Datasets:

CShorten

/

1000-CORD19-Papers-Text

like 2

As an active sensor in the brain, microglia respond to even minor stimuli; however, different types of stimulation may also lead to different actions of microglia and thus be either harmful or beneficial to neuronal survival. In a neonatal mouse MPTP-induced brain injury model, microglia activated by systemic administration of LPS were shown to be neuroprotective. In contrast to the MPTP model, LPS-activated microglia in neonatal mice receiving a stereotaxic injection of ethanol into the striatum were shown to be neurotoxic, and systemic LPS administration in the ethanol-injury model caused a marked increase both in the volume of necrotic lesions and in the number of degenerating neurons in the striatum [168] . Even with the same stimuli, the degree can also determine microglial release of toxic versus protective effectors [169] ; neurotoxic cytokines and ROS were released from microglia only in response to mild neuronal injuries, while trophic microglial effectors such as BDNF and GDNF were up-regulated in response to all degrees of neuronal injury [169] . Additionally, different types of pain resulted in differing activations of microglia [170] .

So far, what we know is that not all microglia respond in the same way, even to the same stimulus, and microglial function is tailored in a context-specific manner [171] . Numerous elements are involved in this context; most likely there are many more beyond what we have discussed here. Identifying these elements and clarifying their interactions or crosstalk with microglia is essential before we are able to design a strategy to control inflammation through the manipulation of microglia. The simple therapy of inhibiting all microglia without differentiating their function in a context-dependent manner surely should be abandoned.

It has been long recognized that the birth of new neurons within the postnatal brain continues throughout life and remains as a potential source of replacement cells in the CNS for the treatment of disease. The microenvironment or the niche in which neural progenitor cells live critically influences the process of neurogenesis, which spans several steps including the proliferation of stem or progenitor cells; the survival of immature or mature neurons; the migration of new neuroblasts to their appropriate locations; and the differentiation of neuroblasts to a neuronal phenotype and the construction of synaptic connectivity [172] . As an important component of the brain microenvironment and due to their invariant participation in most pathological processes in the CNS, microglia are increasingly implicated as a potential non-neural regulator of neurogenesis, as demonstrated by circumstantial evidence [144, 172] . However, just as in the debate over the neuroprotective or neurotoxic nature of microglial activation, whether microglia support or damage the survival and development of neural progenitor cells also remains controversial. On one hand, microglia were shown to play instructive roles during postnatal neurogenesis in the neurogenic niche either by influencing the differentiation of stem cells toward a neuronal phenotype or by directing their migration [144, [173] [174] [175] . On the other hand, multiple studies have demonstrated the deleterious effect of microglial activation on neurogenesis [176, 177] and the effective restoration of neurogenesis though the blockade of microglial activation.

In the two situations of neurogenesis and neuronal survival, similar factors are shared, leading microglia to take supportive or detrimental roles. Among these factors, the most prominent is the microglial activation phenotype that is associated with different cytokine profiles. When acutely activated by either LPS or injury, microglia that release the pro-inflammatory cytokines IL-6, TNF-α or IL-1β usually down-regulate the differentiation or proliferation of neural stem cells or induce the aberrant migration of newborn neurons [178] . This group of inflammatory cytokines has been proven to inhibit neurogenesis [176, 177, 179] ; conversely, blocking antibodies to these pro-inflammatory cytokines (such as IL-6 [177] ) or the use of monocycline to mitigate the microglial activation simply restores neurogenesis [176] . In contrast, microglia that are activated by anti-inflammatory cytokines such as IL-4 or TGF-β increase neurogenesis in vitro or the differentiation of neural stem cells (NSCs) in vivo [180, 181] . Neurotrophins, such as IGF-1, were identified [181] in anti-inflammatory cytokineactivated microglia and were proposed to be one of the mechanisms underlying this pro-neurogenic activity of microglia [182, 183] . However, just like the dual roles in neuroprotection, whether a specific cytokine-activated microglial cell will take a pro-or anti-neurogenic role is also context-dependent. For example, microglial cells activated by IFN-γ, a pro-inflammatory cytokine can be neurotoxic or supportive of neurogenesis, depending on the concentration of IFN-γ [184] . TGF-β, which is considered to be beneficial to neurogenesis, can actually exert a negative influence on neurogenesis when it is chronically produced in the aged brain [185] . Additionally, if other cytokines exist in the same niche simultaneously, the outcome will be determined by the balance among the various cytokines; some authors have concluded that activated microglia are not pro-or anti-neurogenic per se, but the balance between pro-and anti-inflammatory secreted molecules influences the final effect of microglial activation [172, 180] . However, in which situations the microglia will release pro-or anti-inflammatory cytokines is complicated and is affected by multiple factors such as the injury type, the phase of disease or inflammation, and crosstalk with other regulating components, including neural precursors; this is similar to the question of whether microglia will be neuroprotective or neurotoxic. Most likely the same inflammatory scenario that induces neurodegeneration would also inhibit neurogenesis, while a situation that favors neuronal survival would also support neurogenesis. Interestingly, even in a high-inflammation environment, such as two days after a Trimethyltin-induced acute injury in the hippocampus, significant neurogenesis can be detected [186, 187] , suggesting a complicated system of neurogenesis regulation beyond the inflammation scenario.

Cumulative studies have found an age-related decline in neurogenesis, both in the aged adult and in the diseased brain. Because aging may contribute to microglial dysfunction and neurotoxicity, as we discussed previously in this review, one could assume that microglial dysfunction may also be involved in the downregulation of neurogenesis in the aged or diseased brain [188, 189] . Even though very few studies have focused on the effect of microglial dysfunction on neurogenesis, we can still find a clue from Zhu's study that the difference in microglia function patterns between the immature and juvenile brain might be related to a decrease in neurogenesis in the juvenile brain [190] ; however, stronger evidence from the direct comparison of microglia-associated neurogenesis between aged and young brains is needed to support this view.

Another important element regulating the activities of microglia is the T cell, which comes from the peripheral adaptive immune system and enters the CNS by extravasating across the endothelium of the choroid plexus into the cerebrospinal fluid [191] . The interaction of T cells with microglia in the injured spinal cord correlates with enhanced neuronal survival [184] , and rapidly recruited T cells in the middle cerebral artery obstruction (MCAO) model increased hippocampal and cortical neurogenesis by modulating the microglial response and through the production of IGF in the sub-acute phase [192] . Hippocampal neurogenesis was associated with the recruitment of T cells and microglial activation. Immune-deficient mice show impaired neurogenesis in the hippocampus, but this deficiency was attenuated and neurogenesis boosted by T cells recognizing a specific CNS antigen [193] . The cellular source of IFN-γ and IL-4 in vivo is likely to be T cells, therefore it is reasonable to assume that the T cell-mediated immune response is an integral part of the regulation of microglial phenotype or function, and thus can influence neuronal survival or neurogenesis directly or indirectly.

From an increasing number of studies of diverse microglial activity in different experimental sets, we are beginning to appreciate the heterogeneity of microglial functions that have either beneficial or detrimental roles in specific physiological or pathological environments. Whether microglia are committed to one function from the very beginning or if there is any conversion between different phenotypes remains elusive and the factors that initiate this commitment or promote its conversion are far from being clarified. Due to the invariant critical participation of microglia in most diseases, ongoing research to uncover these questions is warranted; before we are sure about the answer, any potential strategies targeting microglia to manipulate inflammation and modify a disease course are unrealistic. Diversity of Salmonella spp. serovars isolated from the intestines of water buffalo calves with gastroenteritis BACKGROUND: Salmonellosis in water buffalo (Bubalus bubalis) calves is a widespread disease characterized by severe gastrointestinal lesions, profuse diarrhea and severe dehydration, occasionally exhibiting a systemic course. Several Salmonella serovars seem to be able to infect water buffalo, but Salmonella isolates collected from this animal species have been poorly characterized. In the present study, the prevalence of Salmonella spp. in water buffalo calves affected by lethal gastroenteritis was assessed, and a polyphasic characterization of isolated strains of S. Typhimurium was performed. RESULTS: The microbiological analysis of the intestinal contents obtained from 248 water buffalo calves affected by lethal gastroenteritis exhibited a significant prevalence of Salmonella spp. (25%), characterized by different serovars, most frequently Typhimurium (21%), Muenster (11%), and Give (11%). The 13 S. Typhimurium isolates were all associated with enterocolitis characterized by severe damage of the intestine, and only sporadically isolated with another possible causative agent responsible for gastroenteritis, such as Cryptosporidium spp., Rotavirus or Clostridium perfringens. Other Salmonella isolates were mostly isolated from minor intestinal lesions, and often (78% of cases) isolated with other microorganisms, mainly toxinogenic Escherichia coli (35%), Cryptosporidium spp. (20%) and Rotavirus (10%). The S. Typhimurium strains were characterized by phage typing and further genotyped by polymerase chain reaction (PCR) detection of 24 virulence genes. The isolates exhibited nine different phage types and 10 different genetic profiles. Three monophasic S. Typhimurium (B:4,12:i:-) isolates were also found and characterized, displaying three different phage types and three different virulotypes. The molecular characterization was extended to the 7 S. Muenster and 7 S. Give isolates collected, indicating the existence of different virulotypes also within these serovars. Three representative strains of S. Typhimurium were tested in vivo in a mouse model of mixed infection. The most pathogenic strain was characterized by a high number of virulence factors and the presence of the locus agfA, coding for a thin aggregative fimbria. CONCLUSIONS: These results provide evidence that Salmonella is frequently associated with gastroenteritis in water buffalo calves, particularly S. Typhimurium. Moreover, the variety in the number and distribution of different virulence markers among the collected S. Typhimurium strains suggests that within this serovar there are different pathotypes potentially responsible for different clinical syndromes. Salmonella spp. found in water buffalo (Bubalus bubalis) herds are a matter of concern since they are responsible for serious economic losses in livestock and are a zoonotic agent responsible for foodborne illness [1] . As for bovine calves, Salmonella-induced diseases in water buffalo calves are characterized by severe gastrointestinal lesions, profuse diarrhea, and severe dehydration [1] . Acute salmonellosis generally induces diarrhea, mucous at first, later becoming bloody and fibrinous, often containing epithelial casts. Ingestion is the main route of infection, although it can also occur through the mucosa of the upper respiratory tract and conjunctiva. The major source of infection in the herd is represented by asymptomatic older animals shedding heavy loads of bacteria through feces. Other sources of infection are contaminated forages and water, as well as rodents, wild winged animals, insects and man [1, 2] . The disease can also cause sudden death without symptoms. Occasionally, the infection is systemic, affecting joints, lungs and/or the central nervous system (CNS) [1] . Moreover, several Salmonella serovars seem to be able to infect water buffalo, mainly affecting 1-12 week old calves, even though reports on salmonellosis in B. bubalis are scarce [1, 3] .

Water buffalo calves are more frequently affected by gastroenteritis than bovine calves, with mortality rates as high as 70% in water buffalo species vs. 50% in bovine [1, 4] . This difference might be due to a greater susceptibility of water buffalo to gastroenteric pathogens, although it also may reflect the lack of appropriate management practices for this animal species. Therefore, water buffalo represents a suitable model to study causative agents of gastroenteritis. In water buffalo, S. enterica serovar Typhimurium can induce a variety of clinical syndromes with different anatomopathological lesions [1, 3] . The severity of the disease can depend on several factors, including host-pathogen interactions, which is highly influenced by the route of infection, the infectious dose, natural or acquired host resistance factors, and the possible presence of other pathogens. Moreover, specific Salmonella virulence factors, frequently located on Salmonella pathogenicity islands (SPIs), prophage regions or virulence plasmids, play a key role in the pathogenesis of the gastroenteritis [5] .

The current study investigated the intestinal contents collected from 248 water buffalo calves affected by gastroenteritis with lethal outcome to: (i) evaluate the prevalence of Salmonella spp., and (ii) perform a polyphasic characterization of the collected isolates of S. Typhimurium.

Salmonella spp. were isolated from 25% of the intestinal contents collected from 248 water buffalo calves affected by gastroenteritis with lethal outcome. Positive samples were detected in subjects bred in 37 of 58 farms (interherd prevalence, 64%). The S. enterica serovars most frequently isolated were Typhimurium (n=13), Muenster (n=7) and Give (n=7). Other recovered serovars were: Derby (n=5), 4 Bovismorbificans (n=4), Newport (n=4), monophasic S. Typhimurium (B:4,12:i:-; n=3), Blockley (n=2), Meleagridis (n=2), Umbilo (n=2), Altona (n=1), Anatum (n=1), Bredeney (n=1), Enterica (−;i;1,2; n=1), Gaminara (n=1), Haardt (n=1), Hadar (n=1), Infantis (n=1), Isangi (n=1), Kottbus (n=1), London (n=1), Muenchen (n=1), and S.II:41;z;1,5 (n=1). Phage-typing of the S. Typhimurium and monophasic Typhimurium strains (Table 1 ) indicated a variable distribution of phage types among strains with nine different phage types of 13 Typhimurium strains, and three different phage types out of three monophasic Typhimurium strains.

This study reports a significant prevalence of Salmonella spp. (25%) in diarrheic water buffalo calves, that are more relevant than those reported in previous studies (11 and 0.8%) [3, 6] . Moreover, in contrast with bovine species where salmonellosis results primarily associated with serovars Dublin and Typhimurium [5] , the extremely variable distribution of the observed serovars confirms the absence of a serovar specifically adapted to water buffalo, as previously suggested [1] . These data provide therefore evidence that Salmonella, particularly S. Typhimurium, can be potentially considered an important pathogen for this animal species. The definitive phage type 104 (DT104), which has often been associated with multiple-antibiotic-resistant strains with ascertained zoonotic potential and, in many countries, has increased over the past two decades [5] , does not seem to be widely spread in water buffalo. Three monophasic S. Typhimurium (B:4,12:i:-) isolates were also found that are S. Typhimurium lacking phase two flagellar antigens that have a rapid emergence and dissemination in food animals, companion animals, and humans. More significantly, the public health risk posed by these emerging monophasic S. Typhimurium strains is considered comparable to that of other epidemic S. Typhimurium [7] .

The diagnostic investigation indicated that non-Typhimurium Salmonella isolates were detected with at least another potential pathogen in 78% of cases ( Figure 1A ). In 35% of cases Salmonella was linked with pathogenic Escherichia coli that were characterized for the presence of virulence factors. Other frequent associations were found with Cryptosporidium spp. (20%) and Rotavirus (10%) ( Figure 1A) . Remarkably, S. Typhimurium was never associated with pathogenic E. coli, while it was isolated sporadically with Clostridium perfringens (strain #82280), Rotavirus (strain #107025), and Cryptosporidium spp. (strain #112) ( Figure 1B) . The presence of more pathogens in the same subject might suggest that, as for other animal species [5] , diarrhea in water buffalo calves can be characterized by a multifactorial etiology. Data from necroscopic examinations of tissues indicated that the lesions caused by S. Typhimurium were characterized by severe damage of the intestine, ranging from congestive to necrotic-ulcerative enterocolitis. In particular, the strains isolated from animals exhibiting the most severe lesions were #16, #92, #233, and #83528. Among these strains, the two DT104 strains were also found, thus supporting the pathogenic role of this phage type. The other Salmonella serovars were instead isolated from subjects exhibiting a variety of different lesions, mostly minor lesions confined to the jejunum, and often (78% of cases) associated with other pathogens. Similarly, the monophasic S. Typhimurium strains were detected either with Rotavirus (strain #154) or st-positive E. coli (strains #175 and #188). These data confirm the pathogenic potential of the serovar Typhimurium for water buffalo calves. On the other hand, the scarcity of observed lesions and the frequent presence of more than one microorganism in the same subject hamper a clear understanding of the potential pathogenic role of the non-Typhimurium Salmonella serovars included in this study.

S. Typhimurium and monophasic S. Typhimurium strains were further characterized by the molecular detection of 24 genes coding for virulence factors. The genetic characterization (Table 2) included five loci (avrA, ssaQ, mgtC, siiD, and sopB) located on SPI 1-5, respectively [8] , eight loci (gipA, gtgB, sopE, sodC1, gtgE, gogB, sspH1, and sspH2) of prophage origin [9] [10] [11] [12] [13] , the gene spvC, located on a virulence plasmid [12] , and nine genes (stfE, safC, csgA, ipfD, bcfC, stbD, pefA, fimA, and agfA) coding for bacterial fimbriae, involved in surface adhesion and gut colonization [5] . As a positive control for the PCR assay, amplification of the chromosomal gene invA was carried out for each strain. All the S. Typhimurium and monophasic Typhimurium isolates displayed the presence of avrA, ssaQ, mgtC, siiD, sopB, sspH2, stfE, ipfD, bcfC, stbD, and fimA genes, and the absence of the sopE gene. Other loci were variably distributed among the strains, with frequency values ranging from 38-92% (Table 1) . On the basis of the presence or absence of the 24 loci included in the study, the 13 strains of S. Typhimurium were subdivided into 10 different genotypes (Table 1) ; however, the isolates with identical genotype displayed different phage types suggesting the presence of 13 different strains. Interestingly, the three monophasic S. Typhimurium strains exhibited three different genotypes (Table 1) .

The following loci: invA, sspH2, stfE, ipfD, bcfC, stbD, fimA, avrA, ssaQ, mgtC, siiD, sopB were present in all the strains; the sopE gene was not found in any of these strains. b NT = not typeable.

The 24 loci-genetic characterization was also extended to the S. Muenster and S. Give isolates to investigate their pathogenic potential because of their large presence in water buffalo calves. In addition they have already been reported to cause saepticemic salmonellosis in cattle and calves [14, 15] . The molecular results (Table 3) indicated that the loci invA, safC, bcfC, fimA and ssaQ were present in all the strains, the genes gipA, gogB, sspH2, sodC1, gtgE, spvC, stfE, ipfD and pefA were not found in any of these isolates, while the remaining loci were variably distributed, with frequency values ranging from 14-86%. In particular, the prophage genes were scarcely present (2 loci in the Muenster serovar, 1 locus in the Give serovar), the plasmidic spvC locus was absent in all the analyzed isolates, while the fimbrial genes and the SPI 1-5 genetic markers were discretely represented (6 loci for the former genes in both serovars, 5 and 4 loci for the latter genes in the serovar Muenster and Give, respectively). Moreover, the molecular profiles allowed to identify 6 different genotypes out of the 7 S. Muenster isolates, and 5 different genotypes out of the 7 S. Give isolates (Table 3) .

Our data confirm the high variability of the Typhimurium serovar [9, 10] , mostly related to virulence factors, and highlight the high discriminating potential of the genotyping technique performed. Our data also suggest that monophasic Typhimurium strains are likely to possess a similarly high degree of genetic variability, particularly linked to virulence markers. Moreover, the presence of virulence markers in the isolated strains of monophasic S. Typhimurium, S. Muenster and S. Give could further support their pathogenic potential. The products of the genes included in the virulotyping assay performed here are known to be important during different stages of infection (Table 2) . However, the distribution of these factors among the tested strains highlights the complexity and the variety of potential mechanisms used by Salmonella to induce disease in the host.

The avrA, ssaQ, mgtC, siiD, and sopB genes are genetic markers for the presence of the SPI 1-5 in all S. Typhimurium strains tested, although their presence does not necessarily implicate the presence of the entire SPI. SPIs are clusters of genes on the chromosome, likely to be horizontally acquired, and variably associated with enhanced invasion and intracellular survival within both phagocytic and non-phagocytic cells. In particular, SPI-5 has been largely associated with the ability to produce enteritis [5] . The S. Typhimurium strains included in this study all displayed the presence of the investigated SPI markers. Interestingly, these loci appeared widely distributed also among the serovars Muenster and Give. The sopE gene is known to favor the entry of Salmonella into host cells and its presence has been correlated with disease in humans [16] and with the epidemic potential of S. Typhimurium strains in cattle [17] . This gene was absent in all the S. Typhimurium strains included in the present study, while was present in all the S. Muenster strains analyzed.

The pefA (plasmid encoded fimbria), agfA (aggregative fimbria A) and spvC (Salmonella plasmid of virulence gene C) genes are all located on plasmids [18] . Five S. Typhimurium isolates tested in the current study possessed both pefA and spvC, two isolates were positive for only spvC, and three isolates were positive for only agfA (Table 1) . These results confirm the presence of more than one virulence plasmid among S. Typhimurium strains isolated from diarrheic water buffalo calves, and suggest horizontal exchange of virulence factors. However, the loci pefA and spvC were absent in all the monophasic S. Typhimurium, S. Muenster and S. Give strains tested. Prophage genes are known to account for most of the variability of closely-related S. Typhimurium strains. Moreover, lysogenic bacteriophages promote changes in the composition of genomic DNA often altering the phenotype of the host [9, 10] . The prophage virulence genes included in this study exhibited a variable distribution among the isolates tested, thus suggesting synergistic and/or redundant effects of these loci on the pathogenicity of Salmonella, likely contributing to the ACTGCGAAAGATGCCACAGA phenotypic variability of this pathogen. These loci were mostly present in S. Typhimurium and monophasic S. Typhimurium rather than in S. Muenster and S. Give isolates. Fimbrial genes appeared widely distributed among all the serovars tested, particularly in S. Typhimurium strains, with frequency values ≥92%, except for the plasmid-borne pefA and agfA genes (with frequency values of 38% and 54%, respectively). These data are consistent with the essential functions of adhesion factors for the attachment and internalization processes that occur during pathogenesis.

To better characterize in vivo virulence, three strains representative of all S. Typhimurium isolates were chosen to perform mixed infections in mice. Animal experiments included the two strains exhibiting the highest and the lowest number of virulence factors (strains #92 and #112, respectively), and strain #16, carrying the same virulotype as strain #92, but that does not harbor the agfA locus (Table 1 ). In the competition assay, strain #92 outcompeted both strains #112 and #16 (CI 0.004; P<0.001, and CI 0.031; P<0.001, respectively). These results were confirmed in a gastrointestinal mouse model of infection, which better resembles the clinical form of salmonellosis in livestock. Using oral inoculation, in the competition assay, again strain #92 outcompeted both strains #112 and #16 (CI 0.009; P<0.001, and CI 0.186; P<0.01, respectively). Our data indicate that among those strains included in the experiment, strain #92 was the most virulent in mice. These competition assays in mice suggest a key role of the agfA gene coding for a thin aggregative fimbria involved in the colonization of host intestinal epithelial cells by attachment to glycoprotein or glycolipid receptors on epithelial cell surfaces. Indeed, the strain which was more virulent in in vivo experiments was characterized by a high number of virulence factors and by the presence of the agfA locus. Moreover, it was isolated from one of the subjects with necrotic-ulcerative enterocolitis.

The presence of this type of fimbria has been reported in clinical human and animal isolates of Salmonella + + ----+ ---2 15228 -+ --------3 66761 -+ --------3 72827 -

Freq. a The following loci: invA, safC, bcfC, fimA and ssaQ were present in all the strains; the genes gipA, gogB, sspH2, sodC1, gtgE, spvC, stfE, ipfD and pefA were not found in any of these strains. [19, 20] . The data presented here suggest that agfA might increase bacterial pathogenicity. Nevertheless, we cannot reject the hypothesis that the mouse model chosen for in vivo experiments could have influenced the virulence phenotype of the tested strains originally isolated from water buffalo calves. Therefore, future studies will be necessary to exclude the possibility that the phenotypic differences observed among the tested Salmonellae are dependent on the animal model or on other virulence factors not included in this study. However, in vivo experiments carried out in mouse models represent a good preliminary source of information on the expression of traits associated with pathogenicity of Salmonella in mammalian species.

This study showed a significant (25%) prevalence of Salmonella spp. in water buffalo calves affected by gastroenteritis with lethal outcome. However, our results did not indicate the existence of a Salmonella serovar specifically adapted to water buffalo and highlighted that S. Typhimurium is the most frequently found serovar. The molecular and phenotypic characterization of the S. Typhimurium isolates provided evidence that within this serovar there are different pathotypes potentially responsible for different clinical syndromes, therefore requiring prophylaxis protocols including the use of specific vaccines for the effective control of salmonellosis in water buffalo calves and possible contamination of the food chain.

This study was carried out in the Campania region, Southern Italy, during 2008-2009, using samples taken from 248 water buffalo calves bred in 58 different farms. The animals were aged between 1-12 weeks old and were all affected by gastroenteritis with lethal outcome. During necropsy, the intestinal lesions were evaluated and the intestinal content of the involved sections was collected and tested for the presence of Salmonella spp. In addition, the presence of E. coli, Eimeria spp., Cryptosporidium spp., Giardia spp., Coronavirus, Rotavirus, and C. perfringens were also determined to investigate their association with Salmonella spp. The isolation of Salmonella spp. was performed according to ISO 6579:2002 [21] . The isolated Salmonella spp. were serotyped according to the Kaufmann-White scheme [22] . Phage-typing of the isolated S. Typhimurium strains was performed by the Italian National Reference Centre for Salmonellosis (Istituto Zooprofilattico Sperimentale delle Venezie).

The presence of Rotavirus and Coronavirus was detected by polymerase chain reaction (PCR) amplification [23, 24] .

Cryptosporidium spp. and Giardia spp. antigens were detected by chromatographic immunoassay (Oxoid, Basingstoke, UK). The presence of Eimeria spp. was examined by flotation technique using saturated saline [25] . E. coli and C. perfringens were isolated according to the protocol reported by Quinn et al. [2] . E. coli hemolytic activity was evaluated by growing colonies on blood agar base, while virulence factors (lt-heat-labile toxin, st-heatstable toxin, stx1-Shiga toxin 1, stx2-Shiga-toxin 2, eaeintimin, cnf-cytotoxic necrotizing factor, and cdt-cytolethal distending toxin) were detected by molecular assays, as previously reported [26] [27] [28] .

Bacterial DNA was extracted from 1 mL of overnight cultures using Chelex 100 Resin (BioRad, Hercules, CA) and used as the template for the PCR detection of genes listed in Table 2 , as described previously [8] [9] [10] [11] [12] [13] 18] . The primers used to amplify the genes sspH1, sspH2, ssaQ, sopB, siiD, stfE, safC, csgA, ipfD, bcfC, stbD, and fimA were designed using the Primer3 software (version 0.4.0; http:// frodo.wi.mit.edu/), and PCR was performed in a final volume of 25 μL containing HotStar Taq Master Mix (Qiagen, Valencia, CA) 1×, 0.4 μM each primer and 1 μL of extracted DNA. The thermal profile included an initial denaturation step at 95°C for 15 min, followed by 35 cycles at 95°C for 30 s, 58°C for 30 s, and 72°C for 1 min, and a final extension step at 72°C for 5 min. Amplification products were visualized under ultraviolet (UV) light after electrophoresis on 3% agarose gels and staining with SYBRsafe (Invitrogen, Carlsbad, CA).

Groups of five age matched (8-10 weeks old) female BALB/c mice used in this study were purchased from Charles River (Calco, Italy). Three strains (S. Typhimurium #16, S. Typhimurium #92, S. Typhimurium #12), representative of the 13 genotypically characterized S. Typhimurium isolates, were selected for an in vivo analysis of virulence by using the Competitive Index (CI) resulting from mixed infections [29] . In particular, two strains were selected that exhibited the highest and lowest number of virulence factors (strains #92 and #112, respectively), and strain #16, carrying the same virulotype as strain #92, but without the locus agfA (Table 1) .

Bacteria were grown overnight at 37°C in Brain Heart Infusion medium (Oxoid, Basingstoke, UK), washed, and diluted in sterile saline. Cultures were alternatively combined in a mixture of equivalent numbers (1:1 ratio) of two of the three selected strains (input). Mice were inoculated intraperitoneally (IP) with a dose of 2×10 4 bacteria or received 20 mg of streptomycin orally (200 μL of sterile solution or sterile saline) 24 h prior of being intragastrically administered with 2×10 7 bacteria. The number of colony-forming units (CFU) contained in the inocula were confirmed by plating serial dilutions and counting colony growth. At 4 (IP) or 7 (os) days after infection, mice were sacrificed, spleens were aseptically removed, and bacteria were counted by plating serial dilutions (output). The ratio of two strains in the input and in the output was evaluated by picking and transferring 200 colonies on selective plates. Antibiotics used were streptomycin and sulfonamide, for which strain 92 and strains 16 or 112 were naturally resistant. The CI was calculated using the formula: CI = output (strain A/strain B)/inoculum (strain A/strain B). Statistical differences between outputs and inputs were determined by Student's t test. All animal handling and sampling procedures were performed under the conditions of the local ethics committee meeting the requirements of Italian legislation. Severe Childhood Malaria Syndromes Defined by Plasma Proteome Profiles BACKGROUND: Cerebral malaria (CM) and severe malarial anemia (SMA) are the most serious life-threatening clinical syndromes of Plasmodium falciparum infection in childhood. Therefore it is important to understand the pathology underlying the development of CM and SMA, as opposed to uncomplicated malaria (UM). Different host responses to infection are likely to be reflected in plasma proteome-patterns that associate with clinical status and therefore provide indicators of the pathogenesis of these syndromes. METHODS AND FINDINGS: Plasma and comprehensive clinical data for discovery and validation cohorts were obtained as part of a prospective case-control study of severe childhood malaria at the main tertiary hospital of the city of Ibadan, an urban and densely populated holoendemic malaria area in Nigeria. A total of 946 children participated in this study. Plasma was subjected to high-throughput proteomic profiling. Statistical pattern-recognition methods were used to find proteome-patterns that defined disease groups. Plasma proteome-patterns accurately distinguished children with CM and with SMA from those with UM, and from healthy or severely ill malaria-negative children. CONCLUSIONS: We report that an accurate definition of the major childhood malaria syndromes can be achieved using plasma proteome-patterns. Our proteomic data can be exploited to understand the pathogenesis of the different childhood severe malaria syndromes. Human malaria caused by Plasmodium falciparum has an estimated annual global disease burden of 300 million clinical episodes, leading to one million deaths [1] [2] [3] [4] . Eighty-five per cent of the cases and 90% of the mortality occurs in sub-Saharan Africa, mostly amongst children [5, 6] . Recent reports point to a reduction of malaria cases in parts of Africa [7] . However, Nigeria, the most populous country of Africa, accounts for a quarter of the global cases and a third of the malaria-attributable childhood deaths [2, 8, 9] .

Cerebral malaria (CM) and severe malarial anemia (SMA) are the major severe disease syndromes in African children with a high level of mortality in the under-five age group. The current WHO case definitions for severe malaria combine P. falciparum blood stage parasitemia with coma, severe anemia or respiratory distress [10] , and it is well documented that there is significant overlap across these syndromes [11] . Despite the fact that these WHO case definitions are sensitive and useful for clinical diagnosis, the pathogenesis of severe disease is not well understood. One disadvantage of the WHO clinical definitions is that they lack the specificity required to carry out studies aimed at understanding the pathogenesis of clinically different forms of childhood malaria.

Previous studies have attempted to define malaria syndromes by studying plasma correlates of severity using reductionist approaches with variable success [12] [13] [14] . Small sample sizes, a lack of validation cohorts and a focus on a small selection of host plasma proteins have limited these studies. To overcome such limitations we use a systems approach to define the plasma proteome profile during malaria infection and identify distinctive patterns that are characteristic of different disease states. Contrary to other proteomic approaches, high-throughput plasma proteome profiling enables simultaneous analysis of a large number of samples. Therefore plasma proteome profiling allows the use of statistical pattern-recognition methods to discover and validate proteome-patterns that discriminate disease states.

We hypothesized that the plasma proteome during malaria infection reflects the molecules that are modulated as the severe status is established. In the present study we show that distinctive plasma proteome-patterns distinguish the different severe presentations of P. falciparum childhood malaria from the uncomplicated cases and also from well or unwell children without malaria.

Parents or guardians of study participants gave informed written consent. This research was approved by the joint ethics committee of the College of Medicine of the University of Ibadan and the University College Hospital Ibadan.

All study participants were recruited under the auspices of the Childhood Malaria Research Group (CMRG) at the 600-bed tertiary hospital University College Hospital (UCH) in the city of Ibadan, Nigeria in west sub-Saharan Africa. Ibadan is a densely populated urban setting with a population of 2.5 million inhabitants. Ibadan has a lengthy 8 months rainy season from March to October with malaria transmission and severe disease present all year round (holoendemic).

The study site is located in the UCH Ibadan Department of Paediatrics. We screen about 12,000 children attending the hospital (ill and well) for malaria parasites per year. Our studies report 11.3% SMA and 19.7% CM admissions in the parasitized children under five years of age [9] .

The participants in this study were recruited during 2006 to 2009 as part of a larger prospective case-control study of childhood severe malaria currently ongoing under the auspices of the CMRG. This case-control study was divided into a Discovery Cohort consisting of those patients recruited during 2006 to 2008 and a Validation Cohort made up of those recruited in the 2008 to 2009 period.

Malaria parasites were detected and counted by microscopy following Giemsa staining of thick and thin blood films [15] . Children with severe malaria were recruited on admission from the Otunba Tunwase Children's Emergency Ward (OTCHEW). Children with uncomplicated malaria were recruited as part of a daily routine malaria parasite screening at the Children's Outpatient Clinics (CHOP). Malaria-negative ill children were recruited either at admission from OTCHEW or from the Department of Paediatrics In-patient wards. Malaria-negative healthy community control children were recruited from local vaccination clinics as well as during school visits across several Ibadan districts.

We recruited children aged from 6 months to 13 years using five participant definitions. The malaria-positive children, the cases, are Uncomplicated Malaria (UM), Severe Malarial Anemia (SMA) and Cerebral Malaria (CM). The malaria-negative children, the controls, are Disease Control (DC) and Community Controls (CC). We followed the WHO criteria for severe P. falciparum malaria [10] . Cerebral malaria cases were defined as children in unrousable coma for at least one hour in the presence of asexual P. falciparum parasitemia with normal cerebrospinal fluid. A Blantyre coma score less than 2 was used to assess coma status. Children with hypoglycemia were excluded from the study. Added to the Table 1 . Characteristics of discovery and validation study groups. strict clinical and laboratory definitions of CM, our study patients recover consciousness after effective antimalarial therapy. We excluded from this study those CM patients who died. Our overall mortality rate for CM is of the order of 10%. Severe malarial anemia cases were defined as conscious children with Packed Cell Volume (PCV) less than 16% in the presence of P. falciparum parasitemia. We excluded from this study those SMA patients who died. Our overall mortality rate for SMA is less than 1%. Uncomplicated malaria cases were defined as febrile children with P. falciparum parasitemia who did not require hospital admission. Our study was designed to discover and validate plasma proteome changes in dichotomous cases for which we only included those children with CM and UM with PCV greater than 20% (Table 1) . We excluded from the study blood culture positive cases. Although we did not carry out blood cultures in all severe malaria patients, the cases recruited into this study are those in whom septicemia was not suspected and who were successfully treated with antimalarial alone. The DC group consists of malaria-negative children with infectious diseases such as meningitis, otitis media, diarrhea and upper respiratory tract infections. It also includes mild to moderately anemic children and children admitted for surgery.

Participants's clinical data were collected using a malariatailored questionnaire designed by the CMRG. A 2.5 ml blood sample was obtained from each participant in an EDTA blood collection tube for subsequent plasma separation. Blood samples were kept on ice and transferred to the central malaria laboratory. Plasma for this study was harvested by centrifugation (1000 g, 10

Packed cell volume (PCV) was measured using the microhaematocrit method [15] . Briefly, Blood was obtained in capillary tubes. Tubes were centrifuged at 12,000 g for 5 minutes. The percentage cell volume compared to the whole tube volume was calculated (i.e. PCV). Mean (6 standard deviation, sd), minimum and maximum PCV for each clinical group are tabulated in Table 1 . For discovery and validation cohort, these data were compared using a one-way multiple ANOVA test (p,0.05).

Malaria parasites were detected and counted by microscopy following Giemsa staining of thick and thin blood films [15] . Malaria Parasite (MP) densities were calculated as follows MP/ Table 1) . The microscopic criterion for declaring a participant to be free of malaria was the absence of parasites in 100 high-power (1000X) fields. One in 10 thick blood films were randomly selected and independently reviewed by local experienced microscopists not part of the research team.

Crude plasma was profiled using Surface Enhanced Laser Desorption/Ionization-Time Of Flight (SELDI-TOF) mass spectrometry. All plasma samples underwent two freeze-thaw cycles prior to analysis. Plasma samples were coded, blinded and randomized before application onto the following solid-phase fractionation surfaces (ProteinChipH arrays Bio-Rad): weak-cation exchange (CM10), strong-anion exchange (Q10) and reverse phase (H50) as previously described [16] . Liquid handling steps were automated using a Biomek 3000 Laboratory Automation Workstation (Beckman Coulter) and a 96 well BioprocessorH (Bio-Rad). Each ProteinChipH 96 well BioprocessorH included 1 quality control plasma standard derived from a single healthy individual, placed at random. Mass spectra were generated on a System 4000 Bio-Rad ProteinChipH mass spectrometer. Spectral peaks corresponding to mass/charge (m/z) clusters were detected and clustered using ProteinChipH Datamanager Client 4.1 software (BioRad). Mass spectrometer calibration was performed using Allin-1 Peptide and Protein calibrants (Bio-Rad). Reproducibility was determined by measuring the inter-ProteinChipH coefficient of variation (CV) for the quality control spectra, based on all peaks in the spectrum with intensity .1 mA. Overall interchip CV for the quality control sample was 20%, consistent with similar studies.

Liquid-phase anion-exchange fractionation of plasma samples was carried out using the ProteinChipH Fractionation Kit (Bio-Rad) according to the manufacturer's instructions with a Biomek 3000 Laboratory Automation Workstation. Six fractions were obtained from each sample eluting at pH 9.0 (f1), pH 7.0 (f2), pH 5.0 (f3), pH 4.0 (f4), pH 3.0 (f5) and organic phase (f6).

We selected subsets of the most relevant mass clusters in the discovery cohort groups using the weighted Kernel-based Iterative Estimation of Relevance Algorithm [17] (wKIERA) that combines a stochastic-search estimation of distribution algorithm with a kernel pattern-recognition method. We then used discovered relevant subsets of mass clusters to build discriminatory predictive models. We adopted a supervised learning approach to derive a classification rule using the Support Vector Machine (SVM) method [18] . Briefly, we used 10-fold cross validation to select parameters for the SVM. For the final model parameters, we selected those that gave the overall highest accuracy across the whole 10-fold cross validation. To obtain robust accuracy estimates for the classifier on the discovery data, we took 100 random re-samplings of the data, using 80% for training and 20% for testing. We selected as a final classifier the one that produced the highest accuracy and was then tested on the validation cohort data. Results were expressed as sensitivity, specificity and accuracy (proportion of correct classifications) and plotted on Receiver Operator Characteristic (ROC) space plots.

Our multivariate statistical tests included testing against age or sex to ascertain that significant pattern changes in the proteome were not dependent on those variables in the population studied.

To visualize the covariance within the mass spectral profiles we used Principal Component Analysis (PCA). PCA encapsulates the covariance within a set of variables by extracting a ranked set of independent factors or principal components. The first 3 components encompass a high proportion (,95%) of the informational content of a multivariate dataset. We plotted each patient with respect to the first 3 components, in 3-dimensional space, color-coding according to patient group.

A total of 946 children participated in this study as part of the discovery and validation case-controlled cohorts. The discovery cohort comprised of 367 malaria-positive children with either Cerebral Malaria (CM), Severe Malarial Anemia (SMA) or Uncomplicated Malaria (UM), and 289 malaria-negative children who were either Disease Controls (DC) or Community Controls (CC) ( Table 1 ). The validation cohort was prospectively recruited after the discovery cohort and comprised 160 malaria-positive children with either CM, SMA or UM, and 130 malaria-negative DC or CC children (Table 1) . PCV and malaria parasite (MP) densities are presented in Table 1 . Consistent with the recruitment criteria, both discovery and validation SMA groups had PCVs below 16% (Table 1 ). There was mild anemia across CM, UM and DC groups in both cohorts, whereas CC had normal mean hematocrit (Table 1) . Parasite densities across all the infected groups were similar (Table 1) .

To compare the proteome-patterns of the study groups, we fractionated plasma samples by three different chromatography procedures on solid-phase surfaces (weak-cationic and stronganionic ion-exchange, and reverse-phase) followed by Time-Of-Flight mass spectrometry. The resulting mass spectra from each of the surfaces contained a series of mass/charge ratio (m/z) peak clusters, each representing a protein of a particular mass. A set of proteins that are present, absent or at a different level in the samples defines a proteome-pattern that may discriminate between two or more of the study groups. To discover such patterns we applied statistical pattern recognition algorithms to the profiles and the selected number of discriminating proteins for each of the pairwise group comparisons is shown in Figure 1 , as the numbers in parentheses (Data S1). We built predictive models with the selected proteome-pattern for each study group comparison using a non-parametric supervised learning statistical framework. The discriminatory accuracy of these predictive models in the discovery cohort groups is shown in Figure 1a . To determine differences for malaria-positive children from healthy malaria-negative children we compared individually the plasma proteome of CM, SMA and UM groups with that of the CC group.

Overall, 22 to 33 proteins composed the discriminatory patterns with accuracies above 90% across the three comparisons (Figure 1a, blue bars) . Twenty-six proteins discriminated healthy from ill (hospital admitted) malaria-negative children (CC vs. DC) with similar accuracy (Figure 1a, green bar) . To examine proteins that are specific to malaria infection we compared each of the malaria-positive groups (CM, SMA, UM) to the DC group, obtaining discrimination accuracies above 80% (Figure 1a . orange bars). Finally, to assess differences between defined malaria syndromes we compared the malaria-positive groups (Figure 1a . yellow bars). In the comparison between CM and SMA, the two major severe syndromes, the accuracy was 70% (24 proteins). Higher accuracies between 70 to 80% were observed when samples from either CM or SMA groups were compared to UM children, using 36 and 54 proteins, respectively.

To validate the accuracy of the discrimination for the discovered plasma proteome-patterns, we tested the predictive models on the validation cohort groups (Figure 1b) . The best predictive model for each group comparison in the discovery cohort was asked to predict the group class in the validation cohort. We found that the predictive models obtained using the discovery cohort had similar accuracy for discrimination in the different group comparisons for the validation cohort (Figure 1b) . We compared the sensitivity and specificity of the predictive models for both discovery and validation cohort groups in ROC space and found them to be similar ( Figure 2) .

We then used Principal Component Analysis (PCA) on the selected proteins to visualize the separation of patient groups. The CC group clustered tightly together (Figure 3 , green spheres). Individual malaria-positive groups showed good separation from the malaria-negative CC group (Figure 3a-c) indicating that regardless of disease severity there are significant differences in the proteomes of the groups. The heterogeneous DC group had a more dispersed cluster pattern with little overlap with the CC group (Figure 3d ). The DC group, despite being distinct, showed different degrees of overlap with the malaria-positive groups (Figure 4a-c) . Of these comparisons, the CM vs. DC patient groups showed the greatest level of cluster dispersion (Figure 4a ) indicating greater covariance in the proteins that define these groups. We then compared the malaria-positive patient groups among themselves (Figure 4d-f ). CM and SMA groups showed overlap at the cluster interface and clearer segregation at the periphery; in the comparison of both severe forms (CM and SMA) with UM we observed that the severe patient groups had compact center clusters surrounded by a more disperse cluster of the UM patient group.

We simplified further the complexity of the plasma proteome by high-throughput liquid-phase anion-exchange fractionation followed by solid-phase weak-cation exchange fractionation prior to protein mass determination in the spectrometer on a subset of the samples. We assessed the discriminatory accuracy of relevant proteins obtained from each of the six anion-exchange fractions ( Figure 5 , f1 to f6) (Data S2). The reduction in the complexity of each fraction of the plasma samples resulted in a larger subset of proteins that improved discrimination between the malaria syndromes. Sets of proteins that distinguish between SMA and CM groups (Figure 5a , f1 to f6 in brackets) slightly outperformed the proteome-pattern from non-fractionated plasma. Sets of proteins differentiated the CM and UM groups with accuracies ranging from 70 to 80% (Figure 5b , f1 to f6 in brackets) and distinguished between SMA and UM with comparable accuracy (Figure 5c, f1 to f6 in brackets) .

We carried out an overall analysis of plasma proteome pattern overlap by comparing the discovered sets of proteins that discriminate UM, CM, SMA (malaria-positive) and DC (malaria-negative) ill children from the malaria-negative well children CC (Figure 6 , f1 to f6). We show that each plasma fraction (f1 to f6) contains a set of proteins that clearly define both the malariapositive and malaria-negative ill children to those malaria-negative well children in the community. Furthermore, we also show that the set of proteins that discriminate SMA and CM from UM have very little overlap across the six plasma fractions (Figure 6 , f1 to f6).

In the present study we carried out a large case-control study of severe childhood malaria, using a discovery cohort to define discriminatory plasma proteome-patterns and a second cohort to validate our findings, at the main tertiary hospital of the city of Ibadan, Nigeria.

We show that proteome-patterns from both crude and prefractionated plasma samples accurately define childhood malaria syndromes in the discovery cohort. We confirmed these findings using a prospectively collected validation cohort. Malaria infection introduces distinguishable changes in the plasma proteome of children as seen by the striking differences between the malarianegative CC and the malaria-positive children groups. The plasma proteome differences are specific for the malaria disease process and not surrogate markers of acute illness, as we are able to accurately distinguish between malaria-negative ill children and malaria-positive groups independently of their disease severity. We have also discovered plasma proteome differences that are specific to each of the childhood malaria syndromes assessed in the present study. Our findings provide a starting point to refine the current WHO definitions of these syndromes, which lack the necessary specificity to further study severe malaria pathogenesis.

We show that assessing the plasma proteome of the major malaria syndromes provides an unbiased discovery of combination of proteins that could be used to deepen our understanding of the pathogenesis of childhood malaria. This is supported by the finding that we can discriminate children with uncomplicated malaria from those with severe malarial anemia or cerebral malaria in both discovery and validation cohorts. These proteomepatterns encapsulate what changes differentiate uncomplicated malaria from the severe cases.

Overall, accuracy of discrimination between the CM and SMA was lower than that in the comparison of each of these syndromes with the UM group. The degree of overlap between CM and SMA goes beyond that expected from strict application of the WHO case definitions used in this study. Nevertheless, the plasma proteome-pattern discriminated with over 70% accuracy between the severe groups. This suggests that beyond common underlying mechanisms, such as acute inflammation, there are significant differences in the pathogenesis of the severe syndromes studied.

Our large cohorts allowed us to statistically validate the patternbased proteome definitions of the major childhood malaria syndromes. Although the mass spectrometry platform used in our study does not provide direct molecular identification, the chromatographic chemistry used and the mass-to-charge (m/z) ratio can be exploited to guide the identification of the set of proteins relevant for discrimination between syndromes. Plasma proteome profiling has been used to define a variety of disease states [16, [19] [20] [21] [22] [23] [24] as there is growing recognition of the advantages of using 'omics'-based methods to achieve sufficient levels of accuracy [24] . Our study showed that complex plasma protein patterns were necessary to discriminate between the different malaria syndromes. This further underlines the advantage of using unbiased high-throughput pattern recognition based methods.

In many infectious diseases, there are clinically important distinctions to be made between different manifestations associated with the same underlying pathogen and malaria clinical syndromes are a clear case in point. The pathogenesis of malaria due to its erythrocytic cycle occurs in the cardiovascular system and it is plausible that proteome changes in organs such as brain, spleen, kidney and bone marrow can be reflected in the plasma proteome. Our study confirms that there are proteome changes characteristic of the clinical malarial syndromes with different level of accuracy. Furthermore, host modulation by the pathogen is likely to generate changing patterns of protein expression associated with the progression of severe malaria syndromes and our current studies are designed to address such changes.

The lack of specific childhood malaria definitions has limited the progress on understanding the pathology of the major severe syndromes. To the best of our knowledge this study is the first to show that a panel of proteins, defined as a proteome-pattern, dissects clinical malaria syndromes. Further identification of the proteins that comprise the proteome-patterns will provide hints to the underlying pathogenesis on each of the syndromes. Furthermore, these proteome-patterns provide a reference point to facilitate the identification of other complex and overlapping severe childhood malaria syndromes.

Data S1 Solid-phase fractionation data. (XLS) Data S2 Liquid-phase fractionation data. (XLS) Predicting pseudoknotted structures across two RNA sequences Motivation: Laboratory RNA structure determination is demanding and costly and thus, computational structure prediction is an important task. Single sequence methods for RNA secondary structure prediction are limited by the accuracy of the underlying folding model, if a structure is supported by a family of evolutionarily related sequences, one can be more confident that the prediction is accurate. RNA pseudoknots are functional elements, which have highly conserved structures. However, few comparative structure prediction methods can handle pseudoknots due to the computational complexity. Results: A comparative pseudoknot prediction method called DotKnot-PW is introduced based on structural comparison of secondary structure elements and H-type pseudoknot candidates. DotKnot-PW outperforms other methods from the literature on a hand-curated test set of RNA structures with experimental support. Availability: DotKnot-PW and the RNA structure test set are available at the web site http://dotknot.csse.uwa.edu.au/pw. Contact: janaspe@csse.uwa.edu.au Supplementary information: Supplementary data are available at Bioinformatics online. Macromolecules such as DNA, RNA and proteins have the ability to form diverse tertiary structures, which enable functionality and thus, life. For many decades, proteins were deemed the global players in the cell until RNA entered the spotlight. For example, RNA structures have been found to be catalytically active, which was assumed to be the privilege of proteins. Furthermore, small RNAs are known to regulate gene expression and RNA viruses employ a plethora of structure elements to invade the host cell.

To gain insight into macromolecule function, one must investigate the structure. The first step in RNA folding is stable base pairing that leads to a secondary structure. As RNA structure formation is of hierarchical nature, secondary structure is the basis for the tertiary fold that produces the functional structure. Especially for RNAs, structure determination by experimental means is an intricate and expensive task. Computational RNA structure prediction is therefore an invaluable tool for biologists. Comparative structure prediction is considered the most reliable approach for computational RNA structure prediction. Single sequence structure prediction is always limited by the accuracy of the underlying folding model. Three main streams have been identified for comparative RNA secondary structure prediction: (i) predict a structure from a pre-computed sequence alignment; (ii) simultaneously compute an alignment and a structure and (iii) alignment-free methods (Gardner and Giegerich, 2004) .

Tools for multiple sequence alignments such as ClustalW (Thompson et al., 1994) are readily available and thus, structure prediction from an alignment is a tempting approach [e.g. RNAalifold (Hofacker et al., 2002) ]. Such methods heavily depend on the sequence conservation and quality of the underlying alignment. However, ncRNAs are conserved rather on the structure level than on the sequence level. The gold standard of RNA comparative structure prediction is the Sankoff approach as it does not rely on a high-quality sequence alignment and captures the structural conservation of ncRNAs. Sankoff (1985) introduced a theoretical dynamic programming algorithm for simultaneous folding and aligning for a set of N sequences that takes O(n 3N ) time and O(n 2N ) space. Practical variants have been derived which more or less retain the Sankoff principle by sacrificing optimality. Alignment-free methods aim to avoid the pragmatic restrictions made in a practical Sankoff approach as well as the reliance on a high-quality alignment [e.g. CARNAC (Perriquet et al., 2003) ]. Note that all of these comparative structure prediction methods exclude the prediction of RNA pseudoknots.

RNA pseudoknots are crossing structure elements with diverse functions. The principle of pseudoknot formation is that bases within a loop region pair with complementary unpaired bases outside the loop. From an algorithmic point of view, even the simplest type of pseudoknot adds considerable computational demands due to crossing base pairs. In fact, the majority of comparative RNA structure prediction methods exclude pseudoknots. Biologists have delivered a wealth of studies, which show that pseudoknots have an astonishing number of diverse functions and occur in most classes of RNA (Staple and Butcher, 2005) . RNA viruses use pseudoknots for hijacking the replication apparatus of the host (Brierley et al., 2007) .

A limited number of RNA comparative structure prediction methods can handle pseudoknots due to the computational complexity. Several of these methods take a sequence alignment as an input. ILM is an algorithm that takes as an input either individual sequences or a sequence alignment (Ruan et al., 2004) . A base pair score matrix is prepared initially and helices are added to the structure in an iterative fashion. In the approach hxmatch, a maximum weighted matching algorithm with combined thermodynamic and covariance scores is used (Witwer et al., 2004) . This program gives the option to be combined with RNAalifold.

KNetFold is a machine learning method, which takes a sequence alignment as an input and outputs a consensus structure allowing pseudoknots (Bindewald and Shapiro, 2006) . Simulfold takes an alignment as an input and simultaneously calculates a structure including pseudoknots, a multiple-sequence alignment and an evolutionary tree by sampling from the joint posterior distributions (Meyer and Miklos, 2007) . Tfold combines stem stability, covariation and conservation to search for compatible stems and subsequently for pseudoknots for a set of aligned homologous sequences (Engelen and Tahi, 2010) . Several comparative structure prediction methods including pseudoknots do not rely on an initial sequence alignment. The graph-theoretical approach comRNA computes stem similarity scores and uses a maximum clique finding algorithm to find pseudoknotted structures (Ji et al., 2004) . SCARNA performs pairwise structural alignment of stem fragments with fixed lengths derived from the probability dot plot (Tabei et al., 2008) .

In the following, a novel comparative approach for predicting structures including H-type pseudoknots called DotKnot-PW will be introduced. The input consists of two unaligned, evolutionarily related RNA sequences. Similarity scores between structure elements will be calculated. Statistically significant pairs will be used to find the set of conserved structure elements common to two sequences, which maximize a combined thermodynamic and similarity score. Using a hand-curated test set of pseudoknotted structures with experimental support, the prediction accuracy of DotKnot-PW will be compared with methods from the literature.

Pseudoknots are functional elements in RNA structures and therefore, the most promising approach for comparative prediction is a structure comparison with less focus on exact sequence matching. In fact, perfect conservation on the sequence level can be more of a curse than a blessing. Especially ncRNAs are known to evolve quickly and so-called consistent and compensatory base pairs in both sequences will give much more confidence for structure conservation than a sequence alignment. One strong point of the DotKnot method for single sequence pseudoknot prediction (Sperschneider and Datta, 2010; Sperschneider et al., 2011) is that the set of possible H-type pseudoknot candidates (and secondary structure elements) is explicitly computed and thus readily available for further investigation. The main steps in the pairwise pseudoknot prediction approach DotKnot-PW are as follows ( Fig. 1) :

(1) Run DotKnot for two unaligned sequences Seq x and Seq y . This returns secondary structure element and Htype pseudoknot candidate dictionaries.

(2) Calculate pairwise base pair similarity scores for secondary structure elements and H-type pseudoknot candidates. Keep significant pairs that have a low estimated P-value.

(3) Use significant pairs to calculate the set of conserved structure elements and pseudoknots for the two sequences that maximizes a combined free energy and similarity score.

The key point of the DotKnot-PW approach is how to score the similarity of stems, secondary structure elements and H-type pseudoknot candidates derived from sequences Seq x and Seq y . Related work has been done for stem finding in unaligned sequences, where stem candidates are assigned a matching score across unaligned sequences, e.g. in SCARNA. Another point is how to assess the significance of a similarity score using Pvalues. These points will be explained in detail in the following section.

For two unaligned RNA sequences Seq x and Seq y , the single sequence prediction method DotKnot (Sperschneider and Datta, 2010; Sperschneider et al., 2011) returns two stem dictionaries D s (x) and D s (y) derived from the probability dot plot. It also returns secondary structure element dictionaries D L s ðxÞ, D L s ðyÞ and D M s ðxÞ, D M s ðyÞ and H-type pseudoknot candidate dictionaries D p (x) and D p (y) (Fig. 1) . To detect conserved structure elements for the two sequences, a pairwise structural comparison is performed. Instead of a full structure-tostructure alignment, which takes O(n 4 ) time and O(n 3 ) space, pairwise base pair similarity scores are calculated using the RIBOSUM85-60 matrix for base pair substitutions (Klein and Eddy, 2003) .

For two given stems s i (x) and s j (y) with fixed lengths in sequences Seq x and Seq y , respectively, the base pair similarity score sim[s i (x), s j (y)] is calculated using an ungapped local structure alignment of the base pairs with the RIBOSUM85-60 matrix. As an example, consider the following optimal ungapped local structure alignment of the two stems with base pair similarity score of sim[s 1 (x), s 2 (y)] ¼ 22.04 using the RIBOSUM85-60 matrix.

)))))--To evaluate the significance of base pair similarity scores instead of the raw score, one has to find out what the underlying probability distribution is. Similar to the case of ungapped local sequence alignments (Karlin and Altschul, 1990) , it is assumed here that the base pair similarity scores follow an extreme value distribution. However, the main difference is that a comparison between fixed-length stem fragments is made. It is important to remember that parameters and K describe the extreme value distribution of optimal local alignment scores in the asymptotic limit of long sequences (Altschul et al., 2001) . Here, the parameters for the generalized extreme value distribution are pre-calculated using maximum Fig. 1 . For two unaligned RNA sequences Seq x and Seq y , DotKnot-PW produces structure element dictionaries derived from the probability dot plot. Similarity scores and P-values are computed to detect conserved elements likelihood fitting of a distribution to the histogram of a large sample of random base pair similarity scores. The maximum likelihood fitting was performed using the ismev package of the R statistical language for a range of stem lengths (see Supplementary Material). The P-value is defined as the probability to obtain a score greater than or equal to the observed score strictly by chance. A stem s i (x) in sequence Seq x and a stem s j (x) in sequence Seq y are a significant pair if the score sim[s i (x), s j (y)] has an estimated P-value less than . Stem pairs with a P-value larger than are not considered in the following.

For two interrupted stems, the base pair similarity score is calculated by deleting bulges and internal loops and scoring stems as consecutive base pairs. Base pair similarity scores for regular and interrupted stems are also calculated if the difference in number of base pairs is less than 5. For example, a stem with one bulge might be a conserved match with a regular stem. A stem s i (x) in sequence Seq x and a stem s j (y) in sequence Seq y are a significant pair if the score sim[s i (x), s j (y)] has an estimated P-value less than .

Calculating the base pair similarity score for two multiloop structures is complex due to the variety of inner loop elements, which may be regular or interrupted stems. A multiloop s M i ðxÞ can be decomposed into an outer stem s o i ðxÞ and a set of inner structure elements

The base pair similarity score sim½s o i ðxÞ, s o j ðyÞ for the outer stems of two multi-loops can be easily obtained from the previously calculated base pair similarity scores. If the outer stem is a conserved match, a local alignment on the set of inner structure elements is used to find the base pair similarity score. Here, gaps are allowed in the local alignment of inner structure elements; however, no gap penalty is used. Let two sets of inner structure elements S i (x) ¼ [s 1 (x),. . ., s n (x)] and S j (y) ¼ [s 1 (y),. . ., s m (y)] be given. Let H(i, j) be the maximum similarity score between a suffix of S i (x) and a suffix of S j (y). The optimal local alignment is calculated as follows:

Hði, 0Þ ¼ 0, 0 i n Hð0, jÞ ¼ 0, 0 j m Hði, jÞ ¼ max 0 Hði À 1, jÞ Hði À 1, j À 1Þ þ sim½s i ðxÞ, s j ðyÞ Hði, j À 1Þ

A multiloop s M i ðxÞ in sequence Seq x and a multiloop s M j ðyÞ in sequence Seq y are a significant pair if the similarity score sim½s M i ðxÞ, s M j ðyÞ has an estimated P-value less than .

A H-type pseudoknot has two pseudoknot stems S 1 and S 2 . The prerequisite for a conserved pseudoknot pair is that both core H-type pseudoknot stem pairs [S 1 (x), S 1 (y)] and [S 2 (x), S 2 (y)] are significant. The base pair similarity score for two H-type pseudoknots p i (x) and p j (y) in sequences Seq x and Seq y , respectively, is the sum of base pair similarity scores for the core pseudoknot stems as well as the base pair similarity score from a gapped local alignment of the recursive secondary structure elements in the loops (as described for multiloops). A pseudoknot p i (x) in sequence Seq x and a pseudoknot p j (y) in sequence Seq y are a significant pseudoknot pair if the similarity score sim[p i (x), p j (y)] has an estimated P-value less than .

The base pair similarity score calculated in the previous sections might not be powerful enough to distinguish true positive conserved structure element pairs from false-positive structure element pairs due to the finite lengths of stems and exclusion of loop sequences in the alignment. Therefore, a dissimilarity score is also used to confirm whether a pair is significant. The dissimilarity for two given structure elements s i (x) and s j (y) in sequences Seq x and Seq y is defined as:

where dissim 1 is the difference in the stem lengths and dissim 2 is the difference in the number of loop lengths. As an example, consider the pseudoknot pair p 1 (x) and p 1 (y) in sequences Seq x and Seq y , respectively, with stems S 1 , S 2 and loops L 1 , L 2 , L3. The pseudoknot pair has dissimilarity of 6. [[[[[[.) )))))) [.[[[[[[[.) ))))). .

A weight is assigned to a significant pair, which is a combination of the free energy, covariation and dissimilarity. The overall weight s of a significant structure element pair [s i (x), s j (y)] in sequences Seq x and Seq y is a combination of the free energy weights w[s i (x)] and w[s j (y)], base pair similarity score sim[s i (x), s j (y)] and dissimilarity dissim[s i (x), s j (y)]: s½s i ðxÞ, s j ðyÞ ¼ Â sim½s i ðxÞ, s j ðyÞ À Â fw½s i ðxÞ þ w½s j ðyÞg À Â dissim½s i ðxÞ, s j ðyÞ Only structure element pairs with positive score s are allowed in the following dynamic programming algorithm. Here, and are set to 0.5 and is set to 1.

Let p x 1 , . . . , p x n be the number of structure elements in the first sequence Seq x and p y 1 , . . . , p y m be the number of structure elements in the second sequence Seq y . Each structure element has a left and right endpoint in the sequence and is a stem, interrupted stem, multiloop or H-type pseudoknot. Structure elements can also be represented as nodes in a graph. In each sequence, the structure elements are ordered by their right endpoints. An edge is drawn between two structure elements in the first and the second sequence if their base pair similarity score has a P-value less than . Given the set of edges between nodes p x 1 , . . . , p x n and p y 1 , . . . , p y m , the goal is to find the set of edges with maximum weight that are non-crossing. This relates to finding the set of non-overlapping structure elements in the two sequences that maximize the score under the requirement that the interval ordering is preserved. A set of structure elements in the first and second sequence, which preserves the interval ordering is called a feasible structure element alignment and must satisfy the following two requirements. Each structure element can be aligned with at most one other structure element in the other sequence. The order of structure elements must be preserved with respect to the alignment. That is, if structure elements p x i and p x j in the first sequence are aligned with p y a and p y b in the second sequence, respectively, the pairs may never overlap: p x i 5p x j^p x a 5p x b (Fig. 2) . Given nodes p x 1 . . . , p x n in the first sequence Seq x and p y 1 , . . . , p y m in the second sequence Seq y , let f(i, a) be the maximum sum of edge weights for nodes between 1 and i in the first sequence and 1 and a in the second sequence such that the edges are non-crossing (i n and a m). The nodes that maximize the sum of edge weights are called an optimal structure element alignment for the two sequences. The optimal structure element alignment is calculated using dynamic programming. For a given structure element p i with start point a i and end point b i , let pre(i) be the non-overlapping predecessor. For each structure element, its predecessor is pre-computed using the sorted list of structure elements. The recursion for calculating the optimal structure element alignment is as follows:

Furthermore, nested structures are taken into account for significant outer stem pairs, which have estimated P-value less than . For each significant outer hairpin loop pair, the optimal structure element alignment of inner elements is computed.

For two unaligned RNA sequences Seq x and Seq y , the single sequence prediction method DotKnot returns structure element dictionaries derived from the probability dot plot. Let n and m be the number of structure elements in sequences Seq x and Seq y , respectively. Calculating the similarity scores and the optimal structure element alignment takes O(nm) time. Furthermore, nested structures are taken into account for significant outer stem pairs, which have estimated P-value less than . Let a be the number of significant stem pairs, where both stems are hairpin loops. For each significant outer hairpin loop pair, the optimal structure alignment of inner elements is computed. In the worst case, this increases time requirements to O(a  nm). The number of structure elements depends on the base composition of the sequence. Empirically, n and m can be observed to grow linearly with the length of the sequence for uniform base distribution (see Supplementary Material). In practice, DotKnot-PW can be expected to run in the order of minutes for sequences shorter than 500 nt.

Many pseudoknot prediction programs have been evaluated using all the entries in the PseudoBase database (van Batenburg et al., 2000) . There are several caveats in this approach. First, the sequences given in PseudoBase are those which exactly harbor the pseudoknot. However, in practice structure prediction algorithms will be applied to longer sequences without prior knowledge of the pseudoknot location. Second, long-range pseudoknot entries appear in a truncated version in the database. Third, some classes of pseudoknots have a large number of entries (such as short H-type pseudoknots in the 3 0 -untranslated regions of plant viruses), whereas more complex types of pseudoknots only have one representative (such as long-range rRNA pseudoknots). Therefore, a hand-curated dataset of pseudoknot structures will be used here.

When it comes to pseudoknots, many structures have been published based on a secondary structure predicted by free energy minimization. These predicted secondary structures are used as a working model and refined using experimental techniques such as chemical and enzymatic probing. However, the native structure remains unsolved unless tertiary structure determination methods such as X-ray crystallography are used. Testing structures that are based on computer predictions with no experimental support creates a bias in the benchmark and will be avoided in this evaluation.

A total of 16 pseudoknotted reference structures from different RNA types were collected, which have strong experimental support. For each reference structure, a supporting set of 10 evolutionarily related sequences was obtained from the RFAM database (Gardner et al., 2010) . Note that for the vast majority of supporting sequences, no experimentally determined structures are available. The average pairwise sequence identities vary from 55% to 99%. Given a reference structure, the performance of prediction algorithms is evaluated in terms of sensitivity (S), i.e. the percentage of base pairs in the reference structure, which are predicted correctly, as well as positive predictive value (PPV), i.e. the percentage of predicted pairs, which are in the reference structure. The Matthews correlation coefficient (MCC) is also reported and is in the range from À1 to 1, where 1 corresponds to a perfect prediction and À1 to a prediction that is in total disagreement with the reference structure. The performance of each method for predicting the reference structure was evaluated as described in Gardner and Giegerich (2004) .

DotKnot-PW was compared with methods that are freely available and use standard input and output formats. The comparative methods are CARNAC, Tfold and hxmatch (with the -A option using RNAalifold). All of these methods return structure predictions for only the reference structure with regards to the support set of evolutionarily related sequences. Tfold and hxmatch take a sequence alignment as the input. ClustalW with the default parameters was used to produce the initial sequence alignment. DotKnot-PW and CARNAC take a set of unaligned sequences as the input. Furthermore, prediction results for the reference sequence (not the supporting sequences) were obtained from the single sequence methods DotKnot (Sperschneider and Datta, 2010; Sperschneider et al., 2011), ProbKnot (Bellaousov and Mathews, 2010) , IPknot (Sato et al., 2011) and RNAfold (Hofacker et al., 1994) . Note that all methods except CARNAC and RNAfold allow pseudoknot prediction.

The results are shown in Table 1 . DotKnot-PW has the highest average MCC of 0.75 for the test sequences. For each reference structure with the 10 support sequences from the corresponding RFAM family, 10 predictions are returned ordered by the combined free energy and similarity score. If only the Fig. 2 . A set of edges with positive scores is given between nodes p 1 ,. . .,p 7 in the first sequence and p 1 ,. . .,p 6 in the second sequence. The goal is to find the best set of non-overlapping structure elements in the two sequences such that the interval ordering is preserved. The optimal structure element alignment, which preserves the interval ordering includes structure elements p 1 , p 4 , p 7 in the first sequence and p 1 , p 4 , p 6 in the second sequence pairwise prediction with highest combined free energy and similarity score is taken, DotKnot-PW has an improved average MCC of 0.81. Tfold and hxmatch have average MCC of 0.6 and 0.59, respectively. CARNAC has average MCC of 0.45 with much higher average specificity than sensitivity. The prediction results for single sequence structure prediction for each of the reference sequences with experimentally determined structures are also shown in Table 1 . Note that this does not include the prediction for the support sequences from RFAM, as no experimentally determined structures are available. All single sequence pseudoknot prediction methods show improved results over using RNAfold. DotKnot has the highest average MCC of 0.76, followed by IPKnot and ProbKnot.

As an example, consider the S15 mRNA pseudoknot that binds to specific proteins in the autoregulation mechanism of ribosomal protein S15 synthesis (Philippe et al., 1995) . For the reference sequence S15 and 10 support sequences from the corresponding RFAM family, DotKnot-PW returns pairwise predictions ordered by the combined free energy and similarity score. The top two pairwise predictions with the highest scores are shown in Figure 3 .

We presented DotKnot-PW for prediction of structures common to two RNA sequences, including H-type pseudoknots. Both DotKnot and DotKnot-PW have been designed as dedicated pseudoknot prediction tools. In the following, important aspects of pseudoknot prediction will be discussed.

Single sequence prediction methods are always limited by the underlying RNA folding model. This may be the set of free energy parameters used by free energy minimization methods or the underlying methodological framework such as maximum expected accuracy methods. DotKnot-PW shows excellent results on H-type pseudoknots with short interhelix loops. For this type of pseudoknots, DotKnot-PW uses free energy pseudoknot parameters by Chen (2006, 2009 ) based on polymer statistical mechanics. Improvements of the accuracy of free energy parameters, both for secondary structures and pseudoknots, will lead to more accurate prediction methods. However, one has to keep in mind that the algorithms themselves must be designed in such a fashion that novel parameters can be efficiently incorporated. The heuristic framework of DotKnot-PW has been designed such that it can incorporate sophisticated free energy parameters for pseudoknots, secondary structures and coaxial stacking. In the future, DotKnot-PW could also use contributions from basic tertiary structure elements such as base triples around the pseudoknot junction or stem-loop interactions.

Pseudoknot prediction algorithms come in two flavors: either they can predict a certain, restricted class of pseudoknots or they do not have a restriction on the type of pseudoknot that can be predicted. For methods using free energy parameters, the inclusion of general types of pseudoknots might be more of a

Each reference structure is given by its ID (see Supplementary Material for dot-bracket notation). The following column gives the method of experimental support (NMR, NMR spectroscopy; X-ray, X-ray crystallography; SC, sequence comparison; MG, mutagenesis; SP, structure probing), length of the sequence and number of pseudoknots. For each reference structure, the corresponding RFAM family ID, average sequence length and average pairwise sequence identity is shown. The * symbol means that the method failed to run. The 'first' prediction for DotKnot-PW is the pairwise prediction with highest combined free energy and similarity score.

curse than a blessing, as no reliable free energy parameters for complex pseudoknots are available. DotKnot-PW has restrictions on the type of pseudoknot that can be predicted. However, this does not always lead to poor prediction results in practice. For example, DotKnot-PW shows the best result for the HDV ribozyme, which is a complex double nested pseudoknot.

The results from the benchmark for structure prediction in Table 1 must be interpreted with care. First, the tested methods can be run with different parameters, possibly producing better results. However, as a typical user has no prior knowledge about the structure, the default parameters for each method are used. Of course, a comprehensive benchmark should include a larger number of structures to obtain a more reliable evaluation. However, in this study, the focus has been on a test set where the structures are supported experimentally. Many structures have been published, which were determined using computational tools and this will inevitably create a bias in a benchmark, and thus they were excluded here.

Here, an extension of the single sequence prediction method DotKnot was presented based on the pairwise comparison of structure elements. This approach called DotKnot-PW is designed as an algorithm for finding the structure including H-type pseudoknots common to two sequences. As shown in Table 1 , DotKnot-PW can greatly improve structure predictions for RNA families when compared with the single sequence prediction using DotKnot. In some cases, a comparative approach might have lower sensitivity than a single sequence prediction; however, this should not generally be judged as 'inferior'. For example, ncRNAs might preserve some integral base pairs throughout evolution and only these will be detected by a comparative approach, which returns the set of base pairs common to a set of evolutionarily related sequences. DotKnot-PW uses a set of unaligned sequences as the input; therefore, no expert user intervention is required. In the future, DotKnot-PW will be extended to include intramolecular kissing hairpins. Furthermore, constrained folding will be implemented to predict a structure subject to constraints, e.g. enforce certain base pairs or regions, which must remain unpaired.

DotKnot-PW has been designed as a dedicated pseudoknot prediction tool and should be applied to RNA sequences where pseudoknotted interactions are suspected in the structures. Prediction accuracy will inevitably decrease for sequences, which are longer than say 400 nt for any single sequence structure prediction method (Reeder et al., 2006) . To achieve reliable results, short sequences should be folded using DotKnot and predictions should be compared with results from other methods from the literature. To gain confidence in predictions, subsequent comparative prediction using DotKnot-PW and other comparative methods is highly recommended. Ideally, experimental verification of computationally predicted pseudoknots should be sought. Fig. 3 . Pairwise prediction results for the S15 mRNA pseudoknot (RFAM family RF00114) with the top two combined free energy and similarity scores. The reference structure is shown at the top and folds into two conformations in dynamic equilibrium: a H-type pseudoknot or a series of hairpins.