL components appropriate to model these covariance matrices, one to five components were fitted 20 times, starting from random values. Table [2](#Tab2){ref-type="table"} shows the percentages of explained information and of converged models. This table shows that the model requires two components, because 20% of the models with three components did not converge to a stable solution.Table 2Number of components for INDSCAL model for human nutritional data set\# components% convergence% variance explained110097.1210099.038099.446099.657099.8The second column shows how many models converged to a stable solution and the third how much information is described by the model The INDSCAL scores of the GTE group (black circles in Fig. [3](#Fig3){ref-type="fig"}a) differ from those of the placebo group (white circles) after 4, 8 and 12 weeks of intervention and from the BL group (grey circle). The first INDSCAL component mainly describes a systematic drift of the GTE group from the region of the plot covered by the placebo group and the second component shows an additional variation in the BMRs of the GTE group, prominent at T4 (Fig. [3](#Fig3){ref-type="fig"}a).Fig. 3INDSCAL model of human nutrition data set. **a** Group scores: the *white circles* indicated the placebo group, the *black* the GTE group and the *grey* the common baseline group (sampled before start of intervention). The region around each score is obtained during model validation and refers to region of plot where 95% scores obtained from resampled models occurred. **b** Loadings with regions of confidence obtained during model validation analogously as for scores. **c** Heat plot of BMRs in *greyscale*; both inserts focus on the relations of TG28 with 29 and of TG54 with 42: these are indicated by the *white* frames The resampling results (confidence intervals around the circles in Fig. [3](#Fig3){ref-type="fig"}a) show that the differences between GTE and placebo are highly significant after 4 and 12 weeks and that after 8 weeks the resampling interval of GTE has only very slight overlap with that of BL. The resampling results of the chemical loadings (intervals around the circles in Fig. [3](#Fig3){ref-type="fig"}b) show that the BMR response to GTE consists of the covariance between metabolites TG28, TG29, TG41 and TG42. The first INDSCAL component is of most interest in this study, because it shows a consistent GTE-associated drift. The heat plot in Fig. [3](#Fig3){ref-type="fig"}c focuses upon these loadings. This heat plot shows BMRs rather than the contributions of the individual lipids to the loadings in Fig. [3](#Fig3){ref-type="fig"}b. The heat plot quantifies the BMRs in greyscale, showing for example that the covariance between TGs 28 and 29 is larger than between TGs 54 and 42, considerably less interpretable from Fig. [3](#Fig3){ref-type="fig"}b alone. Selected lipids: level changes and BMRs {#Sec18} --------------------------------------- The effect of GTE on the plasma lipids is clearly visible in both INDSCAL model representations in Fig. [3](#Fig3){ref-type="fig"}. It is mainly associated with relations between a very small subset of lipids. Most important are the triacylglycerols TG28-29 and TG41-42. Figure [4](#Fig4){ref-type="fig"}a shows that the variance of TG29 (those for TGs 28, 41 and 42 are comparable and not shown) is significantly affected by GTE compared to the control and the BL groups, although the mean group levels of these metabolites did not change (Supplementary Fig. 4). The Pearson correlation coefficients for these lipids did not change between GTE and placebo (see Supplementary Table 5), but the covariances did (Fig. [4](#Fig4){ref-type="fig"}b; Supplementary Fig. 4). The barplot in Fig. [4](#Fig4){ref-type="fig"}b corresponds closely to the INDSCAL scores (Fig. [3](#Fig3){ref-type="fig"}a). The INDSCAL results and the (co)variance plots show that the effect of GTE manifested itself by a systematic increase of the covariance between TG28 and TG29 during the entire study period and an additional increase of the covariance between TG41 and TG42 during the first 4 weeks of intervention, described by the second INDSCAL component. The last is related to a large inter-individual difference in time and magnitude of response at the beginning of the intervention.Fig. 4Variance and covariance of selected metabolites. **a** Variance of TG29 and **b** covariance between TG28 and TG29; *BL* baseline group, *GTE* catechin-enriched green tea extract group, *placebo* placebo group, significantly different: \*\**P* \< 0.05 and \*\*\**P* \< 0.01 Interpretation of the observed GTE effect {#Sec19} ----------------------------------------- The INDSCAL model shows that supplementation of GTE significantly affects relationships between a small subset of triacyloglycerols (see Supplementary Table 6 for the isomer composition). Similar TGs have been reported to play an important role in diet-induced weight loss for metabolic syndrome in a 33-week intervention (Schwab et al. [@CR31]). These changes were not shown by the standard uni- and multivariate statistical analyses, because these focus upon responses in metabolite levels similar for all treated individuals (i.e. PLS-DA or Mann--Whitney tests). Figures [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"} show the observed BMRs relate to an increase in the variation in the levels of selected triacyloglycerols between subjects that received the same intervention. Observed changes in metabolite covariances show that their changes are dependent between metabolites and therefore the observed effect of GTE can be explained on a system biology level. Inter-individual variation in the levels of selected triacylglycerols could be related to individual differences in the activities of transcription factors or enzymes regulating these metabolites (individual phenotype). Depending on the characteristics of the individual phenotype, GTE could induce the increase or decrease of these specific metabolite levels (see Supplementary material for a simulation example). For example, it has been stated that there is a wide variability in the flavonoid O-methylation by catechol-*O*-methyltransferase (COMT), a key enzyme that is hypothesized to be involved in fat oxidation and whose activity may differ between ethnic groups (Westerterp-Plantenga [@CR45]). Alternatively, the increase of inter-individual variation in levels of selected triacyloglycerols can be explained by multiple mechanisms of actions and/or active compounds present in GTE. That may lead to opposing effects of GTE on a network of transcription factors and enzymes and thereby to up- and downregulation of the production of specific metabolites and less controlled ranges of metabolite levels. In this case, metabolic change might be the consequence of a superposition of e.g. changes in dietary fatty acid composition, different mechanisms of TG activation or different effects on the lipid species present in the TGs (Kovacs and Mela [@CR26]; Westerterp-Plantenga [@CR45]). INDSCAL and BMRs in practice {#Sec20} ---------------------------- To extract BMR-related information by standard data analysis methods may be difficult (i.e. PCA) and often even impossible (PLS-DA): these methods have a different focus. This paper shows, by two examples of metabolomic data sets from plant and human nutrition studies, that the BMR-related components of INDSCAL showed an essential aspect of metabolic change that was complementary to that obtained by standard methods. A PCA model of the plant "induced response study" only showed BMR-related change intermingled with level changes like those in Fig. [1](#Fig1){ref-type="fig"}b. However, with INDSCAL these were directly focused upon. In the human nutritional study, INDSCAL revealed increases in the inter-individual variation of four triacylglycerols upon GTE supplementation, while this (or any other) effect of GTE could not be observed by standard data analysis methods. In this study, the BMRs were expressed and included in INDSCAL as covariances, but also other dissimilarity measures such as Pearson or Spearman correlation coefficients can be used for a different focus (Jansen et al. [@CR20]). In fact, multidimensional scaling methods like INDSCAL allow matrices **R**~*k*~ to be filled with many dissimilarity measures, underlined by a valid distance metric (Borg and Groenen [@CR4]). The choice of dissimilarity measure depends on the expected nature of the relationships, such that INDSCAL is a highly flexible tool to find BMRs. Covariance analysis between metabolites, as opposed to correlations, is highly appropriate for studies where responses are expected to be inconsistent between individuals. For example, INDSCAL directly targets the expected variation in the response of different humans to dietary intervention, such as that of GTE. Because not all individuals respond to a dietary supplementation of GTE in the same fashion or degree, covariances rather than levels of these metabolites change when the entire experimental group is observed. The metabolites involved in this effect then also have a different role than in the conventional paradigms in Fig. [1](#Fig1){ref-type="fig"}a, b: triacylglycerols of which the covariances with other metabolites change during dietary intervention could be used a posteriori to select the individuals from the experimental group that have a similar metabolic response. This relates directly to the evolutionary constraints that were already discussed in the introduction: these also rely on responses only present in a subset of the population. The introduction of INDSCAL also makes such patterns available to metabolomics. The literature concerning visualisation of INDSCAL models is sparse (with exceptions like (Chang and Carroll [@CR11])). The representation of **A**~*r*~ in heat maps is---to our knowledge---new and considerably increases the insight into the metabolic relations described by the BMRs, compared to the conventional representation in Fig. [4](#Fig4){ref-type="fig"}. However, the relations between metabolites represent the biochemical reactions within the studied organisms, therefore the INDSCAL loadings would immensely benefit from an interpretation through biochemical pathways. This would connect the Correlation Networks that until now have observed metabolism as a series of pairwise correlations between metabolites (Steuer et al. [@CR38]) with component analyses that simultaneously connect all metabolites to each other. The synergy between Correlation Networks and INDSCAL will be the topic of a follow-up paper. The Between Metabolite Relationships, together with INDSCAL, will therefore greatly enhance the amount of biochemical information that can be obtained from 'omics' experiments. Concluding remarks {#Sec21} ================== Between Metabolite Relationships (BMRs) may reveal systematic changes in biological systems that remain elusive when only metabolite level changes are taken into account. The Individual Differences Scaling (INDSCAL) method is introduced here as a method to analyse these BMRs with component models, which give a system-wide view on the changes in relationships between all compounds measured in a metabolomics study. The results of INDSCAL can support and explain already known metabolic changes, such as those in the "induced plant response" study. They can also provide information that lays beyond the reach of standard data analysis methods in use in metabolomics as in the human nutritional intervention study. The BMRs indicated which relations between metabolites are most prone to a variable response by the biological replicates e.g. by jasmonic acid application (subset of glucosinolates in "induced plant response" study) or by the GTE intervention (subset of triacyloglycerols in human nutritional intervention study). Identification of such changes in metabolite relationships will improve the understanding of possible mechanisms of action of tested interventions. The BMRs, together with INDSCAL, thereby open the door to dedicated analysis of the next generation of questions in systems biology: those that deal with personalized medicine and individual or cohort-specific responses to dietary change. Electronic supplementary material {#AppESM1} ================================= {#SecESM1} Supplementary material 1 (DOC 771 kb) This project was financed by the Netherlands Metabolomics Centre (NMC) which is a part of the Netherlands Genomics Initiative/Netherlands Organisation for Scientific Research. Authors gratefully acknowledge Jorne Troost (NMC, LACDR/Leiden University) for further experimental work on lipid identification and Adrie Dane (NMC, LACDR/Leiden University) for quality control of human nutritional intervention study data set. The 'induced plant response' study was funded by NWO, the Netherlands Organization for Scientific Research through VIDI grant, no. 864-02-001. Open Access {#d29e1415} =========== This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited. The book that this chapter appeared in is out-of-print and difficult to obtain. However, it can be found online in PDF format: <http://publish.uwo.ca/~harshman/abstract.html>. Jeroen J. Jansen and Ewa Szymańska contributed equally to the manuscript.

Thirty-five percent of adults in the United States and United Kingdom have chronic lower limb superficial venous disease. [@JR1900086oa-1] Varicose veins are more common in females, with a predilection toward the older age group and may run in families. A body mass index \>30 kg/m ^2^ is a risk factor for chronic venous insufficiency. [@JR1900086oa-2] Symptoms include limb heaviness, ache, and edema. Skin changes such as spider veins, varicose veins, hemosiderin deposition, inflammation, lipodermatosclerosis, and ulceration often follow in untreated cases. [@JR1900086oa-2] [@JR1900086oa-3] [@JR1900086oa-4] The 2013 National Institute for Health and Care Excellence (NICE) guideline on diagnosis and management of varicose veins (updated March 2018) recommends radiofrequency ablation (RFA) or endovenous laser ablation (EVLA) as first line treatment for truncal reflux. Second-line is ultrasound-guided foam sclerotherapy. Open surgery is indicated only if the other methods are unsuitable. Any incompetent tributaries are preferentially treated in the same session. Compression hosiery should not be used longer than 7 days after intervention, and is first choice only in pregnancy or if the previously mentioned interventions are unsuitable. [@BR1900086oa-5] NICE also issued a specific guideline in 2015 on the use of *n* -butyl-2-cyanoacrylate (NBCA) for varicose veins but did not promote its routine use. [@BR1900086oa-6] Almeida et al reported the first human application of NBCA for incompetent great saphenous veins (GSVs) in 2013. All 38 veins under study were obliterated at 48 hours and 92% at 1 year with minor short-lasting adverse effects. [@JR1900086oa-7] The aim of this systematic review is to assess the efficacy of NBCA in ablating primary truncal varicose veins and eliminating reflux compared with existing endovascular techniques in the immediate, medium, and long-term settings. Secondary outcomes include complications, patient acceptability, and quality of life. Methods ======= Protocol and Search Strategy ---------------------------- This review is registered in PROSPERO database (registration code: CRD42018106323) and followed the PRISMA checklist. [@BR1900086oa-8] [@JR1900086oa-9] One author performed a literature search and data extraction up to October 2018 with no set date range and using established MeSH vocabulary in PubMed, EMBASE, Scopus, Cochrane Library, and ScienceDirect. Search terms were: "varicose vein," "saphenous vein," "glue," " *n* -butyl cyanoacrylate," and " *n* -butyl 2 cyanoacrylate." References and article suggestions by search engines were assessed to identify more relevant studies. Duplicates were removed and further exclusions performed after reviewing abstracts. The chosen manuscripts were then scrutinized while applying inclusion and exclusion criteria. Inclusion and Exclusion Criteria -------------------------------- Human randomized controlled trials (RCTs), cohort studies, and case reports in English language involving the use of NBCA to treat primary truncal varicose veins (i.e., GSV, small saphenous vein \[SSV\], and anterior accessory saphenous vein \[AASV\]) were included. If more than one modality was used, the said manuscript was only included if the data for NBCA could be fully extracted. Studies excluding NBCA glue or comparing NBCA with treatments other than RFA, EVLA, or foam sclerotherapy were excluded. [@JR1900086oa-1] [@JR1900086oa-2] [@JR1900086oa-10] [@JR1900086oa-11] [Supplementary Table 1](#SM1900086oa-1){ref-type="supplementary-material"} (online only) summarizes patient characteristics for inclusion/exclusion. Primary and Secondary Outcomes ------------------------------ Primary outcome was successful obliteration of lumen of target vein, defined as occlusion of the entire treated vein segment with no discrete segments of patency exceeding 5 cm, confirmed on color Duplex ultrasound (DUS) after the procedure. [@JR1900086oa-1] Follow-up DUS assessments at 3 days, 7 days, 1 month, 3 months, 6 months, 1 year, and 2 years were examined. Influence of vein length, diameter, NBCA device, and postoperative compression stockings on early (3 months) and intermediate term (6 months, 1 year) occlusion rate was taken as secondary outcomes. Vein length was taken as a mean value incorporating GSVs, SSVs, and AASVs with no distinction between the three. Where a particular vein diameter was taken at different levels, the mean of these values was calculated. Clinical, Etiological, Anatomical, and Pathophysiological classification and Varicose Clinical Severity Score (VCSS) were used to measure severity of varicose veins at baseline and postintervention. Quality of life was primarily investigated using the Aberdeen Varicose Vein Questionnaire (AVVQ). [@JR1900086oa-2] "Thrombophlebitis" and "abnormal skin reactions" in treatment zones were included with the general term "phlebitis." [@JR1900086oa-12] [@JR1900086oa-13] All thrombus extensions into the deep venous systems were classified as deep vein thromboses (DVTs). Complications common to the three ablation modalities were evaluated. Data Extraction --------------- Any uncertainties in the literature were discussed with the second author and the authors of the original manuscripts where applicable. Risk of methodological bias was explored using the Cochrane Risk of Bias tool for RCTs. [@BR1900086oa-14] [@JR1900086oa-15] Quality assessment was performed using the Downs and Black quality assessment tool (for RCTs) and the National Heart, Lung and Blood Institute: Quality Assessment Tool for Before-After (Pre--Post) Studies With No Control Group (NHLBI-QAT). [@JR1900086oa-16] [@BR1900086oa-17] Statistical Analysis -------------------- Continuous variables were represented by means, standard deviations, and ranges. Categorical variables were shown in actual numbers and percentages. Scatter plots were created using Python version 3.7 (Python Software Foundation, Beaverton, DE). Statistical analysis was done using IBM SPSS Statistics software (IBM Corp. Released 2013. IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY). Spearman\'s correlation and Mann-Whitney U-test were performed on groups of subjects at 3, 6, and 12-month intervals following NBCA treatment. These tests were chosen because continuous variables were not normally distributed. Level of statistical significance was taken as *p*  \< 0.05. Results ======= Description of Studies ---------------------- The PRISMA flowchart ( [Fig. 1](#FI1900086oa-1){ref-type="fig"} ) depicts the choice of manuscripts at different phases. One case report was identified but not reviewed as it contained heterogeneous data. [@JR1900086oa-18] All were published in peer-reviewed indexed scientific journals. There were 3038 participants (3,220 veins). A subgroup of 128 patients were excluded because of the missing data. [@JR1900086oa-19] [@JR1900086oa-20] [@JR1900086oa-21] Of the 2910 patients who were included, 1981 received NBCA, 445 RFA, and 484 EVLA. Comparison of NBCA with RFA and/or EVLA was performed in three RCTs and two retrospective studies. [@JR1900086oa-10] [@JR1900086oa-12] [@JR1900086oa-19] [@JR1900086oa-21] [@JR1900086oa-22] No studies compared NBCA with sclerotherapy, but this was frequently used an adjunctive treatment. Levels of evidence for therapeutic studies were judged using criteria from the Centre for Evidence-Based Medicine. [@JR1900086oa-23] {#FI1900086oa-1} Quality and Risk of Bias Assessment ----------------------------------- ### Randomized Controlled Trials Risk of bias for RCTs is illustrated in [Table 1](#TB1900086oa-1){ref-type="table"} . Bozkurt and Yilmaz pseudorandomized their patients to alternate EVLA and NBCA. This led to a high risk of selection bias. [@JR1900086oa-10] Randomization was better in the VeClose trial and the study by Eroglu and Yasim. [@JR1900086oa-1] [@JR1900086oa-12] [@JR1900086oa-19] The former also included "roll-in cases" so that investigators could achieve familiarity with the NBCA procedure. DUS assessments were not always performed by blinded personnel. Attrition bias was unclear in two RCTs as drop-outs were not formally analyzed. [@JR1900086oa-13] [@JR1900086oa-20] Effect of adjunctive therapies and postoperative compression stockings was not evaluated. Only one performed power analysis. [@JR1900086oa-19] Primary and secondary end points were clearly reported in all RCTs. ###### Traffic light plot illustrating risk of bias of the included RCTs (using the Cochrane risk of bias tool) and Downs and Black quality assessment scores -- -- Abbreviation: RCT, randomized controlled trials. Note: The score for item 27 in the Downs and Black checklist was modified to determine whether power analysis was conducted (yes = 1 point) or not (no = 0 points). So, the maximum score for the checklist was 28 instead of 32. [@JR1900086oa-24] ### Prospective and Retrospective Studies Prospective studies were of a higher methodological quality ( [Supplementary Fig. 1](#SM1900086oa-1){ref-type="supplementary-material"} and [Supplementary Table 2](#SM1900086oa-1){ref-type="supplementary-material"} \[online only\]). Selection bias toward bilateral varicose veins was observed in one prospective and one retrospective study. [@JR1900086oa-25] [@JR1900086oa-26] Another reported a modification of intervention after commencement of data collection which improved the complication rate in the remaining patients. [@JR1900086oa-27] Blinding of assessors was not possible. The loss to follow-up for NBCA was 23.7% in one manuscript. [@JR1900086oa-28] Another started with 34 patients and had 26% loss at 1 month. [@JR1900086oa-13] One prospective and one retrospective study reported percentage occlusion rate only once at 1 month and 1 year respectively despite mentioning several follow-up intervals in the methodology. [@JR1900086oa-13] [@JR1900086oa-22] Coincidentally, the former did not have sufficient patients at the target 3-month interval to formulate strong conclusions. [@JR1900086oa-13] Another study did not differentiate between the short- (1 week) and mid-term (2 months) outcome results, which instead were displayed as combined absolute values. [@JR1900086oa-21] Population and Operative Details -------------------------------- Study characteristics are summarized in [Tables 2](#TB1900086oa-2){ref-type="table"} and [3](#TB1900086oa-3){ref-type="table"} . ###### Characteristics of identified studies ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ First author Year Country Study design, Evidence level Single/Multicenter Comparator Recruitment period NBCA patients Comparator patients Patients excluded Proposed follow-up (mo) Actual follow-up (mo) Definition of varicose vein and/or vein incompetence -------------------------------------- ------ ------------------------ ------------------------------ -------------------- ------------ ---------------------------------------- --------------- --------------------- ------------------- ------------------------- ----------------------- --------------------------------------------------------------------------------------------------------------------------------------------- Bozkurt and Yilmaz [@JR1900086oa-10] 2016 Turkey RCT, 1B Multicenter EVLA December 2013--March 2014 (3 mo) 154 156 EVLA -- 12 12 CEAP C2-C4b with SFJ incompetence and GSV reflux lasting \>0.5 s on DUS. Morrison et al [@JR1900086oa-12] 2017 United States RCT, 1B Multicenter\ RFA March--September 2013 (6 mo) 108 114 RFA -- 12 12 GSV reflux ≥0.5 s on DUS in the standing position. VeClose Trial Eroglu and Yasim [@JR1900086oa-19] 2018 Turkey RCT, 1B -- RFA, EVLA November 2014--June 2015 (7 mo) 168 139 EVLA\ 69 24 24 GSV \>5.5 mm and SSV \>4 mm in diameter 2 cm below the SFJ and SPJ with the patient standing, and reflux \>0.5 s. 149 RFA Proebstle et al [@JR1900086oa-33] 2015 Europe (multinational) Prospective, 2B Multicenter None December 2011--July 2012 (7 mo) 70 -- -- 12 12 Primary GSV incompetence diagnosed clinically +/− visible varicosities and confirmed by DUS. GSV diameter ≥3 mm and ≤10 mm on standing DUS. Kolluri et al [@JR1900086oa-34] 2016 United States Prospective, 2B Multicenter None March--September 2013 (6 mo) 20 -- -- 12 12 Moderate to severe varicosities and venous reflux in the GSV \>0.5 s. Çalık et al [@JR1900086oa-29] 2016 Turkey Prospective, 2B Multicenter None April--September 2014 (5 mo) 181 -- -- 6 7.5 GSV insufficiency with \>0.5 s of reflux. Tekin et al [@JR1900086oa-30] 2016 Turkey Prospective, 2B Single center None January--July 2014 (6 mo) 62 -- -- 6 8 Symptomatic incompetent GSV with a diameter of \>5.5 mm, with or without visible varicosities. Chan et al [@JR1900086oa-25] 2017 China Prospective, 2B Single center None September 2014--October 2015 (13 mo) 29 -- -- 12 9 Retrograde SFJ flow ≥0.5 s on DUS with patient standing. Gibson and Ferris [@JR1900086oa-32] 2017 United States Prospective, 2B Single center\ None October--December 2015 (3 mo) 50 -- -- 1 1 Reflux of \>0.5 s of retrograde flow in a varicose vein in the standing position. WAVES trial Almeida et al [@JR1900086oa-28] 2017 Dominican Republic Prospective, 2B Single center None December 2010 (1 mo) 38 -- -- 36 36 Clinical venous reflux disease in the GSV +/− varicosities, and confirmed by DUS. Eroglu et al [@JR1900086oa-20] 2017 Turkey Prospective, 2B Single center None May--October 2014 (5 mo) 168 -- 12 30 30 GSV diameter \>5.5 mm and a SSV diameter \>4 mm in conjunction with reflux \>0.5 s. Park [@JR1900086oa-13] 2017 South Korea Prospective, 2B Single center None December 2016--February 2017 (2 mo) 34 -- -- 3 3 Saphenous vein with ≥0.5 s of reflux in the standing position with a diameter of at least 3 mm. Koramaz et al [@JR1900086oa-22] 2017 Turkey Retrospective, 2C Single center EVLA May 2013--August 2014 (15 mo) 150 189 EVLA -- 12 12 GSV diameter ≥5.5 mm and ≤15 mm with reflux \>0.5 s. Chan et al [@JR1900086oa-26] 2017 China Retrospective, 2C Single center None September 2014--June 2016 (21 mo) 55 -- -- 12 5 Retrograde flow of \>0.5 s on DUS over the SFJ on standing. Bademci et al [@JR1900086oa-31] 2018 Turkey Retrospective, 2C Single center None September 2015--September 2016 (12 mo) 50 -- -- 12 12 GSV diameter of 5.5--10 mm with reflux \>0.5 s. Yavuz et al [@JR1900086oa-27] 2018 Turkey Retrospective, 2C Single center None April--July 2016 (3 mo) 538 -- -- 12 12 GSV diameter at SFJ of ≥5.5 mm and ≤15 mm on standing. GSV reflux ≥0.5 s on DUS. Yang et al [@JR1900086oa-21] 2019 Canada Retrospective, 2C Single center RFA January 2014--December 2016 (3 y) 106 182 RFA 47 2 2 Not defined. Lane et al [@JR1900086oa-18] 2013 United Kingdom Case report, 4 Single center None March 2012 1 -- -- 6 6 Not defined. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Abbreviations: AASV, anterior accessory saphenous vein; CEAP, Clinical, Etiological, Anatomical, and Pathophysiological; DUS, duplex ultrasound; EVLA, endovenous laser ablation; GSV, great saphenous vein; NBCA, *n* -butyl-2-cyanoacrylate; RCT, randomized controlled trial; RFA, radiofrequency ablation; s, seconds; SFJ, saphenofemoral junction; SSV, small saphenous vein. ###### Intraoperative characteristics of selected studies Author Ablation device GSV count SSV count AASV count Delivery catheter position distal to SFJ/SPJ (cm) Volume of glue used mean ± SD (range) (mL) Mean Vein diameter mean ± SD (range) (mm) Treated segment length mean ± SD (range) (cm) Procedure duration mean ± SD (range) (minutes) Concomitant treatment of tributaries (e.g., foam, phlebectomy) Concomitant GSV and SSV NBCA treatment Concomitant treatment of GSV and SSV with other endovascular modality (e.g., foam) Postoperative compression stockings -------------------------------------- ----------------- ----------- ----------- ---------------------------------------------- --------------------------------------------------- -------------------------------------------- ------------------------------------------- ----------------------------------------------- ------------------------------------------------ ---------------------------------------------------------------- ---------------------------------------- ------------------------------------------------------------------------------------ --------------------------------------------------------------------------------------------------------------------------------- Bozkurt and Yilmaz [@JR1900086oa-10] VariClose 154 0 0 3 -- 7.2 ± 1.8 29.8 ± 5.4 15 ± 2.5 No No No No Evlas 156 0 0 1.5 N/A 7.1 ± 1.6 29.7 ± 8.1 33.2 ± 5.7 No No No Yes × 10 d Morrison et al [@JR1900086oa-12] VenaSeal 108 0 0 5 1.2 (0.4--2.3) 5.6 32.8 (8--61) 24 (11--40) No No No Yes, for 3 d continuously, then 4 more days during waking hours. ClosureFast 114 0 0 -- N/A 5.85 35.1 (6.5--84.5) 19 (5--46) No No No Eroglu and Yasim [@JR1900086oa-19] VariClose 159 9 0 3 -- 7.6 ± 1.9 26.4 ± 6.5 15.3 ± 2.6 No Yes No Yes, elastic compression bandage for 2 d, then Class 1 for 1 mo. ClosureFast 146 3 0 -- N/A 7.8 ± 1.9 27.6 ± 5.3 27.3 ± 7.7 No Yes No Evlas 123 16 0 -- N/A 8 ± 1.9 27.1 ± 5.8 35.0 ± 5.2 No Yes No Proebstle et al [@JR1900086oa-33] VenaSeal 70 0 0 5 -- 7.8 ± 2.1 (6.6--14) 37.6 (7--72) 18.6 (8--74) No No Yes No Kolluri et al [@JR1900086oa-34] VenaSeal 20 0 0 5 1.1 (0.6--2.2) 6.1 31.4 (18--50) 31 (23--46) -- No No -- Çalık et al [@JR1900086oa-29] VariClose 206 9 0 3 0.9 (0.7--2.1) 5.85 31.6 ± 6.1 (23--70) 5.4 ± 2.5 (3--14) Yes Yes Yes Elastic bandages x1 d, no compression stockings. Tekin et al [@JR1900086oa-30] VariClose 62 0 0 5 1.5 7.5 ± 1.5 (5.5--13) 28 (20--40) 17 (9--37) No No No Elastic bandages x1 d, no compression stockings. Chan et al [@JR1900086oa-25] VenaSeal 57 0 0 4 -- 7.1 (3.9--11.4) 27 (17--33) 64 (28--99) Yes No No Yes Gibson and Ferris [@JR1900086oa-32] VenaSeal 48 8 14 5 0.93 ± 0.3 7.7 24 ± 12.8 27 ± 11 (11--55) No Yes No No Almeida et al [@JR1900086oa-28] VenaSeal 38 0 0 3.5 1.3 (0.63--2.25) -- 33.2 ± 9.1 20 (11--33) -- No -- No Eroglu et al [@JR1900086oa-20] VariClose 159 9 0 3 2 7.4 ± 2.3 (5.5--14) 26.3 ± 6.5 (9--43) 15.3 ± 2.5 (10--25) -- No No -- Park [@JR1900086oa-13] VenaSeal 47 16 0 5 1.2 ± 0.3 8.0 ± 3.7 (3.1--18) 37 ± 15 (5--67) 50.4 ± 20.3 (10--95) Yes Yes -- Yes, only those who underwent concomitant procedures (3 d for miniphlebectomy ( *n*  = 15), 7 d for sclerotherapy ( *n*  = 19). Koramaz et al [@JR1900086oa-22] VariClose 150 0 0 3 -- 6.88 ± 1.8 (5.5--15) 31.97 ± 6.83 7 (4--11) No No No No Evlas 189 0 0 0.5 [a](#FN1900086oa-7){ref-type="table-fn"} N/A 7.15 ± 1.77 (5.5--14) 31.65 ± 6.25 18 (14--25) No No No Yes, class 2 × 2 wk. Chan et al [@JR1900086oa-26] VenaSeal 108 0 0 4 -- 6.6 (2.3--11.4) 28 (15--41) 64 (28--116) Yes No No Yes, x 1 mo. Bademci et al [@JR1900086oa-31] VariClose 50 0 0 3 1.5 (1.3--2) 7 (5.5--9) 29.5 (25--36) 25 (20--36) No No No No Yavuz et al [@JR1900086oa-27] VenaBlock 538 0 0 4 0.87 ± 0.15 (0.4--1.39) 6.7 ± 1.7 (5.5--14.6) 25.7 ± 4.9 (10--43) 11.7 ± 4.9 (5--33) No No No No Yang et al [@JR1900086oa-21] VenaSeal 83 17 6 -- 1.8 ± 0.1 -- 43 ± 1 -- -- -- -- No Venefit 289 30 9 -- N/A -- 41 ± 1 -- -- -- -- -- Abbreviations: N/A, not applicable; SD, standard deviation. Note: Gray, comparators; --, no information. Evlas: Evlas Circular fiber EVLA kit (1,470 nm) (Biolas, Ankara, Turkey); Closurefast: Closurefast RFA catheter (VNUS Medical Technologies, San Jose, CA); Venefit: Venefit Targeted endovenous RFA therapy system (Medtronic of Canada Ltd, Vancouver, British Columbia). Distance to superficial epigastric vein. ### *n* -Butyl-2-Cyanoacrylate Mean age of the recruited population was 49.3 years and 64.8% were females. Most procedures from Turkey used the VariClose NBCA system (Biolas, FG Group, Ankara, Turkey). [@JR1900086oa-10] [@JR1900086oa-19] [@JR1900086oa-20] [@JR1900086oa-22] [@JR1900086oa-29] [@JR1900086oa-30] [@JR1900086oa-31] One study used VenaBlock adhesive (Invamed, Ankara, Turkey). [@JR1900086oa-27] The rest utilized the VenaSeal system (Medtronic, Dublin, Ireland). [@JR1900086oa-12] [@JR1900086oa-13] [@JR1900086oa-21] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-28] [@JR1900086oa-32] [@JR1900086oa-33] [@JR1900086oa-34] All procedures commenced by cannulation of the target vein with an introducer needle under ultrasound guidance at the most distal point of reflux. The position of the delivery catheter tip distal to SFJ or SPJ ranged from 3 to 5 cm. The average volume of NBCA glue used was 1.3 mL (range 0.87--2 mL) to treat veins with a mean length of 30.8 cm (range 24--43 cm) and diameter of 7 mm (range 5.6--8 mm). Procedure technique varied depending on the choice of NBCA device. For VenaSeal, two initial 0.09-mL glue aliquots were injected 1 cm apart, followed by 3 cm pullbacks between each trigger pull. Pressure with Ultrasound (US) probe was applied to occlude the SFJ/SPJ before dispensing the first two aliquots to prevent glue from entering the deep venous system. The first two injections were followed by 3 minutes of compression. US probe pressure was applied for 30 seconds after subsequent injections. The VariClose system used a similar technique in terms of initial pressure to occlude the SFJ or SPJ before first injection. The trigger was pressed for 5 seconds while withdrawing the catheter by 10 cm (giving 0.06 mL of glue at 2 cm/s). Pressure over each 10-cm segment of treated vein was applied for 30 seconds. Once the entire vein was treated, a further 30 seconds of pressure over the entire target vein was applied. VenaBlock used a similar method. Recording of duration of NBCA procedures was not standardized. Two prospective studies calculated duration from the time of insertion of NBCA delivery catheter to the time of withdrawal (mean 19.3 minutes). [@JR1900086oa-28] [@JR1900086oa-33] The period from establishing venous access to applying the final bandages was taken as procedure time in another two prospective studies, with an average of 38.7 minutes. [@JR1900086oa-13] [@JR1900086oa-32] An even broader timing interval extended from skin preparation to final bandaging, including phlebectomies (mean 64 minutes). [@JR1900086oa-25] [@JR1900086oa-26] One operator performed the procedures under intravenous sedation, which further extended length of intervention. [@JR1900086oa-13] ### Radiofrequency Ablation Three studies compared NBCA with RFA. [@JR1900086oa-12] [@JR1900086oa-19] [@JR1900086oa-21] The mean age of patients was 51 years and 72.8% were females. The devices used were ClosureFast (VNUS Medical Technologies, San Jose, CA) and Venefit (Medtronic of Canada Ltd, Vancouver, British Columbia, Canada). Both are similar and require perivenous tumescent anesthesia. Procedure duration was recorded in two RCTs and results were conflicting. [@JR1900086oa-12] [@JR1900086oa-19] On one side, NBCA took longer than RFA (24 vs. 19 minutes, *p*  \< 0.01). [@JR1900086oa-1] The other RCT identified a significant reduction in favor of NBCA (15.3 vs. 27.3 minutes, *p*  \< 0.001). [@JR1900086oa-19] Neither documented the actual commencement and completion of recording. ### Endovenous Laser Ablation EVLA was performed on 246 females (50.8%). Mean age was 44.4 years. Evlas Circular fiber EVLA kit (Biolas, Ankara, Turkey) was used in all three studies. It operates at a wavelength of 1,470 nm and uses tumescent anesthesia. Peak temperature reaches 1200°C (compared with 120°C for RFA). One retrospective analysis mentioned the application of manual pressure over the treated vein during laser fiber withdrawal but its benefit in terms of promoting vein closure was not investigated. [@JR1900086oa-22] Compression stockings were prescribed following all EVLA procedures and all agreed that EVLA took significantly longer than NBCA or RFA ( *p*  \< 0.001). [@JR1900086oa-10] [@JR1900086oa-19] [@JR1900086oa-22] Postoperative Success --------------------- ### Occlusion Rate [Fig. 2](#FI1900086oa-2){ref-type="fig"} shows a substantial initial success rate after NBCA ablation followed by RFA and EVLA, respectively. Although limited, the 2-year NBCA data are superior. There is negligible difference between RFA and EVLA plots from 6 months onward. Partial and complete recanalization rates were lowest for NBCA throughout the period of follow-up. {#FI1900086oa-2} ### Complications There were no pulmonary embolic events. Nine cases of postablation DVT were observed in the NBCA group ( [Fig. 3](#FI1900086oa-3){ref-type="fig"} ). [@JR1900086oa-21] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-29] [@JR1900086oa-32] [@JR1900086oa-33] Four DVTs were reported in the RFA group and three following EVLA (endovenous heat-induced thrombi Class 1) without statistical significance. [@JR1900086oa-12] [@JR1900086oa-19] [@JR1900086oa-21] [@JR1900086oa-22] All resolved with or without heparin treatment. Bruising was least in NBCA-treated patients. [@JR1900086oa-1] [@JR1900086oa-25] [@JR1900086oa-26] All RCTs reported a statistically significant lower incidence of ecchymosis in the NBCA group. [@JR1900086oa-1] [@JR1900086oa-10] [@JR1900086oa-19] One explanation is that repeated injections are required for tumescent anesthesia in thermal ablation methods while these are avoided in NBCA. [@JR1900086oa-1] However, one retrospective comparative analysis found that five (2.65%) of EVLA-treated patients developed bruising which did not reach the level of significance compared with NBCA, even though such adverse event was absent in the latter cohort. [@JR1900086oa-22] One prospective and one retrospective study by the same author using NBCA concluded that bruising resulted from stab avulsion sites which was performed in the same sitting. [@JR1900086oa-25] [@JR1900086oa-26] Three studies documented minor point bruising at the access site of NBCA delivery catheter due to residual NBCA being applied close to the entry point. [@JR1900086oa-27] [@JR1900086oa-31] [@JR1900086oa-33] Bleeding and hematoma formation were reported in one patient who underwent NBCA ablation and two post-RFA, the latter being at the site of vein access. [@JR1900086oa-19] [@JR1900086oa-30] Paresthesia was temporary and less frequent in the NBCA group. [@JR1900086oa-10] [@JR1900086oa-12] [@JR1900086oa-21] [@JR1900086oa-22] [@JR1900086oa-25] Seven patients complained of pigmentation at the treatment site after NBCA ablation which improved significantly over 1 year. [@JR1900086oa-10] [@JR1900086oa-13] [@JR1900086oa-31] A higher number was reported after EVLA and were shown to be statistically significant. [@JR1900086oa-22] All were temporary. Phlebitis after NBCA ablation was significantly less than post-RFA or EVLA. [@JR1900086oa-21] [@JR1900086oa-22] One RCT reported the opposite, but failed to reach significance level. [@JR1900086oa-1] Most reactions were transient and self-limiting or resolved with a short course of nonsteroidal anti-inflammatory drugs. [@JR1900086oa-1] [@JR1900086oa-13] [@JR1900086oa-26] [@JR1900086oa-32] [@JR1900086oa-33] Antibiotics were prescribed in two studies. [@JR1900086oa-22] [@JR1900086oa-29] [Supplementary Fig. 2](#SM1900086oa-1){ref-type="supplementary-material"} compares the different NBCA glue products with the proportion of veins having postoperative phlebitis. Although inconsistently and heterogeneously recorded, intraoperative pain experience was least for cyanoacrylate procedures, presumably because of the lack of tumescent anesthesia and heat generation. It was therefore better tolerated. [@JR1900086oa-10] [@JR1900086oa-19] [@JR1900086oa-32] Most subjects returned to work the following day and this was superior to RFA and EVLA. [@JR1900086oa-19] [@JR1900086oa-20] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-32] One patient developed generalized urticaria after the first week of treatment indicating delayed NBCA allergy. This settled with oral antihistamines and steroids. [@JR1900086oa-32] {#FI1900086oa-3} VCSS and Quality of Life Scores ------------------------------- All endovenous ablation modalities exhibited a statistically significant decline in VCSS scores over time. [@JR1900086oa-10] [@JR1900086oa-12] [@JR1900086oa-19] [@JR1900086oa-20] [@JR1900086oa-22] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-27] [@JR1900086oa-28] [@JR1900086oa-29] [@JR1900086oa-31] [@JR1900086oa-33] [@JR1900086oa-34] Two RCTs reported no difference between NBCA and EVLA during follow-up and another favored NBCA at 2 years ( *p*  \< 0.001). [@JR1900086oa-10] [@JR1900086oa-19] [@JR1900086oa-22] Two prospective analyses by Gibson and Park were analyzed separately because they used the revised version of VCSS. [@JR1900086oa-35] Mean baseline scores were 6.5 ± 2.4 (3--14) and 4.3 ± 2.1 (2--13). At 30 days, these improved respectively to 1.8 ± 1.4 (0--6) and 1.2 ± 1.0 (0--5) ( *p*  \< 0.001 and 0.024). [@JR1900086oa-13] [@JR1900086oa-32] The AVVQ was the main reporting modality for quality of life. Its downward decline from baseline was significant, consistent, and similar in all groups. Few manuscripts utilized other quality of life scores including EQ-5D, EQ-5D TTO, CIVIQ, and SF-36. All except SF-36 exhibited a significant improvement from baseline. [@JR1900086oa-1] [@JR1900086oa-12] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-29] [@JR1900086oa-32] [@JR1900086oa-33] [@JR1900086oa-34] Influence of Variables on Occlusion Rate ---------------------------------------- Occlusion rate after cyanoacrylate glue treatment is not influenced by vein length, diameter, dispensing device, or use of postoperative compression stockings ( [Table 4](#TB1900086oa-4){ref-type="table"} ). ###### Analysis of the effect of four variables on occlusion rate of NBCA-treated veins (Spearman\'s correlation, Mann-Whitney U test ^b^ ) *p* -Values ------- ------------- ------- ------- ------- 3 mo 0.728 0.538 0.593 0.564 6 mo 0.423 0.413 0.295 0.521 12 mo 0.931 0.160 0.873 0.240 Abbreviation: NBCA, *N* -butyl-2-cyanoacrylate. Discussion ========== Monomeric cyanoacrylate compounds polymerize upon contact with anionic components of plasma, a process consisting of three distinct phases: initial rapid polymerization with linear increase in tensile forces lasting approximately 10 seconds (phase 1), stable tensile forces lasting approximately 60 seconds (phase 2) followed by a more rapid rise of tensile forces (phase 3). [@JR1900086oa-36] The process of luminal fibrosis after glue injection takes several weeks before it becomes permanent. [@JR1900086oa-37] Adjunctive treatments (phlebectomy or foam sclerotherapy) risk a type 2 error and the confounding potential of these treatments is a subject of future trials. [@JR1900086oa-10] [@JR1900086oa-12] [@JR1900086oa-13] [@JR1900086oa-19] [@JR1900086oa-20] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-27] [@JR1900086oa-29] [@JR1900086oa-33] There were outliers that skewed the NBCA occlusion data at 6 months and 1 year, leading to a dip in success rate at these intervals. [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-30] Bissacco et al reviewed 1,000 NBCA cases in seven studies (two prospective, four retrospective) and found 96.8% of veins occluded at 12 months. [@JR1900086oa-38] Two studies reported NBCA occlusion beyond the 2-year interval, and these were 94.1% at 30 months and 94.7% at 36 months, respectively. [@JR1900086oa-20] [@JR1900086oa-28] Time to complete occlusion was shorter for NBCA than any of the endothermal modalities because veins are instantly occluded by approximation of their intima, while thermal ablation is dependent on vein wall destruction and subsequent fibrosis---a biological process which takes longer. [@JR1900086oa-12] The outcomes of RFA versus EVLA have been extensively studied in previous trials. Using the ClosureFast RFA system on 200 limbs (163 GSVs and 41 SSVs), Choi et al reported 94.6% occlusion in GSV and 94.5% in SSV at 13.9 months, which is similar to our data. [@JR1900086oa-39] A prospective double-blind RCT comparing RFA versus EVLA (159 patients---79 RFA, 80 EVLA) by Nordon et al identified a 100% occlusion at 7 days. The 3-month occlusion rate reached 97% for RFA and 96% for EVLA. There was no significant difference between the groups. [@JR1900086oa-40] In the LARA study, Goode et al reported 95 and 74% occlusion rate for RFA at 10 days and 9 months respectively. For EVLA, these were 95 and 78%. The high failures at 9 months were attributed to incorrect setting on the RFA which improved to 98% upon adjustment. No reasons for EVLA failures were given but the short wavelength of the laser used (810 nm) and pullback speed might be implicated. [@JR1900086oa-41] Recanalization does not necessarily signify return of symptoms as many maintain a good quality of life and anticoagulation does not appear to be a predisposing factor. [@JR1900086oa-12] [@JR1900086oa-25] [@JR1900086oa-27] [@JR1900086oa-29] NBCA is noninferior to RFA in terms of freedom from recanalization. [@JR1900086oa-12] Chan et al found a significant risk in their earlier study with vein diameters ≥8 mm, which was reduced to ≥6.6 mm in a subsequent analysis. [@JR1900086oa-25] [@JR1900086oa-26] This was contradicted in the WAVES study which reported 100% occlusion at 30 days using the same NBCA system. However, the latter allowed operators to inject additional glue in larger veins according to their discretion. [@JR1900086oa-32] Other reported determinants of failure were operator experience, anatomical variation (e.g., aneurysms, junction of large varicosities), [@JR1900086oa-29] development of incompetency in a once competent vein, intraluminal thrombus formation (most relevant for failure after thermal ablation), and missing the vein altogether. [@JR1900086oa-12] [@JR1900086oa-22] [@JR1900086oa-25] [@JR1900086oa-26] [@JR1900086oa-29] There is no officially reported incidence of DVT for NBCA but it is understood to be very low especially if tip of catheter is positioned 5 cm away from the superficial-to-deep vein junction. RFA carries a risk of 0 to 16% and EVLA 0 to 7.7%. Routine postoperative DUS may pick up asymptomatic thrombi. [@JR1900086oa-39] The benefit of anticoagulation for such DVTs is debatable as most resolve spontaneously. Ultrasound guidelines distinguishing thrombus from glue are also lacking. [@JR1900086oa-26] No details about length of stockings were provided (example: thigh high or below knee). Bruising was least after NBCA, particularly when glue injection was stopped 2 cm proximal to the catheter entry site. [@JR1900086oa-27] A modern laser with longer wavelength (1,470 nm) causes less ecchymosis than one with shorter wavelength (810 nm) because it is less damaging to the vessel wall. [@JR1900086oa-2] [@JR1900086oa-41] Prior to this improvement in laser technology RFA was deemed superior to EVLA with regards to postprocedural bruising. [@JR1900086oa-19] [@JR1900086oa-40] [@JR1900086oa-41] Other factors implicated in ecchymosis include the use of tumescent anesthesia, phlebectomies, anticoagulants, body mass index, and ethnicity. [@JR1900086oa-40] Paresthesia typically occurs in 1 to 2% of cases post-RFA and EVLA, and is rare after NBCA. In the latter it is often mild and self-limiting. [@JR1900086oa-2] A few recent studies and case reports address the issue of hypersensitivity reactions causing phlebitis-like signs and symptoms in veins treated with cyanoacrylate glue. Generally these respond well to antihistamines and/or steroids, and may even resolve spontaneously. In those veins requiring excision, histological examination identified features of a type IV hypersensitivity reaction to the glue (foreign body). [@JR1900086oa-42] [@JR1900086oa-43] [@JR1900086oa-44] This is different from the phlebitis encountered after thermal ablation. Patients should be asked about cyanoacrylate allergy preoperatively to minimize risks. [@JR1900086oa-45] This systematic review has some limitations. The comprehensive literature search and data extraction were performed by one author. It excluded mechanochemical ablation and the period of follow-up was short. A meta-analysis would have been ideal but as highlighted in a recent article, the scarcity and heterogeneity of RCTs made this difficult. [@JR1900086oa-38] As most patients were not sedated, double blinding was impossible. Outcome assessors were often the same ones recruiting, carrying out the treatments and/or following-up patients. This was taken into consideration in part by modifying the Cochrane risk of bias tool. [@JR1900086oa-15] Some methodologies opted for an induction period to cater for the "learning curve" but others did not. [@JR1900086oa-12] [@JR1900086oa-34] One major inconsistency was in the duration of procedures. There are no set standards as to when time-keeping should start and stop. The lack of reproducibility makes these measurements unreliable. In terms of patient characteristics, one study included more smokers in the NBCA group and another deviated its protocol to include a patient with higher BMI. [@JR1900086oa-12] [@JR1900086oa-28] No differentiation between unilateral or bilateral treatment of varicose veins was made. [@JR1900086oa-25] [@JR1900086oa-26] "Return to normal activities" needs better definition, as these activities are different in an elderly or morbidly obese patient compared with a healthy fit subject. Reflux is best detected in the standing posture on DUS as recommended by the European Society for Vascular Surgery (2015), but some measured this supine. [@JR1900086oa-2] Lastly, it would be interesting to see a trial addressing NBCA use for varicose veins in anticoagulated patients. Conclusion ========== This systematic review shows the potential benefits of cyanoacrylate glue over RFA and EVLA. Due to its immediate action, occlusion is retained even without postoperative elastic bandages or compression stockings. Patients experienced less pain as there was no tumescent anesthesia, multiple injection sites, or heat involved. Phlebitis is often mild, self-limiting, and attributed to localized skin reaction to the glue. It can be managed conservatively. Procedure times are generally short and patients typically resume work on day 1 or 2. Failure rates are less but longer-term data are required to affirm this. Cyanoacrylate ablation carries less risk of paresthesia, ecchymosis, and eliminates burn injuries. The two most readily available NBCA kits can be used on various lengths and diameters of veins (including bilateral cases of appropriate length with a single vial of glue). [@JR1900086oa-25] Ms Claire Gatt, BSc is the statistician who performed statistical analysis on the data. Ms Bianca-Maria Dimech, BSc Pharm is the second proofreader of the article. **Conflict of Interest** None declared by the authors. Not required as data analysis was performed from previously published data. Supplementary Material ====================== ###### Supplementary Material ###### Supplementary Material

Introduction ============ The petri dish and test tube methods are the two sub types of microbiological inhibition methods. Compared to petri dish methods, the test tube methods are more suitable for high-throughput screening of antimicrobial drugs residues in animal food because it is neither time consuming nor laborious ([@B15]). *Geobacillus stearothermophilus* is the most widely used indicator bacterium in microbiological inhibition methods in terms of test tubes, as it is not easily contaminated, demands high incubation temperature (55°C) and grows faster in a short time (less than 4 h) than other bacteria. Moreover, it is more sensitive to antimicrobial agents, particularly, β-lactam ([@B10]). Additionally, spores of *G. stearothermophilus* are more resistant to adverse factors than vegetative cells and show stable activity for a long time. Therefore, spores of *G. stearothermophilus* can be added into kit's medium during the kits preparation process, which simplifies the detection procedure and prolongs the shelf life of kits. However, *G. stearothermophilus* is not sensitive enough to many commonly used antibiotics except β-lactam ([@B16]). In past years, a number of studies by microbiological inhibition methods in terms of test tubes were developed to improve the sensitivity of *G. stearothermophilus* to different kinds of antibiotics residues in milk. There are brilliant black reduction test (BRT AIM) ([@B13]), Copan milk test ([@B11]), Doveltest SP-NT ([@B1]), Eclipse100^®^ ([@B2]), and Charm^®^ Blue-Yellow II ([@B12]). Among these kits, Charm^®^ Blue-Yellow II can detect more antibacterial drugs including β-lactam, aminoglycosides, tetracyclines, sulfonamides, and macrolides. However, this method is not sensitive enough to aminoglycosides and macrolides, and extremely insensitive to quinolones. The chicken egg and honey are also consumed daily and important for human health. However, little research by microbiological inhibition methods in terms of test tubes is known about chicken egg and honey. Even Premi^®^Test, the test tube method is widely applied for the detection of antibiotics residues in milk, muscle, kidney, egg, honey and feed etc. However, Premi^®^ Test is not considered ideal to detect residual antibiotics in chicken egg and honey, as it does not show enough sensitivity to aminoglycosides, macrolides and quinolones ([@B19]). Therefore the aim of the present study was to develop a new test tube method with *G. stearothermophilus var* C953, which was more sensitive to a different kind of antimicrobial agents especially aminoglycosides, macrolides and quinolones in milk, chicken egg, and honey. Materials and Methods {#s1} ===================== Antimicrobial Standards ----------------------- β--lactam: penicillin G (PEN), cefquinome (CEF); aminoglycosides: neomycin (NEO), streptomycin (STR); tetracyclines: doxycycline (DOX), tetracycline (TET); macrolides: erythromycin (ERY), spiramycin (SPI); sulfonamides: sulfadiazine (SDZ), sulfadimidine (SDM); lincosamides: lincomycin (LIN); quinolones: danofloxain (DAN), enrofloxacin (ENR); trimethoprim (TMP); and chloramphenicol (CAP) were all purchased from Sigma-Aldrich (St. Louis, MO, United States). Drugs for the preparation of antimicrobial solutions were stored and handled according to the manufacturers' instructions before use. In addition, the methods for the preparation of stock solutions and working standard solutions of antibiotics were shown in [Table 1](#T1){ref-type="table"}. ###### Methods for the preparation of stock solutions and working standard solutions of antibiotics. Antimicrobial agents Solvents Diluents ---------------------- -------------------------------------- -------------------------------------- β --lactams Phosphate buffer, pH 6.0, 0.1 mol/L Phosphate buffer, pH 6.0, 0.1 mol/L Aminoglycosides Tris, pH 8.0, 0.01 mol/L Tris, pH 8.0, 0.01 mol/L Tetracyclines HCl, 0.1 mol/L Phosphate buffer, pH 6.0, 0.1 mol/L Macrolides Phosphate buffer, pH 8.0, 0.01 mol/L Phosphate buffer, pH 8.0, 0.01 mol/L Sulfonamides NaOH, 0.1 mol/L Sterilized distilled water Lincosamides Phosphate buffer, pH 8.0, 0.01 mol/L Phosphate buffer, pH 8.0, 0.01 mol/L Quinolones NaOH, 0.1 mol/L Phosphate buffer, pH 8.0, 0.1 mol/L TMP Glacial acetic acid Sterilized distilled water CAP Methanol Sterilized distilled water Test Organism ------------- *Geobacillus stearothermophilus var* C953 was obtained from American Type Culture Centre (ATCC), Rockville, MD, United States. Recovery, Preparation and Conservation of Test Organism ------------------------------------------------------- A freeze-dried strain of *G. stearothermophilus var* C953 was dissolved in sterile physiological saline (0.85% NaCl). A 100 μL of *G. stearothermophilus var* C953 suspension was inoculated into nutrient agar with 0.035 g/L MnSO~4~ ⋅ H~2~O and incubated in incubator for 24 h at 55°C. After three generations recovery, a single culture from nutrient agar with 0.035 g/L MnSO~4~⋅ H~2~O was inoculated into a new same medium and incubated in incubator for 72 h at 55°C. At the end of incubation, the cells were washed from medium by 10% (v/v) dried skimmed milk. After collection, the cells suspension was dispended into amber vials. Aliquots of cell suspensions stored at 4, -20, and -80°C for 6 h respectively step by step. After that, the frozen cells suspension was freeze-dried by freeze vacuum dryer and stored at -80°C until usage. Preparation of Kit's Medium Components -------------------------------------- Plate Count Agar (Becton Dickinson) fortified with glucose (6 g/L; Sigma^®^) was used. The medium was sterilized at 121°C for 15 min. After the medium was cool down to 50 ± 1°C, its pH was adjusted to 7.8 ± 0.1. After that, *G. stearothermophilus var* C953 spore suspension (5 × 10^9^ CFU/L), along with bromocresol purple (0.1 mg/L, Mallinckrodt^®^) and sensitizers such as 50 μg/L trimethoprim (TMP), 40 μg/L chloramphenicol (CAP), 45 μg/L streptomycin (STR) and 60 μg/L enrofloxacin (ENR) were added. A 150 μL of medium was added into each well of microtiter plates by using an electronic pipette (Eppendorf Research^®^Pro) after kit's medium components mixed well. Finally, these microtiter plates were sealed with aluminized film and conserved at 4°C until use. Control Samples --------------- Milk samples were collected from the dairy farm of Huazhong Agricultural University (HZAU), Wuhan, Hubei, China. At the time of samples collection, the cows did not receive any antimicrobial substances in the last 9 weeks and were at postpartum stage (between 60 and 90 days). Because bovine milk presented normal values of chemical composition, total bacterial counts (CFU \< 100,000 mL^-1^) and somatic cell counts (SCC \< 400,000 mL^-1^) ([@B5]) during these days. Milk samples were kept at 4°C for approximately 2 days throughout the experiment. The chicken eggs were collected from laying hens (30 weeks old) with a history of no antimicrobial drugs used either in the form of treatment or growth promoter in last 6 weeks at the chicken farm of HZAU. And chicken eggs were kept at 4°C within 1 week before use. Honey samples were purchased from the local bee farmer and the absence of any antimicrobial substances was confirmed by high performance liquid phase tandem mass spectrometry ([@B5]). Moreover, honey samples were stored at 4°C for less than 1 week before use. Spiked Samples -------------- Spiked samples were prepared from the respective antibiotics working standard solutions in a single step using antimicrobial drugs-free respective antibiotics diluents, milk, homogeneous eggs and diluted honey (spiked levels see [Tables 2](#T2){ref-type="table"}--[5](#T5){ref-type="table"}). In addition, eight concentrations at different levels were prepared for each drug, and 24 replicates were prepared for each concentration. ###### Limit of detection (LODs) of microbiological system in antimicrobial agents' diluents (3.75 h). Antimicrobial agents Spiked levels /(μg/L) EU/CODEX MRL in milk^1,2^/(μg/L) This kit /(μg/L) ---------------------- -------------------------------------- ---------------------------------- ------------------ Penicillin G 0, 1, 2, 3, 4, 5, 6, 8 4 2 Cefquinome 0, 2.5, 5, 10, 20, 40, 60, 80 20 20 Neomycin 0, 25, 50, 75, 100, 150, 200, 300 1500 50 Streptomycin 0, 50, 100, 200, 250, 500, 750, 1000 200 200 Doxycycline 0, 25, 50, 75, 100, 150, 200, 300 0 50 Tetracycline 0, 50, 75, 100, 200, 250, 300, 400 100 100 Erythromycin 0, 10, 20, 30, 40, 50, 75, 100 40 40 Spiramycin 0, 50, 75, 100, 200, 250, 300, 400 200 200 Sulfadiazine 0, 25, 50, 75, 100, 150, 200, 300 100 50 Sulfadimidine 0, 50, 75, 100, 200, 250, 300, 400 100 100 Lincomycin 0, 25, 50, 75, 100, 150, 200, 300 150 150 Danofloxain 0, 50, 75, 100, 200, 250, 300, 400 30 100 Enrofloxacin 0, 50, 100, 180, 200, 220, 250, 280 100 180 1 ( The European Commission (2010) ). 2 Food, 2015 . ###### LODs of microbiological system in milk (3 h). Antimicrobial agents Spiked levels /(μg/L) EU/CODEX MRL in milk^1,2^ /(μg/L) This kit /(μg/L) ---------------------- -------------------------------------- ----------------------------------- ------------------ Penicillin G 0, 1, 2, 3, 4, 5, 6, 8 4 2 Cefquinome 0, 10, 20, 30, 35, 40, 45, 50 20 40 Neomycin 0, 25, 50, 75, 100, 150, 200, 300 1500 50 Streptomycin 0, 50, 100, 200, 220, 250, 280, 300 200 200 Doxycycline 0, 25, 50, 75, 100, 150, 200, 300 0 100 Tetracycline 0, 100, 200, 250, 300, 320, 350 100 300 Erythromycin 0, 10, 20, 30, 40, 50, 75, 100 40 40 Spiramycin 0, 50, 100, 200, 220, 250, 280, 300 200 200 Sulfadiazine 0, 25, 50, 75, 100, 150, 200, 300 100 150 Sulfadimidine 0, 100, 200, 250, 300, 320, 350 100 300 Lincomycin 0, 25, 50, 75, 100, 120, 150, 180 150 120 Danofloxain 0, 50, 75, 100, 200, 250, 300, 400 30 100 Enrofloxacin 0, 100, 200, 300, 400, 430, 450, 480 100 400 1 ( The European Commission (2010) ). 2 Food, 2015 . ###### LODs of microbiological system in chicken egg (3.5 h). Antimicrobial agents Spiked levels /(μg/L) EU/CODEX MRL in chicken egg^1,2^ /(μg/L) This kit /(μg/L) Premi^®^ Test ([@B19]) /(μg/L) ---------------------- -------------------------------------- ------------------------------------------ ------------------ -------------------------------- Penicillin G 0, 1, 2, 3, 4, 5, 6, 8 \- 4 \<2.5 Cefquinome 0, 2.5, 5, 10, 20, 40, 60, 80 \- 40 / Neomycin 0, 25, 50, 75, 100, 150, 200, 300 500 100 / Streptomycin 0, 50, 100, 200, 250, 500, 750, 1000 \- 200 / Doxycycline 0, 25, 50, 75, 100, 150, 200, 300 \- 100 200 Tetracycline 0, 50, 75, 100, 200, 250, 300, 400 200 300 200 Erythromycin 0, 10, 20, 30, 40, 50, 75, 100 150 40 / Spiramycin 0, 50, 75, 100, 200, 250, 300, 400 \- 200 / Sulfadiazine 0, 25, 50, 75, 100, 150, 200, 300 \- 150 \<25 Sulfadimidine 0, 50, 75, 100, 200, 250, 300, 400 \- 300 50 Lincomycin 0, 25, 50, 75, 100, 150, 200, 300 50 50 / Danofloxain 0, 50, 75, 100, 200, 250, 300, 400 \- 100 / Enrofloxacin 0, 50, 100, 200, 300, 400, 500, 600 \- 400 / "-"means no MRL. "/" means not detected. 1 ( The European Commission (2010) ). 2 Food, 2015 . ###### LODs of microbiological system in honey (3.25 h). Antimicrobial agents Spiked levels /(μg/L) Recommended concentration (RC) ([@B4]) /(μg/L) This kit /(μg/L) Premi^®^ Test ([@B19]) /(μg/L) ---------------------- -------------------------------------- ------------------------------------------------ ------------------ -------------------------------- Penicillin G 0, 1, 2, 3, 4, 5, 6, 8 \- 4 5 Cefquinome 0, 2.5, 5, 10, 20, 40, 60, 80 \- 40 25 Neomycin 0, 25, 50, 75, 100, 150, 200, 300 40 50 / Streptomycin 0, 50, 100, 200, 250, 500, 750, 1000 200 \>400 Doxycycline 0, 25, 50, 75, 100, 150, 200, 300 20 100 10 Tetracycline 0, 50, 75, 100, 200, 250, 300, 400 300 10 Erythromycin 0, 10, 20, 30, 40, 50, 75, 100 20 40 15 Spiramycin 0, 50, 75, 100, 200, 250, 300, 400 200 / Sulfadiazine 0, 25, 50, 75, 100, 150, 200, 300 50 150 25 Sulfadimidine 0, 50, 75, 100, 200, 250, 300, 400 300 25 Lincomycin 0, 10, 20, 30, 50, 75, 100, 150 \- 30 25 Danofloxain 0, 50, 75, 100, 200, 250, 300, 400 \- 100 / Enrofloxacin 0, 50, 100, 200, 300, 400, 500, 600 \- 200 200 "-"means no recommended concentration. "/" means not detected . Evaluation Protocol ------------------- The whole evaluation protocol of the kit was shown in [Figure 1](#F1){ref-type="fig"}. Firstly, the number of wells in microtiter plates needed were cut off and their aluminum foil were removed carefully from wells. Secondly, a 50 μL control and spiked samples were added into each well of microplates. Thirdly, the microplates were pre-incubated at room temperature (RT) for 20 min to allow the sample to diffuse through the medium. Fourthly, the remaining sample on the microplates medium surface was eliminated by inverting microplates and the wells were washed thrice with distilled water. Fifthly, the wells were sealed with an adhesive sheet and the microplates having milk and chicken egg samples were incubated in water bath for 10 min at 80°C while the microplates having honey samples were incubated in water bath for 1 h at 45°C. Finally, microtiter plates were incubated in microplates incubator at 65°C until the negative control sample had turned into yellow (approximately 3--4 h). The end-point is determined by visually assessing the color change in wells of microtiter plates. During the incubation period, the wells agar bed can be divided into three theoretical vertical zones, a score is assigned to the sample based on the zone color action pattern. An example is presented in [Figure 2](#F2){ref-type="fig"}. 3 zones yellow and 2/3 yellow = negative (-), 1/2 yellow = detection limit (+/-), 2/3 purple and 3 zones purple = positive (+). {#F1} {#F2} Validation Protocol ------------------- ### Limit of Detection (LOD) The dose--response curves of the antimicrobial agents were established according to the ISO13969: 2003 guidelines. Eight concentrations were prepared with different levels for each drug, and twenty-four replicates were prepared for each concentration. The LOD were estimated as the concentration that was 95% of positive results ([@B8]). ### Specificity and Selectivity One hundred control samples of milk, chicken eggs and honey respectively were analyzed with this kit for the determination of false-positive rate. The sample pre-treatment method was same as described in the "Evaluation Protocol" section. Moreover, the false-positive rate values were calculated as follows: False-Positive Rate = (Numbers of Positive Samples/Total Control Samples) × 100% However, one hundred control samples of each animal origin food spiked at the level of interest (MRL or LOD) were analyzed with this kit for the determination of false-negative rate. The method of sample pre-treatment was similar to described in the "Evaluation Protocol" section. Additionally, the false-negative rate values were calculated as follows: False-Negative Rate = (Numbers of Negative Samples/Total Spiked Samples) × 100% Ruggedness ---------- To determine the ruggedness of this kit, the effects of five factors including five different wells in one microplate, five different microplates in same batch, five different batches microplates, two different breeds (buffalo milk, Holstein milk), three different analysts on the false-positive rate, false-negative rate, sensitivity and detection time were evaluated. The ruggedness experiment was repeated three times for each factor. Moreover, the robustness study focused on seven representative antimicrobial agents of seven different kinds of antibiotics. In addition, the ruggedness of the kit was represented by the coefficient of variations (CVs). Stability --------- The kit stability was determined on the basis of appearance, smell, detection capability, detection time, which were evaluated with same batch kits stored at 4°C over 6 months (0, 7, 15, 30, 60, 90, 120, 150, 180 days). The kits stability experiment was performed for three batches kits. Additionally, the validation experiment focused on seven representative antimicrobial agents of different kinds of antibiotics and milk. Confirmation by Liquid Chromatography -- Tandem Mass Spectrometry (LC/MS-MS) ---------------------------------------------------------------------------- Seven Holstein cows at the stage of postpartum (between 60 and 90 days) and with a history of no antibiotics exposure in last 9 weeks were raised in an ideal environmental condition of standard temperature and humidity at the dairy farm of HZAU (Wuhan, Hubei, China). The seven cows were treated with PEN, STR, SDZ, LIN, and ENR by intramuscular injection, however, TET and ERY by intravenous injection respectively. Three milk samples from each cow were collected and tested for the presence of antibiotics residues at intervals of 0, 24, 48, 72, and 96 h respectively after drugs administration. All samples were analyzed by the kit in present study as described in the "Evaluation Protocol" section and by a multi-residue LC/MS-MS method ([@B9]). Results ======= Detection Capability -------------------- The detection capabilities of the kit used in present study against 13 different antibiotics belonging to seven different groups in respective antibiotics diluents was shown in [Table 2](#T2){ref-type="table"}. It was observed that the LODs of the kit were less than or equal to MRL in milk for β-lactam, aminoglycosides, TET, macrolides, sulfonamides and lincosamides, however, the LODs for DOX and quinolones were higher than MRL in milk. The LODs of the kit for different kinds of antibiotics in milk were given in [Table 3](#T3){ref-type="table"}. It was revealed that the LODs of the kit were less than or equal to MRL in milk for β-lactam, aminoglycosides, macrolides, lincosamides. However, the LODs for tetracyclines, sulfonamides and quinolones were higher than MRL in milk. The detection capability of this kit for different kinds of antibiotics in chicken egg was given in [Table 4](#T4){ref-type="table"}. There are MRLs only for NEO, TET, ERY, LIN in chicken egg. It indicated that the LODs of this kit for all kinds of antibiotics in chicken eggs were same like determined in milk. Moreover, the LODs for NEO, ERY, LIN were less than or equal to MRL in chicken egg. The LODs of this kit for various antibiotics in honey were shown in [Table 5](#T5){ref-type="table"}. In the case of honey, there are no MRLs for antibiotics residues, but the recommended concentration of aminoglycosides, tetracyclines, macrolides, sulfonamides were used as such ([@B4]). It was known that the LODs of this kit for different kinds of antibiotics in honey were similar to those determined in milk. However, the LODs for aminoglycosides, tetracyclines, macrolides, sulfonamides were higher than the recommended concentrations ([@B4]). Specificity ----------- Results showed that the false positive rate of this kit used in milk, chicken egg and honey all were 0%. The false-negative rate results of this kit used in milk, chicken egg and honey were given in [Tables 6](#T6){ref-type="table"}--[8](#T8){ref-type="table"}. It indicated that the false-negative rate of this kit used in three animal foods all were 0%. ###### False negative rates of the kit in milk. Antibiotics MRL /(μg/L) LOD /(μg/L) Spiked concentration /(μg/L) Sample numbers Negative sample numbers False negative rate/% --------------- ------------- ------------- ------------------------------ ---------------- ------------------------- ----------------------- Penicillin G 4 2 4 100 0 0 Cefquinome 20 40 40 100 0 0 Neomycin 1500 50 1500 100 0 0 Streptomycin 200 200 200 100 0 0 Doxycycline 0 100 100 100 0 0 Tetracycline 100 300 300 100 0 0 Erythromycin 40 40 40 100 0 0 Spiramycin 200 200 200 100 0 0 Sulfadiazine 100 150 150 100 0 0 Sulfadimidine 100 300 300 100 0 0 Lincomycin 150 120 150 100 0 0 Danofloxain 30 100 100 100 0 0 Enrofloxacin 100 400 400 100 0 0 ###### False negative rates of the kit in chicken egg. Antibiotics MRL /(μg/L) LOD /(μg/L) Spiked concentration /(μg/L) Sample numbers Negative sample numbers False negative rate/% --------------- ------------- ------------- ------------------------------ ---------------- ------------------------- ----------------------- Penicillin G \- 4 4 100 0 0 Cefquinome \- 40 40 100 0 0 Neomycin 500 100 500 100 0 0 Streptomycin \- 200 200 100 0 0 Doxycycline \- 100 100 100 0 0 Tetracycline 200 300 300 100 0 0 Erythromycin 150 40 150 100 0 0 Spiramycin \- 200 200 100 0 0 Sulfadiazine \- 150 150 100 0 0 Sulfadimidine \- 300 300 100 0 0 Lincomycin 50 50 50 100 0 0 Danofloxain \- 100 100 100 0 0 Enrofloxacin \- 400 400 100 0 0 "-" means no MRL . ###### False negative rates of the kit in honey. Antibiotics Recommended concentration (RC) ([@B4]) / (μg/L) LOD /(μg/L) Spiked concentration /(μg/L) Sample numbers Negative sample numbers False negative rate/% --------------- ------------------------------------------------- ------------- ------------------------------ ---------------- ------------------------- ----------------------- Penicillin G \- 4 4 100 0 0 Cefquinome \- 40 40 100 0 0 Neomycin 40 50 50 100 0 0 Streptomycin 200 200 100 0 0 Doxycycline 20 100 100 100 0 0 Tetracycline 300 300 100 0 0 Erythromycin 20 40 40 100 0 0 Spiramycin 200 200 100 0 0 Sulfadiazine 50 150 150 100 0 0 Sulfadimidine 300 300 100 0 0 Lincomycin \- 30 30 100 0 0 Danofloxain \- 100 100 100 0 0 Enrofloxacin \- 200 200 100 0 0 "-" means no recommended concentration . Ruggedness ---------- Results indicated that three factors of different wells in one microplate, different microplates in same batch, different batches kits had no effect on the ruggedness of the kits. However, different breeds and different analysts had some effect on the ruggedness of kits. Moreover, the CVs of different analysts for false positive rate, false negative rate, detection time, and sensitivity of kits all were less than 4% (see [Table 9](#T9){ref-type="table"}). In addition, the difference of different breeds among false positive rate, false negative rate, detection time and sensitivity of kits were shown in [Table 10](#T10){ref-type="table"}. It indicated that the kit in present study showed weaker sensitivity to different kinds of antibiotics in buffalo milk than those determined in Holstein milk with longer detection time. And the false positive and false negative rates of kits used for detecting antibiotics residue in buffalo milk were higher than 0% and less than 5% while the false positive and false negative rates in Holstein milk all were 0%. However, these performances of this kit used in buffalo milk all were up to the standard requirements of residues screening methods. ###### The CVs of different analysts for false positive rates, false negative rates, detection time and sensitivity of kits. Indexes Different analysts/% ----------------------- ---------------------- ----- Sensitivity/(μg/L) Cefquinome 3.4 Streptomycin 3.4 Tetracycline 3.0 Spiramycin 3.7 Sulfadimidine 3.6 Lincomycin 3.3 Enrofloxacin 3.5 False negative rate/% Cefquinome 3.0 Streptomycin 3.4 Tetracycline 3.2 Spiramycin 3.3 Sulfadimidine 3.6 Lincomycin 3.6 Enrofloxacin 3.5 False positive rate/% 3.8 Incubation time/h 3.6 ###### False positive and negative rates with detection time and sensitivity of kits in different breeds of milk. Indexes Different breeds ----------------------- ------------------ ------------------- ----- **Buffalo milk** **Holstein milk** Sensitivity/(μg/L) Cefquinome 45 40 Streptomycin 220 200 Tetracycline 320 300 Spiramycin 220 200 Sulfadimidine 320 300 Lincomycin 150 120 Enrofloxacin 430 400 False negative rate/% Cefquinome 3 0 Streptomycin 3 0 Tetracycline 4 0 Spiramycin 3 0 Sulfadimidine 4 0 Lincomycin 3 0 Enrofloxacin 4 0 False positive rate/% 4 0 Incubation time/h 3.4 3.0 Stability --------- Results showed that the appearance, smell, detection time, detection capability of this kit had no change over 6 months at 4°C. It indicated that the quality guarantee period of the kit is over 6 months. Confirmation and Quantification of Incurred Samples by LC/MS-MS --------------------------------------------------------------- The results of confirmation and quantification of incurred samples by LC/MS-MS was shown in [Table 11](#T11){ref-type="table"}. It indicated that the samples detected negative with this kit contained antimicrobial drugs residues such as ERY, SDZ, ENR at concentrations lower than LODs of this kit after the LC/MS-MS confirmation. Because LC/MS-MS with a sample pre-treatment of solvent extraction was more sensitive to all kinds of antibiotics than the kit in present study. Additionally, there was no false positive result of the kit. The positive samples, which were confirmed by LC-MS/MS, contained antibiotics residues at concentrations higher than or equal to LODs of this kit. Therefore, the kit in present study was reliable to screen antibiotics residues in incurred samples. ###### Results of confirmation of incurred tissues by LC/MS-MS. Antimicrobial agents Sample numbers This kit LC/MS-MS/ (μg/L) MRL/ (μg/L) ---------------------- ---------------- ---------- ------------------ ------------- Penicillin G 15 N(13) / 4 P(2) 10 Streptomycin 15 N(3) / 200 P(3) 200 P(2) 205 P(3) 212 P(4) 220 Tetracycline 15 N(10) / 100 P(3) 320 P(2) 350 Erythromycin 15 N(5) / 40 N(3) 30 P(4) 46 P(3) 52 Sulfadiazine 15 N(2) / 100 N(7) 110 P(4) 168 P(2) 200 Lincomycin 15 N(10) / 150 P(2) 170 P(3) 187 Enrofloxacin 15 N(3) / 100 N(5) 200 P(4) 400 P(3) 450 "N" means negative results. "P" means positive results. Numbers in brackets means numbers of negative or positive results. "/" means not detected . Discussion ========== Detection Capability -------------------- In past years, several microbiological inhibition methods were developed to detect antibiotics in milk. The detection capabilities of different microbiological inhibition methods in terms of test tubes in milk were shown in [Table 12](#T12){ref-type="table"}. It indicated that the kit in present study was sensitive to β-lactam as previous studies determined. Moreover, the kit was more sensitive to aminoglycosides and macrolides than BRT AIM ([@B13]), Copan milk test ([@B11]), Delvotest SP-NT ([@B1]), Eclipse 100 ([@B2]), Charm Blue Yellow ([@B12]), and Premi^®^Test ([@B19]) at MRL levels. Furthermore, several commercial kits such as BRT AIM ([@B13]), Copan milk test ([@B11]), Delvotest SP-NT ([@B1]), Eclipse 100 ([@B2]), Charm Blue Yellow ([@B12]), and Premi^®^Test ([@B19]) cannot detect quinolones in milk for that *G. stearothermophilus* is extremely insensitive to quinolones. However, the kit in present study was at least ten times more sensitive to quinolones than previously reported studies ([@B14]; [@B12]). And the detection capability of the kit for lincosamides was similar to determined by Delvotest SP-NT ([@B1]), Charm Blue Yellow ([@B12]). Additionally, the LODs for tetracyclines and sulfonamides were slightly higher than Copan milk test ([@B11]), Delvotest SP-NT ([@B1]), Charm Blue Yellow ([@B12]), and Premi^®^Test ([@B19]). ###### The detection capability of different microbiological inhibition methods in term of tubes in milk. Antibiotics EU/CODEX MRL in milk /(μg/L) LOD /(μg/L) --------------- ------------------------------ ------------- ------ ----------- ------- ----------- ----- ------- Penicillin G 4 2 2 3 5 2 2 \<2.5 Cefquinome 20 40 / 100 / / 40 / Neomycin 1500 50 3700 500--2000 9100 100-200 150 / Streptomycin 200 200 6000 1000 10100 300-500 / / Doxycycline 0 100 390 150 260 100 75 100 Tetracycline 100 300 6200 250-500 480 100 100 100 Erythromycin 40 40 630 \>200 750 50 150 \<100 Spiramycin 200 200 / \>2000 18100 200 500 \<125 Sulfadiazine 100 150 5400 50-100 / 50 100 50 Sulfadimidine 100 300 / 100-200 750 25 125 \<25 Lincomycin 150 120 / / / 100 150 / Danofloxain 30 100 / / / / / / Enrofloxacin 100 400 / / 4000 1000-1500 / / 1 Molina et al. (2003) ; 2 Le Breton et al. (2007) ; 3 Beltrán et al. (2015) ; 4 Althaus et al. (2003) ; 5 Linage et al. (2007) ; 6 Stead et al. (2004) . "/" means not detected . Both of chicken egg and honey are important for human health and consumed daily, however, there was few research by microbiological inhibition methods reported about chicken egg and honey. For example, Premi^®^Test is a commercially available kit and widely used for screening of antibiotics residues in milk, muscle, kidney, egg, honey and feed etc. Actually, Premi^®^Test is insensitive to CEF, aminoglycosides, macrolides, LIN and quinolones in chicken egg. However, the kit in present study can detect CEF, aminoglycosides, macrolides, LIN and quinolones in chicken egg, even the LODs for NEO, ERY, LIN were lower than or equal to MRL in chicken egg. Additionally, the LODs of the kit for PEN and DOX were less than or similar to those of Premi^®^Test. But the LODs for tetracyclines and sulfonamides were higher than those determined by Premi^®^Test ([@B19]). When it comes to honey, the LODs of this kit for β-lactam, ERY, LIN, ENR were less than or similar to determined by Premi^®^Test. Additionally, the kit was more sensitive to aminoglycosides, SPI and DAN than Premi^®^Test. However, Premi^®^Test was more sensitive to tetracyclines and sulfonamides than the kit in present study ([@B19]). When compared to previous studies, it was observed that the kit in present study was more sensitive to aminoglycosides, macrolides and quinolones in milk, chicken egg and honey. The CAP can improve the bacteriostatic activity of tetracyclines by synergistic reaction; however, higher concentration of CAP will antagonizes macrolides by competing the subunit 50s site of bacterial ribosomal. Therefore, improvement of the detection capability of the kit in present study for macrolides was operated by lowering CAP concentration in kit's medium. *G. stearothermophilus var* C953 is only sensitive to β-lactam and lincomycin ([@B10]). As a result, in this kit, TMP and CAP was used to improve the sensitivity of the kit to sulfonamides, and tetracyclines separately. At the same time, STR and ENR were used to improve the sensitivity to aminoglycosides, macrolides and quinolones based on the research that improvement of the detection capabilities to ENR by adding moderate concentration of ENR into kits ([@B18]). A small quantity of STR in the kit can improve the sensitivity of the kit to aminoglycosides and also work with macrolides by synergistic reaction. Even a small amount of STR in this kit can work with tetracyclines by the same reaction principle as tetracyclines do. It was the reason that the kit with high pH value was still sensitive to tetracyclines in antimicrobial agents's diluents shown in [Table 1](#T1){ref-type="table"}. Similarly, adding moderate ENR into this kit to improve the detection capability of this kit to quinolones. And the bacteriostatic mechanism of TMP, CAP, STR, and ENR are different, which will produce synergistic reaction, but not antagonism. At same time, the detection capability of this kit to β-lactam and lincosamides was also improved by TMP, CAP, STR, and ENR. Results showed that the LODs of this kit were less than or equal to MRL in milk for β-lactam, aminoglycosides, tetracyclines, macrolides, sulfonamides, lincosamides, however 1.8--3.4 times MRL in milk for quinolones when the kit in present study was used for screening residual antibiotics in respective antibacterial drugs diluents. However, the LODs of the kit for tetracyclines, sulfonamides and quinolones were higher in milk, chicken egg and honey than determined in respective antibacterial drugs diluents. Moreover, the detection capability of the kit for β-lactam, aminoglycosides, macrolides, lincosamides in milk, chicken egg and honey was same as determined in antimicrobial agents diluents. The reasons can be divided into two aspects: the differences among matrix and the detection capability of the kit in present study. The differences among matrix are pH and matrix components. The matrix's pH will affect the bacteriostasis effect of all kinds of antibiotics and the detection time of the kit. In addition, the chicken egg, milk and honey are weak alkaline, weak acidic and acidic matrix separately. According to results, the bacteriostasis of all kinds of antibiotics was almost same in chicken egg, milk and honey. Therefore, the pH of matrix was not the main reason. Moreover, the detection time of the kit in the four matrixes were as follows: 3 h for milk; 3.25 h for honey; 3.5 h for chicken egg; 3.75 h for antimicrobial agents diluents. It indicated that the pH of matrix affected the detection time of the kit obviously. The detection time for the matrix with higher pH was longer while the detection time for the matrix with lower pH was shorter. Additionally, compared to antimicrobial agents diluents, the milk, chicken egg and honey are rich in nutrition, which can promote the growth of bacteria in kit's medium and shorten detection time. It was also reported that dissolution of the final extract in a microbiological growth medium (i.e., Lab Lemco broth) facilitate the bacterial growth cycle and improve the results ([@B19]). Above all, the main reason maybe that the kit in present study was not enough sensitive to tetracyclines, sulfonamides and quinolones. Because improvement of the detection capability of the kit in present study for macrolides was operated by lowering CAP concentration in kit medium. Moreover, a small quantity of TMP, STR, and ENR in kit medium was adopted to avoid false positive result. Therefore, the bacteriostasis of tetracyclines, sulfonamides and quinolones were weaker with a small quantity of sensitizer such as TMP, CAP, and ENR. Then tetracyclines, sulfonamides and quinolones with sensitizer in kit separately cannot completely inhibit the growth of *G. stearothermophilus* spores in kits. Moreover, the part of the spores produced little acid, which cannot support enough acid for bromcresol purple to change color from purple to yellow under the existing nutritional condition of this kit. Thus, it was shown to be antibioitcs residues positive results of tetracyclines, sulfonamides and quinolones. However, negative results of tetracyclines, sulfonamides and quinolones were indicated when this kit was used for detecting antibiotics residues in milk, chicken egg and honey. Because milk, chicken eggs and honey are rich in nutrition, which made the part of the spores to produce enough acid for bromcresol purple to turn into yellow from purple. Therefore, in the future, further study could be conducted to optimize the kit components such as a mixture of nutrients and sensitizers, and sample pre-treatment methods on the basis of the previous research. Specificity ----------- Animal derived food contains natural bacteriostatic substances, which can inhibit the growth of microorganism in microbiological kits and result in false positive results ([@B7]; [@B3]). In this study, the method of pre-permeation at RT was used to prevent excessive natural bacteriostatic substances in animal food from permeating through the kit's medium. BRT AIM and Eclipse 100^®^ had used the similar sample pre-treatment method of pre-permeation at 4°C for 1 h ([@B13]; [@B14]). But the kit in present study did pre-permeation at RT to shorten the pre-permeation time, and thus shorten the whole operation time of the kit. After pre-permeation, the remaining matrix was poured out and then the microplates were cleaned by water, which will remove the impurities on the microplates medium surface. Finally, a small quantity of natural antimicrobial substances infiltrated into the kit during pre-permeation were denatured by water bath at proper temperature for a certain time, which can avoid the false positive results caused by natural bacteriostatic substances in animal food. The microplates having milk and chicken egg were incubated in water bath for 10 min at 80°C, however, the microplates having honey were incubated in water bath for 1 h at 45°C. High temperature can destroy natural antimicrobial substances in animal food. And the incubation temperature and time for milk and chicken egg were 80°C and 10 min separately. However, enzymes especially amylase in honey are extremely unstable to heating. Therefore, the way of incubation at 80°C for 10 min was not compliant to denature natural antimicrobial substances in honey. And the way of incubation at 45°C for 1 h for honey was decided by optimization experiment. In addition, [@B17] detected antibacterial agents in bovine kidney fluid and serum by Premi^®^Test with similar sample pre-treatment. Microbiological kits were incubated in water bath at 80°C for 10 min after adding samples into test well, which effectively inhibit natural antibacterial substances in animal food. Additionally, microbiological kits heated at proper temperature for little time will not affect the sensitivity of the method ([@B17]). Ruggedness ---------- The reproducibility of kits was determined by the experimental materials, preparation process and test operators. Thus, it deserved consideration that the ruggedness of kits in different breeds of animal food, different wells of each microplate, different microplates of same batch, different batches of microplates and different analysts. The CVs of different wells of microplate and different microplates of same batch both were 0%, which indicated that the same standard production process was adopted throughout the whole preparation process of kits. Moreover, the CVs of different batches microplates was also 0%, which revealed that the standard production process was adopted not only throughout the whole preparation process of kits, but also throughout the whole preparation process of *G. stearothermophilus var* C953 spores with the stable performances in kits. The operation results of different operators were not quite different. Because the detection operation flow of this kit was simple with no special training required except the sample procedures according to the instructions the kits. Bovine milk was used as repeatability test because there was a difference in the milk composition of buffalo milk and Holstein milk. Results showed that the false positive rate, false negative rate, detection time and sensitivity were different between buffalo milk and Holstein milk. Because buffalo milk contains more fat, protein and lactose than Holstein milk. Minerals and vitamins in buffalo milk are also dozens of times higher than that of Holstein milk. Therefore, buffalo milk caused more interference to microbiological inhibition methods from matrix than Holstein milk. Stability --------- The stability of kits is important for the transportation, preservation and usage. Results showed that the quality guarantee period of kits was more than 6 months at 4°C. The stability of kits was determined by the production process of kits and the stability of the indicator bacteria. A 150 μL of the culture medium was added into individual wells of microtiter plates using an electronic pipette in a sterile condition. Then these microplates were sealed with aluminized film and stored at 4 °C until use. The purpose of the sealing was to maintain the moisture in kits' medium and prevent the bacteria and CO~2~ in the environment from contaminating the inner medium. Additionally, *G. stearothermophilus var* C953 spores with stable properties were inoculated into kits during the production process of kits and stored in 4 °C. Moreover, the acid-producing ability of the spore and its sensitivity to antimicrobial agents remained unchanged for a long time. Therefore, the medium of this kit was more stable and the shelf life has been extended. Author Contributions ==================== QW, YW, and ZY conceived and designed the experiments. QW, DP, and QL performed the experiments. QW, MS, and AS analyzed the data. QW, ZL, YW, and ZY contributed reagents, materials, and analysis tools. QW wrote the manuscript. All authors discussed the results and commented on the manuscript. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer IM and handling editor declared their shared affiliation. **Funding.** This work was supported by 2018 National Risk Assessment of Quality and Safety of Milk (GJFP201800804) and the Key Project of the Ministry of Agriculture (2011-G5). [^1]: Edited by: Eugenia Bezirtzoglou, Democritus University of Thrace, Greece [^2]: Reviewed by: Ioanna Mantzourani, Democritus University of Thrace, Greece; Beatrix Stessl, University of Veterinary Medicine Vienna, Austria [^3]: This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Erratum to: Mol Genet Genomics DOI 10.1007/s00438-009-0432-z {#Sec1} ============================================================ The Author would like to correct their names in the Author group. Instead of abbreviated form the names should change as: Geetha Govind, Vokkaliga ThammeGowda Harshavardhan, Jayaker Kalaiarasi Patricia, Ramachandra Dhanalakshmi, Muthappa Senthil Kumar, Nese Sreenivasulu, Makarla Udayakumar. The online version of the original article can be found under doi:10.1007/s00438-009-0432-z.

Background ========== Secretory leukocyte protease inhibitor (SLPI) is a member of the chelonianin class of serine protease inhibitors, and is predominantly expressed in secretory epithelial cells of mucosal surfaces, immune cells and has been identified in various tissues \[[@B1]-[@B3]\]. Among serine proteinase inhibitors, SLPI is considered as \"alarm proteinase inhibitor\" that is upregulated during infection or inflammation to compensate for high human neutrophil elastase \[[@B1],[@B4]\]. The C-terminus of SLPI primarily inhibits human elastase, but is capable of inhibiting other serine proteinases such as tryptase and cathepsin G \[[@B5]\]. In addition to its function as an antiprotease, SLPI possesses antimicrobial activity against several bacteria and fungi \[[@B1],[@B4],[@B6]\]. Furthermore, it was shown that SLPI controls cell proliferation by regulation of growth-associated genes such as cyclin D and transforming growth factor (TGF)-β1 \[[@B7]\], modifies the activation of macrophages \[[@B4]\] and regulates the LPS-induced activation of the transcription factor \"nuclear factor kappa B\" (NF-κB) \[[@B8],[@B9]\]. SLPI-deficient mice provided evidence for functional involvement of SLPI in wound healing \[[@B10]\] and lipopolysaccharide (LPS)-mediated inflammation \[[@B11]\]. In context to its role as \"alarm proteinase inhibitor\", SLPI was found to be differentially regulated in inflammatory diseases and cancer. Increased expression or elevated serum levels of SLPI were reported in human sepsis and experimental endotoxemia \[[@B12]\], febrile patients \[[@B13]\], Wegners\'s granulomatosis \[[@B14]\], gastric cancer \[[@B15]\] and pulmonary infection \[[@B16]\]. In contrast, other bacterial or viral infections in lung \[[@B17]\], stomach \[[@B18]\] and cervical epithelial cells \[[@B19]\] were found to be associated with decreased SLPI levels. The underlying mechanisms responsible for the different regulation of SLPI have not been identified, but most likely both microbial and host factors contribute to the up- or downregulation of SLPI in the various diseases. Notably, the reduction of SLPI levels correlated inversely with the severity of inflammation (infiltration of granulocytes) and neutrophil elastase activity in the gastric mucosa of *H. pylori-*infected individuals \[[@B20],[@B21]\] and the bronchoalveolar lavage fluid (BALF) of *Pseudomonas*-infected subjects \[[@B17]\]. Progranulin, also known as acrogranin, proepithelin and PC cell derived-growth factor, is a 68 kDa glycoprotein secreted by many epithelial and immune cells \[[@B22]\]. The full-length protein is subsequently modified by limited proteolysis leading to the generation of 6-25 kDa fragments called granulins \[[@B23]\]. Pathophysiologically, Progranulin has drawn a lot of attention in the last years since it has been identified that mutations of the corresponding *granulin*gene are causally linked to the development of frontotemporal dementia \[[@B24]\]. Individuals with these mutations exhibit tau-negative, but ubiquitin-positive, inclusions in their brain that eventually cause frontotemporal dementia. Both the precursor (Progranulin) and the degraded forms (Granulins) mediate different cellular effects in a variety of pathophysiological conditions such as inflammation, proliferation, carcinogenesis and wound healing \[[@B25]\]. While Progranulin acts as growth factor for epithelial cells, fibroblasts and neurons and has anti-inflammatory properties \[[@B26],[@B27]\], granulins drive inflammation leading to the infiltration of immune cells and induced cytokine expression \[[@B28],[@B29]\]. The conversion of Progranulin to granulins, which is the critical step in the regulation of the balance between both molecular forms, is controlled by SLPI that binds Progranulin and prevents degradation by elastase \[[@B23]\]. The importance of this interaction for the wound healing was demonstrated at the SLPI-deficient mice \[[@B10]\]. The lack of SLPI resulted in higher serine protease-derived activities that were associated with impaired wound healing in these animals \[[@B10]\]. The delayed wound healing was normalized after the addition of Progranulin providing evidence for the importance of the interaction between Progranulin and SLPI. We recently identified a marked down-regulation of mucosal SLPI levels in *H. pylori*-infected subjects \[[@B18]\]. The role of SLPI for the balance between Progranulin and granulins and the high prevalence of mucosal injuries (ulcer, erosions) in *H. pylori*-infected subjects, prompted us to study the expression levels of Progranulin in context to that of SLPI in relation to *H. pylori*status. Considering the role of SLPI for regulating the activity of elastase, we hypothesized that the *H. pylori*-induced reduction of SLPI would lead to a reduction of mucosal Progranulin levels, since the higher elastase activities in the mucosa of *H. pylori*-infected subjects would degrade the molecule into the granulin fragments. In addition, gastric epithelial cells (either infected with *H. pylori*± transfection of SLPI-siRNA) were used as *in vitro*model to prove the proposed hypothesis. Methods ======= Study design and H. pylori status --------------------------------- The study protocol was conducted according to the declaration of Helsinki and approved by the ethics\' committee of the Otto-von-Guericke University (No. of ethical vote: 143/99) as well as government authorities; all participants signed informed consent before entering the study. Details of the protocol (inclusion, exclusion criteria, and demographic data) were described previously \[[@B20]\]. The initial protocol was aimed at studying the interaction of *H. pylori*infection and low-dose aspirin. Briefly, human healthy volunteers were stratified according to the *H. pylori*status leading to the *H. pylori*-positive (*H. pylori*^*+*^, n = 10) and -negative (*H. pylori*^-^, n = 10) group. After successful eradication therapy, 9 out of 10 *H. pylori*-infected individuals agreed to participate in this study after 3 months again composing the *H. pylori*-eradicated (*H. pylori*^*erad*^) group. In order to investigate the potential interaction between Progranulin and SLPI, mucosal and serum levels as well as gene expression levels of Progranulin were analyzed retrospectively in existing samples and tissue specimens in relation to *H. pylori*status and SLPI expression levels published previously \[[@B20]\]. The analysis of Progranulin expression was performed in the \"pre-treatment\" samples, which correspond to day 0 \[[@B20]\] before \"low-dose\" aspirin was taken by the individuals. The study includes a correlation analysis of mucosal Progranulin levels with those of SLPI studied in the same cohorts previously; details concerning the analysis of SLPI were reported recently \[[@B20]\]. Determination of Progranulin expression quantitative RT-PCR and ELISA --------------------------------------------------------------------- Progranulin levels were quantified in the total protein extract from mucosal biopsies at sera stored at -80°C in previous study \[[@B22]\]. Progranulin levels were quantified using the Progranulin ELISA kit (Axxora GmbH, Lörrach, Germany; No: AG-45A-0018PP-KI01) as described by the manufacturer. Protein levels were normalized to ng/μg total protein content of extracted mucosal samples or ng/ml for sera. Corresponding Progranulin m-RNA levels were determined by quantitative RT-PCR using existing cDNA samples stored at -80°C. Quantitative RT-PCR was performed using an iCycler (BioRad, Munich, Germany) and HotStarTaq Master Mix™ (Qiagen, Hilden, Germany) as described \[[@B23]\]. Initial activation of Taq-polymerase at 95°C for 15 min was followed by 40 cycles with denaturation at 94°C for 30 s, annealing at 60°C for 30 s and elongation at 72°C for 30 s. The fluorescence intensity of the double-strand specific SYBR-Green I, reflecting the amount of actually formed PCR-product, was read real-time at the end of each elongation step. Then specific initial template mRNA amounts were calculated by determining the time point at which the linear increase of sample PCR product started, relative to the corresponding points of a standard curve; these are given as arbitrary units (a.u.). Both PCR products were cloned into the pDIRECT™ (Qiagen, Hilden, Germany), and subsequent dilutions of the corresponding plasmid DNA were used to create a standard curve for the RT-PCR. The correlation coefficients of both Progranulin and β-actin standards were \> 0.95. β-actin mRNA amounts were used to normalize the cDNA contents of the different samples. Final data reflect the ratio in a.u. between Progranulin transcript and β-actin transcript levels. The following primers were used for the RT-PCR analysis: ß-actin (fw:5\'-GCC-ATC-CTG-CGT-CTG-GAC-C-3\' rev: 5\'-ACA-TGG-TGG-TGC-CGC-CAG-ACA-G-3\'; 400 bp), and Progranulin (fw:5\`-ATG-GCC-CAC-AAC-ACT-GAG-CAG-G-3\`, rev: 5\`-TCT-GGG-CAG-GGA-GCT-TCT-TTG-C-3\`, 440 bp). Both cDNA fragments included intron-spanning regions resulting in the generation of a larger PCR product from genomic DNA or its exclusion. Therefore, all identified PCR products can exclusively be attributed to the mRNA pool of the sample. Immunohistochemical analysis of Progranulin expression in the gastric mucosa ---------------------------------------------------------------------------- To study the cellular origin of Progranulin expression in antral and corpus mucosa, tissue specimens from all 29 individuals were subjected to immunohistochemical analysis. The pathologist (D.K.) was blinded to the group assignment of samples. Paraffin-embedded biopsy specimens were cut into 3 μm thick sections, mounted on glass slides, and treated with Xylol and dehydrated by standard protocols. For antigen retrieval, specimens were boiled three times in 0.01 M sodium citrate puffer (pH 6.0) for 10 min in a microwave (600W). Incubation with primary polyclonal goat-derived anti-Progranulin antibody (BAF2420, dilution 1:1.000, R&D Systems, Minneapolis, MN, USA) was conducted at 37°C for 35 min and followed by PBS-washing. Positive immunohistochemical reactions were revealed using the iVIEWTM DAB Detection Kit (Ventana, Germany) as chromogen substrate. Finally, the samples were counter-stained with hematoxilin, dehydrated and mounted using DEPEX (Serva, Heidelberg, Germany). For positive control normal prostate tissue was used. For negative control corresponding stainings were performed using unrelated goat antiserum that did not lead to a specific staining (data not shown). Expression of Progranulin was scored for the epithelium of the mucosal surface and gastric glands of the antrum and corpus in 3 representative high power fields (Zeiss Axioskop 50). Staining intensity (SI) and the percentage of positive cells (PP) were assessed using the following semiquantitative score: SI was classified in 0 (no staining), 1 (weak), 2 (moderate) and 3 (strong); PP: 0 (no positive cells), 1 (\< 10%), 2 (10-50%), 3 (51 - 80%), 4 (\> 80%). For each slide the immunoreactive score (IRS) was calculated as \[SI x PP\] with a possible maximum score of 12. Immunohistochemical expression of Progranulin was separately scored for surface epithelium and glands, and then these scores were summarized as \"total score\" that were statistically analyzed among the three groups. The maximum score for epithelial expression of Progranulin was \"24\". Since all type of immune cells showed constantly strong expression of Progranulin, only the number of these infiltrating cells was semiquantitatively assessed. Progranulin-immunoreactive immune cells were evaluated for their quantity in the lamina propria (1 = few, 2 = moderate, 3 = abundant). Therefore, the maximum score of immune cell-related expression of Progranulin was \"3\". Cell Culture and *in vitro*studies ---------------------------------- AGS (CRL-1739) gastric cancer cells were purchased from American Type Culture Collection (ATCC). Cells were maintained in 25 cm^2^cell culture flasks (NUNC GmbH, Wiesbaden, Germany) in a cell incubator at 37°C and 5% CO~2~using RPMI-1640 containing 10% FCS, 100 U/ml Penicillin, 100 μg/ml streptomycin and 100 μg/ml gentamycin (all reagents; PAA, Colbe, Germany). Infection studies were performed using wildtype *H. pylori*strain purchased from ATCC (No. 43504). *H. pylori*was cultivated on selective agar plates (bioMerieux, Marcy I\'Etoile, France) under microaerophilic conditions at 37°C for 2 days, and then resuspended in PBS (pH 7.4). Bacterial suspensions were adjusted based on optical density at 535 nm (OD = 1 corresponds to 1 × 10^9^bacteria). To ensure functional active bacteria, suspensions were microscopically inspected for shape and motility. After washing cells twice with medium without FCS and antibiotics, cells were infected with *H. pylori*at a \"multiplicity of infection\" of 50 in medium lacking antibiotics for 24 h. For siRNA transfection, 4 × 10^5^cells were seeded in complete medium in 6-well plates and cultivated for 24 h. Cells were transfected with either SLPI-siRNA\#1 (No: S100726383) or All-Stars™ negative siRNA control at a final concentration of 3 nM using HiPerfect™ transfect reagent as described by the manufacturer (all reagents, siRNA from Qiagen). Cells were cultivated in the presence of siRNA for another 48 hours at standard conditions, and then infected with *H. pylori*as described above. After completing transfection and/or infection experiments, 0.8 ml of the cell culture medium was collected, centrifuged at 8.000 × g, and the supernatant stored in aliquots at -80°C for analysis. AGS cells were washed three-times with PBS (pH 7.4), and then harvested by PBS (pH 7.4) using a cell-scraper. Cells were washed once (8.000 × g, 4°C, 15 min) and resuspended in 1 ml PBS (pH 7.4). The sample was aliquoted (2 × 500 μl) into two Eppendorf tubes™ (Eppendorf AG, Hamburg, Germany), cells were obtained by centrifugation and the resulting pellets were stored at -80°C until analysis. Three individual experiments (each as duplicate) were performed for all experiments settings. Statistical Analysis -------------------- All data were entered into a database using the Microcal Origin™ 8.0G program package (Northhampton, MA, USA). Data are expressed as raw, median, mean ± standard deviations error (SD), or 95% CI (confidence intervals), if not stated otherwise. Non-parametric Kruskal-Wallis test and Mann-Whitney U test were applied for multiple and pairwise comparisons between groups, respectively. Immunohistochemical data were analyzed by One-way ANOVA (as global test for multiple testing) and LSD as post hoc analysis for pairwise comparisons if global test reached significant level. Correlation analysis was performed by Pearson test. All test were applied two-sided with a level of significance of P \< 0.05. Results ======= Expression of Progranulin in gastric mucosa in relation to *H. pylori*status and SLPI levels -------------------------------------------------------------------------------------------- Progranulin gene expression and corresponding protein levels were identified in all mucosal samples from antrum and corpus as well as serum levels. As shown in figure [1](#F1){ref-type="fig"}, protein levels demonstrated normal distribution, while gene expression levels revealed skewed distribution. Therefore, we decided to apply nonparametric tests for both methodologies. {#F1} *H. pylori*-infected subjects had about 2-fold higher Progranulin protein levels (median: 0.43, range: 0.33-0.63 ng/μg protein) compared to levels after the successful eradication (median: 0.25, range: 0.25-0.46 ng/μg protein) or the unrelated *H. pylori-*negative group (median: 0.34, range: 0.27-0.46 ng/μg protein; p \< 0.05) (Figure [1](#F1){ref-type="fig"}, upper panel). Progranulin protein levels in corpus mucosa (medians: 0.19 - 0.26 ng/μg) and serum samples (medians: 173 - 206 ng/ml) did not differ among the three groups (Figure [1](#F1){ref-type="fig"}, upper panel). Progranulin-mRNA amounts differed significantly in antrum among the three groups. As illustrated in figure [1](#F1){ref-type="fig"} (lower panel), *H. pylori*-negative subjects revealed highest transcript amounts (median: 0.38, range: 0.2-1.3 a.u.), followed by the *H. pylori*-positive subjects (median: 0.17, range: 0.02-0.58 a.u.), and were lowest after eradication (median: 0.06, range: 0.03-0.07 a.u.). Similar results were obtained for corpus mucosa without reaching significance (Figure [1](#F1){ref-type="fig"}, lower panel). To investigate a potential association between mucosal Progranulin and SLPI levels, correlation analysis was performed between both parameters. Please note that data concerning SLPI expression in these cohorts were published previously; therefore these data are not shown in detail in this study \[[@B20]\]. As illustrated in figure [2](#F2){ref-type="fig"}, a significant positive correlation was identified in eradicated subjects, whereas no correlation was seen in both other groups as well as in the combined data set. No correlations between Progranulin and SLPI were identified in corpus mucosa and serum of the three individual groups (data not shown). {#F2} Immunohistochemical localization of Progranulin in the gastric mucosa --------------------------------------------------------------------- As illustrated in figure [3](#F3){ref-type="fig"}, both epithelial and infiltrating immune cells contribute to the mucosal Progranulin expression. Immune cells (granulocytes, lymphocytes) showed constantly high expression of Progranulin except cells of lymphoid follicles. Higher numbers of Progranulin-expressing cells were associated with gastritis in *H. pylori*-infected subjects (Figure [3A+D](#F3){ref-type="fig"}). For the epithelium, strongest expression was observed in the gastric glands followed by the basis of the foveolae mainly in areas of dense inflammatory infiltrate. Surface epithelium between gastric pits showed weak or no expression of Progranulin. Semiquantitative scoring revealed significant higher expression scores of Progranulin for *H. pylori*-infected subjects compared to both other groups in antrum (Figure [4](#F4){ref-type="fig"}, left panel), whereas a tendency was observed for corpus (Figure [4](#F4){ref-type="fig"}, right panel). Furthermore, the number of infiltrating Progranulin-expressing immune cells was significantly higher in both antral and corpus mucosa of *H. pylori*-infected subjects (Figure [4](#F4){ref-type="fig"}). {#F3} {#F4} Expression of Progranulin and SLPI in epithelial AGS cells infected by *H. pylori* ---------------------------------------------------------------------------------- To investigate the regulatory link between SLPI and Progranulin, both molecules were investigated in relation to *H. pylori*infection and siRNA-mediated downregulation of SLPI expression in AGS cells. As demonstrated in figure [5](#F5){ref-type="fig"}, cellular SLPI levels were significantly reduced by 33%, 63%, and 81.3% by *H. pylori*, siRNA, and both factors, respectively. SLPI levels in the supernatant were strongly reduced (-65%) by siRNA, but not by *H. pylori*(Figure [5](#F5){ref-type="fig"}). The analysis of Progranulin levels in the identical samples, revealed no effect of SLPIsiRNA treatment. Both cellular (94.7 ± 9.4%) as well as secreted (109.4 ± 3.3%) Progranulin levels were similar to those of controls. *H. pylori*-infection was associated with elevated Progranulin level in supernatant (353 ± 109%), while cellular levels were found to be slightly reduced (70 ± 5.9%, P \< 0.05). The combined effect of *H. pylori*and SLPI-siRNA approach resulted in similar changes (331 ± 97% and 61.3 ± 8.9% of Progranulin levels in supernatant and lysate, respectively (P \< 0.05, Figure [5](#F5){ref-type="fig"}). {#F5} Discussion ========== Here we demonstrate that (I) the *H. pylori*infection is associated with increased Progranulin levels in the antrum of infected subjects, and (II) that both epithelial and infiltrating immune cells contribute to this phenomenon. Furthermore, we provided evidence that (III) the upregulation of Progranulin seems to be independent of SLPI levels. Considering the central role of the elastase/SLPI equilibrium for the conversion of Progranulin to granulins \[[@B10]\] and the previously identified deregulation of elastase/SLPI expression in *H. pylori*-induced gastritis \[[@B21]\], we anticipated a negative correlation between SLPI and Progranulin for this disease. The *H. pylori*-induced reduction of mucosal SLPI levels resulted in higher elastase activities that were expected to degrade Progranulin leading subsequently to diminished mucosal Progranulin levels. In contrast to our working hypothesis (we expected a negative correlation between mucosal SLPI and Progranulin levels), we identified an increase of mucosal Progranulin levels in the antrum of *H. pylori-*infected subjects. Furthermore, correlation analyses revealed rather a trend or even a positive correlation between both proteins implying that the proposed regulatory link between SLPI and Progranulin is not present in this disease. The fact that increased Progranulin levels were mostly restricted to antral mucosa (except immunohistochemical score of corpus glands) suggests an association of this upregulation with the degree of gastritis. As previously demonstrated, all probands presented antrum-predominant gastritis that was associated with moderate and severe activity scores reflecting the number of infiltrating granulocytes and lymphocytes \[[@B20]\]. As shown in immunohistochemical stainings of the study, immune cells were strongly positive for Progranulin and represent a major source of mucosal Progranulin levels in addition to gastric epithelial cells. Collectively, data of immunohistochemistry correspond to quantitative assessment of Progranulin by ELISA supporting the identified upregulation of Progranulin in *H. pylori*-infection. Interestingly, *H. pylori*-negative subjects revealed significant higher *progranulin*transcript levels, which were associated with lower protein levels, compared to those of the *H. pylori*-positive and -eradicated group. The missing concordance between transcriptional and protein level is not easily explained and remains unclear. One potential explanation might be different regulatory mechanisms of Progranulin expression in gastric epithelial cells of *H. pylori*-negative subjects, who have been negative for the complete life compared to individuals after successful eradication therapy being without *H. pylori*-infection for several months only. As shown recently for mucosal infiltration and by the numbers of Progranulin-expressing immune cells in this study, samples from patients after eradication therapy contained still lymphocytes leading to slightly higher chronicity scores \[[@B20]\] or slightly increased Progranulin scores compared to *H. pylori*-negative subjects. Since in *H. pylori*-positive subjects, two major Progranulin-expressing cell types (epithelial and immune cells) are simultaneously present, *Progranulin*transcript levels can not be assessed individually for each cell type. Despite the missing concordance between protein and transcript levels, it should be emphasized that the mucosal levels of Progranulin were found to be significantly upregulated in *H. pylori*-infected subjects. The results obtained in the AGS cell model do partially not correspond to the *ex vivo*findings. While *ex vivo*data demonstrated an upregulation of Progranulin by *H. pylori*, in the AGS cell model, only the concentration of Progranulin in the supernatant was strongly induced, whereas the cellular expression, analyzed in the lysate, was decreased. There are several aspects that might explain these disconcordant results. In AGS cells, both the intracellular and secreted proportion of Progranulin was separately analyzed. Since in *ex vivo*analysis, both compartments can not be differentiated, the increased Progranulin levels in antral mucosa might reflect both increased secretion and changes in epithelial Progranulin expression. Second, *ex vivo*analysis is performed on complex samples including epithelial and immune cells, whereas the *in vitro*model only mirrors the direct interaction of *H. pylori*to epithelial-derived AGS cells. Third, analyzing the Progranulin expression after 24 hours represents the effects of an acute infection, whereas changes in mucosal biopsies can be considered as long-term effects of an chronic infection that are in a \"steady-state\". Despite these limitations, data from the *in vitro*model allow the conclusion that a down-regulation of epithelial SLPI expression (either by *H. pylori*or siRNA) does not affect the expression of Progranulin in AGS cells. Owing to the low molecular weight of granulins, no method is currently suitable to analyze quantitatively the levels of the Progranulin-derived degradation products. Therefore, no statement can be made concerning the equilibrium between Progranulin and granulins in gastric mucosa that might hypothetically be shifted towards granulins even the Progranulin levels are upregulated. Furthermore, it is of note that SLPI is not the only serine protease inhibitor expressed in the gastric mucosa. Recently, we identified elevated alpha-1 protease inhibitor (A1-PI) levels in the mucosa of *H. pylori*-infected individuals \[[@B30]\]. Since A1-PI can inhibit elastase to a similar extent as SLPI \[[@B7]\], a compensatory mechanism is another potential explanation, while Progranulin is elevated, although SLPI levels are strongly diminished in relation to *H. pylori*infection. The observed association of induced Progranulin levels in context to *H. pylori*infection and its associated gastritis does not allow functional conclusions whether the upregulation has an active regulatory role for the inflammatory process, or it merely reflects the inflammatory conditions of the underlying gastritis. Keeping in mind that Progranulin acts as epithelial growth factor in other diseases \[[@B28],[@B29]\], it is tempting to speculate that the upregulation of Progranulin in *H. pylori-*associated gastritis might be involved in mucosal healing of gastric erosions/ulcers induced by this infection. But at this moment, this remains purely speculative since no functional data are available. Conclusions =========== Taken together data from *in vitro*and *ex vivo*analysis, we can conclude that the proposed regulatory link between SLPI and Progranulin expression seems to be of no or low relevance in context to the *H. pylori*infection. Furthermore, we provide evidence that Progranulin is another molecule the expression of which is upregulated in relation to this infection. Competing interests =================== The authors declare that none of them has financial interests in context to this study. This work was supported by the Deutsche Forschungsgemeinschaft (WE-2170/8-1). Authors\' contributions ======================= TW was involved in the conception and design of the study, analyzing and interpreting data, writing the manuscript and revision of the final version. DK performed immunohistochemical stainings, corresponding semiquantitative analysis and participated in writing the draft. CS and DS performed *in vitro*studies on AGS cells and the corresponding assessment and analysis of Progranulin expression in these samples. GT enrolled the patients groups, performed endoscopic evaluation including sampling biopsies, and contributed in writing the draft. PM was involved in the conception and design of the study, and revised the manuscript for important intellectual content. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/11/63/prepub> Acknowledgements ================ We thank the endoscopy team for their technical assistance, Ursula Stolz, Simone Philipsen (Clinic of Gastroenterology), N. Wiest and C. Kügler (Department of Pathology) for their work.

In the 1950s, Barbara McClintock first discovered transposable elements (TEs) by analyzing genetic stocks of corn that were phenotypically unstable. Her discovery implied that a genetic control exerted by genomes was generally used to regulate TE mobilization. Any loss or decrease of this control would consequently result in severe genetic instabilities due to mobilization of TEs. Just such a genetic instability affecting the genome of *Drosophila melanogaster* under the control of a locus called *flamenco* (*flam*) was first reported in 1983. Focused on *flam*, this review retraces the numerous studies that have been performed from its discovery to the understanding of its ability to survey TEs. A SINGLE GENOMIC MUTATION IS RESPONSIBLE FOR *Gypsy* ACTIVITY, A RETROELEMENT FROM *Drosophila melanogaster* ============================================================================================================ In the 1980s, [@B8] were studying the dominant *ovoD* mutation in *D. melanogaster* . The *Drosophila ovo* gene, which encodes a putative transcription factor (Ovo) with TFIIIA-like zinc fingers, is required for female germline survival and proper oogenesis. The gain of function *ovoD* allele results from an extension of the N-terminal region which gives rise to a neomorphic protein that causes female sterility ([@B25]). [@B8] performed crosses between *OvoD* males and females from a stock of flies from the lab of Madeleine Gans (MG) carrying a *y v f mal* X-chromosome . In the progeny, reversions of the *ovoD* mutation generating recessive *ovo* alleles were frequently observed which allowed fertile daughters to be recovered. Surprisingly, these reversions were also associated with the appearance of mutations in other loci, which could potentially be explained if such crosses were accompanied by the *de novo* mobilization of TEs. [@B26] found that, indeed, a high frequency of *gypsy* insertions was observed in the progeny of this cross and that a hot spot for *gypsy* exists into the *ovo* locus . Insertions of *gypsy* into the *ovo* locus interfere with the coding sequence of the neomorphic allele resulting in a null allele of the gene. Novel gypsy insertions can thus be assayed by the presence of fertile daughters. The gypsy mobilization could then explain both the genetic instability observed in these crosses and the *ovoD* reversion. Also, [@B16] reported a mutator strain (MS) of *D. melanogaster* characterized by an elevated frequency of spontaneous mutations in the germ line up to 10^-3^ - 10^-4^. Mutations were recovered in both sexes and displayed the characteristics of being unstable with frequent reversion to wild type or to new mutant states. When analyzing the localization of a battery of TE families, they found that the genomic distribution of *P*, *mdg1*, *412* (*mdg2*), *mdg3*, and *copia* did not vary among the individuals of this strain. However, this was not the case for *gypsy* (*mdg4*) whose frequency of transposition was high and copy number greatly increased to 30--40 copies. These initial studies identified different mutator lines in which the frequency of *gypsy* insertions is high while several other TE families remain stable ([@B26]; [@B17]; [@B23]). Further work ultimately showed that these *gypsy* instabilities within MS strains resulted from the combination of two factors: the presence of transpositionally active *gypsy* copies, and mutation(s) of loci regulating their transposition ([@B18]). These early studies provided an incredible powerful tool to evaluate *gypsy* activity by assessing the occurrence of fertile females resulting from *ovoD* reversion to a null allele. With the *ovoD* fertility test, one could isolate rare events without having to deal with enormous amount of progeny to score. Interestingly, these tools were created even before the understanding of the mechanism of repression. A β-HETEROCHROMATIC LOCUS CONTROLS SEVERAL RETROELEMENTS: *gypsy*, *ZAM*, AND *Idefix*: ======================================================================================= A mutation responsible for *gypsy* mobilization was identified within the y ***v*** *f mal* chromosome of MG stocks ([@B32]). Genetic mapping localized this mutation at the basis of the X-chromosome at position 65.9 (20A1-3) close to β-heterochromatin where numerous TEs were known to accumulate ([@B44]). The locus was called *flamenco* (*flam*) because it had the ability to make *gypsy* "dance." Non-permissive or permissive alleles of *flam* were defined according to their ability to restrict or allow *gypsy* mobilization, respectively. A fine-scale analysis of *flam* genetic characteristics uncovered that: (i) Its control on *gypsy* activity occurs under a strict maternal effect since transposition is only allowed in the progeny (male and females) of homozygous permissive females even if fathers are non-permissive. (ii) The mutant allele present in the MS strains is essentially recessive. (iii) Transposition is largely a premeiotic event. (iv) Although *ovoD* reversion is primarily controlled by *flam*, it is influenced by other factors such as age and temperature, reversion being higher in young flies grown at 25°C. (v) The effects of *flam* on *gypsy* expression are restricted to the somatic follicle cells that surround the maternal germline ([@B31]). Thus, *flam* function could be viewed as the maternal transmission of some factors preventing *gypsy* transposition. In 1997, an unstable line called Rev was recovered after a PM mutagenesis performed on the line bearing the *w^IR6^* allele ([@B20]; Figure [1A](#F1){ref-type="fig"}). The *w^IR6^* allele is due to the insertion of the non-LTR retrotransposon *I-factor* into the first intron of the *white* gene. It gives an orange eye phenotype to flies ([@B19]). From the PM mutagenesis ([@B37]), a fly with a wild-type red-eye phenotype was recovered and established as a line subsequently called Rev because of the eye phenotype reversion from orange to red. It was found that the *white* locus had suffered a 8.4 kb insertion 3 kb upstream from the *white* start site of transcription (TSS; Figure [1C](#F1){ref-type="fig"}). This insertion corresponded to a novel TE from the *gypsy*-family that was previously uncharacterized and that has been named *ZAM* ([@B20]). *ZAM* did not only insert upstream of *white*. *In situ* hybridization and Southern analyses performed on the Rev genome revealed the presence of some 20 copies of *ZAM*, whereas *ZAM* was not found on the chromosomal arms of the original parental line *w^IR6^* (Figure [1B](#F1){ref-type="fig"}; [@B11]). From Rev, a series of mutations affecting eye coloration has been recovered, most of them affecting the *white* locus (Figure [1A](#F1){ref-type="fig"}). This second event of mutation resulted from the insertion of a novel *gypsy-like* transposable element designated *Idefix* that inserted 1.7 kb upstream of the TSS of the *white* gene. This second mutational event was recovered as a recurrent specific mutation in 11 independent individuals (Figures [1A,C](#F1){ref-type="fig"}; [@B11]). Genome analysis of Rev revealed that this line also suffered a recent and massive invasion of *Idefix* (Figure [1B](#F1){ref-type="fig"}). {#F1} The Rev line brought to light a new genetic model in which the activity of two TEs, *ZAM,* and *Idefix*, could be tested. Thereafter, transgenic flies were established with sensor-transgenes containing the full-length long terminal repeat (LTR) of *ZAM* or *Idefix* linked to the *LacZ* reporter gene. These transgenes provided a convenient read-out for analyzing the control exerted on these elements. Crosses designed to test the influence of the genetic background on these reporter constructs indicated that *ZAM* and *Idefix* responded to two types of controls: one restricting their expression to specific somatic cells of the ovaries and the other silencing their expression in the majority of *Drosophila* lines with only one exception reported in 2003 as being the Rev line ([@B12]). Using these tools, a mutation responsible for the high activity of *ZAM* and *Idefix* was identified in Rev. This mutation was localized at the basis of the X-chromosome close to *flam* (Figure [2](#F2){ref-type="fig"}; [@B12]). Although the mutation was genetically close to *flam*, the Rev line displayed a non-permissive allele of *flam* since *gypsy* was not active in this line and, like in non-permissive lines, only few copies of *gypsy* were detected in Rev. In addition, transgenes carrying fragments of *gypsy* fused to *LacZ* used as reporters of *flam* permissivity were repressed in Rev while *ZAM-LacZ* and *Idefix-LacZ*, reporter transgenes were activated ([@B43]; [@B12]). These findings suggested that *gypsy* regulation was genetically separable from *ZAM* and *Idefix* regulation, and that a second locus existed near *flam* that controlled the activity of *ZAM* and *Idefix*. ![**X-chromosomal deficiencies used for cytogenetic mapping of *COM*.** The chromosomal region is presented at the top. The lines below indicate the deficiencies tested. LacZ staining observed in these lines when *ZAM-LacZ* and *Idefix-LacZ* reporters were tested are indicated on the right. Data reported for *flam* by [@B32] are indicated in the third column. Figure modified from [@B12].](fgene-05-00257-g002){#F2} In , while working on the silencing of testis-expressed Stellate genes by paralogous Su(Ste) tandem repeats in *Drosophila*, [@B3] had reported that double-stranded RNA-mediated silencing might provide the basis for negative control of gene expression. They further proposed that the related surveillance system was implicated in the control of retrotransposons in the germline ([@B3], [@B2]). Around the same time, [@B45] had published that double strand RNAs (dsRNAs) of centromeric heterochromatin repeats in *Schizosaccharomyces pombe* would produce small interfering RNAs (siRNAs) triggering gene silencing and repressing their own transcription ([@B45]). They also suggested that these dsRNAs might silence other loci with homologous sequences. Therefore, we proposed a new hypothesis to account for TE regulation by the heterochromatin region at the base of the X-chromosome whereby vestiges of TEs might produce dsRNAs required for the silencing of *ZAM* and *Idefix* ([@B12]). To illustrate its potential to control over multiple TE families, we referred to this locus as a center required for TE mobilization and proposed to call it COM (Center Organisant la Mobilization; [@B12]). [@B40] reported an additional finding confirming this primary model. Their study demonstrated that *gypsy* did not contain a single binding region for a putative *flam* repressor ([@B40]). They first tested whether the *gypsy* promoter is dispensable for this regulation and swapped it for an alternative promoter from the *yp3* gene expressed in the follicle cells of the ovaries where *gypsy* itself is expressed. They found that a small 59 nucleotide fragment of non-promoter transcribed sequences was sufficient to make a non-*gypsy*-driven transcript sensitive to this regulation. They, then, tested diverse fragments between base 329 and 1072 from the *gypsy* promoter in the same way. They found that any fragment from the *gypsy* 5′-untranslated region (UTR) appeared to be able to target the repression, the only requirement being that *gypsy* sequences were present within the tested transcript. In addition, *gypsy* repression was impeded by *piwi* mutations. Short RNAs from 25 to 27 nucleotides long were also detected. These small RNAs, homologous to sequences within the *gypsy* 5′ UTR, should be able to guide RNA silencing complexes to *gypsy*-containing transcripts. In line with growing body of evidence implicating RNA silencing mechanisms in regulating TE activity, these data supported that *flam* could possibly act through a RNA-dependent mechanism. *flam*: FROM MOLECULAR STRUCTURE TO GENOMIC FUNCTION ==================================================== Cloning of the heterochromatic locus where *flam* and *COM* had been identified proved to be very difficult. Uncertainty in the assembly of repetitive DNA in the early releases of the *D. melanogaster* genome sequence posed difficulties for heterochromatin studies. As a consequence, *flam* localized to a sequencing gap in the Release 1 genome sequence ([@B1]; [@B29]). The group of Alain Pélisson and Alain Bucheton worked very hard in tackling this locus, whose location close to heterochromatin makes its analysis extremely difficult because it is almost impossible to perform meiotic recombination. Furthermore, the repetitive nature of *flam* added to the lack of a discrete transcript produced from the locus prevented the choice of a probe that could have been used to probe cDNA libraries. A helpful tool was provided when N. Prud'Homme generated a P-element-induced mutation *P\[lyB\]* of *flam*. Indeed, ∼100 kB of the genomic DNA flanking the insertion could be analyzed ([@B36]). [@B36] searched for unique sequences that might account for the activity of a gene and identified four of them with transcription units. The closest gene from the P-element insertion, *DIP1*, was assumed to be the best candidate for *flam*, notably because of its double stranded RNA-binding domains. However, all attempts to correlate its function to *gypsy* regulation proved to be unsuccessful ([@B36]). [@B36] further detected some deficiencies permissive for *gypsy* mobilization located \>130 kb away from the P-element insertion, suggesting that sequences responsible for the *flam* function lie large distances away from each other. This lab generated two new alleles of *flam* called *flam* KGP and *flam* BGP. By contrast to the *COM* mutation present in the Rev line, these new alleles brought evidence that certain *flam* mutations have the potential to relieve repression exerted not only on *gypsy* but also on *ZAM*. This study further showed that beyond its function on TE control, *flam* was required somatically for morphogenesis of the follicular epithelium, the tissue where *ZAM*, *Idefix,* and *gypsy* were repressed ([@B20]; [@B43]; [@B27]). These findings indicated that *flam* and *COM* were not always separable, and were in fact a single genomic locus (that will now be referred as *flam*) displaying flexibility in its potential to repress different TE families. A detailed sequence of the TE content in the *flam* region became possible due to improved genome sequence data ([@B9]) and the development of high-resolution TE annotation pipelines ([@B34]; [@B6]). *flam* revealed to be one of the specific regions of the genome with an extremely high local TE density containing 104 different TE insertions from 42 different TE families spanning at \>200 kb of sequence. However, because the high TE density region in the *flam* locus contained a gap in the assembly, the full structure of this locus and its TE content could not be fully determined. Nevertheless, since clear hallmarks of recurrent transposition were detected, inherent mobility of TEs was proposed to explain the high density of TEs in the *flam* region. However, a relatively high incidence of duplicated TE sequences was also identified, suggesting that segmental duplications have played a role in the genesis of the *flam* region. In line with the earlier models, the analysis of global nesting relationships among different TE families led [@B6] to propose that expression of chimeric sequences from regions of high TE density in the β-heterochromatin may simultaneously co-suppress transcripts from multiple euchromatic TE families . A significant breakthrough for *flam* function was achieved in 2007 when [@B7] reported for the first time the existence of discrete small RNA-generating loci that included *flam*. These data were obtained when [@B7] analyzed the control of TEs and its relationship with the Argonaute proteins in *Drosophila*. Three Argonaute proteins, the PIWI proteins Piwi, Aub, and Ago3 had been shown to bind small RNAs ([@B22]). Their mutation was known to affect TE control. Sequencing small RNAs bound by each of these three PIWI proteins from *Drosophila* ovaries, [@B7] found that the majority of the so-called piRNAs were derived from discrete genomic loci including *flam* that were subsequently referred to as piRNA clusters. Among piRNA clusters, *flam* displayed some unique characteristics. First, 94% of its uniquely mapping RNAs were Piwi partners. Second, *flam* produced piRNAs with a marked strand asymmetry that correlated with the strong biased orientation of TEs in the locus. Third, *flam* displayed the potential to produce a high fraction of repressive piRNAs targeting *ZAM*, *Idefix,* and *gypsy* (79, 30, and 33% respectively). The use of *flam* mutations, P(KG00476) and P(BG02658), indicated that a substantial reduction in piRNAs that uniquely map to *flam* was observed in mutant flies whereas piRNAs derived from other piRNA clusters were unaffected. This reduction of *flam* piRNAs was accompanied by a loss of *flam* transcripts and a high increase of the *gypsy* retroelement transcription. From this piRNA sequencing, [@B7] proposed that in ovaries, a pool of primary piRNAs is processed from long single-stranded transcripts encoded by piRNA clusters. These primary piRNAs target sense-transcripts encoded by TEs thereby triggering their degradation. An amplification system starting once the sense transcript has been detected by the primary piRNAs results in production of secondary piRNAs. In their turn, these secondary sense-piRNAs enhance cleavage of anti-sense precursors resulting in amplification of piRNA production. This model has been called the ping-pong model. Although a big step in the understanding of piRNA origin had been made, the model needed to be refined to take into account that piRNAs had been extracted from a mixture of somatic and germ line cellular lineages. *ZAM*, *Idefix,* and *gypsy* had indeed been shown to be active and consequently repressed by *flam* only in the somatic follicle cells ([@B31]; [@B21]; [@B43]). In their study, [@B7] noticed that the amplification cycle detected in ovaries might not operate in somatic follicle cells where Aub and Ago3 were absent. They suggested that, since the vast majority of transposon fragments within *flam* exists in a common orientation, this could lead to the production of anti-sense primary piRNAs processed from a long, unidirectional, precursor transcript. Subsequently, [@B24] sought to determine whether the ping-pong model applied or not in both ovarian germ and somatic follicle cells. By comparing piRNAs from germline and from their somatic support cells, they found distinct piRNA pathways with differing components ([@B24]). A simplified piRNA pathway operates in the somatic lineage in which among the three Argonaute proteins, only Piwi functions. Only primary piRNAs that lack the ping-pong amplification cycle are expressed in these cells ([@B15]). From these studies, it emerged that *flam* was not a classically defined gene producing messenger RNAs with large open reading frames able to encode proteins. By contrast, it had the potential to produce long, unidirectional, non-coding, precursor transcripts containing multiple TE families traversing the locus (Figure [3](#F3){ref-type="fig"}; [@B7]; [@B24]). Thus, although the reason why different lines might display different TE targeting remained elusive, it was then clear that the whole \>180 kb of the *flam* locus could be required to generate piRNAs and to perform multiple TE surveillance. {#F3} Subsequent studies have indicated that piRNA biogenesis requires many other factors than these long TE-containing transcripts and the PIWI proteins. Thus, exhaustive screens were performed to uncover the full repertoire of genes involved in this pathway. *flam*-mediated TE control became the ideal genetic model to validate candidate genes and to elucidate their activity. Indeed, the precise heterochromatic localization of *flam* had been defined from numerous genetic approaches; several of its TE targets were well known like *gypsy*, *ZAM,* and *Idefix*; transgenic tools targeted by *flam* had been constructed; several *flam* alleles with distinct suppressions of either target control were available. To date, numerous studies can be cited in which *flam* has been used to test any gene of interest for its involvement in the somatic piRNA pathway. As few examples see: [@B38], [@B39]), [@B14], [@B33], and [@B28]. *flam* TRANSCRIPTION GENERATES DIVERSE RNA PRECURSORS BEFORE BEING PROCESSED INTO piRNAs ======================================================================================== Although it provided a useful tool to validate candidate genes involved in the piRNA pathway, the mechanism of *flam* transcript did not receive much attention after the sequence analysis of its structure and piRNA production has been reported. For several years, the prevailing model held that the *flam* locus is transcribed as a continuous single stranded RNA spanning \>180 kb. However, this precursor had only been detected through quantitative RT/PCR using primer pairs spanning different regions of *flam* ([@B7]; [@B14]). In 2010, several studies identified Yb-bodies, cytoplasmic structures close to the nuclear membrane of the follicle cells, as sites of primary piRNAs biogenesis ([@B30]; [@B33]; [@B39]). piRNA intermediate-like molecules (piR-ILs) of length varying between 25 and 70 nucleotides were isolated from these structures ([@B39]). They proved to be intermediate molecules between a long precursor whose structure and regulation were still unknown, and mature piRNAs. An important issue that remained to be addressed to go further in *flam* function was to elucidate its transcriptional regulation. [@B35] reported that repressive marks deposited by dSETDB1were required for transcription from all major piRNA clusters including somatic unidirectional clusters like *flam*. In that, dSETDB1 was required for somatic TE control by *flam*. ChIP-seq experiments further indicated that *flam* is actively transcribed by RNA polymerase II and is fairly devoid of the histone mark H3K9me3, a marker of heterochromatic regions ([@B42]). In 2014, new insights into *flam* activity were reported by our group ([@B13]). We identified the promoter of *flam* as an Inr DPE promoter located at 21 502 918, 1743 bp proximal from *DIP1* (flybase version FB2011_08) and showed that its transcriptional activity requires the transcription factor, *Cubitus interruptus* (CI; Figure [3](#F3){ref-type="fig"}). In addition, we found that the *flam* precursor transcript undergoes differential alternative splicing to generate diverse RNA precursors. The intron sizes are extremely diverse and range from 0.7 to 158 kb but the first exon (exon1: 21,502,918 to 21,503,349) was found to be constitutively expressed since it is always present within the processed RNAs. Furthermore, when publicly available RNA-seq libraries were interrogated ([@B42]), piRNAs corresponding to the predicted spliced exon1--exon2 junction were identified. At the same time, piRNAs encompassing exon1/intron1 junction were under-represented in the libraries compared to piRNAs matching the spliced junction. These data indicate that *flam* transcripts are spliced before being processed in piRNAs. RNA FISH experiments indicated that these spliced transcripts are then transferred to the nuclear membrane. Indeed, we further identified a prominent nuclear structure called Dot COM, in which precursor transcripts encoded by *flam* accumulate ([@B10]). Remarkably, this structure is often juxtaposed with Yb bodies and concentrates transcripts from other piRNA clusters. When Yb-bodies are disrupted using mutations of the Armi-Piwi-Yb complex composing Yb-bodies, Dot COM is normally distributed within the nucleus and its morphology unchanged. Overall these last findings suggest the following scenario: at the initial step, *flam* RNA polymerase II transcription is activated by CI in the follicle cells. Transcripts are differently spliced to form a population of RNAs along the \>180 kb region but having in common the presence of the first exon. These RNAs are channeled from their site of transcription to Dot COM at the nuclear membrane in a location facing the Yb-bodies. From here, they are transferred to the cytoplasmic Yb-bodies and processed in piRNAs which in turn *trans*-silence complementary TEs located outside of *flam* (Figure [4](#F4){ref-type="fig"}). At this stage many questions remain to be elucidated: Where does the splicing occur? Can it be co-transcriptional or does it occur in Dot COM? How RNAs are transported from their genomic clusters to Dot COM and then to their piRNA processing center? which factors are required for these processes? {#F4} A HIGH DEGREE OF STRUCTURAL VARIATIONS AFFECTING THE *flam* LOCUS IMPACTS THE GENOMIC TE DISTRIBUTION ===================================================================================================== Despite the molecular data reported above, the link between the presence of TE vestiges in piRNA clusters and their silencing remained to be demonstrated. [@B6] proposed that β-heterochromatin TE nests could act as a trap for new TE invasions providing an "adaptive immunity" to the host genome. It could then be anticipated that different *Drosophila* lines have trapped certain TEs in piRNA clusters and not others, which would potentially explain their differential ability to repress distinct families of TEs. This was indeed what the primary genetic studies of *flam* had suggested for different *Drosophila* lines, displaying different capacities to repress or not the expression of *ZAM*, *Idefix,* and *gypsy*. To test this possibility, [@B46] used the Rev line in which the mutation affecting *flam* releases the silencing exerted on *ZAM* and *Idefix*, but not on *gypsy*. The annotation of *flam* was refined in ISO1A, the line used to generate the genome sequence in which *ZAM*, *Idefix,* and *gypsy* are silenced. Several unknown properties of *flam* were highlighted in this study. We first found that among 52 different TEs present in the *flam* locus, the vast majority (49) are present as a unique copy. This observation supports a key prediction of the transposon trap model that postulates if a TE family is silenced as soon as it inserts *flam*, it should be present only once in the locus. This study also highlighted the high structural dynamics of this locus because numerous differences resulting from deletions, insertions or duplications were identified between different lines. In addition, sequence analysis of the *flam* TEs indicated that many of them correspond to TEs that recently inserted the locus. Among them, 12 new TEs were identified. Interestingly, eight of them were found closely related to TEs from *D. simulans*, *D. sechellia*, *D. yakuba,* or *D. erecta*, consistent with a recent origin from horizontal transfers that occurred between species belonging to the *melanogaster* subgroup ([@B5]). To determine what underlies the difference between *Drosophila* lines that allow or restrict particular TEs to be mobilized, we compared the *flam* structure in ISO1A (restrictive for *ZAM*, *Idefix,* and *gypsy*) and Rev (restrictive for *gypsy* but not *ZAM* or *Idefix*). Importantly, a deletion of the region comprised between X:21638001 and 21684449 was found in Rev that encompasses the unique *ZAM* and *Idefix* copies present in *flam*. This observation provides the first evidence that a strict correlation exists between the presence or absence of TE sequences (i.e., *ZAM* and *Idefix*) within *flam* locus and repression or activity of that particular TE family. These new data highlight how structural variations in piRNA clusters impact the genomic TE distribution across the rest of the genome. Overall, data obtained on *flam* fit with a model of TE invasion and its subsequent genomic control as follows (Figure [5](#F5){ref-type="fig"}). The best genetic background for a TE family to transpose is to enter a virgin genome in which no homologous sequence exists. In such a genome, no regulatory piRNAs are produced that are able to target the new TE. For that reason, horizontal transfer of a TE coming from another species increases the chances that a TE can invade a particular genome. After entering, the newly acquired TE starts replication cycles and its copies insert across the genome. Either by chance, because of relaxed selection, or because of active targeting, a new TE copy will eventually insert into a piRNA cluster. The pool of piRNA precursors produced by this locus will then be changed because of the presence of new sequences brought in by the new TE insertion. These new precursors, transferred to Dot COM and then processed in piRNAs in Yb-bodies will act in *trans* to silence their homologous copies. When this occurs, genomic stability is recovered. Due to their highly repetitive nature, piRNA clusters may subsequently undergo deletion events removing small or large portions of the locus. These deletions can remove TE sequences and may result in sudden bursts of transposition. Thus, periods of stability and instability in global TE dynamics will reflect the mutational events that affect piRNA clusters. {#F5} Conceptually, this dynamics of the *flam* locus provides an RNA-mediated adaptive immunity against TE invasions. Interestingly, this system in *Drosophila* shares striking resemblances with the CRISPR system developed by bacteria to fend off invaders ([@B4]). CRISPR loci (clustered regularly interspaced short palindromic repeats) are typically flanked by CRISPR-associated genes (Cas). The CRISPR-Cas system mediates immune defense involving sequence specific, RNA-mediated targeting of genetic invaders. The first step of the CRISPR-Cas protection occurs when new sequences derived from invading elements like viruses or plasmids are incorporated into the CRISPR locus. This locus is subsequently transcribed and processed into small interfering RNAs that guide Cas nucleases for specific cleavage of complementary sequences. This genome surveillance is thus triggered as soon as a TE, a virus or their derived sequences fall within the trap. It is interesting to note that, for both *flam* and CRISPR loci, these sequences remaining from invasions are transferred to the progeny in which they constitute genetic marks reflecting environmental changes over time. After 40 years of data obtained on *flam*, it is interesting to measure how far we have gone since that time where heterochromatin was considered as a graveyard for TEs. Today, TEs and piRNA clusters in heterochromatin are thought to play fundamental roles in the organization and stability of genomes. The high structural dynamics of *flam* and potentially of the other piRNA clusters appears as a formidable evolutionary tool to remodel both euchromatic and heterochromatic regions, or even to play a role in speciation ([@B41]), by its ability to alternatively constrain or permit TE mobilization. How far will further work on *flam* lead knowledge of heterochromatin function in the years to come? Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited by: *Jeff Schwartz, Griffith University, Australia* [^2]: Reviewed by: *Julius Brennecke, Austrian Academy of Sciences, Austria* [^3]: This article was submitted to Epigenomics and Epigenetics, a section of the journal Frontiers in Genetics.

1. Introduction {#sec1} =============== The heat-shock proteins (Hsps) are a group of highly conserved proteins with major physiological roles in protein homeostasis \[[@B1], [@B2]\]. In most cell types even prior to stress Hsps constitute 1%-2% of total protein, suggesting an important role for these proteins in the biology and physiology of the unstressed cell. These particularly concern regulating the folding and unfolding of other proteins. The proteins are named, however, because they were first identified on the basis of their increased synthesis following exposure to elevated temperatures \[[@B3]\]. Subsequently it has been clearly shown that they can be induced following a variety of stressful stimuli. Some Hsps, such as Hsp90 (each Hsp is named according to its mass in kilodaltons) are detectable at significant levels in unstressed cells, increasing in abundance following a suitable stimulus, whilst others such as Hsp70 exist in both constitutively expressed and inducible forms that is activated by stressful stimuli \[[@B4], [@B5]\]. The dual role of Hsps in both normal and stressed cells, evidently requires the existence of complex regulatory processes which ensure that the correct expression pattern is produced. Indeed, such processes must be operative at the very earliest stages of embryonic development since the genes encoding Hsp70 and Hsp90 have been shown to be amongst the first embryonic genes which are transcribed \[[@B6], [@B7]\]. The induction of Hsps in response to various stresses is dependent on the activation of specific members of a family of transcription factors, the heat-shock factors (HSFs) which bind to the heat-shock element (HSE) in the promoters of the genes encoding Hsps \[[@B8]\]. Four HSFs (HSF1 to −4) have been cloned from a number of organisms and their roles have now been characterised. Only HSF1 and HSF3 have been shown to be involved in regulating Hsps in response to thermal stress whereas HSF2 and HSF4 are involved in Hsp regulation in unstressed cells and their levels are regulated in response to a wide variety of biological processes such as immune activation and cellular differentiation \[[@B8]\]. In general, however, the stimuli which induce such alterations in Hsp gene expression under nonstress conditions are poorly characterized and the mechanisms by which they act are unclear. In this paper, we discuss recent studies indicating that Hsps are not only regulated by HSFs alone, but also by transcription factors which are able to interact or cooperate with HSF1 and modulate the transcriptional regulation of Hsps in response to nonstressful stimuli. More recently, as will be explained later, it has also been reported that HSF2, like HSF1 can also play a role as a stress-inducible factor in promoting the induction of Hsps under certain conditions. 2. Transcriptional Regulation of Hsps by the HSF Family {#sec2} ======================================================= 2.1. HSF1 {#sec2.1} --------- As mentioned previously, HSF1 has been identified as the HSF that mediates stress-induced Hsp gene expression in response to environmental stressors. Such stresses cause HSF1 oligomerization and nuclear translocalization, followed by enhanced DNA binding on the Hsp gene promoters. Recent studies have shown that HSF1 is negatively regulated by Hsp70 and Hsp90, therefore suggesting a negative-feedback loop for the regulation of Hsp70 and Hsp90 genes following a heat-shock response \[[@B8]--[@B10]\]. HSF1 is known to undergo posttranslational modification by various processes including phosphorylation, acetylation, and sumoylation \[[@B8]\]. Both phosphorylation and sumoylation are involved in regulating the transactivation capacity of HSF1 \[[@B8]\]. More recent, whereas p300 has been shown to acetylate HSF1, deacetylation by the NAD+-dependent sirtuin (SIRT1) is involved in the attenuation phase of the heat-shock response by preventing HSF1 acetylation and DNA binding \[[@B8]\]. The kinases responsible for phosphorylating HSF1 on several serine sites include glycogen synthase kinase 3*β* *(*GSK*β*) and c-jun N-terminal kinase (JNK) \[[@B12], [@B13]\]. The cytokine interleukin 6 (IL-6) has been shown to derepress HSF1 by reducing the activity of GSK*β* \[[@B14]\]. However, a positive role of HSF1 phosphorylation in the stress-induced activation of Hsp gene expression is also known to occur. The exact mechanism of this effect has not been fully elucidated, although the protein kinase CK2 seems to be involved in enhancing transcriptional activity and the DNA binding of HSF1 by phosphorylating the threonine 142 residue \[[@B15]\]. It is suggested that activation may also involve dephosphorylation of HSF1 \[[@B12]\]. The key role of HSF1 has been supported by the findings that cells lacking this crucial factor exhibited defects in Hsp induction following exposure to heat shock \[[@B16]\]. Moreover, cells lacking HSF1 were susceptible to apoptotic cell death following exposure to heat stress \[[@B16]\]. In addition, mice lacking HSF1 also had elevated levels of tumour necrosis factor *α* (TNF-*α*), which resulted in increased mortality after endotoxin and inflammatory challenge \[[@B16]\]. Interestingly, HSF1 has been shown to also modulate other genes such as interleukin-1*β* and c-fos \[[@B17], [@B18]\], suggesting a role for HSF1 in regulating stress responsive genes other than those encoding Hsps. More recently, it has been reported that HSF1 also functions in the circadian clock as a circadian transcription factor. The circadian clock enables an organism to adapt to conditions by presetting the area in the brain that controls behavioural changes. Circadian transcription factors are known to be regulated in a timely and rhythmic fashion Thus, using a novel technique of differential display of DNA-binding proteins (DDDPs), HSF1 was shown to be highly rhythmic in its transcriptional activity. Moreover, HSF1 enhanced the expression of Hsps at the onset of the dark phase, when the animals start to be behaviourally active. Furthermore, Hsf1-deficient mice have a longer free-running period and therefore more active than wild-type littermates, suggesting a combined role for HSF1 in the mammalian timekeeping and cytoprotection systems \[[@B19]\]. 2.2. HSF2 {#sec2.2} --------- As mentioned earlier, Hsp gene expression is crucial not only for the survival of cells exposed to extracellular stress stimuli, but also during normal cellular processes such as embryonic development and cellular differentiation. HSF2 has now been described as the factor involved in regulating Hsps under nonstressful conditions. For example, it was previously reported that Hsp70 expression is activated when K562 cells are induced by hemin and this process requires activation of HSF2 \[[@B20]\]. HSF2 exists as two isoforms, HSF2*α* and HSF2*β*, due to alternative splicing, where the HSF2*α* isoform is predominantly expressed in adult tissue, while the HSF2*β* isoform is predominantly expressed in embryonic tissue \[[@B21]\]. HSF2 DNA binding activity is high during early embryogenesis in tissues such as the heart, central nervous system, and testis \[[@B21]\]. The importance of HSF2 in development was recently reported and Hsf2-null mice display gametogenesis defects and brain abnormalities characterized by enlarged ventricles \[[@B22]\]. During mitosis, the genome is well known to be compacted in order for chromosomes to be segregated during cytokinesis. However, some gene promoters such as the inducible Hsp70i (heat stress-induced upregulation) remain uncompacted. The factors that control and prevent this process of compaction or bookmarking have been recently characterized. For example, Hsp70i bookmarking is now known to be mediated by HSF2, which binds this promoter in mitotic cells, recruits protein phosphatase 2A, and interacts with the CAP-G subunit of the condensin enzyme to promote efficient dephosphorylation and inactivation of condensin complexes in the vicinity, thereby preventing compaction at this site \[[@B23]\]. Blocking HSF2-mediated bookmarking by HSF2 RNA interference decreases hsp70i induction and survival of stressed cells in the G1 phase, which demonstrates the biological importance of gene bookmarking. HSF2 has also been shown to be bound to the HSE promoter elements of other heat-shock genes, including Hsp90 and Hsp27, as well as the proto-oncogene c-fos \[[@B24]\]. These data suggest that HSF2 is important for constitutive as well as stress-inducible expression of HSE-containing genes. It is also known that HSF2 can form heterotrimers with HSF1. Following certain stress, HSF1 is activated and HSF1-HSF2 heterotrimers are formed. Heat-shock stress diminishes the levels of HSF2 and restricts heterotrimerization by limiting the availability of HSF2. It has been suggested that HSF1-HSF2 heterotrimerization provides a switch that integrates the transcriptional activation in response to specific stimuli during developmental processes; for review see \[[@B8]\]. 2.3. HSF3 {#sec2.3} --------- HSF3 was originally identified in avian cells and no reports have yet described HSF3 in other organisms. Like HSF1, HSF3 is also heat-stress responsive \[[@B25]\]. However, the threshold temperature required to activate HSF3 and HSF1 are different in that HSF1 is activated by less severe heat shock than HSF3 \[[@B25]\]. Previously, HSF3 was reported to bind to c-Myb, a transcription factor involved in cellular proliferation and required for the G1/S transition of the cell cycle, which also paralleled the expression of Hsp70 \[[@B26]\]. These studies suggest that HSF3/cMyb interaction may be involved in cell cycle-dependent expression of Hsps. Furthermore, more recently, it has been shown that HSF3/c-Myb association is disrupted by direct binding of the p53 tumour suppressor transcription factor to HSF3, resulting in inhibition of Hsp70 expression \[[@B27]\]. 2.4. HSF4 {#sec2.4} --------- In contrast to HSF1 and HSF2 proteins, which are expressed in most tissues, the level of HSF4 protein is very low in many mammalian tissues except in lung and brain \[[@B28]\]. There are at least two isoforms, HSF4a and HSF4b, which are derived by alternative RNA-splicing events. HSF4b is able to activate transcription whereas HSF4a does not and this differential effect has not yet been characterized \[[@B28], [@B29]\]. Interestingly, mutations of HSF4 have been associated with dominant inherited cataracts in human \[[@B30]\]. More recently, HSF4 has been revealed to have a role in regulating lens-specific gamma-crystalline genes during lens development \[[@B31]\]. 3. The Role of Non-HSF Transcription Factors in Modulating Hsp Gene Expression by the STAT and NF-IL6 Pathways {#sec3} ============================================================================================================== The phenotype of mice lacking HSF1 is normal in the absence of stress and expression of Hsp70 and Hsp90 in cells lacking HSF1 is similar to wild-type cells, although they exhibit a defect in the heat-shock response following heat stress \[[@B14]\]. These studies suggest that other HSFs may compensate for the lack of HSF1 and/or that other transcription factors may also be responsible for expression of Hsps under normal growth conditions. Recent studies from our laboratory have identified a separate group of transcription factors that are activated by distinct cytokines and are able to modulate Hsp70 and Hsp90 gene expression. These factors include STAT1, STAT3, and NF-IL6 and their functional roles are described below. The STATs are a family of cytoplasmic transcription factors that mediate intracellular signalling initiated at cytokine cell surface receptors and transmitted to the nucleus. STATs are activated by phosphorylation on conserved tyrosine and serine residues on their C-terminal domains by the Janus kinases (JAKs) and MAP kinase families, respectively. This allows the STATs to dimerise and translocate to the nucleus and thereby regulate gene expression \[[@B32]\]. Interferon-*γ* is a potent activator of STAT1, whilst the interleukin-6 (IL-6) family members including IL-6, leukaemia inhibitory factor (LIF), and CT-1 primarily activate STAT3 \[[@B32]\]. Our laboratory has previously shown that STAT1 and STAT3 have opposing actions on apoptotic cell death in various cell types \[[@B33]\]. We reported that overexpression of STAT1 is able to enhance apoptotic cell death in cardiac myocytes exposed to ischaemia reperfusion (I/R) injury whereas overexpression of STAT3 with STAT1 is able to reduce the levels of STAT1-induced cell death following I/R by modulating the expression of pro- and antiapoptotic genes \[[@B34]\]. Furthermore, these effects on apoptosis require serine-^727^ but not tyrosine-^701^ phosphorylation on the C-terminal transactivation domain of STAT1 \[[@B35], [@B36]\]. Moreover, we have shown that STAT1 is able to modulate the activity of p53 and its effects on apoptosis \[[@B37]\]. These effects involve STAT1/p53 protein-protein interaction with STAT1 acting as a coactivator for p53 \[[@B37]\]. We have also demonstrated that STAT1 is also able to interact with another p53 family member p73 \[[@B38]\]. However, in contrast to STAT1-p53 interaction which enhances p53 transcriptional activity, the STAT1-p73 interaction was shown to reduce p73 functional activity on similar p53-responsive genes \[[@B38]\]. Thus, STAT1 is able to have differential effects on p53/p73 transcriptional activity. A link between p53 activity and the HSF1-heat-shock response pathway has recently been documented by the finding that HSF1 interacts with stress-responsive activator of p300 (Strap) transcription cofactor, a key factor controlling the DNA damage response through its ability to regulate p53 activity \[[@B39]\]. Moreover, Strap augments HSF1 binding and chromatin acetylation in Hsp genes, most probably through the p300 histone acetyltransferase activity of p300 itself. Furthermore, cells depleted of Strap do not survive under heat-shock conditions \[[@B39]\]. Overall, these data indicate that Strap is an essential cofactor that acts at the level of chromatin control to regulate heat-shock-responsive transcription. The cytokine IL-6 is known to stimulate two distinct signalling pathways, resulting in the activation of two different classes of cellular transcription factors \[[@B40]\]. Thus, initial studies showed that a variety of IL-6-inducible genes contained binding sites for a transcription factor named NF-IL6 (nuclear factor IL-6), which showed high homology with the rat-liver nuclear factor C/EBP (CCAAT-enhancer-binding protein), and is therefore also known as C/EBP*β* \[[@B41]\]. Subsequently, another member of the C/EBP family, known as NF-IL*β* or C/EBP*δ*, was identified and shown to form heterodimers with NF-IL6, resulting in a synergistic transcriptional effect \[[@B42]\]. After exposure of cells to IL-6, NF-IL6 is phosphorylated, resulting in its enhanced ability to stimulate transcription \[[@B42]\] whereas NF-IL6*β* is synthesized *de novo* \[[@B42]\]. As mentioned above the second pathway which is stimulated by IL-6 is the JAK/STAT3 signalling pathway. It is generally accepted that the NF-IL6/NF-IL6*β* and STAT3-signalling pathways allow IL-6 to activate two distinct sets of genes, each of which is responsive to one of these pathways. Thus, class 1 acute-phase proteins (such as *α* ~1~-acid glycoprotein, haptoglobin, C-reactive protein, and serum amyloid) contain response elements for NF-IL6 and NF-IL6*β* and these factors have been shown to be involved in the activation of these genes following IL-6 treatment \[[@B33]\]. In agreement with this idea, these genes are stimulated by exposure of cells to IL-1 which also stimulates NF-IL6/NF-IL6*β* activity without affecting STAT3 \[[@B43]\]. In contrast, type 2 acute-phase genes such as fibrinogen, thiostatin, and *α* ~2~ microglobulin are not inducible by IL-1 and lack binding sites for NF-IL6/NF-IL6*β*. Instead, these genes contain binding sites allowing binding of STAT3, which is responsible for activation of these genes in response to IL-6 \[[@B43]\]. 4. Role of STAT1, STAT3, and NF-IL6 Factors in Modulating Hsps {#sec4} ============================================================== We previously reported \[[@B44]\] that IL-6 can induce increased expression of the 90 kDa heat-shock protein (hsp90) in a variety of different cell types. The hsp90*β* gene promoter was shown to be responsive to IL-6 and could also be activated by NF-IL6 or NF-IL6*β*  \[[@B44]\] . Moreover, a short region of the promoter containing an NF-IL6-binding site was essential for activation of the promoter by both IL-6 and NF-IL6. This promoter region could confer responsiveness both to IL-6 and to overexpression of NF-IL6 on a heterologous promoter. These findings suggested that hsp90 was a member of the class of IL-6-responsive genes that were activated by NF-IL6/NF-IL6*β*. Interestingly, this short region of the promoter also contains binding sites for STAT3 and the hsp90 promoter can be activated also by this factor. Moreover, overexpression of NF-IL6 and STAT3 has a synergistic effect on the hsp90 promoter and both these signalling pathways appear to be required for activation of the hsp90 promoter by IL-6 \[[@B45]\]. Despite their synergistic action in IL-6 signalling, however, these two pathways have opposite effects on the heat-shock-mediated regulation of the hsp90 promoter. Thus STAT3 reduces the stimulatory effect of heat shock whereas NF-IL6 enhances it. When applied together, heat shock and IL-6 produce only weak activation of the hsp90 promoter compared with either stimulus alone, indicating that the inhibitory effect of STAT3 on HSF predominates under these conditions \[[@B45]\]. In contrast, IL-1, which activates only the NF-IL6 pathway, synergizes with heat shock to produce strong activation of hsp90 \[[@B45]\]. These results therefore open up a new aspect of hsp90-gene regulation which is additional to and interacts with the heat-shock-activated pathway. Previously, we had also examined whether STAT1 is able to modulate Hsp expression. We showed that IFN-*γ* treatment increases the levels of Hsp-70 and Hsp-90 and also enhances the activity of the Hsp-70 and Hsp-90*β* promoters with these effects being dependent on activation of the STAT1 transcription factor by IFN-*γ* \[[@B46]\]. These effects were not seen in a STAT1-deficient cell line, indicating that IFN-*γ* modulates Hsp induction via a STAT1-dependent pathway. The effect of IFN-*γ*/STAT1 was mediated via the same short region of the Hsp-70/Hsp-90 promoters, which also mediates the effects of NF-IL6 and STAT3 and can bind STAT1 \[[@B46]\]. This region also contains a binding site for the stress-activated transcription factor HSF1. We showed that STAT1 and HSF1 interact with one another via a protein-protein interaction and produce a strong activation of transcription \[[@B46]\]. This is in contrast to the previous finding that STAT3 and HSF1 antagonize one another and we showed that STAT3 and HSF1 do not interact directly. To our knowledge, this was the first report of HSF1 interacting directly via a protein-protein interaction with another transcription factor. Such protein-protein interactions and the binding of a number of different stress and cytokine-activated transcription factors to a short region of the Hsp-90 and Hsp-70 gene promoters are likely to play a very important role in Hsp gene activation by nonstressful stimuli and the integration of these responses with the stress response of these genes. Moreover, our findings that STAT1 can interact with p53 and that both these factors are able to modulate the effects of HSF1 on Hsp expression, suggests different interacting partners of HSF1 may affect HSF1-mediated transcriptional regulation. 5. Linking HSF1, STAT1, STAT3, and NF-IL6 Elevation to Pathological Diseases {#sec5} ============================================================================ A number of disease states have been shown to exhibit elevated levels of Hsps \[[@B47]\]. This includes patients with systemic lupus erythematosus (SLE) who have elevated levels of Hsp90. Interestingly, elevated levels of circulating IL-6 have also been reported in SLE \[[@B48]\], and the levels have been shown to be correlated with disease activity, being highest in patients with active disease. Moreover, spontaneous production of IgG by normal and SLE-derived B lymphocytes in culture can be enhanced by the addition of exogenous IL-6 and inhibited by antibody to IL-6 \[[@B49]\]. These findings therefore suggest that IL-6 might play a role in the pathogenesis of autoimmune diseases. Moreover, infusion of an antibody to IL-6 can relieve disease symptoms in lupus-prone NZB/NZW F1 mice \[[@B50]\]. Furthermore, elevated levels of Hsp90 in SLE correlated with levels of IL-6 and of autoantibodies to Hsp90 \[[@B51]\]. In order therefore to test directly the role of IL-6 in regulating Hsp90 expression *in vivo* we have used mice which have been artificially engineered to express elevated levels of IL-6 either by being made transgenic for extra copies of the IL-6 gene \[[@B52]\] or by inactivation of the gene encoding the transcription factor C/EBP*β* which also results in the elevation of IL-6 levels in these mice \[[@B53]\]. In these experiments, elevated levels of Hsp90 were observed in both the IL-6 transgenic and the C/EBP*β* knock-out mice \[[@B54]\]. Hence, the elevated IL-6 levels induced in these animals are indeed paralleled by increased levels of Hsp90 compared to normal control mice. In addition, it was also observed that in both IL-6 transgenic and C/EBP*β* knock-out animals, elevated hsp90 was associated with the specific production of autoantibodies to Hsp90. It is also of interest that inactivation of the IL-6 gene in the C/EBP*β* knock-out mice resulted in the suppression of Castleman-like disease normally observed in these animals and a reduction in the production of autoantibodies. These results support a model in which elevated levels of IL-6 in SLE patients induce increased levels of Hsp90 protein which in turn results in the production of autoantibodies to this protein. Additionally, IL-10 is also elevated in SLE and IL-10 was demonstrated to enhance Hsp90 gene expression \[[@B55]\]. Therefore, these studies strongly suggest that IL-6 and IL-10 are likely to play a critical role in the regulation of Hsp90 levels and autoantibody production in autoimmune disease states. As described above, NF-IL6 performs diverse functions, participating in the regulation of genes that contribute to the known acute phase response, but also to glucose metabolism, and tissue differentiation, including adipogenesis and hematopoiesis \[[@B56]--[@B58]\]. Hsps and STAT3 have also been found at increased levels in many solid tumours and haematological malignancies \[[@B59], [@B60]\]. Their expression may in part account for the ability of malignant cells to maintain protein homoeostasis even in the hostile hypoxic microenvironment of the tumour. Thus, STAT3 seems to function as an antiapoptotic factor, especially in numerous malignancies, where STAT3 is often constitutively active/phosphorylated and STAT3 activation has been associated with advanced stages of metastatic cancers such as prostate cancer \[[@B61]\]. Furthermore, STAT3 behaves as an oncogene, and is able to transform normal fibroblast cells which can then form tumours in nude mice \[[@B62]\]. Thus, targeting STAT3 activation has been suggested to be an attractive anticancer therapy \[[@B61]\]. Likewise, Hsps allow tumour cell survival, growth, and metastasis, even in growth factor-deprived conditions, by allowing continued protein translation and cellular proliferation \[[@B63], [@B64]\]. Therefore, targeting of Hsps with chemical inhibitors may be beneficial in multiple oncogenic processes. \[[@B65]\]. Further evidence for a link between Hsps and cancer was reported from studies in the HSF1 knockout mice, which showed reduced development of tumours, and HSF1 deficiency rendered cultured cells highly refractory to transformation initiated by mutated RAS or by platelet-derived growth factor-B (PDGF-B) overexpression \[[@B66], [@B67]\]. Similarly, HSF1 depletion decreased viability of multiple human cancer cell lines, but had no effect on normal cells, suggesting that HSF1 provides critical relief to the cellular stresses experienced by cancer cells \[[@B68]\]. It is therefore plausible that the STAT3-Hsp interactions may be one such survival pathway that allows tumour cell survival, growth and metastasis in cancers. 6. Conclusion {#sec6} ============= This review paper demonstrates the modulation of Hsps by a group of transcription factors other than the traditional HSF family under normal nonstressful conditions and also in several disease states. The finding that the responses to these factors occur around the HSF DNA binding site in the Hsp gene promoters, suggests that HSF1 as well as other HSFs are able to interact or cooperate with STATs or NF-IL6 family members. Further studies to identify novel protein interacting partners for HSFs will also provide insight into the regulation of Hsps and other molecular chaperones. Unravelling the mechanistic basis of this cooperation will undoubtedly enhance our understanding of the interdependent relationship between distinct HSFs and their interaction with other factors in the complex regulatory processes which ensure that the correct Hsp expression pattern is produced under different physiological states ([Figure 1](#fig1){ref-type="fig"}). {#fig1} [^1]: Academic Editor: Daniel N. Hebert

Introduction {#Sec1} ============ Distal radius fractures are the most common fractures occurring in childhood and a substantial proportion of these patients will develop malunions initially. Fortunately, malunions in children often show a tremendous remodeling potential and initial treatment can usually be restricted to the reassurance of the parents of the involved child. However, although this is a well-known practice for most doctors treating children with fractures, surprisingly few studies (*n* = 7) are available with quantitative data on the dynamics of remodeling. The time needed for the remodeling process is unknown, which impedes the prediction of outcome and, thus, proper patient information. Reported remodeling times (RT) to full correction vary between a mean of 4 months \[[@CR1]--[@CR3]\] and 5 years \[[@CR4]\] in the literature. In addition, the speed of remodeling has been shown to vary between 0.9° to 2.5°/month \[[@CR5]--[@CR7]\]. Greater angulated fractures tend to remodel at a faster rate \[[@CR5], [@CR7]\]. Hence, the use of a general remodeling speed to predict RT to full correction is not feasible. Friberg, therefore, developed a (exponential) model using the primary malunion angulation (*A*~0~) to describe the residual angulation (*A*~T~) in distal radius malunions \[[@CR5]\]. The model, however, lacks accuracy and is, therefore, only rarely used in orthopedic practice. The aim of the present study is to develop a model which accurately predicts the dynamics of the remodeling process. We use the remodeling data of two previously published studies to modify Friberg's model in order to enhance its accuracy. In addition, we develop a model to calculate the time needed for complete remodeling. These models should allow to provide a more evidence-based patient education and select those malunions that will not sufficiently remodel and require intervention. Patients and methods {#Sec2} ==================== We used data from two published cohorts of children with distal radius fractures with dorsovolar angulation. Cohort A is from a study on the remodeling of malunions of forearm fractures which presents a table with patient data on 36 children \[[@CR4]\]. From this table, were selected the malunions in the distal third of the forearm in dorsovolar dislocation (*n* = 31). Cohort B was derived from a study on the remodeling speed of distal radius fractures with dorsovolar angulation more than 15° (*n* = 32) \[[@CR7]\]. Angle measurements in both cohorts were identical: the central longitudinal intramedullary axis was determined in both the proximal and (angulated) distal fragment. The angle between these two axes was used as the angulation angle. This method was described by Hansen et al. \[[@CR8]\]. From all the included patients, we assessed age at time of fracture, gender, malunion angulation (*A*~0~) in the dorsovolar direction, angulation at follow-up, and time of follow-up (= RT). Because both studies were retrospective, the follow-up times (= RT) differ. The difference between initial malunion angulation (*A*~0~) and angulation at follow-up (*A*~T~) was defined as remodeling, measured in degrees. Using the data from the combined cohort, two models were evaluated: Firstly, a prediction model was formulated based on the findings by Friberg \[[@CR5]\]: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = A_{0} \times e^{ - C \times RT}$$\end{document}$ and, secondly, we modified this model with a second coefficient to study the influence of *A*~0~: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = B \times A_{0} \times e^{ - C \times RT}$$\end{document}$ (the coefficients were calculated using the nonlinear regression function of SPSS, see below). Statistical analysis {#Sec3} -------------------- All data were analyzed using SPSS (version 15.0, SPSS Inc., Chicago, IL, USA). The results are presented as means (standard deviation, SD). Nonlinear regression was used to estimate the coefficients of the models. For the Friberg-based model, we started with the coefficient found in that study. For the modified model, the starting value for the second coefficient was the value found in the study of Jeroense et al. \[[@CR7]\]. The significance of the difference of the parameters and differences between subgroups was tested using the *t* test. To test the precision of the prediction of the models, we compared predicted and observed RT using parametric techniques (*t* test). The best of the two models was subsequently used to estimate time needed to complete remodeling. All tests are two-tailed and considered significant if *p* \< 0.05. Results {#Sec4} ======= Data are based on the analysis of 63 dorsovolar malunions of the distal radius: 31 from the study by Gandhi (A) (cases 1--31 in the patient data table) and 31 patients (32 malunions) from the study by Jeroense (B) (see [Appendix](#Sec11){ref-type="sec"}). There were 38 boys, with a mean age of 8.5 years (range 2--14.5 years). The mean malunion angulation was 25° (SD 7.8), mean remodeling time 22 (SD 18) months, and mean angulation at follow-up 6.7° (SD 5.8). The cohorts showed differences in follow-up time (35 vs. 9 months) and final angulation (see Table [1](#Tab1){ref-type="table"}).Table 1Summary of the data from 62 patientsGandhi cohort (A), *N* = 31Jeroense cohort (B), *N* = 31DifferenceSignificanceAge (years)7.79.11.3 years0.043Remodeling time (months)35925 months0.000Malunion angulation (*A* ~0~)26°24°2.5°0.1Angulation at FU5°8°3.5°0.02Comparison of the two subgroups Prediction of remodeling {#Sec5} ------------------------ ### Friberg's exponential model {#Sec6} Using Friberg's model for the combined cohort, the prediction coefficient was 0.13 \[confidence interval (CI): 0.1--0.16), with a low precision (*R*^2^ = 0.11). Using the model for subgroup analysis (cohorts A and B), we found significant differences in the coefficient of remodeling with B coefficients of 0.06 (95 % CI: 0.068--0.045) and 0.17 (95 % CI: 0.21--0.13), respectively (*p* \< 0.05). ### Modified exponential model {#Sec7} We developed a modified model by adding a second coefficient to modify A~0~. The best fit is the model $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = 0.51 \times A_{0} \times e^{ - 0.034 \times RT}$$\end{document}$ (51 % of the starting angulation and a coefficient of 0.034 for RT). This improves prediction for the combined cohort: *R*^2^ = 0.47. With this model, the subgroups did not differ (Table [2](#Tab2){ref-type="table"}). Adding age or gender did not improve the model. Analysis excluding the four patients older than 12 years of age only marginally influenced the results of this nonlinear regression.Table 2Models of observed remodeling (*A* ~T~) and initial malunion angle (*A* ~0~) and RT (*n* = 63 malunions)ModelDependent variableIndependent variablesModel95 % CI of coefficient of RT*R* ^2^Model of FribergA~T~RT, *A* ~0~$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = A_{0} \times e^{ - 0.13 \times RT}$$\end{document}$0.1--0.160.1Modified modelA~T~RT, *A* ~0~$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = 0.5 \times A_{0} \times e^{ - 0.034 \times RT}$$\end{document}$0.024--0.0440.47*A* ~T~ remodeling angle, *RT* remodeling time, *A* ~0~ initial malunion angulation Precision of the exponential models {#Sec8} ----------------------------------- In both models, the predicted values of remodeling were not significantly different from the observed values. The mean difference between the observed and predicted remodeling based on the Friberg model with the present coefficient was 1.1°. The modified model had a mean difference of 0.07° with the observed values (see Table [3](#Tab3){ref-type="table"}).Table 3Differences between the observed and predicted remodeling of malunions angulations (*n* = 63)Observed remodeling (mean)Predicted remodeling (mean)Difference (*p*)95 % CI of differenceNumber of predictions \<5°Friberg model6.7°5.6°1.1° (*p* = 0.1)−0.2 to 2.541/63Modified model6.7°6.6°0.07° (*p* = 0.89)−0.9 to 1.148/63 Although the mean differences between predicted and observed values of the original Friberg model was small, the SD was substantial. Using the Friberg model, the values in 41/63 fractures were within 5° of predicted values and, in four cases, differed by more than 10°. Using the modified model, the mean difference was 0.07° (SD 4.2°) and with a smaller SD; 48/63 were within 5° (see Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Observed and predicted angulation at follow-up (*A* ~T~) of distal radius malunions. On the horizontal axis is the remodeling time, and on the vertical axis, the observed remodeling angulations (°) are shown next to the predicted angulation (*filled circles*) based on the model $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = 0.5 \times A_{0} \times e^{ - 0.034 \times RT}$$\end{document}$ Time needed for remodeling {#Sec9} -------------------------- The modified model was used to derive a formula for remodeling time. However, since remodeling is an asymptotic function, completed remodeling cannot be determined with the model. For practical purposes, the value of 3° was considered as adequate remodeling. With *A*~T~ = 3°, derivation from the modified model yields the formula $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$= \frac{{{ \ln }\left( {\frac{{A_{0} }}{6}} \right)}}{C}$$\end{document}$ (see [Appendix](#Sec11){ref-type="sec"} for derivation). This formula was used to calculate predicted remodeling times with different coefficients in the modified model using values based on the assessed CI (low--mean--high). The mean and low coefficients resulted in RT longer than described in the literature; only the high coefficient yielded values in accordance with the published results. Using the information on remodeling time described in the literature, this study presents an estimated guess of RT depending on malunions angulation in Table [4](#Tab4){ref-type="table"}.Table 4Estimates of remodeling time of distal radius malunions based on the modified model $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$RT = \frac{{{ \ln }\left( {\frac{{A_{0} }}{6}} \right)}}{C}$$\end{document}$ with fast coefficient *C* and on data from the literatureMalunion angulationExpected remodeling time, mean (range), monthsBased on15°12 (2.5--13)Do study30°36 (30--48)Based on modified model using fast coefficient. Confirmed by Johari and Roth study40°40--50Based on modified model using fast coefficient. Confirmed by Gandhi study Discussion {#Sec10} ========== This study shows that remodeling of distal radius fractures can be described as an exponential function. The use of the original model of Friberg turned out to be less accurate, with a low *R*^2^ and only 41/63 (65 %) of the malunions showed final remodeling within 5°. Whereas in the original study the exponential coefficient was 0.087 \[standard error (SE) 0.058\], the present study found a coefficient of 0.13 (CI: 0.1--0.16). In addition, the two subgroups (cohorts A and B), when calculated according to the Friberg model, showed statistically different values: the oldest (Study A, UK, 1962) has the slowest remodeling (*B* = 0.057), while the most recent (Study B, The Netherlands, 2015) has the fastest (*B* = 0.16). Using the modified model resulted in a more accurate prediction of the remodeling process, with 48/63 (76 %) malunions within 5°, with an *R*^2^ of 0.45. Moreover, when using this modified model, no differences were found between the two subgroups. The exponential model is better than a linear model but intuitively difficult. For practical purposes, a table has been presented with estimates which can be used for prediction. As a rule of the thumb, the estimated time for remodeling would be around 1°/month for distal radius fractures, with 1.5° in the first 6 months. Since the modified model proved to be the most accurate predictor of remodeling, this model was used to derive a formula for the remodeling time. However, we found a discrepancy between the remodeling times calculated with our formula for the mean coefficient compared to earlier studies in the literature. For 15° of malunion, the RT estimates would be between 12 and 38 months, which does not agree with the study of Do et al. \[[@CR1]\], who showed that angulations below 15° correct spontaneously after an average time of 4 months (range 2.5--13 months); apparently, observed remodeling in the literature is faster in the first year than in the presented cohort. The RT calculation using the high coefficient is the best approximation of the literature. Estimates for that value are still longer than the time reported by Johari \[[@CR2]\] (36 months, range 30--48) but agree with Roth et al. \[[@CR3]\], who reported 42 months. Moreover, Gandhi's statement that 95 % of the fractures are corrected after 60 months is correct but might be too conservative. A limitation of this study is that the distal radius fractures studied are a heterogeneous group with some located proximally in the distal third and some distal in that segment. Since the more proximal fractures remodel slower, this may have caused some of the variability found. In addition, the two cohorts have different follow-up times. This has the advantage of having data with a longer time interval for study but, possibly, differences in the early months are less clearly visible. They are from different decades but that should not affect the underlying biological process. Using the original Friberg model, there seems to be a difference in remodeling behavior, but using the modified model, the differences disappeared. Whether this model only describes the study data or can be generalized remains to be tested. A final limitation is that the exponential model is asymptotic and never predicts full remodeling. This suggests that corrective growth is not only longitudinal but also shows a tendency to realign to the anatomical axis. For this, the model might be further expanded. In conclusion, the remodeling process of distal radius malunions in children can be described as an exponential function, with its starting speed dependent on the initial angulation. The current modified model proves to be more accurate than the model derived from the findings of Friberg. In addition, a formula for the prediction of remodeling time, based on the modified model, was described. These models add to our insight of the remodeling process and allow for more evidence-based patient information and optimal planning of eventual surgical intervention. Furthermore, the postulated model could serve as a basis for the description of the correction of other malunions by adaptation of the coefficients in this model. Appendix {#Sec11} ======== Derivation of RT using the modified model. $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{T} = 0.5 \times A_{0} \times e^{ - C \times RT}$$\end{document}$$ For remodeling to 3°: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3 = 0.5 \times A_{0} \times e^{ - C \times RT}$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 = \frac{{A_{0} }}{6} \times e^{ - C \times RT}$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln1 = \ln \left( {\frac{{A_{0} }}{6}} \right) - C \times RT$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$RT = \frac{{\ln \left( {\frac{{A_{0} }}{6}} \right)}}{C}$$\end{document}$$ Table 5Patient data overviewPatientsCaseAge (years)SexMalunion angulation (°)FU angulation (°)RT (months)Cohort A19M32160211F3006034F1506047M3506052M2506068M3006079M1504885M3734894F20448109M330481111M241039127M21436135M24439148M28936159M25736168M25733174M346331812M391227198M3210272010F20427218M16027228M250242310M331324249F351024259M141021265F381218277M331312289M281412297M20012307M130123111M20012Cohort B325.5F181010337F20416346.5M20--1.518359F1594367M1876.5375.5M291110383F191043911F311664010.5M28213418F2576428.5M26153438M30134.544a8F2053.544b"F1703458M29152.5464F491954710F181754810F163294910F177125010F414225110.5F20133529.5M23143.5539F331355411M3143.55511F16227567.5F167165712.5M210235812.5M201385914M15104.56011M28--2176111.5F211116214.5M1682.5*FU* follow-up, *RT* remodeling time The help of DED van der Sluijs, BSc in econometrics, University of Amsterdam, with the analysis is greatly appreciated. Conflict of interest {#FPar1} ==================== Author J. A. van der Sluijs declares that he has no conflict of interest. Author J. L. Bron declare that he has no conflict of interest. Ethical approval {#FPar2} ================ All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent {#FPar3} ================ Informed consent was obtained from all individual participants included in the study.

Cryptic polyadenylation within coding sequences (CDS) or incompletely removed introns produce aberrant transcripts that lack in-frame stop codons[@b1]. Translation of such mRNAs may result in proteins prone to malfunction and deleterious effects on cells[@b2][@b3][@b4]. To mitigate these errors, cells have developed quality-control processes to monitor translating mRNAs and detect aberrant mRNAs, such as those with premature polyA tails within their CDS. Defects in components of the surveillance machineries have been implicated in several types of diseases including neurodegeneration and cancer[@b5][@b6]. The ribosome-associated quality control (RQC) is a mechanism that senses the state of mRNA translation and detects ribosome stalling at the site of defective mRNAs, which results in targeting of both the translating mRNA and nascent peptide for degradation[@b7]. RQC can be divided into several steps, surveillance of the translating mRNA and detection of stalled ribosome, ribosomal subunit dissociation, and degradation of the defective mRNA and nascent peptide. Although the processes of ribosomal subunit dissociation and nascent peptide degradation are well studied[@b8][@b9][@b10][@b11][@b12], the mechanism of surveillance of the translating mRNA and detection of stalled ribosome, in particular the molecular sensors of aberrant mRNAs and their mechanism of action, remain largely unknown. Earlier studies suggested that presence of the polyA sequences within the CDS causes ribosome stalling through interactions between the positively charged peptide (poly-lysine) and the negatively charged exit channel of the ribosome[@b8][@b13][@b14]. However, others showed that at least in mammalian cells RQC at poly-lysine sites is codon-sequence dependent as runs of poly-lysine residues coded by AAA codons induced ribosome stalling much more efficiently than equivalent runs of poly-lysine encoded by AAG codons[@b15]. These results indicate that sensing A-rich mRNA sequence in mammalian cells dominates over general polybasic amino-acid-triggered translational regulation. Nevertheless, the mechanisms by which premature polyA sequences are detected in aberrant mRNAs and the following molecular events leading to ribosome stalling are not known. In yeast, the E3 ubiquitin ligase Hel2 has been implicated in facilitating the earlier steps of RQC at polybasic sequences[@b8]. Notably, Hel2-dependent K63 polyubiquitination is necessary for the initial processes involved in stalled translation surveillance[@b16]. However, the precise functions of Hel2 in detection of stalled ribosomes or its ubiquitination substrates have not been identified. The Zinc Finger Protein 598 (ZNF598) is the human ortholog of Hel2 and contains a RING domain characteristic of E3 ubiquitin ligases and several C2H2-type zinc finger motifs, commonly found in nucleic acid-binding proteins[@b17][@b18]. We previously described ZNF598 protein in a complex with the translation repressor proteins EIF4E2/4EHP and GIGYF2 (ref. [@b19]). Two recent reports showed that ZNF598 is also required for stalling at polyA sequences and linked its E3 ubiquitin ligase activity to translation arrest through ubiquitinating the 40S subunit ribosomal proteins RPS10 and RPS20 (refs [@b20], [@b21]). Here, we reveal that ZNF598 directly binds to the translating mRNA and tRNAs on ribosomes and triggers ribosome stalling and RQC at premature polyA sequences. We further identified RPS3A as an additional substrate of ZNF598 E3 ubiquitin ligase activity, and UBE2D3 as the ZNF598-interacting E2 ubiquitin ligase. Our findings establish a link between the RNA-binding properties and ubiquitin ligase activity of a uniquely conserved protein in monitoring mRNA translation. Results ======= ZNF598 associates with translating ribosomes -------------------------------------------- Human ZNF598 encodes a ubiquitously expressed 904 amino acid (aa) protein containing one *N*-terminal RING domain characteristic of E3 ubiquitin ligases and four *N*-terminal and one *C*-terminal C2H2-type zinc finger motifs ([Fig. 1a](#f1){ref-type="fig"} & [Supplementary Fig. 1](#S1){ref-type="supplementary-material"}). To evaluate its potential role in translational control, we performed polysome profiling using ZNF598 overexpression (ZNF598-OE) or ZNF598 knockout HEK293 cells (ZNF598-KO; [Supplementary Fig. 2](#S1){ref-type="supplementary-material"}). ZNF598-OE induced a shift from heavy polysomes to monosomes and ZNF598-KO induced a shift to heavier polysomes, indicating translational repression ([Fig. 1b,c](#f1){ref-type="fig"}). This effect was independent of EIF4E2 ([Supplementary Fig. 3](#S1){ref-type="supplementary-material"}) and was neither due to general translational repression mediated by phosphorylation of eukaryotic initiation factor 2α EIF2S1/eIF2α ([Fig. 1d](#f1){ref-type="fig"}) under stress condition. Western blot analysis of the polysome fractions showed that a proportion of ZNF598 protein was associated with heavy polysomes in ZNF598-OE cells ([Fig. 1e](#f1){ref-type="fig"}). Size exclusion chromatography also revealed that ZNF598 protein co-fractionated with ribosomes in a ≥2 MDa complex ([Supplementary Fig. 4](#S1){ref-type="supplementary-material"}), suggesting that ZNF598 either interacted transiently with assembled ribosomes or associated with a subset of actively translating ribosomes. Immunofluorescence analysis of ZNF598 upon exposure to arsenite-induced stress did not reveal any change in its cytosolic distribution, unlike many known cytosolic RNA-binding proteins or 18S ribosomal RNA which accumulate in stress granules[@b22][@b23] ([Supplementary Fig. 5](#S1){ref-type="supplementary-material"}). Together, these observations support a role of ZNF598 in translation. ZNF598 binds to RNAs associated with translating ribosomes ---------------------------------------------------------- To investigate ZNF598 function and identify its target RNAs, we performed 4-thiouridine (4SU) photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation (PAR-CLIP)[@b24] in ZNF598-OE HEK293 cells ([Supplementary Figs. 2a and 6a](#S1){ref-type="supplementary-material"}). We observed that ZNF598 cross-linked to tRNAs, mRNAs and rRNAs ([Fig. 2a](#f2){ref-type="fig"}, [Supplementary Fig. 6b--g](#S1){ref-type="supplementary-material"} & [Supplementary Data 1](#S1){ref-type="supplementary-material"}). The average ratio of cross-linked reads annotated as tRNAs, mRNAs and rRNAs was ∼4:2:1. Cross-linked reads derived from mRNAs showed evenly distributed enrichment for CDS over untranslated regions (UTRs) ([Fig. 2a,b](#f2){ref-type="fig"}). The cross-linked mRNA read abundance resembled the overall mRNA abundance in HEK293 cells as determined by polyA mRNA-Seq ([Supplementary Fig. 7](#S1){ref-type="supplementary-material"}). The reads mapped to nuclear encoded cytoplasmic tRNAs originated predominantly from their 5′ halves, which were cross-linked to ZNF598 via their D-loops ([Fig. 2c,d](#f2){ref-type="fig"}). Although ZNF598 protein cross-linked to every cytoplasmic tRNA, cross-linked reads unique to tRNA^Lys^(UUU) were ∼10-fold enriched relative to their total cellular abundance, whereas cross-linked reads to tRNA^Lys^(CUU) only displayed an average twofold enrichment ([Fig. 2e](#f2){ref-type="fig"} & [Supplementary Data 2](#S1){ref-type="supplementary-material"}). Cross-linked reads to rRNA ([Fig. 2f](#f2){ref-type="fig"}) originated predominantly from 5S (pos. 96--121) and 18S rRNAs (pos. 686--707 and 745--778) and to a lesser extent from 5.8S and 28S ([Supplementary Figs. 8 and 9](#S1){ref-type="supplementary-material"}). Taken together, the cross-linked RNA targets and positional cross-linking spectra indicate that a fraction of ZNF598 protein was bound to translating ribosomes. In light of the enrichment of cross-linking to (AAA)-decoding tRNA^Lys^(UUU), we hypothesized that ZNF598 may have a role in detecting premature polyA tails or protein-folding problems of poly-lysine rich proteins that would induce ribosome stalling and RQC pathway[@b7][@b13]. The transcript abundance of ZNF598 is ∼10-fold below the average of transcripts encoding ribosomal proteins but similar in abundance to other proteins implicated in RQC ([Supplementary Data 3](#S1){ref-type="supplementary-material"}). Considering that direct molecular contacts are required for photo-cross-linking of ZNF598 to the CDS of mRNAs, tRNAs and rRNAs, we conclude that ZNF598 protein is associated with translating ribosomes and intimately monitors the CDS of translated mRNAs and/or identity of tRNAs occupying the ribosome and triggering RQC upon encounter with premature polyA tails. ZNF598 initiates RQC at premature polyA sequences ------------------------------------------------- In yeast, Hel2 facilitates ribosome stalling at both poly-lysine and poly-arginine polybasic amino-acid coding sites[@b8][@b16][@b25][@b26]. Polybasic peptides are lysine- (AAA or AAG codons) and/or arginine-rich (CGU, CGC, CGA, CGG, AGG or AGA codons). To determine whether the amino acid or the mRNA sequence is responsible for ribosome stalling, we generated stable HEK293 reporter cell lines for parental (CTR) and ZNF598-KO expressing GFP and mCherry fusion proteins separated by a 12 aa spacer composed of various lysine, arginine or control threonine-serine codon repeats ([Fig. 3a](#f3){ref-type="fig"}). We observed fourfold decreased expression of the fusion protein separated by a (AAA)~12~ poly-lysine-coding sequence as compared with either control poly-threonine-serine-coding (ACT AGC)~6~, poly-lysine-coding (AAG)~12~, poly-arginine (AGA)~12~, or poly-arginine-coding combinations of CGC, CGA and CGG. Importantly, expression of the (AAA)~12~ reporter reverted to the level of the (ACT AGC)~6~ control in ZNF598-KO cells ([Fig. 3b](#f3){ref-type="fig"} & [Supplementary Fig. 10b](#S1){ref-type="supplementary-material"} left panels), demonstrating ZNF598-dependent repression of the (AAA)~12~ reporter. The expression of the reporter fusion proteins was also studied in ZNF598-OE and its empty vector control (EV) using transient reporter plasmid transfection, revealing further decreased (AAA)~12~ reporter expression as compared with the (ACT AGC)~6~ control ([Fig. 3b](#f3){ref-type="fig"} & [Supplementary Fig. 10b](#S1){ref-type="supplementary-material"} right panels). The increased repression of the (AAA)~12~ reporter in ZNF598-OE cells also suggests that ZNF598 protein is sub-stoichiometric to ribosomes at standard conditions. Translational repression of the (AAA)~12~ reporter was not associated with an imbalance of the GFP to mCherry signal ratio ([Fig. 3b](#f3){ref-type="fig"}) or accumulation of truncated fusion protein ([Supplementary Fig. 10b](#S1){ref-type="supplementary-material"}), which is consistent with activation of the RQC pathway and destruction of the nascent reporter GFP segment[@b8]. These results demonstrate that whereas both poly-lysine and poly-arginine sequences induce ribosome stalling and RQC in yeast[@b8][@b16][@b26], only poly-lysine, encoded by the AAA codon induces RQC in mammalian cells in a ZNF598-dependent manner. Therefore, sensing A-rich mRNA sequence in mammalian cells dominates over general polybasic amino-acid-triggered translational regulation, as reported previously[@b15]. Interestingly, although there are 7,433 (AAA)~2~ and 175 (AAA)~3~ consecutive codons in human CDS, multiple consecutive (AAA) codons are scarce and do not exceed more than four ([Supplementary Fig. 11](#S1){ref-type="supplementary-material"} and [Supplementary Data 4](#S1){ref-type="supplementary-material"}). This implies that ZNF598 protein may be directly involved in detecting aberrant CDS containing ≥12 nt polyA during translation, and/or senses the simultaneous occupation of the ribosomal A, P and E sites with tRNA^Lys^(UUU), and/or a translating ribosome conformational transition as a consequence of the above. The multiple C2H2-type zinc fingers positioned along the protein may be implicated in such function. The *N*-terminal RING domain is critical for ZNF598 function ------------------------------------------------------------ To identify the protein domains of ZNF598 critical for triggering premature polyA-dependent ribosome stalling and RQC, we complemented the ZNF598-KO cells with full-length or a series of truncated mutants of ZNF598 ([Fig. 3c](#f3){ref-type="fig"}). Although the *C*-terminal C2H2 domain (aa 864--904) was dispensable for RQC of the reporter without compromising RNA cross-linking, the *N*-terminal RING domain (lacking in the 81--904 truncation) was required for RQC ([Fig. 3d](#f3){ref-type="fig"}), albeit RNA cross-linking and thus translating ribosome binding was unaffected by deletion of the RING domain ([Fig. 3e](#f3){ref-type="fig"} & [Supplementary Fig. 12](#S1){ref-type="supplementary-material"}). These functional observations were corroborated by polysome profiling, which show the requirement of the RING domain and unstructured central domain, but not the *C*-terminal C2H2 motif for translational repression of the endogenous mRNAs ([Supplementary Fig. 13](#S1){ref-type="supplementary-material"}). Ubiquitination of ribosomal proteins is required for RQC -------------------------------------------------------- RING finger proteins coordinate the transfer of ubiquitin to substrate proteins by the recruitment of E2 ubiquitin ligases[@b27]. Although Hel2-dependent ubiquitination has been implicated in the initial surveillance process of stalled ribosomes in yeast[@b16], the target substrates of Hel2 or its corresponding E2 ubiquitin ligase were not known. The *N*-terminal domains of Hel2, including the RING domain, are the most conserved in ZNF598 ([Supplementary Fig. 1](#S1){ref-type="supplementary-material"}), suggesting a similar E3 ligase activity for ZNF598. To identify proteins targeted by ZNF598-dependent ubiquitination, we used ubiquitin remnant immuno-affinity profiling[@b28] in control, ZNF598-OE and ZNF598-KO HEK293 cells. Three small ribosomal subunit proteins RPS3A (eS1), RPS10 (eS10) and RPS20 (uS10), and the heat-shock protein HSPH1 were enriched more than fourfold in ZNF598-OE cells and decreased more than fourfold in ZNF598-KO cells as compared with CTR cells ([Fig. 4a](#f4){ref-type="fig"} and [Supplementary Data 5](#S1){ref-type="supplementary-material"}). In addition to priming the target proteins to proteasome degradation, ubiquitination also has a regulatory function in diverse molecular pathways such as DNA repair[@b29], anti-viral immunity[@b30] and signal transduction[@b31]. Evolutionarily conserved, regulatory ubiquitination of the small ribosomal subunit proteins induced by inhibitors of translation elongation has previously been reported in mammalian cells[@b32]. Western blot analysis revealed that overexpression or depletion of ZNF598 failed to affect the expression of RPS3A, RPS10 and RPS20 proteins ([Supplementary Fig. 14](#S1){ref-type="supplementary-material"}), indicating that the ZNF598-regulated ubiquitination of small ribosomal subunit proteins is likely a signalling event rather than inducing protein degradation by the proteasome. Further biochemical experiments were performed to investigate the functional consequence of ZNF598-mediated ubiquitination of RPS3A, RPS10 and RPS20. We generated stable cell lines expressing the *C*-terminus Myc-DDK-tagged wild-type or mutant RPS3A, RPS20 or RPS10 in which lysines that are subject to ubiquitination were substituted by arginine. For RPS10 and RPS20, where multiple lysines were ubiquitinated (RPS10; K138, K139 and RPS20; K4, K8), we also created double mutants. Western blot analysis confirmed that the mutant proteins were at least as abundantly expressed as the wild-type proteins ([Supplementary Fig. 15a--c](#S1){ref-type="supplementary-material"}). However, only RPS10 K138R, and the double mutant K138R/K139R, partially impaired RQC for the (AAA)~12~ reporter, whereas mutations in RPS20 and RPS3A failed to effect (AAA)~12~ reporter expression ([Supplementary Fig. 15d](#S1){ref-type="supplementary-material"}). This indicates that downstream ribosomal protein ubiquitination of RPS10 contributes to RQC. Recent reports also identified ZNF598-dependent ubiquitination of ribosomal proteins RPS10 and RPS20 as ZNF598-stimulated ubiquitin conjugates; although these studies differ with respect to the relative importance of the two proteins as well as their ubiquitination sites[@b20][@b21]. Both studies showed the requirement for ubiquitination of RPS10 in polyA-induced ribosome stalling. Nevertheless, while Juszkiewicz and Hegde[@b21] observed only a partial effect in RPS20 K4R/K8R double mutant, Sundaramoorthy, *et al*.[@b20] observed that both RPS20 and RPS10 mutants displayed enhanced readthrough of the polyA sequence. We detected RPS3A as an additional target of ZNF598-regulated ubiquitination, which is surprising considering its distance from RPS10 and RPS20 in the ribosome[@b33][@b34]. However, as mutations in RPS3A did not impact the polyA-induced ribosome stalling, it is likely that ubiquitination of this protein occurs or has a role in downstream events such as during ribosomal subunit dissociation. RING family E3 ubiquitin ligases catalyse the transfer of ubiquitin from an E2 enzyme to target proteins[@b35]. To identify the E2 ligase for ZNF598 we performed immunoprecipitation (IP) in ZNF598-OE HEK293 cells followed by mass spectrometry (IP/MS) analysis of the immunoprecipitate. We identified the 97% identical UBE2D2 and UBE2D3 E2 ligases among the most significant hits ([Fig. 4b](#f4){ref-type="fig"}, [Supplementary Figs. 16 and 17a](#S1){ref-type="supplementary-material"} & [Supplementary Data 6](#S1){ref-type="supplementary-material"}). Their interactions with ZNF598 were verified by IP and western blot analysis, which revealed that UBE2D3 (the most abundant homologue in HEK293 cells; [Supplementary Fig. 17b](#S1){ref-type="supplementary-material"}), but not UBE2D2, specifically interacted with ZNF598 ([Supplementary Fig. 17c,d](#S1){ref-type="supplementary-material"}). UBE2D3 also co-fractionated with ZNF598 and ribosomal proteins in ≥2 MDa complexes ([Supplementary Fig. 18](#S1){ref-type="supplementary-material"}), but failed to co-fractionate in heavy complexes with the ΔRING mutant (aa 81--904) of ZNF598 ([Supplementary Fig. 12](#S1){ref-type="supplementary-material"}). In addition, knockdown of UBE2D3, but not UBE2D2, partially abrogated RQC on the (AAA)~12~ reporter ([Fig. 4c,d](#f4){ref-type="fig"}) or translational repression of endogenous mRNAs in ZNF598-OE cells ([Supplementary Fig. 19](#S1){ref-type="supplementary-material"}). These results underscore the requirement of the E3 ligase activity of ZNF598 and highlight the regulatory ubiquitination of the ribosomal proteins RPS3A, RPS10 and RPS20 in ribosome stalling and RQC. Although ZNF598 was originally identified as a component of the ZNF598/GIGYF2/EIF4E2 complex, we showed that ZNF598-dependent translational repression was independent of the cap-binding EIF4E2 protein. Considering that EIF4E2 represses mRNA translation[@b19], we propose that this complex may have a role in downstream degradation of premature polyadenylated mRNAs via displacement of the canonical cap-binding protein EIF4E by the repressive EIF4E2 homolog. The sub-stoichiometric abundance of ZNF598 as well as other proteins implicated in RQC sensing deleterious amino-acid repeats, premature stop codons or truncated non-polyadenylated mRNAs, suggest that subpopulations of error-sensing ribosomes distributed randomly among translating polysomes divide the labour of detecting faulty mRNAs. The study of molecular and functional ribosome heterogeneity[@b36] will provide further direction in elucidating mechanisms of RQC. Although the human genome encodes 864 proteins with RING domain and 2,474 proteins containing at least one C2H2 domain, only three additional proteins, TRIM23, ZNF645 and CBLL1, carry a combination of RING and C2H2 domains (<http://www.ebi.ac.uk/interpro/protein/>), but none is as strongly evolutionarily conserved as ZNF598 ([Supplementary Fig. 20](#S1){ref-type="supplementary-material"}). In as much these proteins contribute to other possibly less conserved ubiquitination-dependent RNA pathways outside of translation, it poses an intriguing question. In summary, we showed that ribosome stalling and RQC in mammalian cells at premature polyA-containing mRNAs involved recognition of the tRNA^Lys^(UUU) and/or the mRNA (AAA) codon repeats by the unique RNA-binding and E3 ligase protein ZNF598 (see model; [Fig. 4e](#f4){ref-type="fig"}). Methods ======= Cell lines and culture conditions --------------------------------- Flp-In T-REx 293 cells (Thermo Fisher Scientific, R78007) were grown in high glucose Dulbecco\'s Modified Eagle\'s Medium (DMEM) (Thermo Fisher Scientific, 11965118) supplemented with 10% v/v fetal bovine serum (FBS), 100 U ml^−1^ penicillin, 100 μg ml^−1^ streptomycin, 2 mM [L]{.smallcaps}-glutamine, 100 μg ml^−1^ zeocin and 15 μg ml^−1^ blasticidin. Presence of mycoplasma contamination in cells was tested by mRNA-Seq. Cell lines inducibly expressing 3xFlag-ZNF598 or Flag/HA-tagged truncated ZNF598 variants were generated as described previously[@b37] and selected and maintained in media supplemented with 100 μg ml^−1^ hygromycin. Expression of tagged proteins was induced for 24 h by the addition of doxycycline at 1 μg ml^−1^ final concentration. In the text, figures and tables, parental Flp-In T-REx 293 cells are labelled as CTR and the CRISPR-Cas9-mediated knockout as ZNF598-KO. 3xFlag-ZNF598 cells are labelled as ZNF598-OE. Because they were grown in different media than the CTR and ZNF598-KO cells, ZNF598-OE cells were always compared with control cells expressing 3xFlag EV, labelled as EV. For experiments with truncated variants, cells expressing full-length or truncated variants of ZNF598 were established using ZNF598-KO cells. Antibodies and RNA interferences -------------------------------- The following antibodies were used: mouse anti-β-tubulin (Sigma, T4026; 1:5,000 dilution), mouse anti-α-tubulin (Santa Cruz, sc-23948; 1:1,000 dilution), mouse anti-β-actin (Sigma, A5441; 1:1,000 dilution), mouse anti-Flag M2 (Sigma, F3165; 1:2,000 dilution), mouse anti-RPS6 (Cell Signaling, C-8; 1:2,000 dilution), rabbit anti-RPS3A (Abcam, ab171742; 1: 2,000 dilution), rabbit anti-RPS10 (Abcam, ab151550; 1: 2,000 dilution), rabbit anti-RPS20 (Abcam, ab133776; 1: 2,000 dilution), rabbit anti-Rack1 (Cell Signaling, 4716; 1:2,000 dilution), rabbit anti-EEF2 (Cell Signaling, 2332; 1:1,000 dilution), rabbit anti-4EHP (GeneTex, GTX103977; 1:500 dilution), mouse anti-HA (Fisher; 50-103-0108; 1:2,000 dilution), rabbit anti-UBE2D2 (Abcam, ab155088; 1:2,000 dilution), mouse anti-UBE2D3 (Abcam, ab58251; 1:2,000 dilution), mouse anti-GFP (Clontech, 632375; 1:1,000 dilution), rabbit anti-ZNF598 (Abcam, ab135921; 1:500 dilution), rabbit anti-ZNF598 (a gift from Jianxin Xie at Cell Signaling Technology; 1:1,000 dilution), rabbit anti-ZNF598 (GeneTex, GTX119245; 1:500 dilution). For ZNF598 the Cell Signaling antibody was used in most experiments, unless stated otherwise. Uncropped images of all of the immune-blots in this manuscript are shown in [Supplementary Figs 21 and 22](#S1){ref-type="supplementary-material"}. The following siRNA and shRNAs were used: ON-TARGETplus Non-targeting Control Pool (Dharmacon, D-001810-10-05), UBE2D2 siRNA SMARTpool (Dharmacon, L-010383-00-0005), UBE2D3 siRNA SMARTpool (Dharmacon, L-008478-00-0005), Non-Targeting shRNA Controls (Sigma, SHC002) and EIF4E2 shRNA (Sigma, TRCN0000152006). Lentivirus production --------------------- A total of 8 × 10^6^ 293FT (Thermo Fisher Scientific, R70007) cells were cultured in a 10-cm dish for 24 h in high-glucose DMEM supplemented with 10% v/v FBS. Medium was replaced by OptiMEM (Thermo Fisher Scientific, 51985091) 30 min before transfection. Lentivirus particles were produced by transfecting the 293FT cells using Lipofectamine 2000 (Thermo Fisher Scientific, 11668019) and 10 μg shRNA plasmid, 6.5 μg psPAX2 (Addgene, plasmid 12260) and 3.5 μg pMD2.G (Addgene, plasmid 12259) packaging plasmids. 5 h post transfection, the medium was replaced with fresh high-glucose DMEM supplemented with 10% v/v FBS. Supernatant was collected at 48 h post transfection, replaced with fresh medium and harvested again after 24 h. Viral particles were cleared by filtration (45 μm; Fisher Scientific, 09-720-005) and virus titre was measured by colony formation assay using 293FT cells. The multiplicity of infection was adjusted to ∼5. Virus solution was stored at −80 °C without cryopreservative in 1 ml aliquots or used to infect the cells directly in the presence of 6 μg ml^−1^ polybrene (Sigma, H9268). Generation of ZNF598 knockout cell lines ---------------------------------------- CRISPR-Cas9-mediated genome editing was performed as described previously[@b38]. We designed three small guide RNAs (sgRNAs) cognate to the coding region of ZNF598 gene: 5′-CTACTGCGCCGTGTGCCGCG, 5′-GAAAGGTGTACGCATTGTAC, and 5′-TACGCATTGTACAGGTGAGC. The following primers were used for the PCR-genotyping: sense primer1, 5′-GGAGGCGGAGGCGGCGGCAGC; anti-sense primer1, 5′-CCCCGCCCTGGGTGGCCCCACC; sense primer2, 5′- GGGGGTCCCATCCCAGTCCTGC; anti-sense primer2, 5′- CCTGGCCCCAGCATTGGTGCACC. PCR products were cloned using the Zero Blunt PCR Cloning Kit (Thermo Fisher Scientific, K270040) and 11 clones were sequenced per cell line to verify successful genome editing. Plasmid construction -------------------- For overexpression, ZNF598 cDNA was cloned into pcDNA5/FRT/TO/3xFlag/N-term plasmid (a gift from Dr Jernej Ule) by digestion with *Kpn*I and *Not*I and subsequent use of T4 DNA ligase. pENTR4 (Thermo Fisher Scientific, A10465) plasmids for expression of ZNF598 full-length or truncated variants were generated by PCR amplification of the respective sequences adding attB1 and attB2 recombination sites to the primers followed by recombination using the Gateway BP recombinase (Thermo Fisher Scientific, 11789020). pENTR4 plasmids were recombined into the pFRT/TO/Flag/HA-DEST destination vector (Thermo Fisher Scientific) using the GATEWAY LR recombinase (Thermo Fisher Scientific, 11791020). Reporters expressing GFP, mCherry, or GFP-mCherry fusion proteins separated by 12 amino-acid spacers were cloned into pcDNA5/FRT (Thermo Fisher Scientific, V6010-20) by digestion with BamHI and NotI and subsequent use of T4 DNA ligase. For expression of ribosomal proteins, wild-type cDNAs were cloned into pLenti-C-Myc-DDK-IRES-Puro (Origene) plasmid by digestion with AscI and MluI. Further mutagenesis was performed using QuikChange Lightning Multi Site-Directed Mutagenesis Kit (Agilent). Fluorescence-activated cell-sorting analysis (FACS) --------------------------------------------------- CTR and ZNF598-KO cells with stable, and EV and ZNF598-OE cells with transient, expression of GFP, mCherry and GFP-mCherry fusion proteins separated by 12 amino-acid spacers were sorted and analysed by flow cytometry (LSR II, BD Biosciences). Polysome profiling ------------------ A total of 20 × 10^6^ cells in a 15-cm plate were pretreated with cycloheximide (100 μg ml^−1^; BioShop Canada, CYC003) for 5 min, collected by centrifugation at 4 °C for 5 min and lysed in 500 μl hypotonic buffer containing 5 mM Tris-HCl, pH 7.5, 2.5 mM MgCl~2~, 1.5 mM KCl, complete EDTA-free protease inhibitor cocktail (Roche, 04693159001), 100 μg ml^−1^ cycloheximide, 2 mM DTT, 200 U ml^−1^ RNasin (Promega, N2111), 0.5% v/w Triton X-100, and 0.5% v/w sodium deoxycholate using 1.5 ml microtubes. The lysates were cleared by centrifugation at 20,000 × g for 5 min at 4 °C. Total RNA concentration in the supernatant was measured by NanoDrop 2000 (Thermo Fisher Scientific) at 254 nm supernatant, and the equivalent of 300 μg of RNA was diluted to a final 500 μl volume and separated on 12 ml of 10--50% sucrose gradient by ultracentrifugation at 230,500 × g for 2 h in an SW40 rotor (Beckman Coulter) at 4 °C. Fractions of 700 μl were collected using an ISCO gradient fractionation system and the OD254 was continuously recorded with a Foxy JR Fractionator (Teledyne ISCO) during the collection process. Size exclusion chromatography ----------------------------- HEK293 cells overexpressing 3xFlag-ZNF598 (20 × 10^6^) were collected by centrifugation and resuspended in 600 μl of lysis buffer containing 25 mM HEPES-KOH, pH 7.4, 150 mM KCl, 75 mM KOAc, 2 mM MgCl~2~, 0.5% NP40, supplemented with complete EDTA-free protease inhibitor cocktail and 1 mM NaF, 1 mM Na~3~VO~4~ and 1 mM β-glycerophosphate phosphatase inhibitor. The lysate was clarified by centrifugation at 15,000 × g for 10 min at 4 °C. 5 mg of the protein extract was brought to a total volume of 500 μl of lysis buffer (final concentration 10 μg μl^−1^) and directly loaded onto a 24 ml Superose 6 column (HR 10/300, GE Healthcare Life Sciences) pre-equilibrated with lysis buffer and run in the same buffer at a flow rate of 0.5 ml min^-1^. Molecular mass calibration was carried out by using the Gel Filtration HMW Calibration Kit (Healthcare Life Sciences, 28-4038-42). Fluorescence microscopy ----------------------- Immunofluorescence and RNA-FISH experiments were performed following the protocol described previously[@b23]. For stress granule assays, HEK293 cells expressing 3xFlag-ZNF598 were grown on chamber slides. Arsenite was added to the cells at a final concentration of 400 μM and incubation was continued for 30 min at 37 °C. Chamber slides were hybridized overnight at 40 °C in hybridisation buffer containing 20 nM LNA-modified oligoT probe labelled with ATTO647N for detection of polyA and a cocktail of four different anti-28S rRNA LNA-modified oligodeoxynucleotides labelled with ATTO550 at 10 nM each for detection of 28S rRNA[@b23]. The slides were subsequently incubated for 1 h with anti-Flag antibody (Sigma, F3165) followed by incubation with a DAPI and Alexa Fluor 488-labelled goat anti-mouse IgG H+L (Thermo Fisher Scientific, A11001) for 1 h at RT. Images were recorded on the Olympus VS110 and processed using Visiopharm Integrated Systems Inc. software. PAR-CLIP -------- PAR-CLIP was performed using a single RNase A digestion step and using anti-Flag-M2 magnetic beads (Sigma, M8823) for IP as described previously[@b39]. PAR-CLIP cDNA libraries were sequenced on an Illumina HiSeq 2500 instrument, and data were analysed using the PAR-CLIP suite[@b39]. Reads mapping to mRNAs with d1 T-to-C sequence transitions and ≥20 nt were extracted and mapped to human genome using STAR-2.5.2a. For the metagene analysis, gencode gtf file (gencode.v19.chr_patch_hapl_scaff.annotation.gtf) was used to calculate the logarithm of average read coverage for 3′ UTR, CDS and 5′ UTR for the 2,000 mRNAs with highest coverage. Relative UTR and CDS sizes were calculated based in their average size in all mRNAs expressed in HEK293. Average coverage was calculated for each UTR and CDS independently according to the actual length of the region and number of bins (10 for 5′ UTR, 60 for CDS and 20 for 3′ UTR) and additionally represented in the upper panel as percent of total gene coverage. Reads mapping to tRNAs with d1 T-to-C sequence transitions and ≥20 nt were extracted and compared with tRNA sequencing (Gogakos, T & Tuschl, T, Characterizing expression and processing of precursor and mature human tRNAs by hydro-tRNAseq and PAR-CLIP, manuscript in preparation) data set obtained by hydro-tRNAseq[@b40] for differential expression. The analysis was conducted with the R/Bioconductor package edgeR[@b41] v. 3.14.0. The read counts were normalized using the weighted trimmed mean of M values[@b42] and normalized for library size. The differences were tested using the Fisher\'s exact test and the read count variation was estimated using tagwise or common dispersion. Differences were considered significant below a false discovery rate of 5%. Reads mapping to rRNAs with d0, d1 T-to-C and d1 other and ≥15 nt were mapped independently using bowtie2 (ref. [@b43]) to rRNA sequences, and bedtools[@b44] were used to obtain the coverage across each rRNA. polyA mRNA-sequencing --------------------- Oligo(dT)-selected RNA was converted into cDNA for polyA mRNA-sequencing using the Illumina TruSeq RNA Sample Preparation Kit v2 according to the instructions of the manufacturer and sequenced on an Illumina HiSeq 2500 platform using 100 nt single-end sequencing. Co-IP and mass spectrometry (IP/MS) ----------------------------------- Cell lines expressing 3xFlag and 3xFlag-ZNF598 from one 15-cm plate (∼20 × 10^6^) were lysed in NP40 lysis buffer composed of 50 mM HEPES-KOH, pH 7.5, 150 mM KCl, 2 mM MgCl~2~, 2% v/v NP40, 0.5 mM DTT, 1 × complete EDTA-free protease inhibitor cocktail and 1 × PhosStop (Roche, 04906837001) and the Flag-tagged protein was immunoprecipitated from the lysate with anti-FLAG-M2 magnetic beads. Magnetic beads were washed three times with washing buffer composed of 50 mM HEPES-KOH, pH 7.5, 300 mM KCl, 2 mM MgCl~2~, 0.5% v/v NP40 and three times with high-salt washing buffer composed of 50 mM HEPES-KOH, pH 7.5, 500 mM KCl, 2 mM MgCl~2~, 0.5% v/v NP40. The immunoprecipitates were eluted with Flag peptide and separated on NuPAGE Novex 4--12% Bis-Tris protein gels (Thermo Fisher Scientific, NP0322BOX), which were run shortly with the dye front 12 mm from the loading pocket. The experiment was performed for three biological replicates per condition. The protein bands were excised, carefully washed and followed by reduction (DTT final concentration of 5 mM for 30 min at 55 °C) and alkylation (iodoacetamide final concentration of 2 μg ml^−1^ for 15 min at room temperature). Proteins were digested overnight with Endopeptidase Lys-C (Wako) and trypsin (Sequencing Grade, Promega). Peptides were extracted, desalted[@b45] and analysed by nano LC-MS/MS (Dionex Ultimate 3000 coupled to a Q-Exactive Plus, Thermo Fisher Scientific). Data were processed using MaxQuant v. 1.5.3.28 (ref. [@b46]). Proteins quantitated in two out of three biological replicates for at least one condition and with Welch's *t*-test false discovery rate \<5% were analysed using the statistical software Perseus[@b47]. Co-IP assays ------------ Parental Flp-In T-REx 293 cells (6 × 10^6^ cells grown in a 10-cm plate) were transiently transfected with pcDNA5.A-3xFlag-ZNF598 and control EV (3xFlag) using Lipofectamine 2000 according to the manufacturer's instructions. After 24 h, cells were washed twice with cold PBS and collected in 1 ml of cold lysis buffer composed of 40 mM HEPES-KOH, pH 7.5, 0.3% CHAPS, 120 mM NaCl, 1 mM EDTA supplemented with complete EDTA-free protease inhibitor cocktail, and RNase A (10 μg/ml), and incubated for 30 min on ice. Lysates were cleared by centrifugation at 20,000 × g for 15 min at 4 °C. The protein concentration of lysates was quantified using the Bradford assay and 1 mg of lysate protein was used for IP. Before IP, the lysates were cleared again by incubating with 50 μl of 50% protein G agarose fast flow beads (EMD Millipore, 16--266) for 2 h at 4 °C with gentle agitation. The pre-cleared lysates were centrifuged at 3,000 × g for 1 min at 4 °C and the supernatant was incubated with 1 μl of anti-Flag M2 antibody in 1 ml total volume on an end-over-end rotator for 3 h at 4 °C. Subsequently, 50 μl of 50% protein G agarose beads were added to the lysate/antibody and the mixture was incubated at 4 °C overnight on an end-over-end rotator. Beads were washed three times with 500 μl of lysis buffer and the immunoprecipitated complex was eluted from the beads by boiling in 30 μl of 2 × SDS loading buffer composed of 100 mM Tris-HCl, pH 6.8, 4% w/v SDS, 0.2% bromophenol blue, 20% v/v glycerol and 200 mM DTT for 10 min at 65 °C. Ubiquitin remnant immunoaffinity profiling ------------------------------------------ Proteome-wide ubiquitination sites were identified by using the PTMScan Ubiquitin Remnant Motif (K-ε-GG) Kit (Cell Signaling Technologies, 5562). Ubiquitinated peptides were enriched and identified by immunoprecipitation using a bead-conjugated monoclonal antibody generated against the sequence CXXXXXXK-ε-GGXXXXXX, where X is any amino-acid except cysteine and tryptophan, and K-ε-GG is a di-glycine moiety bound to the ε amino group of a lysine residue. Enrichment of ubiquitinated peptides in conjunction with liquid chromatography (LC) tandem mass spectrometry (MS/MS; EasyLC 1200 coupled to a Fusion Lumos operated in HCD high/high mode, Thermo Fisher Scientific) was performed according to the manufacturer's instructions. In brief, for each experiment ∼1--2 × 10^8^ cells were grown to 90% confluency. Cells were harvested, washed with 1 × PBS, and lysed in 10 ml of freshly prepared urea lysis buffer (20 mM HEPES-KOH, pH 8.0, 9 M urea, 1 mM Na~3~VO~4~, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM iodoacetamide). Using a microtip, the cell lysate was sonicated by three 15 s bursts at 15 W. The lysate was cleared by centrifugation for 15 min at 20,000 × g at room temperature. The cleared supernatant contained about 20 mg of total protein and was subjected to reduction (DTT final concentration of 5 mM for 30 min at 55 °C) and alkylation (iodoacetamide final concentration of 2 μg ml^−1^ for 15 min at room temperature) followed by overnight trypsinization (TPCK Treated, Worthington Biochemical Corporation). Peptides were desalted and purified using a Sep-Pak C18 column (WAT051910, Waters Corporation). Peptides were eluted with 20 ml Solvent B (0.1% TFA, 40% acetonitrile), frozen in liquid nitrogen, and lyophilized for 2 days to assure complete TFA removal. For immunoaffinity purification, the lyophilized peptides were dissolved in 1.4 ml IAP buffer containing 50 mM MOPS-NaOH, pH 7.2, 10 mM Na~2~HPO~4~, 50 mM NaCl. The peptide solution was added to a microtube of the anti-K-ε-GG motif antibody beads and immunoprecipitated on a rotator for 2 h at 4 °C. After IP, the bead slurry was centrifuged at 2,000 × g for 30 s and the supernatant was removed. The beads were washed three times with 1 ml of IAP buffer followed by two washes with 1 ml 1 × PBS. Peptides were eluted in 100 μl of 0.15% TFA, concentrated and desalted by stage tip chromatography, and subjected to nano LC-MS/MS analysis in technical replicate. LC-MS/MS data were queried against a Uniprots Human database (March 2016) using MaxQuant v. 1.5.3.28. Sequence alignment and domain analysis -------------------------------------- Amino-acid sequences were obtained from the NCBI database. ZNF598 proteins: human, NP_835461.2; mouse, NP_898972.1; *Xenopus*, NP_001119543.1; *Drosophila*, NP_611932.2; *Arabidopsis*, NP_566094.2: *S. cerevisiae*, NP_010552.3. UBE2D family proteins: human UBE2D1 protein, NP_003329.1; human UBE2D2, NP_003330.1; human UBE2D3, NP_871621.1; human UBE2D4, NP_057067.1. Trim23 proteins: human, NP_001647.1; mouse, NP_109656.1; *Xenopus*, XP_002934252.2. CBLL1 proteins: human, NP_079090.2; human ZNF645, NP_689790.1; mouse, NP_001240776.1; *Xenopus*, NP_001123714.1; *Drosophila*, NP_001260593.1. Protein sequence alignments were carried out using PRALINE[@b48][@b49] (<http://www.ibi.vu.nl/programs/pralinewww/>). Conservation analyses were carried out using Clustal Omega (<http://www.clustal.org/omega/>) in combination with Trex Newick Viewer (<http://www.trex.uqam.ca/index.php?action=newick&project=trex>) to obtain the radial tree. Domains analyses were performed via Interpro (≤<http://www.ebi.ac.uk/interpro/>). Disorder profile of proteins was mapped via DISOPRED3 (ref. [@b50]). Data availability ----------------- The NCBI SRA accession numbers for the sequencing data reported in this paper are SRR5137268 and SRR5137269 for ZNF598 PAR-CLIP experiments and SRR5137268 for HEK293 polyA mRNA-Seq. [Supplementary Data](#S1){ref-type="supplementary-material"} are also available on <https://rnaworld.rockefeller.edu/ZNF598>. The additional data that support the findings of this study are available from the corresponding author upon request. Additional information ====================== **How to cite this article:** Garzia, A. *et al*. The E3 ubiquitin ligase and RNA-binding protein ZNF598 orchestrates ribosome quality control of premature polyadenylated mRNAs. *Nat. Commun.* **8**, 16056 doi: 10.1038/ncomms16056 (2017). **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Material {#S1} ====================== ###### Supplementary Information ###### Supplementary Data 1 ###### Supplementary Data 2 ###### Supplementary Data 3 ###### Supplementary Data 4 ###### Supplementary Data 5 ###### Supplementary Data 6 The authors thank Edna Matta-Camacho, Pudchalaluck Panichnantakul, Connie Zhao (The Rockefeller University Genomics Resource Center) and Svetlana Mazel (The Rockefeller University Flow Cytometry Resource Center) for technical assistance, Witold Filipowicz (Friedrich Miescher Institute) for critical reading of the manuscript and Wen Li and Joachim Frank (Columbia University) for valuable discussions. This work was supported by CIHR MOP-7214 (N.S.) and NIH R01GM104962 (T.T.). A.G. held a postdoctoral fellowship from the Basque Government. S.M.J. is a recipient of CIHR postdoctoral Fellowship. C.M. was supported by the German Academic Exchange Service (DAAD). M.A. is a recipient of McGill Graduate Excellence Fellowship. C.C. is supported by FRQS and FRM postdoctoral fellowships. The Sohn Conferences Foundation and the Leona M. and Harry B. Helmsley Charitable Trust are acknowledged for mass spectrometer instrumentations. T.T. is cofounder and advisor to Alnylam Pharmaceuticals. The remaining authors declare no competing financial interests. **Author contributions** A.G. and S.M.J. designed, performed experiments, analysed data and wrote the manuscript; C.M. performed the microscopy and ubiquitin remnant immunoaffinity profiling; C.C. performed size exclusion chromatography; T.G. contributed to tRNA and codon bias analysis; P.M. developed the bioinformatics tools for metagene analysis; M.A. performed some co-immunoprecipitation assays; M.S. provided unpublished reagents; H.M. performed mass spectrometry analysis; T.T. and N.S. provided general oversight, analysed data, and edited the manuscript. {#f1} {#f2} ![The RING domain of ZNF598 is essential for ribosome stalling at polyA residing within coding sequences.\ (**a**) Schematic diagram of the reporter constructs sandwiching a polybasic oligopeptide track between the fluorescent GFP and mCherry (mCh) fusion protein. (ACT AGC)~6~ \[(ThrSer)~6~\] encoded a neutrally charged amino-acid tract that served as a control. (**b**) Detection of GFP and mCherry fluorescent signals by FACS analyses in samples from (**a**) shown as relative cell numbers. Each experiment was performed in triplicates. (**c**) Domain structures of ZNF598 full-length and truncation mutants, with numbers referring to the position of amino acids. (**d**) Detection of GFP and mCherry fluorescent signals by FACS analyses in samples expressing full-length or truncated versions of ZNF598 and transiently transfected with GFP-mCherry reporter with (ACT AGC)~6~ and (AAA)~12~ linkers. Each experiment was performed in triplicate. (**e**) Upper panel: autoradiograph of cross-linked, ^32^P-labelled, RNA- Flag/HA-ZNF598 immunoprecipitate. Flag/HA-tagged full-length ZNF598 or truncated versions were separated by SDS-PAGE after 4SU PAR-CLIP. Lower panel: Anti-HA Western blot analyses of the cross-linked RNA-protein immunoprecipitates; HC, antibody heavy chain.](ncomms16056-f3){#f3} {ref-type="supplementary-material"} for a complete list of all detected peptides. (**b**) Volcano plots of the quantitative proteomic analysis of the ZNF598 interactome. The *t*-test difference based on label free quantitation for each detected protein is plotted against the negative logarithmic *P* value of a Welch's *t*-test. The intensity based absolute quantitation (iBAQ)[@b51] values correspond to the sum of all the peptide intensities divided by the number of observable peptides of a protein and are represented by point size. Proteins with a permutation-based FDR-value of \<5% and *t*-test difference \>0 are labelled in red and represent putative ZNF598 interactors (see also [Supplementary Fig. 15](#S1){ref-type="supplementary-material"} and [Supplementary Data 6](#S1){ref-type="supplementary-material"}). (**c**) Analysis of siRNA-mediated knockdown of UBE2D2, UBE2D3 or both UBE2D2 and UBE2D3 in HEK293 cells by western blot. C indicates mock transfection. (**d**) Detection of GFP and mCherry fluorescent signals by FACS analyses in samples from (**c**), for reporter constructs containing (ACT AGC)~6~ and (AAA)~12~ linkers. Each experiment was performed in triplicate. (**e**) Model for ZNF598-dependent ribosome stalling and RQC at cryptic polyadenylated protein-coding mRNAs.](ncomms16056-f4){#f4} [^1]: These authors contributed equally to this work.

INTRODUCTION {#sec1-1} ============ Surgical site infections (SSIs) are either superficial or deep and may involve the organs, or spaces accessed during an operation. The reported incidence of SSIs in coronary artery bypass grafting (CABG) surgery ranges between 0.3% and 8%.\[[@ref1]\] There is a strong suggestion that an impairment of vascular supply of the sternum may be one of the most important factors influencing the incidence of deep sternal wound infection (DSWI). Several studies have studied the risk factors for SSIs including DSWI in cardiac surgery. These risk factors included obesity, diabetes mellitus, chronic obstructive pulmonary disease (COPD), connective tissue disease, steroid use, smoking, peripheral vascular disease and renal insufficiency. In addition, intraoperative factors (e.g., use of bilateral internal mammary arteries \[BIMA\] grafting, prolonged cardiopulmonary bypass \[CPB\] duration) and postoperative variables (e.g., prolonged mechanical ventilation, reoperation for bleeding, postoperative transfusions and gastrointestinal, nephrological and respiratory complications) have been shown to be associated with DSWI.\[[@ref2][@ref3][@ref4]\] The risk for sternal wound infection (SWI) is increased if cardiac surgery involves internal thoracic arteries grafting and a valve procedure, or use of a ventricular assist device.\[[@ref5][@ref6]\] Leg wound infections at donor sites account for \>70% of cases with severe infection following cardiac surgery.\[[@ref7]\] Cardiac SSIs increase the length of hospital stay (LOS) and increase treatment costs in proportion to the severity of the infection. These costs increase by 3.8%, 14.7% and 29.4% in mild, moderate and severe infections respectively.\[[@ref7]\] Treatment is often confounded by the emergence of antibiotic-resistant pathogens and in addition, substantial proportions of these infected patients are elderly and have co-existing medical problems. In the past, such elderly patients with significant comorbidities would not have been considered for surgery.\[[@ref8]\] As the population ages, it is reasonable to assume that older and sicker patients will be admitted for surgery, and this will inevitably increase the risk and incidence of SSIs.\[[@ref9]\] Within our institution, the infection rate of postoperative wounds has been under surveillance by the infection control team since 2005. According to our preliminary data of rates of SSIs among isolated CABG patients, this was found to be significantly disproportionate to other regional institutions. As a result, we decided to audit, study and identify likely perioperative risk factors among our patients who have undergone isolated CABG between 2012 and 2013.\[[@ref10]\] The most important step in the management of wound infection is prevention, and this is best done by identifying risk factors. The present study was carried out in our centre to identify the incidence of wound infections following isolated CABG and identify the risk factors that may be associated with SSIs in our center. None of these perioperative risk factors have been studied previously in our population undergoing isolated CABG. Our pool of patients originates from a general Middle Eastern population, known to have a high prevalence of diabetes and obesity, a sedentary lifestyle with a lack of exercise, which represents a change of lifestyle following the discovery of oil in the region.\[[@ref10][@ref11][@ref12][@ref13][@ref14]\] MATERIALS AND METHODS {#sec1-2} ===================== Definitions of Infection {#sec2-1} ------------------------ Sternal SSI was defined according to the SSI criteria of the US Centers for Disease Control and Prevention. Sternal infections occurring within 30 days after surgery can include the following types: (1) Superficial incisional (infection above the sternum with no bony involvement); (2) deep incisional (infection involving the sternum); and (3) organ/space (site-specific infection such as mediastinitis).\[[@ref8][@ref15]\] Leg SSI was defined as redness, swelling, increased pain, excessive bleeding or discharge at the incision site among the patients who had undergone CABG. All SSI cases were diagnosed by attending physicians and confirmed by the nosocomial infection control committee. Patients who did not have any SSI formed the control group. Study Design {#sec2-2} ------------ From January 2012 to December 2013, 357 isolated CABG procedures were performed at our cardiac center. Totally, 40 postoperative patients diagnosed with an SSI (including sternal SSI, leg SSI and double SSI \[both sternum and leg\]), in accordance with the CDC criteria for SSI surveillance, were selected randomly. These 40 patients formed our study group (SSI) (*n* = 40, group I). This group was matched according to age, sex, nature of procedure and timing within the above study period with a control group (non-SSI) of 40 postoperative patients who did not suffer from an SSI (non-SSI group: *n* = 40, group II). The eighty selected CABG patients\' data were collected and analyzed retrospectively. Eight potential risk variables were compared between groups I and II. Potential Risk Factors {#sec2-3} ---------------------- Eight possible perioperative risk factors were analyzed and included the following: A prolonged LOS (LOS by days) which was arbitrarily taken as beyond a 30 days stay, a previous diagnosis of diabetes mellitus, impaired estimated glomerular filtration rate (eGFR \< 60 ml/min) taken as a sign of renal insufficiency, urgency of surgery (i.e. surgery done within 24 h of diagnosis of surgical coronary artery disease), the use of BIMA for grafting, impaired ejection fraction (EF) including moderate and severe (EF% \<45%), prolonged CPB duration taken as \>2 h and an elevated body mass index (BMI) that is, \>25. Data Analysis {#sec2-4} ------------- The risk factors for infection were assessed by univariate analysis. Discrete variables were assessed using Chi-squared analysis or Fisher\'s exact test. Variables were assessed using two tailed Student\'s *t*-test. All variables suggested by the univariate analysis were entered into a stepwise binary logistic regression analysis model. The chosen level of significance was 5%. All analysis was performed using the Statistical Package for the Social Sciences (SPSS) 19.0. IBM Corporation. RESULTS {#sec1-3} ======= All of the 80 patients who enrolled during the study period underwent isolated CABG only. Among the eight potential risk factors studied, the factors that had significant differences between the SSI study group and non-SSI control group were an impaired eGFR (*P* = 0.011, odds ratio \[OR\]: 3.8) and an impaired EF% (*P* = 0.015, OR: 5.1) \[Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}\]. ###### Association of exposure with SSI  ###### Potential preoperative risk factor for cardiac surgical infection for underwent CABG, 2012-2013\*  Patient\'s LOS (days), urgency of surgery, BIMA grafting, prolonged CPB duration and an increased BMI had no significant influence on the incidence of wound infection \[[Table 1](#T1){ref-type="table"}\]. Of the 40 patients in the SSI group, 22 were male, and 18 were female that was equivalent to the non-SSI group. The mean age for the SSI group was 59 versus 61 for the non-SSI group. There was no significant statistical difference in the age and sex of both groups. Only 2 patients in the SSI group stayed in the hospital beyond 30 days. None of the non-SSI group had a prolonged LOS. The difference was statistically insignificant. Thirty-five patients (87%) in the SSI group were previously diagnosed with diabetes mellitus (type 1 or type 2) while 30 patients (75%) in the non-SSI group were diabetic. Hence, diabetes was not found to be a preoperative predictor of SSIs (*P* - 0.156). Urgent isolated CABG was performed on 6 patients (15%) in the SSI group compared to only 4 patients in the non-SSI group. There was no statistically significant difference with a *P* = 0.499 and an OR: 1.5. Seventeen patients (42%) in the SSI group had a prolonged CPB time of more than 2 h duration while 13 (32%) of patients in the non-SSI group had a prolonged CPB time. Again, prolonged CPB duration was not found to be an intraoperative predictor of SSI (*P* - 0.356). Bilateral internal mammary harvesting was performed on 6 patients in the SSI group while 11 patients in the non-SSI group had bilateral mammary harvesting performed. Bilateral mammary harvesting was not found to be a risk factor for SSI with a *P* - 0.1/OR: 2.1. DISCUSSION {#sec1-4} ========== Surgical site infections are a manifestation of an imbalance between microbial growth and host\'s defenses. The Surgical stress response imposes an impairment of these defenses.\[[@ref16]\] Loop *et al*., described several risk factors for sternal wound complications in cardiac surgery. Bilateral internal mammary harvesting, diabetes, obesity, blood transfusion and operative time were considered significant risk factors for sternal wound complications.\[[@ref17]\] Other authors had described other risk factors for SSIs in cardiac surgery, with conflicting findings. Preoperative hospital admission duration, antibiotic prophylaxis use, surgical urgency, reoperation, surgical time, CPB duration, amount of blood transfused, postoperative blood loss, chest re-exploration, rewiring of a sterile sternal dehiscence, duration of mechanical ventilation and days of treatment in the intensive care unit were described as other perioperative factors contributing to the development of SSIs.\[[@ref18]\] Our cardiac center had performed 377 isolated CABG procedures over the designated study period of which, 67 patients developed an SSI. This was an incidence rate of 17%. This elevated incidence of infection prompted us to perform this study in an attempt to analyze perioperative surgical and patient risk factors that might contribute to this rate. Following a literature review, we decided to study eight variables \[[Table 2](#T2){ref-type="table"}\] in our cohort of patients who originate from a population, known to have a high prevalence of prediabetes, diabetes mellitus, obesity and pursues a modern, sedentary lifestyle.\[[@ref10]\] The Bahraini population has become increasingly modernized, over the last 40 years, resulting in a transformation from an active lifestyle to one that lacks physical activity, sunlight exposure and has acquired unhealthy dietary patterns. These social factors have led to a higher prevalence of chronic obesity, insulin resistance, prediabetes, and type 2 diabetes.\[[@ref11][@ref12][@ref13][@ref14][@ref19]\] We found that an impaired renal function and/or impaired left ventricular ejection fraction (LVEF) are statistically significant patient characteristics for acquiring SSIs. Interestingly, other risk factors like diabetes, bilateral mammary harvesting and an elevated BMI were found to be statistically insignificant risk factors. Sakamoto *et al*. had previously concluded that patients in a poor perioperative condition, that is, in a poorly perfused state and requiring hemodynamic supportive devices, were more likely to develop DSWI.\[[@ref20]\] We believe, having an impaired LVEF could be considered a marker for poor tissue perfusion in the perioperative period. Since our study did not identify those patients who had had the aid of hemodynamic support devices, we think these patients would be included in our cohort that had a severely impaired EF. A moderate or severely impaired EF can lead to a state of generalized poor perfusion, which would hinder wound healing. Hence, we correlate our findings with those of Sakamoto *et al*. Renal failure was found by some authors, to be a significant risk factor for mediastinitis and hemorrhage after cardiac surgery and isolated CABG.\[[@ref6][@ref20][@ref21]\] We studied patients with an impaired preoperative estimated GFR, that is, chronic kidney disease (CKD) stage 2 or more; an eGFR \< 60 ml/min. We found this to be a statistically significant risk factor. CKD impairs immunity and the healing process through hyperuremia, presence of anemia of chronic disease, previous multiple blood product transfusions and a long-term indwelling dialysis catheter. Such catheters can harbor microorganisms that become a source of infection. Several authors have identified obesity/overweight (BMI \> 25) as a major risk factor.\[[@ref6][@ref22][@ref23]\] Surprisingly, our study found that obesity and being overweight were not statistically significant risk factors. This could be because of the high prevalence of obesity in both our study and control groups (87% and 72% respectively) which is a reflection of its high prevalence in the general population.\[[@ref10][@ref11][@ref12][@ref13]\] In Bahrain, a study by Hubail and Culligan showed that the prevalence of a BMI ≥25.0 kg/m^2^ was 56.4% in males and 79.7% in females among the general Bahrain population.\[[@ref19]\] Obesity is a known modifiable risk factor for coronary artery disease. The high prevalence of obesity and the fact that our study is a study of coronary artery disease might have led to a selection bias. Both of these factors might have affected our study finding regarding obesity. We conducted an extensive literature review with respect to diabetes and the incidence of SSIs.\[[@ref24]\] We found that the diabetes is one of the major risk factors for post CABG SSIs. The increased infection rate in diabetes has been attributed to the impairment of neutrophil chemotaxis, phagocytosis, adherence plus the glycosylation of collagen matrix proteins - all of which lead to weakened antibacterial defenses and delayed wound healing.\[[@ref14]\] We studied the preoperative diagnosis of diabetes regardless of type and control, anticipating a correlation between being diabetic and acquisition of an SSI. However, our results showed no correlation between a preoperative diagnosis of diabetes and CABG SSI. This again was surprising, and we assume this would be because of the high prevalence of diabetes in both our study and control groups (87.5% and 75%, respectively). Studies of diabetes in Bahrain indicate prevalence rates of 25.8% in males and 36.4% in females. These are considered to be among the highest in the world.\[[@ref19]\] In addition, we practice tight intraoperative and postoperative glucose control which may have contributed to this lack of correlation. Hence, we suggest that further regional studies should focus rather on the preoperative control of diabetes, for example correlating the preoperative level of HBA1C with CABG SSIs. The role of surgical urgency as a risk factor for developing SSI in cardiac surgery is controversial. Sakamoto *et al*. studied surgical urgency and found it to be a significant factor for DSWIs,\[[@ref20]\] while Ku *et al*.,\[[@ref25]\] could not identify such a correlation. We looked at surgical urgency as a risk factor, defining it for isolated CABG as the performance of CABG within 24 h of diagnosis of coronary artery disease that required surgical intervention and/or unscheduled CABG performed out of normal working hours in our center. About 15% of our study group were done as an emergency versus 10% in the control group. Our univariate analysis could not show a correlation between urgency of surgery and SSI. This may be due to that our preoperative preparation for these urgent cases is very similar to that of elective CABG, that is, bathing, chest hair shaving, and iodine preparation. In addition, preoperative hemodynamic stabilization with assist devices (e.g., Intra-Aortic Balloon Pump (IABP)) and medical management allows time for adequate preoperative preparation. Another possible explanation is our use of an identical antibiotic prophylaxis regimen for all cardiac surgical cases, elective or urgent. Kouchoukos *et al*.\[[@ref26]\] and Grossi *et al*.\[[@ref27]\] reported that the use of bilateral mammary grafting in isolated CABG did significantly increase the incidence of SWIs; attributed to the diminished blood supply to the sternum resulting in impaired healing. Saso *et al*.\[[@ref28]\] found a reduction in SWIs in isolated CABGs with skeletonized bilateral internal mammary harvest rather than pedicled harvests. The practice in our center is pedicled dissection of internal mammary harvests. We found no significant relationship between bilateral mammary harvesting and SSIs. We did not specifically address deep versus superficial SWIs but rather SSIs in general. Another limitation is the univariant correlation analysis performed by our study that did not investigate the concordance of diabetes and bilateral mammary harvesting on SSI incidence. Cardiopulmonary bypass causes immunosuppression. The lungs\' role of macrophageal scavenging is bypassed, in addition to the release of immunosuppressive immunomodulators.\[[@ref29]\] We studied the relationship between CPB duration and SSIs, we arbitrarily considered 2 h to represent a prolonged duration of exposure to CPB. Sakamoto *et al*. and Minohara *et al*. studied CPB duration and SWIs and found no relation.\[[@ref20][@ref30]\] Our study concurs with their findings. About 42% of our study group had a prolonged CPB, with 32% in the control group. This was found to be insignificant. Our explanation; that CPB-induced immunomodulation might be event-related rather than time-dependent. CPB triggers the complement cascade and activates cytokines like C3a and TGF-β~1.~ Some studies have shown a relationship between prolonged perioperative hospital stay and SSIs.\[[@ref31][@ref32][@ref33]\] We looked at prolonged perioperative hospital stay (\>30 days) and found only 2 patients from our study group had a prolonged hospital stay. However, controversy remains regarding the role of a prolonged perioperative hospital stay. CONCLUSION {#sec1-5} ========== In our Middle Eastern population, we found that a poor preoperative clinical condition manifested as a moderately or severely impaired left ventricular and/or renal impairment were significant preoperative risk factors for acquiring SSIs in patients undergoing isolated CABG. Interestingly, obesity, diabetes, bilateral mammary harvesting, prolonged CPB time, urgency of surgery and length of perioperative hospital stay were found to be statistically insignificant risk factors in this small study. We recognize limitations in our study that include the small sample size and the univariant analysis of individual known risk factors. We did not investigate the possibility of the concordance of these factors. Our study also omitted the investigation of other known risk factors, e.g., blood transfusions, smoking and other comorbidities like COPD and hypertension. We suggest that future regional studies should carry out multivariant analysis on these and other risk factors in a larger cohort. We postulate that several factors do play a role in the acquisition of post CABG SSIs and possibly have an additive/cumulative effect on the incidence of SSIs. The authors gratefully acknowledge the work of the Shk. Moahmmad Al Khalifa Cardiac Center statistician Ms. Na Lian for her support in the data analysis of our study. **Source of Support:** Nil **Conflict of Interest:** None declared.

Introduction {#Sec1} ============ In contrast to mammals, lampreys show spontaneous and successful functional recovery after a complete spinal cord injury (SCI) and this is in part due to their impressive ability for axonal regeneration^[@CR1]--[@CR8]^. But, even in lampreys, not all descending neurons of the brain are able to regenerate their axons through the site of injury after a complete spinal cord transection^[@CR4],[@CR9]--[@CR12]^. The lamprey brainstem contains approximately 30 large individually identifiable descending reticulospinal neurons that vary greatly in their ability for axonal regeneration after SCI, even when their axons run in similar paths in a spinal cord that is permissive for axonal regrowth^[@CR4],[@CR12],[@CR13]^. Some identifiable descending neurons of lampreys are considered "good regenerators" (i.e. they regenerate their axon more than 55% of the times; the I3, I4, I5, B2, B5 and B6 neurons) and others are considered "bad regenerators" (i.e. they regenerate their axon less than 50% of the times; the M1, M2, M3, I1, I2, B1, B3, B4 and Mth neurons)^[@CR4],[@CR6],[@CR12]^. This indicates that interactions with the extrinsic spinal cord environment and intrinsic differences between descending neurons affect their regenerative abilities after SCI. Recent work has also shown that identifiable descending neurons of lampreys that are known to be "bad regenerators" slowly die after a complete SCI and are also "poor survivors"^[@CR12],[@CR14],[@CR15]^. The death of these neurons after SCI appears to be apoptotic as indicated by the appearance of TUNEL labelling and activated caspases in their soma^[@CR14]--[@CR18]^. This offers a convenient vertebrate model to study the inhibition or promotion of neuronal survival and axonal regeneration in the same in vivo preparation and at the level of single neurons. In mammals, SCI leads to a massive release of aminoacidergic neurotransmitters (glycine and GABA:^[@CR19],[@CR20]^; glutamate:^[@CR21]--[@CR23]^). Excessive glutamate release after SCI is responsible for excitotoxicity and neuronal death^[@CR21],[@CR22]^. High extracellular glutamate levels result in excessive activation of glutamate receptors, triggering massive Ca^2+^ influx into cells, which leads to neuronal death^[@CR24]^. Extracellular glycine could also contribute to glutamate excitotoxicity^[@CR20]^, since it is a co-agonist of the *N*-methyl-D-aspartate glutamate receptor^[@CR25]^. The phenomenon of excitotoxicity has been mainly studied in intrinsic spinal cord cells; however, retrograde damage to neurons is also likely due to the fact that Ca^2+^ ions gain access to the axoplasm of damaged axons^[@CR26]^. In contrast to glutamate, it has been reported that GABA could have neuroprotective effects after different types of central nervous system (CNS) damage^[@CR27]--[@CR31]^. The activation of pre-synaptic GABAB receptors causes inactivation of voltage-dependent Ca^2+^ channels (see^[@CR32]^), which could prevent the influx of Ca^2+^ ions due to glutamate release. In addition, it has been shown that GABA can modulate and promote neurite outgrowth in vitro or during development (for reviews see^[@CR33],[@CR34]^). However, a role for GABA and GABAB receptors in neuroprotection and especially in axonal regeneration after SCI has not been reported yet. In lampreys, glutamate induces an inhibition of neurite outgrowth in reticulospinal neurons in vitro due to Ca^2+^ influx^[@CR35]^. Electrophysiological studies have also suggested that low intracellular Ca^2+^ levels due to downregulation of Ca^2+^ channels could facilitate axonal regeneration in axotomized descending neurons of lampreys^[@CR36]^. More recently, we have reported that, as in mammals, there is a massive release of glutamate, GABA and glycine from most spinal cord neurons close to the lesion site following a complete SCI^[@CR37]--[@CR39]^. Between 1 and 3 days after the injury, we observed the extracellular accumulation of GABA in the form of "*halos*" around some axotomized axons of descending neurons close to the site of injury. Statistical analyses revealed a significant correlation between GABA accumulation and a higher survival ability of the corresponding identifiable descending neurons^[@CR37]^. An electrophysiological study in the spinal cord of lampreys has also found a correlation between higher GABAergic inhibition and a better recovery of function in spinal lesioned animals^[@CR40]^. These data prompted us to hypothesize that, in lampreys, increased GABA signalling after SCI could be favouring the recovery process by promoting survival and axonal regeneration of descending neurons. Here, we address this question for the first time in vivo in any vertebrate and provide gain and loss of function evidence showing that endogenous GABA, acting through GABAB receptors, promotes survival and axonal regeneration of identifiable descending neurons after SCI in lampreys. Materials and methods {#Sec2} ===================== Animals {#Sec3} ------- All experiments involving animals were approved by the Bioethics Committee at the University of Santiago de Compostela and the *Consellería do Medio Rural e do Mar* of the *Xunta de Galicia* (License reference JLPV/IId; Galicia, Spain) or the Institutional Animal Care and Use Committee at the Marine Biological Laboratory (Woods Hole, MA) and were performed in accordance to European Union and Spanish guidelines on animal care and experimentation or the National Institutes of Health, respectively. During experimental procedures, special effort was taken to minimize animal suffering and to reduce the use of animals. Animals were deeply anaesthetized with 0.1% MS-222 (Sigma, St. Louis, MO) in lamprey Ringer solution before all experimental procedures and euthanized by decapitation at the end of the experiments. Mature and developmentally stable larval sea lampreys, *Petromyzon marinus* L. (*n* = 115; between 95 and 120 mm in body length, 5 to 7 years of age), were used in the study. Larval lampreys were collected from the river Ulla (Galicia, Spain), with permission from the *Xunta de Galicia*, or provided by Lamprey Services, Inc. (Ludington, MI, USA) and maintained in aerated fresh water aquaria at 15--23 °C with a bed of river sediment until their use in experimental procedures. Lampreys were randomly distributed between the different experimental groups. SCI surgical procedures {#Sec4} ----------------------- Animals were assigned to the following experimental groups: control unlesioned animals or animals with a complete spinal cord transection that were analyzed 1 week post-lesion (wpl), 2 wpl, 4 wpl, 10 wpl or 12 wpl. Within the 2, 10 and 12 wpl groups, the injured animals were assigned to either control or treatment groups. Table [1](#Tab1){ref-type="table"} summarizes the number of animals assigned to each experimental group and condition. Each experiment was carried out in at least two different batches of animals. Complete spinal cord transections were performed as previously described^[@CR41]^. Briefly, the rostral spinal cord was exposed from the dorsal midline at the level of the 5th gill by making a longitudinal incision with a scalpel (\#11). A complete spinal cord transection was performed with Castroviejo scissors and the spinal cord cut ends were visualized under the stereomicroscope. After spinal transections, the animals were returned to fresh water tanks and each transected animal was examined 24 h after surgery to confirm that there was no movement caudal to the lesion site. Then, the animals were allowed to recover in individual fresh water tanks at 19.5 °C and in the dark.Table 1Table showing the number of animals included in each experimental group and also the total number of identifiable descending neurons that were included in the analysesAnimalsTotal number of neurons included in the analysesM1M2M3I1I2I3I4I5B1B2B3B4B5B6MthChanges in gabab1 expressionControl71111128--5859--109--791 wpl710101412--12969--139--11144 wpl61012129--6665--65--57GABA treatment (2 wpl)Control6\*1111101291112712111212111212Treated568896995101010107910Baclofen treatment (caspase activation, 2 wpl)Control7\*1312111491314814131414121414Treated71213131471413614131414131313GABOB treatment (12 wpl)Control15303029292430302930303030303030Treated14262524282428272825242628232824Baclofen treatment (axonal regeneration, 12 wpl)Control7121213141214141114121414141213Treated11212121221922222122222222222222Gabab1 morpholino (ISH, 2 wpl)Control36----6----------------------Treated47----8----------------------Gabab1 morpholino (axonal regeneration, 10 wpl)Control9181818181818181818181818181818Treated13262626252626262626262626262626Total115Please note that in the in situ hybridization experiments, only the neurons that were unequivocally identified in at least two brain sections were included in the quantifications. In the FLICA experiments, six animals were used as controls for both the GABA and baclofen treatments and an extra animal was used as a control only for the baclofen treatment (asterisks) In situ hybridization {#Sec5} --------------------- For gabab1 in situ hybridization, the head of the animals was fixed by immersion in 4% paraformaldehyde (PFA) in 0.05 M Tris-buffered saline (TBS; pH 7.4) for 12 h. Then, the brains were dissected out, washed and embedded in Neg 50^TM^ (Microm International GmbH, Walldorf, Germany), frozen in liquid nitrogen-cooled isopentane, sectioned on a cryostat in the transverse plane (14 μm thick) and mounted on Superfrost Plus glass slides (Menzel, Braunschweig, Germany). In situ hybridization with a specific riboprobe for the gabab1 subunit of the sea lamprey gabab receptor (GenBank accession number KX655780; see Suppl. Figure [1](#MOESM1){ref-type="media"}) was conducted as previously described^[@CR42]^. Briefly, brain sections were incubated with the sea lamprey gabab1 DIG-labelled probe at 70 °C and treated with RNAse A (Invitrogen, Massachusetts, USA) in the post-hybridization washes. Then, the sections were incubated with a sheep anti-DIG antibody conjugated to alkaline phosphatase (1:2000; Roche, Mannhein, Germany) overnight. Staining was conducted in BM Purple (Roche) at 37 °C. In situ hybridization experiments were performed in parallel with animals from the different experimental groups (control, 1 wpl, 2 wpl and 4 wpl) and the colorimetric reaction was stopped simultaneously for all sections from the different groups of animals. Drug treatments {#Sec6} --------------- The following drugs were used to treat the animals following the complete spinal cord transection: GABA (Sigma; Cat\#: A2129; MW: 103.12 g/mol), GABOB (a GABA analogue; Sigma; Cat\#: A56655; MW: 119.12 g/mol) and baclofen (a GABAB receptor agonist). Baclofen was acquired from two different companies: for the experiments of caspase activation, we used baclofen from Molekula (Newcastle, UK; Cat\#: 31184509; MW: 213.66 g/mol), and for the experiments of axonal regeneration, we used baclofen from Carbosynth (Berkshire, UK; Cat\#: FB18127; MW: 213.66 g/mol). The drugs were applied in the water where the animals were left after the SCI surgical procedures (GABA at a concentration of 500 µM, GABOB at a concentration of 50 µM and baclofen at a concentration of 125 µM). The concentrations of baclofen and GABA were selected based on previous in vitro electrophysiological studies in lampreys^[@CR43]^. Since GABA does not easily cross the blood--brain barrier, it was applied at a high concentration and only in the first days after the injury when the spinal cord is still disrupted. We assumed that GABOB and baclofen also cross the blood--brain barrier as in mammals, since the blood--brain barrier of lampreys is similar to that of higher vertebrates^[@CR44],[@CR45]^. While we do not know the final concentration of the drugs in the CNS, we were confident that this application route allows access to the CNS as there were changes in the swimming behaviour of unlesioned animals in pilot experiments using baclofen and GABOB at these concentrations (not shown). Since these drugs are water soluble, control lesioned and non-treated animals were left in fresh water only. The animals that were analyzed for caspase activation 2 wpl were treated with GABA or baclofen during 4 days from the day of injury and replacing the drug and water every day during those 4 days. The animals that were analyzed for axonal regeneration 12 wpl were treated with GABOB or baclofen during the 12 weeks replacing the drug and the water four times each week. The animals were always kept in the dark during the drug treatments to prevent light degradation of these drugs. Morpholino treatment {#Sec7} -------------------- Application of morpholinos was performed as previously described in ref.^[@CR13]^. Briefly, the spinal cord was transected at the level of the 5th gill (see surgical procedures), and morpholinos (20 µg in lamprey internal solution: 180 mM KCl, 10 mM HEPES, pH 7.4; designed by GeneTools, LLC; Philomath, OR) were added at the time and site of SCI soaked in a small piece of Gelfoam (Pfizer; New York, NY). These included an active splicing-blocking gabab1 morpholino (5′-ACGTCTGCAACGGAGAGTCATGAGA-3′) generated against the boundary between the second intron and the second exon of the partial sea lamprey gabab1 sequence (Suppl. Figure [1](#MOESM1){ref-type="media"}), and a 5-base pair mismatch gabab1 negative control morpholino (5′-ACcTCTcCAACcGAGAcTCATcAGA-3′). During recovery, the morpholinos are retrogradely transported to the cell soma of descending neurons where they can knockdown the expression of the target mRNA^[@CR13],[@CR46]--[@CR48]^. Animals were allowed to recover for 10 wpl to analyze the effect of gabab1 knockdown (KD) in axonal regeneration of identifiable descending neurons. In situ hybridization was used to control the efficacy of the gabab1 morpholino KD in animals processed at 2 wpl. Detection of activated caspases in whole-mounted brain preparations {#Sec8} ------------------------------------------------------------------- The Image-iT LIVE Green Poly Caspases Detection Kit (Cat. No. I35104, Invitrogen, USA) was used to detect activated caspases in identifiable descending neurons (the M1, M2, M3, I1, I2, I3, I4, I5, B1, B2, B3, B4, B5, B6 and Mth neurons; see Suppl. Figure [2A](#MOESM2){ref-type="media"}) of larval sea lampreys 2 weeks after the complete spinal cord transection and the GABA or baclofen treatments. This kit contains 1 vial (component A of the kit) of the lyophilized FLICA reagent (FAM--VAD--FMK). The reagent associates a fluoromethyl ketone (FMK) moiety, which can react covalently with a cysteine, with a caspase-specific aminoacid sequence (valine--alanine--aspartic acid (VAD)). A carboxyfluorescein group (FAM) is attached as a fluorescent reporter. The FLICA reagent interacts with the enzyme active centre of an activated caspase via the recognition sequence, and then attaches covalently through the FMK moiety. Experiments for the detection of activated caspases in whole-mounted brain preparations were done as previously described^[@CR16]--[@CR18]^. Briefly, brains from control and treated 2 wpl animals were dissected out and immediately incubated in 150 µL of phosphate buffered saline (PBS) containing 1 μL of the 150× FLICA labelling solution at 37 °C for 1 h. Then, the brains were washed with PBS. Brains were fixed in 4% PFA in PBS for 2 h and 30 min at 4 °C. Next, the brains were washed with PBS, mounted on Superfrost Plus glass slides, and mounted with Mowiol. Retrograde labelling of descending neurons with regenerated axons {#Sec9} ----------------------------------------------------------------- At 10 (morpholino treatments) or 12 (GABOB and baclofen treatments) wpl, a second complete spinal cord transection was performed 5 mm below the site of the original transection and the retrograde tracer Neurobiotin (NB, 322.8 Da molecular weight; Vector; Burlingame, CA) was applied to the spinal cord lesion with the aid of a Minutien pin (\#000). The animals were allowed to recover at 19.5 °C with appropriate ventilation conditions for 7 days to allow the transport of the tracer from the application point to the neuronal soma of identifiable descending neurons (the M1, M2, M3, I1, I2, I3, I4, I5, B1, B2, B3, B4, B5, B6 and Mth were analyzed; see Suppl. Figure [2A](#MOESM2){ref-type="media"}). Since the original SCI also was a complete spinal cord transection, only neurons whose axons regenerated at least 5 mm below the site of injury were labelled by the tracer. Brains of these larvae were dissected out, and the posterior and cerebrotectal commissures of the brain were cut along the dorsal midline, and the alar plates were deflected laterally and pinned flat to a small strip of Sylgard (Dow Corning Co., USA) and fixed with 4% PFA in TBS for 4 h at room temperature. After washes in TBS, the brains were incubated at room temperature with Avidin D-FITC conjugated (Vector; Cat\#: A-2001; 1:500) diluted in TBS containing 0.3% Triton X-100 for 2 days to reveal the presence of Neurobiotin. Brains were rinsed in TBS and distilled water and mounted with Mowiol. Imaging and quantifications {#Sec10} --------------------------- An Olympus photomicroscope (AX-70; Provis) with a 20× Apochromatic 0.75 lens and equipped with a colour digital camera (Olympus DP70, Tokyo, Japan) was used to acquire images of brain sections from the in situ hybridization experiments. Images were always acquired with the same microscope and software settings. For the quantification of the level of gabab1 positive signal in identifiable descending neurons, first we established the intensity rank of positive colorimetric in situ signal. For this, we analyzed 10 random images from different descending neurons of control and lesioned animals. The "histogram" function in Image J shows the number of pixels in each image in a range of intensity from 0 to 255. With these images, we compared the intensity values in regions with clear in situ signal and the intensity values in regions without in situ signal. Based on this, we established a value of 179 as the lower limit to consider the colorimetric in situ signal as positive. Then the number of pixels of positive in situ signal was quantified for each section of each identified descending neuron. In brain sections, the identification of some of the specific descending cells becomes more difficult than in whole-mounts. Thus, only the cells that were unequivocally identified in at least two different sections were included in the quantifications (the M1, M2, M3, I1, I3, I4, I5, B1, B3, B4, B6 and Mth neurons; see Suppl. Figure [2A](#MOESM2){ref-type="media"}). Then, we calculated the average number of positive pixels per section for each individual neuron (see Table [1](#Tab1){ref-type="table"}) and these data were used for statistical analyses. The experimenter was blinded during quantifications. The quantification of the intensity of FLICA labelling was done as previously described^[@CR18]^. Briefly, photomicrographs were acquired with a spectral confocal microscope (model TCS-SP2; Leica, Wetzlar, Germany). Images were always acquired under the same microscope conditions for control or treated animals. Quantification of mean fluorescent intensity (mean grey value) of each identifiable neuron was done using the Fiji software. In whole-mounted brain preparations, the specific descending neurons are easily identifiable based on their morphology and rostro-caudal and dorso-ventral anatomical location. The experimenter was blinded during quantifications. The data from each individual identifiable neuron (see Table [1](#Tab1){ref-type="table"}) were used for statistical analyses. The percentage of neurons with regenerated axons (labelled by the Neurobiotin tracer) with respect to the total number of analyzed neurons (see Table [1](#Tab1){ref-type="table"}) was calculated for each type of identifiable descending neuron using an Olympus microscope or a Zeiss AxioImager Z2 microscope. The percentage of neurons with regenerated axons with respect to the total number of analyzed neurons in each animal was also calculated and these data were used for statistical analyses. The experimenter was blinded during quantifications. For the figures, images were taken with the Olympus microscope or the spectral confocal microscope (model TCS-SP2; Leica). After quantifications, contrast and brightness were minimally adjusted with Adobe Photoshop CS4 or CS6 (Adobe Systems, San José, CA, USA). Figure plates and lettering were generated using Adobe Photoshop CS4 or CS6 (Adobe Systems). Statistical analyses {#Sec11} -------------------- Statistical analysis was carried out using Prism 6 (GraphPad software, La Jolla, CA). Data were presented as mean ± S.E.M. Normality of the data was determined only when *n* numbers were higher than 10 by using the D'Agostino-Pearson omnibus test, and the homoscedasticity was determined by the Brown-Forsythe test. The in situ hybridization data that were normally distributed and homoscedastic were analyzed by a one-way ANOVA. Post-hoc Dunnett's multiple comparison tests were used to compare pairs of data. In situ hybridization data that were not normally distributed (or when the *n* numbers were lower than 10) were analyzed by a Kruskal--Wallis test and post-hoc Dunn's multiple comparisons test. The results of control vs. treatment groups were analyzed by a Student's *t*-test (for normally distributed data) or a Mann--Whitney *U* test (for non-normally distributed data). The in situ hybridization data after morpholino application were analyzed by a Mann--Whitney *U* test. The significance level was set at 0.05. In the figures, significance values were represented by different number of asterisks: 1 asterisk (*p* value between 0.01 and 0.05), 2 asterisks (*p* value between 0.001 and 0.01), 3 asterisks (*p* value between 0.0001 and 0.001) and 4 asterisks (*p* value \<0.0001). Results {#Sec12} ======= Increased expression of the gabab1 subunit in identifiable descending neurons after SCI {#Sec13} --------------------------------------------------------------------------------------- GABAB receptors are obligate heterodimers formed by gabab1 and gabab2 subunits^[@CR49]^. In previous work, we reported the expression of the gabab1 and gabab2 receptor subunits in identifiable descending neurons of adult sea lampreys under normal conditions^[@CR42]^. Here, we used gabab1 in situ hybridization first to confirm that this receptor is also expressed in identifiable descending neurons of mature larval sea lampreys (Suppl. Figure [2B](#MOESM2){ref-type="media"}; Fig. [1a, c, e](#Fig1){ref-type="fig"}) and then to quantify changes in its expression after SCI (Fig. [1a--g](#Fig1){ref-type="fig"}; Suppl. Figure [3](#MOESM3){ref-type="media"}). The M1, M2, M3, I1, I3, I4, I5, B1, B3, B4, B6 and Mth neurons were included in the analyses (see Material and Methods). This revealed a significant increase in the expression of the gabab1 subunit in the M2 (ANOVA, *p* = 0.0049), M3 (ANOVA, *p* = 0.002), I1 (Kruskal--Wallis, *p* = 0.0009), I3 (Kruskal--Wallis, *p* = 0.0097), B1 (Kruskal--Wallis, *p* = 0.015) and B3 (Kruskal--Wallis, *p* = 0.0178) neurons (Fig. [1g](#Fig1){ref-type="fig"}; Table [2](#Tab2){ref-type="table"}) in 1 wpl animals as compared to control unlesioned animals. Subsequent power calculations (using 80% power) indicated that the sample sizes were appropriately powered. Although a similar trend was observed for the M1, I4, I5, B4, B6 and Mth neurons in 1 wpl animals as compared to control unlesioned animals, statistical analyses did not reveal significant changes in the expression of the gabab1 subunit in these neurons (Suppl. Figure [3](#MOESM3){ref-type="media"}; Table [2](#Tab2){ref-type="table"}). At 4 wpl, the expression of the gabab1 subunit was not significantly different to control unlesioned animals in all identifiable descending neurons and returned to control levels (Fig. [1g](#Fig1){ref-type="fig"}; Suppl. Figure [3](#MOESM3){ref-type="media"}; Table [2](#Tab2){ref-type="table"}). This shows that the complete SCI induced an acute increase in the expression of the gabab1 subunit in descending neurons, which, together with the accumulation of GABA around the axons of identifiable neurons^[@CR37]^, supports the possible role of endogenous GABA as a neuroprotective and pro-regenerative molecule after SCI in lampreys.Fig. 1Changes in the expression of the gabab1 subunit in identifiable descending neurons after a complete SCI.**a**, **c** and **e** Photomicrographs of transverse sections of the brain showing the expression of the gabab1 transcript in descending neurons of control animals. **b**, **d** and **f** Photomicrographs of transverse sections of the brain showing the expression of the gabab1 transcript in descending neurons of lesioned animals at 1 wpl. **g** Graphs showing significant changes (asterisks) in the number of gabab1 positive pixels per section of the soma of identifiable descending neurons. The mean ± S.E.M. values are provided in Table [2](#Tab2){ref-type="table"}. Scale bars: 20 µmTable 2Mean ± S.E.M. values of the number of gabab1 positive in situ pixels/section in identifiable descending neurons of control and injured animalsGabab1 positive pixels/sectionControl1 wpl4 wplM1258,170 ± 86,006320,350 ± 78,104148,657 ± 43,920M2237,552 ± 61,602618,158 ± 176,397131,367 ± 19,089M3144,448 ± 56,162423,439 ± 93,05985,245 ± 16,702I1140,858 ± 94,768627,179 ± 137,246176,392 ± 74,810I339,783 ± 17,599204,914 ± 49,09961,945 ± 34,492I468,678 ± 11,645147,613 ± 44,90345,723 ± 7,610I537,027 ± 21,46362,221 ± 13,38054,634 ± 23,587B1198,893 ± 57,286429,418 ± 43,503202,694 ± 75,156B3184,112 ± 51,392488,947 ± 86,189161,864 ± 70,596B4183,470 ± 49,454317,944 ± 79,980274,012 ± 88,414B6222,515 ± 105,034456,303 ± 87,734273,466 ± 101,442Mth201,230 ± 84,794395,530 ± 90,345139,834 ± 31,750Refers to Fig. [1](#Fig1){ref-type="fig"} and Suppl. Figure [2](#MOESM2){ref-type="media"} GABA and baclofen treatments inhibit caspase activation in descending neurons after SCI {#Sec14} --------------------------------------------------------------------------------------- To test our hypothesis, we first analyzed the effect of GABA and baclofen (GABAB agonist) treatments in caspase activation in identifiable descending neurons after a complete SCI using FLICA labelling (Fig. [2a--i](#Fig2){ref-type="fig"}). As previously shown, in control lesioned animals, there is a statistically significant correlation between the intensity of FLICA labelling and the long-term survival and regenerative abilities of identifiable neurons (not shown;^[@CR18],[@CR50]^). At 2 wpl, animals treated with GABA or baclofen during 4 days showed a significant inhibition of caspase activation (fluorescence intensity of FLICA labelling) in identifiable descending neurons as compared to control animals (GABA: Student's *t*-test, *p* \< 0.0001; baclofen: Student's *t*-test, *p* \< 0.0001; Fig. [2j, k](#Fig2){ref-type="fig"}). This suggests that GABA can inhibit apoptosis in descending neurons after SCI by activating GABAB receptors.Fig. 2GABOB and baclofen treatments inhibit caspase activation in identifiable descending neurons.**a**, **d** and **g** Photomicrographs of whole-mounted brains showing identifiable descending neurons with intense FLICA labelling in control animals. **b**, **e** and **h** Photomicrographs of whole-mounted brains showing identifiable descending neurons with a reduction in FLICA labelling in GABA-treated animals. **c**, **f** and **i** Photomicrographs of whole-mounted brains showing identifiable descending neurons with a reduction in FLICA labelling in baclofen-treated animals. **j** Graph showing significant changes (asterisks) in the level of caspase activation (intensity of fluorescent FLICA labelling) after the GABA treatment. **k** Graph showing significant changes (asterisks) in the level of caspase activation (intensity of fluorescent FLICA labelling) after the baclofen treatment. Rostral is up in all photomicrographs. Scale bars: 100 µm GABOB and baclofen long-term treatments promote axonal regeneration in descending neurons after SCI {#Sec15} --------------------------------------------------------------------------------------------------- Then, we studied the long-term effect of increasing GABAergic signalling in axonal regeneration after a complete SCI. Retrograde neuronal tract-tracing with Neurobiotin showed that a treatment with either GABOB (Fig. [3](#Fig3){ref-type="fig"}) or baclofen (Fig. [4](#Fig4){ref-type="fig"}) during 12 weeks post-lesion significantly promoted axonal regeneration of identifiable descending neurons after a complete SCI as compared to control animals (GABOB: Student's *t*-test, *p* = 0.0129 (Fig. [3h](#Fig3){ref-type="fig"}); baclofen: Mann--Whitney *U* test, *p* = 0.0004 (Fig. [4h](#Fig4){ref-type="fig"})). The baclofen used in these experiments (from Carbosynth) was also tested to confirm that it had the same effect in the activation of caspases as the baclofen acquired from Molekula. This baclofen also inhibited caspase activation significantly in identifiable descending neurons as compared to control animals in a different set of experiments (Mann--Whitney *U* test, *p* \< 0.0001; not shown). This shows that an increase in GABAergic signalling through GABAB receptors promotes axonal regeneration after a complete SCI.Fig. 3A long-term GABOB treatment promotes axonal regeneration of identifiable descending neurons.**a**, **c** and **e** Photomicrographs of whole-mounted brains showing different reticulospinal populations with regenerated identifiable neurons in control animals, as identified by retrograde labelling. **b**, **d** and **f** Photomicrographs of whole-mounted brains showing different reticulospinal populations with an increased number of labelled (regenerated) identifiable neurons in treated animals. **g** Graph showing the percentage of regenerated neurons (with respect to the total number of analyzed neurons) for each identifiable cell in control and GABOB-treated animals. **h** Graph showing significant changes (asterisks) in the percentage of regenerated neurons per animal after the GABOB treatment (control: 37.27 ± 3.33%; GABOB: 49.79 ± 4.16%). Arrows indicate descending neurons that regenerated in GABOB-treated animals but not in controls animals. Rostral is up in all photomicrographs. Scale bars: 50 µmFig. 4A long-term baclofen treatment promotes axonal regeneration of identifiable descending neurons.**a**, **c** and **e** Photomicrographs of whole-mounted brains showing different reticulospinal populations with regenerated identifiable neurons in control animals. **b**, **d** and **f** Photomicrographs of whole-mounted brains showing different reticulospinal populations with an increased number of labelled (regenerated) identifiable neurons in treated animals. **g** Graph showing the percentage of regenerated neurons (with respect to the total number of analyzed neurons) for each identifiable cell in control and baclofen-treated animals. **h** Graph showing significant changes (asterisks) in the percentage of regenerated neurons per animal after the baclofen treatment (control: 46.17 ± 7.15%; baclofen: 77.91 ± 3.56%). Arrows indicate descending neurons that regenerated in baclofen-treated animals but not in controls animals. Rostral is up in all photomicrographs. Scale bars: 100 µm Endogenous GABA signalling through GABAB receptors promotes axonal regeneration after SCI {#Sec16} ----------------------------------------------------------------------------------------- To test whether endogenous GABA also promotes regeneration by activating GABAB receptors, we decided to use morpholinos to knockdown the expression of the gabab1 subunit in descending neurons after a complete SCI (Fig. [5](#Fig5){ref-type="fig"}). First, we used in situ hybridization to confirm that the active gabab1 morpholino is able to knockdown the expression of the gabab1 mRNA in identifiable neurons of 2 wpl animals (Fig. [5a--e](#Fig5){ref-type="fig"}). As an example, we analyzed the M1 (Fig. [5a, b](#Fig5){ref-type="fig"}) and the I1 neurons (Fig. [5c, d](#Fig5){ref-type="fig"}). The active gabab1 morpholino was able to significantly knockdown the expression of the gabab1 mRNA in identifiable descending neurons as compared to the gabab1 mismatch control morpholino (M1: Mann--Whitney *U* test, *p* = 0.0111; I1: Mann--Whitney *U* test, *p* = 0.0057; Fig. [5e](#Fig5){ref-type="fig"}). Then, neuronal tract-tracing showed that the treatment with the active gabab1 morpholino significantly inhibited axonal regeneration of identifiable descending neurons 10 weeks after a complete SCI as compared to the animals treated with the gabab1 mismatch control morpholino (Mann--Whitney *U* test, *p* = 0.0133; Fig. [5f--m](#Fig5){ref-type="fig"}). This confirms that, in lampreys, endogenous GABA promotes axonal regeneration of descending neurons after a complete SCI by activating GABAB receptors.Fig. 5Gabab1 morpholino treatments inhibit axonal regeneration of identifiable descending neurons.**a**, **c** Photomicrographs of transverse sections of M1 (**a**) and I1 (**c**) neurons showing the expression of the gabab1 transcript in control animals. **b**, **d** Photomicrographs of transverse sections of M1 (**b**) and I1 (**d**) neurons showing the decreased expression of gabab1 transcript in gabab1 morpholino-treated animals. **e** Graphs showing significant changes (asterisks) in the number of gabab1 positive pixels per section in the soma of M1 and I1 neurons after the gabab1 morpholino treatment. **f**, **h** and **j** Photomicrographs of whole-mounted brains showing different reticulospinal populations with regenerated identifiable neurons in animals treated with the control morpholino. **g**, **i**, **k**: Photomicrographs of whole-mounted brains showing different reticulospinal populations with fewer regenerated identifiable neurons in animals treated with the active gabab1 morpholino. **l** Graph showing the percentage of regenerated neurons (respect to the total number of analyzed neurons) for each identifiable cell in control and active gabab1 morpholino-treated animals. **m** Graph showing significant changes (asterisk) in the percentage of regenerated neurons per animal after the morpholino treatment (control mismatch morpholino: 43.89 ± 3.26%; gabab1 active morpholino: 33 ± 5%). Arrows indicate descending neurons that regenerated in gabab1 morpholino-treated animals but not in controls animals treated with the mismatch control morpholino. Rostral is up in photomicrographs (**f**) to (**k**). Scale bars: black, 20 µm; white, 50 µm Discussion {#Sec17} ========== Here, we have provided gain and loss of function data, using pharmacological and genetic treatments, showing that endogenous GABA signalling through GABAB receptors promotes neuronal survival and axonal regeneration of identifiable descending neurons of lampreys after a complete SCI. The analysis of the changes of expression of the gabab1 subunit in response to a complete SCI revealed a significant increase in the expression of this subunit in some identifiable descending neurons (with other neurons showing a similar trend). As stated in the introduction, massive glutamate release and the subsequent activation of glutamate receptors lead to an increase in Ca^2+^ influx into cells, which causes excitotoxicity and neuronal death after SCI^[@CR21],[@CR22],[@CR24],[@CR51]^; see^[@CR52]^. GABAB receptors can cause the inactivation of voltage-dependent Ca^2+^ channels (see^[@CR32]^). Therefore, this increase in the expression of GABAB receptors could compensate for the influx of Ca^2+^ into axotomized descending neurons caused by massive glutamate release. The acute increase in the expression of the gabab1 subunit in descending neurons and the massive release of GABA after SCI^[@CR37]^ appears as one of the mechanisms favouring neuronal survival and axonal regeneration after SCI in lampreys. As far as we are aware, no study has analyzed the expression of gabab subunits after SCI in mammals. Only a few mammalian studies have looked at changes in the expression of this receptor following other types of nervous system injuries (sciatic nerve ligation:^[@CR53]^; traumatic brain injury:^[@CR54]^; ulnar nerve transection:^[@CR55]^; cerebral ischaemia:^[@CR56]^). In contrast to the present results in lampreys, these studies showed that the expression of GABAB receptors decreases after the injury in different regions of the brain^[@CR54]--[@CR56]^. This could be a key difference between regenerating and non-regenerating animals, since axons of the later do not show good regenerative abilities after CNS injuries. Interestingly, and in agreement with the results in lampreys, Huang and colleagues^[@CR56]^ reported that an elevation in the protein expression of GABAB receptors in the cerebral cortex promotes neuroprotection after ischaemic damage. There is some controversy on the topic of whether descending neurons of the brain of mammals die after SCI. Some studies have shown the death of brain neurons after SCI^[@CR57]--[@CR63]^. On the other hand, two recent reports did not find evidence of the death of corticospinal neurons after SCI^[@CR64],[@CR65]^, and suggested that these neurons only suffer atrophy but do not die^[@CR65]^. In any case, the death or atrophy of descending neurons of mammals appears to involve apoptotic mechanisms as shown by the appearance of TUNEL labelling and activated caspase-3 immunoreactivity in these neurons after the injury at spinal levels^[@CR60]--[@CR62]^. Recent work in lampreys has also shown that identifiable descending neurons known to be "bad regenerators" are actually "poor survivors" after a complete SCI^[@CR8],[@CR12],[@CR14],[@CR50]^. These neurons enter in a process of slow and delayed death after SCI^[@CR8],[@CR12]--[@CR18]^ that is initiated by caspase activation in the injured axon at spinal levels^[@CR17],[@CR18]^. The death of these neurons also occurs through apoptotic mechanisms as shown by the appearance of activated caspases^[@CR15]--[@CR18]^, TUNEL labelling^[@CR14],[@CR15]^ and Fluoro-Jade® C labelling^[@CR12],[@CR18]^. Recent results have shown that there is a significant correlation between the intensity of caspase activation 2 wpl and the long-term regenerative^[@CR18]^ and survival^[@CR50]^ abilities of identifiable descending neurons of lampreys after SCI. Present results indicate that the activation of GABAB receptors by GABA/baclofen can inhibit caspase activation after SCI in identifiable descending neurons, which is a key step to preventing the development of apoptosis and promoting neuronal survival. Previous work in other models of CNS injury also showed that a baclofen treatment can inhibit caspase activation (model of kainic-acid-induced seizure in rats:^[@CR29]^; models of ischaemic brain injury in rats:^[@CR27],[@CR31]^; model of chemical hypoxia in retinal ganglion cells in rats:^[@CR66]^). Our study shows that the activation of GABAB receptors can also prevent apoptosis after a traumatic SCI. Of major importance is the fact that our results also support the role of GABA as a molecule that promotes true axonal regeneration of descending neurons through the site of a complete SCI. Behavioural analyses were not performed to establish a relationship between increased regeneration and improved functional recovery after the treatments. First, because in our case, control animals usually reach the highest level of recovery when using the Ayers test (see ref.^[@CR39]^) and also because the treated animals were in the drugs until the day of analysis. Experiments using a gabab1 morpholino demonstrated that endogenous GABA acts as a pro-regenerative factor after SCI by activating GABAB receptors. The morpholino experiments suggest that GABA might promote regeneration by activating GABAB receptors expressed in the axotomized descending neurons. But, we cannot rule out the possibility that GABA could also promote the regeneration of descending neurons indirectly by inhibiting other cells expressing GABAB receptors, like intrinsic spinal cord neurons^[@CR39],[@CR42]^ that could have also taken the morpholino in our experiments. Our data agree with previous in vitro or developmental studies regarding the role of GABA and GABAB receptors in neurite outgrowth^[@CR67]^. López-Bendito and coworkers^[@CR67]^ showed that the GABAB antagonist CGP52432 decreases the length of the leading process in migrating inhibitory neurons in brain slice cultures of mice. Also, both GABA and baclofen stimulate retinal ganglion neurite outgrowth in *Xenopus* cultures, and the GABAB antagonist CGP54262 shortened the developing optic projection in vivo^[@CR68]^. But, as far as we are aware, our results are the first in vivo demonstration showing that GABA promotes axonal regrowth after a CNS injury by activating GABAB receptors. Present and previous^[@CR37]^ results indicate that the GABAergic system of lampreys responds successfully to a SCI to limit retrograde degeneration and promote the regeneration of descending pathways. Conclusion {#Sec18} ========== We have revealed a major role of GABA and GABAB receptors in promoting the survival and regeneration of individually identifiable descending neurons of lampreys following a complete SCI. Now, it would be of interest to decipher the underlying mechanisms behind the neuroprotective and pro-regenerative effect of GABA. Based on previous results in lampreys showing a negative effect of Ca^2+^ in neurite outgrowth^[@CR35],[@CR36]^, a decrease in Ca^2+^ levels due to the activation of GABAB receptors could be one of the key events in the inhibition of apoptosis and activation of axonal regeneration by GABA. In future studies, it might be also interesting to analyze changes in gene expression elicited by GABA signalling and the activation of GABAB receptors to reveal new pathways involved in axonal regeneration and neuronal survival in lampreys. The present results provide further support for the idea suggesting that the lesioned spinal cord is a "new spinal cord"^[@CR69]^ and the importance of understanding the changes that occur after SCI in different neurotransmitter systems in the brain and in the spinal cord above and below the site of injury. This study adds to previous work revealing anatomical^[@CR37]--[@CR39],[@CR70],[@CR71]^ and physiological^[@CR40],[@CR72],[@CR73]^ changes in different neurotransmitter systems above and below the lesion in recovered lampreys and highlights the importance of understanding these changes before applying neuropharmacological interventions in SCI patients. Specially, when drugs affecting neurotransmission might not only modulate locomotor circuits, but also affect the process of neuronal regeneration and recovery (e.g. serotonin inhibitors/toxins: see refs.^[@CR72],[@CR74]^; GABOB/baclofen: present results). Our results provide a strong basis to translate this knowledge to mammalian models of SCI for the development of new therapies for patients with SCI. A recent large observational cohort study has found that the early administration of gabapentinoids (which are administered as anticonvulsants for SCI patients) improves motor recovery following SCI^[@CR75]^. Interestingly, baclofen is also already in use in the clinic, even for the treatment of SCI patients with spasticity^[@CR76]^ or neuropathic pain^[@CR77]^, which could facilitate the clinical translation of similar results in pre-clinical models of SCI. Electronic supplementary material ================================= {#Sec20} Supplementary Figure 1 Supplementary Figure 2 Supplementary Figure 3 Supplementary figure legends These authors contributed equally: Antón Barreiro-Iglesias, María Celina Rodicio Edited by: A. Verkhratsky **Electronic supplementary material** **Supplementary Information** accompanies this paper at (10.1038/s41419-018-0704-9). **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Grant sponsors: Spanish Ministry of Economy and Competitiveness and the European Regional Development Fund 2007--2013 (Grant number: BFU2014-56300-P) and Xunta de Galicia (Grant number: GPC2014/030). D.R.-S. was supported by a fellowship from EMBO (Ref.: 7010) to carry out a short-term stay at the laboratory of JRM. A.B.-I. was supported by a grant from the Xunta de Galicia (Grant number: 2016-PG008) and a grant from the crowdfunding platform *Precipita* (FECYT; Spanish Ministry of Economy and Competitiveness; grant number 2017-CP081). The authors would like to acknowledge the following individual donors of the crowdfunding campaign in *Precipita*: Blanca Fernández, Emilio Río, Guillermo Vivar, Pablo Pérez, Jorge Férnandez, Ignacio Valiño, Pago de los Centenarios, Eva Candal, María del Pilar Balsa, Jorge Faraldo, Isabel Rodríguez-Moldes, José Manuel López, Juan José Pita, María E. Cameán, Jesús Torres, José Pumares, Verónica Rodríguez, Sara López, Tania Villares Balsa, Rocío Lizcano, José García, Ana M. Cereijo, María Pardo, Nerea Santamaría, Carolina Hernández, Jesús López and María Maneiro. The authors thank the staff of *Ximonde* Biological Station for providing lampreys used in this study, and the Microscopy Service (University of Santiago de Compostela) and Dr. Mercedes Rivas Cascallar for confocal microscope facilities and help. We also thank the Director of the Central Microscopy Facility at the Marine Biological Laboratory, Louie Kerr, for technical assistance and the Marine Biological Laboratory in Woods Hole (MA) for providing support for these experiments. This article is dedicated to the memory of José Manuel Pérez Cancela (20/11/1975--05/03/2018) from the *Ximonde* Biological Station. Conflict of interest {#FPar1} ==================== The authors declare that they have no conflict of interest.

Introduction {#Sec1} ============ Almost one century ago D'Arcy Thomson proposed that the spatiotemporal alterations in tissue mechanics inevitably alter its mechanoenvironemtal properties. These local biphasic mechanical properties, such as stiffness and fluidity, determine the system response to generated forces. This theory has been accepted widely^[@CR1]--[@CR7]^. Rheological techniques are rarely used for medical diagnosis of living tissue due to the invasive nature of rheological tests such as indentation experiments, torsional resonators and oscillatory shear testing devices^[@CR8]--[@CR10]^. These methods do allow for exploration of the viscoelastic properties of the medium in a wide range of frequencies, including the ultra-low frequency range (less than 1 Hz). Although a rheometer can assess the viscoelastic behavior of a medium in ultra-low frequency ranges, it is a time consuming and cumbersome process to use it in such a low frequency range^[@CR11]--[@CR13]^. Over a decade ago, non-invasive elastography methods were developed with the capability of remotely inducing shear waves in tissue and measuring its elasticity by recording the deflection response by MRI^[@CR14]^ and ultrasound^[@CR15]^. Later the *in vivo* application of elastography in measuring the stiffness of different tissues and organs like liver^[@CR16]--[@CR18]^, breast^[@CR19],[@CR20]^, brain^[@CR21],[@CR22]^, heart^[@CR23],[@CR24]^ and muscle^[@CR25],[@CR26]^ were reported. Although elasticity measurements were the focus of these efforts, a few studies considered the estimation of both elasticity and viscosity of *in vivo* tissues^[@CR20],[@CR22],[@CR27],[@CR28]^. While it is possible to use shear wave elastography methods for *in vivo* cases, the frequency range to explore the viscoelasticity of tissue is much narrower than with rheology methods^[@CR20],[@CR22],[@CR27]^. With shear wave elastography methods, reaching the ultra-low frequency range is almost impossible because there is always a tradeoff between the resolution of the resulting map and the frequency of vibration and shear waves^[@CR29]^. In this paper, we introduce the noninvasive, Loss Angle Mapping (LAM) method, which is based on measuring the local displacement and strain behaviors under constant stress as a function of frequency. Technical details were explained in our previous work^[@CR30]^. The LAM method can monitor the viscoelastic properties of the tissue with high resolution due to its high accuracy in displacement measurement at the micrometer level. High frame rate ultrasound strain imaging, which is the essential part of the LAM method, allows capturing the local viscoelastic parameters *in vivo* like breast tissue. The main components of the LAM test are a compression mechanism that is used to exert an approximately step-force on the tissue and a high-frame rate ultrasound system for monitoring the internal local strain responses. In other words, a step-force is used as a stimulus for a certain amount of time, and the transient strain response, which is governed by viscoelastic properties of the medium, is monitored by analyzing a sequence of radiofrequency (RF) data during the excitation^[@CR31],[@CR32]^. In the LAM method the local tissue behavior in the sub-Hertz frequency range is used because at this frequency range the local biphasic behavior of tissue is more evident compared to other frequency ranges^[@CR33]--[@CR37]^. Results {#Sec2} ======= Viscoelastic gel phantom {#Sec3} ------------------------ Gel phantom can be used as a simplified mechanical model for breast tissue. Gel consists of collagen type I matrix that is saturated in water^[@CR38]^, similar to breast tissue. The matrix peptide chain in this type of collagen is responsible for the dense electric charge associated with the hydrophilic properties of collagen fibers and culminates in the viscoelastic properties of the medium^[@CR39]^. The same mechanism can happen in breast tissue in which the glycoproteins are responsible for viscoelastic properties due to their hydrophilic nature^[@CR39]^. The viscosity of the fluid for these two media, however, is different. The solid matrix in both gel and tissue makes a porous structure which helps move the fluid when the medium is compressed or under load^[@CR40]^. Fluid viscosity is responsible for the viscoelastic response of the media or tissue. In addition the hydrogen crosslinks between the fibers in a solid matrix can trigger the viscoelastic response^[@CR40]^. To verify the performance of the proposed model-free method, a viscoelastic inclusion phantom was made. The inclusion part of the phantom was made with 25.14 grams of gelatin (Sigma-Aldrich, St. Louis, MO), 60 ml propylene glycol (Sigma-Aldrich, St. Louis, MO), and 4 grams cellulose (ultrasound scattering; Sigma-Aldrich) in distilled water for a total volume of 300 ml. The background part of the phantom was made with 32.3 grams gelatin, 30 ml Vanicream Lite (Pharmaceutical Specialties, Inc., Rochester, MN), 6 grams cellulose (ultrasound scatterer; Sigma-Aldrich) and potassium sorbate (preservative; Sigma-Aldrich) in distilled water with a total volume of 600 ml^[@CR30]^. The inclusion phantom dimensions were 7.5 cm × 5.5 cm × 5.5 cm (L × W × H), with the cylindrical inclusion having a diameter of 1.5 cm. An ultrasound B-mode image of the gel phantom can be seen in Fig. [1](#Fig1){ref-type="fig"}. For demonstration purposes, two points were selected: point 1 in the background and point 2 in the inclusion. The results of the temporal strain profile for these points can be seen in Fig. [2a](#Fig2){ref-type="fig"}. Loss angle profiles for these 2 points were obtained in the frequency range less than 10 Hz (Fig. [2b](#Fig2){ref-type="fig"}) and less than 0.35 Hz (c). Similar to the above mentioned 2 points the viscoelastic map created by processing all the spatial points at a frequency of 0.033 Hz using the LAM method as shown in Fig. [2(d)](#Fig2){ref-type="fig"}.Figure 1Ultrasound Verasonics B-mode image of the gel phantom with the inclusion part indicated with a dashed line. 1 = a point in the background; 2 = a point in the inclusion.Figure 2Profiles obtained on the inclusion gel phantom. (**a**) Temporal strain profile of point 1 (background, red) and point 2 (inclusion, blue) in the phantom imaged in Fig. [1.](#Fig1){ref-type="fig"} (**b**) Loss angle profile of point 1 (background, red) and point 2 (inclusion, blue) in a frequency range less than 10 Hz. (**c**) Loss angle profile of point 1 (background, red) and point 2 (inclusion, blue) in a frequency range less than 0.35 Hz. (**d**) Viscoelastic map produced at 0.033 Hz based on the LAM method. ***In vivo*** patient study {#Sec4} --------------------------- ### **Patient study** {#Sec5} The phantom study results encouraged us to apply the LAM method on breast patients. The patient study was approved by the Institutional Review Board of the Mayo Clinic, Rochester MN, and informed consent was signed by each enrolled patient. This study was also compliant with the Health Insurance Portability and Accountability Act (HIPPA) in the Mayo Clinic. A total of 156 female patients with visible breast lesions in US images were recruited at Mayo Clinic from November 2014 to September 2016. The data from the first five patients were used to test the device and algorithm and were eliminated from the final study. Thus, a total of 151 breast patients were included in the study. Amongst them after applying the motion compensated cross-correlation metric (MCCC), 45 patients were selected for analysis. The rest of the patients were rejected. The mean patient age was 56 ± 15 years within the age range of 25--85 years. Breast lesions were categorized using the Breast Imaging Reporting and Data System (BI-RADS). Figure [3](#Fig3){ref-type="fig"} illustrates the distribution of lesion type in this patient population and Table [1](#Tab1){ref-type="table"} shows the BI-RADS distribution among them.Figure 3Distribution of lesion type in patient population.Table 1Distribution of patients.BI-RADS23456Number of patients08102424 **BI-RADS determination of lesions.** The BI-RADS value was determined for each lesion according to the sonographic features found in clinical US images obtained during a clinical procedure at Mayo Clinic. The patient's eligibility for biopsy was decided based on this value. For BI-RADS 5 and 6, a biopsy is always prescribed; as such, cases are highly suggestive of malignancy. A BI-RADS value of 3 or 4 is challenging as this covers a wide range of suspicion, including low, intermediate, and moderate. In our study, all the BI-RADS 3 and BI-RADS 4 were biopsied. The LAM technique was performed after determination of the lesion BI-RADS and location, and prior to the biopsy procedure. **Histology.** Surgical excision biopsy or US-guided core needle biopsy was performed as a part of clinical care and the histology results for all patients in this study were available. In the latter cases, five core biopsy samples of each lesion were acquired by one of our certified radiologists. An experienced Mayo Clinic pathologist with more than 15 years of experience provided the histopathological diagnosis. Surgical histopathology was considered conclusive over core needle biopsy. **LAM results.** As mentioned before, 156 patients were recruited for the study. The statistical results related to the whole patient population are depicted in supplementary part of this paper. Amongst those patients, 45 patients passed, the criterion set by MCCC metric. Only these female patients with a suspicious breast lesion (15.93 ± 8 mm) visible in ultrasound images were considered in this study. The mean and median age for this group was 56 and 55 years, respectively. The youngest participant was 25 and the oldest was 85 years old. All the patients underwent a biopsy procedure after the LAM test. Figures [4](#Fig4){ref-type="fig"}--[7](#Fig7){ref-type="fig"} illustrate examples of the loss angle maps of various malignant and benign lesion types. In all of these images, the LAM map is created at the frequency of 0.033 Hz. for each patient. we examined the clinical image, the corresponding B-mode image as seen using the I-Q data from the programmable ultrasound machine (Verasonics), an overlay of the estimated loss angle, a loss angle map with unreliable areas excluded (white areas in Figs [4(d)](#Fig4){ref-type="fig"},[5(d)](#Fig5){ref-type="fig"},[6(d)](#Fig6){ref-type="fig"} and [7(d)](#Fig7){ref-type="fig"}), a graph of the normalized applied stress, and representative temporal normalized strain curves from the lesion and normal tissue areas and corresponding estimated loss angle as a function of frequency.Figure 4Imaging of a benign tumor. (**a**) Ultrasound clinical B-mode image of a benign tumor. (**b**) Ultrasound Verasonics B-mode image of the same tumor. (**c**) 2D color map of the loss angle overlay on the B-mode image produced at 0.033 Hz based on the LAM method. (**d**) 2D color map of the loss angle produced at 0.033 Hz. (**e**) Normalized strain temporal behavior of the two specified points in (**d**) accompanied with a normalized stress profile. **(f)** Spectral behavior of the two specified points in (**d**) in a frequency range less than 1 Hz.Figure 5Imaging of a malignant tumor. (**a**) Ultrasound clinical B-mode image of a malignant tumor. (**b**) Ultrasound Verasonics B-mode image of the same tumor. (**c**) 2D color map of the loss angle overlay on the B-mode image produced at 0.18 rad/s (\~0.03 Hz) based on the LAM method. (**d**) 2D color map of the loss angle produced at 0.03 Hz. (**e**) Normalized strain temporal behavior of two specified points in (**d**) accompanied with the normalized stress profile. **(f)** Spectral behavior of two specified points in (**d**) in a frequency range less than 1 Hz.Figure 6Imaging of a malignant tumor. (**a**) Ultrasound clinical B-mode image of a malignant tumor. (**b**) Ultrasound Verasonics B-mode image of same tumor. (**c**) 2D color map of the loss angle overlay on B-mode image produced at 0.033 Hz based on the LAM method. (**d**) 2D color map of the loss angle produced at 0.033 Hz. (**e**) Normalized strain temporal behavior of two specified points in (**d**) accompanied with the normalized stress profile. **(f)** Spectral behavior of two specified points in (**d**) in a frequency range less than 1 Hz.Figure 7Imaging of benign tumor. (**a**) Ultrasound clinical B-mode image of  benign tumor. (**b**) Ultrasound Verasonics B-mode image of the same tumor. (**c**) 2D color map of the loss angle overlay on the B-mode image produced at 0.033 Hz based on the LAM method. (**d**) 2D color map of the loss angle produced a0.033 Hz. (**e**) Normalized strain temporal behavior of two specified points in (**d**) accompanied with the normalized stress profile. (**f**) Spectral behavior of two specified points in (**d**) in a frequency range less than 1 Hz. Among 45 patients, 27 cases were diagnosed as benign and 18 as malignant. LAM method, Eqs ([1](#Equ1){ref-type=""} and [2](#Equ2){ref-type=""}), is used to estimate the complex modulus parameters (storage, loss) as was illustrated in Figs [4](#Fig4){ref-type="fig"}--[7](#Fig7){ref-type="fig"} for delta, δ. The overall results are depicted in following figures. Figure [8](#Fig8){ref-type="fig"} shows the result based on the contrast, average and standard deviation measured in tumor part.Figure 8Summary of loss angle modulus parameters: Storage, Loss and Delta in 45 breast lesions. In the next step, the Logistic regression is applied on all of these aforementioned parameters shown in Fig. [8](#Fig8){ref-type="fig"}. The relevant ROC (Receiver Operating Characteristic) curve for each analysis is illustrated in Fig. [9](#Fig9){ref-type="fig"}.Figure 9ROC curve for 45 breast lesions. (**a**) ROC based on measuring the contrast. (**b**) ROC based on lesion average (**c**) ROC based on standard deviation. The measured sensitivity is 77.8%, and the estimated specificity is 96.3%. The accuracy comes to 88.9%. The Area under Curve, AUC, is 0.94. The standard error is 0.04. 95% Confidence interval is 0.82 to 0.99. Detailed information is shown in Table [2](#MOESM1){ref-type="media"} in the supplementary part. It should be noted that, the BI-RDAS results were considered in the aforementioned statistical results. Discussion {#Sec6} ========== Due to biphasic properties of soft tissue, there is always some phase lag between applied stress and the resulting strain. In addition to storage and loss modulus magnitude, the LAM method can leverage the phase lag between these two parameters to study the local viscoelastic properties of living tissue in more details^[@CR31],[@CR41]^. The resulting contrast based on the storage and loss modulus or their ratio can be even more conspicuous in a lower frequency range. Mizuno *et al*.^[@CR11]^ reported that nonequiliburium fluctuations were observable in cytoskeletal networks in a frequency range less than 10 Hz using an active rheometer. Implementing micro rheology on cardiac thin filaments also demonstrated that, in a frequency range less than 10 Hz, both the storage and loss modulus had been elevated by increasing the Ca^2+[@CR12]^. In addition, it has been shown that in living cells, more complex behavior at low frequency ranges can occur^[@CR42]^. Rheometry techniques also have been used on variety of tissue samples to validate the newly developed elastography techniques^[@CR43]--[@CR45]^. In the most recent study, the rheometer results confirmed the higher contrast in viscosity in the frequency range less than 1 Hz^[@CR45]^. Benign breast tumors are generally more viscous with higher phase lag, Fig. [8(g)](#Fig8){ref-type="fig"}, than the malignant ones, particularly with fibroadenoma cases^[@CR31],[@CR32]^. The extra cellular matrix (ECM) is mainly made of a collagen fiber network to which proteoglycan molecules are attached^[@CR46]--[@CR49]^. The primary part of these molecules is the dense and hydrophilic sulfate groups^[@CR50]^. Increasing the concentration of proteoglycan, and consequently the surrounding water molecules in ECM, can cause an increase in viscosity^[@CR31]^. The collagen fibers in ECM are interconnected toward their ends but sparsely connected in the middle, with strong covalent cross-links, which are responsible for the elastic response of the tissue to stress^[@CR31]^. According to electron microscopy observation of malignant breast lesions such as infiltrating ductal carcinoma, the number of sulfated proteoglycan molecules, which are mainly responsible for viscosity, declines at least five times per unit in connective tissues. Thus, while viscosity decreases in malignant lesions, they can be stiffer than surrounding tissue due to increased collagen fibers and covalent cross-links^[@CR46],[@CR49],[@CR50]^. There is, however, no sign of elevation in viscosity in the formation course of this kind of lesions^[@CR46],[@CR49]^. The common feature in benign solid lesions, such a fibroadenomas, is higher collagen density compared to surrounding tissues. An increase in the number of fibers leads to a greater number of proteoglycan molecules and decreased inter-fiber distance. The lower distance between collagen fibers will cause a greater H-bonded cross-link density. Generally, H-bonds help in stabilizing the ECM matrix by keeping the helical shapes within fibers. This functionality of H-bonded cross-links has an important role in creating an elastic restoring force immediately after being stressed. However, these fragile bonds break and reform during the stress application and some of the strain energy is dissipated. This causes the strain response delay in returning to the initial steady state, (usually in less than 5 seconds^[@CR31]^). In other words, this is the reason for viscoelastic response of the external compression, which occurs at a very low frequency range based on its retardation time. Thus, collagen density determines the strength and density of inter-fiber H-bonded links. Some fibrous tumors are not necessarily stiffer than surrounding regions, but due to higher cross-link density, can create a longer delay for a full strain response^[@CR32],[@CR46]--[@CR50]^. This phenomenon is the basis of viscoelastic methods like LAM method. Thus, in the case of fibroadenomas and other collagenous benign lesions, the lesion can be stiffer than the surrounding tissues but they are more viscous in contrast to malignant lesions^[@CR31],[@CR32],[@CR49]^ Fig. [8(g)](#Fig8){ref-type="fig"} also showed that phenomena. To observe such differences, it is necessary to evaluate tissue's viscoelastic response in a period of 1--10 s, which corresponds to a frequency range less than 1 Hz^[@CR32],[@CR34]^. This work on breast patients showed that considering the viscoelastic behavior of tissue by estimating the complex elasticity parameters like storage, loss and delta in an ultra-low frequency range could provide reliable biomarkers for differentiation of benign and malignant tumors in breast patients. The distinctive characteristic of the LAM method is its sensitivity to viscoelastic properties of breast tissue in this range of frequency. This characteristic correlates well with the biochemical construction of benign and malignant breast lesions. The sensitivity and specificity of this method in differentiation of benign and malignant lesions were 77.8% and 96.3% respectively. A possible contributing factor in misclassifications is measurement error. Lesion mobility can adversely affect the quality of recorded data. Patient motion during a test can also negatively affect data quality. Some patients were not able to hold their breath for the entire duration of data collection, which resulted in tissue motion. Cardiac motion was not considered as a major source of error as the frequency of such motion is around 1 Hz, which is somewhat outside the frequency band of these ultra-low frequency range calculation, less than 0.05 Hz. Due to all aforementioned reasons, we tried to apply more restrictions on our motion measurement algorithm. The initial patient population that was considered for this study was 156 patients. The statistical outcomes of these patients are shown in Figs [S1](#MOESM1){ref-type="media"} and [S2](#MOESM1){ref-type="media"}, Tables [S1](#MOESM1){ref-type="media"} and [S2](#MOESM1){ref-type="media"} in supplementary part. However, applying MCCC algorithm in order to detect the out-of-plane and non-axial motion while doing compression was a great assistance to identify cases with mostly axial compression with minimal out-of-plane motions. While this is done retrospectively, if implemented to operate in real-time, such metric can be utilized to ensure proper lesion compression and data collection. Reconsidering these data and using MCCC algorithm resulted in 45 cases out of 156 cases as they have been discussed already. In addition, a multi-parameters analysis using the features obtained from the all creep data (i.e. storage, loss modulus and delta) in conjunction with contrast, average and standard deviation measurement, provided and enhanced classification of the breast tumors in this study, Figs [8](#Fig8){ref-type="fig"} and [9](#Fig9){ref-type="fig"}. Future studies would include expanding the cohort and studying the sources of error in our measurement. This multi factorial analysis is another advantage of the LAM method. LAM can be utilized in conjunction of other ultrasound-derived lesion characteristics such as BI-RADS and size. Comparing Tables [2](#Tab2){ref-type="table"} and [3](#Tab3){ref-type="table"}, with and without considering BI-RADS respectively shows that both the specificity and sensitivity have been elevated from 89.29% to 96.3% and 68.42% to 77.78% respectively. The combined value of LAM and ultrasound imaging features promise better diagnosis without requiring more expensive and contrast-based imaging modalities such as contrast-enhanced magnetic resonance imaging.Table 2Classification table when using MCCC, BI-RADS number and tumor size (Cut-off value = 0.5).Actual groupPredicted groupPercent correct01Benign cases26196.30%Malignant cases41477.78%Percent of cases correctly classified88.89%**ROC curve analysis**Area under the ROC curve (AUC)0.938Standard Error0.03995% Confidence interval0.824 to 0.988Table 3Classification table when using MCCC but not BI-RADS number or tumor size (Cut-off value = 0.5).Actual groupPredicted groupPercent correct01Benign cases25389.29%Malignant cases61368.42%Percent of cases correctly classified80.85%**ROC curve analysis**Area under the ROC curve (AUC)0.868Standard Error0.05695% Confidence interval0.738 to 0.949 Another important aspect of this study is the automated ROI (Region of Interest) selection to remove the subjectivity associated with manual selection of data processing. In other words, all of the patient data was processed in a consistent manner without human input for selecting the region of interest or changing other parameters involved in data processing. First, the eligibility of data was tested for out-of-plane motions and then the processing began automatically either in selecting the lesion from normal part or in setting the creep legitimate time duration in which the monotonous increasing of compliance curve was counted. This procedure went through all the collected patient data. It should be noted that as the results show, the LAM method could be a very effective method in the context of non-invasive diagnostic/prognostic mechanical testing especially for soft tissue; however, it may not be a solid candidate for mechanical property estimation in general. In addition, the LAM method was observed to present a solid contrast between benign and malignant lesions at frequency range less than 0.033 Hz, however it needs more comprehensive studies to find the optimum frequency with highest contrast in this ultra-low frequency range. This would be explored in future work. Methods {#Sec7} ======= **LAM technique** {#Sec8} ----------------- In this patient study, an automated compression device was used to apply a ramp-and-hold force excitation for a duration of time automatically set based on monotonically elevation of creep response. The ultrasound probe is part of this device for recording the viscoelastic response of the underlying tissue. This device is portable, light-weight and easy to use for patient studies to explore the tissue dynamics under external stress^[@CR30]^. For the gel study, 8 N force with the speed of 16 N/s ramp was used. For the patient study, 2--4 N force with 8--12 N/s ramp were applied. To monitor the local viscoelastic response, an ultrasound system (Verasonics, Inc., Kirkland, WA, USA) with a linear array transducer (L11-4v, Verasonics, Inc., Kirkland, WA, USA) was used. In these experiments, the ultrasound center frequency was 6.43 MHz and the frame rate was 200 Hz during the acquisition time of 12 seconds. To track the fast tissue deformation under compression, a two-dimensional autocorrelation method was employed for a particle velocity calculation from adjacent frames. The integration of the particle velocity in time resulted in a displacement estimation^[@CR51]^. After performing the displacement estimation for all consecutive IQ data, the gradient of the resulting displacement was computed to measure the corresponding local strain^[@CR30]^. For patient studies, an experienced sonographer with more than 28 years of experience in breast ultrasound assisted with manual delineation of the lesion in B-mode US images obtained from the Verasonics system at the beginning of each patient data acquisition. Using the radiofrequency data, the loss angle maps were reconstructed offline at a frequency range less than one Hertz. Supplementary video is available online. ### **Dynamic complex measurement** {#Sec9} It was shown that the dynamic complex modulus in the frequency domain, *E*\*(*ω*), can be directly derived using the strain time data (local creep response) in a model-free fashion with some assumption about the experimental creep response^[@CR13],[@CR52],[@CR53]^, Eq. [1](#Equ1){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}^{\ast }(\omega )=\frac{i\omega }{i\omega J(0)+\frac{{e}^{-i\omega t(N)}}{\eta }+{\sum }_{n=1}^{N}(\frac{J(n)-J(n-1)}{t(n)-t(n-1)})({e}^{-i\omega t(n-1)}-{e}^{-i\omega t(n)})},n=1:N$$\end{document}$$where J(n) represents the strain data sampled at time point with index *n*. The parameter $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta $$\end{document}$ represents the steady state viscosity which is estimated by extrapolation of strain data to *t* → *∞*. Hence, the loss angle can be directly derived as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta =arctan(\frac{{E}_{l}^{\ast }(\omega )}{{E}_{s}^{\ast }(\omega )})$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{l}^{\ast }(\omega )$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{s}^{\ast }(\omega )$$\end{document}$ are the imaginary and the real parts of the complex shear modulus, corresponding to the loss and storage modulus, respectively, and δ is the loss angle which represents the phase difference between the storage and loss modulus of the medium due to its viscoelastic properties^[@CR13],[@CR52]^. ### **Measurement** {#Sec10} **Strain Quality Assessment, SQA.** Our main assumption in this study is the linear viscoelastic behavior of the material being tested (phantoms or tissue). Applying an approximate step force results in a temporal strain response, or creep response, which increases monotonically. This behavior is called a normal strain response. Applying this method for *in vivo* studies, however, requires introduction of additional constraints to avoid estimation of unrealistic values. Here, we define a new parameter as Strain Quality Assessment (SQA). SQA verifies two parameters: the total strain and the slope of the final part of the creep response. For each point in the viscoelasticity map reconstruction domain, SQA assigns a value of one for points that have both a positive total strain value and positive final slope, and zero if either of these criteria is violated. All the white areas in Figs [4(d)](#Fig4){ref-type="fig"}, [5(d)](#Fig5){ref-type="fig"}, [6(d)](#Fig6){ref-type="fig"} and [7(d)](#Fig7){ref-type="fig"} are excluded due to SQA. **Contrast:** Contrast refers to a comparison between features of a lesion to those of the surrounding, or background, tissues. For *in vivo* studies, measuring contrast parameters is important for diagnostic purposes. In some cases, measuring the contrast is even more important than measuring the parameter values^[@CR32]^. In order to measure contrast we use the Eq. ([3](#Equ3){ref-type=""}).$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Contrast=\frac{{S}_{lesion}-{S}_{background}}{({S}_{lesion}+{S}_{background})/2}=\frac{Difference}{Average}$$\end{document}$$ In this equation *S*~*lesion*~ is the mean value in the lesion area and *S*~*background*~ represents the mean value of the normal background tissue surrounding the lesion. To measure contrast, the mean value of S for the lesion and background parts should be determined. This is performed by first determining the lesion boundaries from a registered B-mode image within the region of interest (ROI). The automatic ROI was applied to remove the subjectivity about lesion and normal tissue values. In addition, the creep duration was selected automatically based on monotonic increasing compliance curve. As it has been emphasized before all data in this manuscript was analyzed at 0.033 Hz. **Rejection of data based on MCCC metric:** The displacement field obtained from the phase sensitive speckle tracking was utilized to stretch each frame back to its original location, as explained in^[@CR54]^. Briefly, the normalized cross correlation was performed between the pre-compressed and motion-compensated post-compressed echoes in the lesion area. This value served as a quality metric to assess the uniaxiality of the induced motions. Given substantial decorrelation that may occur during compression, a minimum normalized cross correlation value of 10% was chosen to declare a successful uniaxial deformation. Using this criterion, cases/acquisitions that did not meet a required threshold value of MCCC metric were excluded from the analysis. ### **Statistical Analysis** {#Sec11} MedCalc Statistical Software version 15.8 (MedCalc Software bvba, Ostend, Belgium; <https://www.medcalc.org>; 2015) was employed for statistical analysis. The receiver operator curve (ROC) was used to find the best diagnostic discrimination threshold for the estimated loss angle contrast in comparison with the pathology outcomes. The confidence intervals for sensitivity and specificity were found using bias-corrected and accelerated bootstrapping of 1000 trials. Resulting sensitivity, specificity and area under the curve were reported. The Wilcoxon analysis was performed with 95% confidence interval to represent a statistically significant difference for all analysis. All multi-parameter analyses were performed using a logistic regression method in MedCalc. Supplementary information ========================= {#Sec12} Supplementary **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Alireza Nabavizadeh and Mahdi Bayat contributed equally Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1038/s41598-019-41885-9. The authors would like to express their appreciation to Mr. Duane Meixner for ultrasound scanning of the breast patients, and Mr. Randall Kinnick for his technical support. This work was supported by grant R01CA168575 from the National Cancer Institute (NCI) and the National Institutes of Health (NIH). A.N. phantom experiment, patient data collection, writing computer code, analyzing LAM data, statistical analysis, writing the manuscript. M.B. patient data collection, statistical analysis, writing comuter code and analyzing LAM data. V.K. patient data collection. A.G. patient data collection and statistical analysis. J.W. patient data collection. A.A. funding acquisition, leading human study, study design, team supervision, and manuscript editing. M.F. funding acquisition, leading the project, study design, and editing manuscript. Competing Interests {#FPar1} =================== The authors declare no competing interests.

Introduction {#s1} ============ Articular cartilage is a highly specialised connective tissue in joints. Its main function is to provide a smooth, lubricated surface for articulation and to take up and distribute high loads. Its remarkable dimensional stability and mechanical properties are due to the composition of its extracellular matrix. The load-bearing function is based on the high osmotic pressure created by negatively charged glycosaminoglycans, which are predominantly aggrecan molecules. In addition, the fibrillar collagen network, mainly composed of type II collagen, provides the tissue with its tensile resistance. As the only cell type in articular cartilage, chondrocytes are entirely responsible for maintaining the metabolic balance of matrix proteins. Accordingly, it has been shown that mechanical forces affect chondrocyte metabolic activity (for a review, see [@pone.0036964-Grodzinsky1]). More precisely, *ex vivo* and *in vitro* models of chondrocyte mechanobiology have generally shown that static compression inhibits the expression of cartilage matrix proteins whereas dynamic compression regimens enhance them [@pone.0036964-Buschmann1]--[@pone.0036964-Jones1]. In this context, mechanotransduction is the molecular process by which cells convert mechanical force into biochemical signalling. Little is currently known regarding the sequence of biochemical events that are involved in mechanotransduction and that eventually result in the modulation of the chondrocyte phenotype. It is therefore necessary to assess the signalling and regulatory pathways activated during mechanical signal transduction in chondrocytes. In this study, we employed microarray analysis to investigate the overall changes in chondrocyte gene expression in response to dynamic compression. We used a cell model system consisting of isolated mouse chondrocytes embedded within an agarose hydrogel. We have previously used these constructs to develop experimental procedures to analyse the effects of compression at the mRNA level (using reverse transcription-polymerase chain reaction experiments) and to determine the phosphorylation state of signalling molecules (using Western blotting) [@pone.0036964-Bougault1], [@pone.0036964-Bougault2]. Here, our study was designed to identify candidate genes involved in the early response of chondrocytes to compression. ![Chondrocytes embedded in agarose gel maintain a well-differentiated phenotype.\ Expression of extracellular matrix proteins, integrins and the Sox9 transcription factor were analysed on Western blots of chondrocytes cultured in 3D for 3 days (d3) or 6 days (d6). The presence or absence patterns of proteins at day 6 are representative of 3 independent experiments. A: Type II (Col II) and type IX (Col IX) collagens accumulate and become cross-linked by day 6. Procollagen II forms (pro) and mature collagen II chains \[α1(II)\] are present. Beta (β) indicates cross-linked α1(II) dimers. Collagen IX chains \[α1(IX)\] are present. Asterisks (\*) indicate cross-linked collagens. B: Type I collagen (Col I) and α11 integrin (Itgα11) are not or only faintly detected, whereas the chondrocyte-specific α10 integrin (Itgα10) is present at day 6. Passaged chondrocytes cultured in monolayer were used as positive controls (Ctrl) for Col I and α11 integrin immunorevelations. Procollagen I (pro) and mature collagen I chains α1(I) and α2(I) are indicated. C: Sox9 chondrogenic transcription factor increases with the duration of culture.](pone.0036964.g001){#pone-0036964-g001} Taken together, the results presented here indicate that the mitogen-activated protein kinase (MAPK) and the transforming growth factor (TGF)- β pathways are involved in the early response of chondrocytes to dynamic compression. The microarray analysis revealed that only 20 transcripts were modulated more than 2-fold. At a fold modulation threshold of 1.4, an extended list of candidate genes included 325 candidate mechanosensitive genes, of which 85% were down-regulated. This global down-regulation may indicate a general control mechanism for a rapid response to dynamic compression. Many of the observed modulated genes are known to be mechanosensitive in other biological contexts. In addition, modulation of genes or transcripts involved in various aspects of cellular physiology was observed. Our integrated analysis provides new molecular insight into how chondrocytes respond to mechanical forces. {#pone-0036964-g002} Results {#s2} ======= Maintenance of the chondrocyte phenotype and cartilage-characteristic matrix deposition in an agarose hydrogel {#s2a} -------------------------------------------------------------------------------------------------------------- To investigate the early effects of dynamic compression on gene expression of fully differentiated chondrocytes, we used a previously described cell model system [@pone.0036964-Bougault1], [@pone.0036964-Bougault2]. Briefly, mouse chondrocytes were embedded in agarose just after their isolation from cartilage and these chondrocyte-agarose constructs were cultured for 6 days to allow extracellular matrix deposition. Under these conditions, chondrocytes are viable and their proliferation was confirmed by an increase in DNA content (about 1.5 fold, data not shown). Furthermore, they maintain their round morphology and type II collagen and aggrecan accumulate at the cell periphery [@pone.0036964-Bougault1], [@pone.0036964-Bougault2]. Western blotting was used to obtain more detailed information on the matrix proteins and integrin receptors present in the chondrocyte-agarose constructs just before application of dynamic compression ([Figure 1](#pone-0036964-g001){ref-type="fig"}). {#pone-0036964-g003} As expected, after 3 days of culture, chondrocytes synthesised type II collagen, but mainly in the procollagen form ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel A). Fibrillar collagens such as type II collagen are synthesised as precursor forms that must be cleaved to produce the mature triple helical collagens capable of packing into fibrils (for a review, see [@pone.0036964-Canty1]). After 6 days of culture, mature-form type II collagen was the predominant form and showed interchain covalent cross-links ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel A). All the enzymes necessary for the post-translational maturation of collagen were therefore active in the 3D scaffolds. In addition, we investigated type IX collagen, which is a minor non-fibrillar collagen present in hyaline cartilage. Western blot analysis confirmed the presence of covalent cross-links between collagen molecules in the chondrocyte-agarose constructs after 6 days of culture ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel A). To demonstrate the absence of critical proteins that could cause the chondrocytes to transduce mechanical signals in a non-characteristic way, we looked for type I collagen, the classical marker of fibroblasts and dedifferentiated chondrocytes. No type I collagen was detected in Western blots on the chondrocyte-agarose constructs, but it was detected in the positive controls, i.e. extracts of mouse chondrocytes cultured in monolayer ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel B). Therefore, before the compression experiments, chondrocytes synthesise mature and cross-linked extracellular matrix components that are part of the typical collagen network in cartilage. {ref-type="table"}. (B) Real-time PCR analysis on three independent experiments confirmed DNA microarray results for eight selected genes. Bars represent the compression-induced gene expression modulation (mean +/− SD), either up-regulation (red) or down-regulation (green) (\* p\<0.05, \*\* p\<0.01).](pone.0036964.g004){#pone-0036964-g004} Integrin transmembrane receptors connect the extracellular matrix to the intracellular cytoskeletal network and are expected to play an important role in cellular responses to mechanical forces. The main collagen-binding integrin on chondrocytes in cartilage is α10β1 integrin, whereas α11β1 integrin is more characteristic of mesenchymal tissues. Thus, α10 and α11 are good markers for evaluating the status of the chondrocyte phenotype [@pone.0036964-Gouttenoire1]. Integrins, probably along with other surface proteins, were removed from the cell surface after enzymatic isolation of chondrocytes from cartilage ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel B). α10 was re-expressed at the end of the culture period in agarose, whereas α11 could only be faintly detected. We also monitored another differentiation marker: Sox9, a transcription factor required for cartilage formation ([Figure 1](#pone-0036964-g001){ref-type="fig"} Panel C). In mouse chondrocytes, high levels of Sox9 protein correlate with type II collagen synthesis and a well-differentiated phenotype, whereas dedifferentiated as well as hypertrophic chondrocytes lack Sox9 [@pone.0036964-Lefebvre1]. Thus, after a 6 day culture period, robust Sox9 expression together with α10 integrin expression further confirmed that chondrocytes were highly differentiated. 10.1371/journal.pone.0036964.t001 ###### Results from DNA microarray analysis: gene expression levels in compressed samples were compared to uncompressed control samples (fold change \>2 and p-value \>0.01). {#pone-0036964-t001-1} RNA ID PROTEIN ID GENE NAME FOLD CHANGE ADJUSTED P-VALUE ----------- ------------ -------------------------------------------------------------------------------- ------------- ------------------ **UP-REGULATED GENES** NM_010234 Q6PCX9 Proto-oncogene protein c-fos;Fos 9.27 2E-04 NM_007913 Q9WVQ1 Early growth response protein 1;Egr1 3.90 1E-04 NM_010444 Q9DBG7 Nuclear receptor subfamily 4 group A member 1;Nr4a1 2.84 4E-04 NM_010499 P17950 Immediate early response gene 2 protein;Ier2 2.63 2E-05 NM_010118 Q9JLB2 Early growth response protein 2;Egr2 2.60 6E-03 NM_007570 Q04211 Protein BTG2;Btg2 2.28 5E-04 NM_010591 Q6SJQ0 Transcription factor AP-1;Jun 2.27 1E-02 NM_008036 P46935 Protein fosB;Fosb 2.16 4E-03 NM_008416 Q61136 Transcription factor jun-B;Junb 2.00 8E-04 **DOWN-REGULATED GENES** NM_175284 Q149J3 Frizzled homolog 10;Fzd10 0.28 7E-07 NM_030696 Q8BL66 Monocarboxylate transporter 4;Slc16a3 0.31 2E-05 NM_027864 Q61468 Polypeptide N-acetylgalactosaminyltransferase 14;Galnt14 0.36 3E-05 NM_026358 Q8VI64 Ovary-specific acidic protein;Osap 0.38 8E-06 NM_138741 Q9D994 Serum deprivation-response protein;Sdpr 0.40 2E-03 AJ293625 Q9D8T7\* SRA stem-loop-interacting RNA-binding protein (mitochondrial);Slirp 0.41 4E-06 AK020134 Metastasis associated lung adenocarcinoma transcript 1 (non-coding RNA);Malat1 0.41 1E-03 AK032986 Q8BQ86 WD repeat-containing protein 60;Wdr60 0.46 7E-03 NM_023190 Q11011 Apoptotic chromatin condensation inducer in the nucleus;Acin1 0.47 3E-04 NM_018857 Q70KY4 Mesothelin, cleaved form;Msln 0.48 7E-05 NM_146112 Q6Y7W8 PERQ amino acid-rich with GYF domain-containing protein 2;Gigyf2 0.50 3E-03 In conclusion, the chondrocytes in our agarose model system were well-differentiated and did synthesise mature, cross-linked extracellular matrix components as well as integrins, before we applied dynamic compression. {#pone-0036964-g005} Detection of MAPK pathway and Smad2 activation due to dynamic compression {#s2b} ------------------------------------------------------------------------- The mitogen-activated protein kinase (MAPK) pathways involving ERK1/2 and p38 are implicated in chondrocyte mechanotransduction [@pone.0036964-Fanning1]--[@pone.0036964-Hung1]. We therefore investigated MAPK activation in the chondrocyte-agarose constructs to select the appropriate duration of dynamic compression for the characterisation of mechanotransduction events. Chondrocyte-agarose constructs underwent dynamic compression for 5, 15 or 30 min ([Figure 2](#pone-0036964-g002){ref-type="fig"}) and Western blots were used to examine phosphorylation levels of ERK1/2 and p38. ERK1/2 phosphorylation was observed primarily after 15 min of compression (average 2-fold increase when compared to uncompressed samples) but high variability in the phosphorylation rates impaired the statistical significance of the results ([Data S1](#pone.0036964.s001){ref-type="supplementary-material"}). Following 5 min of compression, p38 phosphorylation stimulation was low but highly reproducible (p\<0.05). After 15 min, this activation seemed stronger (3-fold induction), but again, the response was highly variable ([Data S1](#pone.0036964.s001){ref-type="supplementary-material"}). 10.1371/journal.pone.0036964.t002 ###### Primers used for real-time PCR analysis (reference gene: *Rpl13a*). {#pone-0036964-t002-2} GENE NAME PRIMER SEQUENCE ----------- ----------------- ---------------------------- *Rpl13a* S atccctccaccctatgacaa AS gccccaggtaagcaaactt *Fos* S gggacagcctttcctactacc AS gatctgcgcaaaagtcctgt *Egr1* S ccctatgagcacctgaccac AS tcgtttggctgggataactc *Nr4a1* S ctgtccgctctggtcctc AS aatgcgattctgcagctctt *Ier2* S ttgaatctcagggtcgaactc AS ggtagtgaaacggccttgaa *Btg2* S gcgagcagagactcaaggtt AS ccagtggtgtttgtaatgatcg *Jun* S agggacccatggaagttttt AS tttttctaggagttgtcagattcaaa *Fzd10* S tgctgcctgtgcataaactt AS cccccaggaaagctctttag *Galnt14* S tactatgcagctcggccttt AS caggttcagcctgttctcaa Focal adhesion kinase (FAK) is a non-receptor tyrosine kinase whose phosphorylation is generally detected within minutes after application of mechanical strain in a variety of cell types, including chondrocytes [@pone.0036964-Lee1]. However, in our chondrocyte-agarose model system, we found no compression-induced increase in FAK phosphorylation ([Figure 3](#pone-0036964-g003){ref-type="fig"}). Since TGF-β pathways activation was once reported as part of the cartilage response to mechanical strain [@pone.0036964-Neu1], we analysed Smad phosphorylation with or without compression of the chondrocytes in agarose. Mechanical stimulation promoted Smad2 phosphorylation, mainly after 5 and 15 min of dynamic compression, whereas there were no differences in the phosphorylation levels of Smad1/5/8 between compressed and uncompressed samples ([Figure 3](#pone-0036964-g003){ref-type="fig"}). Results from four independent experiments revealed the relatively high intensity (2.5-fold induction) and the great reproducibility (p\<0.01) of this early event of Smad2 activation ([Figure 3](#pone-0036964-g003){ref-type="fig"}). Finally, the time-dependent activation of the MAPK and canonical TGF-β/Smad pathways demonstrated that the compression regimen we applied to our chondrocyte-agarose model system was sufficient to trigger a cellular response at the molecular level. These pathways, independently or in synergy, may induce changes in the expression of genes that are important in the early responses of chondrocytes to mechanical signals. Confirmation of the mechanosensitive character of members of the AP-1 transcription factor family and *Egr1* {#s2c} ------------------------------------------------------------------------------------------------------------ Next, we undertook a microarray analysis to detect gene expression modulation in response to 30 min of dynamic compression. Observed gene expression regulation was thus putatively downstream the activation of the MAPK and TGF-β/Smad pathways observed after only 15 min of compression. An extensive microarray analysis was performed on four independent pairs of compressed and uncompressed experiments. We focused on highly responsive genes: a 2-fold change threshold for up- and down-regulation was applied (p\<0.01). Of the 20 transcripts with a difference of 2-fold or more, 9 transcripts corresponded to up-regulated genes and 11 to down-regulated genes ([Table 1](#pone-0036964-t001){ref-type="table"} and [Figure 4](#pone-0036964-g004){ref-type="fig"}). Interestingly, 8 of the up-regulated genes were known transcription factors. The most responsive gene was *Fos*, with a compression-induced over-expression of about 9-fold (p\<0.001) and *Egr1* showed a compression-induced up-regulation of about 4-fold (p\<0.001). Furthermore, the concomitant stimulation of *Jun, Junb* and *Fosb*, which are all genes coding for members of the AP-1 transcription factor family, and *Egr1* has also been reported to occur as an early event in diverse models of compression for skeletal cells [@pone.0036964-Fitzgerald1]--[@pone.0036964-Papachristou1]. Identification of new candidate mechanosensitive genes {#s2d} ------------------------------------------------------ In addition to those mentioned above, genes showing a difference of 2-fold or more with a p-value of less than 0.01 are listed in [Table 1](#pone-0036964-t001){ref-type="table"} ([Figure 4](#pone-0036964-g004){ref-type="fig"}, Panel A). Regarding the up-regulated transcription factor-encoding genes, *Egr2* and *Btg2* are members of the early growth response gene family and *Nr4a1* encodes a nuclear receptor. The last up-regulated gene in the list was *Ier2*, another early gene inducible by growth factors. Therefore, all up-regulated genes in this list are already known as "immediate early genes". The list of the down-regulated genes appeared more diversified. *Slc16a3*, *Galnt14*, *Osap* and *Slirp* code for proteins involved in cell metabolism, *Acin1* and *Msln* are genes related to cell death, *Sdpr/Cavin-2* encodes a caveolar protein, *Malat1* corresponds to a non-coding RNA and *Fzd10* and *Gigyf2* encode signalling molecules. No information is available in databanks on the predicted protein encoded by *Wdr60*. To validate the expression profiles obtained by microarray analysis, real-time PCR was used to compare the mRNA expression levels in compressed and control chondrocytes. We examined eight genes and confirmed the same gene expression modulation pattern as the microarray analysis ([Figure 4](#pone-0036964-g004){ref-type="fig"}, panel B). These observations indicated that our experimental procedure reliably identified putative mechanosensitive genes. In addition, the microarray analysis revealed many other candidate mechanosensitive genes when the fold change threshold was lowered from 2 to 1.4 (p\<0.01, [Data S2](#pone.0036964.s002){ref-type="supplementary-material"}). This extended dataset included 325 genes, with 48 up-regulated (i.e. 15%) and 277 down-regulated genes (i.e. 85%). The early response of chondrocytes to compression is thus generally characterised by down-regulation of gene expression ([Figure 5](#pone-0036964-g005){ref-type="fig"} Panel A). The presence of numerous transcription factors on the short list of highly responsive genes suggests that our cell model system was suitable for exploring the early events of mechanotransduction. We sought to further confirm this hypothesis by using the extended dataset of mechanosensitive genes. Hence, we analysed this extended list using PANTHER classification system to cluster candidate genes into relevant categories regarding signal transduction. From the extended dataset, 212 coding transcripts were eligible for functional annotation, of which 41 were up-regulated and 171 were down-regulated proteins ([Data S3](#pone.0036964.s003){ref-type="supplementary-material"}). Transcription factors were the most over-represented class of proteins (36 proteins, p\<0.001) and when pooled with DNA-, RNA- and nucleic acid-binding proteins (63 proteins), they represented 30% of the modulated proteins detected here ([Figure 5](#pone-0036964-g005){ref-type="fig"} Panel B). In addition, 20 proteins belonged to the protein class grouping kinases, phosphatases and kinase regulators, and 25 proteins belonged to the protein class grouping receptors and receptor-binding proteins ([Figure 5](#pone-0036964-g005){ref-type="fig"} Panel B and [Data S4](#pone.0036964.s004){ref-type="supplementary-material"}). Since these data strongly suggest that chondrocytes are involved in signal transduction mechanisms, the dataset of the 212 functionally annotated proteins was further analysed using the PANTHER and Pathway Express systems to identify over-represented signalling pathways. Several signalling pathways, such as Wnt, TGF-β, or MAPK pathways, were prominent, although statistical support was modest (data not shown). Altogether, our results demonstrate the relevance of the extended list of modulated genes for identifying new actors or targets involved in chondrocyte mechanotransduction. Discussion {#s3} ========== Validation of the chondrocyte-agarose construct as a model for identifying the mechanosensitive response typical of chondrocytes {#s3a} -------------------------------------------------------------------------------------------------------------------------------- The aim of this study was to explore the molecular-level response of chondrocytes to dynamic compression using a model system we previously developed [@pone.0036964-Bougault1], [@pone.0036964-Bougault2]. Because sensing and response to external mechanical stimuli by cells is controlled by cell-matrix interactions, we carefully examined --- before performing the compression experiments --- the extracellular matrix proteins and cellular receptors synthesised by chondrocytes in agarose. Western blot analysis extended our previous immunohistochemistry studies [@pone.0036964-Bougault1], [@pone.0036964-Bougault2] and confirmed that chondrocytes produced a cartilage-characteristic matrix during the pre-culture period. Regarding type II and type IX collagen production, the presence of cross-links in the newly formed matrix indicated that these chondrocytes were able to synthesise enzymes necessary for proper maturation and stabilisation of collagen molecules and their packing into collagen fibrils. In addition, in our model system, chondrocytes expressed the collagen-binding integrin α10 [@pone.0036964-Camper1]. Therefore, the chondrocyte-agarose model system used in this study made it possible to examine the molecular events underlying mechanotransduction, which probably occur during typical chondrocyte-cartilage matrix interactions. Since chondrocyte response to mechanical stimulation is affected if chondrocytes dedifferentiate prior to compression [@pone.0036964-Wiseman1], [@pone.0036964-Das1], we also carefully examined the chondrocyte phenotype in our model system. Western blot analysis of type I, II and IX collagens, α10 and α11 integrins and Sox9 extends our previous studies [@pone.0036964-Bougault1], [@pone.0036964-Bougault2] and confirmed that chondrocytes maintain a well-differentiated phenotype in our model system. Agarose hydrogel cultures have already been used to enhance chondrocytes in other models [@pone.0036964-Benya1], [@pone.0036964-Aydelotte1]; the challenge here was to use freshly isolated mouse cells and to obtain a complete differentiated phenotype after one week of culture. Suitability of our model system for studying the early events of mechanotransduction {#s3b} ------------------------------------------------------------------------------------ In the chondrocyte-agarose constructs, the levels of phospho-FAK were not significantly different in compressed compared to uncompressed cells, contradicting numerous published results showing a rapid activation of FAK following various mechanical stimuli. One possible explanation is that, in contrast to our chondrocyte-agarose constructs, cells where not embedded in a 3D environment. It is well known that cells in 3D systems form matrix adhesions that are not the same as their 2D counterparts [@pone.0036964-Wozniak1]. We detected transient activation of ERK1/2 and p38 in response to mechanical stress, as expected from previous studies [@pone.0036964-Fanning1]--[@pone.0036964-Hung1]. We also found that *Fos* and *Jun* family members and *Egr-1* gene expressions were activated after 30 min of compression, shortly after the primary activation of the MAPK pathway. These results are very consistent since *Fos*, *Jun* and *Egr-1* are downstream targets of the MAPK pathway activated by compression in chondrocytes [@pone.0036964-DeCroos2], [@pone.0036964-Papachristou2]. In addition to *Fos*, *Fosb*, *Jun* and *Junb*, the *Atf3* gene was also stimulated 1.47-fold by compression ([Data S2](#pone.0036964.s002){ref-type="supplementary-material"}). This modulation is in good agreement with the modulation observed for AP-1 genes since Atf3, a transcription factor known to be induced in stress responses, forms heteromers with Jun members for its transcriptional activities [@pone.0036964-Hai1]. Overall, the microarray analysis revealed that very few gene expression levels were modulated more than 2-fold, suggesting that dynamic compression triggered modest regulatory events. All the up-regulated genes in this list are already known as "immediate early genes". Examination of the very early events of dynamic compression reduces the risk of interpreting the result of feedback signalling. The presence of numerous transcription factors among the 20 most responsive genes was consistent with a high frequency of genes with a \>1.4-fold change in expression that code for proteins linked to signal transduction and gene expression regulation. These results further demonstrate that our model system is useful for studying mechanotransduction early events. Characteristic TGF-β signalling is activated by dynamic compression {#s3c} ------------------------------------------------------------------- Only a few studies have reported activation of TGF-β/Smad signalling as an early event in cellular mechanotransduction. Osteoblasts and the Saos-2 osteoblastic cell line respond to mechanical stimulation by increasing the activation of bone morphogenetic protein (BMP) receptor substrates, Smad1/5 [@pone.0036964-Mitsui1]--[@pone.0036964-Kido1]. Likewise, Smad2/3 phosphorylation increases when umbilical cord progenitor cells are stretched [@pone.0036964-Turner1]. Regarding chondrocytes, only one immunohistochemistry study has shown Smad2/3 activation in specific regions of bovine articular cartilage subjected to 5 min of shear stress [@pone.0036964-Neu1]. In our study, a Western blot analysis showed that Smad2, but not Smad1/5/8, was activated by dynamic compression, thus confirming that activation of TGF-β/Smad signalling represents an early response of chondrocytes to mechanical loading. Chondrocytes cultured in agarose gel secrete TGF-β [@pone.0036964-Tschan1] and this protein is secreted by various cells --- including chondrocytes --- as part of a latent complex that associates with matrix proteins such as fibrillin, proteoglycans, and fibronectin [@pone.0036964-Chaudhry1]--[@pone.0036964-Hyytiainen1]. One component of the latent complex, the latency-associated protein, interacts directly with integrins, especially αvβ5. Myofibroblasts cultured on stiff matrices can exert tension on the latent complex through integrins, causing conformational changes and the release of sequestered TGF-β in an active form [@pone.0036964-Wells1]. Although we did not measure the release of active TGF-β, it is possible that dynamic compression on chondrocyte-agarose constructs causes the mechanically driven release of soluble TGF-β which then binds to its receptor and subsequently triggers signalling as exemplified by Smad2 phosphorylation. Regardless of the exact mechanism of TGF-β activation in our cell model system, the microarray analysis confirmed the involvement of TGF-β signalling in the chondrocyte response to dynamic compression. For instance, *Htra1*, a gene coding for a serine protease that inhibits TGF-β signalling [@pone.0036964-Oka1] and *Arkadia/Rnf111*, a gene coding for an ubiquitin ligase involved in Smad2/3 regulation [@pone.0036964-Mavrakis1], were down-regulated under dynamic compression (1.47-fold and 1.66-fold, respectively, [Data S2](#pone.0036964.s002){ref-type="supplementary-material"}). Moreover, *Cyr61* was up-regulated by 1.64-fold. *Cyr61* is an important regulator of chondrogenesis and a member of the CCN family that includes connective tissue growth factor (Ctgf) [@pone.0036964-Wong1]. *Cyr61*, like *Ctgf*, is up-regulated in fibroblasts cultured under mechanical stress within a 3D collagen gel [@pone.0036964-Schild1]. Because *Cyr61* expression is activated as an early response to TGF-β [@pone.0036964-Brunner1], it is possible that the observed *Cyr61* up-regulation results, at least in part, from the activation of TGF-β signalling triggered by dynamic compression. Clearly, the interplay between growth factors, growth factor signalling and mechanotransduction is highly complex. Dynamic compression induces a general down-regulation of gene expression in chondrocytes {#s3d} ---------------------------------------------------------------------------------------- The microarray analysis revealed that around 85% of the 325 mechanosensitive identified genes were down-regulated. One possible explanation for the observed trend towards a reduction in RNA levels is an increase in mRNA decay. Interestingly, two genes coding for major proteins involved in RNA degradation, *Btg2* and *Zfp36*, were one of the relatively few genes up-regulated following dynamic compression (2.28-fold and 1.76-fold, respectively, [Data S2](#pone.0036964.s002){ref-type="supplementary-material"}). Btg2, a member of the Btg/Tob family of proteins, is a general activator of mRNA decay [@pone.0036964-Mauxion1], and Zfp36 binds to unstable mRNA and promotes their degradation [@pone.0036964-Clement1]. Zfp36 has been proposed as an inducible attenuator of growth factor signalling, by promoting degradation of rapidly induced genes and thus restricting the cell\'s responsiveness to stimulation [@pone.0036964-Amit1]. Btg/Tob factors are thought to facilitate the rapid switch to a new gene expression program by speeding up the degradation of previously made mRNAs [@pone.0036964-Mauxion1]. For example, Btg2 activates BMP signalling [@pone.0036964-Park1]. Down-regulation of gene expression may therefore represent a general mechanism in the early response of chondrocytes to mechanical stress. Dynamic compression affects various aspects of chondrocyte physiology {#s3e} --------------------------------------------------------------------- Independently of the general down-regulation observed in gene expression, careful examination of the extended list of modulated genes indicates that some of them have already been identified as mechanosensitive genes involved in different aspects of cellular physiology like in cartilage, e.g. *Biglycan (Bgn)*, an extracellular matrix protein [@pone.0036964-Wang1], *Mmp9*, a matrix metalloprotease [@pone.0036964-Kisiday1], *Cyr61*, a regulator of chondrogenesis from the CCN family [@pone.0036964-Schild1], *Cited2*, a transcription co-regulator playing a key role in shear-induced regulation of MMPs in chondrocytes [@pone.0036964-Yokota1], or in other tissues, e.g. *Thrombomodulin (Thbd)*, a gene coding for an anticoagulant factor [@pone.0036964-Rossi1], *Lmo4*, a fluid flow-responsive transcription factor [@pone.0036964-Ni1], *Ptgs1/Cox1*, a cyclooxygenase involved in the production of prostaglandin E2 [@pone.0036964-Wang2], or *Ahnak*, a protein involved in Ca^2+^ signalling pathways and regulated exocytosis [@pone.0036964-Kessler1], [@pone.0036964-Borgonovo1]. The regulation of these genes reflects diverse cellular responses to mechanical stimulation. Interestingly, a subset of modulated genes, including *Pcm1* [@pone.0036964-Keryer1], *Nek1* [@pone.0036964-Shalom1], *Smo* [@pone.0036964-Chen2], *Cdk5rap2* [@pone.0036964-Barrera1], *Spop* [@pone.0036964-Chen3], *Dync2h1* [@pone.0036964-Ocbina1], *Syne1/Nesprin1* [@pone.0036964-Dawe1], *Topors* [@pone.0036964-Chakarova1] and Wnt signalling molecules [@pone.0036964-Lancaster1] such as *Fzd10, Sfrp1, Rspo3/Cristin1*, are linked to ciliary function ([Data S2](#pone.0036964.s002){ref-type="supplementary-material"}). The primary cilium has long been hypothesised to function as an antenna for chondrocytes to sense the biomechanical environment, as in renal cells [@pone.0036964-Nauli1], [@pone.0036964-Praetorius1]. Using the same chondrocyte-agarose constructs as those used here, Wann *et al.* have just provided the first direct experimental evidence that the primary cilium mediates mechanotransduction through control of calcium signalling in compressed chondrocytes [@pone.0036964-Wann1]. Previously, using bovine chondrocyte-agarose constructs and confocal microscopy, McGlashan *et al.* showed that the application of cyclic compression affects cilia length in a time-dependent manner [@pone.0036964-McGlashan1]. In addition, mechanical forces have been reported to play a role in primary cilia assembly/disassembly *in vitro* in other cell types [@pone.0036964-Iomini1], [@pone.0036964-Resnick1]. These observations are correlated with *in vivo* studies, where the presence or absence of cilia is linked to the intensity of shear stress in blood vessels [@pone.0036964-VanderHeiden1]. Therefore, the mechanosensitivity observed here for the subset of cilium-related genes may represent an early signal triggered by chondrocytes to adapt the length and/or function of the primary cilium in response to mechanical loading. Nevertheless, part of the RNA transcriptome corresponds to RNAs that do not code for proteins, referred to as non-coding RNAs (ncRNAs) [@pone.0036964-vanBakel1]. Microarray screening identified two down-regulated long ncRNAs in compressed chondrocytes: *Xist* and *Malat1* (1.57-fold and 1.42-fold, respectively, [Data S2](#pone.0036964.s002){ref-type="supplementary-material"}), which are two of the three large non-coding transcripts present in mammalian nuclei [@pone.0036964-Hutchinson1]. Furthermore, *Dicer1*, a gene coding for an endoribonuclease that processes pre-miRNAs into siRNAs [@pone.0036964-Kim1], was down-regulated by 1.81-fold upon compression. In particular, recent studies have shown that miRNA can control expression of alternative splicing regulators [@pone.0036964-Kalsotra1] and *Malat1* can control the activity of some miRNAs [@pone.0036964-Tripathi1]. This is particularly interesting because alternative splicing events have been recorded in bone following mechanical loading [@pone.0036964-MantilaRoosa1]. These findings suggest that modulation of ncRNA expression is part of the molecular response to mechanical stress. Moreover, it is possible that these ncRNAs participate in the regulation of pre-mRNA splicing in response to compression. Concluding remarks {#s3f} ------------------ The aim of this study was to perform an integrated analysis of mechanotransduction in chondrocytes at the gene and protein level. The originality of our analysis was to investigate early molecular events triggered by dynamic compression. Our study reveals that, in addition to the well-known involvement of the MAPK-signalling pathway in the chondrocyte mechanotransduction response, TGF-β signalling may also play a prominent role. In addition, our microarray analysis results provide new molecular insight into how chondrocytes sense dynamic compression. The candidate mechanosensitive genes identified here can serve as starting points for future investigations of mechanotransduction in chondrocytes. The availability of genetically modified mice offers an opportunity to study the impact of gene modification in chondrocyte mechanotransduction using the cell model system presented here. Ultimately, identifying candidate mechanosensitive genes can provide important information not only for the molecular understanding of mechanotransduction in chondrocytes, but also for cartilage engineering. For example, agarose (or agarose-alginate) hydrogels constitute clinically potential scaffolds for autologous chondrocyte implantation [@pone.0036964-Barlic1] and mechanical conditioning can be used to stimulate *in vitro* chondrocyte biosynthesis in 3D scaffolds before implantation. Therefore, mechanosensitive targets can help optimise mechanical conditioning for cartilage reconstruction. Materials and Methods {#s4} ===================== Ethics statement {#s4a} ---------------- Mouse care and treatment were conducted in accordance with institutional guidelines in compliance with national and international laws and policies. This study was specifically approved by our local ethics committee (Authorization n°69387416 given by the French Prefecture du Department du Rhone). Antibodies {#s4b} ---------- For type I and type II collagens, polyclonal rabbit antibodies against mature collagens were used (Novotec; references 20151 and 20251, respectively; used at 1∶2000 and 1∶5000, respectively). Monoclonal antibody (mAb) against collagen IX (23-5D1; 1∶6000) was a gift from Bjorn Olsen (Boston, MA). Polyclonal antibodies against α10 (1∶2000) and α11 (1∶4000) integrins were from Cartela AB (a gift from Evy Lundgren-Akerlund, Lund, Sweden). Antibodies against Phospho-Smad1/5/8 (\#9511), Phospho-Smad2 (\#3101), Phospho-ERK1/2 (\#9101), Phospho-p38 (\#9251), Smad2/3 (\#3102), ERK1/2 (\#9102), p38 (\#9212) and anti-rabbit IgG horseradish peroxidase (HRP)-linked antibodies were purchased from Cell Signaling Technology (all 1∶1000). Rabbit mAb to Smad1 (1649-1) and Smad5 (1682-1) were from Epitomics and used both 1∶1000 in mixture. Anti-Sox9 polyclonal antibody (AB5535, 1∶2000) and anti-FAK monoclonal antibody (clone 4.47, 1∶5000) were purchased from Millipore. Polyclonal rabbit antibodies against phosphoY397-FAK (1∶1000) were obtained from Biosource-Invitrogen. Anti-actin monoclonal antibodies (A5060, 1∶800) were purchased from Sigma-Aldrich. Anti-mouse (170-6520) or rabbit (170-6518) IgG-alkaline phosphatase conjugates and anti-mouse IgG-HRP conjugates (170-6516) were purchased from Bio-Rad, all used 1∶5000. Chondrocyte isolation and 3D culture {#s4c} ------------------------------------ Embryonic mouse chondrocytes were isolated from the costal cartilage of day 17.5 post-coitum mice. Like articular cartilage, rib cartilage is a hyaline-type cartilage. Immediately after enzymatic isolation, cells were embedded in 2% agarose gels at a density of 2×10^6^ cells/mL as described [@pone.0036964-Bougault2]. Chondrocyte-agarose gels were punched to form cylindrical constructs of 13 mm in diameter and 3 mm in thickness. They were then cultured in the wells of Biopress™ compression plates (Flexcell international) for 6 days in 5% CO~2~ at 37°C. The Dulbecco\'s modified Eagle\'s medium/Ham\'s F-12 culture medium was changed daily as previously detailed [@pone.0036964-Bougault2]. Serum was progressively substituted with insulin-transferrin-selenium and cultures were gradually supplemented with ascorbic acid (up to 20 µg/mL). Used as positive controls of dedifferentiation, other mouse chondrocytes were cultured in monolayer for one week, passaged once and cultured for another week. DNA content {#s4d} ----------- DNA quantification was performed using the Hoechst 33258 (Fluka) DNA stain. The calibration curve was obtained using a DNA standard solution (Invitrogen). Application of dynamic compression {#s4e} ---------------------------------- Chondrocyte-agarose constructs were subjected to compression using a previously characterised model system [@pone.0036964-Bougault2], [@pone.0036964-Bougault1]. The FX-4000C Flexercell Compression Plus System (Flexcell International) was used to apply dynamic compressive strain to agarose gels. Compressed constructs were subjected to cyclical compression ranging from 20 kPa to 40 kPa in a square waveform at a frequency of 0.5 Hz ([Figure 2](#pone-0036964-g002){ref-type="fig"}) for 5, 15 or 30 min. Control constructs were uncompressed. Protein extraction and analysis by Western blotting {#s4f} --------------------------------------------------- Protein extraction from the agarose gels was performed with special care to avoid any modification in the phosphorylation state of proteins [@pone.0036964-Bougault2]. For Western blotting, proteins were separated on 10% or 4--12% polyacrylamide gradient mini-gels and transferred to PVDF membranes (Millipore). The membranes were probed with the appropriate primary antibodies, washed and incubated with HRP- or alkaline phosphatase-conjugated anti-mouse or anti-rabbit IgG. After multiple washes, bound antibodies were detected on x-ray films using a Bio-Rad Immun-star or WesternC chemiluminescent substrate. The membranes probed with antibodies to collagens or integrins were sequentially re-probed after stripping (Re-Blot Plus Strong, Chemicon). A final re-probing with anti-actin antibodies served as a loading control. The membranes probed with antibodies to phospho-proteins were stripped and re-probed with antibodies that recognise all forms of the protein in question. Phosphorylation levels were quantified by densitometry using ImageQuant software (Molecular Dynamics). For each protein, the ratio of phospho-protein band intensity to the total protein band intensity was calculated and mechanically-induced phosphorylation was normalised to uncompressed controls. DNA microarray analysis {#s4g} ----------------------- Total RNA was extracted from chondrocyte-agarose constructs as previously described [@pone.0036964-Bougault2]. To ensure a sufficient quantity of RNA, extractions from six similar constructs were pooled. To ensure quality of RNA in each sample, integrity and purity were assessed using a capillary electrophoresis system (Agilent Bioanalyser, Agilent Technologies). DNA microarray analysis was performed on four independent experiments to compare gene expression levels between compressed (30 min compression) and uncompressed (control) constructs. Hybridisation was carried out following the Two-Colour Microarray-Based Expression Analysis protocol (Agilent Technologies) and 500 ng or 1 µg of purified total RNA were used for linear amplification. The resulting labelled cRNA from a compressed sample was co-hybridised with the labelled cRNA of the corresponding control sample to the Agilent Mouse Genome CGH Microarray 44 K probe set (Agilent Technologies). Each co-hybridization was performed several times starting from different total RNA preparations and using a dye swap. Each microarray contained 44,000 sequences spanning the whole mouse genome and control probes. The microarrays were scanned using an Innoscan 700 Microarray Scanner (Innopsys) at 532 nm (for detection of the Cy3 dye) and 635 nm (Cy5 dye). The resulting image was analysed using Mapix v3.1 software. The signal intensity of each spot was acquired and non-exploitable spots were filtered out. The statistical analysis and normalisation steps were done using the Limma (Linear Models for Microarray Data) package [@pone.0036964-Smyth1] in the statistical language R [@pone.0036964-Ihaka1]. The "global Loess" function was applied to the data to correct for bias. Normalised data were then averaged between direct and swapped comparisons to calculate values of differential expression and expression level. A classification of statistically significant modulations was obtained using a moderated Student\'s *t*-test with a Bayesian false-discovery rate approach [@pone.0036964-Lnnstedt1]. Analysis of genes associated with cell function was carried out using the PANTHER (Protein ANalysis THrough Evolutionary Relationships) classification system (<http://www.pantherdb.org>) and Pathway-Express (<http://vortex.cs.wayne.edu/Projects.html>) profiling system to identify protein categories or biological pathways which may be associated with modulated gene expression (with M\>0.5 and p\<0.01). Confirmation of modulation in gene expression by real-time PCR {#s4h} -------------------------------------------------------------- Real-time PCR analysis was performed on three independent experiments as previously described [@pone.0036964-Bougault2]. Levels of gene expression were determined by using the comparative Ct method with *RPL13a* gene as the endogenous control. Primer pairs used in this study are described in [Table 2](#pone-0036964-t002){ref-type="table"}. Dissociation curves were conducted at the end of each run to verify the absence of DNA contamination. Student\'s *t*-test (paired, two-tailed) was used for statistical analysis. Supporting Information {#s5} ====================== ###### ERK1/2 and p38 transient compression-induced activation in chondrocyte-agarose constructs. [Legend:]{.ul} Chondrocytes cultured in agarose for 6 days underwent dynamic compression (+) or were not compressed (−) for the indicated times and phosphorylation levels of ERK1/2 and p38 were analysed on Western blots. (A) Representative blots. (B) Densitometric analysis was performed on four (5 and 15 min) or three (30 min) independent experiments. The phospho-MAPK to total MAPK ratio was calculated and mechanically induced phosphorylation was normalised to uncompressed controls. Bars represent the compression-induced phosphorylation modulation (mean fold change +/− SD), either up-regulation (red) or down-regulation (green) (\* p\<0.05). (TIF) ###### Click here for additional data file. ###### Results from DNA microarray analysis: gene expression levels in compressed samples were compared to uncompressed control samples (fold change \>1.4 and p-value \>0.01). (XLS) ###### Click here for additional data file. ###### Results from DNA microarray analysis: modulated coding transcripts eligible for functional annotation (PANTHER analysis). (XLS) ###### Click here for additional data file. ###### Results from DNA microarray analysis: modulated coding transcripts sorted by protein class (PANTHER analysis). (XLS) ###### Click here for additional data file. We thank Daniel Hartmann (Novotec, Lyon, France) for providing the polyclonal antibodies against type I and II collagens and Evy Lundgren-Akerlund (Cartela AB, Lund, Sweden) for providing the monoclonal antibody against α10 integrin and polyclonal antibodies against α11 integrin. We also thank the technical facilities of SFR BioSciences Gerland-Lyon Sud (US8/UMS3444) for the quantitative PCR analyses. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work was financially supported by the CNRS, Université de Lyon, and ANR TecSan 2006. CB was supported by the French Ministry of Research. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: CB EAF FMG. Performed the experiments: CB EAF AP LH. Analyzed the data: CB EAF DH. Wrote the paper: CB EPG MDC FMG.

A number of individuals are struggling with the category of low-quality-status medically unexplained symptoms (MUSs)[@b1] and the morbidity rate is 1.6--70%, 2.4--87% and 4.6--18% in young, middle aged and elderly populations dividually from 1966 according to the investigation of MUSs[@b2]. Meanwhile, MUSs has been defined as "suboptimal health" (*Yajiankang* in China) in traditional Chinese medicine explained as a borderline state between health and disease. To our disappointment, suboptimal health is more difficult to be diagnosed with a biological disease because of only vague changes in function but no clear signs of organic disease[@b3], which present as low energy level, loss of vitality, altered sleeping patterns and so on[@b4]. It could be parallel with symptoms of chronic fatigue syndrome (CFS)[@b5][@b6] or "THE THIRD STATE" or "GRAY STATE" raised by the former Soviet scholar prospectively. Also state of suboptimal health included several different subtypes, and as a subtype, psychological suboptimal health has attracted more attentions. Psychological suboptimal health is a prevalent state with a pathophysiological mechanism that is extremely complicated and poorly understood. Although it exhibits objective symptoms without a specific disease and it cannot reach the standards of psychiatric diagnosis such as depression and anxiety neurosis estimated by scores on diagnostic scales, the 10th edition of international Classification of diseases (ICD-10), Classification and Diagnostic Criteria of Mental Disorders in China-Third-Edition (CCMD-3), the 4th edition Diagnostic and Statistical Manual of Mental Disorder (DSM-IV)[@b7], for instance, but we must not ignore potential hazards. As the intermediate state between mental health and psychological disease, the emblematical symptoms indicating someone immersed in the state contain out of humor, panic, negative emotion, easy to get angry, losing interest, insomnia, impaired concentration and so on. What's more, psychological suboptimal health, can result in crippling quality of life and raising costs in medical due to frequent, unnecessary visits to healthcare facilities for checkups and diagnoses. In virtue of potential damage and ambiguity in pathomechanism, psychological suboptimal health has garnered increasing attention and has been described in experimental reports and defined as "subthreshold depression"[@b8][@b9] or "subthreshold obsessive-compulsive disorder"[@b10], the concepts of which are very similar to psychological suboptimal health. In addition, Blackwood has divided chronic fatigue syndrome (CFS) into two states of psychological and physical in the survey[@b11] and the psychological state of CFS is parallel to psychological suboptimal health. These researchers laid particular emphasis on epidemiologic characteristics, and unfortunately, studies on psychological suboptimal health and the pathogenetic mechanisms involved are rare relatively. What's more, a diagnostic criterion that effective and widely accepted has not been established at home and abroad. In China, scholars and doctors prefer to use a variety of scales and questionnaires to diagnose the intermediate state, including the Symptom Checklist 90 (SCL-90), Cornell Medical Index (CMI), mental functions decline index health assessment (MDI), or other self-made evaluations, combining with subjective judgment. To a certain extent, these approaches are authentic for diagnosis but at the same time, rate of missed diagnosis and misdiagnosis is not satisfying, owing to inconformity indigestibility of scales, concealment of patients, doctors relying too much on experiences and subjective judgment. So diagnostic methods that objective, high reliability and easy to operate need to be developed urgently. Through the approach of evaluating the significant differences at the molecular level, novel biomarkers or a biomarker panel then further could be discovered in the plasma samples from patients with psychological suboptimal health and healthy controls. And they would be used in clinical diagnosis after further validation and evaluation. Moreover, metabolomics technologies are the principal approaches for diseases biomarkers discovering. Systems biology[@b12] including genomics, proteomics, and metabolomics can be utilized in research of diseases[@b13][@b14][@b15]. As an important component of systems biology, metabolomics technologies have become a powerful tool and platform for detecting endogenous small compounds[@b16][@b17] as candidate biomarkers closely related to pathological and physiological processes of diseases and carrying rich information concerning metabolism as key pathways[@b18]. It may help to unravel the mechanisms of disease occurrence and progression on the metabolic level[@b19]. Also major metabolomics technologies were based on H-nuclear magnetic resonance (^1^H-NMR)[@b20][@b21][@b22], liquid chromatography−mass spectrometry (LC−MS)[@b23], and gas chromatography−mass spectrometry (GC−MS)[@b24][@b25]. Furthermore, it was noteworthy that ^1^H-NMR is the earliest method used in metabolomics analysis with the advantages of possessing a rapid, non-destructive, high-throughput system[@b26], and still is widely used to detect biomarkers of diseases for clinical diagnosis[@b27]. In common sense, medicine should be applied to improve clinical symptoms, but few chemical drugs was suitable. As a well-known traditional Chinese prescription, Baihe Dihuang Tang (BDT) is described initially in "Synopsis of Golden Chamber" (Jinkui *Yaolue*) consisting of two herbal medications: lily bulb (Bulbus Lilii) and rehmannia root (Radix, Rehmanniae). It is used to treat mental instability, absentmindedness, insomnia, and dysphoria in clinical. These major symptoms are closely associated with early depression disorder[@b28] and also perform in psychological suboptimal health state. Furthermore, BDT has been widely used and significantly improved the symptoms of psychological suboptimal health due to a deficiency of *yin (Yin Xu*), according to the theory of TCM and also BDT was applied as the intervention measure in our experimentation. As far as we know, just several published papers were involved in the research of suboptimal health state with meatabolomics and achieved some results[@b29][@b30][@b31] but the study of the psychological suboptimal health state taking advantage of metabolomics technology is almost a blank. In the present study, plasma metabolomics based on ^1^H-NMR coupled with multivariate statistical analysis are used for investigating metabolites with significant differences at a molecular level and screening potential biomarkers. What the goal is to develop a biomarker panel from the biomarkers through correlation analysis, drug intervention of BDT and evaluation of diagnostic ability that can be used for clinical diagnosis ultimately. A biomarker panel would provide support for objective diagnostic laboratory tests for psychological suboptimal health. Results ======= Clinical information of participators ------------------------------------- According to the scale and clinical diagnosis, 22 patients being in state of psychological suboptimal health and 23 volunteers acting as the healthy control group were screened. From the SCL-90 scores of 143.9 ± 22.6 and 90 as the mean ± SD form and the filter factors mentioned above, a significant difference between two groups was confirmed in clinical. The basic clinical data for the participators are shown in [Table 1](#t1){ref-type="table"}. ^1^H-NMR spectra of plasma -------------------------- To identify the small endogenous molecules in plasma and survey the level varieties in different states, all samples were processed, and typical Carr-Purcell-Meiboom-Gill(CPMG) ^1^H-NMR spectra of plasma from groups of psychological suboptimal health was depicted ([Fig. 1](#f1){ref-type="fig"}). 32 metabolites were identified according to the Human Metabolome Database (HMDB: <http://www.hmdb.ca/>), the Chenomx NMR suite (Chenomx Inc, Edmonton, AB, Canada) and previously published references[@b32][@b33][@b34]. For a better visualization, the vertical scales for the 2D spectra, including 1H-1H correlation spectroscopy (1H--1H COSY) and 1H--13C heteronuclear multiple quantum correlation (1H--13C HMQC) spectra ([Supplementary Figures S1 and S2](#S1){ref-type="supplementary-material"}) were adjusted based on metabolite peaks. Plasma spectra from healthy controls and the BDT group are shown in [Supplementary Figures S3 and S4](#S1){ref-type="supplementary-material"}. The metabolites identified in the spectra were listed in [Table 2](#t2){ref-type="table"}. Several amino acids, glucose, organic acids, lipids, choline were demonstrated in the spectra. Validation and assessment of the differences between groups ----------------------------------------------------------- With the purpose of demonstrating significant differences not only in the clinical scale scores, we analyzed the NMR spectra information using multivariable statistics. Metabolome difference by comparing the numerical integration was observed and partial least squares discrimination analysis (PLS-DA)-based profiling was employed to explore the intrinsic differences between the groups of psychological suboptimal health and mental health. The samples from different groups were separated and classified into two distinct clusters presented in the PLS-DA score plot ([Fig. 2A](#f2){ref-type="fig"}); each point represents an individual sample (to show the group clusters). The model parameters (R^2^X = 0.541, R^2^Y = 0.949, Q^2^ = 0.755) and the validated model (permutation number: 200) indicated no over fitting ([Fig. 2B](#f2){ref-type="fig"}), supporting the result. All of the results indicated the existence of differences between the two groups and the reliability of diagnosis according to the method with scales mentioned previously. Discovery and screening of potential biomarkers ----------------------------------------------- To identify changed metabolites and considering the high information content and complexity of the spectra, orthogonal partial least squares discriminant analysis (OPLS-DA) was used to amplify the subtle differences due to the abnormal state of psychological suboptimal health. The supervised model of OPLS-DA could develop a better separation into two clusters and contribute to the discovery of biomarkers. The group of psychological suboptimal health exhibited a perfect separation from the healthy controls in the OPLS-DA scores plot ([Fig. 2C](#f2){ref-type="fig"}), as well as in permutation tests and CV-ANOVA (*p* \< 0.05). All parameters indicating the model quality were listed in [Supplementary Table S1](#S1){ref-type="supplementary-material"}. The metabolites responsible for a significant contribution to the separation of two groups were indicated in the corresponding S-plot ([Fig. 2D](#f2){ref-type="fig"}) and marked with number containing glutamine, N-acetyl-glycoproteins, TMAO, citrate, phenylalanine, valine, isoleucine, tyrosine and glucose. The specific change trends that higher levels of glutamine, N-acetyl-glycoproteins, TMAO, citrate, tyrosine and phenylalanine and lower levels of valine, isoleucine, and glucose were revealed in [Table 3](#t3){ref-type="table"}. Furthermore, a heatmap plot with different color that green stands for low level and the red is opposite was constructed, from which we could observe the trends more visually ([Fig. 3](#f3){ref-type="fig"}). Screening biomarker panel ------------------------- ### Correlation analysis of potential biomarkers To investigate the relationship among the potential biomarkers, the levels in the plasma samples from patients and healthy controls were correlated using Spearman's correlation ([Fig. 4A](#f4){ref-type="fig"}) with Metaboanalyst 3.0[@b35], an online data tool. A positive correlation indicated the relationship of the metabolites with certain pathways that exerted influence on the state of psychological suboptimal health and was distinguished with a red color, whereas a negative correlation suggested the metabolites may be from different pathways and was indicated with a blue color[@b36]. Analysis of the correlation among these potential biomarkers can be used to identify a biomarker panel. Remarkably, citrate was positively correlated with phenylalanine, glutamine, tyrosine and TMAO between patients of psychological suboptimal health and healthy controls. In additional, phenylalanine levels were positively correlated with N-acetyl-glycoproteins, glutamine, tyrosine, TMAO and citrate. Further analysis using Pattern Hunter with Spearman coefficients was applied to identify the correlation between groups of healthy control and patients. Phenylalanine, glutamine, tyrosine, TMAO, N-acetyl-glycoproteins and citrate have been demonstrated a positive correlation, whereas a negative correlation of isoleucine, valine and glucose was observed between the two groups of different groups ([Fig. 4B](#f4){ref-type="fig"}). Correlation analysis of plasma metabolites displaying significant difference was performed to gain insight into the pathogenic characteristics and pathways involved. With a purpose of selecting biomarkers that were positively correlated with state of psychological suboptimal health and forming a biomarker panel, 6 metabolites containing phenylalanine, glutamine, tyrosine, TMAO, N-acetyl-glycoproteins and citrate were selected and defined as a biomarker panel from the 9 potential biomarkers. ### Drug intervention and validation Based on the significantly decreased frequency of clinical symptoms and scores of SCL-90 after treatment for 4 weeks (*P* \< 0.05), BDT exerted an obvious effect on improvement of symptoms, and 22 patients in state of psychological suboptimal health improved markedly These results are shown in [Table 1](#t1){ref-type="table"}. To obtain an overview of the metabolic responses to the actions of BDT, the PLS-DA (R^2^X = 0.15, R^2^Y = 0.941, Q^2^ = 0.531) trajectories ([Fig. 5A](#f5){ref-type="fig"}) of all of the spectra from plasma samples containing healthy controls, pre- and post-BDT-treated groups were analyzed and separated into three clusters as outstanding differentiation. In the scores plot, the BDT treatment group was close to the healthy control group. The trend of transformation suggested the disturbance of the plasma metabolic profile of patients and stabilization after BDT administration. The validated model indicated no over fitting ([Fig. 5B](#f5){ref-type="fig"}). Using the strategy mentioned previously, as could be observed in the PLS-DA scores plot (R^2^X = 0.403, R^2^Y = 0.894, Q^2^ = 0.687) (Figure S5A) and the validated model that indicated no over fitting (Figure S5B), the psychological suboptimal health group and the BDT-treatment group were clearly seen as separated. The OPLS-DA model ([Fig. 5C](#f5){ref-type="fig"}) and corresponding S-plot ([Fig. 5D](#f5){ref-type="fig"}) indicated that the levels of the potential biomarkers tended to recover to a normal level. The levels of valine, glutamine, TMAO and phenylalanine changed significantly and reversed to normal levels after BDT treatment (*P* \< 0.01, *P* \< 0.05). And the metabolites changed significantly mentioned above were labeled with number ([Fig. 5D](#f5){ref-type="fig"}). The *t*-test results of significant differences in these potential biomarkers and their changes after BDT administration are shown in [Table 3](#t3){ref-type="table"}. Permutation tests and CV-ANOVA (*p* \< 0.05) were also performed. All parameters indicating the model quality are listed in [Supplementary Table S1](#S1){ref-type="supplementary-material"}. As a result, BDT treatment showed the obvious effect on the biomarker panel that levels of glutamine, TMAO, and phenylalanine that changed significantly and also citrate, tyrosine and N-acetyl-glycoproteins exhibited a trend to normal levels. As a TCM for treating mental and emotional diseases, BDT drug intervention could demonstrate the high correlation between the biomarker panel and pathomechanism of psychological suboptimal health to a limited extent. Diagnostic capability evaluation of biomarker panel --------------------------------------------------- Biomarkers with higher sensitivity and specificity are expected to be developed. ROC analysis was applied to evaluate the accuracy of diagnosis based on the identified potential biomarkers or combinations and the area under the curve (AUC) of ROC; 0.5 \< AUC \< 0.7, 0.7 \< AUC \< 0.9, AUC \> 0.9 explain a low, fair, and superior accuracy of diagnosis, respectively. For most of the biomarkers, AUCs were \<0.8 ([Supplementary Figure S6 and Table S2](#S1){ref-type="supplementary-material"}), indicating a poor prediction probably in virtue of the inability of a single metabolite to predict a disease accurately or a small sample size. By selecting the metabolites demonstrating an AUC \> 0.7, some combinations of potential biomarkers, including the biomarker panel mentioned above that could provide higher predictive power than single one, were examined. Finally, the AUC of the biomarker panel reached 0.971. The ROC curves and AUCs of the combinations are shown in [Fig. 6](#f6){ref-type="fig"} and [Table 4](#t4){ref-type="table"}. The AUC of the biomarker panel containing 6 metabolites indicated the highest predictive ability and the highest correlation with psychological suboptimal health. In this study, methods of statistical analysis, correlation analysis, drug intervention and the ROC analysis were united, and a biomarker panel tightly correlated with psychological suboptimal health was identified and demonstrated. Combined with all the analysis, these findings revealed that the biomarker panel is reliable and robust and possess a perfect ability to separate psychological suboptimal health. In future, it would be a better diagnostic approach in clinical. Discussion ========== As we have known, few studies focus on establishing an objective and accurate diagnostic method of psychological suboptimal. Scales and questionnaires in public or self-made are applied in clinic widely, whereas an more credible standard of diagnosis has not been formulated yet. The existing circumstances of lack of objective laboratory diagnosis for early detection and curative effect evaluation index may bring about adverse effects in disease prevention such as depression or. As an exploration, this study applied NMR metabolomics in investigating the state of psychological suboptimal health that meaning "not yet ill" for the first time with the purpose of seeking out potential biomarkers or a biomarker panel highly related with the state and setting it as a laboratory diagnostic method in clinical. In this study, we discovered that a set of altered metabolites including amino acid (isoleucine, valine, phenylalanine, glutamine, and tyrosine), energy metabolism-related molecules (citrate and glucose) and other metabolism molecules (N-acetyl-glycoproteins and TMAO) that would be the potential biomarkers. A deeper insight of the internal relationship among the potential biomarkers and metabolic mechanisms closely related with state of psychological suboptimal should be gained and biological significance of potential biomarkers should be analyzed. We constructed systematic metabolic pathway analysis based on information obtained from the Kyoto Encyclopedia of Genes and Genomes Web site ([www.genome.jp/kegg/](http://www.genome.jp/kegg/)) and would be discussed in further details below. As a mental and emotional disorder, the out of control metabolic pathway highly interrelated with the state of psychological suboptimal health may relate with nervous system. And some perturbed significantly metabolites involved in neurotransmission including phenylalanine, tyrosine, valine and isoleucine were observed indeed. Phenylalanine is an essential amino acid absorbed from food that can be metabolized primarily in the liver into tyrosine, which is then used in dopamine (DA), norepinephrine (NE) and epinephrine synthesis in the nervous system and the adrenal medulla[@b37]. Disorder of phenylalanine metabolism s delays the process of phenylalanine translating into tyrosine and contributes to overbalanced levels of phenylalanine and tyrosine, and the equal phenomenon was also observed in the plasma of subjects in the psychological suboptimal health group in this study. Furthermore, researchers have shown that a higher level of phenylalanine would induce damage in the nervous system, resulting in hypokinesia, depression and psychogeny[@b38]. Previous research also suggested that phenylalanine was a large neutral amino acid that could affect 5-HT synthesis[@b39][@b40]. So we could deduce that a higher level of phenylalanine accompanying physical symptoms would indicate a state of psychological suboptimal health and imply that damages to the nervous system were in progress, and if it was ignored, mental disorder would emerge soon. In generally, valine and isoleucine are called branched-chain amino acids (BCAAs) because of their aliphatic side-chains. The decreased concentration of BCAAs in plasma could be an indication of the abnormal release of brain 5-HT that is highly related to central fatigue[@b41][@b42], which is in conformity with common symptoms of psychological suboptimal health in clinical that easy to get fatigued and memory deterioration. Also some metabolites at abnormality levels may be the precursor of neurotoxicity in nervous system, in this research, the major endogenous molecule we found was glutamine. As reported previously, glutamate is the primary excitatory neurotransmitter in the mammalian brain[@b43]. Through glia cells, glutamate is converted to glutamine and released into the extracellular fluid from which it is reabsorbed into presynaptic terminals and converted back to glutamate via the action of neuronal glutaminase. Glutamine and glutamate are inter-converted between neurons and astrocytes, which is necessary for glutamine homeostasis[@b44]. It induces neurotoxicity and is related to the neurobiology of depression if excessively released[@b45][@b46]. Also the increased level of glutamine in plasma may be a compensatory adaptation to counteract glutamate-induced neurotoxicity. This is in agreement with previous hypotheses[@b47][@b48]. Individuals in state of psychological suboptimal health are struggling with the condition of lack of vitality in clinical, in most instances and the appearance may indicate that metabolic disturbance of energy is highly related the pathomechanism. Citrate, as a potential biomarker which is related to energy metabolism, is a dominant intermediate of the tricarboxylic acid cycle (TCA). The higher level of citrate in the plasma samples of the subjects in the state of psychological suboptimal health is indicative of TCA cycle dysfunction. Also levels of the BCAAs containing valine and isoleucine declined sharply, suggesting their consumption in large quantities for energy supply[@b49], meanwhile isoleucine deficiency is marked by muscle tremors. Moreover an organism would be forced to produce ATP by anoxic respiration to adapt to the supply/demand imbalance because of deficient energy and the decreased level of glucose can be considered an indicator of the severity of the supply/demand imbalance. All the analysis of abnormal metabolic pathways energy related showed close relationship with clinical symptoms. Loss of appetite, a common symptom of psychological suboptimal health, has shown contact with abnormalities in gut microflora. Trimethylamine N-oxide (TMAO) is an oxidation product of trimethylamine (TMA) and a common metabolite in animals and human. In particular, TMAO is biosynthesized endogenously from TMA, which is derived from choline obtained from dietary lecithin or dietary carnitine. Several previous clinical studies have indicated that depressed patients display a disturbance of gut microflora, including concentration changes of metabolites such as TMAO, DMA and dimethylglycine[@b50]. Previous research also demonstrated that plasma choline is derived from phosphorylcholine by phosphotransferase. TMA could be converted by choline via gut microbiota and then detoxified through flavin monooxygenase in the liver, forming TMAO[@b51]. Therefore, it is plausible that the state of psychological suboptimal health caused a disturbance in gut microbiota colonies. Furthermore, we observed a higher level of N-acetyl-glycoproteins in the group of patients with psychological suboptimal health although most of the broad protein was eliminated by the method presented above and the resonances were suppressed by the CPMG pulse sequence[@b52]. Acetyl-glycoproteins are acute-phase proteins that can act as inflammation mediators[@b53] and the levels of these proteins increase immediately in response to external or internal challenges such as infection, inflammation, and stress[@b49] that are believed to be the cause of the state. Alterations in the levels of N-acetyl-glycoproteins may indicate that people have been suffering in an extreme environment and are developing psychological suboptimal health. This analysis would be the proof of close connection between N-acetyl-glycoproteins and extraneous factors leading to disease. All of the analysis above would be the foundation and deep proof of the relationship among the metabolites and pathological mechanisms as well as incentives. These metabolic changes and the associated pathways provide insights into the mechanisms involved in the development and progression of psychological suboptimal health. Furthermore, for the purpose of screening more representative biomarkers, methods of correlation analysis for selecting biomarkers as a biomarker panel and drug intervention for validating the close internal relations between the biomarker panel and the state were united. Then a biomarker panel containing phenylalanine, glutamine, tyrosine, citrate, N-acetyl-glycoproteins and TMAO was identified and high correlation with the state of psychological suboptimal was also demonstrated. As following, the ROC curve analysis for evaluation of clinical diagnosis ability was carried out. Small AUC of single one metabolite showed low diagnostic capability for the reason of small sample size or one metabolite cannot reflect comprehensively. But biomarker panel displayed the highest AUC (0.971) that show perfect diagnostic and recognition capability of psychological suboptimal health and would be used as an innovative diagnosis method. Finally, although a biomarker panel was sought out using ^1^H-NMR metabolomics, but a large number of clinical samples should be collected and technologies of GC-MS and LC-MS should be used to quantify these metabolites of the biomarker panel for the ultimate goal that the biomarkers can be applied in clinical diagnosis. Materials and Methods ===================== Ethical statement ----------------- All control and psychological suboptimal health subjects provided informed consent prior to the collection of any data. This research was approved by the Ethical Committee of the First Hospital of Shanxi Medical University in Taiyuan and was conducted according to the principles expressed in the Declaration of Helsinki. Written informed consents from all recruited participants were acquired. Subjects and assessment ----------------------- In this study, patients being the state of psychological suboptimal health (31--60 years) were filtrated from the traditional Chinese Medical Department of the First Affiliated Hospital of Shanxi Medical University as Baihe Dihuang Tang treatment group. Then age-and sex-matched mental health subjects were recruited to be the healthy controls. Briefly, patients were screened by items as follows:(1) totally scored ≥9 and ≤250 diagnosed by the scale of SCL-90; (2) cardinal symptom on the diagnostic criteria for deficiency of yin referring to the diagnosis curative standard of TCM disease; (3) not on any narcoleptic drugs within one year; (4) no mental disease, pregnancy, organic disease and allergic to TCM. The healthy controls should meet the standards: (1) score of SCL-90 should be at the point of 90; (2) no any previous history of neurological; (3) no systemic medical illness. Sample size calculation ----------------------- In the design of clinical trials, the number of participants was determined by the manipulators and the participators were made up of 30 patients and 30 healthy controls. Through screening outpatients in the hospital and recruiting healthy volunteers for 2 months, 30 patients and 30 healthy controls were included into the trial through the assessment standard mentioned above. Unfortunately, 8 patients were lost during the 4-week intervention with BDT with the potential reasons of the following: (1) medication cycle of 4 weeks was a little bit longer; (2) unable to endure the slow onset of TCM drug action; (3) not follow the doctor's advice and take other drugs not allowed in the trial. Moreover, 7 healthy controls fell off for the possible reasons followed: (1) suffering from a cold, inflammation or other diseases at the point of collecting samples; (2) not want to take part in this trial continuously; (3) not get to the hospital because of some unexpected situation. So at the end of the trial, samples of 22 patients and 23 healthy controls were used for analysis. BDT preparation process and dosage ---------------------------------- The medicinal plants used to prepare BDT decoction were purchased from the Chinese herbal medicine market in the city of An-guo, Hebei Province and authenticated by Professor Xue-mei Qin from Modern Research Center for Traditional Chinese Medicine, Shanxi University. The preparation was done in traditional Chinese Medical Department of the First Affiliated Hospital of Shanxi Medical University, where the standard machine and manipulators were performed according to the guidelines. Each dosage of BDT containing Lily bulb (30 g) and Rehmannia root (20 g) were soaked in water (1:8, w/v) for 30 min at room temperature and boiled for 1 h. The aqueous extract were filtered and concentrated to the volume of 200 mL, and then divided in two parts with the package automatically. The BDT was administrated to the patients with one dosage every day for 4 weeks and drinking or seafood was strictly prohibited in the case of the interference with this protocol. This clinical work was performed strictly and correctly in the First Affiliated Hospital of Shanxi Medical University, which is a first-class hospital with national clinical trials research center of new drugs (GCP center). Also the hospital has ethics committee and this work had been permitted. The manipulators of the research have been engaged in clinical work for many years, specializing in the treatment of mental disorders and participated clinical trials of new drugs on many occasions. Experimental program had been designed by the manipulators and they ensured the standardization of the experimental process according to the Good Clinical Practice. Plasma sample collection ------------------------ After the patients had fasted, 5 mL of venous blood was collected from all subjects in the psychological suboptimal health group into 10 mL heparin sodium tubes before and after 4 weeks of treatment. Blood was also collected from healthy controls before 4 weeks in the morning. Samples were centrifuged at 1250 × g for 15 min at 4 °C, after which each plasma sample was divided into equal aliquots and stored at −80 °C for subsequent analysis. Sample preparation ------------------ Plasma Samples were thawed at 0 °C in an ice-water mixture. First, 450 μl of plasma was mixed with 900 μl of analytical pure methanol, vortexed for 2 min, and then centrifuged at 16172 × g for 20 min at 4 °C to pellet proteins. After that, 1000 μl of supernatant was transferred into an EP tube. Another 900 μl of analytical pure methanol was added again, and the mixture was centrifuged at 16172 × g for 20 min at 4 °C for outright protein removal. Finally, a total of 1800 μl of supernatant was dried under nitrogen gas, and the dried samples were completely dissolved in 600 μl phosphate buffer solution in 100% D2O (0.2 M Na2HPO4/NaH2PO4, pD = 7.4) containing TSP (0.025%) to minimize chemical shift variations. The samples were then centrifuged (16172 × g, 10 min, at 4 °C) to eliminate any precipitates, and 550 μl of supernatant was transferred into 5 mm NMR tubes for NMR analysis[@b47]. Metabolic profiling data acquisition ------------------------------------ A Bruker 600 MHz AVANCE III NMR spectrometer (Bruker Biospin, Rheinstetten, Germany) was used to receive the ^1^H-NMR spectra of plasma samples, operating at a ^1^H frequency of 600.13 MHz and a temperature of 298 K. A one-dimensional (1D) Carr-Purcell-Merboom-Gill (CPMG, RD--90− (τcp−180−τcp) -acquisition) with water suppression and a total spin-spin relaxation delay of 320 ms was set for the plasma analysis. The ^1^H NMR spectrum for each sample consisted of 64 scans requiring 2.7 min of acquisition time with the following parameters: spectral width = 12019.2 Hz, spectral size = 65536 points, pulse width(90) = 14.0 μs, and relaxation delay (RD) = 1.0 s. FIDs were Fourier transformed with LB = 0.3 Hz. For a good signal dispersion and visualization, two-dimensional (2D) NMR spectra for the selected samples were also recorded using a 298 k on Bruker 600 MHz AVANCE III NMR spectrometer, including 1H--1H correlation spectroscopy (COSY) and 1H--13C heteronuclear multiple quantum coherence (HMQC). The 2D 1H-1H COSY experiments were acquired in magnitude mode (Bruker pulse sequence cosygpqf) at 600 MHz with 2k data points in F2 and 256 increments in F1, using spectral widths of 6602.1 and 6601.5 Hz in both dimensions. A total of 25 transients were collected with an acquisition time of 0.155 s. The relaxation delay was 1.5 s, the 90 pulse width was 14.0 μs, and the receiver gain 203. And also the 2D 1H-13C HMQC experiments were acquired in magnitude mode (Bruker pulse sequence hmqcgpqf) at 600 MHz with 1 k data points in F2 and 256 increments in F1, using a spectral width of 6602.1 Hz in ^1^H dimension and 36219.4 Hz in the ^13^C dimension. A total of 110 transients were collected with an acquisition time of 0.078 s. The relaxation delay was 1.5 s, the 90 pulse width was 14.0 μs, and the receiver gain 203. NMR data preprocessing ---------------------- The baseline and phase pretreatment of the acquired 1H NMR files were set manually with MestReNova software (Mestrelab Research, Santiago de Compostella, Spain). All of the spectra were referenced to the chemical shift of TSP located at δ 0.00 ppm. After the regions of δ 4.70--5.20 and δ 3.34--3.37 ppm were removed to eliminate the influence of water and methanol, the spectra were segmented at δ 0.01 intervals across the chemical shift range of 0.5 to 9.00 ppm. To reduce significant concentration differences between the samples, the integral values from each spectrum were normalized to a sum of all of the integrals in a spectrum, and date matrices were constructed for further multivariate analysis[@b54][@b55]. Data analysis ------------- Prior to statistical analysis, all resulting integral data from ^1^H-NMR metabolomics analysis were imported into SIMCA-P13.0 (Umetrics, Sweden) for multivariate analysis. Partial least squares discrimination analysis (PLS-DA) was conducted to distinguish different groups in a supervised manner. Parameters for model fitness (R^2^) and predictive ability (Q^2^) with leave-one-out cross validation and the response of the permutation test (200 cycles) should be used to evaluate whether the model is established or not because of the small number of samples[@b56]. Furthermore, a supervised pattern recognition approach known as an orthogonal projection to latent structures discriminant analysis (OPLS-DA) was used to improve the classification of the different groups while screening biomarkers. With an aim to discover the potential variables contributing to the differentiation, we generated an S-plot for the OPLS-DA model used to define metabolites significantly contributing to the separation of groups. On the basis of the variable importance in the project (VIP) threshold of 1 (VIP ≥ 1.00), a number of metabolites responsible for the difference in metabolic profiles of different groups could be obtained. In parallel, the metabolites identified by the OPLS-DA were validated at a univariate level using *t*-test (SPSS 17.0) with the critical *p* value set to 0.05 in order to detect the main metabolites that were significantly different leading to the class discrimination. A system statistical metabolic correlation analysis was further applied to display the relationships between these certain metabolite integrals[@b57]. Metabolite intensities relative to the sum of the total spectral integral were used as variables, and Spearman's correlation coefficient was calculated among those variables using Java. An absolute value of the correlation coefficient indicates a statistically significant relationship among these potential biomarkers. Positive values masked in the pixel map are shown with red colors, and negative values are indicated with blue colors. A receiver operating characteristic (ROC) curves was carried out to further evaluate the performance of the metabolites selected by *t*-test in clinical diagnosis. The area under the curve (AUC) was used to evaluate diagnostic psychological suboptimal health state values in the clinic. Additional Information ====================== **How to cite this article**: Tian, J.-s. *et al.* Discovery, screening and evaluation of a plasma biomarker panel for subjects with psychological suboptimal health state using ^1^H-NMR-based metabolomics profiles. *Sci. Rep.* **6**, 33820; doi: 10.1038/srep33820 (2016). Supplementary Material {#S1} ====================== ###### Supplementary Information This work was supported by the National Natural Science Foundation of China (No. 81441096), the project of Shanxi provincial health and Family Planning Commission (No. 2014ZY01), and the science and technology project of Shanxi province (201603D321077). Furthermore, we thanks to Prof. Xiang Zhang from the Department of Chemistry of the University of Louisville, US for his friendly revision on this manuscript. **Author Contributions** X.-m.Q. and G.-h.D. design of the study and writing the protocol. Y.-f.W. and L.Z. collection of samples and observation of clinical curative effect. X.-t.X. and H.X. collection and analysis of data and drafting of the manuscript. J.-s.T. conception and design, interpretation of data, writing and revising and final approval of the manuscript submitted. X.Z. friendly revision on the type and gramma. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6} ###### General characteristic of the participants at baseline and at the end of the Baihe Dihuang Tang intervention (4 weeks) and the healthy controls.   Psychological suboptimal health Healthy controls ------------- --------------------------------- ------------------ ------------ Sample size 22 22 23 Sex (M/F) 4/18 4/18 5/18 Age (year) 48.7 ± 5.3 48.7 ± 5.3 49.3 ± 4.6 SCL-90 143.9 ± 22.6 112.4 ± 11.5\*\* 90 M: male; F: Female; SCL-90: The Symptom Checklist 90. \*\**P* \< 0.01 before and after 4 weeks. ###### Peak attribution of the main marked metabolites in ^1^H-NMR spectra of plasma samples. Key Metabolites Moieties δ^1^H/ppm and multiplicity/Hz Key Metabolites Moieties δ^1^H/ppm and multiplicity/Hz ----- --------------- ----------------------- ------------------------------- ----- ------------------------ ----------------------- ------------------------------- 1 2-OH-butyrate γCH~3~ 0.90 (t, 7.47) 17 Isoleucine δCH~3~ γ'CH~3~ 0.94 (t, 7.4) 1.02 (d, 7.0) 2 3-OH-butyrate γCH~3~ αCH 1.20 (d, 6.60) 4.15 (m) 18 Lactate βCH~3~ αCH 1.33 (d, 6.9) 4.12 (q, 6.9) 3 Acetate βCH~3~ 1.93 (s) 19 Leucine δCH~3~ αCH~2~ 0.96 (d) 3.73 (m) 4 Acetoacetate CH~3~ CH 2.28 (s) 3.48 (s) 20 Lipids CH~3~ (CH~2~)~n~ 0.84 (t) 1.28 (m) 5 Alanine βCH~3~ CH 1.48 (d, 7.3) 3.77 (m) 21 Methionine γCH~2~ S-CH~3~ 2.62 (t, 7.58) 2.14 (s) 6 Betaine N(CH~3~)~3~ CH~2~ 3.27 (m) 3.89 (s) 22 Methylamine CH~3~ 2.61 (s) 7 Carnitine N(CH~3~)~3~ 3.21 (s) 23 Methanol CH~3~ 3.36 (s) 8 Choline N(CH~3~)~3~ 3.20 (s) 24 N-acetyl-glycoproteins CH~3~ 2.04 (s) 9 Citrate Half CH~2~ Half CH~2~ 2.54 (d, 16.1) 2.65 (d, 16.2) 25 Phosphatidylcholine N(CH~3~)~3~ 3.22 (s) 10 Creatine N-CH~3~ CH~2~ 3.93 (s) 3.04 (s) 26 Phenylalanine 2 and 6-CH 3 and 5-CH 7.33 (m) 7.42 (m) 11 Cysteine CH CH~2~ 3.97 (dd) 3.06 (m) 27 Proline αCH~2~ βCH~2~ 1.99 (m) 2.34 (m) 12 Glutamine αCH βCH~2~ 2.16 (m) 2.45 (m) 28 Pyruvate CH~3~ 2.38 (m) 13 Glutamate βCH~2~ γCH~2~ 2.07 (m) 2.35 (m) 29 Trimethylamine oxide CH~3~ 3.26 (m) 14 Glycine αCH~2~ 3.57 (s) 30 Tyrosine 3 and 5-CH 2 and 6-CH 6.90 (m) 7.19 (m) 15 glycerol CH~2~ CH 3.67 (m) 3.78 (m) 31 Valine γCH~3~ γ'CH~3~ 0.99 (d, 7.1) 1.05 (d, 7.0) 16 Histidine 2-CH 4-CH 7.68 (s) 7.10 (s) 32 Glucose C~1~H 5.22 (d, 3.7) 4.64 (d, 8.0) ^a^s: singlet, d: doublet, t: triplet, q: quartet, m: multiplet, dd: doublet of doublet. ###### Metabolites selected as biomarkers characterized in plasma profile and their change trend after Baihe Dihuang Tang treatment. No. Metabolites Shift chemical[a](#t3-fn1){ref-type="fn"} Patients *vs.* Control[b](#t3-fn2){ref-type="fn"} *P* value Treated *vs.* before Treatment[b](#t3-fn2){ref-type="fn"} *P* value Metabolism Pathway ----- ------------------------ ------------------------------------------- --------------------------------------------------- ------------ ----------------------------------------------------------- ------------ ------------------------ 1 valine 1.00 (d) 1.05 (d) ↓ 2.26E-03\* ↑ 3.02E-05\* Amino acid metabolism 2 Isoleucine 0.94 (t) 1.02 (d) ↓ 3.04E-02\* ↑ 2.09E-01 Amino acid metabolism 3 Glutamine 2.45 (m) 2.16 (m) ↑ 2.64E-04\* ↓ 4.56E-03\* Amino acid metabolism 4 Citrate 2.54 (d) ↑ 1.13E-02\* ↓ 7.59E-02 TCA cycle 5 Glucose 4.66 (d) ↓ 1.99E-03\* ↑ 2.06E-01 Glucose metabolism 6 Trimethylamine oxide 3.26 (s) ↑ 8.76E-03\* ↓ 4.99E-10\* Methylamine metabolism 7 N-acetyl-glycoproteins 2.04 (s) ↑ 1.13E-02\* ↓ 6.88E-01 inflammatory responses 8 Tyrosine 6.90 (m) 7.19 (m) ↑ 9.15E-04\* ↓ 1.63E-01 Amino acid metabolism 9 Phenylalanine 7.36 (m) 7.42 (m) ↑ 1.22E-03\* ↓ 8.60E-03\* Amino acid metabolism ^a^Multiplicity definitions: s, singlet; d, doublet; t, triplet; m, multiplet. ^b^Metabolites with"↑/↓" means increased/decreased, "\*" means dates significant differences. ###### Area under the curves of the biomarker combinations. Biomarkers Area Std. Error Asymptotic Sig. Asymptotic 95% Confidence Interval ----------------------- ------- ------------ ----------------- ------------------------------------ ------- A 0.814 0.068 0.000 0.680 0.948 A + B 0.882 0.053 0.000 0.778 0.987 A + B + C 0.880 0.051 0.000 0.781 0.979 A + B + C + D 0.890 0.048 0.000 0.797 0.984 A + B + C + D + E 0.924 0.040 0.000 0.845 1.002 A + B + C + D + E + F 0.971 0.020 0.000 0.931 1.011 A, Citrate; B, Glutamine; C, Tyrosine; D, Phenylalanine; E, TMAO; F, N-acetyl-glycoproteins.

{#sp1 .830} {#sp2 .831} {#sp3 .832} {#sp4 .833} {#sp5 .834} {#sp6 .835} {#sp7 .836} {#sp8 .837} {#sp9 .838} {#sp10 .839} {#sp11 .840} {#sp12 .841} {#sp13 .842} {#sp14 .843} {#sp15 .844} {#sp16 .845} {#sp17 .846}

INTRODUCTION ============ The treatment choice for patients with gastric cancer is radical gastrectomy with esophagojejunal anastomosis and lymphadenectomy when involves to upper body or cardia of the stomach. Anastomotic leaks are life-threatening complications that can increase postoperative mortality. In total gastrectomy, anastomotic leaks are more common than subtotal gastrectomy, with reported incidence rates ranging from 4% to 17%. The mortality rate is still high (12% to 50%) despite traditional standard management such as conservative treatment with nil per os (NPO), abscess drainage, and re-exploration with surgical repair \[[@B1][@B2][@B3]\]. Subsequent surgical repair is difficult in patients with anastomotic leaks because of the associated high operative mortality. Endoscopic self-expandable metal stent (SEMS) replacement has recently begun to be accepted as a primary treatment of patients with esophagojejunal anastomotic leaks after total gastrectomy. Endoscopic SEMS replacement has 2 significant problems: embedded stents due to tissue hyperplasia and stent migration in the patient with anastomotic leaks. Endoscopists should carefully choose the type of SEMS to avoid these issues. Some endoscopists may prefer to use a partially covered SEMS to prevent migration and reduce leakage between the stent and esophageal wall. However, embedded stents are emerging as a common obstacle that can be difficult to extract when using partially covered SEMS and even a conventional fully covered SEMS. Embedded stents may occur with other issues including secondary stricture, perforation, and bleeding, despite the successful removal of stents using various methods \[[@B4][@B5]\]. Choosing the right type of stent to avoiding stent migration and embedding is crucial to improve successful endoscopic treatment in patients with anastomotic leaks. Conventional fully and partially covered SEMS may not be suitable considering migration and embedding for endoscopic treatment. In this study, we used a benign fully covered SEMS with an anchoring thread and thick silicone covering membrane to prevent stents from becoming embedded and migrating in patients with gastric cancer and esophagojejunal anastomotic leaks after total gastrectomy. The clinical outcomes, including effectiveness and complications, were evaluated. MATERIALS AND METHODS ===================== Patient selection ----------------- We performed a retrospective review of the data of patients who underwent benign fully covered SEMS placement with the Shim\'s technique for anastomotic leaks after total gastrectomy to treat gastric cancer between January 2009 and December 2016 at the Chonnam National University Hwasun Hospital. Overall, 1,018 patients with gastric cancer underwent total gastrectomy; anastomotic leaks occurred in 24 (2.4%) cases. Open surgery was performed in 679 cases, including 10 (1.5%) anastomotic leaks, and laparoscopic surgery was performed in 339 cases, including 14 (4.1%) anastomotic leaks. Anastomotic leak treatment included endoscopic stent replacement (n=14, 58.3%), endoscopic hemoclips (n=2, 8.3%), surgical repair (n=3, 12.5%), and conservative treatment (n=5, 20.8%). Of these patients, 14 were enrolled in this study ([Fig. 1](#F1){ref-type="fig"}). We obtained consent from all patients and the study was approved by the Institutional Review Board of the Chonnam National University Medical School, Gwangju, Korea (CNUHH-2015-113). {#F1} Embedded stent and migration ---------------------------- Conventional fully covered SEMS with anchoring thread has been commonly used endoscopic stent for treating esophagojejunal anastomotic leaks after total gastrectomy in Korea. However, this stent can cause embedding into esophageal mucosa by granulation tissue before extraction of stent. We had experience case of anastomotic leaks with a severely embedded stent after use of a conventional fully covered SEMS (Hanarostent^®^; M.I. Tech, Seoul, Korea). The embedded stent was removed with great difficultly. The rat tooth grasped at the distal end of stent, and the stent was peeled backward through the inner side of the stent under fluoroscopic control. This patient\'s leaks resolved after stent retrieval, but severe stricture occurred mid-esophagus. The patient ultimately underwent a second surgery for severe stricture after 1 year because of a persistent stricture despite repeated endoscopic balloon dilation ([Fig. 2](#F2){ref-type="fig"}). {#F2} Fully covered SEMS with an anchoring thread and thick covering membrane ----------------------------------------------------------------------- We used a benign fully covered SEMS with a thick membrane and a silk thread (benign Hanarostent^®^, fully covered esophageal SEMS with Shim\' technique; M.I. Tech) to treat anastomotic leaks in the next patient. This stent has a specially designed silicone membrane that was modified in thickness (0.2--0.24 mm) and a long end of the silicone membrane (5 mm) to prevent overgrowth and ingrowth of granulation tissue. This stent was also applied with an anchoring thread, using the so called Shim\'s technique to prevent migration \[[@B6]\]. The Shim\'s technique involves attaching the silk thread at the proximal end of the stent, which can then be moved through the nose and attached to the ear lobe with tape, similar to the method for endoscopic nasobiliary drainage. Endoscopic stent replacement ---------------------------- Anastomotic leaks were diagnosed via esophagram using gastrografin or endoscopy. Abdominal and chest computed tomographies (CTs) were performed to evaluate fluid collection or abscess formation in the abdomen or chest cavity. Endoscopic benign stent placements were performed for the following indications: 1) endoscopic confirmed leaks at esophagojejunal anastomosis site, 2) hemodynamically stable patients with systolic blood pressure greater than 90 mmHg, and 3) no residual cancer after total gastrectomy. The methods used for endoscopic stent replacement are described in [Fig. 3](#F3){ref-type="fig"}. Patients were consciously sedated with intravenous midazolam and fentanyl during the procedure. The stent length was 4 cm longer than the size of the leaks to ensure sufficient coverage of the esophageal and jejunal margins of leakage sites. The stent was carefully deployed to locate the middle portion of the stent at the leakage site through a guide wire under endoscopic and fluoroscopic control. After stent deployment, the silk thread was removed through the nose and fixed to the patient\'s ear lobe using tape. Simple chest radiography was checked to confirm the stent position the day after the procedure and then at 1-week intervals. Per-oral intake began with a small amount of liquid, which then progressed to a regular diet. If an abscess or fluid collection was found on abdominal or chest CT, abscess drainage was performed with intravenous broad-spectrum antibiotics. Follow-up endoscopy was performed to check the healing status of the leak site and stent position within 3 weeks based on the results of simple chest radiography and clinical findings. The endoscopist can directly observe the healing status of anastomotic leaks through the space between stent and esophageal lumen without embedding stent. In malnourished patients, an additional feeding tube was inserted within stent under endoscopic guidance to improve nutritional status and decrease the incidence of leakage in the space between the esophageal wall and the stent. All stents were extracted with/without a fluoroscopic guide after checking complete closure or other complications under endoscopic visualization. Follow-up endoscopy was scheduled 3--6 months after leak closures, and then at yearly intervals. {#F3} Statistical analysis -------------------- The Student\'s t-test and χ^2^ test were conducted for continuous and categorical variables. Continuous variables were expressed as mean (standard deviation) and categorical variables, as a percentage (%). Data were analyzed using SPSS version 21.0 (SPSS Inc., Chicago, IL, USA). RESULTS ======= Baseline characteristics of patients ------------------------------------ A total of 14 patients underwent endoscopic benign fully covered SEMS placement for esophagojejunal anastomotic leaks after total gastrectomy to treat gastric cancer ([Table 1](#T1){ref-type="table"}). Patients consistent of 12 (85.7%) men and 2 (14.3%) women with a mean age of 69.3±8.8 years. The mean body mass index was 23.6±3.1. Co-morbidities were hypertension (5, 35.7%), diabetes mellitus (4, 28.6%), coronary artery disease (2, 14.3%), and chronic obstructive lung disease (3, 21.4%). The approaches used for total gastrectomy were open (6, 42.9%) and laparoscopic surgery (8, 57.1%). The mean largest tumor size was 35.7±29.4 mm. Tumor stages were Ia (6, 42.9%), Ib (3, 21.4%), II (2, 14.3%), and III (3, 21.4%) ###### Baseline characteristics of patients with stents for anastomotic leaks after total gastrectomy  Variables Total (n=14) -------------------------------- ---------------------------------- ---------- Age (yr) 69.3 (8.8) Male sex 12 (85.7) Body mass index (kg/m^2^) 23.7 (3.2) Comorbidity Diabetes mellitus 4 (28.6) Hypertension 5 (35.7) Coronary artery disease 2 (14.3) Chronic obstructive lung disease 3 (21.4) Approach of total gastrectomy Open 6 (42.9) Laparoscopic 8 (57.1) Method of lymph node resection D1 6 (42.9) D2 8 (57.1) Largest tumor size (mm) 35.7 (29.4) Pathologic grade G1 5 (35.7) G2 2 (14.3) G3 7 (50.0) Tumor stage Ia 6 (42.9) Ib 3 (21.4) II 2 (14.3) III 3 (21.4) Data are shown as number (%) or mean (standard deviation). Characteristics of anastomotic leaks and endoscopic stent placement ------------------------------------------------------------------- Clinical and endoscopic characteristics are summarized in [Table 2](#T2){ref-type="table"}. Clinical inflammatory findings of patients were fever (6, 42.9%), leukocytosis (5, 35.7%), and elevated C-reactive protein (8, 57.1%) before endoscopic treatment. The feeding status of patients were NPO (2, 14.3%), sips of water (5, 35.7%), soft diet (4, 28.6%), and regular diet (3, 21.4%) before endoscopic treatment. The mean size of leaks was 13.1 mm (range, 3--30 mm). The duration from operation to stent insertion was 10.7 days (range, 3--35 days). Technical success of stent insertion was achieved in all patients. All patients received benign fully covered SEMS with a thick membrane and a silk thread. The lengths of stent were 8 cm (2, 14.3%), 10 cm (8, 57.1%), and 12 cm (4, 28.6%). ###### Endoscopic and clinical characteristics of anastomotic leakage  Variables Total (n=14) ------------------------------------------------------- -------------------------------------- ----------------------------------- ---------- Clinical findings before stent insertion Fever (≥38°C) 6 (42.9) Leukocytosis (≥10,000/mm^2^) 5 (35.7) High C-reactive protein (≥10 mg/dL) 8 (57.1) Feeding status NPO 2 (14.3) Water only 5 (35.7) Soft diet 4 (28.6) Regular diet 3 (21.4) Endoscopic findings of leaks Size of leak (mm) 13.1 (3--30) Grade of leak 13 (81.3) Small (\<1/3 of anastomosis site) 7 (50.0) Large (≥1/3 of anastomosis site) 7 (50.0) Endoscopic stent insertion Technical success of stent insertion 14 (100.0) Length of stent (cm) 8 2 (14.3) 10 8 (57.1) 12 4 (28.6) Diameter of stent (22 mm) 14 (100.0) Time interval from operation to stent insertion (day) 10.7 (3--35) Data are shown as number (%) or value (range). NPO, nil per os. Therapeutic outcomes and complications -------------------------------------- The clinical outcomes and associated variables are summarized in [Table 3](#T3){ref-type="table"}. Twelve (85.7%) patients had complete resolution from anastomotic leaks after stent extraction, while 2 (14.3%) patients had treatment failure. Two patients died 2 weeks postoperatively, one patient had lung empyema with pneumonia before stent insertion that progressed to acute respiratory distress syndrome, and the other had intraperitoneal abscess that progressed to septic shock. The time interval from stent insertion to extraction was 32.3 days (range, 18--49 days). The overall complication rate was 14.3% (n=2). Embedded stent and migration, which were major complications associated with SEMS, were absent. One patient had a jejunal active ulcer by the distal tip of stent but experienced complete healing after antiulcer treatment. Delayed stricture at large leak area of anastomosis site was found in 1 patient and successfully resolved after repeated endoscopic balloon dilatation. ###### Therapeutic outcomes and complications  Variables Total (n=14) --------------------------------------------------------- ------------------------ ---------- Complete closure of leak 12 (85.7) Time intervals from stent insertion to extraction (day) 32.3 (18--49) Complications 2 (14.3) Embedded stent 0 (0) Migration 0 (0) Bleeding 0 (0) Jejunal ulcer 1 (7.1) Stricture at leak site 1 (7.1) Secondary stent insertion 0 (0) Secondary operation 0 (0) Endoscopic follow-up within 3 weeks 10 (71.4) Complete closure 2 (20.0) Improvement of leak 8 (80.0) Endoscopic repositioning 5 (35.7) Additional feeding tube 4 (28.6) Combined fluid drainage Abdominal abscess 6 (42.9) Lung empyema 5 (35.7) Duration of admission (day) 58.0 (16--164) Treatment at intensive care unit 7 (50.0) Mortality 2 (14.3) Septic shock 1 (7.1) Pneumonia 1 (7.1) Data are shown as number (%) or value (range). Clinical courses after endoscopic stent placement ------------------------------------------------- Per-oral intake was started with a small amount of liquid 24 hours after stent placement, which then progressed to a regular diet. Intravenous broad-spectrum antibiotics were injected in all patients. Combined fluid or abscess drainages were performed for intraperitoneal abscess (6, 42.9%) and lung empyema (5, 35.7%). Improvement of inflammation was seen 2 weeks after stent replacement. Laboratory changes according to endoscopic stent placement are summarized in [Table 4](#T4){ref-type="table"}. Ten (71.4%) patients received follow-up endoscopy to check the healing status of the leak site and stent position 3 weeks postoperatively. All leaks showed signs of improvement (8, 80.0%) or complete closure (2, 20.0%) under follow-up endoscopy. Endoscopic repositioning was performed in 5 (35.7%) patients due to slight proximal stent displacement. All stents were removed under endoscopy after confirming complete healing of leaks. No additional endoscopic stent placements were needed. Four (28.6%) patients underwent additional feeding tube placement into the deep portion of the jejunum below the stent under endoscopic guidance because of poor appetite and food material remaining in their drainage ([Fig. 4](#F4){ref-type="fig"}). ###### Laboratory changes according to endoscopic stent placement  Variables Before stent 3 days later 1 week 2 weeks 3 weeks --------------------------- -------------- -------------- -------------- ------------- ------------- Fever (°C) 37.6±0.8 37.0±0.8 36.6±0.6 36.4±0.2 36.3±0.3 White blood cell (/mm^2^) 11,615±7,432 11,061±3,736 13,065±5,719 9,975±2,748 7,220±1,225 C-reactive protein 14.9±8.7 9.1±5.1 7.5±5.6 5.3±5.2 3.8±2.9 Data are shown as mean±standard deviation. {#F4} DISCUSSION ========== Anastomotic leak after total gastrectomy are more frequent and can be a fatal major complication with a 3-fold higher mortality rate than patient without leaks \[[@B7]\]. Surgical repair has been difficult in patients with anastomotic leaks because of their poor general condition and high mortality. Endoscopic treatment using a SEMS has become widely used as a primary treatment for esophagojejunal anastomotic leaks after total gastrectomy because of its lower invasiveness and favorable outcomes. However, the reported therapeutic efficacy and incidence rates of adverse events of SEMS placement for anastomotic leaks have been inconsistent, with complete closure rates ranging from 23% to 100% \[[@B8][@B9][@B10][@B11]\]. These variable clinical results for SEMS placement are mainly caused by the type of stent, which influences 2 significant SEMS-related issues: embedded stent due to tissue hyperplasia and stent migration. The endoscopist should consider these issues before choosing the type of stent to be used for treating leaks. The fully covered SEMS with a thick membrane and a silk thread is a novel stent for treating patients with benign condition such as perforation, stricture, and fistula because it can completely prevent tissue hyperplasia and stent migration. To our knowledge, this study is the first clinical study evaluating the feasibility of this stent for treating anastomotic leaks after total gastrectomy in patients with gastric cancer. Stent migration frequently occurs in patients with anastomotic leaks because no portion of the stent is anchored. The migrated stent may cause secondary severe complications such as obstruction or perforation at the distal bowel, along with sealing failure at the leak site. Higher migration rates have been reported for a fully covered SEMS (26%--87%) compared with a partially covered SEMS (0%--20%) \[[@B9][@B10][@B12][@B13]\]. The Shim\'s technique simply uses a silk thread to prevent the migration of the stent. Using a fully covered SEMS with the Shim\'s technique was a useful method to completely prevent distal stent migration in all patients. However, the Shim\'s technique requires the additional process of unthreading through the nose and induced discomfort in some patients. Embedded stent can occur due to tissue hyperplasia from mucosal irritation at the end portions of the stent. A partially covered SEMS is always associated with embedded stents. The occurrence with a fully covered SEMS is variable, with reported incidence rates ranging from 8% to 21% \[[@B10][@B14]\]. Embedded stents are major adverse events that can increase the difficulty of stent removal when treating patients with benign disease. It can also cause secondary complications including hemorrhage, obstruction, stent fracture, perforation, fistulas, and abscess formation \[[@B5][@B15]\]. Tissue hyperplasia or granulation tissue formation is related to the type of stent and the duration of stent placement. Tissue hyperplasia can start to develop during the second week from the insertion of a conventional fully covered SEMS. The stent then becomes completely embedded into the esophageal mucosa by the 4th--6th week. Massive granulation tissue can result in partial esophageal stricture by the 8th week or later \[[@B16]\]. Various methods have been introduced to successfully remove stent in patients with embedded stent, such as stent-in-stent (SIS) technique, argon plasma coagulation, and the overtube technique \[[@B4]\]. Some clinicians prefer to use partially covered SEMS rather than a fully covered SEMS, as the risk for migration is lower, and hyperplasia at the uncovered portion may decrease leakage between the stent and the esophageal wall. Such a stent should still be able to be removed using various removal methods, such as SIS technique despite the occurrence of embedded stent. However, we believe that partially covered SEMS should no longer be used in patients with anastomotic leaks because the novel stent described here can completely prevent stent migration and embedding. The fully covered SEMS with a thick membrane and a silk thread was first developed in 2007 for treating patients with benign esophageal disease such as stricture. To minimize tissue responses to the stent, a thicker (0.2--0.24 mm) and longer (5 mm) fully covered silicone membrane was used to prevent overgrowth and ingrowth of the granulation tissue. Additionally, the Shim\'s technique using a silk thread was applied to prevent migration. This clinical study sought to determine the type of stent need to treat anastomotic leaks after total gastrectomy and showed improved complete healing rates without major complications, compared with previous studies using conventional stents \[[@B4][@B9][@B11][@B13]\]. The novelty of this stent lies in its perfect prevention of the occurrence of distal migration and embedded stent as 2 major problems of conventional stent. In addition, it is possible to safely remove the stent after checking sealing leaks under direct endoscopic view through the space between the esophageal wall and the stent without embedding. The only disadvantage of this stent is the patient\'s discomfort due to the silk thread coming from the nose (as with a Levin tube). This study was limited in its small sample size and a retrospective study design. Therefore, further investigation is warranted. In conclusion, the endoscopic placement of a benign fully covered SEMS with a thick membrane and a silk thread is an effective and safe novel treatment for patients with anastomotic leaks total gastrectomy to prevent stent migration and embedding. The endoscopist should carefully choose the type of stent, considering the risk of stent migration and embedding before treating anastomotic leaks. **Author Contributions:** **Conceptualization:** C.S.B.**Data curation:** L.S.H., M.D.S.**Formal analysis:** L.S.H.**Funding acquisition:Investigation:** L.W.S.**Methodology:** L.W.S.**Project administration:** C.S.B.**Resources:** J.M.R., R.S.Y., P.Y.K.**Software:** J.G.M.**Supervision:** P.Y.K.**Validation:** J.Y.E.**Visualization:** J.G.M.**Writing - original draft:** J.G.M.**Writing - review & editing:** C.S.B. **Conflict of Interest:** No potential conflict of interest relevant to this article was reported. [^1]: ^\*^Gum-Mo Jung and Seung-Hyun Lee were equally contributed as 1st author.

1. Introduction {#sec1-jof-05-00074} =============== Coccidioidomycosis, commonly known as Valley Fever, is caused by the soil-dwelling fungi *Coccidioides immitis* and *C. posadasii*. As is common for fungi that infect humans and animals, species of *Coccidioides* are dimorphic in terms of their life cycle, growing saprobically as multicellular filaments on non-living organic matter and, upon entry into a host lung, growing pathogenically in a yeast-like spherule stage \[[@B1-jof-05-00074]\]. The endospores formed by spherules can disseminate within a host but are not transmissible to new hosts. Outside a living host, species of *Coccidioides* form arthroconidia (asexual spores) that can become airborne via soil disturbance and can be inhaled by potential hosts \[[@B2-jof-05-00074]\]. Although coccidioidomycosis presents in about 40% of patients as a pulmonary infection, a chronic and disseminated form of coccidioidomycosis, often resulting in lifelong treatment, occurs in roughly 5% of patients \[[@B3-jof-05-00074],[@B4-jof-05-00074]\]. Although arthroconidia can be detected in soil and in dust, the specific niches of *Coccidioides* species are unknown \[[@B5-jof-05-00074]\]. The disease has been characterized as endemic to arid environments that include the Southwestern United States, Central and South America and Mexico \[[@B6-jof-05-00074]\]. In the United States, highly endemic hotspots have been documented in California and Arizona \[[@B7-jof-05-00074],[@B8-jof-05-00074],[@B9-jof-05-00074]\]. Although less prevalent, Valley Fever is present in Nevada, Colorado, Texas, New Mexico, Utah and Washington state \[[@B10-jof-05-00074],[@B11-jof-05-00074],[@B12-jof-05-00074],[@B13-jof-05-00074]\]. Molecular and phenotypic analyses have resulted in *Coccidioides* being divided into two species: *C. immitis*, known primarily from California, and *C. posadasii*, recognized originally as the non--California group \[[@B6-jof-05-00074]\]. While clinical diagnosis and treatment of coccidioidomycosis have not depended on identifying isolates to species, medical and scientific communities are becoming more cognizant of the potential need to recognize previously undetected phenotypic, morphological, ecological, and genetic differences between species. Multi-locus genetic analyses and whole-genome studies have attempted to map the distribution of these species. To date, however, the isolates studied have been mainly limited to California, Arizona, Central America, Mexico, and Texas \[[@B14-jof-05-00074],[@B15-jof-05-00074]\], and the genetics of isolates from New Mexico have not been examined. The study reported here employed genetic analyses of 18 isolates collected from patients diagnosed with coccidioidomycosis from diverse locations across New Mexico and the Four Corners region. Although Southern New Mexico has been recognized as part of the endemic region for *Coccidioides* \[[@B12-jof-05-00074],[@B16-jof-05-00074]\], several of the isolates examined here were from Northern and Central New Mexico, suggesting a broad range for *Coccidioides* that includes the Four Corners region. Also noteworthy is the fact that isolates of both *C. immitis* and *C. posadasii* were obtained from patients in New Mexico with *C. immitis* being obtained from a patient who resides in San Juan County, NM. The risk for coccidioidomycosis has been reported as much higher among some ethnic groups, particularly African Americans and Filipinos \[[@B17-jof-05-00074]\]. In these ethnic groups, the risk for disseminated coccidioidomycosis is ten-fold that of the general population \[[@B18-jof-05-00074]\]. Health records from patients in our study suggest the possibility that Native American Indians represent an additional risk group for disseminated disease. 2. Materials and Methods {#sec2-jof-05-00074} ======================== 2.1. Sample Collection and Patient Information {#sec2dot1-jof-05-00074} ---------------------------------------------- Eighteen human clinical specimens from seventeen patients diagnosed with coccidioidomycosis obtained between 2013 and 2017 were submitted to the New Mexico Department of Health (NMDOH) Scientific Laboratory Division (SLD). Coccidioidomycosis is a reportable condition in New Mexico. The NMDOH collects information on all confirmed and probable cases (case status definitions follow CSTE standards) who reside in New Mexico at the time of diagnosis to look for potential risk factors via chart review and data extraction. When possible, patient information was noted with specific interest in type of infection, residency, travel history, occupation, race/ethnicity, gender, age, and medical history ([Table 1](#jof-05-00074-t001){ref-type="table"}). 2.2. Molecular Methods {#sec2dot2-jof-05-00074} ---------------------- DNA was extracted at the NMDOH SLD from *Coccidioides* isolates using the PrepMan^®^ Ultra Reagent method (Applied Biosystems, Foster City, CA) and stored at −80 °C until further processing. The DNA extractions were transferred to the Natvig Laboratory at the University of New Mexico where a 1/10 DNA dilution of each preparation was used for PCR amplification of three AFToL (Assembling the Fungal Tree of Life)-designated nuclear gene regions in addition to a serine proteinase gene region diagnostic for species ([Table 2](#jof-05-00074-t002){ref-type="table"}). For the serine proteinase, MCM7, and RPB1 genes, primers were designed to amplify and sequence regions that we determined from comparisons of sequences in GenBank to be diagnostic in distinguishing between the two *Coccidioides* species. Serine proteinase, specifically, was chosen as a target based on results presented by Koufopanou et al. \[[@B19-jof-05-00074]\]. ITS amplification and sequencing involved the entire ITS1-5.8S rRNA-ITS2 region. We designed a primer pair, with one primer anchored in a sequence unique to *C. posadasii*, to amplify a mitochondrial intron sequence from *C. posadasii* but not *C. immitis*. A second primer set was designed to amplify a portion of the first cox1 exon along with an upstream intergenic region in both *Coccidioides* species but with the potential to differentiate *Coccidioides* from other Onygenales ([Table 2](#jof-05-00074-t002){ref-type="table"}). All PCR reactions began with an initial step at 95 °C for 5 min. This was followed by 35 cycles of 94 °C for 30 s, a gene-specific annealing temperature ([Table 2](#jof-05-00074-t002){ref-type="table"}) for 30 s, then 72 °C for 45 s with a final extension at 72 °C for 7 min. PCR products were cleaned with ExoSAP-IT (Thermo Fisher Scientific, Waltham, MA, USA) before DNA sequencing using BigDye v3.1 (Applied Biosystems, Foster City, CA, USA) chain termination with the Big Dye STeP protocol \[[@B20-jof-05-00074]\]. Forward and reverse sequences were assembled and edited with Sequencher 5.1 (Gene Codes, Ann Arbor, MI, USA). Sequences were deposited in GenBank under accessions MH748760--MH748777 for serine proteinase; MH748742--MH748759 for MCM7; MH748724--MH748741 for RPB1; and MH725244--MH725261 for ITS ([Table 3](#jof-05-00074-t003){ref-type="table"}). 2.3. Phylogenetic Analysis {#sec2dot3-jof-05-00074} -------------------------- Sequences for the four gene regions were aligned individually using Clustal Omega version 1.2.45 \[[@B21-jof-05-00074]\]. For outgroup sequences, each alignment included the appropriate homologous gene region from *Aspergillus steynii* from the GenBank accessions ([Table 3](#jof-05-00074-t003){ref-type="table"}). This species was chosen as an outgroup because it had a clear homolog for each of the four gene regions examined. The serine proteinase in particular, appears to undergo rapid evolution and perhaps gene loss, and as a result, it was difficult to identify clear orthologs in many other species of Eurotiomycetes. Tree-building analyses employed maximum likelihood analysis with PHYLIP (version 3.695) DNAMLK and parsimony analysis with PHYLIP DNAPARS \[[@B22-jof-05-00074]\]. In each case, tree building employed 1000 bootstrap datasets. Analyses were done on each of the four gene alignments separately, as well as on a concatenated alignment of all four genes. The concatenated alignment has been deposited at TreeBase (Submission ID 24737). 2.4. Ethics Statement {#sec2dot4-jof-05-00074} --------------------- All patient data analyzed for this study were anonymized. 3. Results {#sec3-jof-05-00074} ========== All four gene regions examined were capable of distinguishing between the two *Coccidioides* species, based on comparisons with sequences reported in Genbank. Three of the isolates were revealed to be *C. immitis*, while the remaining 15 were *C. posadasii*. Isolates NM3006, NM9443, and NM9737 were clustered together as a *C. immitis* clade in single-locus trees and in the concatenated four-gene phylogeny ([Figure 1](#jof-05-00074-f001){ref-type="fig"} and [Figure S1](#app1-jof-05-00074){ref-type="app"}). Both NM9443 and NM9737 came from a single patient in Utah whose type of infection was unknown ([Table 1](#jof-05-00074-t001){ref-type="table"}). Isolate NM3006 was from a 60-year-old male from San Juan County, New Mexico, with disco-vertebral osteomyelitis due to *C. immitis*. Human coccidioidomycosis due to *C. posadasii* was represented by twelve cases of pulmonary infections (one also had a facial lesion), one of osteomyelitis, and one knee infection. The type of infection associated with the remaining patient was unknown ([Table 1](#jof-05-00074-t001){ref-type="table"}). Five of eight infections for which patient ethnicity was known occurred in Native Americans, while two of the eight occurred in African Americans ([Table 1](#jof-05-00074-t001){ref-type="table"}). In addition to the four nuclear genes examined, we explored the possibility that mitochondrial DNA (mtDNA) regions might be useful in distinguishing between *C. immitis* and *C. posadasii*. In comparisons of the sequences available in GenBank, it appeared that the first intron of the cytochrome c oxidase I (cox1) gene possesses a large insertion/deletion that separates the two species. This allowed the design of PCR primers that resulted in amplification of a sequence in all 15 *C. posadasii* isolates but not isolates of *C. immitis* ([Figure 2](#jof-05-00074-f002){ref-type="fig"}). 4. Discussion {#sec4-jof-05-00074} ============= The isolates employed in this study came from individuals across New Mexico, with several isolates coming from the northwestern part of the State (San Juan County, NM, USA; [Table 1](#jof-05-00074-t001){ref-type="table"}). This is a surprising result given that species of *Coccidioides* are expected to occur primarily in the southern portions of New Mexico, where the environment is similar to the Sonoran Desert regions of Arizona and California. Although 14 of the 15 patients who were infected with *C. posadasii* were residents of New Mexico, two reported recent travel to Arizona ([Table 1](#jof-05-00074-t001){ref-type="table"}). New Mexico sees far fewer cases of coccidioidomycosis than its neighbor Arizona (approximately 47 cases/yr compared to 9680 cases/yr from 2008--2014) \[[@B23-jof-05-00074]\]. There are three non-mutually-exclusive possible reasons for this: (1) physicians in New Mexico do not have a thorough understanding of the disease for diagnosis and treatment, (2) *Coccidioides* is most common in less populated areas of the state, and/or (3) the low human population density across most portions of the state means that there are fewer targets for infection and less human-caused soil disturbance on a wide scale (a known factor in generating the airborne spores that cause infections). From a Knowledge, Attitudes, and Practices Survey of New Mexican physicians in 2010, 72% were not confident in their ability to diagnose and 70% were not confident in their ability to treat coccidioidomycosis. There is a clear need for an increased understanding of disease ecology, including the environmental conditions related both to where the fungus grows and factors that aid in dispersion to better inform our healthcare professionals of the distribution in New Mexico. An increased awareness will help avoid complications from delayed diagnosis and inappropriate treatment that can result in severe disease progression and even death. This point is driven home by a news report of several recent severe cases of human coccidioidomycosis in Southern New Mexico, all of which resulted in delayed diagnosis that prevented timely treatment \[[@B24-jof-05-00074]\]. Native Americans and African Americans represent only 11% and 2.5% of the population of New Mexico, respectively \[[@B25-jof-05-00074]\], so it was surprising to see five of eight infections for which patient ethnicity was known occurred in Native Americans, and two of the eight occurred in African Americans ([Table 1](#jof-05-00074-t001){ref-type="table"}). The increased risk for coccidioidomycosis among African Americans is well known \[[@B26-jof-05-00074]\]. The risk among Native Americans is not documented in the literature. Although one study \[[@B27-jof-05-00074]\] found Southwest Native Americans in regions of high *Coccidoides* endemism (Lower Sonoran desert) to have high rates of positive response to coccidioidin skin tests, that study did not find that rates were higher than for non-Native Americans in the same region. A separate study found a possible correlation between diabetes and coccidioidomycosis-associated death among Native Americans \[[@B28-jof-05-00074]\]. We acknowledge that any or all of multiple factors could contribute to increased risk of coccidioidomycosis. These include living in remote rural areas with high airborne dust loads, working in agriculture or construction, as well as comorbidities such as diabetes and other health conditions. Nonetheless, our results suggest a need for health professionals in New Mexico and the Four Corners region to be aware of potential at-risk groups, as well as a need for a better understanding of locations of *Coccidioides* endemism. Coccidioidomycosis incidence is on the rise in highly endemic areas such as Arizona and California and in more sparsely populated reporting regions including New Mexico, Nevada, and Utah \[[@B11-jof-05-00074]\]. One reason for this increase may be that clinicians are becoming more aware of the disease. Other hypotheses include changes in testing practices, increased travel or relocation to endemic areas, and/or growth of the "at-risk" immunosuppressed population (although coccidioidomycosis can infect healthy individuals). Climatic factors, such as temperature and moisture, in addition to increases in human activities such as construction that produce dust from soil disturbance, can result in increased spore dispersal \[[@B29-jof-05-00074]\]. Our genetic analysis of isolates collected from patients diagnosed with coccidioidomycosis in New Mexico provides a foundation for future exploration of distribution, incidence, and susceptibility of patients in New Mexico and the American Southwest Four Corners region. Because *C. immitis* has been considered to be the "California species" and the vast majority of *C. immitis* infections occur in California, the presence of *C. immitis* among our isolates was unexpected. The reported range for *C. immitis* does, however, include locations in Washington state and Northeastern Utah \[[@B13-jof-05-00074],[@B30-jof-05-00074]\]. Acknowledging that the *C. immitis* isolates we examined could reflect infections acquired as a result of travel outside the Four-Corners region, it is entirely possible that the range of *C. immitis* includes Southern Utah and/or Northern New Mexico. Related to the question of species distributions, we note that although hybridization between *C. immitis* and *C. posadasii* has been reported \[[@B31-jof-05-00074],[@B32-jof-05-00074]\], the fact that all four nuclear genes and the mitochondrial region examined for our isolates agreed in terms of species separations would suggest that any introgression of genes across species would be minimal for the isolates we examined. Our observation that *C. immitis* was present among isolates obtained from patients in New Mexico argues that it is important for health professionals and researchers to have rapid methods to distinguish between the two species. This is true in part because, while differences in distribution of the two species have become increasingly clear in the past decade, the ecological niche difference between the species is unknown \[[@B33-jof-05-00074],[@B34-jof-05-00074]\]. Although physicians do not currently rely on speciation for diagnosis and treatment, this may well change in the future with the increasing discovery of genetic and phenotypic differences between the species. For example, *C. immitis* has a tendency to grow faster than *C. posadasii* on high-salt media \[[@B6-jof-05-00074]\], which suggests there may be other growth differences in physiology that affect virulence and ecology. Rapid methods to distinguish between the two *Coccidioides* species such as the PCR-based method we employed here with mtDNA, along with similar methods that can be used to detect species of *Coccidioides* in environmental DNA samples, should prove valuable in future studies. Given the reported hybridization between *C. immitis* and *C. posadasii* cited above \[[@B31-jof-05-00074],[@B32-jof-05-00074]\], we acknowledge that species assignments made based on a single gene region should be viewed as tentative. The authors would like to thank the New Mexico Department of Health (NMDOH) State Laboratory Division, General Microbiology Section, for providing sample DNA, and Pamela Morden (NMDOH) and Terry Torres-Cruz (Penn State University) for helpful comments and insights. Supplementary materials can be found at <https://www.mdpi.com/2309-608X/5/3/74/s1>. Figure S1. Phylogenetic trees for individual gene regions: Serine protease, MCM7, RPB1, and ITS. ###### Click here for additional data file. P.S.H., M.I.H. (equal contributors) and D.O.N. conceived and completed the molecular analyses. P.S.H. and D.O.N. acquired funding. P.L. obtained isolates and DNA. S.M. analyzed patient histories. All authors participated in writing the manuscript. This work was funded in part by a University of New Mexico (UNM) Graduate and Professional Student Association High Priority Grant to P.S.H. Additional funding was provided by the UNM Biology Department's Graduate Research Allocations Committee Research Award and by the University of New Mexico's Sevilleta LTER Summer Graduate Student Fellowship program (P.S.H.) (National Science Foundation awards DEB 1655499 and DEB 1440478). We thank the UNM Department of Biology's Molecular Biology Facility for help with sequencing and analysis, supported by the UNM Center for Evolutionary & Theoretical Immunology (CETI) under National Institutes of Health grant P30GM110907. Data analysis was aided by computing resources of the Center for Advanced Research Computing, supported in part by the National Science Foundation. The authors declare no conflict of interest. {ref-type="app"} (with all four genes agreeing in terms of species separation). Both *C. posadasii* and *C. immitis* isolates were present among those obtained from patients in New Mexico (shown in red). As expected, based on previous analyses of isolates from California, Arizona and Texas, most NM isolates were from *C. posadasii* (*C. immitis* being known primarily from CA). One patient infected with *C. immitis* (isolates NM9443 and NM9737) was a resident of the Four Corners region of Utah, while another (isolate NM3006) was from the Four Corners region of NM with no apparent travel history to California. GenBank accession numbers are given in [Table 3](#jof-05-00074-t003){ref-type="table"}. Bootstrap values (percentage of 1000 replicates) greater than 60% are shown for maximum likelihood analysis above the branches and for parsimony analysis below the branches. The tree was rooted with *Aspergillus steynii*. Maximum likelihood and parsimony analyses performed without sequences from *A. steynii*, and employing midpoint rooting, separated *C. immitis* and *C. posadasii* clades with 100% bootstrap support and placed the root between the two species (results not shown). Arrows indicate branches leading to isolates with identical sequences.](jof-05-00074-g001){#jof-05-00074-f001} {#jof-05-00074-f002} jof-05-00074-t001_Table 1 ###### Isolate and patient information. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Isolate Isolate ID ^a^ County or State of Residency Source Type of Infection Age Sex Race/Ethnicity Significant Past Medical History Occupation Travel History ----------------------------- ---------------- ------------------------------ ------------------------------------------------- ----- ----- ------------------ ---------------------------------- ------------------------------- ------------------------- ***C. immitis* Isolates** NM3006 *Ci* San Juan, NM Bone\ 60 M American Indian Yes Railroad employee None Vertebral osteomyelitis NM9443 ^b^ *Ci* Utah Ear 51 M Unk ^c^ Unk Unk Unk NM9737 ^b^ *Ci* Utah Ear 51 M Unk Unk Unk Unk ***C. posadasii* Isolates** NM0369 *Cp* Bernalillo, NM Bronchial wash\ 66 M Unk Yes Unk None Pulmonary infection NM4233 *Cp* Bernalillo, NM Tissue\ 48 M Unk Yes Rug merchant Tucson, AZ, 3 wks prior Pulmonary infection NM9861 *Cp* Bernalillo, NM Tissue\ 30 F Unk Yes Unk None Pulmonary infection NM3894 *Cp* Bernalillo, NM Fluid\ 77 M Unk No None None Knee infection NM0317 *Cp* Chaves, NM Ear\ 62 M African American Yes Retired. Prior military in CA None Pulmonary infection and facial lesion NM4297 *Cp* Eddy, NM Sputum\ 63 M White Yes Mining engineer Travel to Kansas Pulmonary infection NM0071 *Cp* Lea, NM Nasal sputum\ 28 M African American Yes Incarcerated patient N/A Pulmonary infection NM3957 *Cp* McKinley, NM Pleural fusion\ 42 M Unk Unk Unk Unk Pulmonary infection NM9837 *Cp* McKinley, NM Nasal sputum\ 51 M American Indian Yes Welder Recent work Phoenix, AZ Pulmonary infection NM5945 *Cp* McKinley, NM Tissue\ 48 F American Indian Yes Unk None Pulmonary infection NM4708 *Cp* San Juan, NM Bronchial wash\ 40 F American Indian Yes Unk None Pulmonary infection NM0459 *Cp* Socorro, NM Tissue\ 50 F Unk Unk Unk None Pulmonary infection NM7898 *Cp* Torrance, NM Bronchial lavage\ 48 M Unk Unk Incarcerated patient None Pulmonary infection NM8725 *Cp* New Mexico Fluid/Unk Unk Unk Unk Unk Unk Unk NM8945 *Cp* Arizona Bone\ 62 M American Indian Yes Unk None Pulmonary infection in childhood; osteomyelitis ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^*Ci* = *C. immitis*, *Cp* = *C. posadasii*; ^b^ NM9443 and NM9737 were from the same patient; ^c^ Unk = unknown. jof-05-00074-t002_Table 2 ###### Primers used in gene amplification. DNA Region Direction Sequence Annealing Temp. --------------------------- ------------------------------- ------------------------------ ----------------- ITS Forward 5′ CTTGGTCATTTAGAGGAAGTAA 3′ 50 °C Reverse 5′ TCCTCCGCTTATTGATATGC 3′ Serine Proteinase Forward 5′ ATAGAGACCACGCAGAAGGC 3′ 55 °C Reverse 5′ AGCTGTCACGGATGGTATCG 3′ MCM7 Forward 5′ TGGTTATAGCGCGATTCTCC 3′ 50 °C Reverse 5′ CGAGTCGTTATACCTCGAACG 3′ RPB1 Forward 5′ CCGCGTCATTATTTAAGCATC 3′ 55 °C Reverse 5′ AGCGTATTCACCAACTTCTC 3′ *C. posadasii* Intron ^a^ Forward (P2F) 5′ TCAAATCATGTGTAATATGTGG 3′ 50 °C Reverse (P2R) 5′ GTTGACCATAAAGAAAAGTTGG 3′ cox1 exon ^b^ Forward (P1F) 5′ ATAAAATAAACTACGATTTGCG 3′ 50 °C Reverse (P1R) 5′ GATTGCATGAGCTGTAATAATAC 3′ ^a^ This primer pair amplifies an intron sequence from *C. posadasii* but not C. *immitis* (see Results). ^b^ This primer pair amplifies a region in both *C. posadasii* and *C. immitis* that includes a portion of the first cox1 exon along with an upstream intergenic region (see [Section 3](#sec3-jof-05-00074){ref-type="sec"}). jof-05-00074-t003_Table 3 ###### GenBank accession numbers for sequences employed in phylogenetic analysis. Isolate ITS Serine Proteinase MCM7 RPB1 -------------------------------------------------------------------------------------------- ------------------ ------------------- ------------------ ---------------- **Sequences from New Mexico and Four Corners Region *Coccidioides* Isolates (This Study)** NM3006 MH725248 MH748764 MH748742 MH748724 NM9443 MH725258 MH748774 MH748757 MH748739 NM9737 MH725259 MH748775 MH748758 MH748740 NM0071 MH725244 MH748760 MH748749 MH748731 NM0317 MH725245 MH748761 MH748748 MH748730 NM0369 MH725246 MH748762 MH748759 MH748741 NM0459 MH725247 MH748763 MH748746 MH748728 NM3894 MH725249 MH748765 MH748754 MH748736 NM3957 MH725250 MH748766 MH748756 MH748738 NM4233 MH725251 MH748767 MH748743 MH748725 NM4297 MH725252 MH748768 MH748752 MH748734 NM4708 MH725253 MH748769 MH748744 MH748726 NM5945 MH725254 MH748770 MH748751 MH748733 NM7898 MH725255 MH748771 MH748747 MH748729 NM8725 MH725256 MH748772 MH748755 MH748737 NM8945 MH725257 MH748773 MH748750 MH748732 NM9837 MH725260 MH748776 MH748745 MH748727 NM9861 MH725261 MH748777 MH748753 MH748735 **Sequences from Existing GenBank *Coccidioides* Entries** RMSCC 2394 AATX01000513.1 AATX01000326.1 AATX01000203.1 AATX01000264.1 224--846 35438--36110 43422--44141 433820--434247 H538.4 AASO01000085.1 AASO01002025.1 AASO01002210.1 AASO01003054.1 10339--10766 35115--35787 4641--5360 10945--11568 RS AAEC03000009.1 AAEC03000008.1 AAEC03000005.1 AAEC03000010.1 4725--5351 2537048--2537720 3687326--3688045 828825--829252 CPA 0020 ABIV01003320.1 ABIV01000896.1 ABIV01000569.1 ABIV01001762.1 16973--17598 3088--3760 311--1030 5431--5860 CPA 0001 ABFO01003353.1 ABFO01000797.1 ABFO01001238.1 ABFO01003988.1 1--346 ^a^ 2878--3503 665--1337 ABFO01001237.1 6129--6558 12376--12757 ^a^ RMSCC 3700 ABFN01001891.1 ABFN01000336.1 ABFN01000290.1 ABFN01001137.1 3121--3746 1--638 5242--5961 412--841 RMSCC 2133 ABFM01000924.1 ABFM01000717.1 ABFM01000464.1 ABFM01000297.1 22927--23552 3481--4153 1904--1262 6563--6992 RMSCC 3488 ABBB01000249.1 ABBB01000240.1 ABBB01000156.1 ABBB01000255.1 33179--33804 371029--371701 145000--145719 527482--527911 Silveira KM588216.1 ABAI02000152.1 ABAI02000361.1 ABAI02000102.1 34354--35026 12054--12773 54472--54901 **Sequences from Existing GenBank *Aspergillus steynii* Entries** IBT 23096 MSFO01000033.1 MSFO01000001.1 MSFO01000005.1 MSFO01000005.1 3264--2712 3181742--3182796 1831768--1834777 2491035--2491482 ^a^ The complete CPA0001 MCM7 gene is split across two contig accessions.

Neurocysticercosis is a leading cause of acquired epilepsy in the developing world ([@R1]*,*[@R2]). The disease occurs when larvae of the pork tapeworm, *Taenia solium*, encyst in the human brain; this process causes a broad range of neurologic signs and symptoms, including seizures, headache, obstructive hydrocephalus, encephalitis, stroke, and cognitive and other mental health disorders ([@R3]*,*[@R4]). Neurocysticercosis is endemic in poor rural communities in Latin America, sub-Saharan Africa, and Asia, where pigs can access and ingest human feces ([Figure 1](#F1){ref-type="fig"}). However, the disease is also of increasing public health concern in the United States, especially in the immigrant population and among persons who have traveled to regions where cysticercosis is endemic ([@R5]). {#F1} The World Health Organization designates cysticercosis as a neglected tropical disease (NTD) and has called for international efforts to strengthen surveillance ([@R6]*--*[@R8]). The disease remains neglected partly because the scale of the problem has not been well defined ([@R2]). In most disease-endemic regions, population-level data are sparse because surveillance for neurocysticercosis is nonexistent and diagnostic neuroimaging is typically unavailable. In the United States, there is an opportunity to collect population-based data on neurocysticercosis because of the large immigrant population at risk for infection, the widespread availability of neuroimaging, and the well-established disease surveillance infrastructure. However, only Alaska, Arizona, California, New Mexico, Oregon, and Texas require reporting of neurocysticercosis. Death rates due to neurocysticercosis in the United States have been reported previously ([@R9]), but national-level assessments of neurocysticercosis that use population--based data are lacking. The objective of our study was to evaluate the frequency and total associated charges for hospitalizations due to neurocysticercosis in the United States and to compare these against other tropical diseases of potential importance in the United States. Methods ======= Data Source ----------- We analyzed hospital discharge data contained in the Nationwide Inpatient Sample (NIS) for 2003--2012 ([@R10]*,*[@R11]). The NIS, a stratified weighted sample of short-term and nonfederal hospitals, is designed to approximate a 20% sample of all community hospitals in the United States. As of 2012, 47 states participated in reporting discharge data to the NIS (only Alabama, Delaware, Idaho, and the District of Columbia had not participated), creating a sample representing 95% of the national population. The NIS is the largest collection of longitudinal inpatient care data in the United States and holds information on ≈8 million hospitalizations from \>1,000 hospitals each year ([@R10]). NIS data are de-identified and include information on demographics, diagnostic and procedural codes, length of stay, discharge status, total charges, and expected payees associated with each hospitalization. Case Definitions ---------------- We based our case definitions for hospitalization on diagnostic and procedural codes from the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM). The ICD-9-CM code listed in the first diagnostic field is intended to capture the primary reason for hospitalization. However, there is no specific ICD-9-CM code for neurocysticercosis, so coding patterns may vary. For example, a hospitalization for neurocysticercosis might be coded with a first diagnostic field of 123.1 (cysticercosis) or with a neurologic code, such as 345.9 (epilepsy unspecified), in combination with 123.1 in a different diagnostic field. We used 2 case definitions in this analysis. The first was a conservative case definition for reporting hospitalizations associated with neurocysticercosis. This definition required the ICD-9-CM code for cysticercosis (123.1) in any of the 15 available diagnostic fields and a supporting diagnostic or procedural code associated with a clinical manifestation of neurologic disease in any of the first 5 diagnostic or procedural fields ([Table 1](#T1){ref-type="table"}). We used individual ICD-9-CM codes and coding groups defined by Clinical Classification Software to define these additional diagnostic or procedural codes ([@R13]). This conservative case definition was designed to reduce the likelihood of including hospitalizations for persons carrying an existing diagnosis of neurocysticercosis who were hospitalized for an unrelated condition. ###### Supporting diagnostic or procedural codes from ICD-9-CM used for conservative case definition for reporting hospitalizations associated with neurocysticercosis\* CCS code Diagnosis or procedure ------------------ ---------------------------------------------------------------------- Diagnostic code 76 Meningitis 77 Encephalitis 78 Other CNS infection and poliomyelitis 83 Epilepsy; convulsions 84 Headache; including migraine 85 Coma; stupor; and brain damage 90 Inflammation; infection of eye 95 Other nervous system disorders 109 Acute cerebrovascular disease 111 Other and ill-defined cerebrovascular disease 112 Transient cerebral ischemia 245 Syncope 650 Adjustment disorders 651 Anxiety disorders 652 Attention-deficit, conduct, and disruptive behavior disorders 653 Delirium, dementia, and amnestic and other cognitive disorders 656 Impulse control disorders, NEC 657 Mood disorders 658 Personality disorders 659 Schizophrenia and other psychotic disorders 662 Suicide and intentional self-inflicted injury 670 Miscellaneous mental disorders Procedural codes 1 Incision and excision of CNS 2 Insertion; replacement; or removal of extracranial ventricular shunt 177 Computerized axial tomography (CT) scan head 198 Magnetic resonance imaging 199 Electroencephalogram (EEG) \*A complete list of ICD-9-CM codes used in this study is provided in the online Technical Appendix (<http://wwwnc.cdc.gov/EID/article/21/6/14-1324-Techapp1.pdf>). ICD-9-CM, International Classification of Diseases, 9th Revision, Clinical Modification; CCS, Clinical Classification Software groupings of ICD-9-CM codes ([@R12]); CNS, central nervous system; NEC, not elsewhere classified. The second case definition was designed to facilitate comparison of hospitalizations for cysticercosis with those for the 16 other NTDs and malaria. The case definition for cysticercosis included all hospitalizations with an ICD-9-CM diagnostic code for cysticercosis (123.1) in any of the first 15 diagnosis fields, but it did not require an additional supportive diagnostic or procedural code. Similarly, the case definitions for the other tropical diseases in the comparative analysis relied on ICD-9-CM codes specific to the disease without a requirement for an additional supportive diagnostic or procedural code. This approach ensured consistency of case definitions across the various diseases at the expense of greater specificity. We assumed that the likelihood of capturing unrelated hospitalizations was similar for the diseases we compared. We excluded Buruli ulcer from our comparison because there is no ICD-9-CM code specific for this disease. However, to our knowledge, Buruli ulcer has not been reported in the United States ([@R14]). We did not report hospitalizations for rabies, African trypanomiasis, or dracunculiasis because the numbers of hospitalizations were too low (\<10/year) to provide accurate estimates. A list of ICD-9-CM codes used in all case definitions is provided in the [Technical Appendix](#SD1){ref-type="local-data"}. Statistical Methods ------------------- To account for the sampling design of the NIS, we analyzed all data by using the survey family of commands in Stata 13 (StataCorp LP, College Station, TX, USA). We applied hospital discharge weights provided in the NIS to estimate total national hospitalizations on the basis of the stratified sample. All sampled hospitals, regardless of whether they had a patient who was hospitalized with neurocysticercosis, were included for calculation of SEs and CIs. We examined neurocysticercosis hospitalizations by patient age, sex, race, place of service, discharge status, and length of stay; US region; associated diagnostic and procedural codes; and hospitalization charges. State-level assessment was not possible because of the sampling and stratification strategy used in the NIS. Mean annual hospitalization rates were calculated as the weighted number of hospitalizations per 100,000 population on the basis of the US Census Bureau data for each year during the study period ([@R15]). Age- and sex-adjusted rates were calculated by using the direct standardization method and the 2005 US Census population as the reference population. We used Gaussian family generalized linear models with logarithmic function link within the Stata survey framework to estimate the crude and adjusted mean length of stay and mean hospitalization charges. We first constructed univariate generalized linear models to evaluate demographic variables of interest, retaining those that were significant at the p\<0.2 level (Wald test) in the final multivariate models. The independent categorical variables we evaluated were sex, age, race, hospital region, and year of hospitalization. Once we built the final models, we estimated the mean length of stay and mean hospitalization charges for diagnoses and procedures commonly seen with neurocysticercosis (i.e., seizures, obstructive hydrocephalus, headache, stroke, mental health disorder, encephalitis/meningitis, cerebral edema, syncope, neuroimaging, ventricular shunt management, and central nervous system surgery) by individually introducing dummy variables encoding these clinical variables into the models. Inflation-adjusted charges were used in all models. Hospital Charges ---------------- We analyzed hospital charges that were billed to private insurance, Medicaid, Medicare, and other sources from the payer's perspective. Charges represent the amount that hospitals billed for services, not the actual cost of providing these services. Generally, total charges did not include professional fees, noncovered charges, or charges incurred in the emergency department unless the patient was admitted directly from the emergency department into the hospital. We adjusted all charges for inflation by using the Consumer Price Index, setting the base year to 2012. Results ======= During 2003--2012, an estimated 23,266 hospitalizations (95% CI 21,741--24,792) in the United States were assigned an ICD-9-CM code of 123.1 in any of the first 15 diagnostic fields. Of these hospitalizations, 18,584 (95% CI 17,322--19,846), approximately 80% of the total, met our case definition of hospitalization due to neurocysticercosis. The number of hospitalizations due to neurocysticercosis per year ranged from a high of 2,247 in 2006 to a low of 1,495 in 2012. The largest proportion of hospitalizations due to neurocysticercosis occurred in the western region (n = 8,026, 42.9% \[95% CI 39.2%--46.7%)\], followed by the southern region (n = 5,860, 31.8% \[95% CI 28.6%--35.1%\]), the northeastern region (n = 2,902, 15.5% 95% CI \[13.5%--17.6%\]) and the midwestern region (n = 1,796, 9.8% \[95% CI 8.2%--11.7%\]). We found distinct differences in the mean annual incidence rates of hospitalization stratified by age, sex, and race ([Table 2](#T2){ref-type="table"}). The mean annual incidence of hospitalization was highest in 20- to 44-year-old age group (1.04 hospitalizations/100,000 population). Hospitalization rates were 33% higher among male patients than female patients. The age- and sex-adjusted mean annual incidence of hospitalizations was highest among Hispanics (2.50 hospitalizations/100,000 population); the rate was 35 times higher than that for non-Hispanic whites, 10 times higher than that for blacks, and 8 times higher than that for Asian/Pacific Islanders. Unadjusted rates by race were similar: Hispanic, 2.57/100,000; white, 0.06/100,000; black, 0.23/100,000; and Asian/Pacific Islander, 0.26/ 100,000. ###### Number and rate of hospitalizations for neurocysticercosis in the United States, by demographic group, 2003--2012\* Characteristic† No. hospitalizations (SE) \% All hospitalizations (95% CI) Rate (95% CI)‡ -------------------------- --------------------------- ---------------------------------- ------------------- Age, y   \<20 1,493 (103) 8.0 (7.1--9.1) 0.18 (0.16--0.21)   20--44 10,827 (394) 58.3 (56.5--60.1) 1.04 (0.97--1.12)   45--64 4,357 (232) 23.5 (22.0--25.0) 0.56 (0.51--0.62)   ≥65 1,889 (136) 10.2 (9.0--11.5) 0.49 (0.42--0.56) Sex   M 10,377 (373) 56.3 (54.5--58.2) 0.70 (0.65--0.75)   F 8,043 (349) 43.7 (41.8--45.5) 0.52 (0.48--0.57) Race/ethnicity   Hispanic 12,030 (551) 74.0 (71.5--76.3) 2.50 (2.27--2.73)   White 1,530 (104) 9.4 (8.2--10.7) 0.07 (0.06--0.08)   Black 900 (95) 5.5 (4.5--6.8) 0.25 (0.21--0.30)   Asian/Pacific Islander 377 (61) 2.3 (1.7--3.2) 0.31 (0.23--0.39) Overall 18,584 (644) 0.61 (0.57--0.66) \*National estimates were based on the Nationwide Inpatient Sample, by using diagnosis code 123.1 from the International Classification of Diseases, 9th 
Revision, Clinical Modification.
†Missing data not presented
‡Rate for age and sex are unadjusted. Rate for race/ethnicity is adjusted for age and sex by direct standardization method by using 2005 US Census data. Rates expressed as mean annual incidence per 100,000 population. Length of Stay, Total Charges, and Payees ----------------------------------------- The mean length of hospitalization was 6.0 (95% CI 5.7--6.4) days and did not show a significant trend over the study period (p = 1.0). Total inflation-adjusted hospitalization charges over the 10-year study period were US \$908,238,000 (95% CI US \$814,483,000--\$1,001,992,000), increasing 27% from US \$72,560,000 in 2003 to US \$91,959,000 in 2012. The mean charge per hospitalization was US \$50,976 (95% CI US \$47,492--\$54,716), increasing 50% over the 10-year study period from US \$41,874 in 2003 to US \$62,986 in 2012. After we adjusted for demographic variables, mean length of stay and mean hospitalization charges were substantially higher for male patients, middle-aged adult patients, and patients from the western region (online Technical Appendix). Publically funded insurance (Medicaid or Medicare) was the primary payer in 40% of the hospitalizations ([Table 3](#T3){ref-type="table"}). ###### Source of admission, disposition, and expected payer for hospitalizations due to neurocysticercosis, United States, 2003--2012\* Characteristic No. hospitalizations (SE) \% All hospitalizations (95% CI) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------- ---------------------------------- Source of admission   Emergency department 9,436 (484) 74.7 (72.3--76.9)   Routine 2,210 (145) 17.5 (15.6--19.6)   Transfer 947 (80) 7.5 (6.4--8.7) Disposition   Routine 15,693 (562) 84.5 (83.1--85.8)   Transfer 1,617 (107) 8.7 (7.7--9.8)   Home health 834 (81) 4.5 (3.8--5.4)   Against medical advice 205 (33) 1.1 (0.8--1.5)   Died 218 (32) 1.2 (0.9--1.6) Expected primary payer   Medicare 2,025 (139) 10.9 (9.7--12.3)   Medicaid 5,543 (316) 29.9 (27.8--32.1)   Private insurance 4,335 (206) 23.4 (21.4--25.4)   Self-pay 4,753 (224) 25.6 (23.8--27.5)   Other payer 1,883 (160) 10.2 (8.9--11.6) \*National estimates were determined on the basis of the Nationwide Inpatient Sample, by using diagnosis code 123.1 from the International Classification of Diseases, 9th Revision, Clinical Modification. Associated Diagnoses and Procedures ----------------------------------- The most common diagnosis group associated with hospitalization for neurocysticercosis was epilepsy/convulsions, which occurred in 57.3% of hospitalizations, followed by obstructive hydrocephalus (17.7%) and headache (12.4%) ([Table 4](#T4){ref-type="table"}). After we controlled for year and patient demographics, the diagnoses associated with the longest mean length of stay and the highest mean charges were encephalitis/meningitis (12.2 days and US \$78,984) and hydrocephalus (11.4 days and US \$79,084). Diagnostic codes for syncope and headache were associated with the shortest stays and lowest charges (3.4 days and US \$20,017 and 3.8 days and US \$19,893, respectively). Procedure codes for shunt management (insertion, removal, or repair) were associated with a mean length of stay of 16.3 days and mean hospitalization charges of US \$86,272; codes for brain surgery (central nervous system incision or excision) were associated with a mean length of stay of 10.3 days and mean hospitalization charges of US \$89,893. Only 17% of hospitalizations included a procedural code for either computed tomography scans or magnetic resonance imaging of the head. ###### Diagnostic and procedure codes for hospitalizations due to neurocysticercosis, United States, 2003--2012\* Associated diagnoses and procedures No. hospitalizations (SE) \% All hospitalizations (95% CI) Mean length of stay, d† Mean charges, US\$† ------------------------------------------------ --------------------------- ---------------------------------- ------------------------- -------------------------- Diagnoses   Epilepsy; convulsions 10,652 (360) 57.3 (55.3--59.3) 5.4 (4.2--6.9) 33,058 (21,846--50,023)   Obstructive hydrocephalus 3,292 (208) 17.7 (16.2--19.3) 11.4 (8.1--16.2) 79,084 (46,139--135,552)   Headache, including migraine 2,308 (126) 12.4 (11.3--13.6) 3.8 (2..9--4.9) 19,893 (12,422--31,857)   Cerebrovascular disease 1,650 (121) 8.9 (7.9--10.0) 7.5 (5.3--10.6) 45,183 (26,027--78,436)   Mental health disorder 1,843 (132) 9.9 (8.8--11.2) 6.4 (4.7--8.8) 24,436 (13,578--43,979)   Encephalitis/meningitis 1,033 (81) 5.6 (4.8--6.4) 12.2 (8.1--18.6) 78,983 (46,851--133,151)   Cerebral edema 931 (79) 5.0 (4.3--5.9) 7.5 (5.6--10.0) 40,639 (23,449--70,429)   Syncope 573 (56) 3.1 (2.6--3.7) 3.4 (2.6--4.6) 20,017 (11,934--33,577) Procedures   Neuroimaging, CT of head or MRI 3,087 (330) 16.6 (13.9--19.8) 6.1 (4.8--7.7) 34,905 (22,177--54,937)   Ventricular shunt, insert, remove, or repair 1,661 (137) 8.9 (7.9--10.2) 16.3 (10.6--25.1) 86,272 (48,313--154,054)   CNS incision or excision 1,499 (111) 8.1 (7.1--9.2) 10.3 (7.9--13.5) 89,893 (56,625--142,709) \*National estimates were based on the Nationwide Inpatient Sample, by using diagnosis code 123.1 from the International Classification of Diseases, 9th Revision, Clinical Modification. CNS, central nervous system; CT, computed tomography; MRI, magnetic resonance imaging.
†Mean length of stay and mean inflation-adjusted hospitalization charges for diagnostic and procedure codes after adjusting for year, patient age, sex, race, and hospital region. Diagnostic and procedure codes were evaluated individually as independent variables in the final generalized linear models built for length of stay and charges. Comparison of NIS Data for Cysticercosis with that for NTDs and Malaria ----------------------------------------------------------------------- The frequency of and total charges for hospitalizations due to cysticercosis exceeded those for all other NTDs combined ([Figure 2](#F2){ref-type="fig"}). During 2003--2012, an estimated 23,266 (95% CI 21,741--24,792) hospitalizations were associated with a diagnosis code for cysticercosis, resulting in US \$1,149,044,000 in total hospital charges (95% CI US \$1,038,730,000--\$1,259,357,000). In contrast, there were 20,029 hospitalizations and US \$1,043,109,000 in total charges for all of the other NTDs combined ([Table 5](#T5){ref-type="table"}). {#F2} ###### Hospitalizations and total charges for neglected tropical diseases and malaria, United States, 2003--2012\* Disease Hospitalizations Total charges -------------------------------------------------- ------------------ ---------------- --------------- ------------ -------------- Cysticercosis 23,266 (778) 21,741--24,792 1,149 (56) 1,039--1,259 Malaria 14,319 (434) 13,469--15,169 387 (18) 351--423 Echinococcosis 3,919 (170) 3,586--4,252 206 (16) 174--237 Soil-transmitted helminth--associated infections 3,256 (151) 2,959--3,552 201 (19) 162--239 Dengue 2,644 (135) 2,379--2,909 89 (9) 70--107 Leprosy 2,055 (135) 1,791--2,319 94 (9) 76--111 Lymphatic filariasis 1,836 (106) 1,629--2,044 86 (9) 68--103 Schistosomiasis 1,811 (120) 1,576--2,046 101 (12) 78--125 Chagas disease 1,686 (151) 1,389--1,982 118 (17) 84--152 Leishmaniasis 1,022 (92) 841--1,203 52 (7) 38--66 Trachoma 649 (69) 514--784 20 (4) 13--28 Foodborne trematode--associated infections 610 (60) 492--729 41 (7) 28--54 Onchocerciasis 380 (47) 287--473 29 (12) 5--53 Yaws 161 (28) 106--216 7 (2) 3--11 \*National estimates were determined on the Nationwide Inpatient Sample by using diagnostic codes from the International Classification of Diseases, 9th Revision, Clinical Modification. A complete list of ICD-9-CM codes used in this study is provided in the online Technical Appendix (<http://wwwnc.cdc.gov/EID/article/21/6/14-1324-Techapp1.pdf>). Discussion ========== The study findings demonstrate that neurocysticercosis poses considerable health and economic problems in the United States, especially among the Hispanic population. Over the 10-year study period, \>18,500 hospitalizations for neurocysticercosis occurred, totaling hospital charges of \>US \$908 million, of which 40% was billed to publicly funded insurance programs. Hospitalization stays were prolonged and expensive, reflecting the complicated nature of acute disease management. Hospitalizations and associated charges for cysticercosis exceeded the totals for malaria and for all of the other NTDs combined. The hospitalization rates we report in this nationwide study are comparable to those reported in previous state- or county-level studies, providing support for the case definition we used ([@R12],[@R15]*--*[@R20]). Because there is no ICD-9-CM diagnostic code specific for neurocysticercosis, the case definitions varied slightly among these studies. The estimated overall hospitalization rate of 0.65/100,000 population that we report falls between the rates previously observed in California (0.8--1.1 hospitalization/100,000 population) and Oregon (0.2--0.5 hospitalizations/100,000 population) ([@R12],[@R15]*--*[@R18]). Risk for hospitalization was highest among Hispanic, male, and young to middle-aged adult patients in all studies. Nearly three quarters of all patients hospitalized for neurocysticercosis in the United States were Hispanic. The Hispanic population is the largest minority group in the United States and among the fastest growing US population groups. Because the hospitalization rate for the Hispanic population is 36 times greater than that of the non-Hispanic white population, the effect of neurocysticercosis on the US economy is likely to increase substantially in the coming years. The US Census Bureau projects that the Hispanic population will grow from 53 million in 2012 to \>78 million by 2030 ([@R21]). Without changes in the rate of hospitalization or the increase in mean hospitalization charges, there could be \>1,900 hospitalizations and US \$250 million total charges related to neurocysticercosis among Hispanics alone in the year 2030. Changing immigration patterns may also bring an influx of cases in persons from other regions of the world where neurocysticercosis is endemic, particularly Asia and sub-Saharan Africa. Several hospital-based studies have shown that seizures are the most frequent reason for hospitalization for neurocysticercosis ([@R3]*,*[@R4]*,*[@R22]). In this study, epilepsy was the most frequent diagnosis associated with hospitalization for neurocysticercosis; it was coded in more than half of all hospitalizations for the disease. Seizures in neurocysticercosis are typically amenable to therapy with antiepileptic and anti-inflammatory drugs, resulting in relatively uncomplicated and short hospital stays. In contrast, more severe disease may require intensive interventions and longer hospitalizations, resulting in higher charges ([@R23]*--*[@R25]). While diagnoses of obstructive hydrocephalus or encephalitis/meningitis occurred in ≈20% of persons hospitalized for neurocysticercosis, these more severe presentations accounted for 40% of the total charges incurred. We report hospitalization diagnostic codes that may not represent the distribution of disease manifestations experienced by individual patients. For example, although a diagnostic code for headache was listed for 11% of hospitalized patients, only patients with headaches associated with underlying pathology requiring acute intervention, such as obstructive hydrocephalus, are likely to be admitted and therefore represented in this study. Even then, the diagnosis of headache may be underrepresented. There were twice as many hospitalizations with diagnostic codes for hydrocephalus and encephalitis than for headache, although both of these manifestations would be expected to be associated with headache ([@R22]). Similar caution is suggested in interpreting the frequency of other diagnoses presented here. It may seem contradictory that only 17% of hospitalizations had a procedural code for neuroimaging. However, because most imaging for neurocysticercosis would be expected to occur in the emergency department before admission, the infrequent coding for neuroimaging may reflect exclusion of these procedural codes from the hospital discharge summary. This study documents the substantial costs of hospitalizations due to neurocysticercosis in the United States, but the true effect of neurocysticercosis on the US health care system is likely much greater. Only those emergency department visits that result directly in inpatient admission are captured in the hospital discharge databases in the NIS. In Oregon ([@R15]), over 40% (31/72) of all patients with neurocysticercosis were seen only in the emergency department and were not admitted to the hospital. While nonadmissions likely represent cases of less clinical severity, substantial charges are still incurred in the emergency department and in outpatient follow-up. Neurocysticercosis is also likely to be substantially underdiagnosed and misdiagnosed because of the lack of a definitive diagnostic test and limited provider awareness of the disease. Neurocysticercosis also often results in chronic disease that requires outpatient follow-up with infectious disease or neurology specialists, none of which is captured in this study. Management of neurocysticercosis may involve long-term antiepileptic therapy, prolonged regimens of antiparasitic drugs and high-dose corticosteroids, monitoring and repair of ventriculoperitoneal shunts, and treatment of frequent complications resulting from these interventions ([@R26]*,*[@R27]). A chart review at the outpatient neurology clinic in a Houston hospital showed that 2% of all patients were seen for management of neurocysticercosis ([@R28]). A few states are now collecting comprehensive claims data covering health care provided in inpatient, outpatient, and long-term care settings. Data from these programs could provide more complete information about health care and associated costs related to management of neurocysticercosis in all settings. The high neurocysticercosis hospitalization rate we noted in young adults and men suggests substantial indirect costs to the US domestic workforce. Loss of worker productivity should also be considered in the overall costs of neurocysticercosis. The use of administrative databases, such as the NIS, to obtain data for this study does have drawbacks, including several limitations we already described. An additional drawback to using the NIS was the inability to identify multiple hospitalizations for a single person, which precludes the ability to estimate the prevalence or incidence of disease. Although the number of states participating in NIS has grown over the years, several states still do not participate in reporting hospital discharge data. In addition, the regional sampling structure of the NIS does not allow for accurate state-level estimates, limiting the ability to identify specific states whose populations are at increased risk for neurocysticercosis. Furthermore, the lack of in-depth demographic and clinical information in the NIS limits the type of questions that can be addressed. For example, knowing the country of birth or travel history of patients with neurocysticercosis could help understand their source of exposure. Although the primary purpose of this study was to evaluate hospitalizations for neurocysticercosis, we also compared hospitalizations for cysticercosis with those for other NTDs and malaria. Our findings showed that the number of hospitalizations for cysticercosis was nearly 2 times the number for malaria, and the associated hospital charges were nearly 3 times higher. In addition, hospitalizations and charges for cysticercosis were higher than those for all other NTDs we evaluated combined. This comparative analysis was not meant to be exhaustive; we recognize that many factors other than hospitalization contribute to the public health effect of any particular disease. However, the markedly higher number of hospitalizations and associated charges related to cysticercosis, compared with those for other NTDs and malaria in the United States, merits attention and further exploration. ###### **Technical Appendix.** Diagnostic codes related to neurocysticercosis, region groups for states included in the Nationwide Inpatient Sample, and regression output for hospitalization charges and hospital stay. *Suggested citation for this article*: O'Neal SE, Flecker RH. Hospitalization frequency and charges for neurocysticercosis, United States, 2003--2012. Emerg Infect Dis. 2015 Jun \[*date cited*\]. <http://dx.doi.org/10.3201/eid2106.141324> Dr. O'Neal is an assistant professor of public health and preventive medicine at Oregon Health & Science University, Portland, Oregon, USA. His primary research interest is the epidemiology and control of *Taenia solium* infection. Dr. Flecker is currently obtaining a Masters of Public Health (epidemiology/biostatistics) at Oregon Health & Science University. He concentrates his research on emerging zoonotic diseases.

All relevant data are within the manuscript and its Supporting Information files. Introduction {#sec001} ============ Oral squamous cell carcinoma (OSCC) is the most lethal type in head and neck squamous cell carcinoma (HNSCC) in the world. Over the past decades, the incidence rate of OSCC has increased among younger generations \[[@pone.0213463.ref001]--[@pone.0213463.ref003]\]. In spite of considerable advances in surgery, radiotherapy and chemotherapy, the 5-year survival rate for OSCC has not improved markedly because patients still frequently arise loco-regional recurrence and lymph node metastasis \[[@pone.0213463.ref004]--[@pone.0213463.ref006]\]. Hence, finding new biomarker(s) and therapeutic molecule(s) is urgent. Recently, a growing evidence indicates that microRNAs (miRNAs) contribute to the initiation and development of oral cancer \[[@pone.0213463.ref007]--[@pone.0213463.ref009]\]. Therefore, exploring unique miRNAs and related molecular pathways underlying OSCC aggressive will provide advantages to improve therapeutic efficacy, as well as to design more effective treatment strategies. Previously, we established a dysregulated signature of eighty-four miRNAs from OSCC clinical samples using a miRNA microarray \[[@pone.0213463.ref010]\]. From this analysis, we found that miR-450a was significantly overexpressed in tumor tissues than that in corresponding adjacent normal tissues. MiR-450a is an intragenenic miRNA clustered with miR-542-5p, miR-542-3p, miR-503, miR-450b-5p, miR-450b-3p, and miR-424 on chromosomal location Xq26.3. High expression level of miR-450a performs as a potential oncogene in laryngeal squamous cell carcinoma and breast cancer \[[@pone.0213463.ref011], [@pone.0213463.ref012]\]. Up-regulated miR-450a is found in mesenchymal part of epithelial-to-mesenchymal transition (EMT)-activation in human endometrial carcinosarcoma \[[@pone.0213463.ref013]\]. Contrarily, downregulation of miR-450a is required in hepatocellular carcinoma carcinogenesis \[[@pone.0213463.ref014]\]. Although miR-450a was reported to be dysregulated in different cancer types, its functions and underlying mechanisms have not been elucidated, especially in oral cancer. Transmembrane proteins (TMEMs) are a group of novel proteins, which have key roles in cell differentiation and tumorigenesis in many cancers, such as pancreatic cancer, prostate cancer, ovarian cancer, and renal cell carcinoma \[[@pone.0213463.ref015]--[@pone.0213463.ref019]\]. However, the role of TMEM182 in oral cancer is unknown. In this study, we demonstrated that miR-450a may function as an oncogene by directly targeting 3'-untranslated region (UTR) of TMEM182 and reduce its expression in OSCC. Downregulation of TMEM182 or overexpression of miR-450a has the same effect on reducing cellular adhesion of OSCC cells. In addition, we also found that miR-450a expression was increased by TNF-α through ERK1/2-dependent pathway, more than via NF-κB. Taken together, these findings may provide understanding into oral carcinogenesis and suggest new therapeutic opportunities in this cancer. Materials and methods {#sec002} ===================== Human samples {#sec003} ------------- The study protocol was approved by the Research Ethics Committee of National Health Research Institutes (EC1040409-E) and Institutional Human Experiment and Ethic Committee of National Cheng Kung University Hospital (HR-97-100) for the use of clinical materials for research purpose. Thirty-five paired primary OSCC and their adjacent non-tumorous epithelial samples were obtained from patients with curative surgery at the National Cheng Kung University Hospital (Tainan, Taiwan) from 1999 to 2010. All human tissues were snap-frozen in liquid nitrogen. Total RNA was extracted by miRNeasy Mini Kit (Qiagen, \#217004) followed by instruction manual. The patient's backgrounds and clinical parameters were summarized in [S1 Table](#pone.0213463.s004){ref-type="supplementary-material"}. The prognostic value of miR-450a among The Cancer Genome Atlas (TCGA) HNSCC cohort was analyzed through SurvMicro database (<http://bioinformatica.mty.itesm.mx:8080/Biomatec/Survmicro.jsp>) and was uploaded to GEO (Access number GSE36682). Forty paired of OSCC patients cDNA microarrays analysis were performed according to our previous study deposited in GEO (Accession number GSE37991 and GSE45238) \[[@pone.0213463.ref010]\]. Cell culture {#sec004} ------------ Cultured conditions for all human OSCC cell lines were summarized as previously described \[[@pone.0213463.ref015]\]. Human oral keratinocytes (HOK) were purchased from ScienCell (Carlsbad, CA, USA) and cultured according to the manufacturer's instructions. All cells were maintained at 37°Clin a 5% CO~2~ atmosphere properly. Cytokine and chemical inhibitor treatment {#sec005} ----------------------------------------- Cells were incubated at low-serum conditional medium for 24 h, before TNF-α (10ng/ml) addition as indicated times. For intrinsic pathway analysis, cells were incubated with each of the following inhibitors: 1 μM for human dysplasia oral keratinocyte (DOK) and 10 μM for human tongue cancer cells (SAS) of NFκB inhibitor (Calbiochem, 481406), 30 μM U0126 (ERK inhibitor)(Cell Signaling, 9903), and 30 μM SB203580 (p38 inhibitor)(Cell Signaling, 5633). RNA extraction and reverse-transcription PCR (RT-PCR) {#sec006} ----------------------------------------------------- Total RNA was extracted from OSCC cell lines using TRIzol reagent (Life Technologies, Gaithersburg, MD) according to the manufacturer's instructions. RNA concentration and purification were checked by NanoDrop ND-1000 spectrophotometer. First-strand cDNAs were synthesized by NxGen^™^M-MuLV reverse transcriptase with oligo dT~12-18~ primer (Invitrogen, Carlsbad, CA). Gene expression analyses were assayed on a Biometra T3000 thermocycler (Biometra GmbH, Germany) as following conditions: 95°C for 5 min, followed by 35--40 cycles of amplification (95°C for 30s, 60°C for 30s, and 72°C for 30s), and 72°C for 10 min. GAPDH was used as a loading control. PCR products were subjected to electrophoresis on 2% agarose gel and visualized on UVP GDS-8000 Bioimaging System (UVP, CA, USA) with 0.01% of SYBRSafe (Invitrogen, Carlsbad, CA, USA) inner staining. Primer sequences are listed in [S2 Table](#pone.0213463.s005){ref-type="supplementary-material"}. Quantitative real-time PCR (qPCR) {#sec007} --------------------------------- Mature miR-450a and RNU44 internal control levels were analyzed by miRNA-specific stem-loop primers and TaqMan Universal PCR Master Mix (Applied Biosystems) on an Applied Biosystems StepOne Plus real-time PCR system. Fold changes were calculated by using2^-ΔΔCt^ method using control and reference normalized. Primer sequences are listed in [S2 Table](#pone.0213463.s005){ref-type="supplementary-material"}. Plasmids {#sec008} -------- A wild type 3′-UTR of TMEM182 containing the miR-450a binding sites (3\'UTR-WT) and truncated 3′-UTR fragment with deleted miR-450a binding sites (3\'UTR-DEL) were constructed into the XhoI/XbaI sites of pmiRGLO firefly luciferase-expressing vector (Promega, WI, USA). For gene knockdown experiments, the shRNA clones of TMEM182 (sh182 \#1 & \#2) and empty vector pLKO_TRC (shCTRL) were obtained from the National RNAi Core Facility (Academia Sinica, Taiwan). Human TMEM182 cDNA was sub-cloned into empty vector pCDH-CMV-GFP puro+ (vehicle) (System Biosciences) at EcoRI/BamHI sites and termed as TMEM182-flag. Human TMEM182 cDNA was sub-cloned into empty vector pEGFPN1 (BD Biosciences Clontech's) (vehicle) at XhoI/BamHI sites, termed as TMEM182-GFP. The sequence data were compared against the National Center for Biotechnology Information (NCBI) database using BLAST and miRBase (<http://www.mirbase.org/>). Kyte-Doolittle hydrophobicity analysis was accessed to predict TMEM182 transmembrane portions using ExPASY (<https://web.expasy.org/protscale/>). List primer sequences in the following [S2 Table](#pone.0213463.s005){ref-type="supplementary-material"}. Protein extraction and western blot {#sec009} ----------------------------------- Cell lysates were prepared as previously reported.\[[@pone.0213463.ref020]\] Equal amounts of protein lysates were separated by 10\~12% SDS polyacrylamide gels and transferred to poly-vinylidene fluoride (PVDF) membrane (Pall Life Sciences, Glen Cove, NY). Immunoblotting was performed with specific antibodies against TMEM182 (ab177360; Abcam). α-tubulin (sc-23950; Santa Cruz) and GAPDH (GeneTex, GTX100118) were used as internal controls. Signals from HRP-conjugated secondary antibodies were visualized by enhanced chemiluminescence (ECL) detection system (PerkinElmer, Waltham, MA) and chemiluminescence was exposed onto Kodak X-Omat film (Kodak, Chalon/Paris, France). Microscopic examination and immunofluorescence staining {#sec010} ------------------------------------------------------- Optic microscopy with 40 magnification objective lens was used for cell morphology and adhesion assays. Images were analyzed with ImageJ software. Scale bars were indicated in panels. In immunofluorescence staining, SAS cells were transfected with TMEM182-GFP or empty vector plasmids for 36 h, following by gently fixed with 4% paraformaldehyde and permeabilized with phosphate buffered saline (PBS) containing 0.1% Triton X-100. Slides were stained with mouse polyclonal antibody against E-cadherin antibody (1:1000, Cell signaling 3195) at 4°C overnight. 2^nd^ donkey polyclonal antibody conjugated with TexasRed (Santa Cruz, 1:1000) were used for 1 h at room temperature and mounted using mounting solution (ImmunoTech). DAPI (Roche) was used as counterstain for 1 h. Slides were examined by Leica TCS SP5 and analyzed with ImageJ software. Transfection {#sec011} ------------ Transient transfections of 10 nM miR-450a mimics (Life technologies, AM17100) and 10 nM control scramble oligonucleotide (Life technologies) into DOK and SAS cells were performed using Lipofectamine RNAiMAX (Life technologies) according to the manufacturer's instructions. For transfection of the other plasmids, cells were transiently transfected using Lipofectamine 2000 (Invitrogen, CA, USA) according to the manufacturer's protocol. Adhesion and invasion assay {#sec012} --------------------------- 1x10^4^ of DOK cells or 2x10^4^ of SAS cells were seeded into the fibronectin (2 mg/ml; Corning, 356008) or matrigel^™^ (2 mg/ml; BD Biosciences) pre-coated 96-well plate, and incubated at 37°C for 1 h. After rinsed, attached cells were stained with 0.1% crystal violet and evaluated by measuring the absorbance at 595 nm in a Microplate reader (Molecular Devices, CA, USA). Invasion assays were performed as previously described \[[@pone.0213463.ref020]\]. Briefly, the invasion ability was determined using 24-well insert-based assays (BD Biosciences, Franklin Lakes, NJ). The upper insert, with 8 μm pore size, was coated with a density of 40 μg/well of Matrigel Basement Membrane Matrix (BD Biosciences). 2.5 x 104 cells were suspended in medium containing 10% NuSerum, and then added onto the upper insert. After incubating for 24 hours at 37°C, the cells that invaded through the Fluoro-Blok membrane were stained with propidium iodine, and fluorescence images were taken. The invasive cell numbers were then counted with Analytical Imaging Station software package (Imaging Research, Ontario, Canada). Luciferase reporter assay {#sec013} ------------------------- DOK and SAS cells were transfected with100 ng of TMEM182 3′-UTR wild-type (WT) or truncated (DEL) pmirGLO reporter plasmid and transfected with 10 nM of miR-450a mimics or control oligonucleotide (scramble) with Lipofectamine 2000 according to the manufacturer's instructions. The activity of luciferases was determined at 48 h post transfection with Dual Luciferase Reporter Assay System (Promega, USA) as described by the manufacturer's protocol. Luminometry readings were obtained using an Orion L luminometer (Berthold). Statistical analysis {#sec014} -------------------- All quantitative results were from at least three independent experiments and reported as the mean±SEM. A linear correlation and Pearson correlation were used to investigate the association between 2 variables. Differences of various groups were assessed by one-way analysis of variance (ANOVA) and paired Student's t-tests, unless otherwise stated. Recurrent analysis was calculated using Pearson Χ^2^ test. Kaplan-Meier method was using to calculate prognostic values with the log-rank test. All statistic values were carried out using GraphPad Prism V. 4.01 (San Diego, CA). *P*\<0.05 were considered statistically significant and represented as \**P*\<0.05; \*\**P*\<0.01; \*\*\**P*\<0.001. Results {#sec015} ======= miR-450a mediates cellular adhesion and invasion in OSCC {#sec016} -------------------------------------------------------- To examine the expression pattern of miR-450a which screened from OSCC microarray data \[[@pone.0213463.ref010]\], we used quantitative real-time PCR to measure the miR-450a expression in another thirty-five clinical OSCC specimens. We found that the expression levels of miR-450a were significantly higher in 35 OSCC tumors compared with their corresponding normal samples (p\<0.0001, [Fig 1A](#pone.0213463.g001){ref-type="fig"}). Similar results were observed in OSCC cell lines. The expression levels of miR-450a in these OSCC cell lines were significantly higher than the normal human oral keratinocyte HOK ([Fig 1B](#pone.0213463.g001){ref-type="fig"}). To study the potential functions of miR-450a in OSCC, we introduced DOK cells with miR-450a mimics. Specifically, we observed a morphological change, from a rounded shape into a spindle-like shape, in comparison with the scramble control transfectants ([Fig 1C](#pone.0213463.g001){ref-type="fig"}). According to the alteration of cell morphology, we hypothesized that miR-450a might regulate cell adhesion. To test this hypothesis, DOK and SAS cells were subjected to adhesion assays on various components of the extracellular matrix (ECM). Overexpression of miR-450a in DOK and SAS cells showed significant decreases in adhesion ability on fibronectin and matrigel ([Fig 1D and 1E](#pone.0213463.g001){ref-type="fig"}). Furthermore, overexpressed miR-450a increased OSCC cells invasion capacity ([Fig 1F](#pone.0213463.g001){ref-type="fig"}) and had a poor prognosis in HNSCC patients ([Fig 1G](#pone.0213463.g001){ref-type="fig"}). These results indicate that miR-450a may have oncogenic effects in OSCC cells. Augmented miR-450a reduced the cellular adhesion and consequently induced invasion in oral carcinogenesis. {#pone.0213463.g001} TMEM182 is directly targeted by miR-450a {#sec017} ---------------------------------------- In order to determine the downstream target genes of oncogenic miR-450a in OSCC, we performed genome-wide gene expression analysis using miR-450a transfected DOK and SAS cells. Our strategy for collection of miR-450a downstream target genes is presented in [Fig 2A](#pone.0213463.g002){ref-type="fig"}. Compared with control cells, a total of 16878 and 17000 genes were downregulated in miR-450a transfected DOK and SAS cells, respectively. These genes were then analyzed the putative binding sites of miR-450a in their 3'-UTR by microRNA.org database. Through this step, we found 455 genes in DOK group and 498 genes in SAS group with miR-450a binding sites in their 3'-UTR, respectively. Combining the results of these two sets, we identified 256 of common genes which were not only downregulated in miR-450a transfected DOK and SAS cells, but also with miR-450a binding sites. Next, to verify the clinical significance of these genes, we intersected the 256 genes with 40 pairs of OSCC patients which expression data deposited in GEO (accession number GSE37991) by our previously study \[[@pone.0213463.ref010]\]. Finally, we got a set of 12 genes which were downregulated in OSCC tumors compared with their corresponding normal samples ([S3 Table](#pone.0213463.s006){ref-type="supplementary-material"}). However, when we correlated these 12 genes with miR-450a expression, we found that 7 genes were positively correlated with miR450a expression in OSCC tumors. This result doesn't meet our expectation. Only 5 genes showed a negative correlation with miR-450a expression in OSCC patients. Among these 5 genes, TMEM182 is the best negatively correlation with miR-450a ([S3 Table](#pone.0213463.s006){ref-type="supplementary-material"} and [Fig 2B](#pone.0213463.g002){ref-type="fig"}). Therefore, we focused on the TMEM182 as a possible target gene regulated by miR-450a in OSCC and for further study. {#pone.0213463.g002} Overexpression of miR-450a mimics in DOK and SAS cells decreased TMEM182 expression at mRNA and protein levels compared with miR-scramble control in OSCC cells ([Fig 2C](#pone.0213463.g002){ref-type="fig"}). To confirm whether TMEM182 is a direct targeted by miR-450a, we performed a luciferase reporter assays using a vector encoding the partial sequence of the 3'-UTR of TMEM182 either including (WT) or excluding (DEL) the miR-450a binding sites ([Fig 2D](#pone.0213463.g002){ref-type="fig"}). We observed that luciferase intensity was significantly reduced by transfection of wild-type 3'-UTR of TMEM182, but not in cells transfected with the TMEM182 3'-UTR containing the deleted miR-450a binding sites ([Fig 2E](#pone.0213463.g002){ref-type="fig"}). We also analyzed the TMEM182 mRNA levels using the same clinical OSCC specimens in [Fig 1a](#pone.0213463.g001){ref-type="fig"} and found that the TMEM182 expression was lower in OSCC tissues than in their corresponding normal samples ([Fig 2F](#pone.0213463.g002){ref-type="fig"}). To consolidate our findings, we correlated the expression levels of miR-450a and TMEM182 in clinical OSCC specimens and found a strong inverse correlation between the miR-450a and TMEM182 ([Fig 2G](#pone.0213463.g002){ref-type="fig"}). Taken together, our results demonstrated that TMEM182 is a direct target of miR-450a. TMEM182 regulates cell adhesion and invasion in OSCC cells {#sec018} ---------------------------------------------------------- To determine the effect of TMEM182 on cell adhesion, we performed loss-of-function studies by specific shRNAs in DOK and SAS cells. Levels of TMEM182 mRNA were repressed by shTMEM182 transfectants ([Fig 3A](#pone.0213463.g003){ref-type="fig"}). TMEM182 knockdown in DOK and SAS significantly reduced their cell adhesion ability towards to fibronectin and matrigel ([Fig 3B](#pone.0213463.g003){ref-type="fig"}). Whereas, overexpression of TMEM182 in DOK and SAS cells ([Fig 3C](#pone.0213463.g003){ref-type="fig"}) not only promoted the cell adhesion ability ([Fig 3D](#pone.0213463.g003){ref-type="fig"}), but also suppressed the invasion of DOK and SAS cells ([Fig 3E](#pone.0213463.g003){ref-type="fig"}). Our data suggested that TMEM182 may play a role in regulating OSCC invasion and adhesion abilities. {#pone.0213463.g003} MiR-450a enhances cell adhesion throughTMEM182 downregulation {#sec019} ------------------------------------------------------------- Next, we tried to investigate whether miR-450a induced TMEM182 downregulation plays a major role on cell adhesion. For this purpose, we generated a 3'-UTR-lacking TMEM182 vector and transfected it alone or in combination with miR-450a in DOK and SAS cells. As expected, miR-450a decreased endogenous TMEM182 expression and restoration of 3'-UTR-lacking TMEM182 successfully rescued miR-450a-decreased endogenous TMEM182 expression ([Fig 4A](#pone.0213463.g004){ref-type="fig"}). Moreover, the restoration of 3'-UTR-lacking TMEM182 rescued the adhesion ability suppressed by miR-450a but suppressed the invasion ability induced by miR-450a ([Fig 4B and 4C](#pone.0213463.g004){ref-type="fig"}). Our results indicated that miR-450a-mediated TMEM182 function was crucial for regulating attachment ability of OSCC cells. On the basis of functional sequences annotations from NCBI database and Kyte-Doolittle hydrophobicity analysis ([S1 Fig](#pone.0213463.s001){ref-type="supplementary-material"}), there are four predictable hydrophobic regions which aligned in the protein sequences of TMEM182. This result implies that tmem182 may be a membrane protein. To prove that, GFP linked TMEM182 was generated to observe TMEM182 expression ([Fig 4D](#pone.0213463.g004){ref-type="fig"}) and location in cells ([Fig 4E](#pone.0213463.g004){ref-type="fig"} and [S2 Fig](#pone.0213463.s002){ref-type="supplementary-material"}). TMEM182 appeared at lateral membrane zones; particularly at cell-cell contact sites on the plasma membrane ([Fig 4E](#pone.0213463.g004){ref-type="fig"}). Taken together, these data demonstrate that miR-450a-mediated TMEM182 functions are crucial in regulating cell adhesion and invasion in OSCC cells. {#pone.0213463.g004} TNFα triggered-miR-450a attenuates TMEM182 expression in OSCC {#sec020} ------------------------------------------------------------- Previous study has shown that TNF-α could downregulate TMEM182 transcript in 3T3-L1 adipocytes \[[@pone.0213463.ref021]\]. However, little is known about the underlying mechanism between TNF-α and TMEN182 in oral cancer. Thus, we hypothesized that miR-450a may involve in the TNF-α-induced TMEM182 downregulation. To address this, we first tested the ability of TNF-α on TMEM182 expression in OSCC cells. We found that TNF-α treatment indeed significantly reduced the mRNA level of TMEM182 in DOK and SAS cells ([Fig 5A](#pone.0213463.g005){ref-type="fig"}). Simultaneously, we also observed that TNF-α treatment not only induced miR-450a expression ([Fig 5B](#pone.0213463.g005){ref-type="fig"}) but also decreased TMEM182 expression by targeting miR-450a binding site at TMEM182 3\'-UTR ([Fig 5C](#pone.0213463.g005){ref-type="fig"}). However, transfection of TMEM182 3'-UTR containing the deleted miR-450a binding sites (3'UTR-DEL) abolished the TNF-α-induced miR-450a binding to TMEM182 3'-UTR ([Fig 5C](#pone.0213463.g005){ref-type="fig"}), indicated that TMEM182 is directly targeted by miR-450a upon TNF-α treatment. Furthermore, overexpression of TMEM182 reversed the cell adhesion reduced by TNF-α treatment and suppressed the invasion ability induced by TNF-α ([Fig 5D and 5E](#pone.0213463.g005){ref-type="fig"}). These evidences showed that TMEM182 was mediated by miR-450a in response to TNF-α. {#pone.0213463.g005} TNFα induces miR-450a expression via endogenous ERK and NFκB pathways {#sec021} --------------------------------------------------------------------- Some pathways, such as ERK1/2, NF-κB, or p38, are known to be the principle downstream signaling of TNF-α \[[@pone.0213463.ref022]\]. To further delineate the mechanisms of TNF-α-signaling to miR-450a expression, we examined the relevance of ERK1/2, NF-κB, or p38 signaling activated by TNF-α. As shown in [Fig 6A](#pone.0213463.g006){ref-type="fig"}, ERK inhibitor (PD98059) and NF-κB inhibitor (NF-κBi), but not p38 inhibitor (SB20219), significantly blocked the TNF-α-induced miR-450a expression in both DOK and SAS cell lines. Under this condition, ERK inhibitor (PD98059) and NF-κB inhibitor (NF-κBi) can effectively inhibit the ERK1/2 and NFkB activation ([S3A Fig](#pone.0213463.s003){ref-type="supplementary-material"}). Furthermore, ERK inhibitor (PD98059) and NF-κB inhibitor (NF-κBi) not only rescued the TNF-α-induced TMEM182 downexpression ([Fig 6B](#pone.0213463.g006){ref-type="fig"}) but also rescued the TNF-α-induced adhesion reduction ([Fig 6C](#pone.0213463.g006){ref-type="fig"}). Otherwise, we found that the activities of ERK and NFkB were initiated at 10 mins after TNF-α treatment and the activity could sustain to 60 mins. However, the level of miR-450a does not rise significantly until 30 mins ([S3B and S3C Fig](#pone.0213463.s003){ref-type="supplementary-material"}). Therefore, we think that miR-450a expression is an immediate downstream target of ERK/NFkB signaling pathway. Our data suggested that TNF-α-induced miR-450a and reduced cell adhesion by decreasing TMEM182 via intrinsic ERK1/2 and NF-κB pathways ([Fig 6D](#pone.0213463.g006){ref-type="fig"}). {#pone.0213463.g006} Discussion {#sec022} ========== MiR-450a is identified as one of prognostic markers for adrenocortical tumor, aristolochic acid nephropathy and type 2 diabetes \[[@pone.0213463.ref023]--[@pone.0213463.ref025]\]. Emerging data indicated that miR-450a controls a divergent function on the osteoblastic differentiation of human mesenchymal stem cells by targeting STAT1 and rats' adipogenesis by targeting WISP2 \[[@pone.0213463.ref026], [@pone.0213463.ref027]\]. However, there are no data to suggest a connection between miR-450a and oral cancer in recent investigations. In this study, gain-of function approach indicated that miR-450a restoration significantly inhibited cell adhesion and enhanced invasion ability in DOK and SAS cells, suggesting miR-450a is an onco-miR and play an important role in OSCC cellular adhesion on matrix. For elucidation the molecular mechanisms and identification the putative targets of miR-450a in OSCC, we performed genome-wide gene expression analysis using miR-450a transfected DOK and SAS cells. In silico analysis the expression signatures of miR-450a transfectants in OSCC cells and OSCC patients, we found that twelve genes were downregulation in clinical specimens and that they were candidate targets of miR-450a. Among these genes, transmembrane protein TMEM182 showed a best inverse correlation with miR-450a. Downregulating TMEM182 using shRNA suppressed cell adhesion in a manner comparable to miR-450a overexpression. Moreover, the restoration of TMEM182 potently rescued the adhesion ability suppressed by miR-450a, suggesting that miR-450a mainly acts as a novel regulator in mediating OSCC adhesion by targeting membrane protein TMEM182. *TMEM182* gene encodes an entirely 229-amino-acid protein, which is predicted to consist of four putative membrane-spanning regions ([S1 Fig](#pone.0213463.s001){ref-type="supplementary-material"}). It is highly evolutionary conserved among different species \[[@pone.0213463.ref021]\]. Even TMEM182 plays important roles in adipogenesis, myogenesis, and glaucoma \[[@pone.0213463.ref018], [@pone.0213463.ref021], [@pone.0213463.ref028]\], however, its working mechanisms were still unknown. Dissolution of junctional connection, detachment to ECM, and migration are key steps of OSCC loco-regional invasion \[[@pone.0213463.ref029]--[@pone.0213463.ref031]\]. Our findings demonstrated that overexpression of TMEM182 increased OSCC adhesive ability and restrained its invasiveness. Moreover, restoration of TMEM182 completely rescued the cellular attachments suppressed by miR-450a in vitro. Thus, decreased cell-matrix adhesion might enhanced the cellular contraction and thereby facilitate tumor migration and invasion. On the other hand, disassembly of cell-cell interaction is occurrence at the early stage of OSCC invasion \[[@pone.0213463.ref030]\]. Cell adhesion molecules, such as integrin, cadherin family, and immunoglobulin superfamily, play a role in cell-cell interactions and involved in the process of tumor invasion and metastases \[[@pone.0213463.ref030], [@pone.0213463.ref032]--[@pone.0213463.ref034]\]. Loss of these cell adhesion molecules is associated closely with invasion and could be used for the prognostic purposes in oral cancer \[[@pone.0213463.ref035]--[@pone.0213463.ref038]\]. In this study, immunofluorescence data demonstrates that TMEM182 appeared at lateral membrane zones; particularly at cell-cell contact sites on the cell membrane. These results suggest that TMEM182 may play a role in cell-cell interaction and cell-extracellular matrix adhesion as a result of involving in the process of tumor invasion. However, little studies have described the functions of TMEM182 or their relationship between cell-cell interaction and cell-extracellular matrix adhesion. The detail mechanisms remain to be elucidated. Recent studies present that inflammatory factors, including TNF-α, are potential prognostic biomarkers for OSCC \[[@pone.0213463.ref039], [@pone.0213463.ref040]\]. Our findings supported that TNF-α activated endogenous ERK1/2 and NF-κB pathways to induce miR-450a expression. Current studies reveal that TNF-α induced EMT to promote OSCC invasion through NF-κB pathway by targeting at well-known Snail and Id2 \[[@pone.0213463.ref041]--[@pone.0213463.ref043]\]. It is worth noting that the miR-450a expression induced by TNF-α primarily through ERK1/2 activation rather than through NF-kB pathway. As a consequence, TMEM182 was downregulated by miR-450a to increase OSCC cells invasion. To our knowledge, this study is the first to describe the roles of ERK1/2 and NF-κB in TNF-α-induced miR-450a expression in human OSCC. Upregulation of miR-450a could reduce cellular adhesion to matrix by targeting TMEM182 and enhance tumor invasion. Therefore, TNF-α/miR-450a/TMEM182 signaling axis may be a novel potential target for clinical intervention in oral cancer. Supporting information {#sec023} ====================== ###### Hydrophobic region prediction of TMEM182. Protein length of human TMEM182 precursor (229a.a.) was analyzed by ExPasy database. Total four transmembrane regions were indicated by Arabic numbers in order and Alphabet letter S. S indicated as a potential signal sequence. (PDF) ###### Click here for additional data file. ###### Fluorescence of TMEM182 and E-cadherin in OSCC cells. Epifluorescence images of SAS cells transiently transfected with TMEM182-GFP/ Vehicle and co-immunolabeled endogenous E-cadherin with divided channels. **(A)** Vehicle (Green) expression was spread all over the cells. **(B, E)** Membrane marker Ecadherin (Red) was localized at intracellular junctional areas. **(C, F)** Nuclei were labeled with DAPI (Blue). **(D)** TMEM182-driven GFP (green) located at cell-cell contact sites on the lateral membrane and endoplasmic reticulum. Scale bars were indicated in panel. (PDF) ###### Click here for additional data file. ###### Effects of ERK and NFκB pathways on TNF-α. **(A)** OSCC cells were pre-incubated with either DMSO vehicle (-), ERK inhibitor (ERKi, 30 μM), NF-κB inhibitor (NF-κBi, 10 μM) for 6 h and then treated with 10 ng/ml of TNF-α for another 24 hrs, followed by measurements of TMEM182 expression by western blotting. GAPDH was used as an internal control. **(B)** Western blotting analysis of ERK and NFκB activity after TNF-α treatment in SAS cells at indicated time. GAPDH was used as an internal control. (C) miR-450a expression level in SAS cells treated with TNF-α using qRT-PCR and normalized to RNU44. Results were represented as mean±SEM;\*\**P*\<0.01, \*\*\**P*\<0.001. (PDF) ###### Click here for additional data file. ###### Patients' cinicalpathological analysis. (DOC) ###### Click here for additional data file. ###### Sequences of primers. (DOCX) ###### Click here for additional data file. ###### Twelve of candidate genes downregulated by miR-450a. (DOC) ###### Click here for additional data file. En-Wei Hsing carried out his thesis research under the auspices of the Structural Biology Program, Institute of Bioinformatics and Structural Biology, National Tsing Hua University, Hsinchu, Taiwan. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Background {#Sec1} ========== The African pygmy hedgehog (*Atelerix albiventris*) has become a very popular pet in Poland in the last few years. It is smaller than the European hedgehog and is a member of the insectivore family *Erinaceidae*, subfamily *Erinaceinae*. African pygmy hedgehogs are domesticated animals and live for about 5--7 years in captivity. They possess 36 brachyodontic teeth: 2(3/2,1/1,3/2,3/3) with the first incisors being notably longer than the rest, and are spaced apart \[[@CR1]\]. In studies of hedgehogs at histopathology of surgically resected tumor or necropsy approximately 40 % of hedgehogs aged from 1 month to 3 years were diagnosed with neoplastic disease \[[@CR2], [@CR3]\]. The most common histologic types of tumors are mammary gland adenocarcinoma, lymphoma, and oral squamous cell carcinoma \[[@CR2], [@CR4], [@CR5]\]. The digestive tract, including the oral cavity, is the third most common site of neoplastic disease in hedgehogs \[[@CR4]\]. Chemotherapy and radiotherapy protocols have not yet been established for the African pygmy hedgehog, and therefore, surgical resection is currently the best treatment in cases where the tumor is benign, well separated from the healthy tissue and without metastases. Medical knowledge, veterinary care, and the awareness of African pygmy hedgehog owners are ever increasing. The average life span of domesticated animals is prolonged compared with wild animals. This situation predisposes domesticated hedgehogs to more frequent development of tumors, including oral cavity. In addition, it is well known that periodontal disease, tooth root abscesses, and various neoplasms (e.g. squamous cell carcinoma, lymphosarcoma) occur frequently in African pygmy hedgehogs \>3 years old \[[@CR4]\]. Peripheral odontogenic fibroma (previously named as fibromatous epulis of periodontal ligament origin) is a peripheral odontogenic neoplasm, indistinguishable clinically from fibrous hyperplasia, most common in dogs, and rarely occurring in cats. The prognosis following surgical removal is good \[[@CR6], [@CR7]\]. To our knowledge, this is the first case report of surgical resection of a peripheral odontogenic fibroma in the African pygmy hedgehog. The significance of this case report is that it will enable veterinary clinicians to familiarize themselves with the surgical resection of benign oral tumors (peripheral odontogenic fibroma) in the African pygmy hedgehog and consider the peripheral odontogenic fibroma as other primary neoplasm of oral cavity in this species. Case presentation {#Sec2} ================= A 5-year-old male African pygmy hedgehog showed two erythematous, round small tumors protruding from the oral cavity. One tumor was visible on the closed mouth on the right side of the head (Figs. [1](#Fig1){ref-type="fig"} and [2](#Fig2){ref-type="fig"}), and the other was located on the left side, inside the oral cavity, above the molar teeth (Fig. [3](#Fig3){ref-type="fig"}). Both tumors were well separated from the gum, pedunculated, soft texture, uneven surfaces, painless. The neoplastic lesion on the left side had a diameter of approximately 5 mm, was pale pink, while the lesion on the right side had a diameter of approximately 11 mm and on its surface small foci of hyperemia were observed. There was no deformation of hard tissue of splanchnocranium during clinical examination. The pet owner complained of problems with food and water intake, because of the neoplastic tumors located on the surface of maxilla. The animal's appetite remained good. Physical examination revealed the animal was in very good condition with a rectal temperature of 36.8 °C.Fig. 1Preoperative view of oral tumors located above the molar teeth, on both sides of the maxilla in an African pygmy hedgehogFig. 2Preoperative view of oral tumors located above the molar teeth, on both sides of the maxilla in an African pygmy hedgehogFig. 3Preoperative view of oral tumors located above the molar teeth, on both sides of the maxilla in an African pygmy hedgehog Anesthesia was induced with 30 mg/kg of ketamine hydrochloride (Bioketan; Vetoquinol Biowet, Gorzow Wielkopolski, Poland) and 0.15 mg/kg of medetomidine hydrochloride (Domitor; Zoetis, Florham Park, NJ, USA) intramuscularly (IM). To prevent hypersalivation, atropine sulfate (Atropinum Sulfuricum; Polfa, Warszawa, Poland) at 0.03 mg/kg IM was administered. The preferred method for sedating small animals is gas anesthesia with isoflurane or sevoflurane \[[@CR8]\], but in this case using a facial mask was impossible because of the location of the tumors. The animal was placed on a heating mat to prevent hypothermia. A portable veterinary monitor (MEC-1200-Vet; Mindray, Shenzhen, China) was used to constantly monitor the patient's breathing rate and heart rate. Surgical resection of tumors and bleeding were controlled simultaneously using a surgical knife electrocoagulation system (Martin System 2000; Gebrüder Martin GmbH & Co, Tuttlingen, Germany). The results of surgical resection of the oral tumors are shown in Figures (Figs. [4](#Fig4){ref-type="fig"} and [5](#Fig5){ref-type="fig"}).Fig. 4Postoperative view of the excised oral tumors using surgical knife electrocoagulationFig. 5Postoperative view of the excised oral tumors using surgical knife electrocoagulation After surgical procedures, atipamezole hydrochloride (Antisedan; Zoetis) 0.3 mg/kg IM and meloxicam (Metacam; Boehringer Ingelheim, Ingelheim, Germany) 0.2 mg/kg subcutaneously were administered \[[@CR9]\]. The patient's condition was monitored until it reached full consciousness. The excised tumors underwent fixation in 7 % buffered formalin, were embedded in paraffin blocks, and 6 μm slides were cut. The preparations were stained using the standard hematoxylin-eosin method \[[@CR10]\], and subsequently evaluated using light microscopy using WHO guidelines for the evaluation of oral cavity tumors. Photomicrographs of the preparations were obtained using computer-amplified image analysis and an optical microscope (Olympus BX53; Olympus, Tokio, Japan). Histopathological analysis was conducted using the cell^\^^A software (Olympus Soft Imaging Solution GmbH, Münster, Germany). Results and discussion {#Sec3} ====================== Histopathological examination of lesions showed epithelial covered, well-vascularized, cellular fibroblastic tissue comprised of small spindle to stellate fibroblasts with small dark basophilic nuclei dispersed in a dense collagen matrix. Few mononuclear inflammatory cells, areas of hard tissue corresponding with areas of mineralization, and branching cords or islands of epithelium with peripheral basal stratum were also observed (Fig. [6a](#Fig6){ref-type="fig"}, [b](#Fig6){ref-type="fig"}, and [c](#Fig6){ref-type="fig"}). All tumors were removed with an appropriate (approximately 2--5 mm) clinical surgical margin and evaluated later by histopathology. The observed pattern is characteristic for peripheral odontogenic fibroma \[[@CR6], [@CR7]\]. Peripheral odontogenic fibromas have been extensively reported in a variety of domestic mammals and humans \[[@CR11]--[@CR16]\]. However, to our knowledge, there is no information concerning peripheral odontogenic fibroma in the African pygmy hedgehog.Fig. 6Histopathological pattern of peripheral odontogenic fibroma in the pygmy hedgehog. **a** Small spindle to stellate fibroblasts immersed in eosinophilic dense collagen matrix with branching cords and islands of odontogenic epithelium and peripheral palisading of same epithelium. **b** A small amount of mononuclear inflammatory cells accompanying the peripheral odontogenic fibroma. **c** Areas of hard tissue corresponding with areas of mineralization within the tumor According to the literature, the most common tumors of the gastrointestinal tract in hedgehogs are oral squamous cell carcinoma, intestinal adenocarcinoma, acinic cell carcinoma, metastatic hepatocellular carcinoma, fibrosarcoma, plasmocytoma, and lymphoma \[[@CR4], [@CR17]--[@CR19]\]. Raymond and Garner \[[@CR4]\] diagnosed 35 (53 %) neoplasms in a group of 66 hedgehogs. In three of the 35 animals, more than one type of tumor was present. The authors revealed that 85 % of the tumors were malignant \[[@CR4]\]. Raymond et al. \[[@CR20]\] indicated the presence of malignant mast cell tumor in adult African hedgehog. This tumor was located subcutaneously, along to ventral part of the neck with metastasis to local lymph node \[[@CR20]\]. In other studies, the authors revealed the evidence of a probable retrovirus associated with multicentric sarcomas in two 3 years old hedgehogs, male and female \[[@CR21]\]. In our case, the investigated tumor was benign. Nonetheless, its location in the oral vestibule, which caused the animal's discomfort, pain, and risk of hemorrhage, was an indication for surgery. Conclusions {#Sec4} =========== An early and accurate diagnosis is essential for positive prognosis, curative treatment, and fast recovery in hedgehogs. The resection of oral cavity tumors in the African pygmy hedgehog carried out in this case report can be successfully applied by veterinary clinicians. The established protocol is safe for the patient and provides the best solution for mild proliferative lesions of the oral cavity. To our knowledge this is the first report of surgical resection and histological description of oral tumors (peripheral odontogenic fibroma) in the African pygmy hedgehog. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** AW-B carried out anesthesia during surgery, took postoperative care of the animal, and drafted the manuscript. MJ carried out surgical resection of the oral tumors and helped draft the manuscript. IJ participated in the preparation and consolidation of histopathological samples and helped draft the manuscript. MN carried out pathomorphological studies. All authors read and approved the final manuscript. Publication supported by Wrocław Centre of Biotechnology, programme the Leading National Research Centre (KNOW) for years 2014-2018.

**Core tip:** Analytical performance of the serum creatinine assays is the critical determinant of estimated glomerular filtration rate (eGFR) accuracy. The most widely used compensated Jaffé creatinine assay suffers from a non-specific bias from pseudo-creatinine chromogens (glucose, ketones), which is not the case with the costly enzymatic assays. We evaluated the influence of creatinine methodology on the performance of chronic kidney disease (CKD)-Epidemiology-calculated eGFR for CKD diagnosis/staging in diabetic patients. Our results indicate that compensated Jaffé creatinine procedure, in spite of the glucose-dependent bias, is not inferior to enzymatic creatinine in CKD diagnosis/staging and therefore may provide a reliable and cost-effective tool for the renal function assessment in diabetic patients. INTRODUCTION ============ Global prevalence of diabetes mellitus is rising progressively\[[@B1]\]. Chronic morbidity, associated with various debilitating complications, increased risk for adverse health-outcomes and significant impact regarding both the working ability and quality of life identify diabetes as one of the greatest health-care and socio-economic challenges worldwide. Appropriate strategies to tackle diabetes epidemic include education and lifestyle interventions, evidence-based clinical management as well as the screening for and monitoring of diabetes and/or diabetes-related complications using state-of-the art diagnostic tools. Diabetic kidney disease (DKD) is one of the most prevalent chronic complications of diabetes and the most common single cause of end-stage renal failure\[[@B1],[@B2]\]. It has been amply evidenced that appropriate interventions at an early stage of DKD can be efficient in preventing and/or delaying the progression of kidney disease and improving patient outcomes. Thus, the regular screening for DKD has became one of the cornerstones of diabetes care. Current clinical guidelines recommend at least an annual screening of DKD in patients with type 1 diabetes with a duration above 5 years, in all patients with type 2 diabetes and in all hypertensive diabetic patients\[[@B3]\]. Once detected, DKD is treated according to clinical guidelines and further monitored at regular intervals\[[@B2],[@B3]\]. Two simple laboratory tests are used for both the screening and staging of CKD in diabetes: Urinary albumin excretion (UAE) and serum creatinine-based estimated glomerular filtration rate (SCr-eGFR). Abnormal UAE has long been identified as a sensitive marker of the glomerular basal membrane damage, which is one of the early pathophysiological events in the development of DKD\[[@B4]\]. However, a significant decline in eGFR is a common finding in a notable proportion of diabetic patients with normal UAE, probably reflecting a diversity in the natural history of DKD\[[@B5]\]. Thus, the pathophysiology of DKD has shifted from the "albuminuric paradigm"\[[@B6]\], and the accumulated evidence implicating the progressive renal function decline as an equally relevant pathway identified reliable and accurate laboratory testing for serum creatinine and SCr-eGFR as a very important issue for the diagnosis, staging and monitoring of CKD in diabetic patients. SCr has been used as a cost-effective and practical marker of kidney function for decades, despite severe limitations due to both biological and analytical variability\[[@B7]\]. A handful of biological factors such as age, gender, ethnicity and nutritional habits substantially influence serum creatinine levels, while partial tubular reabsorption and secretion of creatinine further compromise its use as the glomerular filtration marker\[[@B8],[@B9]\]. Nevertheless, SCr-based estimation of GFR by the use of appropriate predictive equation remains the recommended surrogate marker for the assessment of kidney function, since the actual measurement of GFR, due to its complexity and high costs, is not available outside the specialized clinical settings. Current guidelines from the Kidney Disease Improving Global Guidelines (KDIGO) CKD Working Group recommend the use of the chronic kidney disease-Epidemiology Collaboration Group (CKD-EPI) equation\[[@B2]\]. CKD-EPI equation offers an improved reproducibility and accuracy at higher GFR levels (\> 60 mL/min per 1.73 m^2^), which is the most prominent disadvantage of the previously recommended Modification of Diet in Renal Disease equation\[[@B10]\]. Analytical performance and specificity of SCr assay are critical determinants of the eGFR accuracy\[[@B11]\]. The relationship between SCr and GFR is exponential, therefore, errors in SCr measurements resulting from imprecision and bias could strongly impact eGFR results and result in misclassification of the patients regarding their kidney function\[[@B2]\]. Despite standardization and harmonization by the calibration traceable to isotope-dilution-mass spectrometry (IDMS), the non-specific bias from pseudo-creatinine chromogens (glucose, proteins, ketone bodies) is still affecting the most widely used compensated Jaffé alkaline picrate colorimetric creatinine assay\[[@B11],[@B12]\]. Enzymatic creatinine methods are free from these interferences, but far more expensive and therefore not widely used. High-volume routine enzymatic creatinine testing may introduce a substantial financial challenge for the laboratories, even in the otherwise fairly resourced health-care systems\[[@B13]\]. Several analytical and clinical studies advocated the replacement of the compensated Jaffé with enzymatic creatinine assays in order to improve reliability of the eGFR, especially in the diabetic population, which is expected to have an increased amount of interfering substances in serum. However, recently published risk-analysis study, using both analytical and biological variability criteria, revealed a low risk for misclassification of CKD based on Jaffé-SCr-eGFR results in the general population\[[@B13]\], while the clinical impact in diabetic population remains unclear. The aim of this study was to evaluate the influence of creatinine methodology on the performance of CKD-EPI-calculated eGFR for CKD evaluation and staging in a large cohort of diabetic patients. MATERIALS AND METHODS ===================== Fasting blood samples were taken from diabetic patients attending our clinic for their regular annual examination, including laboratory measurement of SCr and eGFR. Samples from the patients with concomitant infection, limb-amputation and malignancies, as well as the pregnant patients and the patients with severe kidney disease (stage 5, according to KDIGO-2012 classification) were not included in the study. A subset of samples of the patients with severe hyperglycaemia were included in order to evaluate the interference of glucose on the CKD classification across various eGFR categories. Serum creatinine was measured by both IDMS-traceable compensated Jaffé (cJ-SCr) and enzymatic (e-SCr) (Beckman Coulter, Inc., Pasadena, California, United States) procedures with intra-assay imprecision (CV) of 1.58% and 1.39%, respectively. Hexokinase (Beckman Coulter, Inc., Pasadena, California, United States) and NGSP-traceable immunoturbidimetric assays (Tina Quant, Roche, F.Hoffmann-La Roche, Basle, Switzerland) were used for plasma glucose and HbA~1c~ measurement. Assay-specific SCr-eGFR was estimated by the 4-variable CKD-EPI equation using respective creatinine values\[[@B10]\]. UAE was measured by an automated immunoturbidimetric procedure (Beckman Coulter, Inc., Pasadena, California, United States) in fresh spot urine samples. Urinary creatinine was measured in the same samples and UAE results expressed as the urinary albumin/creatinine ratio. Staging of albuminuria and CKD, as well as risk assessment for CKD progression was carried out according to KDIGO-2012 criteria (Table [1](#T1){ref-type="table"}). ###### Kidney Disease Improving Global Guidelines-2012 Prognostic Categories of Chronic Kidney Disease according to estimated glomerular filtration rate and albuminuria (adapted from the Reference 2) **Albuminuria categories \[albumin/creatinine (mg/mmol)\]** ---------------------------------------- ------------------------------------------------------------- --------------------------- --------------------------- ----------- eGFR categories (mL/min per 1.73 m^2^) G1, ≥ 90 Low risk Moderately increased risk High risk Normal/high G2, 60-90 Low risk Moderately increased risk High risk Mildly decreased G3a, 45-59 Moderately increased risk High risk Very high risk Mildly to moderately decreased G3b, 30-44 High risk Very high risk Very high risk Moderately to severely decreased G4, 15-29 Very high risk Very high risk Very high risk Severely decreased G5, \< 15 Very high risk Very high risk Very high risk Kidney failure eGFR: Estimated glomerular filtration rate. The results were analyzed in the entire population and in sub-groups according to albuminuria (Table [2](#T2){ref-type="table"}). Normality of distribution was tested by the Kolmogorov-Smirnov test and the significance of differences between the groups was assessed by the Kruskal-Wallis and Mann-Whitney test, as appropriate. Comparison between the creatinine methods in the study population was tested by Passing-Bablok regression analysis. Specific SCr-eGFR data were compared by Bland Altman analysis, and their agreement regarding clinical CKD staging was evaluated by inter-rater agreement (kappa-analysis). Statistical analyses were performed using MedCalc for Windows, version 9.4.2.0 (MedCalc Software, Ostend, Belgium). *P* \< 0.05 was defined as the threshold of significance. ###### Clinical characteristics of the study subjects across the Kidney Disease Improving Global Guidelines-2012 categories of albuminuria **Category of Albuminuria** ------------------------------------ ----------------------------------------------------------------------------- ----------------------------------------------------------------------------- -------------------------------------------------------------------------------- N (M/F) 372 (212/198) 166 (87/79) 72 (38/34) Age (yr) 63 (19-88) 68 (18-88) 60[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (29-85) Glucose (mmol/L) 9.0 (8.8-9.3) 9.4 (8.9-10.0) 7.4[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (5.7-8.9) HbA~1c~ (%) 7.4 (7.3-7.6) 7.5 (7.3-7.7) 7.8 (7.4-8.5) HbA~1c~ (mmol/mol) 59 (57-61) 59 (57-61) 62 (58-71) e-SCr (μmol/L) 69[b](#T2FN2){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (66-72) 75[a](#T2FN1){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (72-77) 100[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (88-137) cJ-SCr (μmol/mol) 70[b](#T2FN2){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (67-72) 77[a](#T2FN1){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (72-81) 108[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (87-140) e-SCr-eGFR (mL/min per 1.73 m^2^) 91[b](#T2FN2){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (88-93) 85[a](#T2FN1){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (80-88) 60[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (39-77) cJ-SCr-eGFR (mL/min per 1.73 m^2^) 90[b](#T2FN2){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (87-92) 83[a](#T2FN1){ref-type="table-fn"},[d](#T2FN3){ref-type="table-fn"} (76-86) 55[a](#T2FN1){ref-type="table-fn"},[b](#T2FN2){ref-type="table-fn"} (39-73) Age is expressed as median (range) and other variables as median (95%CI of median). *P* \< 0.001 *vs* A1; *P* \< 0.001 *vs* A2; *P* \< 0.001 *vs* A3. HbA~1c~: Glycosylated hemoglobin; SCr: Serum creatinine; eGFR: Estimated glomerular filtration rate; e: Enzymatic; cJ: Compensated Jaffé; M: Male; F: Female. The study was approved by the Merkur University Hospital Ethics Committee. Due to the retrospective nature of the study, with the *post-hoc* selection of anonymized samples from the routine laboratory visits, patient's informed consent was not obtained. RESULTS ======= A total of 648 Caucasian diabetic patients (337 males) was included in this study. No gender-related differences were observed in the clinical and biochemical parameters, except significantly lower creatinine levels in females (*P* \< 0.001, data not shown). There was a significant increase of SCr and a decrease of eGFR, as measured/estimated by both methods across the categories of albuminuria (Table [2](#T2){ref-type="table"}). Fasting plasma glucose was significantly lower in the A3 subgroup only, while HbA~1c~ levels showed no differences regarding albuminuria (Table [2](#T2){ref-type="table"}). eGFR and creatinine results did not differ significantly depending on creatinine methodology in either category of albuminuria (*P* = 0.228, 0.2306 and 0.7553 for A1, A2 and A3 category, respectively; Mann-Whitney test). Passing-Bablok regression analysis revealed a small, but significant constant difference between the enzymatic and compensated Jaffé SCr assays \[y = -2.8095 (95%CI: -3.8125 to -1.6066) + 1.0476 × (95%CI: 1.0328-1.0625)\] across a wide range of creatinine values (Figure [1](#F1){ref-type="fig"}). This was accompanied by a minor, but significant creatinine assay-dependent difference in SCr-eGFR values \[Bland Altman: y = 1.5154 (95%CI: 1.1635-1.8674; lower limit: -7.4276; upper limit: 10.4585) *P* \< 0.001\] (Figure [2](#F2){ref-type="fig"}). The severity of both acute and chronic hyperglycaemia was identified as the significant predictor of between-method SCr-eGFR bias (Spearman\'s rho = -0.363 and -0.369 for fasting plasma glucose and HbA~1c~, respectively, *P* \< 0.001). {#F1} {#F2} Inter-rater agreement analysis showed an excellent agreement (weighted kappa = 0.918; 95%CI: 0.894-0.94) between the method-specific SCr-eGFRs when classifying subjects into KDIGO-2012 CKD-stages. However, some cases were classified differently between CKD stages depending on the creatinine method used for eGFR calculation (Table [3](#T3){ref-type="table"}). Compared to e-SCr-eGFR-based CKD classification, 58/648 (9%) patients were re-classified into a different CKD stage when cJ-SCr-based eGFR was used. The majority of these (54/648; 8%) were re-classified into a more advanced stage of CKD (positive discordance), with 23 (3.5%) cases re-classified into the clinically significant eGFR category indicating mildly to moderately decreased kidney function (\< 60 mL/min per 1.72 m^2^). Among these, 7 cases (1%) had A1 stage of albuminuria, whereas the rest of clinically significant positive discordant cases had more advanced stages of albuminuria. On the other hand, 8/648 (1%) of patients were re-classified into a less-advanced CKD stage when compensated Jaffé-SCr-eGFR was used (negative discordance), with only 2 cases being re-classified between the 3A and 2 eGFR categories. A1 and A2 stage of albuminuria was detected in each one of these two cases. ###### Reclassification of the estimated glomerular filtration rate based chronic kidney disease stage according to enzymatic and compensated Jaffé creatinine values **e-SCr-eGFR-based CKD stage** ----------------------------- ------------------------------------ ------------------------------------- ------------------------------------ ------------------------------------ ------------------------------------ ----------------------------------- ------------ cJ-SCr-eGFR-based CKD stage 1 272[1](#T3FN1){ref-type="table-fn"} 5[2](#T3FN2){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 277 (42.7) 2 25[3](#T3FN3){ref-type="table-fn"} 206[1](#T3FN1){ref-type="table-fn"} 2[2](#T3FN2){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 233 (36) 3a 0[1](#T3FN1){ref-type="table-fn"} 23[3](#T3FN3){ref-type="table-fn"} 54[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 77 (11.9) 3b 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 5[3](#T3FN3){ref-type="table-fn"} 28[1](#T3FN1){ref-type="table-fn"} 1[2](#T3FN2){ref-type="table-fn"} 34 (5.2) 4 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 0[1](#T3FN1){ref-type="table-fn"} 1[3](#T3FN3){ref-type="table-fn"} 26[1](#T3FN1){ref-type="table-fn"} 27 (4.2) Total 297 (45.8) 234 (36.1) 61 (9.4) 29 (4.5) 27 (4.2) 648 Concordant; Negatively discordant; Positively discordant cases. CKD: Chronic kidney disease; cJ-SCr-eGFR: Compensated Jaffé serum creatinine-based estimated glomerular filtration rate. There was a significant difference in fasting plasma glucose values regarding concordance of CKD staging, with higher glucose values for positive- and lower glucose values for negative discordant subjects, in comparison to concordant sub-group (11.2 ± 4.3 *vs* 7.5 ± 1.8 *vs* 8.9± 2.1 mmol/L, *P* \< 0.001). HbA~1c~, indicating a chronic level of hyperglycaemia, showed an identical pattern (8.4 ± 2.3% ± /69 ± 25 mmol/mol *vs* 6.6 ± 0.7%/49 ± 7.3 mmol/mol *vs* 7.8 ± 1.7%/62 ± 19 mmol/mol; *P* \< 0.01). We analyzed the frequency of discordances according to the level of hyperglycaemia, by using the fasting plasma glucose cut-off of 17.0 mmol/L, which was reported to significantly influence SCr results obtained by the colorimetric Jaffé procedure\[[@B7]\]. Positively discordant results were more prevalent in the sub-group of patients with fasting plasma glucose above (*n* = 59), than below (*n* = 589), 17.0 mmol/L glucose cut-off (20% *vs* 8%, *χ*^2^ = 11.968, *P* = 0.0025). However, in general, patients with eGFR \< 60 mL/min per 1.73 m^2^ had lower fasting plasma glucose than those with eGFR \> 60 mL/min per 1.73 m^2^ (7.3 ± 1.9 mmol/L *vs* 9.2 ± 2.0 mmol/L, *P* \< 0.0001). DISCUSSION ========== In this study, we attempted to evaluate the influence of creatinine methodology on the performance of CKD-EPI-calculated eGFR for CKD staging in a large cohort of diabetic patients. Our results indicate an overall excellent agreement in CKD staging (kappa = 0.918) between the Jaffé serum creatinine- and enzymatic serum creatinine-based CKD-EPI-eGFR, with 9% of discordant cases. As compared to the enzymatic creatinine, the majority of discordances (8%) were positive, *i.e*., associated with the more advanced CKD stage re-classification, whereas only 1% of cases were negatively discordant if Jaffé creatinine was used for eGFR calculation. Plasma glucose was identified as a significant determinant of between-method bias. Kidney function is rather uniquely affected by diabetes. Elevated GFR, known as hyperfiltration, is a common finding in new-onset diabetes, probably as a consequence of hyperglycaemia and related metabolic and endocrine disturbances. Hyperfiltration, being considered as an early sign of DKD\[[@B14]\], is declining with the progression of diabetes and the intensive diabetes treatment was found to be effective in reducing the risk for the progression to DKD by delaying the GFR decline in both type 1 and type 2 diabetes\[[@B15],[@B16]\]. Thus, specific features of DKD in diabetes indicate the need for an accurate and reliable method for GFR estimation over the wide range of GFR. Our previous study showed that CKD-EPI-eGFR, with improved accuracy in the GFR range above 60 mL/min per 1.73 m^2^, represented a superior surrogate marker of GFR in diabetic patients, particularly those with normoalbuminuria and hyperfiltration implicating its use as a reliable screening tool for an early renal impairment in diabetes\[[@B17]\]. Either compensated Jaffé or enzymatic creatinine assay, traceable to the reference IDMS procedure, is needed for eGFR-CKD-EPI calculation\[[@B2]\]. Analytical interference of the glucose and other reducing substances in the alkaline picrate Jaffé creatinine assay has long been identified\[[@B7]\]. Several method improvements, including modified spectral kinetics and standardization to the IDMS reference procedure with a mathematical adjustment of results to compensate for interferences (compensated Jaffé assays), have remarkably improved the accuracy of the method. Nevertheless, the non-specificity remained a matter of concern in selected patient subpopulations, such as subjects with diabetes\[[@B8]\]. Enzymatic creatinine assays offer improved specificity and several authors argued that Jaffé method should be entirely abandoned, particularly in the diabetic population. It was reported that CKD-EPI-eGFR showed better concordance to the measured GFR, empowering further the enzymatic method as a method of choice for serum creatinine measurement in diabetic patients\[[@B15]\]. However, evidence supporting this proposition is based either on cross-sectional method-comparison studies including a limited number of patients, or simulation studies using analytical bias extracted from inter-laboratory comparisons, with no data regarding the clinical outcome-associated risk\[[@B18]-[@B21]\]. Our results demonstrate a minor, but significant glucose-dependent positive bias between the serum creatinine levels measured by compensated Jaffé and enzymatic procedure, with a mirroring effect regarding respective eGFRs, but the key question of this study was the clinical relevance of the observed difference. State-of-the-art strategies for laboratory test evaluation implicate not only analytical, but also clinical performance together with clinical- and cost-effectiveness as essential interactive components of the overall diagnostic test utility assessment\[[@B22]\]. In a recently published outcome-based study, Schmidt et al\[[@B13]\] reported on a very low risk for patient outcomes due to the miss-classification of CKD stages with Jaffé creatinine assay in the general population. Our study reveals that most of the discordant cases in diabetic subjects were positive, *i.e*., Jaffé method was likely to classify 8% of the patients into a more advanced CKD stage than the enzymatic method. Among these, 23/648 cases were classified into the clinically significant stage 3A, while only 7 positively discordant cases with normal UAE (1%) were re-classified between the low4 and moderately increased risk for the CKD progression, according to KDIGO guidelines (Table [1](#T1){ref-type="table"}). The frequency of positively discordant cases was 2.5 times greater in the sub-group of patients with plasma glucose above the 17.7 mmol/L, previously reported as a threshold for a significant analytical interference\[[@B7]\]. It is important to emphasize that both fasting plasma glucose and HbA~1c~ in positively discordant sub-group was far above the glycaemic recommendations for adults with diabetes, requiring immediate interventions to reduce the hyperglycaemia regardless of the kidney function\[[@B3]\], which is known to be afflicted by the renal hypoperfusion in acute hyperglycaemic episodes\[[@B23],[@B24]\]. Apart from assessing frequency, our main goal was to evaluate the clinical consequences of the discordant Jaffé creatinine-dependent CKD staging. In general, clinical guidelines recommend the optimization of glycaemic control and blood pressure as the treatment strategy to prevent and/or delay the development/progression of CKD in diabetic patients\[[@B3]\]. Patients with stage 3a CKD (eGFR range 45-60 mL/min per 1.73 m^2^) should be referred to a nephrologist for further evaluation, their diet and medication adjusted and eGFR monitoring intensified to twice a year. The more advanced stages of CKD require further specialized care and intensified monitoring, and metformin should be discontinued in patients with stage 3b CKD (30-45 mL/min per 1.73 m^2^). However, KDIGO-2012 guidelines recommend repeated eGFR measurement within 3 mo for all subjects with eGFR \< 60 mL/min per 1.73 m^2^, in order to confirm the classification\[[@B2]\]. Considering the recommended diabetes guidelines, our positively discordant group at the 60 mL/min per 1.73 m^2^ eGFR level (3.5%) would not experience any harm from being classified into more advanced CKD stage according to the Jaffé creatinine-based eGFR. On the contrary, clinical intervention to improve glycaemic control was identified as being immediately needed for all these patients, and, if KDIGO-2012-recommended confirmatory eGFR was to be done after 3 mo, presumably reached glycaemic goals would allow more accurate CKD classification not only by the Jaffé method. Limited design of our study did not allow the insight into actual follow-up data, but the well-established ameliorating effect of improved glycaemic control on renal function is likely to elicit the consequent beneficial effect on eGFR *via* biological mechanism(s)\[[@B23]\], regardless of the creatinine method used. Furthermore, the majority of the positively discordant cases had A2 and A3 stages of albuminuria, confirming an increased or high risk of CKD progression in these patients\[[@B2]\]. Significant improvements in CKD screening strategies were enabled by creatinine standardization and automated eGFR reporting by clinical laboratories\[[@B11],[@B12]\]. However, recommended practice is still not implemented in many clinical settings, indicating a lack of appropriate clinical care with significant medical and financial consequences\[[@B25]\]. One of the reasons for the reluctance to implement the guidelines is the concern of increased costs of specific creatinine testing. Our results show that low-cost Jaffé creatinine assay can be safely used for the CKD screening in patients with diabetes, despite confirmed positive analytical bias in comparison to enzymatic assay, provided that compensated Jaffé assay, traceable to the reference IDMS procedure, and CKD-EPI-calculated eGFR are used. The frequency of positively discordant CKD classification is very low and associated with severe hyperglycaemia, where the appropriate interventions to attain glycaemic goals are warranted regardless of CKD. Concomitant presence of increased albuminuria in the majority of discordant cases further diminishes the clinical practice consequences, which may, in turn, be beneficial for the patients in terms of increasing frequency of eGFR monitoring and intensifying efforts to control hyperglycaemia and hypertension. Due to a very low frequency, positive discordance is not likely to present a great burden for the health-care providers. On the other hand, a very low proportion of negatively discordant cases (1%) at the 60 mL/min per 1.73 m^2^ eGFR level indicate a negligible possibility to miss the CKD diagnosis, which could be the most prominent clinical problem affecting patient outcomes. Namely, CKD, being a major risk factor for cardiovascular disease and overall mortality, should be detected as early as possible, since appropriate interventions can substantially reduce risks and improve patient outcomes\[[@B2]\]. Limitations of our study include cross-sectional design and singular ethnicity. A large cohort of diabetic patients stratified according to albuminuria and degree of hyperglycaemia, as well as clinical outcome-based, rather than method-comparison-based approach can be regarded as the study advantages. In conclusion, results from our study indicate that compensated Jaffé creatinine procedure, in spite of the glucose-dependent bias, is not inferior to enzymatic creatinine in CKD diagnosis/staging and therefore may provide a reliable and cost-effective tool for the renal function assessment in diabetic patients. COMMENTS ======== Background ---------- Diabetic kidney disease (DKD) is one of the most prevalent chronic complications of diabetes and the most common single cause of end-stage renal failure. Current clinical guidelines recommend regular screening for and staging of DKD by the laboratory assessement of both urinary albumin excretion rate and estimation of glomerular filtration rate from serum creatinine (SCr-eGFR). Research frontiers ------------------ Analytical performance of the serum creatinine assays is the critical determinant of eGFR accuracy. The most widely used compensated Jaffé creatinine assay suffers from a non-specific bias from pseudo-creatinine chromogens (glucose, ketones), which is not the case with the costly enzymatic assays. Several studies advocated the replacement of the compensated Jaffé with enzymatic creatinine assays in order to improve reliability of the eGFR, especially in the diabetic population, which is expected to have an increased amount of interfering substances in serum. However, recently published risk-analysis study, using both analytical and biological variability criteria, revealed a low risk for misclassification of CKD based on Jaffé-SCr-eGFR results in the general population, while the clinical impact in the diabetic population remains unclear. Innovations and breakthroughs ----------------------------- This study evaluated the influence of creatinine methodology on the performance of CKD-EPI-calculated eGFR for CKD diagnosis/staging in diabetic patients. The results indicate that compensated Jaffé creatinine procedure, in spite of the glucose-dependent bias, is not inferior to enzymatic creatinine in CKD diagnosis/staging and therefore may provide a reliable and cost-effective tool for the renal function assessment in diabetic patients. Applications ------------ Significant improvements in CKD screening strategies were enabled in the last decade by creatinine standardization and automated eGFR reporting by clinical laboratories. However, recommended practice is still not implemented in many clinical settings, indicating a lack of appropriate clinical care with significant medical and financial consequences. One of the reasons for the reluctance to implement the guidelines is the concern of increased costs of specific creatinine testing. This results show that low-cost Jaffé creatinine assay can be safely used for the CKD screening in patients with diabetes, despite confirmed positive analytical bias in comparison to enzymatic assay, provided that compensated Jaffé assays, traceable to the reference IDMS procedure, and CKD-EPI-calculated eGFR are used. This finding is particularly valuable for the primary health-care facilities, since available and cost-effective screening for DKD may substantially improve patient outcomes and reduce overall costs associated with more advanced complications of diabetes. Peer-review ----------- The paper is far from being a useful tool for the clinical management of CKD patients, it shows an interesting new approach to be validated in a prospective way and bigger sample size in order to clarify the potential use in specific clinical situations of CKD patients. Manuscript source: Invited manuscript Specialty type: Endocrinology and metabolism Country of origin: Croatia Peer-review report classification Grade A (Excellent): 0 Grade B (Very good): B, B Grade C (Good): 0 Grade D (Fair): 0 Grade E (Poor): 0 Institutional review board statement: The study was approved by the Ethics Committee of the Merkur University Hospital, Zagreb, Croatia. Informed consent statement: This study was waived by IRB due to the retrospective nature of the study, with the post-hoc selection of anonymized samples from the routine laboratory visits, according to creatinine and glucose results. Conflict-of-interest statement: No potential conflicts of interest relevant to this article are reported. Data sharing statement: No additional data are available. Peer-review started: November 2, 2016 First decision: December 1, 2016 Article in press: March 2, 2017 P- Reviewer: D'Orazio P, Zhao JB S- Editor: Qi Y L- Editor: A E- Editor: Li D [^1]: Author contributions: Lovrenčić MV designed the study and wrote the article; Biljak VR and Lovrenčić MV analyzed the data, Biljak VR and Božičević B performed the laboratory part of the study; Blaslov K and Duvnjak LS performed the clinical part of the study; all authors discussed the results and approved the final version of the manuscript. Correspondence to: Marijana Vučić Lovrenčić, PhD, Department of Laboratory Medicine, Merkur University Hospital, Zajčeva 19, HR-10000 Zagreb, Croatia. <vucic@idb.hr> Telephone: +385-1-2353861 Fax: +385-1-2353847

The organ size is controlled through changes in rates of cell growth and division[@b1][@b2]. The underlying mechanisms of organ size control, such as Insulin/IGF signaling (IIS) and Target of Rapamycin complex 1 (TORC1) signaling pathways, have been intensively investigated by using proliferating and morphologically simple cells, such as those in epithelial monolayers of the Drosophila imaginal disc[@b3][@b4]. However, it is still largely unknown how the size of postmitotic and morphologically complex cells such as neurons scale with body size and which molecules control such scaling. We therefore explored these scaling mechanisms in Drosophila neurons. For our investigation, we used the dendritic arbor of one class of sensory neurons in Drosophila adults as a readout ([Figure 1A and 1B](#f1){ref-type="fig"})[@b5][@b6][@b7][@b8][@b9]. Dendrites are the antennae of neurons that receive and integrate sensory and/or synaptic inputs[@b10][@b11]. Our model neuron in this study, previously designated as v\'ada, is one of the dendritic arborization (da) neurons, whose arbor for adult life is regenerated during pupal stages and entirely covers the lateral plate (pleura) in the abdomen ([Figure 1A--1C](#f1){ref-type="fig"})[@b7][@b8]. Results ======= Scaling of dendritic arbors of the wild-type da neuron with body size --------------------------------------------------------------------- To examine how body-size changes affect the size and the branching pattern of the dendritic arbor of the wild-type neuron, we starved larvae beyond 91--96 hr after egg laying (AEL). This is because it has been shown that larvae exposed to such starvation after the "critical weight" stop body growth, but develop to become fertile adults that are smaller than normal, without any developmental delay[@b12][@b13][@b14][@b15]. Under this mild starvation condition, the size of the dendritic arbor was significantly decreased in proportion to the decreased body size ([Figure 1D--1G](#f1){ref-type="fig"}). Importantly, the number of branch endings was not changed and the branch density (total length/arbor size and ending number/arbor size) was significantly increased ([Figure 1H--1K](#f1){ref-type="fig"}). We also quantified the branching complexity, by assigning Strahler orders to individual segments (intervals between branching points; see [Figure 1L](#f1){ref-type="fig"})[@b16][@b17][@b18]. The numbers of individual order segments were similar under the fed and starved conditions ([Figure 1M](#f1){ref-type="fig"}), while 2^nd^ to 4^th^ order segments became shorter under the starved condition ([Figure 1N](#f1){ref-type="fig"}). These results are consistent with the hypothesis that the wild-type neuron can respond commensurately to the decrease in body size and/or the starvation and can form a "miniature" dendritic arbor by tuning its dendritic segment length. Neurons with defective IIS/TORC1 signaling pathways show dendritic "undergrowth" -------------------------------------------------------------------------------- The above result suggested that the wild-type neuron is able to scale with the body size and/or the nutrition condition, while keeping the branching complexity intact. We wondered if the IIS/TORC1 pathways ([Supplementary Figure S1A](#s1){ref-type="supplementary-material"}) participate in this scaling, and whether defects in the pathways would affect the size and the branching pattern of the dendritic arbor in adults, when larvae were raised under the normal food condition. Neurons with disruptions of *Drosophila insulin receptor* (*dinr*) or *Akt* downsized and simplified the dendritic arbors, as evidenced by decreases in the arbor size, the ending number, and the total length ([Supplementary Figure S1B--S1D and S1H--S1J](#s1){ref-type="supplementary-material"}). In contrast, the branch density (total length/arbor size and terminal number/arbor size) was not significantly altered from that of the wild-type neuron ([Supplementary Figure S1K and S1L](#s1){ref-type="supplementary-material"}). A loss of function mutation of *tor*, overexpression of a dominant negative form of Phosphoinositide 3-kinase (PI3K), or a knockdown of *raptor* encoding an essential component of TORC1 not only decreased the arbor size, but also the branching complexity ([Supplementary Figure S1E--S1G](#s1){ref-type="supplementary-material"}). Thus this "undergrowth" phenotype is distinct from the "miniature" dendrite of the wild-type neuron under the starved condition ([Figure 2L](#f2){ref-type="fig"}), suggesting that a regulatory mechanism other than the IIS/TORC1 pathways may contribute to the dendritic scaling in the normal da neuron. *CHORD* mutant neurons form miniature dendrites ----------------------------------------------- To explore the hypothetical mechanism of the dendritic scaling, we conducted a forward genetic screen by employing the mosaic analysis with a repressible cell marker (MARCM) system[@b19]. To facilitate the generation of mosaic clones, we made "SOP-FLP" lines that express FLP recombinase in sensory organ precursors (SOPs; see details in Methods). In our screening under the fed condition, we found a mutant chromosome that generated miniature dendritic patterns in homozygous neurons ([Figure 2B and 2C](#f2){ref-type="fig"}). Indeed a number of quantitative features indicated that the arbor of this mutant neuron was a proportionally scaled down miniature or a microcopy of the wild-type neuron ([Figure 2E--2I](#f2){ref-type="fig"}): decreases in the arbor size and the total dendritic length, an unaltered ending number, an increase in branch density, profiles of the Strahler-order analysis similar to those of the wild-type neuron under the starvation condition (compare [Figure 1M--1N](#f1){ref-type="fig"} with [Figure 2J--2K](#f2){ref-type="fig"}), and an unaltered distribution of angles at individual branching points (data not shown). In a word, the mutant neuron reduced the dendritic segment length ([Figure 2K](#f2){ref-type="fig"}), without simplifying the branching pattern, which is reminiscent of how the wild-type neuron constructed its arbor under the starved condition ([Figure 1F](#f1){ref-type="fig"}). Whole-genomic sequencing and complementation mapping using Drosophila deficiency stocks identified a 1 bp deletion that leads to a frameshift in the *CHORD*/*morgana* gene (denoted *CHORD^2^* hereafter; [Figure 2A](#f2){ref-type="fig"}). CHORD is an evolutionarily conserved co-chaperone of HSP90, and it negatively regulates Rho-kinase (Rok) activity to suppress overduplication of centrosomes[@b20][@b21]. Introduction of the 4757-base-pair (bp) genomic fragment including the *CHORD* gene almost fully rescued the above phenotypes to normal, demonstrating that *CHORD* is the causative gene for the miniature dendrite phenotype ([Figure 2D and 2E--2K](#f2){ref-type="fig"}). The miniature arbor of the *CHORD* mutant neuron may mimic a default state of the wild-type neuron -------------------------------------------------------------------------------------------------- One possible explanation of the *CHORD* phenotype is that the mutant neuron no longer interprets favorable or unfavorable extracellular conditions (either the nutrition and/or body size) and generates the miniature arbor as an invariable default. Alternatively, the *CHORD* mutant neuron misreads the size of the abdominal lateral plate; consequently, the arbor occupies only a portion (e.g., 50% instead of 100%) of the body surface. To distinguish these possibilities, we examined dendritic arbors of *CHORD* mutant neurons under the starved condition ([Figure 3A and 3B](#f3){ref-type="fig"}). Neither the size nor the branching complexity significantly changed between the two nutrition conditions ([Figure 3C--3I](#f3){ref-type="fig"}), showing that the *CHORD* mutant neuron formed the miniature arbor irrespective of the extracellular condition. This result suggested that the latter possibility was less likely. It appears that the wild-type neuron possesses a CHORD-dependent mechanism that extends the branch segment length beyond a preset value in response to a favorable environment. Elongation/retraction rate of terminal branches is critical for reproducing the miniature phenotype of the *CHORD* mutant neuron -------------------------------------------------------------------------------------------------------------------------------- We next investigated the dynamics of dendrite formation whereby the *CHORD* mutant neuron produced the miniature arbor during pupal stages. First we found that the mutant arbor underwent a persistent increase in complexity and size, but the overall appearance had already become miniature-like before eclosion ([Supplementary Figure S2A--S2H](#s1){ref-type="supplementary-material"}). Then we performed time-lapse recordings of the wild-type and mutant neurons at 70--75 hr after puparium formation (APF), when dendritic arbors were under active construction, and dissected the dendrite dynamics of elongation/retraction and branching. We quantified frequencies of elongations and retractions of terminal branches (branches of Strahler order 1, see [Figure 1L](#f1){ref-type="fig"}) and found that these parameters were not altered in *CHORD* mutant neurons ([Supplementary Figure S2I and S2J](#s1){ref-type="supplementary-material"}). We also used a Fano Factor as a quantitative descriptor of branch tip dynamics (see details in Methods)[@b22] and observed that there was no significant difference between the wild-type and *CHORD* mutant neurons ([Supplementary Figure S2K](#s1){ref-type="supplementary-material"}). In contrast, the rates of branch--length changes (μm/min) were significantly decreased in the mutant neurons ([Supplementary Figure S2L](#s1){ref-type="supplementary-material"}). In this sensory neuron, most of the branches are collaterals that sprout from stalks, rather than bifurcations of branch tips[@b9]. Therefore, we quantified the frequency of lateral branching and found that wild-type and *CHORD* mutant neurons showed similar values ([Supplementary Figure S2M](#s1){ref-type="supplementary-material"}). Altogether, dendritic dynamics of *CHORD* mutant neurons are very similar to that of wild-type neurons, with the exception that the rates of change of branch lengths decrease in the mutant neurons. We then asked whether these changes can explain the miniature phenotype of the mutant neurons or not. We addressed this question with a computer-assisted simulation of dendrite growth using experimental values. The elongation and retraction of the terminals was represented as the addition and removal, respectively, of one unit segment per 1 minute ( = 1 step in simulation) at the terminal of the dendrites, and lateral branching was reproduced by the addition of unit segments to the existing branches[@b23][@b24][@b25]. The probabilities of elongation, retraction, and lateral branching were determined by experimental data (see details in Methods and [Supplementary Table S3](#s1){ref-type="supplementary-material"}). We defined the unit length of a segment as 0.9 μm or 0.75 μm, based on the experimental data for the wild-type or *CHORD* mutant neurons, respectively ([Supplementary Figure S2L](#s1){ref-type="supplementary-material"}). Representative images of the wild-type and mutant model neurons closely resembled those of in vivo neurons ([Supplementary Figure S2N](#s1){ref-type="supplementary-material"}). In fact, our quantification showed decreases in the arbor size and the total length, an unaltered ending number, and an increase in the branch density of the mutant model arbors, compared to the wild-type model arbors ([Supplementary Figure S2O--S2S](#s1){ref-type="supplementary-material"}). Thus, these results support the hypothesis that subtle but significant differences in the rates of branch-length changes play a critical role in scaling the dendritic arbor, and that the miniature arbor of the *CHORD* mutant neuron can be attributed to the scaled-down elongation/retraction dynamics of dendritic branches. Finally, we examined whether the dendrite dynamics in starved wild-type animals is comparable to that of *CHORD* mutant neurons. We found that the wild-type neurons under the starved condition showed significantly decreased rates of branch--length changes compared to those under the fed condition, while other parameters were not significantly changed ([Supplementary Figure S2I--S2M](#s1){ref-type="supplementary-material"}), very much resembling *CHORD* mutant neurons under the fed condition. These results strengthened our hypothesis that coordinated regulation of the elongation/retraction dynamics of branches contributes to scaling of the dendritic arbor. Genetic interaction between CHORD and the TORC2 component Rictor ---------------------------------------------------------------- We next asked how body size/nutrition state is conveyed to CHORD. As a candidate of such an upstream molecule, we examined the involvement of TORC2, another TOR complex, for the following reasons: (1) TORC2 is activated by association with ribosomes in a growth factor dependent manner, thus it may mediate between the extracellular environment and growth capacity of individual cells[@b26][@b27][@b28]. (2) TORC2 regulates the actin cytoskeleton and controls dendrite pattern formation[@b27][@b28][@b29][@b30]. We assessed the phenotype of mutant neurons of *rictor*, which encodes an essential and specific component of TORC2 ([Figure 4A and 4B](#f4){ref-type="fig"}). *rictor* mutant neurons showed complicated but intriguing phenotypes; that is, combined features of undergrowth and miniaturization. They showed a decrease in the arbor size, the total length, and the ending number ([Figure 4D--F](#f4){ref-type="fig"}), which is characteristic of the undergrowth phenotype. On the other hand, they also exhibited features of the miniature phenotype: increases in the branch density (both the total length/arbor size and the ending number/arbor size, as shown in [Figure 4G and 4H](#f4){ref-type="fig"}) and a decrease in segment length ([Figure 4J](#f4){ref-type="fig"}). Importantly, overexpression of the *CHORD* transgene in *rictor* mutant neurons partially restored the features of the miniature phenotype ([Figure 4G--4J](#f4){ref-type="fig"}), but not those of the undergrowth phenotype ([Figure 4D--4F](#f4){ref-type="fig"}). In contrast to this genetic interaction between CHORD and TORC2 (Rictor), *CHORD* overexpression did not rescue simplified phenotype of the *dinr* knockdown ([Supplementary Figure S1M and S1N](#s1){ref-type="supplementary-material"}). All these results could be explained by the hypothesis that CHORD functions downstream of TORC2, being at least partly separate from TORC1. The partially separate nature was further supported by our KD experiment using S2 cells ([Supplementary Figure S1O](#s1){ref-type="supplementary-material"}). Even when *CHORD* was knocked down, phosphorylation of S6K (a readout of TORC1 activity) was up-regulated in an insulin-dependent manner, as in the control knockdown cells. We then asked whether the TORC2 activity is altered under the nutrient-limited condition that was employed to decrease body size. We starved larvae for 8 hrs beyond 91--96 hr AEL and analyzed the phosphorylation of Akt at Ser505 as an established readout of the TORC2 activity[@b31]. The level of S505 phosphorylation was significantly and reproducibly reduced under this condition compared to that in well-fed flies ([Figure 4K](#f4){ref-type="fig"}). All these results suggest the possibility that TORC2 may communicate the extracellular conditions (body size and/or nutrients) to CHORD proteins ([Figure 4L](#f4){ref-type="fig"}). Discussion ========== This study provides novel mechanistic insights into size control of neuronal dendritic arbors. To sample sensory input precisely, certain types of neurons should adjust their dendritic arbor size to cover the receptive field completely. We showed that there are two distinct ways of downsizing dendritic arbors when the field size is reduced: one way is to arrest both growth and branching during maturation of dendrite morphologies, causing the "undergrowth" phenotype. The other way is to regulate the elongation of dendrites (more specifically speaking, the length of branch segments) selectively, thereby making a "miniature" form ([Figure 2L](#f2){ref-type="fig"}). Our results support the notion that dendritic growth and branching are controlled by at least partly separate mechanisms, which is also seen in other instances such as the development of postembryonic dendritic architecture of motorneurons[@b32]. At the molecular level, the IIS/TORC1 pathways control both growth and branching to avert underdevelopment, whereas CHORD and TORC2 tunes the segment length to realize proportional scaling of the entire arbor ([Figure 4L](#f4){ref-type="fig"}). Ablation of the TORC2 component Rictor in mouse Purkinje cells causes multiple structural changes of dendritic arbors, including a decrease in total dendrite length and an increase in the number of primary branches[@b30]. It will require further investigation to address how the overall branching complexity and the segment length is affected in the absence of Rictor, and to fully characterize the loss of function phenotype of CHORD in this subtype of neurons. CHORD was originally discovered in plants as a key player in the innate immune response[@b33][@b34]; in animals, CHORD negatively regulates Rho kinase activity, thereby suppressing overduplication of centrosomes[@b20][@b21]. *CHORD* is also expressed in tissues that are populated by postmitotic cells, such as brains[@b35], and has been proposed to function beyond regulating cell division. Here, we revealed that CHORD regulates the size of dendritic arbors: cells defective in CHORD showed a decreased elongation/retraction rate of terminal branches; thus, by extrapolation, the normal function of CHORD must be to accelerate the elongation and retraction of terminal branches. Rho-kinase (Rok) is reported to inhibit neurite outgrowth, dendritic branching, and spine formation[@b36]. Therefore, we pursued the possibility that Rok acts downstream of CHORD. However, we couldn\'t find any genetic and biochemical evidence for the interaction between Rok and CHORD ([Supplementary Figure S3](#s1){ref-type="supplementary-material"}; see details in the legend). Instead, our results suggested that CHORD is functionally related to the TORC2 signaling pathway, which regulates the actin cytoskeleton[@b28]. It has been reported in mammalian fibroblasts that phosphorylation by TORC2 facilitates folding of some AGC kinases, such as Akt and conventional PKCs, and stabilizes them, and that newly synthesized, unphosphorylated Akt is protected by HSP90 from degradation[@b37][@b38]. Therefore, CHORD may contribute as a co-chaperone to the stability of the AGC kinases by recruiting HSP90 to those clients, or it may employ its own chaperone activity[@b39] to perform this task. This hypothetical role of CHORD might underlie the partial rescue of the *rictor* mutant neuron by *CHORD* overexpression. Comparative anatomical studies have reported that dendrites of some types of neurons, such as sympathetic neurons, become larger and increase branching complexity as the brain or the body size increases across species[@b40][@b41], which we designate overgrowth as opposed to undergrowth. In contrast, other types of neurons such as somatosensory thalamocortical projection neurons increase their dendritic arbor size with larger brains, while preserving key features of the dendritic branching pattern[@b42][@b43], which is proportional magnification. These two distinct ways of scaling, overgrowth/undergrowth and miniaturization/magnification, might be regulated by IIS/TORC1 and TORC2/CHORD, respectively, and may contribute to neuronal cell-type specific information processing. It should be noted that "scaling growth" of dendritic arbors does take place during larval development, but mechanistically it is distinct from the arbor scaling at postlarval stages, which we reported in this study. Larval da neurons regulate their growth in coordination with the expanding body whose mass increases by approximately 200-fold during the complete larval development[@b18][@b44]. This is accomplished by increasing both branching numbers and the total length, where IIS/TORC1 plays a critical role[@b29][@b45], but *CHORD* appears to be dispensable ([Supplementary Figure S2](#s1){ref-type="supplementary-material"}), and TORC2 is required for dendritic tiling for larval da neurons[@b29]. On the other hand, the da neuron during pupal development first prunes its dendrites and then starts constructing arbors for adult life; thus this neuron completes this task in a body whose volume has been predetermined by the nutritional status during larval development. It is this task that requires the role of CHORD, which fits the dendritic arbor to the final adult body size. We speculate that this hypothetical CHORD-dependent mechanism somehow senses body size by receiving as-yet undetermined signals, either local or systemic, from other tissues. Candidate tissues for sources of such signals would be either those adjacent to the neuron, such as lateral tergosternal muscles, abdominal histoblasts (epitherial cells), and glial cells, and/or tissues that secrete growth factors, such as fat body and insulin producing cells[@b8][@b45][@b46][@b47][@b48]. Therefore the dendrite scaling of the adult da neuron might provide a useful model system to study interactions between neurons and other organs. Methods ======= Molecular cloning ----------------- Six tandem repeats of a 20-base-pair (bp) sequence that includes a proneural binding site (\[scE1\]~6~)[@b49] were used to build a *SOP-FLP* transgene. The *hsp70* minimal promoter and the entire coding sequence of *flp* from the UAS-FLP vector (DGRC) were inserted between \[scE1\]~6~ and a SV40-polyA sequence. This *SOP-FLP* transgene was cloned into the pHStinger vector, or a pUAST vector from which the UAS sequence and *hsp70* minimal promoter had been removed. To generate a rescue construct, the 4754-bp sequence (3R: 20009616-20013777 in version FB2013_05) that includes *CHORD* was amplified from *yw* genomic DNA and cloned into the vector pCasper. To generate UAS shRNA for each gene, we followed a protocol previously described[@b50]. The target sequences of the shRNAs are as follows: 5′-CACCGAGTTCCTCAACATCAA-3′ and 5′-TTCGACCTGGATGACATTAAA-3′. These shRNAs were cloned into the vector pUAST-attB. All constructs were injected, in accordance with standard protocols, to generate transgenic fly lines. Drosophila strains ------------------ Detailed genotypes of the animals and clones are described and summarized in [Supplementary Table S1](#s1){ref-type="supplementary-material"}. To visualize dendrites and /or express transgenes, we used the following *Gal4* drivers: *Gr28b.c*[@b51][@b52] and *Gal4^5-40^*[@b53]. To express fluorescent proteins, we used *UAS-mCD8:GFP* (\#5137 of the Bloomington Stock Center) or *UAS-Venus-pm*[@b44][@b54][@b55]. Other strains used were *dinr^339^*[@b56], *Akt^q^*[@b57], *Tor^ΔP^*[@b58], *rok^2^*[@b59], *UAS-dicer2* (\#60009 of Vienna Drosophila RNAi Center), *UAS-raptor^RNAi^* (HMS00124/\#34814 of the Bloomington Stock Center), *UAS-mCherry^RNAi^* (\#35785 of the Bloomington Stock Center), *UAS-CHORD^RNAi^* (this study), *UAS-Dp110\[D954A\]* (\#25918 of the Bloomington Stock Center), *UAS-rok^RNAi^* (GD1522/\#3793 and KK107802/\#104675 of the Vienna Drosophila RNAi Center; JF03225/\#28797, HMS01311/\#34324, and GL00209/\#35305 of the Bloomington Stock Center), *UAS-hsp90^RNAi^* (HMS00899/\#33947 of the Bloomington Stock Center) and *UAS-Rok.CAT^48.2^* [@b60]. MARCM-based forward genetic screen ---------------------------------- The *piggyBac* insertion collection with FRT insertion[@b61] was used for genetic screening. We crossed individual insertion stocks to "*SOP-FLP* based MARCM-ready" fly stocks (for 2^nd^-chromosome left arm screen, *Gal4^5-40^ UAS-Venus:pm SOP-FLP^\#42^; tubP-Gal80 FRT40A*; for 3^rd^-chromosome left arm screen, *SOP-FLP^\#42^; Gal4^109(2)80^ UAS-mCD8:GFP SOP-FLP^\#73^/CyO; tubP-Gal80 FRT2A*; and for 3^rd^-chromosome right arm screen, *hsFLP UAS-mCD8:GFP; Gal4^109(2)80^ UAS-mCD8:GFP SOP-FLP^\#73^/CyO; FRT82B tubP-Gal80*). We then mounted adult abdomens, and dendrites were imaged under a laser scanning microscope. We screened 1537 *piggyBac* stocks and isolated 3 stocks (*LL04611*, *LL04133*, and *LL03277*) that showed the "miniature" phenotype, and 19 that showed the "undergrowth" phenotype ([Supplementary Table S2](#s1){ref-type="supplementary-material"}). Both *LL04611* and *LL04133* were homozygous lethal. Mapping and identification of the mutation that was responsible for the "miniature" phenotype --------------------------------------------------------------------------------------------- Starting from the above *LL04611* and *LL04133* stocks of the *piggyBac* collection, we generated precise excision strains. These precise excision strains were still homozygous lethal, and MARCM clones of those strains showed the miniature dendritic phenotypes. We performed a complementation test for lethality between *LL04611* and *LL04133*, and found that they belong to the same complementation group. These results suggested that the lethality and the "miniature" dendrite phenotype were caused by a background mutation(s) that was not linked to the piggyBac insertion. We backcrossed the excision strains to the control *FRT82B* stock three times and used the progeny for the phenotypic analysis below. To identify the mutated gene that gave rise to the phenotype, we determined whole-genomic sequences of the two mutant strains with an Illumina next-generation sequencer and directly compared them with each other. For the sequencing, we prepared genomic DNA from larvae that were heterozygous for *LL04611* or *LL04133* and employed paired-end Illumina sequencing technology. We also determined the sequences of *LL02779* and *LL00232*, whose MARCM clones did not show a miniature phenotype, as negative controls. Then we mapped the sequence data to the wild-type reference genome and identified variants relative to the wild-type reference genome by using Burrows-Wheeler Aligner (BWA) and SAMtools. In addition to the whole genome sequencing, we performed conventional deficiency mapping of the lethality by using the Bloomington Deficiency Kit, and mapped the mutation responsible for the lethality (and possibly the clone phenotype) within a \~45 kb region in the right arm of the 3rd chromosome. Within this \~45 kbp mapped region, we found only one mutation unique to *LL04611* and *LL04133*: a 1 bp deletion in the *CHORD/morgana* gene that caused a frame shift (denoted as *CHORD^2^*), which was confirmed by the Sanger method. Image collection and image analysis ----------------------------------- Imaging da neurons in whole-mount adults was done as described earlier[@b7]. Briefly, we collected adult females within 12-hour after eclosion unless described otherwise, washed them in 0.7% NaCl and 0.3% Triton X-100, removed the heads and legs of adult flies, and mounted the abdomens in 50% glycerol on slides, between spacers made of vinyl tape. All of the images were acquired using a Nikon C1 laser-scanning confocal microscope or a Zeiss LSM 510 META laser-scanning confocal microscope. For quantification of dendritic patterns of da neurons, live imaged dendrites were first traced using Simple Neurite Tracer plugin in Fiji and tracing data was exported as csv files. Quantification of total length, terminal number, total length/arbor size, and terminal number/arbor size and analysis of Strahler order were performed using Excel. Arbor size was quantified by analyzing skeletonized tracing images by using the convex-hull selection in Fiji. The data of different genotypes were compared by one-way ANOVA with Tukey\'s HSD post hoc analysis, or by Student\'s t-test using KaleidaGraph (version 4.0; Synergy Software). Time-lapse recording -------------------- Time-lapse recordings of branch formation ([Supplementary Figure S2I--S2M](#s1){ref-type="supplementary-material"} and [Supplementary Video S1--S3](#s1){ref-type="supplementary-material"}) were performed as described previously[@b9]. Briefly, each pupa between 70--74 hour after puparium formation (APF) was taken out of its puparium and mounted on a 35-mm glass-bottomed dish. In mounting, we folded legs and put abdomens on the dish and tilted them to retain an appropriate angle. All images were collected at 2-minute intervals for 1-hour with a 2-μm Z-step using a Nikon C1 laser-scanning confocal microscope. After image acquisition, each pupa was kept at 25°C, and their survival was confirmed to at least the pharate stage or the adult stage. Acquired movies were traced manually using Simple Neurite Tracer plugin in Fiji and the frequency and rate of elongation/retraction were quantified using Excel. As an unbiased indicator of branch dynamics, a Fano factor (FF) value was calculated based on branch tip length measurements. The FF is defined as the variance in the measured length of an individual branch across all imaging sessions, divided by the mean length of that branch[@b22]. We quantified neurons whose FF was above 0.35, which is defined as dynamic neurons[@b22]. Reconstruction of dendrite dynamics in a computer-aided simulation ------------------------------------------------------------------ Dendrite growth dynamics was modeled in two-dimensional space following previous mathematical models[@b23][@b24][@b25]. The 'elongation' and 'retraction' were represented by the addition and removal of 1 unit segment, respectively. As a length of 1 unit segment, we employed the 0.9 μm and 0.75 μm to recapitulate the wild-type and *CHORD* mutant neurons, respectively. Lateral branching was reproduced by the addition of unit segments to the existed branches. The angle of elongation was assumed to follow a homogenous distribution from −7.5° to +7.5°, while that of branching was assumed to follow a normal distribution with average 97° and standard deviation 38°. The initial condition was as follows: 2 branches with 20-segment lengths sprout from the origin in anti-parallel directions along the y axis and bifurcate with angle 30°, and each of these 4 branches elongate 10-segments in length. Other parameters used in the simulation are summarized in [Supplementary Table S3](#s1){ref-type="supplementary-material"}. Production of antibodies, RNAi, and Western blotting ---------------------------------------------------- Guinea pigs were immunized with GST proteins that had been fused to the N-terminal 200 amino acids of CHORD to generate anti-CHORD antibody. For RNAi experiments, S2 cells were cultured with 20 μg/ml dsRNA of the full-length CHORD or GFP-coding sequence for 3 days. Lysates were separated on a 8% polyacrylamide gel and transferred to PVDF (Millipore). Antibodies were used at the following concentrations: guinea pig anti-CHORD, 1:1000 (this study); rabbit anti-phospho-Myosin light chain 2 (Ser-19), 1:1000 (Cell Signaling \#3671); rabbit anti-phospho-Akt (Ser-505), 1:1000 (Cell Signaling \#4054); rabbit anti-Akt, 1:1000 (Cell Signaling \#9272); mouse anti-actin, 1:1000 (Millipore MAB1501). Signals were detected with ECL plus (Amersham). To obtain stronger signals, we employed Can Get Signal Immunoreaction Enhancer solution (TOYOBO). Author Contributions ==================== K.S. carried out most of the experiments; K.F., T.N., M.O., T. Usui and M.K. helped some experiments. K.S. and T. Uemura wrote the paper. Supplementary Material {#s1} ====================== ###### Supplementary Information Recording of dendritic dynamics of the wild-type neuron under the fed condition ###### Supplementary Information Recording of dendritic dynamics of the CHORD mutant neuron ###### Supplementary Information Recording of dendritic dynamics of the wild-type neuron under the starved condition ###### Supplementary Information Supplementary Figures and Tables The reagents were provided by the Drosophila Genetic Resource Center at Kyoto Institute of Technology, the Bloomington Stock Center, Y.N. Jan, B. Yi, T. Chihara, K. Emoto, L.M. Powell, B. Lemaitre, S. Cohen, R. Niwa, T. Nishimura, R. Karess, N. Fuse, S. Hayashi, and J. Zallen. We thank O. Nishimura, N. Fuse, and R. Ueda for help with next-generation sequencing analysis, S. Yonehara for use of the DNA sequencer, N. Yamamoto for discussion about the neuron size, J.A. Hejna for his efforts to polish the manuscript, and M. Futamata, Y. Miyake, K. Shimizu, J. Mizukoshi, and K. Oki for their technical assistance. We also thank the members of Comparative Genomics Laboratory in National Institute of Genetics for technical and computational assistance. This study was supported by a Grant-in-Aid for Scientific Research (A) and for Scientific Research on Innovative Areas "Mesoscopic neurocircuitry" (22115006) to T. Uemura, by the Funding Program for Next Generation World-Leading Researchers (NEXT Program) to M. K., by Research and Education Platform on Dynamic Living Systems, by NIG Collaborative Research Program (2011-A48), and by a Grant-in-aid for Scientific Research on Innovative Areas 'Genome Science' and 'Comprehensive Brain Science Network' from the Ministry of Education, Science, Sports and Culture of Japan. K.S. was a recipient of a Fellowship of the Japan Society for the Promotion of Science for Young Scientists. {#f1} ![*CHORD* mutant neurons form "miniature" dendrites.\ (A) Whole-genome sequencing identified a 1 bp deletion in the *CHORD* gene unique to the mutant strain that was isolated from our forward genetic screening (bottom). We denoted this allele as *CHORD^2^* hereafter. Note that we determined the sequences of flies heterozygous for the mutation. (Top) The domain structure of CHORD protein[@b20]. The 1 bp deletion caused a frame shift that resulted in a premature stop codon (M172 → STOP) in the 2^nd^ CHORD domain. (B--D) Representative images of MARCM clones of the wild-type neuron (B), the *CHORD^2^* mutant neuron (C) and the *CHORD^2^* mutant neuron to which a genomic fragment including the wild-type *CHORD* gene was introduced (D). The *CHORD^2^* mutation is early larval lethal, and note that all of the data throughout this study are from single neurons homozygous for individual mutations in otherwise heterozygous animals, using the MARCM system (hence, a mosaic analysis), except for the data from RNAi. Scale bars, 50 μm. (E--K) Quantification of branching patterns of dendritic arbors. (E) Dendritic arbor size. (F) Total length of dendritic branches. (G) The number of endings of dendritic branches. (H and I) The branch density: total length/arbor size (H) and ending number/arbor size (I). (J and K) The number (J) and length (K) of branches of each order. All data are presented as means ± standard deviation (SD). \*p \< 0.05, \*\*p \< 0.01, and \*\*\*p \< 0.001. Blue asterisks indicate statistically significant differences of the cohort from the wild-type neuron; and orange asterisks indicate statistically significant differences of the cohort from the *CHORD* mutant neuron. NS: Statistically not significant (*P* \> 0.05). (L) Schematic representation of two partly distinct ways of downsizing dendritic arbors. In neurons with defective IIS/TORC1 signaling pathways showed the final arbor pattern similar to the proximal arbor of the normal neuron (left). In contrast *CHORD* mutant neurons proportionally downsize the original structure, making a miniature form (right).](srep04415-f2){#f2} {#f3} {ref-type="supplementary-material"}. All data are presented as means ± standard deviation (SD). \*p \< 0.05, \*\*p \< 0.01, and \*\*\*p \< 0.001. Blue asterisks indicate statistically significant differences of the cohort from the wild-type neuron; and orange asterisks indicate statistically significant differences of the cohort from the *rictor* mutant neuron. NS: Statistically not significant (*P* \> 0.05). (L) Hypothetical molecular complexes that regulate dendrite growth and branching in response to the extracellular environment (the body size and/or the nutrition condition). The environmental cue(s) controls CHORD downstream of TORC2, which tunes the segment length (growth); on the other hand, the cue(s) also regulates the IIS/TORC1 pathway and affects both growth and branching.](srep04415-f4){#f4}

Letter: Since the initial report of infection with the coronavirus disease 2019 (COVID-19) in December 2019, the number of confirmed COVI-19 cases has exceeded 1.6 million globally.[@bib1] Without a vaccine in sight and given the success of social distancing, we expect the outbreak to last longer than originally projected. Social distancing while at work remains a challenge of us all. Mid-March, we implemented separate call teams, so that each of us only comes into contact with 4 to 5 coworkers at maximum over the course of the pandemic. These changes are necessary to maintain the workforce required to continue to be able to provide mechanical thrombectomy (MT) for our patients. Every stroke care system must prepare for the worst-case scenario involving a surge that completely overwhelms the capacity of that system to function. Over the last several weeks, we have been learning from our colleagues around the nation---first, news out of Seattle and then New York City, and now most recently from Detroit, with greater than 800 providers testing positive for COVID-19 at Henry Ford. With approximately 25% of health care providers being infected with the virus, we are faced with the challenge of providing the care our patients desperately require while ensuring the safety of our health care providers.[@bib2] We must be mindful that our selfless tendencies as dedicated frontline providers do not become our biggest vulnerability. This challenge requires that we adjust our mindset and also place at the forefront the safety of ourselves and our teams. We make up one of the smallest and most subspecialized units in our hospitals. As such, we are easily incapacitated with quarantine or illness; who will take care of our community then? This safeguarding requires major changes to our pre-COVID-19 workflow, to which the current pandemic has added multiple layers of complexity. Engagement of multiple specialties, including emergency department (ED) physicians, stroke neurologists, anesthesiologists and neurointensivists, is required. Fears and anxieties are normal human responses to a pandemic, and each must be acknowledged and addressed, never dismissed. In the acute stroke setting, the stakes are greatest because every minute counts. Although it may seem that there is not enough time to get it right, we must remember that this is also no time to get it wrong. That could mean being placed on diversion and not having the capacity to continue to treat our patients with stroke. It begins with not overly burdening an already-tenuous hospital system. In the current COVID-19 crisis, there is critical bed shortage at greater level of care hospitals. We must therefore reduce the number of "futile transfers" from community hospitals and ensure that only patients who are likely to receive MT are transferred to the thrombectomy-capable center. Turning down a transfer could be the difference between being able to accept the next. This will entail obtaining advanced imaging at spoke hospitals to confirm the presence of proximal large vessel occlusion (LVO) before transport. Patients ruled out for LVO can remain at the spoke for routine care, even those who receive intravenous thrombolysis, particularly those spokes that are primary stroke center certified. Even for patients in whom an LVO is confirmed, the criteria for transfer to the thrombectomy capable center should be decided on by a multidisciplinary team. During a pandemic and in the context of bed shortages, plus the prospect of bringing in vulnerable patients into an environment in which they might have a greater likelihood of acquiring the virus, hard decisions must be made. Accept the hard truth that your center should be more stringent in your criteria for MT, particularly for patients who don\'t fall within the guideline recommendations. As a team, discuss how you will handle those falling outside the "trial criteria"---very elderly patients, patients with mild, yet disabling stroke symptoms, and patients with distal occlusions. Only patients with a high likelihood of receiving thrombectomy should be transferred. Once the patient with LVO who is an MT candidate arrives to your ED, all teams must be in alignment. Pre-thrombectomy screening for COVID-19 is a major challenge; often these patients are unable to provide a history and collateral information is typically lacking. The Society of NeuroInterventional Surgery currently recommends that patients with unknown COVID-19 status should be screened for fever and respiratory symptoms.[@bib3] Given that approximately 35% of patients with COVID-19 are asymptomatic and the growing awareness that asymptomatic patients are able to transmit the virus, this screening might not be sufficient.[@bib4] It may therefore be prudent to consider is to obtain a computed tomography (CT) scan of the chest to evaluate for underlying infiltrates at the time of obtaining a CT angiogram and perfusion. Although extending the CT to capture the entire chest entails additional radiation, identifying a lung infiltrate suggestive of likely COVID-19 infection in an asymptomatic patient affords the health care team the ability to don the proper precautions before the procedure. Another question that has sparked major debate is whether to intubate patients before MT. Some proponents of an "intubate-all" approach argue that it protects the small and highly specialized MT team from exposure. Although all valid points, this approach fails to consider the other facets and team members along the way. Thrombectomy in and of itself is a low-risk procedure for contracting the virus, whereas the intubation and extubation are by far the greatest-risk components and would incur additional risk to your ED and neuro-intensive care unit (ICU) colleagues, respectively. In addition, intubating an elderly patient adds morbidity, and the airway management would require negative-pressure rooms in both the ED and the neuro-ICU. These are resources expected to be either in very short supply or not available at all during peak surge. The lowest-risk pathway to the system as a whole is to get the LVO patient through the thrombectomy awake and cooperative without requiring pre-procedure intubation. However, the greatest potential risk to both the anesthesia and MT teams is in the event the thrombectomy unexpectedly transforms into an aerosol-generating procedure. A patient coughing or vomiting mid-procedure, would entail the worse possible scenario not just by the greatest-risk exposure to the providers but also in the delay it would incur. Neuroangiography suites are positive-pressure rooms (exceptions would be hybrid angio/operating room suites), necessitating a clearing out of the room by personnel following an aerosol-generating procedure like an intubation. Thus, the "lowest-risk" pathway is not necessarily the safest. Finding the balance of risk to patient versus provider and resource use (negative-pressure rooms and personal protective equipment \[PPE\]) is delicate, and the decision of which patient should undergo preprocedural intubation is not straightforward. The Society of NeuroInterventional Surgery recommends a lower threshold for intubation for patients with suspected COVID-19, however, without specific criteria for intubation.[@bib3] We suggest that those patients at risk for converting an otherwise non--aerosol-generating procedure (thrombectomy) to a greater-risk procedure due to airway compromise should be intubated before MT. These would include those with severe stroke symptoms, patients with receptive aphasia, any signs of respiratory distress, or vertebrobasilar occlusion should be intubated before MT. Intubation ideally takes place in a negative-pressure room in the ED. Patients should then be transferred to the ICU with the same ventilator so that a closed circuit can be maintained. The emergence of a faster COVID-19 test that provide results within minutes will provide us with a more efficient and reliable way to rapidly triage patients with LVO and avoid unnecessary intubations. However, until that test becomes widely available, following vigilant screening and maximizing precautions will be of paramount importance to protect our health care providers. Another consideration is modification to your angio-suites for team protection. At our center, we have designated one of the biplanar rooms a "COVID-19" room, in which patients with suspected or confirmed COVID-19 receive MT. The door to this room has been sealed off, making it impermeable to aerosolized particles to completely isolate it from the control room. We have rehearsed an elaborate protocol to deliver additional supplies and devices into that room should they be needed via the hallway, as well as donning/doffing of PPE for room entry/exit protocol. Due to staff and PPE shortages, only essential personnel should be allowed into the COVID room during the procedure. This means that only one nurse, one technologist, the interventionists, and anesthesiologist are the only staff member to be in the room. Once the procedure is finished, patients are transferred to the ICU, where they are extubated in a negative-pressure room. Remarkable challenges lie ahead of us, but we remain optimistic knowing that our field is graced with tremendously devoted, talented, and innovative people. Our solutions may not always be perfect, but these are also imperfect times, and we may have to do the best with what we can. However, one certainty remains: on reflecting back on these unprecedented times, it will be known that we answered the call for our patients.

Introduction {#Sec1} ============ Heart failure is an ever-growing disease with a growing number of patients in end-stage disease, requiring heart transplantation. However, there is a shortage of organs available to replace failing hearts in these patients. Organs from other species, such as swine, have been proposed to meet the demand in these situations as they are anatomically similar to human hearts, genetically manipulatable, have short breeding cycles and are readily available. However, the arduous immunologic barriers between cross-species transplantation has limited its immediate use. Heterotopic cardiac transplantation in the intra-abdominal position in a large animal model has been essential in the progression of the field of cardiac transplantation. Our group has over 10 years of experience in cardiac xenotransplantation with pig to baboon models, the longest xenograft of which survived over 900 days, with rejection only after reducing immunosuppression^[@CR1]^. This abdominal model facilitates immunologic monitoring through the period from implantation until graft rejection at reduced cost and complexity compared to orthotopic (life-supporting) models, with the additional advantage that, since the native heart remains in place, rejection of the heterotopic xenograft does not result in primary hemodynamic compromise and/or death. Here, we demonstrate the approach to implantation of a cardiac graft into the intra-abdominal position in a baboon recipient for the study of transplantation and briefly highlight our model's ability to provide insight into not only xenotransplantation but across disciplines. We include details that have provided us with consistent success in this model; performance of the anastomoses, de-airing of the graft, implantation of a long-term telemetry device for invasive graft monitoring, and ideal geometric positioning of the heart and telemetry device in the limited space of the recipient abdomen. We additionally detail surveillance techniques to assess long-term graft function. This heterotopic model, namely that it provides a readily reproducible method for long-term and whole-organ cardiac perfusion without compromising the recipient, should be seen as a standard model for testing iterative improvements in immunosuppression regimens and xenograft genetic manipulations for the further enhancements of cardiac xenotransplantation and allotransplantation at large. While we describe this in the context of transplantation from our extensive experience using this model, considerations are otherwise similar in any other large animal model. As this model uniquely provides *in vivo* assessment of whole organ function without compromising host physiology, it can be used for assessing cardiac physiology across disciplines, where other models have failed or are limited. Materials {#Sec2} ========= Specific pathogen-free (SPF) baboons of either sex weighing 15--30 kg (2--3 years of age) from Oklahoma University of Health Sciences (Norman, OK) were housed in a clean pathogen-free facility and were used as recipients. 6 to 8 week-old genetically modified swine of either sex, with an established genetic backbone known to produce prolonged xenograft survival, alpha 1--3 galactosyltransferase gene knockout (GTKO) and overexpression of human CD46 (hCD46) and thrombomodulin (hTBM), GTKO.hCD46.hTBM, were used as donors (Revivicor Inc., Blacksburg, VA) as our standard donor^[@CR1]^. However, we have also demonstrated success in pigs that additionally express human transgenes for thromboregulation (endothelial protein C receptor, tissue factor pathway inhibitor), complement inhibition (decay accelerating factor), and cellular immune suppression (hCD39, hCD47). SPF baboons were selected for low non-gal antibody titers as previously published^[@CR2]^. Critical materials are listed in Table [1](#Tab1){ref-type="table"} and the immunosuppression regimen has been previously described^[@CR1],[@CR3]--[@CR5]^.Table 1Additional information on critical materials for heterotopic cardiac transplantation: while these are suggestions based on materials we have used, there are likely other suitable alternatives.Heterotopic Cardiac Xenotransplantation-Important MaterialsNameManufactererReference \#9 Fr Cardioplegia Aortic Root CannulaMedtronic (USA)20012Bladder Irrigation Tubing for CardioplegiaBaxter (USA)2C4041Extension IV Tubing Seticumedical (USA)12656-28PlegisolHospira Inc. (USA)409796905Data Sciences International (DSI) Telemetry Device L21DSI (USA)DSI L2110 Fr Hickmann Tunneled Triple Lumen CatheterBARD (USA)606560 Methods {#Sec3} ======= All procedures described here have been approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Maryland School of Medicine. All methods were carried out in accordance with relevant guidelines and regulations. Heterotopic cardiac transplantation technique {#Sec4} --------------------------------------------- The intra-abdominal heterotopic model (HHTx) is a two-anastomosis system utilizing arterial supply from the infrarenal aorta of the recipient baboon to perfuse the coronaries of the donor heart with drainage of the cardiac graft through the donor pulmonary artery remnant anastomosed to the intra-abdominal inferior vena cava of the baboon recipient (Fig. [1](#Fig1){ref-type="fig"}). A list of critical materials is provided in Table [1](#Tab1){ref-type="table"}. A brief description is provided followed by step-by-step instructions for performing HHTx. A supplemental video is also provided (supplementary video [1](#MOESM1){ref-type="media"}).Figure 1Intraabdominal placement of a cardiac xenograft with pressure telemetry monitor in the apex. IVC-inferior vena cava, Ao-Aorta, PA-pulmonary artery, EKG-electrocardiogram leads from telemetry device. Of note, whereas the telemetry device depicted here only has one pressure sensor and it is placed in the left ventricle at the apex, the pressure can be placed in the right atrium, pulmonary artery or aorta as well, depending on which hemodynamic parameters of interest are to be studied. Image Copyright: Tim Phelps JHU/AAMM, 2020. Briefly, the cardiac donor is prepped and draped sterilely, and a midline sternotomy is performed. Pericardium is opened and major vessels are isolated. Silk ties are placed around both superior (SVC) and inferior (IVC) vena cavae. Cold blood cardioplegia is administered through a 9 Fr aortic root canula after ligating the SVC and applying a vascular cross clamp on the aorta. The heart is decompressed by venting the IVC and left atrium or pulmonary vein and cardiectomy is performed. The heart is placed on ice during backtable preparation. The IVC is ligated and pulmonary vein common channel is created and over sewn (Fig. [2](#Fig2){ref-type="fig"}). The cardiac graft is now readily for transplantation into the abdomen.Figure 2Backtable preparation of the heart for transplantation. Ao-ascending aorta, PA-pulmonary artery, Pvv-Pulmonary vein common channel, Cava-superior and inferior vena caval junction. Image Copyright: Tim Phelps JHU/AAMM, 2020. Retroperitoneal exposure of the infrarenal abdominal aorta and IVC of the recipient is performed and proximal and distal control of these vessels are obtained. Aortotomy and cavotomies are placed approximately 2 centimeters (cm) distal to the renal vessel takeoffs. The aorta-aorta anastomosis is carried out before the pulmonic-caval anastomosis (Fig. [3](#Fig3){ref-type="fig"}).Figure 3Aortic and Pulmonic Anastomosis to the Recipient. Abdomen. Image Copyright: Tim Phelps JHU/AAMM, 2020. Aortic and Pulmonic Anastomosis to the Recipient. Abdomen. Image Copyright: Tim Phelps JHU/AAMM, 2020. Step-by-step performance of HHTx transplantation is as follows, with clinical pearls based on observation and experience denoted at relevant points of the procedure. All procedures described here have been approved by the Institutional Animal Care and Use Committee (IACUC) of our residing institution. Here we describe these steps in the context of pig-to-baboon xenotransplantation, however, this can be conducted similarly in baboon heterotopic allotransplantation. ### Procurement of the Cardiac graft {#Sec5} #### Preparation of the cardiac donor {#Sec6} The donor is sedated with 10 milligrams per kilogram (mg/kg) of Ketamine intramuscular (I.M.) and 2 mg/kg xylazine for transfer to the operating room. Clinical pearl*In swine, xylazine is used as an adjunct to ketamine, as many are somewhat refractory to the sedative effects of ketamine. While xylazine is known to cause second degree heart block and hypotension requiring atropine in some cases, this has not been our experience, although the true incidence is not known. Alternatively Ketamine and 0.12 mg/kg IV of Dexmedetomidine or Telazole 1 mg/kg can be used as an adjunct to ketamine*^[@CR1]^. A peripheral intravenous catheter is placed for medication administration. A 24 gauge angiocatheter in an ear or forelimb vein is preferred. Alternatively, a percutaneous femoral venous cannula can be placed if peripheral IV access is difficult to obtain. Additionally, a femoral arterial line is placed in either groin after appropriate sterility is obtained. Induction of anesthesia is performed with isofluorene (1--1.5%). Routinely, both prophylactic amiodarone and lidocaine are given at 2 mg/kg and 1 mg/kg, respectively. A 6--7 mm cuffed endotracheal tube is used for oral tracheal intubation. Anesthesia is maintained with isofluorene, with a goal minimal alveolar concentration (MAC) of 1.0--1.2. Paralytics such as succinylcholine and opiates such as fentanyl are given as needed to maintain paralysis and analgesia, respectively after ensuring appropriate levels of anesthesia. Clinical pearl*In our experience, sevofluorene requires high doses for anesthetic effect in swine and thus the cardio suppressive effects of inhaled anesthetic agents is accentuated. However, there is some evidence that in the setting of experimentally induced MI models, sevoflurane increases the risk of fatal ventricular arrythmias. In regards to intubation, we sometimes elect to place a surgical tracheostomy as the swine anatomy can make oral tracheal intubation difficult and put the swine at risk for laryngospasms*^[@CR6],[@CR7]^. *In terms of antiarrhythmics, lidocaine is well tolerated, however, amiodarone often causes bradycardia and hypotension. This can be mitigated by slow infusion, however, in severe cases this is refractory to most agents except calcium chloride, which can be given as a 100 mcg push. Amiodarone should be given only after hemodynamic stability is assured, an arterial line has already been placed and the surgeon is aware that it will be administered.* Place EKG leads to the skin of the chest for cardiac monitoring, as lateral as possible to remain away from the surgical field. Following intubation and induction of general anesthesia, the surgical field is prepped, with a series of three alternating scrubs with 10% betadine and 70% alcohol solution from the angle of the mandible to the upper abdomen at least 4 cm below the xyphoid process and laterally to the proximal arms. #### Exposure of the donor cardiac graft {#Sec7} The skin is incised from above the sternal notch to below the xyphoid process using a 10-blade or cutting electrocautery. It is deepened through the subcutaneous fat and muscle using coagulating electrocautery. The sternal midline is identified by feeling the sternal edges in the rib interspaces and should be from associated muscle and tissue. Clinical pearl*One must attend to the most superior portion of this incision. There is thick muscle to be cleared from the keel at the manubrium circumferentially and posteriorly.* The xyphoid process may be amputated with electrocautery. The retrosternal space is cleared from cardiac adhesions and pericardium using finger dissection and blunt scissors working superiorly from the subxyphoid space. Electrocautery may be used to take down diaphragmatic attachments. An oscillating sternal saw or thick shears is used divide the sternum at the midline, taking care to protect the right ventricle from injury. Clinical pearl*The manubrium's keel is too thick to perform a midline sternotomy at this portion. Once resistance is met at the superior edge of the sternum, this area should be avoided and the sternum should be incised laterally in one direction or the other as the sternal saw is not able to cut this keel. Thus, soft tissue and muscle should be cleared here along this path as well. However, caution must be taken as the innominate vein is just deep to this area, as damage can cause devastating bleeding.* Adequate hemostasis is ensured of the divided sternal edges utilizing electrocautery and bone wax if preferred. An appropriately sized sternal retractor is placed for exposure and the pericardium is carefully divided. Take care to avoid irritation of the myocardium to prevent cardiac arrythmias. Pericardial sutures are placed with 2-0 silk to create a well, if needed. A plane is created through sharp dissection or electrocautery between the ascending aorta and pulmonary artery to accommodate eventual cross-clamp of the aorta. The superior vena cava is dissected inferior to the azygos vein and tagged with two 2-0 silk ties. Clinical pearl*In swine, the azygos vein is small, posterior, and fragile. During dissection of the superior vena cava (SVC), care must be taken to avoid injuring or avulsing the azygos vein located on the posterior medial surface of the SVC. Alternatively, this can be ligated without consequence with surgical clips or tie.* Similarly, the IVC is isolated and two silk ties are placed on a tension-free tag #### Cardioplegia and arrest of donor cardiac graft {#Sec8} A purse-string or U-stitch of 5-0 Prolene is placed in the ascending aorta for securing of the cardioplegia cannula. This should be at least 2 cm distal to the aortic root. The cardioplegia cannula is placed and secured with a Rummel tourniquet. Clinical pearl*Great care should be taken to prevent puncture of the posterior wall of the aortic root with the aortic cannula by pointing the tip of the needle bevel side anterior and pointed about 15 degrees inferiorly toward the aortic root. Hypotension can prevent proper resistance and distensibility of the aorta, increasing the risk of posterior wall puncture. Similarly, ensuring that systolic pressures are not higher than 120 mmHg prior to cannulation prevents excessive bleeding and reduces the risk of aortic dissection.* 500 un/kg of systemic heparin is administered. Ensure at least three minutes between administration and cross-clamp. Ensure the readiness of cardioplegia tubing and solution, sterile ice, and the entire team for cross-clamp and arrest. In rapid sequence, ligate the cranial superior vena cava, place the aortic cross-clamp, vent the right atrium by dividing the inferior vena cava, and vent the left atrium either through incising the left atrial appendage or the inferior pulmonary veins. Immediately upon venting, begin running cardioplegia and fill the mediastinum with sterile ice slush. Clinical pearl*It is important to prevent distention of the right and left ventricle with venting, as described above, for proper arrest and myocardial protection. Additionally, in order to facilitate optimal cooling of the heart on ice, the pleural spaces can be opened bilaterally allowing drainage of blood and cardioplegia. This prevents excess fluid from accumulating in the mediastinum and inefficient myocardial cooling.* 30 cc/kg of cardioplegia is administered through the cardioplegia cannula (Plegisol with 50 millimoles (mmol) sodium bicarbonate 8.4%, 50 mmol potassium chloride). Ensure that the heart is soft and relaxed with appropriate pressure (20 mmHg) and distention of the aortic root. Clinical pearl*Cardioplegia should be kept on ice. Additionally, great care must be taken to ensure no air is present, as even small amounts of air can cause ischemia. Cardioplegia should be administered through pressure-resistant tubing (Table* [*1*](#Tab1){ref-type="table"} *for our preference) and a pressure bag. To ensure appropriate root pressures an aortic cannula arterial line can be placed, but this can be cumbersome. To a trained surgeon, this can be easily assessed by palpation of the aortic root.* #### Donor cardiectomy {#Sec9} The superior vena cava is ligated and divided. The division of the inferior vena cava is completed from the venting step during cardioplegia administration, if necessary. The aorta is incised distal to the cardioplegia cannula at the junction of the aortic root and arch. The pulmonary artery is incised just proximal to the bifurcation. Cardiectomy is completed by dividing the remaining pulmonary veins along the pericardial reflection. ### Backtable Preparation of the Cardiac Graft {#Sec10} #### Closure of caval orifices {#Sec11} The superior and inferior vena cava are tied with 2-0 silk prior to cardiectomy. However, alternatively this can be done on the backtable. (Fig. [2](#Fig2){ref-type="fig"}). #### Closure of pulmonary vein orifices {#Sec12} A common orifice is created between the pulmonary veins at the level of the left atrium. This common orifice is closed in two layers with a back-and-forth running 5-0 Prolene suture (Fig. [2](#Fig2){ref-type="fig"}). #### Preparation of the aorta and pulmonary artery for anastomosis {#Sec13} The donor aorta and pulmonary should be beveled about 20--30 degrees from the axis parallel to the recipient IVC and abdominal aorta. It should also be an appropriate length to minimize kinking and tension (see Fig. [3](#Fig3){ref-type="fig"}). While this can be approximated on the backtable, it likely will need to be refined once the donor heart is placed in the recipient abdomen just prior to the anastomosis. ### Exposure of the Recipient Abdominal Aorta {#Sec14} #### Preparation of the graft recipient {#Sec15} The recipient is sedated with 10 mg/kg of Ketamine I.M. for transport to the operating room. Preoperatively, a tunneled triple lumen central venous catheter for administration of induction immunotherapy and maintenance immunotherapy, anticoagulation and other medications postoperatively is placed in either internal jugular veins (see Table [1](#Tab1){ref-type="table"} for model number). Our Immunosuppression has been previously described^[@CR1],[@CR3]^. One to two peripheral intravenous catheters for access for medication and fluid administration if needed and the central venous catheter is insufficient. 18--20 gauge catheters in forearm veins are preferred. EKG leads are placed to the skin of the chest for cardiac monitoring, as lateral as possible to remain away from the surgical field. A foley catheter is placed into the bladder for decompression and urine output monitoring Following intubation and induction of general anesthesia, prepare the skin with a series of three alternating scrubs with 10% betadine and 70% alcohol solution from the xyphoid process to distal to the pubic symphysis across the bilateral groins for sterile femoral access. Clinical pearl*We mirror our anesthesia strategy to the swine donor's, as we believe the anesthesia strategy should mirror the cardioprotective approach.* #### Safe entrance into the abdomen {#Sec16} A midline laparotomy is performed from the xyphoid to the pubic symphysis. The skin is incised using a 10 blade and deepened to the fascia using electrocautery. Two forceps are used to elevate fascia and make a small (3 mm) sharp incision through fascia and peritoneum to enter the abdomen. Extend this incision to its full length using a finger or malleable retractor to protect the intra-abdominal contents (bowel) from transmitted electrocautery. Clinical pearl*Great care should be taken to prevent electrocautery injury to the bowel. It is best to use Metzenbaum scissors for initial entry through the fascia into the abdomen. Additionally, the dome of the bladder should be cared for and avoided on the inferior portion of the incision. Despite a foley catheter, its peritoneal reflection can still be present above the pubic symphysis. While we have never encountered this, injuries to the bowel or bladder should be promptly repaired in 2 layers, the inner layer with an absorbable suture such as a 2-0 vicryl and a permanent suture such as a 2-0 silk for the outer layer. Missed bowel injuries can cause fatal abdominal sepsis in an immunosuppressed recipient.* The bowel is protected with wet lap pad or towel and gently retracted cranially using an abdominal retractor. The retroperitoneal space is opened inferior to the kidneys and at least 5 cm of inferior vena cava and abdominal aorta are exposed for accommodation of partially occlusive vascular clamps The inferior vena cava and aorta are encircled with proximal and distal vessel loops, respectively, for proximal and distal control of the vessels. ### Two-Anastomosis Implantation of the Cardiac Graft {#Sec17} #### Aorto-aortic anastomosis {#Sec18} A partially occlusive vascular clamp is placed on the superior portion of the infrarenal abdominal aorta after administration of 300 un/kg of heparin. See special notes for step 4.1.2. The aortotomy is created sharply in a longitudinal fashion with an 11 blade, to a length of approximately 1 centimeter. Adjust this to accommodate the size of the donor ascending aorta with Pott's scissors. Clinical pearl*Take care not to cause a dissection of the abdominal aorta during this incision or injure the lateral or posterior walls during this incision. It is very important to remain partially occlusive on the abdominal aorta during the anastomosis to prevent prolonged totally occlusive clamp times. Additionally, variant anatomy sometimes yields an infrarenal abdominal aorta that is too small to accommodate an anastomosis and a suprarenal clamp must be placed instead. While this should be avoided, sometimes it is not possible but clamp times should be minimized to prevent kidney and spinal cord ischemia.* An end-to-side, donor aorta to recipient infrarenal aorta, anastomosis using a running 5-0 Prolene on an RB-2 needle. Alternatively, the anastomosis can be created with two 5-0 prolene sutures as well. Clinical pearl*The aortic anastomosis is performed prior to the pulmonary-caval anastomosis for technical reasons. We have found that the resultant orientation of the xenograft is more optimal and the sewing of the anastomosis is easier this way. If, when orienting the heart and creating the abdominal aortotomy and cavotomy, it is found to be more suitable to begin with the pulmonary anastomosis, one may do so. However, we have not found this to be the case.* #### Pulmonary artery-caval anastomosis {#Sec19} A partially occlusive vascular clamp to the infrarenal IVC is placed after completion of the aorto-aortic anastomosis. Clinical pearl*It is very important to remain partially occlusive during the anastomosis as prolonged totally occlusive clamp times of the IVC can contribute to bowel edema. See comments of vascular clamping noted in step 4.1.2.* The inferior vena cava is opened sharply in a longitudinal fashion to a length of approximately 1 cm. It is adjusted to accommodate the size of the donor pulmonary artery remnant. The incision can be made using Pott's scissors or an 11 blade. Clinical pearl*Take care not to injure the lateral or posterior walls of the cava during this incision. Additionally, the cavotomy is usually placed slightly superior to the aortotomy's position as to accommodate proper placement of the heart in the abdomen to prevent kinking or tension*. The xenograft is placed in a position that will not cause tension on the aorto-aortic anastomosis and trim the donor pulmonary artery to the shortest possible length to create an end-to-side caval anastomosis in this position. An end-to-side, donor pulmonary artery to recipient IVC, anastomosis is performed using a running 5-0 Prolene on an RB-2 needle. The xenograft can be temporarily flipped laterally over the axis of the anastomosis in order to expose the contralateral edge of the anastomosis if needed. Alternatively, similar to the arterial anastomosis, two 5-0 Prolene anastomosis can be performed, depending on surgeon preference. Partially occlusive vascular clamps can be placed superiorly or inferiorly oriented, depending on exposure, angle and orientation of the cardiac graft (Fig. [3](#Fig3){ref-type="fig"}) ### Reperfusion of the cardiac graft {#Sec20} An 18-gauge angiocath needle is used to access the apex of the left ventricle for de-airing. This site will ultimately be used for implantation of the telemetry pressure probe. Administer systemic sodium bicarbonate 1 mmol/kg, lidocaine 1 mg/kg, and mannitol 0.3 gm/kg in preparation for reperfusion. The IVC clamp is removed first to allow for retrograde filling of the right ventricle, followed by release of the aortic cross clamp. Clinical pearl*Be prepared for electric cardioversion in the case of fibrillation. Hearts of swine are prone to ventricular arrhythmias and is very common upon reperfusion. Prompt unsynchronized internal fibrillation of 10--15 Joules should be initiated.* ### Implantation of the telemetry device {#Sec21} A pocket is formed in the lateral abdominal wall, between the fascia and muscle of the external oblique, to accommodate a telemetry device. The telemetry device is placed in this pocket and the fascia closed over the device, for soft tissue coverage of the device, prior to implanting the pressure probe and EKG leads. The pressure probe is placed in the apex of the left ventricle, preferably at the prior site of de-airing. Secure the probe with a 5-0 Prolene purse string using felt pledgets. Clinical pearl*Notably, some telemetry devices come with two pressure probe channels. The pressure probes can be safely placed in the left or right atrial appendage or ascending aorta, if other parameters are desired.* The first EKG lead to the left ventricle is placed with a simple 5-0 Prolene stitch. The second lead may be free in the peritoneal cavity as a ground. ### Closure of the Recipient Abdomen {#Sec22} The midline fascia is closed with running 0 Prolene suture. Take care to avoid catching the leads in the closure and protect bowel during closure. Close the remaining incision in two layers using absorbable suture (2-0 Vicryl followed by 3-0 vicryl at the skin is our preference). Clean and dry the incision and apply skin glue. Dress the wound as desired for 48 hours. Recover the animal from general anesthesia. Place on systemic anticoagulation (e.g., heparin) 24 hours after surgery, once surgical risk is deemed minimal. The goal ACT is twice the level of baseline or aPTT 60--80. ### Transabdominal Myocardial Biopsies {#Sec23} The recipient is placed under anesthesia as previously described, along with administration of prophylactic antiarrhythmics. The point of maximal cardiac impulse is located and the smallest incision at the skin and underlying fascia, as required to safely perform a myocardial punch biopsy, is performed to expose the heart. Any scar tissue, overlying bowel or omentum is cleared from the heart and a point for biopsy is located. Similar for telemetry pressure lead placement (as described above), a 5-0 Prolene purse string using felt pledgets is performed around the site of planned punch biopsy A full thickness punch biopsy is performed using a 2-3 Fr punch biopsy tool at the center of this purse string The purse string is tied down in a standard fashion and hemostasis is ensured Closure of the abdominal wall is closed as previously described above Long-term evaluation of the cardiac graft {#Sec24} ----------------------------------------- The primary mechanisms for evaluation of results are the following: transabdominal palpation of the xenograft, transabdominal ultrasonography of the xenograft, direct vascular, atrial or ventricular pressure and EKG monitoring utilizing the implanted telemetry device and serial serum laboratory values including complete blood count and troponin. Transabdominal myocardial punch biopsies are performed a maximum of two times over the lifetime of the recipient, at least one month apart, as limited by our local IACUC for histopathologic surveillance (described in detail above). Palpation allows the easy grading by feel of the contractility of the xenograft. A system of 0 to 4+ is utilized, where 0 indicates no contractility and 4+ indicates full contractility (supplementary video 2). On transabdominal ultrasonography, the contractility of the left and right ventricles, the presence or absence of left ventricular thrombus and/or left ventricular hypertrophy may be assessed (supplementary video 2). Left ventricular thrombus (Fig. [4](#Fig4){ref-type="fig"}) is common but may be partially or completely resolved with the administration of a systemic heparin infusion, which is recommended to be performed routinely.Figure 4Left ventricular thrombus in porcine xenograft. Progressive left ventricular hypertrophy may indicate ongoing xenograft rejection and will generally be seen prior to depressed left ventricular function. Elevated and increasing troponin, particularly in conjunction with thrombocytopenia, can increase the suspicion for rejection. The most precise measure, however, is pressure tracings of the telemetry device, which we have previously characterized^[@CR8]^. We have found the most sensitive measure of rejection is an increase in left ventricular end diastolic pressure (LVEDP). Cardiac contractility ceases after the pulse pressure is reduced to \<10 mmHg. We have also previously characterized intracardiac thrombus as an indicator of rejection (Fig. [4](#Fig4){ref-type="fig"})^[@CR4]^. Results and Discussion {#Sec25} ====================== We have extensive experience with this model over the last 10 years. With establishment of this model and consistent long-term survival, we can now take observations made *in vitro* that can be rigorously tested. Additionally, this model can yield clinical insights regarding allotransplantation, immunology and cardiac specific tissue injury. We have been able to characterize and expand CD4 + CD25 + FoxP3+ regulatory (Treg) T-cells and demonstrate their suppressive effects onto xenografts, recipient B and T-cell populations and their potential role in allotransplantation^[@CR9]--[@CR11]^. Additionally, we have shown that Rapamycin, a currently clinically approved immunosuppressive drug in allotransplantation, promotes enrichment of functional Treg cells with immunoregulatory properties^[@CR12]^. Lastly, we have extensively characterized transgenic pigs for the use in cardiac xenotransplantation and identified early markers for rejection that are applicable to not only cardiac xenotransplantation but also as a generic marker of tissue injury relevant to other fields of study and are graft specific for this model^[@CR13]--[@CR16]^. Lastly, we have extensively studied co-stimulation blockade and B-cell depletion's role in xenotransplantation, which has transformed the field and extended survival not only in cardiac, but also kidney, liver and islet cell xenotransplantation^[@CR17]--[@CR19]^. There are several critical steps in this procedure. Smooth procurement of the xenograft, with adequate myocardial protection, is the first critical step. Anastomosis in the recipient abdomen in a way that avoids narrowing or kinking of either of the two anastomoses, but particularly the pulmonary artery-caval anastomosis, is the next. Finally, maintenance of xenograft contractility in a normal sinus rhythm is necessary for coronary perfusion and ultimately xenograft survival. During procurement of the xenograft, the cardioplegia must be administered under pressure. The heart must be adequately vented and the output should become clear. The distention of the aortic root can be assessed manually, as can the distention of the xenograft itself. Generally, if there is difficulty running the cardioplegia or distention of the graft or root, extending the incision in the inferior vena cava, left atrial appendage, and/or pulmonary veins will aid in venting and resolve this difficulty. Once transferred into the recipient abdomen, care in the performance of the anastomoses, such that the vascular lumens to not become narrowed, is critical. The geometry of the xenograft can be assessed during implantation. The more common error in this step is creating a pulmonary-artery caval connection in which the pulmonary artery remnant is too long, and allows the xenograft to fold on itself and kink the anastomosis. Finally, utilization of both chemical and electrical means to establish normal sinus rhythm early following implantation is imperative. The limitations of this model are in its non-life-supporting nature. The native recipient heart supplies complete cardiac output, hemodynamic support and end-organ perfusion. The intra-abdominal xenograft does not contribute to this support. Rather, it remains in continuity with the recipient's vascular supply, allowing continuous perfusion of the xenograft with recipient blood and exposure to the full recipient immune response (Fig. [5](#Fig5){ref-type="fig"}).Figure 5Perfusion of Blood Through the Heterotopic Cardiac Graft: The graft is perfused via its coronary vessels during graft diastole and aortic valve cooptation, supplied from the recipient aorta. Perfused blood from the graft is then drained into the right atrium via the coronary sinus. Solid arrows denote recipient native blood flow. Dotted arrows represent graft blood flow in systole (outward from graft major vessels) and diastole (into graft atria/ventricles). However, this non-life supporting model allows the evaluation of cardiac transplant rejection, testing of different porcine genetic modifications and immunosuppressive regimens, without life-threatening physiologic perturbations of the recipient. Life-sustaining studies require the recipient to be placed on cardiopulmonary bypass just to undertake the transplantation. Additionally, these recipients require critical care postoperatively, as perturbations of the cardiac graft can compromise supporting perfusion for all vital organs. This model requires no critical care and can be immediately transferred to a non-intensive setting similar to preoperative housing. The incidence of cardiovascular disease causing end-stage heart failure requiring transplantation is growing^[@CR20]^. Genetically modified porcine organs that have been tested in this model and that are now being used in orthotopic studies are increasingly meeting preclinical efficacy requirements for their eventual use^[@CR21]--[@CR23]^. While survival has been obtained for up to 6 months, there is much still to learn regarding the physiologic and immunologic mismatch between swine and baboon (and human for that matter) as these hearts eventually still succumb to end-stage failure by an unknown mechanism^[@CR24]--[@CR27]^. Additionally, porcine endogenous retrovirus (PERV) knockout swine have been produced out of concern for the theoretical transmission of PERV (although there has been no evidence of human transmission and PervC null animals are being produced regularly without gene editing)^[@CR28]--[@CR31]^. All things considered, this model continues to be relevant for demonstrating pre-clinical efficacy, safety and testing iterative changes in genetic constructs of porcine xenografts. Further successes in reproducing this model in other institutions may enable continued enhancement of porcine cardiac xenotransplantation for the eventual use in humans, but also push the boundaries of discovery as has been limited by other disease models across disciplines. For example, knowledge gained by insights in xenotransplantation, an arduous immunological barrier in excess of human allotransplantation, can shed light onto strategies to abrogate problems faced by dysregulation in immunity such as antibody mediated rejection or graft versus host disease. Additionally, as this model uniquely provides *in vivo* assessment of whole organ function without compromising host physiology, it serves as a way to assess cardiac physiology beyond the limitations of what current models can provide^[@CR32]--[@CR33]^. Supplementary information ========================= {#Sec26} Supplementary Video 1. Supplementary Video 2. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-66430-x. M.M.M. designed, supervised and conducted the experiments; interpreted data and wrote the manuscript. C.E.G., A.K.S. and T.Z. conducted the experiments, analyzed data and contributed to writing of manuscript and assembling figures and tables. L.D. contributed to writing the manuscript. K.H. and P.C.C. contributed in surgical procedures. F.S. and B.L. contributed to day-to-day animal care and helped with surgical care. D.A. provided the genetically engineered pigs. I.T. directed cared for the animals in this study and helped review the manuscript. A.H. conducted many of the assays in this study and helped review the manuscript. All authors critiqued and the manuscript and assembled the video. This study was supported by funding from United Therapeutics, Inc and the NIAID, NIH grant 5U19090959-10. David Ayares is an employee of Revivicor, Inc. There are no conflicts of interest to disclose by any other authors.