Datasets:

suolyer

/

pile_pubmed-central

like 1

Introduction {#Sec1} ============ The finding of fresh monomers^[@CR1],[@CR2]^ and the development of active catalysts^[@CR3],[@CR4]^ are the central topics in synthetic polymer chemistry. Carbonyl sulfide (COS), a key intermediate of the atmospheric sulfur cycle and the most abundant sulfur-containing gas in the troposphere, causes haze, acid rain, and ozonosphere damage^[@CR5]^, and is also a one-carbon (C~1~) heterocumulene and structural analog of carbon dioxide (CO~2~). Utilizing COS to copolymerize with epoxides is a emerging atom-economic and versatile approach to produce functional sulfur-containing polymers^[@CR6]--[@CR11]^. In contrast, traditional synthesis of sulfur-containing polymers often involves the condensation of thiols with phosgene and ring-opening polymerization (ROP) of cyclic thiocarbonates that are generally derived from thiols and phosgene^[@CR6]^. Recent synthetic advances^[@CR6]--[@CR11]^ provide metal catalytic strategies for making a variety of COS-derived poly(monothiocarbonate)s that have good solubility in organic solvents, superior optical properties, and excellent chemical resistance^[@CR6],[@CR7]^. Zinc-cobalt(III) double-metal cyanide complexes^[@CR8]^ and (salen)CrX/onium salts (Lewis bases, LBs, Fig. [1a](#Fig1){ref-type="fig"})^[@CR9]--[@CR11]^ have been discovered to catalyze the COS/epoxide copolymerization. Unfortunately, metal-contaminated or colored copolymers are resulted and severely impede their applications in optical, optoelectronic, photochemical, or biomedical materials^[@CR12]--[@CR14]^. Organic Lewis pairs composed of triethylborane (TEB) and LBs, including amidine, quinidine, quaternary onium salts could generate poly(monothiocarbonate)s from COS and epoxides^[@CR15]^ (Fig. [1b](#Fig1){ref-type="fig"}). However, the TEB/LB pairs often led to relative broad polydispersity (PDI) and higher molecular weights than the calculated one. Meanwhile, only TEB, which is toxic and spontaneous combustion in air, was effective to COS/epoxide copolymerization.Fig. 1The catalyst systems for COS/epoxide copolymerization. **a** Metal catalysts (zinc-cobalt(III) double-metal cyanide complexes and (salen)CrX/onium salts), **b** TEB/LB pairs (triethylborane/Lewis bases), **c** TU/base pairs in this study (TU-1: diisopropyl thiourea; TU-2: 1-cyclohexyl-3-phenylthiourea; TU-3: 1-\[3,5-bis(trifluoromethyl) phenyl\]-3-cyclohexylthiourea; DBU: 8-diazabicyclo\[5.4.0\]undec-7-ene; MTBD: *N*-methyl-1,5,7-triazabicyclododecene; ^*t*^Bu-P~4~: **P4**, 1-*tert*-butyl-4,4,4-tris(dimethylamino)-2,2-bis \[tris(dimethylamino)- phosphoranylidenamino\]-2λ5,4λ5-catenadi(phosphazene); ^*t*^Bu-P~2~: **P2**, 1-*tert*-Butyl-2,2,4,4,4-pentakis(dimethylamino)-2λ5,4λ5-catenadi(phosphazene); and ^*t*^Bu-P~1~: **P1**, *tert*-butylimino-tris(dimethylamino) phosphorene) The above-mentioned catalysts have the coordination bonds that are responsible to the activation of the monomers. Whereas very few of reports suggested an anionic copolymerization process involved C~1~ monomers. For example, Nozaki et al., disclosed that \[PPN\]Cl could solely catalyze the carbon-disulfide (CS~2~)/propylene sulfide copolymerization^[@CR16]^. Feng and Gnanou et al. presented that alkoxide/benzyl alcohol (BnOH) could effectively initiate the CO~2~/epoxide copolymerization^[@CR17],[@CR18]^. However, anionic copolymerization of COS with epoxides remains unexplored. In contrast with the CO~2~/epoxide copolymerization that is often expected to attain fully alternating structure and no production of side cyclic carbonate (i.e., 100% polycarbonate)^[@CR19]--[@CR24]^, the chemistry of COS/epoxide copolymerization is more complicated^[@CR6]^. One is the possible occurrence of oxygen/sulfur exchange reactions (O/S ERs), which cause the production of CO~2~, and thiirane intermediate, will produce randomly distributed dithiocarbonate and carbonate units in the final copolymer^[@CR8],[@CR25]--[@CR27]^. The other is that the copolymerization of structurally asymmetric COS with a terminated epoxide, will generate four consecutive monothiocarbonate diads, i.e.,: head-to-tail (H--T), tail-to-head (T--H), tail-to-tail (T--T), and head-to-head (H−H) diads^[@CR6]^. As a result, metal-free catalyst for anionic COS/ epoxide copolymerization should avoid O/S ER, attain highly regioselectivity involved two asymmetric monomers and be precisely controlled by varying the monomer to initiator ratios under mild condition. Herein, we have developed a living copolymerization of COS with various epoxides with high activity, using commonly available thioureas (TUs) and organic LBs (DBU, MTBD, **P4**, **P2**, and **P1**, Fig. [1c](#Fig1){ref-type="fig"}). This catalyst system was developed based on the hypothesis that the cooperative catalytic process of Lewis pairs composed of TU and base, undergoing a non-covalent mode to activate and stabilize the alcohol initiator/chain end for controlling the anionic copolymerization^[@CR28]--[@CR31]^. The resulting poly(monothiocarbonate)s have perfectly alternating structure with regioregularity, controlled molecular weights and narrow PDIs, and are colorless (Supplementary Fig. [1](#MOESM1){ref-type="media"}). Results {#Sec2} ======= Anionic COS/propylene oxide (PO) copolymerization {#Sec3} ------------------------------------------------- Previous studies have shown that the organic bases, e.g.: amidine and guanidine widely employed as the cocatalysts for metal complex for catalyzing CO~2~(COS) /epoxide copolymerization^[@CR6],[@CR32]--[@CR34]^. In this scenario, the organic bases often exhibited no activity towards CO~2~/epoxide copolymerizations^[@CR19]--[@CR24],[@CR32]--[@CR34]^. Since COS is more reactive than CO~2~ and the expected thiocarbonate anion \[OC(=O)S^−^\] is more nucleophilic than carbonate anion \[OC(=O)O^−^\]^[@CR6]^, we performed the sole catalysis of LBs (DBU, MTBD, **P4**, **P2**, and **P1**) for COS/PO copolymerization as controls (entries 1--5 in Table [1](#Tab1){ref-type="table"}) for comparatively studying the catalytic performance of the designed Lewis pairs of TU/LB (entries 9--18 in Table [1](#Tab1){ref-type="table"}). Unexpectedly, we observed that DBU could solely catalyze COS/PO copolymerization in a high PO/DBU feed ratio of 250/1 at ambient temperature (25 °C) for a long time of 24 h (entry 1, Table [1](#Tab1){ref-type="table"}). Poly(propylene monothiocarbonate)s (PPMTCs) were obtained with a number-average molecular weight (*M*~n~) of 38.3 kg/mol with production of 7% of cyclic monothiocarbonate. **P4**, a phosphazene with the strongest basicity^[@CR35]^, was active for the anionic ROP of epoxides, cyclic esters, and caprolactam^[@CR36]--[@CR39]^, was also effective to the COS/PO copolymerization with a TOF of 23 h^−1^ even under a PO/**P4** feed ratio (1000/1). However, the copolymer selectivity was only 60%, and 8% polyether was detected (entry 2 in Table [1](#Tab1){ref-type="table"}, Supplementary Fig. [2](#MOESM1){ref-type="media"}). The sole catalysis of **P2**, **P1**, and MTBD for the copolymerization exhibited high copolymer selectivity (≥96%), but low TOFs (1--4 h^−1^, entries 3--5 in Table [1](#Tab1){ref-type="table"}). Concurrently, high reaction temperatures (≥70 °C) or the use of TBD (at 20 °C) only afforded cyclic products (entry 6 in Supplementary Table [2](#MOESM1){ref-type="media"}; entry 10 in Supplementary Table [1](#MOESM1){ref-type="media"}). PPMTCs obtained by these organic bases had *M*~n~s of 29.4--46.0 kg/mol and PDIs of 1.20--1.58. This result exclusively suggested that the COS/PO copolymerization could be performed in an anionic manner.Table 1COS/PO copolymerization catalyzed by TU-1/organic base pairs at 25 °CEntry^a^LB\[PO\]: \[LB\]: \[TU-1\]: \[I\]TOF (h^--1^)^b^Copolymer selectivity^c^Alternating degree (%)^c^T--H diad content (%)^d^O/S ER product^d^*M*~n~ (kg/mol)^e^PDI^e^1DBU250:1:0:0493/7100\>99N.F.38.31.352**P4**1000:1:0:02360/4092/8\>99N.F.29.41.583**P2**250:1:0:0498/2100\>99N.F.45.11.144**P1**250:1:0:01\>99100\>99N.F.9.01.055MTBD250:1:0:0396/4100\>99N.F.46.01.206DBU2000:1:0:11092/8100\>99N.F.13.91.207**P4**2000:1:0:15599/1100\>99N.F.42.51.238**P2**2000:1:0:12999/1100\>99N.F.35.61.159**P4**2000:1:1:04197/3100\>99N.F.31.31.1610**P2**1000:1:1:024\>99100\>99N.F.34.31.1611DBU250:1:1:0793/7100\>99N.F.30.21.1312MTBD250:1:1:0495/5100\>99N.F.36.11.1613^f^MTBD1000:1:1:01098/2100\>99N.F.98.41.1414**P4**4000:1:1:175\>99100\>99N.F.21.31.2115**P4**4000:1:5:511298/2100\>99N.F.9.91.1316**P2**2000:1:1:135\>99100\>99N.F.19.31.1317DBU2000:1:1:12294/6100\>99N.F.11.31.1918MTBD2000:1:1:11797/3100\>99N.F.12.01.23The copolymerization results under other conditions (including controls), representative NMR spectra and corresponding calculations are in Supplementary Figs. [2](#MOESM1){ref-type="media"}--[11](#MOESM1){ref-type="media"} and [35](#MOESM1){ref-type="media"}, and the Methods^a^Reactions were run at 25 °C in neat PO (1.0 ml; COS: PO = 1.2: 1) in a 10 ml autoclave, 24 h^b^Turnover of frequency (TOF), (Mol epoxide consumed)/(mol LB h), PO conversion was determined by ^1^H NMR spectroscopy^c^Determined by ^1^H NMR spectroscopy. The copolymer selectivity is the molar ratio of the copolymer/cyclic product; The alternating degree is the molar percentage of the monothiocarbonate unit in the polymer chain^d^Determined by ^13^C NMR spectroscopy. O/S ER = oxygen-sulfur exchange reaction. N.F. = not found (dithiocarbonate and carbonate units)^e^Determined by gel permeation chromatography in THF, calibrated with polystyrene standards^f^72 h The activity and selectivity of the above LB-catalyzed COS/PO copolymerization were clearly enhanced by introducing TU-1 (entries 9--12 in Table [1](#Tab1){ref-type="table"}), while TU-1 did not catalyze the reaction alone (entry 11 in Supplementary Table [1](#MOESM1){ref-type="media"}). For example, **P4**/TU-1 and **P2**/TU-1 pairs afforded improved TOFs (41 and 24 h^−1^), copolymer selectivity (97/3 and \>99%) even in high monomer/pair feed ratios of 2000/1/1 and 1000/1/1, respectively, (entries 9--10, Table [1](#Tab1){ref-type="table"}). Concurrently, COS/PO copolymerization via the catalysis of DBU/TU-1 or MTBD/TU-1 pairs led to a slight improvement of TOF (4--7 h^−1^) and the copolymer selectivity (93/7 and 95/5) while the **P1**/TU-1 pair had low TOF (1 h^−1^), but perfect copolymer selectivity (\>99%) (entry 5, Supplementary Table [1](#MOESM1){ref-type="media"}). Of note, the use of TU-1 caused a slight decrease of *M*~n~ (30.2--36.1 kg/mol) and narrower PDIs of 1.13--1.16, suggesting the generation of larger amounts of active centers when TU-1 was introduced. Of special interest, MTBD/TU-1 pair catalysis provided a copolymer with a remarkably high *M*~n~ of 98.4 kg/mol with PDI of 1.14 after 72 h reaction (entry 13, Table [1](#Tab1){ref-type="table"}), which can be ascribed to a longer lifespan of the active species. The addition of benzyl alcohol (BnOH) to the COS/PO copolymerization catalyzed by either single LB (entries 6--8 in Table [1](#Tab1){ref-type="table"}, entries 3, 6, and 8 in Supplementary Table [1](#MOESM1){ref-type="media"}) or TU-1/LB pairs (entries 14--18 in Table [1](#Tab1){ref-type="table"}, entries 2, 4, 7, and 9 in Supplementary Table [1](#MOESM1){ref-type="media"}) led to a dramatic improvement of TOF without the sacrifice of the copolymer selectivity, even in a rather low catalyst loading. For example, adding equimolar BnOH to **P4** and **P4**/TU-1-catalyzed copolymerizations presented improved TOFs of 55 and 75 h^−1^, respectively, (entries 7 and 14 in Table [1](#Tab1){ref-type="table"}). In addition, high loading of TU-1 and BnOH (**P4**/TU-1/BnOH = 1/5/5) led to a dramatic increase of TOF to 112 h^−1^ (entry 15 in Table [1](#Tab1){ref-type="table"}). Introducing equimolar BnOH to the DBU/TU-1 pair led to a clear promotion of TOF (22 h^−1^) even in a low PO/DBU/TU-1 feed ratio of 2000/1/1(entry 17 in Table [1](#Tab1){ref-type="table"}). In contrast, no copolymers were afforded in the absence of BnOH in PO/DBU feed ratio of 2000/1 under the same reaction conditions (entry 12 in Supplementary Table [1](#MOESM1){ref-type="media"}). Actually, LB could activate BnOH rapidly through H-bond interaction to form alkoxide anion, which initiate the copolymerization^[@CR28]--[@CR31]^. Hence, the stronger the basicity of the LB, the higher TOFs of the COS/PO copolymerization. This was evident by the TOF deceasing with the basicity orders of **P4** \> **P2** \> DBU ≥ MTBD \> **P1**. Being totally different from the (salen)CrCl complexes and TEB/LB pairs^[@CR9],[@CR15]^, the TU-1/LB pairs could catalyze COS/PO copolymerization in a low COS/PO feed ratio of 1.2/1, even in a feed ratio of 1.05/1 (entries 1--2, Supplementary Table [1](#MOESM1){ref-type="media"}). Since COS did not homopolymerize, the copolymerization stopped once PO was nearly completely consumed. The resultant copolymers had 100% of alternating degree (Supplementary Figs. [3](#MOESM1){ref-type="media"}--[11](#MOESM1){ref-type="media"}), indicative of much faster COS activation than successive PO enchainment. O/S ER was also effectively suppressed for all samples in Table [1](#Tab1){ref-type="table"} and Supplementary Table [1](#MOESM1){ref-type="media"}. Moreover, the control experiment ruled out the possible route of ROP of the cyclic monothiocarbonate formed (Supplementary Fig. [12](#MOESM1){ref-type="media"}). Chain microstructures of COS/PO copolymers {#Sec4} ------------------------------------------ The perfectly alternating structure and regioregularity of the PPMTCs were also revealed by MALDI-TOF MS spectroscopy (Fig. [2](#Fig2){ref-type="fig"}). Figure [2a](#Fig2){ref-type="fig"} showed one distribution of α-OH, ω-OH-terminated \[H + (PO + COS)~m + n~ + (PS + COS) + PO + OH + K^+^\] copolymer, i.e., a single PPTMC with two --OH end groups and one dithiocarbonate unit (*M*~n~: 6.5 kg/mol; PDI: 1.15). Furthermore, high-resolution ^1^H(^13^C) NMR spectra (Supplementary Fig. [13a, b](#MOESM1){ref-type="media"}) revealed that the copolymer contained two secondary --OH end groups with minimal regio-defect (one dithiocarbonate unit)^[@CR40]^. These results were indicative of inclusively regioselective attack of the sulfur anion to the CH~2~ site of PO, and thus the sole production of the T--H diad via the organocatalysis^[@CR40]^. In the presence of BnOH, only one distribution of α-OBn, ω-OH copolymer-terminated \[BnO + (PO + COS)~n~ + H + K^+^\] (Fig. [2b](#Fig2){ref-type="fig"}) with \>99% T--H diad content (Supplementary Fig. [13c, d](#MOESM1){ref-type="media"}) was obtained without dithiocarbonate unit (*M*~n~: 4.4 kg/mol; PDI: 1.14), meant that BnOH was a very efficient initiator and depressed the production of the dithiocarbonate unit.Fig. 2Analysis of PPTMCs by MALDI-TOF MS spectra. GPC curves and *M*~n~ of PPTMCs and *M*~n~ vs. repeated unit number plots are inserted. **a** PPTMC was synthesized without BnOH, afforded α-OH, ω-OH-terminated \[H + (PO + COS)~m + n~ + (PS + COS) + PO + OH + K^+^\] copolymer with one dithiocarbonate unit (*M*~n~: 6.5 kg/mol; PDI: 1.15); **b** PPTMC was synthesized with BnOH, afforded α-OBn, ω-OH copolymer-terminated \[BnO + (PO + COS)~n~ + H + K^+^\] (*M*~n~: 4.4 kg/mol; PDI: 1.14). Reaction conditions: PO/DBU/TU-1(BnOH) = 100/1/1(/1), 25 °C, 3.5 h Living COS/PO copolymerization catalyzed by Lewis pairs {#Sec5} ------------------------------------------------------- Remarkably, the TU/LB pair catalysis allowed for the copolymerization exhibiting the living features. Take the TU-1/DBU pair-catalyzed COS/PO copolymerization with or without using BnOH (Fig. [3a, b](#Fig3){ref-type="fig"}) as instances: linear increase of *M*~n~ with PO conversion, narrow PDIs (1.11--1.12, 1.16--1.19) to high PO conversion (89% and 83%, respectively). Simultaneously, the decay in monomer concentration follows zero-order kinetics under various loading of the TU-1/DBU pair in the presence or absence of BnOH or using TU-1/**P2** pair without adding BnOH (Supplementary Fig. [15](#MOESM1){ref-type="media"}). In addition, the determined molecular weights by GPC and NMR (i.e.: *M*~n~^GPC^ and *M*~n~^NMR^) were in well agreement with the calculated *M*~n~^Theo^ that increased with the increase of the \[PO\]/\[BnOH\] molar ratio in a good linear fashion (Fig. [3c](#Fig3){ref-type="fig"}, Supplementary Table [3](#MOESM1){ref-type="media"}, and Figs. [17](#MOESM1){ref-type="media"} and [18](#MOESM1){ref-type="media"}) while keeping narrow PDIs (1.17--1.19). This result confirmed that *M*~n~s of the resultant copolymers could be tuned by changing the \[PO\]/\[BnOH\] feeding ratio in the presence of TU-1/DBU pair. Of interest, further investigation showed that the use of exogenous water (even relatively large amounts, no BnOH added) could also effectively initiate the copolymerization and control the molecular weights of the resultant copolymers without changing the chain microstructure (Supplementary Table [4](#MOESM1){ref-type="media"} and Figs. [20](#MOESM1){ref-type="media"}--[22](#MOESM1){ref-type="media"}). As a result, such TU/LB pairs are robust for the copolymerization under mild conditions.Fig. 3Living character of Lewis pair-catalyzed COS/PO copolymerization. **a** Linear plots of *M*~n~ of copolymer vs. PO conversion in the absence of BnOH, \[DBU\] = \[TU-1\] = 0.2 M, 25 °C; **b** Linear plots of *M*~n~ of copolymer vs. PO conversion in the presence of BnOH (\[BnOH\] = 0.08 M), \[DBU\] = \[TU-1\] = 0.2 M, 25 °C; **c** the effect of the \[PO\]/\[BnOH\] ratio on *M*~n~s of the COS/PO copolymers \[calculated value (blue line), and determined by GPC (red line) and NMR (purple line)\]; **d** GPC traces of the PPTMCs before and after chain extension We further explored the chain extension reaction via tandem synthesis (Fig. [3d](#Fig3){ref-type="fig"} and Supplementary Fig. [23](#MOESM1){ref-type="media"}). PO and COS were copolymerized firstly using TU-1/DBU pair without using BnOH (\[PO\]: \[COS\]: \[TU-1\]: \[DBU\] = 50: 60: 1: 1) at 25 °C for 10 h. After slowly venting the unreacted COS, PO was totally consumed according to the ^1^H NMR spectra, and the resultant PPTMC exhibited unimodal peak with a *M*~n~ of 8.6 kg/mol and a PDI of 1.19. Then, 1.0 ml PO and 1.18 g COS (COS: PO = 1.2: 1) were added into the reactor and sealed for another 20 h reaction at 25 °C. PO conversion was 88% according to ^1^H NMR spectra. The resultant copolymer had a *M*~n~ of 22.3 kg/mol with a PDI of 1.26, as revealed by that GPC curve shifted overall to high molecular weight. The result of the chain extension reaction also confirmed the living mode of TU-1/DBU pair-catalyzed COS/PO copolymerization. Kinetic and mechanistic study {#Sec6} ----------------------------- The cooperative catalysis of TU-1/LB pair for the COS/PO copolymerization was firmly supported by the kinetic studies (Fig. [4a](#Fig4){ref-type="fig"}). The apparent rate constant (*k*~app~) obtained from the slopes of the best-fit lines to the plots of PO conversion vs. time, is well proportional to \[TU-1\] + \[DBU\] in the absence or presence of BnOH, suggesting that the TU-1/DBU pair behaved as a discreet catalyst species \[(*a*) in Fig. [6](#Fig6){ref-type="fig"}\]. As expected, *k*~app~ of the TU-1/**P2** pair was 13.6 ± 0.3\*10^−2^ h^−1^ and higher than that of TU-1/DBU pair (5.7 ± 0.8\*10^−2 ^h^−1^) for the copolymerization under the same conditions due to the stronger basicity of **P2**. This result is in agreement with the binding constant via ^1^H NMR dilution (see Methods, Supplementary Fig. [24](#MOESM1){ref-type="media"}) that was 10.3 ± 2.5 for TU-1/DBU pair and 21.2 ± 2.2 for TU-1/**P2** pair in equilibrium in CDCl~3~ at 25 °C (Fig. [4b](#Fig4){ref-type="fig"}). Because the H-bond interaction could be weakened by elevating the temperature^[@CR30]^, the TU-1/DBU pair catalysis for COS/PO copolymerization at high temperatures (≥55 °C) produced considerable amounts of the cyclic products (Supplementary Table [2](#MOESM1){ref-type="media"}), which is consistent with the catalysis involving only the bases for the coupling reaction of COS with PO^[@CR41]^.Fig. 4The determination of the apparent rate constant (*k*~app~) and the binding constant of TU-1/base pairs. **a** The plot of the *k*~app~ vs. \[Base\] + \[TU-1\], in neat PO (14.5 M, 14.5 mmol), \[TU-1\] = \[base\] = 0.1, 0.15, and 0.2 M; **b** The plot of the chemical shifts of two binding systems with the concentration of TU-1 and base, \[base\] = \[TU-1\] = 0.01--0.06 M in CDCl~3~. *K*~eq~ is the binding constant (at 297 K), and *k*~app~ is the apparent rate constant. Errors of *k*~app~ and *K*~eq~ are exactly the errors in the slope of the line, determined from linear regressionFig. 5The specific recognition of TU-1 to PO and DBU to COS in TU-1/DBU pair via hydrogen-bonding interaction. **a** ^1^H NMR spectra of TU-1, TU-1/COS (1/excess), TU-1/PO (1/60), **b** ^1^H NMR spectra of DBU, DBU/COS (1/excess), DBU/ PO (1/1), 0.5 M TU-1 (DBU) in CDCl~3~Fig. 6Proposed anionic copolymerization mechanism. Chain initiation and growth routes are presented upon the cooperative catalysis of TU-1/DBU pair in the presence of BnOH and H~2~O. Species (*a*)--(*d*) were revealed by Fig. [5](#Fig5){ref-type="fig"} and Supplementary Figs. [24](#MOESM1){ref-type="media"}--[26](#MOESM1){ref-type="media"}; copolymer distributions (*e*) and (*f*) were evidenced by Fig. [2](#Fig2){ref-type="fig"} We have further studied the combining capabilities of TU-1, DBU, COS, and PO in CDCl~3~ using ^1^H NMR spectra (Fig. [5](#Fig5){ref-type="fig"} and Supplementary Fig. [25](#MOESM1){ref-type="media"}). In a high concentration of TU-1 and DBU (0.5 M), the H-bond interaction was clearly revealed by the proton signal of NH group (5.65 ppm) of TU-1 disappearing while the six protons (NCH~2~) of DBU became chemically equivalent (Supplementary Fig. [25](#MOESM1){ref-type="media"}). Of interest, TU-1 was shown to solely activate PO and did not interact with COS in CDCl~3~ (Fig. [5a](#Fig5){ref-type="fig"}). Conversely, DBU could only activate COS owing to a clear deshielding effect of protons of NCH~2~ of DBU (from 2.38 to 2.77 ppm) and appearance of chemical equivalent protons of NCH~2~ of DBU rather than activate PO (Fig. [5b](#Fig5){ref-type="fig"}). Such supramolecular specific recognition of TU-1 to PO and DBU to COS in TU-1/DBU pair promoted the copolymerization cooperatively. The introduction of BnOH into the polymerization system led to the prior formation of the BnOC(=O)S^−^...DBUH^+^ owing to the deprotonation of BnOH by DBU, generated species (*b*) and (*c*) in Fig. [6](#Fig6){ref-type="fig"}, as revealed by ^1^H NMR spectra (Supplementary Fig. [26](#MOESM1){ref-type="media"})^[@CR42]^. Since PO was activated by TU-1 through H-bond \[species (*d*) in Fig. [6](#Fig6){ref-type="fig"}, revealed by Fig. [5b](#Fig5){ref-type="fig"}\], the chain growth could be accelerated (Table [1](#Tab1){ref-type="table"}). On the other hand, in the absence of BnOH, trace water (i.e., R=H, Fig. [6](#Fig6){ref-type="fig"}) could also initiate the copolymerization via the similar activation route of BnOH by TU/LB pair. Such H~2~O initiation led to the formation of the end --S(O=C)--OH group, which was thermodynamically unstable and thus converted to --SH group via decarboxylation process^[@CR6],[@CR40]^. Since the end --SH group has the stronger acidity than --OH group, it was rapidly deprotonated, generating bifunctional initiator for further chain growth, a copolymer with two end secondary --OH groups and one dithiocarbonate unit (*f*) was produced, which was clearly revealed by Fig. [2a](#Fig2){ref-type="fig"} (Supplementary Figs. [13](#MOESM1){ref-type="media"} and [22](#MOESM1){ref-type="media"}). The impact of Lewis pairs on COS/epoxides copolymerizations {#Sec7} ----------------------------------------------------------- Different Lewis pairs including other thioureas were also investigated for the copolymerization of COS with several epoxides. Two other thioureas, TU-2 and TU-3 were synthesized^[@CR43]^ and successfully utilized for the COS/PO copolymerization (entries 1--4 in Table [2](#Tab2){ref-type="table"}). H-bond interactions of both TU-2 and TU-3 with DBU are similar to that of the DBU/TU-1 pair (Supplementary Fig. [28](#MOESM1){ref-type="media"}). The TU-3/DBU pair was more active than TU-1 (TU-2)/DBU pairs for the copolymerization, but with a slightly lower copolymer selectivity of 93%. Both TU-2 and TU-3 paired with **P2** showed the same TOFs (29 h^−1^, entries 3--4, Table [2](#Tab2){ref-type="table"}) for the COS/PO copolymerization. Totally, TU-3 with more electron-withdrawing group resulted in the production of slightly more amounts of cyclic products. In addition, two epoxides, glycidyl phenyl ether (PGE) and cyclohexene oxide (CHO) were copolymerized with COS by using **P2**/TU-1, **P4**/TU-1 pairs, affording fully alternating copolymers (entries 5--8, Table [2](#Tab2){ref-type="table"}). Thereof, the COS/PGE copolymerization were fully regioselective with T--H diad content of \>99% (Supplementary Fig. [29](#MOESM1){ref-type="media"}), while COS/CHO copolymerization could proceed at 40 °C, afforded well-defined copolymer with perfect alternating degree and \>99% copolymer selectivity (Supplementary Fig. [30](#MOESM1){ref-type="media"}). These results illustrate the use of several low-cost thioureas for the copolymerization of COS and epoxides with different structures.Table 2The copolymerization of COS with various epoxides by using various TU/LB pairsEntry^a^EpoxideTU/LBEpoxide: LB: TU-1TOF (h^--1^)^b^Copolymer selectivity^c^Alternating degree (%)^c^T--H diad content (%)^d^O/S ER product^d^*M*~n~ (kg/mol)^e^PDI^e^1POTU-2/DBU500:1:1697/3100\>99N.F.21.91.152POTU-3/DBU500:1:11093/7100\>99N.F.21.91.183POTU-2/**P2**1000:1:12998/2100\>99N.F.19.81.104POTU-3/**P2**1000:1:12996/4100\>99N.F.32.21.205PGETU-1/**P2**1000:1:12296/4100\>99N.F.29.31.266PGETU-1/**P4**4000:1:15296/4100\>99N.F.17.51.187^f^CHOTU-1/**P2**250:1:12\>99100---N.F.15.81.198^f^CHOTU-1/**P4**250:1:15\>99100---N.F.13.41.16Representative NMR spectra are in Supplementary Figs. [29](#MOESM1){ref-type="media"}, [30](#MOESM1){ref-type="media"}*PGE* glycidyl phenyl ether, *CHO* cyclohexene oxide^a--e^Reaction conditions and characterization methods were the same with Table [1](#Tab1){ref-type="table"}^f^40  °C. Discussion {#Sec8} ========== We have described the synthesis of perfectly alternating and regioregular poly(monothiocarbonate)s from COS and epoxides by employing thioureas and organic bases under mild conditions. Of importance, the use of thioureas and BnOH led to living/controlled COS/epoxide copolymerization, improved catalytic activity and copolymer selectivity than previous systems. One of the key features of such metal-free catalyst systems is that it can be applied to a variety of epoxides. Due to the ease of tailoring the molecular structures of these organic thioureas and bases, this strategy is a promising alternative to get metal-free well-defined sulfur-containing polymers with high activity and selectivity. Our ongoing efforts are to seek a better understanding of the mechanistic aspects of such catalytic processes, and to develop chiral TU/LB pairs for the copolymerization. Methods {#Sec9} ======= Materials {#Sec10} --------- Propylene oxide (PO), glycidyl phenyl ether (PGE), and cyclohexene oxide (CHO) were purified by distillation after stirring with calcium hydride for 3 days. 1,3-Diisopropyl-2-thiourea (TU-1) was purchased from Sigma Aldrich and sublimeed before use. 1-cyclohexyl-3-phenylthiourea (TU-2) and 1-\[3,5-bis(trifluoromethyl) phenyl\]-3-cyclohexylthiourea (TU-3) were synthesized according to literature^[@CR43]^. ^1^H NMR spectra of TU-2 and TU-3 are shown in Supplementary Fig. [27](#MOESM1){ref-type="media"}. *Tert*-butylimino-tris(dimethylamino) phosphorene (^*t*^Bu-P~1~, **P1**, 97%), 1-*tert-*Butyl-2,2,4,4,4-pentakis(dimethylamino)-2λ^5^, 4λ^5^-catenadi (phosphazene)(^*t*^Bu-P~2~, **P2**, \~2.0 M in THF) and 1-*tert-*butyl-4,4,4-tris(dimethylamino)-2,2-bis \[tris(dimethylamino)- phosphoranylidenamino\]-2λ^5^, 4λ^5^-catenadi(phosphazene) (^*t*^Bu-P~4~, **P4**, \~0.8 M in hexane) were purchased from Sigma and used directly. 1,5,7-Triazabicyclo\[4,4,0\]dec-5-ene (TBD) was purchased from Aldrich Chemical Co, which were purified by dissolving in toluene over CaH~2~, filtering after stirring overnight, and removing the solvent. Benzyl alcohol (BnOH), *N*-methyl-1,5,7-triazabicyclododecene (MTBD) and 1,8-diazabicyclo\[5,4,0\]undec-7-ene (DBU) were purchased from Alfa Aesar Chemical Co. and Aldrich Chemical Co., respectively, which were purified by distillation over distillation over CaH~2~ and stored in an inert gas (N~2~)-filled glove box. Sodium hydride (95%) was purchased from Sigma and used directly. Carbonyl sulfide (COS) (99.9%) was purchased from the APK (shanghai) Gas Company LTD and used as received. Representative procedure for copolymerization reactions {#Sec11} ------------------------------------------------------- All polymerizations were carried out in glove box under N~2~ atmosphere unless otherwise specified. A 10-ml autoclave with magnetic stirrer was first dried in an oven at 110 °C overnight, then immediately placed into the glove box chamber. After keeping under vacuum for 1--2 h, the reaction vessel was put into the glove box under nitrogen atmosphere. The copolymerization of COS with PO described below is taken from entry 17 in Table [1](#Tab1){ref-type="table"} as an example. TU-1 (1.4 mg, 0.007 mmol) was added to the reactor firstly. PO (1.0 ml, 14 mmol) was then carefully added into the vessel. Afterwards, DBU (1.05 μl, 0.007 mmol) and BnOH (0.75 μl, 0.007 mmol) were added into the reactor, respectively. The reactor was sealed and then taken out for charging with the set amounts of COS. The copolymerization was performed at 25 °C for 24 h. Then, the reactor was cooled in ice-water bath, and the unreacted COS was slowly vented. An aliquot was taken from the resulting crude product for the determination of the PO conversion and the molar ratio of copolymer/cyclic products by ^1^H NMR spectrum. Traces of acetic acid were then added for ^1^H NMR spectrum, in order to prevent degradation of the crude product. Next, the crude product was quenched with HCl in ethanol (1 mol/l). The crude product was dissolved with CH~2~Cl~2~ and then precipitated in cold methanol. The product was collected by centrifugation and dried in vacuum at 40 °C until a constant weight. Characterization {#Sec12} ---------------- ^1^H and ^13^C NMR spectra were performed on a Bruker Advance DMX 400 MHz or 600 MHz spectrometer. And chemical shift values were referenced to TMS at 0 ppm for ^1^H NMR and ^13^C NMR. The number-average molecular weight (*M*~n~) and molecular weight distribution (*Ð* = *M*~w~/*M*~n~) of the resultant copolymers were determined with a PL-GPC220 chromatograph (Polymer Laboratories) equipped with an HP 1100 pump from Agilent Technologies. The GPC columns were eluted with THF with 1.0 ml/min at 40 °C. The sample concentration was 0.4 wt. %, and the injection volume was 50 μl. Calibration was performed using monodisperse polystyrene standards. Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometric measurements were performed on a Waters MALDI Micro MX mass spectrometer, equipped with a nitrogen laser delivering 3 ns laser pulses at 337 nm. Dithranol (97%, Alfa), was used as the matrix. CH~3~COOK ( ≥ 98%, Aladdin) was added for ion formation. Binding constant studies {#Sec13} ------------------------ Equations used for binding studies^[@CR44]^:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{For}}\,{\mathrm{dilution}}:\frac{{\delta _0 - \delta _i}}{{[A]_0}} = - 2K_{{\mathrm e}{\mathrm q}}(\delta _0 - \delta _i) + K_{{\mathrm e}{\mathrm q}}(\delta _0 - \delta _\infty )$$\end{document}$$where: *δ*~0~ is the chemical shift of the *ortho*-protons of pure TU-1; \[*A*\]~0~ is concentration of TU-1 or a base; *δ*~i~ is the chemical shift of the *ortho*-protons of TU-1 in the solution when \[TU-1\] = \[base\] = \[*A*\]~0~; *δ*~∞~ is the chemical shift *ortho*-protons of the "pure complex" TU-1 with a base, (*δ*~0~ -- *δ*~∞~) is a constant; *K*~eq~ is the binding constant between TU-1 and a Base. Following a similar procedure reported by Matthew K. Kiesewetter^[@CR29]^, NMR dilution experiments were carried in CDCl~3~ with the concentration of \[TU-1\] = \[base\] varied from 0.01 M to 0.06 M in CDCl~3~. DBU and **P2** were employed as bases respectively. The ^1^H NMR spectra was shown in Supplementary Fig. [24](#MOESM1){ref-type="media"}. The binding constants (*K*~eq~) were determined from the slope of the linear forms of the binding equation (above). And the error in *K*~eq~ is exactly the error in the slope of the line, which can be determined from linear regression. Calculation of copolymer selectivity, PO conversion and TOF {#Sec14} ----------------------------------------------------------- Copolymer selectivity and PO conversion were calculated based on the ^1^H NMR spectrum of the crude product. Taking entry 11 in Table [1](#Tab1){ref-type="table"} as an example, spectrum of the crude product was showed in Supplementary Fig. [35](#MOESM1){ref-type="media"}. Protons with chemical shifts at 4.80, 3.52, 3.26 and 1.51 ppm belong to the methenyl (**d**), methene (**e**), and methyl (**f**) groups, respectively, the peaks in 5.16, 3.08, and 1.36 ppm \[also seen in Supplementary Fig. [5](#MOESM1){ref-type="media"}, a purified PPTMC\] belong to the methenyl (**a**), methene (**b**), and methyl (**c**) groups in the copolymer. And the area ratio of these two parts was taken as the copolymer selectivity. The corresponding peaks in 3.02, 2.77, and 2.46, 1.32 ppm belong to methenyl (**h**), methene (**g**) and methyl (**i**) in PO which was not consumed. On the base of ^1^H NMR spectra, we have:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Copolymer}}\,{\mathrm{selectivity}} = \frac{{A_{5.16}}}{{A_{5.16} + A_{4.82}}}\times 100\%$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{PO}}\,{\mathrm{conversion}} = \left( {1 - \frac{{A_{2.77}}}{{A_{5.16} + A_{4.82} + A_{2.77}}}} \right)\times 100\%$$\end{document}$$ Thus,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{TOF}}\,({\mathrm h}^{ - 1}) = {\mathrm{PO}}\,{\mathrm{conversion}}\times \frac{{[{\mathrm P}{\mathrm O}]/[{\mathrm L}{\mathrm B}]}}{t}$$\end{document}$$ Data availability {#Sec15} ----------------- The authors declare that the data supporting this study are available within the paper and its Supplementary Information File. All other data is available from the authors upon reasonable request. Electronic supplementary material ================================= {#Sec17} Supplementary Information Peer Review File **Electronic supplementary material** **Supplementary Information** accompanies this paper at 10.1038/s41467-018-04554-5. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work was supported by the National Science Foundation of the People's Republic of China (no. 21774108) and the Distinguished Young Investigator Fund of Zhejiang Province (LR16B040001). We thank Prof. Jun-Peng Zhao (South China University of Technology, China) and Prof. Donald J. Darensbourg (Texas A&M University, USA) for discussion and suggestions. X.-H.Z. conceived, designed and directed the investigations, and revised the manuscript. C.-J.Z. carried out all experiments and analyses, contributed much for the sole catalysis of organic bases for the copolymerization and wrote the draft. H.-L.W. and J.-L.Y. carried out some NMR, MALDI-TOF-MS characterizations and analyses, and discussed with Y.L. for mechanism and kinetic study. Competing interests {#FPar1} =================== The authors declare no competing interests.

Introduction {#s1} ============ There is increasing interest in the use of the pre-surgical window study design to determine the efficacy of novel and repurposed cancer therapeutics in a time- and cost-effective manner. Such a strategy relies on the accurate quantification of biomarkers, which act as surrogates for longer term clinical outcomes and disease response. In endometrial cancer, this has traditionally involved comparing a baseline endometrial biopsy with tumor sampled from the hysterectomy specimen following intervention. This does, however, potentially raise a methodological issue. The endometrial biopsy samples the cancer *in situ* and is thus an accurate representation of tumor biology. The hysterectomy specimen, by contrast, is subject to a variable period of hypoxia once the uterine arteries are clamped during surgery and before it is removed from the body, followed by cold ischemia until formalin fixation occurs. Many of the biomarkers interrogated as outcome measures in endometrial cancer window studies are activated and deactivated through phosphorylation and dephosphorylation, events which have been shown to be transient and highly sensitive to hypoxia ([@B1]). Indeed, as little as 10 min of anoxia has been shown to be sufficient to induce significant biochemical alterations ([@B2]). Expression of phospho-Akt (protein kinase B) in colorectal tumors was found to be completely absent from surgically resected specimens where there had been an interruption to the blood supply of 20 min or more despite being present in the same tumor sampled earlier by biopsy ([@B1]). In breast cancer, the size of the sample has also been found to be of importance, with loss of pERK1/2 expression occurring in larger specimens ([@B3]). This was likely to be the consequence of the slow penetration of formalin and delays in the formation of stabilizing cross links between formaldehyde and proteins, which would have prevented their degradation ([@B4]). Reliance on surgically excised specimens for accurate readouts of tumor biology could, therefore, be risky ([@B5]). If these findings were replicated in endometrial cancer, they would potentially call into question results from earlier window studies that determined drug efficacy on the basis of immunohistochemical expression of proteins in the hysterectomy specimen ([@B6]--[@B8]). Many of these lacked contemporaneous control groups for comparison and when these have been incorporated, reductions in biomarker expression in both the active drug and untreated arms suggest that reduced protein expression may be common to all surgically excised malignancies ([@B9]). This study sought to determine whether there is a significant difference in immunohistochemical expression of commonly studied biomarkers in endometrial cancer window studies, including Ki-67, phosphorylated markers of the PI3K-Akt-mTOR and insulin signaling pathways and hormone receptors between an endometrial biopsy taken immediately prior to the start of surgery and the hysterectomy specimen. Differences in expression of endometrial cancer stem cell markers were also explored as these are likely to be increasingly used in future window studies as surrogate outcome measures. The impact of intra-tumoral hypoxia and delays in fixation of tissues on protein expression were also studied. Materials and Methods {#s2} ===================== Patient and Tissue Selection ---------------------------- Tumor tissue was obtained from patients recruited into PREMIUM, a placebo-controlled, randomized trial of pre-surgical metformin for the treatment of atypical hyperplasia and endometrioid adenocarcinoma of the endometrium at five hospitals in the North West of England ([@B10]). The study found no effect of metformin on endometrial cancer cell proliferation, as determined by Ki-67 expression, with short-term treatment. The trial was approved by the North West Research Ethics Committee (14/NW/1236) and prospectively registered on the UK (ISRCTN 88589234) clinical trial database. All participants provided written, informed consent. Samples included an endometrial biopsy taken with a vacuum aspiration device immediately prior to the start of surgery and representative tumor blocks taken from the hysterectomy specimen itself. Patients with two matched samples taken on the same day were included in the final analysis, regardless of the allocated trial treatment arm. After being obtained, the endometrial biopsies were immediately formalin fixed in theater and subsequently embedded in paraffin. The hysterectomy specimen was either placed straight away in formalin or was transferred dry to the local pathology department for a directed tumor biopsy before being fixed and paraffin embedded. The tumor biopsy was used for other research projects. Four-micrometer sections were cut from the tumor block using a microtome and mounted onto histological glass slides before undergoing immediate Ki-67 immunohistochemistry. This slide preparation technique was used as it has been previously demonstrated to be the most reliable and reproducible method for quantifying Ki-67 expression ([@B11]). In order to conserve tumor tissue, expression of all other markers was determined on tumor microarrays (TMAs). These were constructed using triplicate cores from representative areas of the tumor identified by an experienced gynecological histopathologist (JS) on hematoxylin and eosin stained slides. Immunohistochemistry -------------------- Immunohistochemistry was performed using the Leica Bond Max (Leica Biosystems, Wetzlar, Germany) and heat-induced epitope retrieval, unless otherwise stated. Full details of the antibodies and conditions used are given in [Table 1](#T1){ref-type="table"}. Primary antibody incubation was for 1 h with the exception of the hormone receptors where incubation was for 20 min. Primary antibody detection was performed using the Refine Detection Kit (Leica Biosystems), which utilizes a rabbit anti-mouse IgG secondary antibody and anti-rabbit poly-HRP IgG antibody and includes 3,3\'-diaminobenzidine (DAB) as a chromogen. Staining for hormone receptors was performed in the clinical histopathology laboratory at Manchester University NHS Foundation Trust, using the automated Ventana BenchMark ULTRA IHC / ISH Staining Module (Ventana, Tucson, AZ, USA) and a horseradish peroxidase linked secondary antibody, with DAB as a chromogen and a substrate and copper enhancer. Counterstaining of all slides was performed using hematoxylin and negative (isotype) and positive controls were included for each antibody ([Table 1](#T1){ref-type="table"}). ###### Antibodies and conditions used for immunohistochemistry. **Antibody** **Manufacturer** **Catalog number** **Dilution** **Host** **Antigen retrieval** **Blocking step?** **Positive control tissue** --------------------- ------------------ -------------------- -------------- ------------------- ----------------------- ---------------------- ----------------------------- Ki-67 (MIB-1 clone) Dako X0931 1:100 Monoclonal mouse EDTA pH9 Included-casein Tonsil pAkt (Ser 473) Cell signaling \#4060 1:50 Rabbit monoclonal EDTA pH9 Not included Breast cancer p4EBP1 (Thr37/46) Cell signaling \#2855 1:800 Rabbit monoclonal EDTA pH9 Included Colon cancer pIR (Y1361) Abcam ab60946 1:1000 Rabbit polyclonal EDTA pH9 Included Placenta pIGF1R Abcam ab39398 1:50 Rabbit polyclonal Citrate buffer pH6 Included Placenta ER (SP1) Ventana 790--4,324 RTU Rabbit monoclonal EDTA pH 8.4 Included-ultraviolet Uterus PR (1E2) Ventana 790--2,223 RTU Rabbit monoclonal EDTA pH 8.4 Included-ultraviolet Uterus CD133 Miltenyi 130-090-422 1:25 Mouse monoclonal Citrate buffer pH6 Included-casein Colon cancer ALDH BD Biosciences BD 611194 1:100 Mouse monoclonal EDTA pH9 Included-casein Liver HIF-1α BD Biosciences BD 610959 1:50 Mouse monoclonal EDTA pH9 Not included Tonsil CA-IX Novus NB100-417 1:2000 Rabbit polyclonal EDTA pH9 Not included Renal cell carcinoma *P4EBP1, phospho-4EBP1; pIR, phospho-insulin receptor; pIGF1R, phospho-insulin like growth factor 1 receptor; ER, estrogen receptor; PR, progesterone receptor; ALDH, aldehyde dehydrogenase; RTU, ready to use; EDTA, Ethylenediaminetetraacetic acid*. Immunohistochemical Scoring --------------------------- Slides were digitized using the Leica SCN400 Slide Scanner (Leica Microsystems, Wetzlar, Germany). Semi-automated scoring was performed using Definiens Developer software, in which an optimized solution is created to accurately identify staining of different intensity and is applied to manually selected endometrial cancer glands. The correct classification of staining within cells was checked manually by two independent researchers (SK and ZM), blinded to sample type. One researcher scored all tumors, while the second independently scored a proportion (20%) to ensure consistency. The use of the semi-automated Definiens Developer software to quantify immunohistochemical expression was more time efficient than manual scoring and associated with greater reliability and reproducibility of the scores obtained ([@B11]). Quantification of Ki-67 staining was performed using the hot spot approach, in which the three areas of greatest Ki-67 expression across the whole slide were selected at x10 magnification, in accordance with previously published recommendations ([@B11]). The percentage of positively stained nuclei was recorded, regardless of staining intensity. For all other markers, all malignant glands within the triplicate cores on the TMAs were scored in their entirety. With the exception of the cancer stem cell activity markers, CD133 and ALDH, staining was quantified using an H-score, which is the product of staining intensity (0 = none, 1 = weak, 2 = moderate, 3 = strong) and the percentage of cells of that intensity and has a maximum value of 300. Cells were regarded as positive if there was evidence of staining within the nucleus only (ER, PR, and HIF1α), nucleus and cytoplasm (pAkt, p4EBP1) or at the cell membrane and/or cytoplasm (pIR, pIGF1R, CA-IX). CD133 and ALDH were scored as the percentage of cells with positive apical membrane or cytoplasmic staining, respectively, regardless of staining intensity, in keeping with previous work ([@B12], [@B13]). Data Collection --------------- Demographic and pathological data were collected by interview and from electronic and paper medical records. Where pathology services were off site, specialist gynecological pathologists were not available to perform directed tumor biopsies. Instead, the hysterectomy specimen was placed directly into formalin and potentially remained in the fixative for longer than those specimens removed at centers where pathologists were present onsite. Day of surgery was included as a surrogate marker of duration of exposure of the hysterectomy specimen to formalin, with specimens removed at the end of the working week placed in the fixative for longer due to the weekend break. Statistical Analysis -------------------- Data were summarized using means and standard deviations. The immunohistochemical scores of endometrial biopsies and corresponding hysterectomy specimens were compared using Wilcoxon signed-rank test. Inter-observer variability was assessed using the intra-class correlation coefficient. Comparisons of continuous, normally distributed data were performed using the Student *T*-test and of ordinal variables by Pearson\'s correlation coefficient. A *p* ≤ 0.05 was considered statistically significant, with asterisks used to denote significant results as ^\*^*p* ≤ 0.05, ^\*\*^*p* ≤ 0.01, ^\*\*\*^*p* ≤ 0.001 and ^\*\*\*\*^*p* ≤ 0.0001. The statistical analysis was conducted using SPSS version 23, Stata version 14 and Graph Pad Prism 7. Results {#s3} ======= Matched endometrial biopsies and hysterectomy specimens of sufficient size and quality to be suitable for analysis were available for 75 patients. Assessment of inter-observer variability found excellent agreement between the two scorers with kappa values of 0.846 and 0.758 for the assessment of IHC staining in endometrial biopsies and hysterectomy specimens, respectively. Effect of Tumor Sampling Technique on Commonly Studied Biomarkers in Endometrial Cancer --------------------------------------------------------------------------------------- Ki-67 expression was significantly lower in the hysterectomy specimen compared with the corresponding endometrial biopsy (*p* \< 0.0001, [Figure 1](#F1){ref-type="fig"}). Mean Ki-67 expression in the endometrial biopsy was 41.8% (SD 18.6%) compared with 33.2% (SD 18.7%) in the hysterectomy specimen. {#F1} A significant reduction in expression of phosphorylated markers of the PI3K-Akt-mTor and insulin signaling pathways was also found in the hysterectomy specimen (all *p* \< 0.0001). The mean H-score for pAkt in the endometrial biopsy compared with the hysterectomy specimen was 42.6 (SD 40.0) vs. 3.3 (SD 4.8), 79.3 (SD 40.7) vs. 6.7 (SD 21.5) for p4EBP1, 200.8 (SD 38.2) vs. 55.1 (SD 50.9) for pIR and 221.8 (SD 40.1) vs. 126.2 (SD 82.7) for pIGF1R. In many cases, expression of markers was completely lost in the hysterectomy specimen ([Figure 2](#F2){ref-type="fig"}). {#F2} Similar results were found when expression of the hormone receptors ER and PR were compared between the two tumor sampling techniques ([Figure 3](#F3){ref-type="fig"}). The mean H-score for ER expression decreased from 271 (SD 48.2) in the endometrial biopsy to 211.6 (SD 69.5) in the hysterectomy specimen (*p* \< 0.0001), whilst PR expression decreased from 206.9 (SD 67.1) to 143.1 (SD 69.7, *p* \< 0.0001). Despite lower expression of the estrogen receptor in the hysterectomy specimen, there was no discrepancy in overall ER status between the two tumor samples. This was not the case when expression of the progesterone receptor was considered; loss of PR expression in the hysterectomy specimen resulted in the misclassification of two out of 63 tumors (3.2%) as being PR negative. {#F3} In contrast to the above, expression of markers of cancer stem cell activity, CD133 and ALDH, were unaffected by sampling technique. Mean expression of CD133 was 3.3% (SD 4.7%) in the endometrial biopsy and 2.7% (SD 3.7%) in the hysterectomy specimen (*p* = 0.48). There was a non-significant increase in ALDH expression in the hysterectomy specimen, with a mean of 64.8% (SD 24.4%) of cells staining positive compared with 58.4% (SD 26.3%) in the endometrial biopsy. The magnitude of loss of expression of Ki-67, pAkt, p4EBP1, pIR, pIGF1R, ER, and PR in the hysterectomy specimen was closely correlated with baseline expression in the endometrial biopsy. The higher the expression in the endometrial biopsy, the greater the loss of expression in the hysterectomy specimen (all *p* ≤ 0.001). The extent to which loss of expression of each individual biomarker was associated with loss of expression of the other biomarkers was examined ([Table 2](#T2){ref-type="table"}). Loss of expression of p4EBP1 in hysterectomy specimens was associated with significant reductions in expression of Ki-67 (*r* = 0.24, *p* = 0.05, 68 patients) and pIGF1R (*r* = 0.32, *p* = 0.007, 71 patients). Loss of expression of pIR correlated with a reduction in expression of pAkt (*r* = 0.23, *p* = 0.05, 70 patients), whilst loss of expression of pIGF1R was associated with reductions in expression of ER (*r* = 0.29, *p* = 0.02, 63 patients) and PR (*r* = 0.36, *p* = 0.004, 61 patients). Loss of PR expression also correlated with loss of expression of pIR (*r* = 0.33, *p* = 0.008, 62 patients), pIGF1R (*r* = 0.36, *p* = 0.004, 61 patients) and, in particular, ER (*r* = 0.65, *p* \< 0.0001, 61 patients). ###### Pairwise correlation matrix of differences in biomarker expression between endometrial biopsies and hysterectomy specimens. ---------------------------------------- ---------------------------------------- ------------------------------------------- ------------------------------------------- ------------------------------------------- -------------------------------------------------- --------- Ki-67 diff 0.14\ pAkt diff 0.2665 0.24\ 0.03\ p4EBP1 diff 0.05[^\*^](#TN1){ref-type="table-fn"}\ 0.7869 68 −0.02\ 0.23\ 0.22\ pIR diff 0.9067 0.05[^\*^](#TN1){ref-type="table-fn"}\ 0.0672 70 0.17\ −0.07\ 0.32\ 0.10\ pIGF1R diff 0.1666 0.5769 0.007[^\*\*^](#TN2){ref-type="table-fn"}\ 0.4272 71 0.13\ −0.04\ −0.14\ 0.22\ 0.29\ ER diff 0.3360 0.7363 0.2763 0.08 64 0.02[^\*^](#TN1){ref-type="table-fn"}\ 63 0.17\ 0.03\ 0.12\ 0.33\ 0.36\ 0.65\ PR diff 0.2059 0.8361 0.3662 0.008[^\*\*^](#TN2){ref-type="table-fn"}\ 0.004[^\*\*^](#TN2){ref-type="table-fn"}\ \<0.0001[^\*\*\*\*^](#TN3){ref-type="table-fn"}\ 62 61 61 ---------------------------------------- ---------------------------------------- ------------------------------------------- ------------------------------------------- ------------------------------------------- -------------------------------------------------- --------- *Results are presented as Pearson\'s correlation coefficient, p-value and number of samples compared*. *p ≤ 0.05*, *p ≤ 0.01*, *p ≤ 0.0001*. Loss of biomarker expression in the hysterectomy specimen was compared against clinico-pathological variables ([Table 3](#T3){ref-type="table"}). Loss of expression was not associated with age, BMI, exposure to metformin or placebo, depth of myometrial invasion or need for adjuvant therapy (all *p* \> 0.05). ###### Correlation between loss of expression in hysterectomy specimen and clinico-pathological variables. **Ki-67 diff** **pAkt diff** **p4EBP1 diff** **pIR diff** **pIGF1R diff** **ER diff** **PR diff** ---------------------------------- ---------------- ------------------------------------------ --------------------------------------- -------------- ----------------- ------------------------------------------ ------------- **AGE** \<60 years 9.4 (16.7) 24.2 (24.2) 55.9 (45.0) 138.6 (62.9) 105.3 (102.4) 100.7 (69.3) 79.9 (86.7) ≥60 years 8.2 (15.7) 45.1 (42.2) 78.9 (46.8) 149.7 (60.2) 89.4 (99.8) 42.1 (68.1) 56.8 (88.3) p value 0.78 0.01[^\*^](#TN4){ref-type="table-fn"} 0.06 0.49 0.54 0.004[^\*\*^](#TN5){ref-type="table-fn"} 0.34 **BMI** \<30kg/m^2^ 6.3 (16.7) 43.4 (46.7) 65.5 (53.6) 149.0 (58.4) 109.6 (96.0) 61.1 (57.1) 48.5 (81.0) ≥30kg/m^2^ 10.2 (15.3) 35.9 (33.1) 76.4 (42.2) 144.6 (62.9) 83.8 (102.8) 58.4 (83.0) 74.5 (91.8) *p*-value 0.33 0.47 0.36 0.76 0.28 0.88 0.24 **GRADE** 1+2 9.5 (15.2) 39.6 (35.2) 72.7 (44.3) 147.3 (61.2) 92.1 (98.5) 60.8 (59.9) 68.1 (67.7) 3 3.5 (18.9) 35.0 (54.5) 67.9 (62.0) 141.4 (60.6) 106.0 (112.3) 52.7 (127.4) 59.8 (70.5) *p*-value 0.34 0.79 0.8 0.76 0.7 0.85 0.76 **STAGE** 1+2 9.1 (16.4) 43.1 (40.0) 73.2 (46.5) 145.8 (61.5) 95.7 (97.9) 63.2 (59.3) 71.4 (84.4) 3 4.8 (11.7) 15.6 (19.7) 63.7 (52.7) 149.3 (59.3) 87.2 (117.1) 36.7 (132.9) 36.4 (87.6) p value 0.38 0.002[^\*\*^](#TN5){ref-type="table-fn"} 0.6 0.86 0.82 0.57 0.32 **DEPTH OF MYOMETRIAL INVASION** \<50% 8.2 (15.4) 37.7 (36.8) 72.8 (54.0) 140.5 (71.7) 76.2 (107.8) 60.6 (57.8) 65.6 (98.7) ≥50% 9.1 (18.0) 42.3 (44.1) 69.6 (37.5) 157.0 (35.8) 117.0 (83.6) 61.2 (95.5) 63.2 (72.1) p value 0.83 0.66 0.77 0.21 0.08 0.98 0.92 **LVSI** Absent 9.6 (15.4) 42.5 (36.4) 79.4 (47.4) 150.0 (64.4) 91.9 (102.1) 63.1 (59.3) 70.6 (94.8) Present 5.4 (18.7) 32.3 (45.9) 50.3 (44.2) 139.2 (50.4) 90.0 (100.2) 54.8 (105.4) 46.9 (69.7) *p*-value 0.42 0.38 0.02[^\*^](#TN4){ref-type="table-fn"} 0.46 0.94 0.76 0.31 **LN METS** Absent 8.9 (16.4) 41.2 (39.9) 73.4 (46.0) 148.2 (59.6) 94.0 (98.4) 60.5 (74.2) 65.0 (92.0) Present 2.7 (13.4) 21.6 (31.4) 46.8 (71.4) 133.0 (75.2) 63.7 (132.0) 65.0 (82.0) 60.9 (34.2) p value 0.43 0.2 0.46 0.65 0.61 0.91 0.85 **ADJUVANT THERAPY** No 8.7 (13.8) 41.4 (37.5) 70.4 (48.9) 139.2 (69.2) 82.4 (104.2) 56.9 (59.0) 59.9 (97.3) Yes 8.3 (19.2) 37.1 (42.2) 72.9 (47.7) 156.6 (46.9) 103.4 (96.6) 65.4 (89.7) 71.0 (78.9) p value 0.93 0.67 0.83 0.21 0.39 0.67 0.63 *Results are presented as mean difference in expression between endometrial biopsy and hysterectomy specimen (SD), LN mets lymph node metastases*. *p ≤ 0.05*, *p ≤ 0.01*. Impact of Poor Fixation on Loss of Immunohistochemical Expression ----------------------------------------------------------------- In order to investigate the extent to which delays in achieving adequate tissue fixation were responsible for loss of immunohistochemical expression in the hysterectomy specimen, differences in staining were correlated with specimen characteristics and tissue handling ([Table 4](#T4){ref-type="table"}). No clear association was found between loss of expression and tumor size, specimen weight, day of the week on which the surgery was performed or whether pathology services were located on- or offsite (all *p* \> 0.05). The difference in expression between the endometrial biopsy and hysterectomy specimen was smaller if the uterus was removed by laparotomy than total laparoscopic hysterectomy, although the result was not statistically significant. Open surgery is generally associated with a shorter hypoxic time for the uterus than minimal access techniques. There was a trend toward smaller differences in immunohistochemical expression between endometrial biopsies and the hysterectomy specimen if the uterus had been bisected prior to being placed in formalin, although, with the exception of p4EBP1, this did not reach statistical significance. Whilst there was no overall correlation between loss of expression and tumor-serosal distance, this relationship was strengthened when only unopened uteri were considered. ###### Correlation between loss of immunohistochemical expression and specimen characteristics and tumor handling. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **Ki-67 diff** **pAkt diff** **p4EBP1 diff** **pIR diff** **pIGF1R diff** **ER diff** **PR diff** -------------------------------------------------- ---------------- ------------------------------------------ --------------------------------------- --------------- ----------------- --------------- --------------- Type of hysterectomy TLH 9.2 (19.0) 43.9 (41.8) 81.5 (49.0) 139.7 (62.3) 112.4 (101.3) 62.1 (66.7) 74.9 (98.6) TAH 7.9 (14.6) 39.1 (38.2) 67.9 (47.9) 156.8 (53.8) 79.0 (102.4) 59.3 (81.4) 57.2 (84.2) *p-*value 0.77 0.64 0.27 0.25 0.19 0.89 0.48 Uterus bisected No 10.1 (14.5) 36.2 (42.1) 84.7 (38.8) 148.7 (65.4) 114.9 (92.9) 63.8 (84.4) 71.8 (99.5) Yes 7.1 (17.2) 41.4 (35.5) 59.5 (51.6) 144.0 (56.7) 74.5 (104.2) 54.9 (59.8) 53.7 (70.9) *p*-value 0.44 0.58 0.02[^\*^](#TN6){ref-type="table-fn"} 0.75 0.08 0.63 0.4 Tumour-serosal distance (mm) −0.06 0.01 −0.01 0.09 −0.03 0.07 −0.01 0.66 0.95 0.96 0.45 0.79 0.63 0.94 65 65 68 68 68 59 58 Tumour-serosal distance (unopened specimens, mm) −0.15 −0.13 −0.06 0.15 0.04 0.11 0.03 0.43 0.47 0.76 0.41 0.81 0.56 0.86 31 31 32 32 32 29 30 Tumour size (largest dimension, mm) −0.01 −0.34 0.07 −0.09 0.11 −0.05 −0.11 0.94 0.01[^\*\*^](#TN7){ref-type="table-fn"} 0.62 0.52 0.43 0.73 0.48 51 52 54 54 55 46 43 Specimen weight (g) 0.08 −0.29 −0.07 −0.01 0.06 0.02 −0.11 0.57 0.05[^\*^](#TN6){ref-type="table-fn"} 0.64 0.97 0.66 0.88 0.46 52 50 52 52 51 45 44 Day of surgery 0.12 0.01 −0.04 −0.11 −0.04 −0.18 −0.21 0.35 0.97 0.75 0.35 0.73 0.16 0.12 66 67 70 70 70 61 60 Pathology service On site\ 9.2 (16.3)\ 41.5 (40.0)\ 69.6 (47.5)\ 146.3 (57.1)\ 90.4 (101.6)\ 53.2 (71.7)\ 56.5 (87.3)\ Off site\ 3.5 (11.3)\ 17.7 (16.4)\ 91.1 (42.0)\ 146.5 (90.5)\ 132.5 (82.6)\ 103.7 (72.1)\ 113.5 (78.8)\ p value 0.23 0.006[^\*\*^](#TN7){ref-type="table-fn"} 0.21 1.00 0.25 0.10 0.09 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ *Results are presented as mean difference in expression between endometrial biopsy and hysterectomy specimen (SD) or Pearson\'s correlation coefficient, p-value, number of samples compared*. *p ≤ 0.05*, *p ≤ 0.01*. Impact of Hypoxia on Loss of Immunohistochemical Expression ----------------------------------------------------------- To determine the impact of hypoxia of the uterine tissues on loss of immunohistochemical expression of commonly studied biomarkers in endometrial cancer window studies, expression of HIF-1α and its downstream effector CA-IX were compared in the hysterectomy specimen to the corresponding endometrial biopsy. Paradoxically, despite the uterine blood supply being clamped for at least 20 min during surgical excision and obvious discoloration of the uterus as a result, expression of both hypoxia markers was significantly lower in the hysterectomy specimen than in the endometrial biopsy (both *p* \< 0.0001, [Figure 4](#F4){ref-type="fig"}). {#F4} Discussion {#s4} ========== The immunohistochemical expression of commonly studied biomarkers in endometrial cancer window studies, including Ki-67, phosphorylated markers of the PI3K-Akt-mTOR and insulin signaling pathways and hormone receptors, is significantly lower in the hysterectomy specimen compared with an endometrial biopsy taken immediately prior to the commencement of surgery. In contrast, expression of markers of cancer stem cell activity, such as CD133 and ALDH, are unaffected by tumor sampling technique. Investigation of the extent to which loss of protein expression was due to surgically induced hypoxia was hampered by lower levels of immunohistochemical staining of hypoxia inducible factor and its downstream effector in the hysterectomy specimen. This was despite observational evidence of reduced perfusion for over 30 min, which would normally result in a significant increase in expression of these hypoxia inducible proteins ([@B14], [@B15]). The extent to which biomarker expression was lost in the hysterectomy specimen was closely related to baseline expression of these proteins in the endometrial tumor and was reduced if the uterus was bisected before being placed in formalin. These findings suggest that delays in achieving adequate fixation of tissues are of critical importance in driving loss of detectable protein expression in surgically excised specimens. Formalin penetrates tissues at a rate of only 1 mm/h, meaning that it can be several hours before it has reached and fixed tumor tissue located within the uterine cavity ([@B2]). In breast cancer, a time interval of 16 h or more between surgical excision of a tumor and its fixation has been shown to be sufficient to result in a significant reduction in expression of Ki-67 ([@B16]). Similarly, prolonged fixation (\>2 days) has also been shown to be detrimental to the expression of proliferation markers, highlighting the importance of developing and adhering to guidelines of the optimization of pre- and intra-fixation parameters. As the results of this study have also shown, immediate dissection and sectioning of large surgical specimens is necessary to ensure proper fixation ([@B17]). Whilst the insensitivity of CD133 and ALDH to tumor sampling technique may initially appear surprising, hypoxia has previously been shown to enrich for cancer stem cells ([@B18]). Expression of both of these markers at both the mRNA and protein level as well as the proportion of CD133^+ve^ and ALDH^high^ cells detected by flow cytometry is increased in response to the growth of glioblastoma and ovarian cancer cell lines in 1% oxygen ([@B18]--[@B20]). Any loss of expression through poor fixation is, therefore, likely to be counteracted by the hypoxia-induced increase in expression of these markers within endometrial cancer cells. The fact that expression of the hypoxia markers HIF-1α and CA-IX were significantly lower in the hysterectomy specimens compared with the endometrial biopsies suggests that the cancer stem cell markers studied were more stable at the protein level and resistant to delays in fixation. In order to avoid the issue of loss of protein expression in hysterectomy specimens, the use of endometrial biopsies for post-treatment immunohistochemical analysis is to be encouraged, as occurred in the aforementioned PREMIUM trial ([@B10]). The smaller size of a biopsy sample means that fixation can occur much more rapidly and, as a consequence, more accurately reflects tumor biology. The findings reported here mirror those seen in breast cancer, where immunohistochemical expression of pAkt and pERK1/2 was found to be significantly lower in surgically excised specimens than in core-cut biopsies and, indeed, was found to be almost absent in a number of cases ([@B3]). Whilst Pinhel et al. found no significant difference in expression of hormone receptors in breast cancers removed surgically or sampled by biopsy, their analysis was based on only 29 samples and there was a trend toward lower expression of ER in excised specimens. From a clinical point of view, the concern has been raised that inappropriate decisions regarding adjuvant treatment may be made if hormone receptor status is determined solely in surgical specimens. It has been estimated that up to 9% of breast cancer cases may be denied the benefits of hormone therapy due to false negative immunohistochemistry results ([@B5]). For window studies, smaller changes in protein expression are used to determine drug efficacy, meaning that the hysterectomy specimen can no longer be considered adequate to be used in the assessment of primary and secondary outcomes in these kinds of trials in endometrial cancer. This potentially calls into question the results of earlier studies which reported a reduction in immunohistochemical staining of Ki-67 and phosphorylated markers of the PI3K-Akt-mTOR and MAPK/ERK pathways with exposure to metformin and anastrazole in the pre-surgical window ([@B6], [@B7], [@B21]). Future studies should be designed to compare expression of biomarkers in endometrial biopsies taken prior to and following drug treatment, with the latter performed immediately prior to the start of the hysterectomy procedure as in the case of the PREMIUM study ([@B10]). Whilst concerns may be raised that "blind" endometrial biopsies risk sampling error by failing to analyze a representative proportion of the tumor, this risk may be minimized by obtaining a generous sample under general anesthetic. In the current study, only cases for whom a representative endometrial biopsy was taken immediately prior to the start of the hysterectomy procedure were included. For 75 of the 88 women (85%) participating in the PREMIUM trial, the endometrial biopsy was of sufficient size and contained tumor of similar morphology to that of the hysterectomy specimen, as determined by a specialist gynecological pathologist. In three cases it was not possible to obtain an endometrial biopsy due to cervical stenosis and in the remaining 10 cases, the biopsy was either too small or not representative of the tumor overall. It would therefore appear to be both feasible and appropriate to use endometrial biopsies rather than the hysterectomy specimen for analysis of trial outcomes in window studies. A disadvantage is that it does not allow assessment of the myometrium, and may contain minimal stroma or normal endometrium for comparison, although this may not be relevant if the effect of drugs on the tumor glandular component only is being assessed. The strengths of this study include the comparison of a range of biomarkers used as outcome measures in endometrial cancer window studies in a large sample of endometrial biopsies and matched hysterectomy specimens. Scoring was performed using semi-automated software, which has been shown to be more reliable and reproducible than manual scoring ([@B11]). Only limited data were available regarding the handling of tissues once they had been surgically removed from the body; in particular, details of the time interval before the specimen was placed in formalin (the cold ischemia time) and the length of time it remained in the fixative before sectioning were not recorded. Surrogate measures were therefore used in their place to investigate the relationship between loss of immunohistochemical staining and delays in or prolongation of fixation, including specimen and tumor size and tumor-serosal distance. The extent to which these variables correlate with tumor fixation have not, however, been determined. Whilst records were kept of the number of surgical specimens opened to obtain a directed tumor biopsy prior to being placed in formalin were kept, the number of uteri that were bisected but then did not have a biopsy taken due to insufficient surplus tumor were not documented. This was known to have happened on at least several occasions. The association between formalin penetration and loss of immunohistochemical staining may have been strengthened if these data had been available. It is possible to limit the cold ischemia time, particularly if fresh, frozen tissue is not required, by placing specimens directly into formalin and bisecting large specimens, such as the uterus, to reduce the time required for the fixative to penetrate deep tissues. If fresh tissue is required, ideally a pathologist should be available within the operating room to perform the necessary sampling, otherwise appropriately trained deputies should be present to prevent "research sampling" impacting clinical diagnostic assessment. Correlation between the degree of loss of protein expression in the hysterectomy specimen and surgically induced hypoxia was not possible due to the presence of lesser degrees of staining of hypoxia markers in these samples. This may well be the result of poor fixation of these tissues. Alternative methods of determining intra-tumoral hypoxia, however, are either invasive or expensive, thereby negating their use ([@B22]). Regardless of the reason for the loss of protein expression in the hysterectomy specimen, the end result is a sample that no longer represents the tumor from which it was obtained, especially for phosphorylated markers. Conclusion {#s5} ========== Immunohistochemical staining of commonly studied biomarkers in endometrial cancer window studies, including Ki-67, phosphorylated markers of key carcinogenic pathways and hormone receptors, is significantly lower in the hysterectomy specimen than in an endometrial biopsy taken immediately prior to surgery. Surgically induced hypoxia and, in particular, poor fixation of large specimens are likely to be responsible. In order for methodologically robust results to be generated from future endometrial cancer window studies, endometrial biopsies should be used for post-treatment analyses. Ethics Statement {#s6} ================ This study was carried out in accordance with the recommendations of North West Research Ethics Committee (14/NW/1236) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the North West Research Ethics Committee. Author Contributions {#s7} ==================== EC obtained funding for the study and was its Chief Investigator, designed the study, supervised study execution, contributed to data interpretation, edited the manuscript, and study guarantor. SK designed the study, recruited to the study and performed clinical and laboratory study procedures, data interpretation and wrote the manuscript. SK, ZM, and VS undertook immunohistochemical scoring on tumors annotated by JS. All authors provided critical comment and approved the final version of 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. We would like to thank all patients who participated in the trial and the clinical staff at recruiting hospitals for their support in conducting this work. Assistance with slide preparation was kindly provided by the Histology Department at the Cancer Research UK Manchester Institute. Slide scanning was performed by the Advanced Imagining Facility. EC and SK were funded through a National Institute for Health Research (NIHR) Clinician Scientist Fellowship (NIHR-CS-012-009). VS was funded through a Wellcome Trust/Wellbeing of Women Research Training Fellowship (Ref 098670/Z/12/Z) and this article presents independent research funded by the NIHR, supported by the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-20007) and facilitated by the Greater Manchester Local Clinical Research Network. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health. [^1]: Edited by: Rebecca Stone, Johns Hopkins Medicine, United States [^2]: Reviewed by: Sarah M. Temkin, Virginia Commonwealth University, United States; Linda Rosenbaum Duska, University of Virginia, United States [^3]: This article was submitted to Women\'s Cancer, a section of the journal Frontiers in Oncology

Sir, The need and characteristics of an ideal nomenclature of psychotropic drugs and the arrival of neuroscience-based nomenclature (NbN) are a landmark in psychopharmacology.\[[@ref1][@ref2]\] NbN has medicolegal importance too. Psychotropic drugs lack specificity in indication. Antipsychotics can be used in schizophrenia, bipolar mood disorders, depressive disorders, tic disorders, and also obsessive--compulsive disorders. However, as expert witness in the court, one may get into difficulty for the very same reason. I was once asked by the honorable court as to why I had prescribed antipsychotic medication (risperidone) for a patient with obsessive--compulsive disorder when she did not have psychotic symptoms. I had used risperidone as an augmenting strategy. The lawyer of the patient\'s husband submitted that the patient was receiving antipsychotic medication and so, it was a case of schizophrenia and hence a ground for divorce could be considered. It was a difficult task explaining to them the varied indications of a particular class of psychotropic drugs. I was also asked as to why we call the drug as "antipsychotic" when we use it for indications other than psychosis. The nomenclature system of "Medication defining the diagnosis" may thus cause difficulties in the court. NbN can circumvent these difficulties. Therefore, it is a welcome change that many journals have started adopting this system.\[[@ref3]\] The replacement of indication-based nomenclature by pharmacologically driven nomenclature to some extent reduces the stigma associated with psychiatric diagnosis. The need of moving from a disease-centered model to a drug-centered model of how psychotropic drugs work and its implications on research and clinical practice was posited by Moncrieff and Cohen in 2009.\[[@ref4]\] Financial support and sponsorship {#sec2-1} ================================= Nil. Conflicts of interest {#sec2-2} ===================== There are no conflicts of interest.

1. Introduction {#sec1} =============== Nontuberculous mycobacteria (NTM) are ubiquitous in the environment and are responsible for several diseases in animals and humans known as mycobacterioses \[[@B1]--[@B3]\]. A study published in 2013 has shown that *M. fortuitum, M. gordonae, M. kansasii,*and *M. peregrinum*are frequently present in the aquatic environment of surface waters in the Moravian region of the Czech Republic. In contrast, the most frequent NTM present in aquaria with ornamental fish were *M. marinum* \[[@B4]\]. *Mycobacterium marinum* is the cause of chronic systemic infections in fish and an occasional cause of granulomatous skin lesions in humans. Infections in humans result in skin and soft tissue infections characterized by their predilection for the upper extremities, often following minor trauma to hands with a history of typical exposure to aquarium tanks \[[@B5], [@B6]\]. Diagnosis is usually made after biopsy and culture of the lesion, but microbiologists should be alerted to the possibility of *M. marinum* if the injury originated from an aquarium or the water environment \[[@B7]\]. Here we report two cases of *M. marinum* infection in humans ([Table 1](#tab1){ref-type="table"}). Previously described species-specific qPCR targeting the *erp* and IS*2404* genes together with a conventional culture method was used for the detection of *M. marinum* in clinical specimens of two infected humans ([Table 2](#tab2){ref-type="table"}). Simultaneously, epidemiology screening for possible sources in patients\' aquaria via VNTR fingerprinting of *M. marinum* isolates was carried out. 2. Materials and Methods {#sec2} ======================== 2.1. Sample Collection {#sec2.1} ---------------------- A total of 49 samples were taken from two humans (*n* = 9), ornamental fish (*n* = 20), and the aquarium environment (*n* = 20; biofilms, water, and plants). Animal and environmental samples were examined within the framework of an epidemiological study carried out in the aquaria of infected aquarists. 2.2. Cultivation and Isolate Characterization {#sec2.2} --------------------------------------------- All fish and environmental samples were subjected to decontamination with N-acetyl-L-cysteine followed by cultivation at 25°C, 30°C, and 37°C for 8 weeks \[[@B8]\]. In case of clinical samples decontamination step was excluded. Mycobacterial isolates were grown on solid Löwenstein-Jensen and Ogawa media (Trios, Prague, Czech Republic) and in liquid Sula medium (Trios). Clinical specimens were subjected to microscopic examination after Ziehl-Neelsen (ZN) staining for the detection of acid fast bacilli (AFB). Mycobacterial isolates were identified by their phenotypic properties (temperature preference, growth rate, and pigment production in the dark and upon exposure to light), microscopic examination after ZN staining, and sequence analysis of the *16S rRNA* gene \[[@B9]\]. Subsequently, VNTR analysis according to a previously published method was used to fingerprint isolates identified as *M. marinum* \[[@B10]\]. 2.3. DNA Isolation and Molecular Detection of *Mycobacterium marinum* {#sec2.3} --------------------------------------------------------------------- DNA isolation from human, fish, and environmental samples was based on a protocol described previously \[[@B11]\]. The isolated DNA was analyzed according to a previously described *erp*/IS*2404* qPCR assay, which enables species-specific detection of *M. marinum* \[[@B12]\]. 3. Patient Clinical Presentation {#sec3} ================================ 3.1. Patient 1 {#sec3.1} -------------- A Twenty-five-year-old male reported cleaning his injured left knee with a brush usually used to clean the aquarium, which resulted in infiltration and subsequently visible crusts together with two subcutaneous resistances caused by *M. marinum*. The patient was initially treated with clarithromycin. Subcutaneous resistance did not lessen after nine weeks; therefore, treatment was continued with ethambutol for an additional nine weeks. 3.2. Patient 2 {#sec3.2} -------------- The second case involved cutaneous infection of a 60-year-old male with a professional interest in fish breeding. He reported injury to the tip of his index finger, which occurred during aquarium clearing. He was misdiagnosed with rheumatoid arthritis and treated with methylprednisolone for a period of 3 months, which resulted in a rare systemic spread of *M. marinum* from the primary site into the joint and bones of the index and ring finger, forearm, testis, and epididymis. The patient was initially treated with a combination of clarithromycin and ethambutol. Because of the insufficient clinical response in the area of the right hand an eight-month-long treatment based on results of sensitivities to a combination of amikacin, ciprofloxacin, ethambutol, cycloserine, and pyrazinamide was administered. Due to the damage to the testis an orchiectomy was also carried out. Because of long-lasting pain and damage the distal part of the ring finger was surgically removed. 4. Results and Discussion {#sec4} ========================= Generally, three main histopathological patterns of *M. marinum* infection are distinguished: granulomatous nodular or diffuse inflammation with mixed granulomas; abscesses with mild granulomatous reactions; and subcutaneous (patients 1) or deep dermal granulomatous inflammation \[[@B13]\]. Tenosynovitis, septic arthritis, bursitis, or osteomyelitis was also reported in the literature to follow from deep subcutaneous infection \[[@B14], [@B15]\]. *M*. *marinum* infection may have the potential for systemic dissemination in humans as has been reported earlier and was shown in the second patient \[[@B6]\]. Conventional microbiological methods used for *M. marinum* diagnostics are slow and rely solely on phenotypic characteristics, which can be very similar to different causative agents. Clinicians in general have to consider other causes besides *M. marinum* including deep fungal infection; therefore, a diagnostic ladder is needed in this kind of patients. Delayed diagnosis is the main cause of the patients\' serious side effect. Rapid and accurate molecular diagnosis methods are essential. Rapid detection of mycobacteria using conventional broad range PCR assay was previously described in the literature \[[@B16]\]. The advantage of this approach is the ability to detect multiple mycobacterial species. Generally, conventional PCR could show weakness such as proneness to contamination or lack of sensitivity when used for detection in clinical specimens where pathogen is present in limited amount \[[@B17]\]. Sensitive and specific method for the direct detection of *M. marinum* in clinical specimens suitable to overcome this limitation would be beneficial. So far, there have been very few reports on the detection of *M. marinum* directly from infected tissue without previous culturing \[[@B12], [@B18]\]. The molecular based approach used in this study represents a fast, sensitive, and specific method for the detection of *M. marinum*, in comparison to more time-consuming conventional methods. The protocol used here enabled us to complete the analysis of a sample, including controls, in approximately 6 hrs. The applied qPCR assay enabled the detection of *M. marinum* in clinical specimens collected from both the infected humans prior to successful cultivation ([Table 2](#tab2){ref-type="table"}). Rare systemic spread of *M. marinum* in the second patient was proven by qPCR, microscopy, and cultivation. Furthermore, direct qPCR detection was successful in the case of three culture-negative samples. Screening for the presence of *M. marinum* in the aquarium environment of both patients was carried out. The qPCR analysis of the patients\' ornamental fish and aquarium environment revealed positivity for*M. marinum* in eight fishes (*n* = 3, patient 1; *n* = 5, patient 2). The number of viable cells in qPCR-positive samples was possibly lower due to the applied decontamination procedure because only one fish from each patient was later culture-positive for *M. marinum*. As reported earlier, microbiological stains give positive results in less than 25% of cases and do not correlate with the severity of *M. marinum* infection \[[@B14]\]. Although *M. marinum* is described to be present in the water environment, it was detected only in fish. From the epidemiological aspect it has been proven that the *M. marinum* strain isolated from fish had the same VNTR profile as the clinical isolate originating from the fish owner and thus it can be concluded that the aquaria were the source of infection in the case of both aquarists ([Table 3](#tab3){ref-type="table"}). 5. Conclusion {#sec5} ============= With regard to health, it is important that many NTM species classified as potentially pathogenic survive not only in the water environment, but also in infected fish. Contact zoonosis may occur particularly in risk groups such as aquaculture and fishery professionals, fish processors, ornamental fish hobbyists, and also consumers. The diagnosis of infections is usually hampered by the unfamiliarity of clinicians with disease agents derived from aquatic species. Fish tank exposure is the source of most cases of cutaneous *M. marinum*infections. It should be considered as a possible threat to humans reporting close contact with this environment and may be prevented by the use of waterproof gloves by persons with acute or chronic open skin lesions. This work was supported by Grant no. MZE0002716202 and Grant "AdmireVet" no. CZ 1.05/2.1.00/01.0006-ED0006/01/01 from the Ministry of Education, Youth and Sports of the Czech Republic. ###### Cutaneous infection of*Mycobacterium marinum* in two patients. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Patient no. Age/sex Case background Drug therapy ----------------- --------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------ --------------------- ------------------------------------------------------------------------------------------------------------------------------------------------- ----- \(1\) 25/m Hobby aquarist who cleaned injured knee with a brush usually used to clean the aquarium Clarithromycin Infiltrate with visible crusts together with two subcutaneous resistances on the left knee\ Yes Extirpation of subcutaneous nodule because positivity for *M. marinum* in the sample of exploratorily excised cutaneous tissue was not detected Clarithromycin\ Treatment changed due to the insufficient clinical response in the area of subcutaneous resistance Ethambutol \(2\) 60/m Patient with a professional interest in fish breeding reported finger injury during aquaria cleaning Methylprednisolone\ Suspected rheumatoid arthritis due to swelling of the index finger joint\ Yes Methotrexate Immunosuppression contributed to the hematogenous spread of infection Clarithromycin\ Insufficient clinical response in the area of the right hand Ethambutol Amikacin\ Severe right-hand oligoarthritis\ Ciprofloxacin\ Osteomyelitis of the affected metacarpophalangeal joints\ Ethambutol\ Orchiectomy due to inflammation and necrosis in the epididymis with continuous spreading to the testis\ Cycloserine\ Surgical removal of the ring finger Pyrazinamide ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ###### The presence of *Mycobacterium marinum* in clinical specimens studied using the *erp* and IS*2404* qPCR assay. Sample *erp*/IS*2404*  qPCR ZN microscopy Cultivation Isolate identity --------------------- ---------------------- --------------- ------------- ------------------ -------------- Patient 1 Subcutaneous nodule Left knee 1.6 × 10^7^ AFB \+ *M. marinum* Cutaneous tissue Left knee − − − − Patient 2 Sperm Testis − − −   Puncture Metacarpal joint 2.8 × 10^5^ AFB \+ *M. marinum* Pus Ring finger 5.2 × 10^6^ AFB \+ *M. marinum* Blood Vein 6.2 × 10^3^ − − − Subcutaneous nodule Right forearm 4.1 × 10^5^ − \+ *M. marinum* Tissue Epididymis 2.8 × 10^7^ AFB − − Tissue Testis 1.2 × 10^5^ AFB − − Quantification of *M.  marinum* is shown in genome equivalents per g of tissue or mL of liquid sample. AFB: acid fast bacilli. ###### VNTR analysis of *Mycobacterium marinum* isolates. Isolate source VNTR loci^a^ --------------------- -------------- --- --- --- --- --- ---- --- --- Patient 1 Subcutaneous nodule 2 2 3 3 2 2 NA 3 4 Fish 2 2 3 3 2 2 NA 3 4 Patient 2 Puncture 2 2 3 4 2 2 2 3 4 Pus 2 2 3 4 2 2 2 3 4 Subcutaneous nodule 2 2 3 4 2 2 2 3 4 Fish 2 2 3 4 2 2 2 3 4 ^a^NA: no amplification. As Locus 9, Locus 14, and VNTR1132 failed in amplification from any isolates, they are not listed here. [^1]: Academic Editor: Tomasz Jagielski

Background ========== Atherosclerosis is a chronic inflammatory disease in large arteries; it can cause life-threatening cardiac or cerebral vascular events, such as myocardial infarction and ischemic stroke. The main features of atherosclerotic lesions are that they contain subintimal lipid deposition in certain parts of the arteries, and they have the abnormal proliferation of smooth muscle cells or fibrous matrix components \[[@b1-medscimonit-25-9949],[@b2-medscimonit-25-9949]\]. This causes a healthy arterial wall to thicken and lose elasticity, eventually leading to atherosclerotic plaque formation. Macrophage accumulation for depositional lipid elimination in the subendothelial region of the arterial wall is a hallmark of atherosclerosis. These cells laden with lipid facilitate inflammatory responses within the wall of the artery that can cause diverse fatal consequences, such as rupture, hemorrhage, and calcification \[[@b3-medscimonit-25-9949],[@b4-medscimonit-25-9949]\]. In early stages, atherogenesis is propelled by the infiltration of monocytes underneath the vascular endothelium. After their differentiation into macrophages, they tend to clear the excess depositional lipids and lipoproteins in the neointima, causing the macrophages to become engorged with lipids and terminally lacking the ability to emigrate from the plaque \[[@b5-medscimonit-25-9949],[@b6-medscimonit-25-9949]\]. Indeed, this process contributes to the development of a local arterial inflammatory state and thus generates a chronic and complicated atherosclerotic lesion \[[@b7-medscimonit-25-9949]\]. The macrophages are present in various lesion regions, such as the plaque shoulder, next to the necrotic core, next to blood vessels, or in the zone next to the rigid and calcified arterial zone \[[@b8-medscimonit-25-9949]\]. Nevertheless, wherever they are in the atherosclerotic plaque, macrophages are sensitive to the complex micro-environment and they are continuously facing stimuli, or external signals, including cytokines, calcium, and iron \[[@b9-medscimonit-25-9949]--[@b11-medscimonit-25-9949]\]. Thus, macrophage phenotypes could be influenced by these factors, which manifest different patterns called polarization or activation. In turn, macrophages also influence and mostly control the atherosclerosis-related processes from initiation to maturity. In recent years, several studies have revealed the role of genetic changes or genetic transcription in the modulation of macrophage phenotype in atherosclerosis. Macrophages can display a continuous phenotype described by different gene expression profiles with the ability to switch from one phenotype to another based on external stimuli. Genetic factors have important impacts on the regulation of macrophage phenotypes involved in the pathogenesis of atherosclerosis. The expression level of a few factors positively correlates with the expression of macrophage markers such as LXRα, PPARγ, and KLF4 \[[@b9-medscimonit-25-9949],[@b12-medscimonit-25-9949],[@b13-medscimonit-25-9949]\]. Therefore, transcription factors, or post-transcriptional regulators, and other related signaling molecules play key roles in the control of macrophage differentiation and polarization; significant changes in expression levels of distinct transcripts clearly occur \[[@b14-medscimonit-25-9949],[@b15-medscimonit-25-9949]\]. As mentioned above, evidence has shown that the potential therapeutic targets for atherosclerotic disease may be hiding under the abnormally variable transcriptome. Therefore, it is important to explore the possible mechanisms underlying global changes in gene expression during the physiopathological process of macrophages, which may help to provide more reliable and effective early diagnostic molecular markers for blocking and controlling the progression of atherosclerotic diseases. Microarray technology is an efficient tool for obtaining accurate data on general gene expression levels, and it has been widely used to study overall patterns of gene expression changes in different diseases. These microarrays offer a new approach to study disease-related genes and provide evidence for prognosis and potential molecular targeted therapy \[[@b16-medscimonit-25-9949]\]. Recently, large amounts of gene chip data have been uploaded to public databases, and using these specific data provide a chance to deeply study the molecular mechanisms in diseases. The present study aimed to re-analyze the data of GSE7074 (a data set derived from the GEO database, which contains the required transcriptome data) using bioinformatics methods to identify significant differentially expressed genes (DEGs) of resident macrophages in human normal tissue (alveolus, spleen, liver) and in atherosclerotic plaque. After PCA analysis, functional annotation and pathway enrichment were then performed, followed by the construction of protein-protein interaction (PPI) networks, analysis of key nodes, and the regulatory relationship prediction between microRNA/transcription factors (TFs) and DEGs. These analyses allowed us to discover key genes and to develop new insights for understanding the role of macrophages in atherosclerotic plaque formation, as well as suggesting potential new directions in the study of atherosclerosis. Material and Methods ==================== Microarray data --------------- The microarray data of GSE7074 were downloaded from Gene Expression Omnibus (GEO); the platform for this dataset is GPL4868, AMC-MAD Homo sapiens 19k ver 1.0. The dataset contained 30 samples, including atherosclerotic (carotid arteries) and nonatherosclerotic (lung, liver, and spleen, macrophages were isolated and derived from the patients without any clinical symptoms of systemic inflammation or malignancies) tissues, that were uploaded by Eijgelaar WJ and Horrevoets AJ \[[@b17-medscimonit-25-9949]\]. Series matrix files and platform information were downloaded, and expression profile data preprocessing was performed using the R software package (R version 3.5.0). Data preprocessing included converting and rejecting the unqualified data, calibration, filling in of missing data, and standardization. PCA analysis, identification of DEGs, and hierarchical clustering analysis -------------------------------------------------------------------------- Principal component analysis (PCA) is a dimension reduction approach, which can effectively represent the effects of the measurements or dataset consisting of a large number of interrelated variables \[[@b18-medscimonit-25-9949]\]. PCA for this dataset was conducted using R software, and the results were visualized. After the probe ID was converted into a universal name for genes (official gene symbol), gene differential expression analysis was performed using the limma package (from the Bioconductor, *<http://www.bioconductor.org/>*), in R as well. Gene expression levels of samples with an adjusted P value \<0.05 and \|log2-fold change (FC)\| \>0.58 were considered as valid DEGs. Clustering analysis was then performed using the pheatmap package (from the Comprehensive R Archive Network, *<https://cran.r-project.org>*). Functional and pathway enrichment analyses of DEGs -------------------------------------------------- The DAVID online analysis database (*<https://david.ncifcrf.gov/>*) provides functional classification and annotation analyses of candidate genes \[[@b19-medscimonit-25-9949]\], and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis offers a way to detect the potential various types of biochemistry pathways. The functional and pathway enrichment analyses were performed using the DAVID database and the KOBAS database (*<http://kobas.cbi.pku.edu.cn>*), respectively. To comprehensively evaluate the relevant pathways, or biological procedures, of macrophages in atherosclerotic plaque, Gene Ontology (GO) and pathway enrichment analyses on DEGs were carried out through the online tools with a threshold of P values \<0.05. Protein--protein interaction (PPI) network construction ------------------------------------------------------- The STRING database (*<http://string-db.org/>*) is an interactive online tool for identifying interactions or associations between proteins \[[@b20-medscimonit-25-9949]\]. In the present study, STRING was used to predict the interactions among proteins encoded by DEGs. A combined score ≥0.4 of PPI pairs was considered significant and was chosen to be visualized by constructing a PPI network. The proteins in the network act as the "nodes" and nodes in the central positions may be core roles, which mean that they can be the pivotal candidate proteins/genes with crucial biological functions in the regulatory network. Moreover, the hub nodes were identified by a high score based on the method of degree ranking. Network module analysis ----------------------- The plugin of MCODE in Cytoscape \[[@b21-medscimonit-25-9949]\] was used to analyze the significant module in the PPI network. The MCODE algorithm can calculate the number of close interactions and enrichments of the subnet module, and higher scores of the module presented indicate that the interaction relations inside the module were more closed. According to the analysis, the modules with score ≥5 and node ≥5 were selected. Then, pathway enrichment for the DEGs in modules was analyzed. Prediction of miRNA-target gene regulatory relations ---------------------------------------------------- The miRNA and target genes regulatory network analysis was performed using the Enrichr \[[@b22-medscimonit-25-9949]\] (*<http://amp.pharm.mssm.edu/Enrichr/>*). The predicted miRNAs and target DEGs were identified based on the network geometric test. With the threshold of P value \<0.05, the regulatory network between miRNA and target DEGs was visualized using the Cytoscape. Prediction of transcription factor (TF)-target network ------------------------------------------------------ The regulatory network between TFs and target DEGs were predicted using Enrichr, with the threshold of P value \<0.05, and significant TFs and target DEGs relationships were visualized using Cytoscape. Results ======= PCA analysis and identification of DEGs in human atherosclerotic macrophages ---------------------------------------------------------------------------- To better define the heterogeneity of gene expression levels between human atherosclerotic macrophages and nonatherosclerotic macrophages, a principal component analysis was performed in R. The result was visualized using a 3D PCA scatterplot ([Figure 1](#f1-medscimonit-25-9949){ref-type="fig"}), which indicated that most of the nonatherosclerotic sample data in the present study had an obvious distribution difference compare to the atherosclerotic samples (points of different colors represent samples from different groups), and this could offer a guide for the subsequent study. After the macrophage expression microarray dataset GSE7074 was standardized, the data were divided into the atherosclerotic macrophage group or normal tissue macrophage group, including 8 atherosclerotic samples and 22 normal samples from liver, lung, and spleen. After the objective dataset had been screened for gene differential expression analysis (adjusted P value \<0.05, \|log2-fold change (FC)\| \>0.58), 236 DEGs were obtained, including 215 downregulated genes and 21 upregulated genes (results of gene differential expression analysis are shown in [Table 1](#t1-medscimonit-25-9949){ref-type="table"}). The multiple differential expression genes screened out from the dataset are shown in [Figure 2](#f2-medscimonit-25-9949){ref-type="fig"}. Next, hierarchical cluster analysis was performed to correctly distinguish DEGs between atherosclerotic and nonatherosclerotic macrophage samples based on significantly different expression patterns ([Figure 3](#f3-medscimonit-25-9949){ref-type="fig"}), indicating that the results were reliable and suitable for subsequent analyses. Enrichment analyses of DEGs --------------------------- Biological annotation and KEGG pathway enrichment of the DEGs were performed using the DAVID online analysis tool and KOBAS online database (functional classification, annotation analyses, and the potentially related biochemistry pathways of candidate genes), respectively. With a threshold of P\<0.05, 22 GO terms and 25 KEGG pathways were enriched from 54 terms and 182 pathways. The results are presented in [Figures 4](#f4-medscimonit-25-9949){ref-type="fig"}[](#f5-medscimonit-25-9949){ref-type="fig"}--[6](#f6-medscimonit-25-9949){ref-type="fig"}, and mainly involved 'visual perception' (GO Biological Process \[BP\] term, P=6.21E-06), 'phototransduction' (GO BP term, P=1.39E-04), 'photoreceptor disc membrane' (GO Cellular Component \[CC\] term, P=0.001328569), 'integral component of plasma membrane' (GO CC term, P=0.007973519), and the KEGG pathways of 'Purine metabolism' (P=0.001546974), 'Phototransduction' (P=0.002534266), and 'Metabolic pathways' (P=0.002834313) categories. The top 10 significant results of the GO term and KEGG pathway enrichment analysis of DEGs in atherosclerotic macrophages are shown in [Tables 2](#t2-medscimonit-25-9949){ref-type="table"} and [3](#t3-medscimonit-25-9949){ref-type="table"}. Analyzing all DEGs using a PPI network and modules identification ----------------------------------------------------------------- Considering the comprehensive nature of network construction and the whole process, all DEGs from the groups were sent to the STRING website to obtain the PPI information (associations between the candidate genes/proteins). After processing, a subset of 222 nodes with 186 interactions was drawn out from STRING. A complex network of chosen DEGs was constructed after removing the isolated nodes, as shown in [Figure 7](#f7-medscimonit-25-9949){ref-type="fig"}. The top 10 leading degree hub nodes in the network were: RHO, F5, BEST1, POLR2F, PDE6G, HIST1H2BL, PDE6A, ABCA4, CNGA3, and AIPL1. Among them, there were 6 nodes with degrees ≥8, only RHO, F5, and BEST1, the 3 most significant downregulated genes ranked degree scores ≥9, which were considered to be critical in this network ([Table 4](#t4-medscimonit-25-9949){ref-type="table"}). Further, from the PPI network, only 2 subnet modules were selected with score ≥5 (module A=7.429 and module B=5). As shown in [Figure 8](#f8-medscimonit-25-9949){ref-type="fig"}, module A consisted of 8 downregulated nodes with 26 edges, which were significantly enriched in 'Phototransduction' (hsa04744), 'Purine metabolism' (hsa00230), and 'ABC transporters' (hsa02010). Module B was made up of 5 downregulated nodes with 10 edges, and the genes were significantly enriched in 'Complement and coagulation cascades' (hsa04610). RHO and BEST1 were in module A, and F5 was counted in module B. miRNA/TF prediction and analysis -------------------------------- In the miRNA-target regulatory network, 16 miRNAs were predicted in the DEGs with the threshold of P\<0.05, and 4 miRNAs were highlighted with degrees ≥30, including hsa-miR-4640-3p, hsa-miR-3177-5p, hsa-miR-1908, hsa-miR-663. Among them, hsa-miR-4640-3p and hsa-miR-3177-5p had the highest scores (19.23667544 and 11.46519428, respectively), which could target PPI network-crucial genes F5 and BEST1, respectively. TFs were also predicted from the DEGs; 5 TFs were predicted, including EGR1, CLOCK, GATA3, WRNIP1, and PBX3, among which early growth response-1 (EGR1) was predicted to target 3 downregulated DEGs, and others were predicted to target 2. Based on the obtained miRNA-target pairs, and TF-targets, we constructed integrated networks using Cytoscape software, as shown in [Figures 9](#f9-medscimonit-25-9949){ref-type="fig"}, [10](#f10-medscimonit-25-9949){ref-type="fig"}. Discussion ========== Recent studies have revealed the functional and ontogenetic diversity of tissue-resident macrophages, and data acquired by high-throughput sequencing has confirmed the transcriptional and epigenetic programs of these macrophages \[[@b23-medscimonit-25-9949],[@b24-medscimonit-25-9949]\]. In contrast to most of the general histocytes which have the same transcriptional program regardless of where they reside, these tissue-resident macrophages possess evolutionary diversity and distinct functions in maintaining homeostasis, and they also exhibit extensive plasticity during disease progression. Here, the gene data profiles present 3 different regions of normal tissue-resident macrophages: Kupffer cells, alveolar macrophages, and splenic macrophages. Compared to Kupffer cells, splenic macrophage and alveolar macrophage populations are essential parts of the tissue immune compartment and tend to be involved in anti-inflammatory settings and the maintenance of a steady state. Many pathological studies have revealed a series of changes in blood vessels during atherogenesis and indicated that blood-derived inflammatory cells like monocytes/macrophages have a key role. Tissue culture studies with endothelial cells, vascular smooth muscle cells, and monocytes/macrophages suggested possible pathways of disease initiation and progression in atherosclerosis. Thus, this chronic inflammatory disease involving monocytes/macrophages aggregation in the vessel wall is inevitably related to tissue-resident microenvironmental changes, cellular stress, and monocyte phenotypic diversity. This may explain the obvious difference between the 3 kinds of nonatherosclerotic macrophages and atherosclerotic macrophages in PCA analysis. In the present study, we focused on exploring the link between normal tissue-resident macrophages phenotypic changes as well as functional diversity and atherosclerotic macrophages due to specific tissue microenvironmental changes or external stimuli, from macrophages resident in the human alveolus, spleen, and liver compared with carotid plaque. R software was used to analyze the dataset; 236 DEGs were identified, and most of them are downregulated genes. Specially, RHO, F5, and BEST1 were considered as important genes. With high degrees in the PPI network, F5 could be targeted by hsa-miR-4640-3p and BEST1 could be targeted by hsa-miR-3177-5p, and the transcription factor EGR 1 was predicted to target most DEGs. Meanwhile, both RHO and F5 were enriched in Golgi membrane cellular component, and both RHO and BEST1 were enriched in visual perception biological process and integral component of plasma membrane cellular component. We believe that these genes may be closely related to atherosclerotic plaque formation and development. As integral membrane proteins, bestrophins (Best) were first found by genetic linkage of human BEST1 to a juvenile form of macular degeneration called 'best disease' (best vitelliform macular dystrophy) \[[@b25-medscimonit-25-9949]\]. Belonging to a protein family, consisting of 4 members (bestrophin 1--4), BEST1 has been functionally proposed to be a regulator of Ca^2+^-activated Cl^−^ channel in heterologous studies \[[@b26-medscimonit-25-9949]--[@b28-medscimonit-25-9949]\]. Recent research showed the role of BEST3 in TNFα-induced human endothelial inflammatory response and the underlying molecular mechanisms, showing that overexpression of BEST3 significantly attenuated TNFα-induced expression of adhesion molecules and chemokines, and subsequently inhibited the adhesion of monocytes to human umbilical vein endothelial cells, suggesting that BEST3, a member of the bestrophins family, may be a novel approach for the treatment of vascular inflammatory diseases \[[@b29-medscimonit-25-9949]\]. Experiments using downregulated bestrophins *in vivo* support the hypothesis that bestrophins play key roles in the cGMP-dependent Ca2^+^-activated Cl^−^ current for initiation of vasomotion, and the specific downregulation of bestrophin-3 induces a secondary reduction of bestrophin-1 and -2 expressions \[[@b30-medscimonit-25-9949]\]. This means that BEST1, along with other members of the bestrophin family, may exert synergistic effects in the maintenance of normal vasomotion and vascular homeostasis. Disrupting the balance leads to vascular lesions, and abnormal vasomotion is believed to be a precursor to atherosclerotic disease \[[@b31-medscimonit-25-9949],[@b32-medscimonit-25-9949]\], which may be the initiation of atherosclerosis. Thus, BEST1 is likely to be involved in the progression of atherosclerotic inflammation, just like the role of BEST3 in the endothelial inflammatory response. We also predicted that BEST1 could be targeted by hsa-miR-3177-5p, which has been shown to have important regulatory effects on the expression of the 5-HT1A receptor in HEK-293 cells \[[@b33-medscimonit-25-9949]\]. The 5-HT1A receptor is a subtype of serotonin receptor system (5-HT receptor), and serotonin modulates vascular tonus and mediates development and rupture of atherosclerotic plaques \[[@b34-medscimonit-25-9949],[@b35-medscimonit-25-9949]\]. Whole-blood levels of hsa-miR-3177-3p were also identified to serve as a biomarker for vasospasm \[[@b36-medscimonit-25-9949]\]. miR-3177 was proved to take part in the process of vascular tonus or vasospasm modulating. Considering the specific regulatory function of BEST1 in vasculature as described above, it is no coincidence that both of miR-3177 and BEST1 can participate in the regulation of vasomotion, and the regulatory relationship between them seems to be clear. Thus, it was speculated that hsa-miR-3177-5p may take part in the regulation of vasomotion by targeting BEST1, and the expression disorder of miR-3177-5p/BEST1 can cause abnormal vasomotion and may be subsequently involved in the occurrence of atherosclerosis. Upregulation of miR-3177-5p and low expression of BEST1 in atherosclerotic macrophages are likely to represent the molecular phenotype of this vascular disease. F5 (Factor V) is part of the common pathway and plays a critical role in blood coagulation. It is a large, heavily glycosylated, single-chain protein that is homologous to factor VIII and shares a similar domain organization \[[@b37-medscimonit-25-9949],[@b38-medscimonit-25-9949]\]. Interestingly, defects in F5 may result in either hemorrhagic (FV deficiency) or thrombotic phenotypes. The most prevalent form of factor V deficiency is caused by an inherited genetic defect in the F5 gene sequence (congenital factor V deficiency or parahemophilia), and these mutations typically reduce protein expression or cause a loss of protein function \[[@b39-medscimonit-25-9949]\]. In particular, the missense mutation in the factor V position 506 by glutamine is found in patients presenting with venous thrombosis \[[@b40-medscimonit-25-9949]\], called Factor V Leiden (FVL). Early in 2007, a clinical trial proved that Factor V Leiden mutation can influence the progression of atherosclerosis \[[@b41-medscimonit-25-9949]\], and a recent study has confirmed that homozygosity for the FVL mutation in mice leads to enhanced arterial thrombosis and atherosclerosis \[[@b42-medscimonit-25-9949]\]. The mutation of F5 ultimately alters the normal gene expression, and it participates in the regulation of both anticoagulation function and coagulation function. Our results showed that F5 gene expression level is downregulated in atherosclerotic macrophages. Recent evidence indicates that the mutation of this gene is likely associated with atherogenesis, and the mutations lead to gene expression deficiency or severe biological function impairment of the protein product. It was also reported that when monocytes extravasate by crossing the vascular endothelium, they differentiate into macrophages and gradually lose F5 expression \[[@b43-medscimonit-25-9949]\]. However, there has been little research on the function of F5 in atherosclerosis, and it is unclear whether the simple change of F5 expression has a direct effect on promoting internal progression, thrombosis, or rupture of atherosclerotic plaque. In addition, miRNA-4640, which was identified in 2011 by next-generation sequencing of small RNAs in breast cancer \[[@b44-medscimonit-25-9949]\], was predicted to target F5. As a relatively new miRNA under study, miR-4640 has yet to be investigated in the context of physiology and disease. Simultaneously enriched in GO term 'integral component of plasma membrane cellular component' with BEST1 in the present analysis, another integral membrane protein, rhodopsin, was first found in the rods of the retina and belongs to the G-protein-coupled receptors (GPCRs). Rhodopsins are members of class A subfamily that are the largest group of GPCRs. Studies have revealed that some rhodopsin-like G protein-coupled receptors play different roles in the pathogenesis of atherosclerosis. For instance, apelin (APJ) is a class A, Rhodopsin-like G protein-coupled receptor and is widely expressed in the cardiovascular system and central nervous system. A number of studies have shown that there is a close relationship between APJ system and atherosclerosis: in the Ang II-induced atherosclerotic apolipoprotein E-deficient mouse model, Chun et al. \[[@b45-medscimonit-25-9949]\] found that apelin signaling blocked atherogenesis by activating NO synthase (eNOS) phosphorylation pathways, and the APJ system was also found to be involved in the development of atherosclerosis by affecting vascular smooth muscle cells \[[@b46-medscimonit-25-9949]\]. Low levels of plasma apelin were correlated with the severity of carotid plaque vulnerability, whereas it was positively related to the stability of atherosclerotic plaque \[[@b47-medscimonit-25-9949],[@b48-medscimonit-25-9949]\]. Another class A or rhodopsin-like G-protein-coupled receptor, neuropeptide Y (NPY) receptor, plays a crucial role in various biological processes. Abnormal regulation of NPY is involved in the development of a wide range of diseases, including hypertension, atherosclerosis, and metabolic disorders \[[@b49-medscimonit-25-9949]\]. Our results also show that the expression level of rhodopsin is downregulated in atherosclerotic macrophages. This suggests that rhodopsins have potential inhibitory effects on atherosclerosis, but this conclusion is controversial, and more investigations are needed to disclose the mechanisms and pathways mediating the various processes in atherosclerosis carried out by the archetype of the largest class of GPCRs in the human genome. In the present study, among the 5 predicted TFs, early growth response-1 (EGR1) had the most targeted-DEGs in the TF-target regulatory network. EGR1 is thought to play an important role in the pathogenesis of atherosclerosis, and critical work in the identification of the biological link between EGR1 and atherosclerosis came from experiments by the laboratories of McCaffrey et al. \[[@b50-medscimonit-25-9949]\], who found that transcripts for EGR1 were upregulated in human atherosclerotic plaque and the lesions of the LDL (low-density lipoprotein) receptor-deficient mice fed with a high-fat diet. Moreover, immunohistochemistry localized EGR1 to macrophages, endothelial cells, and vascular smooth muscle cells (SMC). Within the atherosclerotic plaque, increased EGR1 expression in the lesion was positively correlated with elevated expression of several known EGR1 target genes, such as TNF, ICAM-1, and M-CSF, suggesting that EGR1 is transcriptionally active in human atheroma \[[@b50-medscimonit-25-9949]\]. In the present study, EGR1 was predicted to target 3 DEGs: ERBB3 (erb-b2 receptor tyrosine kinase 3), AR (androgen receptor), and NAB1 (NGFI-A binding protein 1). It was shown that ERBB signaling plays a role in the control of proinflammatory activation of monocytes \[[@b51-medscimonit-25-9949]\]. Experiments using the ARKO mouse model suggest an atheroprotective role for AR \[[@b52-medscimonit-25-9949]\]. EGR-1 is the partner identified for NAB1, and it appears that NAB1 serves as a repressor for the EGR-1-responsive expression of proinflammatory associated genes in many cells \[[@b53-medscimonit-25-9949],[@b54-medscimonit-25-9949]\]. Specific function of ERBB3 and NAB1 remain unclear, as the EGR-1 downstreams influence an inflammatory reaction in macrophages in atherosclerotic plaque. Nonetheless, from the current evidence, it is highly likely that these unascertained downstream interacting partners of EGR1 also participate in the process of atherogenesis; thus, whether and how these genes contribute to atherogenesis suppression involved in the EGR1 associate-regulatory network is a promising direction of research in the future. Taken together, these findings provide definitive mechanistic support for the link between transcription factor EGR1 and the pathogenesis of atherosclerosis, and suggest that the present analytic results have theoretical significance. However, the present study has some limitations: the number of specimens in the existing database limited our efforts to include more samples in the study, and a finite number of macrophage examples were included in the analysis, which might affect the values of these results and findings. The dataset GSE7074 had been used in another study \[[@b55-medscimonit-25-9949]\]. Compared to the previous research, our results demonstrated distinct DEGs. A possible reason for this difference may be that macrophage samples used in the previous study contained peripheral blood-derived macrophages from patients with atherosclerosis, which is obviously distinct from the *in situ* atherosclerotic tissue-derived macrophages used in the present study. Furthermore, the previous study used the online differential gene analysis platform GEO2R, but we utilized the rigorous code of R software for data processing, and the criteria and thresholds defined in differential gene screening are not all the same. The findings obtained from the microarray method still need experimental verifications, and the clinical significance of the DEGs and predicted miRNAs/TFs in disease need further elucidation. Conclusions =========== We found 236 significantly mutual DEGs from transcriptomic comparisons of human atherosclerotic and nonatherosclerotic macrophages covering the majority of the human body's internal environment. Specially, 3 highlighted downregulated genes (RHO, F5, and BEST1) in the PPI network were identified as potential suppressors or diagnostic biomarkers for atherosclerosis formation and development. Moreover, hsa-miR-3177-5p was predicted as the potential upstream regulator for BEST1 and might be involved in the progression of atherosclerosis, and the transcription factor EGR1 was the potential facilitator in atherogenesis and could serve as an underlying target for atherosclerosis treatment. These findings provide the theoretical basis for the future study in this disease and provide insight into the underlying molecular mechanisms within the initiation and progression of the disease. However, more work is needed to better define the specific role of these genes in tissue-specific metabolites and macrophage function in different tissues, along with how they are changed during inflammation and the general procedure of atherosclerosis formation. **Source of support:** This work was supported by the National Natural Science Foundation of China (grant no. 81571144) and the Independent Research Project of Natural Science Foundation of Tianjin City (grant no. 16JCZDJC35700) {#f1-medscimonit-25-9949} {#f2-medscimonit-25-9949} {#f3-medscimonit-25-9949} {#f4-medscimonit-25-9949} {#f5-medscimonit-25-9949} {#f6-medscimonit-25-9949} {#f7-medscimonit-25-9949} {#f8-medscimonit-25-9949} {#f9-medscimonit-25-9949} {#f10-medscimonit-25-9949} ###### Screening DEGs acquired from analyzed data. Gene names (upregulated DEGs) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ASAP2, ARL4C, SNRK, ITGA10, DISP1, SFRP2, NAB1, LOC339803, FCGBP, LRP12, LRRC29, AGPAT4-IT1, COL16A1, GHRL, AMMECR1L, BVES-AS1, ASAP3, STPG1, B4GALT5, TRPM2, FAM64A **Gene names (downregulated DEGs)** C14orf2, ZNRD1, SPRYD7, BTD, LHX3, NSUN5, FAM118B, ZNF460, RAD51, AGTRAP, FAM20C, PSMB10, GPX3, PKIG, CYFIP2, TNFRSF25, LRRC4C, AMOTL2, EFNA5, PTHLH, ELAC1, BRD2, U2SURP, NOX4, SOX14, RP1, ALDH1A1, HS3ST3A1, HMGN4, AR, OLFML2B, HOXB5, KCNK1, PID1, EYA3, C14orf169, OGDH, ENAM, OXA1L, ZNF608, IGSF9B, KRT8P12, TBCE, DENR, SENP3, PDE6A, POLR3B, IL2RB, PIK3CB, IL17RA, PROS1, SLC1A7, SEC23IP, ATP9A, ZNF609, STARD3, RPS6KA3, TUBA8, LOC158402, HCRP1, TMEM38A, NDUFB10, JPH2, VIPR1, FAM189A2, SRR, DLK1, SHROOM2, MPO, FGF11, NPTXR, CD302, STMN1, DRD5, CYP4F8, KCNQ1OT1, CD79A, ZNF277, GBA, PRTFDC1, MAD2L2, VPS11, MYO15A, PDCD10, MYL9, IQCC, TNFAIP6, C1orf159, HAX1, FSCN1, FBXO21, SNORA71B, FBXO24, GPD1L, FPR3, UBL7-AS1, ARNTL2, NDRG4, P2RX4, DRD3, RNASE7, PEX11B, HDLBP, PROP1, HOXC6, FAM204A, MIS18A, KCNJ2, NRSN2, GCNT3, SNAPC2, SLC52A1, SPARCL1, POLR2F, PAX4, PRDM1, CRTAC1, RHO, XXYLT1, KIAA1324, MOGAT3, POLD4, DGAT1, RNLS, CRYBB1, ATP13A1, ZNF496, EIF2B4, NOCT, CHGB, DNAJC2, SLC25A40, LOC340085, ZNF154, FAM131A, HIST1H3E, NOC3L, CCNJ, ARL6IP5, NKX2-8, XCL1, XKR8, ERBB3, VPREB3, ICOS, IL3, GOSR2, OR5L2, KIFC1, BCO1, DUSP22, F5, CLCNKB, CDC42EP2, ACSS2, ZNF148, NLRP1, PDE6G, SCD5, XCR1, SLC39A10, CNTLN, HCRTR2, LINC00115, TULP2, CCM2, SGO1, VIPR2, FUT8, WNT2B, PDPK1, AP1S1, ZNF593, TSTA3, IFT20, MGAT3, AIPL1, POLD2, OR1A1, NAA20, ABCA4, TRAF4, ATP6V1G1, LRRC8A, SPSB4, HIST1H2BL, EFHD1, HIC1, MAP3K12, ROM1, PSMD6-AS2, LOC100507547, YOD1, NDUFA3, MEF2C-AS1, LHX2, SH2B1, AES, KCNE5, CNGA3, CAPZB, SEMA3F, C2, ARNTL, DNMT3L, OPLAH, ARHGEF17, MNS1, ANXA9, TFPI, DYNC2LI1, ICE2, GCLC, PRO1596, BEST1 ###### The top ten results of GO enrichment analysis of differentially expressed genes. GO ID Term Count P-value ------------- ---------------------------------------- ------- ------------- GO: 0007601 Visual perception 13 6.21E-06 GO: 0007603 Phototransduction, visible light 4 1.39E-04 GO: 0097381 Photoreceptor disc membrane 4 0.001328569 GO: 0005887 Integral component of plasma membrane 28 0.007973519 GO: 0016056 Rhodopsin mediated signaling pathway 3 0.008862666 GO: 0042622 Photoreceptor outer segment membrane 3 0.016475551 GO: 0016477 Cell migration 7 0.018312903 GO: 0000139 Golgi membrane 14 0.022394358 GO: 0007596 Blood coagulation 7 0.02453119 GO: 0010629 Negative regulation of gene expression 6 0.025500076 ###### The top ten results of KEGG pathway enrichment analysis of differentially expressed genes. Term ID Input.count P-value ------------------------------------- ---------- ------------- ------------- Purine metabolism hsa00230 7 0.001546974 Phototransduction hsa04744 3 0.002534266 Metabolic pathways hsa01100 22 0.002834313 Homologous recombination hsa03440 3 0.003054876 Pyrimidine metabolism hsa00240 5 0.00367646 RNA polymerase hsa03020 3 0.003951027 Complement and coagulation cascades hsa04610 4 0.007059075 cAMP signaling pathway hsa04024 6 0.012042269 Glutathione metabolism hsa00480 3 0.013195502 mTOR signaling pathway hsa04150 5 0.015864191 ###### Top 6 genes with the degree ≥8 in the protein-protein interaction (PPI) network. Gene LogFC Adj.P Degree ----------- -------------- ------------- -------- RHO −1.325868022 0.035374995 10 F5 −1.545198153 1.11E-05 9 BEST1 −3.456209744 2.66E-08 9 HIST1H2BL −1.977145687 0.006584092 8 PDE6G −1.59907313 0.026771841 8 POLR2F −1.305954127 0.014398162 8 LogFC -- log2-fold change; Adj.P -- adjusted P-value. [^1]: Study Design [^2]: Data Collection [^3]: Statistical Analysis [^4]: Data Interpretation [^5]: Manuscript Preparation [^6]: Literature Search [^7]: Funds Collection [^8]: Weihan Wang and Kai Zhang contributed equally to this work

###### Strengths and limitations of this study - This is the first Norwegian valuation study with composite time trade-off and discrete choice experiment undertaken on a scale large enough to meet the recommendations of the most recent EuroQol (EQ) five dimensions protocol. - Sampling strategy designed to both ensure representativeness of the final sample according to geographical region, age, sex and educational level and increase the number of experience-based valuations. - Data collection restricted to EQ protocol, primarily developed for hypothetical health state valuation, but with the additional aim of collecting experience-based valuations. Study design does not allow for the assessment of methods other than those described in the EQ-valuation technology protocol. - Restricted samples for comparisons of experience-based valuations. - High respondent burden experienced in interviews limits the scope for addressing additional methodological questions. Introduction {#s1} ============ Economic evaluation undertaken by the Norwegian Institute of Public Health (NIPH) and the Norwegian Medicine Agency increasingly informs decisions about the introduction of new drugs and other health technologies in Norway.[@R1] The Norheim Committee[@R3] and Magnussen Working Group[@R4] proposed methods to enhance the quality of economic evaluation, thereby further strengthening the role of economic evaluation in decision-making. The Ministry of Health followed up these proposals in a 2016 White Paper to Parliament on principles for priority setting in healthcare.[@R5] Given the important role and impact of economic evaluation, it is important that the methods it incorporates, including cost--utility analysis, are consistent with societal values regarding publicly financed healthcare. Economic evaluation, when taking into account societal values, often takes the form of cost--utility analyses with the estimation of the incremental cost per quality-adjusted life year (QALY) gained.[@R6] QALY takes the integral of health-related quality of life (HRQoL) over time, with HRQoL represented on a scale where 1 indicates a preference equal to that for full health and 0 implies a health state equal to that of being dead. Values are typically derived using general population surveys where respondents consider the relative undesirability of different health states described using instruments such as the EuroQol five dimensions (EQ-5D).[@R7] After assigning values to health states described by an instrument, QALYs are calculated by multiplying the health state value by the length of time spent in each. Evaluation of alternative technologies then involves comparison of incremental QALYs gained over incremental costs for new vs existing technologies. Several instruments are available to calculate QALYs, of which the EQ-5D is by far the most widely applied both internationally and in Norway.[@R8] The EQ-5D, a trade mark of the EuroQol Research Foundation, is available in over 150 languages[@R11] in the self-complete paper version,[@R12] and national value sets and normative data exist for over 20 countries.[@R7] It is brief, widely tested, and includes five important aspects of health (mobility, self-care, usual activities, pain/discomfort and anxiety/depression), with the most recent version having five levels (5 L) from no problems to severe problems. The EQ-5D is considered highly acceptable to most patient groups and feasible for application where a short-form general measure of health is required. The instrument has had widespread application in research including clinical trials, population health surveys, in both Norwegian[@R26] and Swedish National Quality Registries,[@R27] and more recently as a healthcare quality indicator as part of the National Health Service for England and Wales Patient-Reported Outcomes Measures programme.[@R14] The Norwegian Medicines Agency recommends the use of EQ-5D in all technology assessments and the use of a 5 L tariff for studies where the 5 L version has been used.[@R2] In the absence of a Norwegian tariff, the 2018 Agency guidelines currently recommend the use of the EQ-5D-5L tariff for England[@R14] where EQ-5D-5L has been used. However, criticism has been levelled at the English tariff including concerns with data quality in which serious deficiencies were revealed.[@R28] The English 5 L tariff followed an early protocol, which has since been updated with the aim of improving data quality and interview techniques. Following these concerns, and in contrast to recommendations of the Norwegian Medicines Agency, National Institute for Health and Care Excellence continues to recommend the use of the 3 L tariff over the 5 L tariff, with 5 L values mapped onto 3 L where appropriate.[@R29] The EQ-5D is widely used in Norway, including the national quality registers where it is the most widely used patient-reported outcome measure. A national EQ-5D-5L value set and scoring algorithm is highly anticipated and will enhance the validity of economic evaluations in Norway. Norwegian EQ-5D users have largely relied on the EQ-5D-3L scoring algorithm from the UK,[@R30] with a crosswalk-based approach[@R31] for studies that have used the 5 L version. Crosswalk-based approaches have several limitations related to issues with data dependency and differences in scale range, and are an interim solution pending a national 5 L value set.[@R31] Cross-national comparisons of national EQ-5D-5L value sets also suggest that there might be substantial differences across countries[@R13] with culture and values having a role.[@R35] Values for health for the 5 L version of the EQ-5D, that are representative for the Norwegian general population, will enhance the validity and legitimacy of economic evaluation in Norway. With few exceptions,[@R36] existing EQ-5D value sets are based on the general population valuing hypothetical health states, which follows recommendations that economic evaluation should include societal preferences.[@R39] In recent years, there has been some criticism levelled at this approach, questioning the validity of health state valuations from a general population lacking the adequate experience or knowledge of the health states which they value in the form of hypothetical health states.[@R40] An alternative approach, as recommended by Sweden's Dental and Pharmaceutical Benefits Agency,[@R42] involves individuals valuing their own health state to give experience-based values or basing their valuations on other forms of experience. The debate on whether to use hypothetical or experience-based values is to a certain extent a normative issue, relating to what we aim to maximise.[@R43] However, there are a number of empirical questions pertaining to experience-based values. Arguably, patients have a better understanding of the consequences of reduced health on quality of life.[@R41] On the other hand, they may have trouble imagining life in full health, may under-report impact of disease due to adaptation or changes in expectations over time,[@R44] or may be less inclined to value their current health state as a state that is worse than being dead. Experience-based valuations, if better understood and elicited from representative samples of the general population may, however, be suitable for inclusion as societal values. The feasibility of collecting experience-based values, the assessment of how those with less than perfect health value their current health state and other health states in general, and how different forms of experience may influence health state valuations, are new areas for research to which this study will contribute.[@R48] The project will derive a Norwegian EQ-5D-5L value set representative of region, age, sex and level of education composition in the Norwegian adult general population. Furthermore, the study will allow for comparisons of experience-based and hypothetical health state valuation. Methods and analysis {#s2} ==================== Values for EQ-5D-5L health states will be obtained by electronic data collection including computer assisted face-to-face, one-to-one interviews and the use of composite time trade-off (cTTO) and discrete choice experiments (DCE).[@R49] The latest EQ-5D-5L protocol will be followed including EQ-valuation technology (EQ-VT). Sampling {#s2-1} -------- Respondents must be aged 18 years or older, resident in Norway and proficient in Norwegian. Following EQ-VT protocol, sample size is set to a minimum of 1000 individuals with each valuing 10 health states, which gives the recommended 10 000 responses.[@R50] Additional 300--500 interviews, based on the oversampling of those with less than perfect health, will increase the number of valuations of experienced health states. Norway is a Northern European country with a population of slightly more than 5 million, and a universal healthcare system. The population covers a comparatively large land mass, and for many there may be several hours travel time to the nearest hospital or large city. Urbanisation has further contributed to variation in demographic characteristics at the regional level. These factors combined with local culture, politics and traditions mean that geographical considerations are important to the design of the study. The study will use a combination of multistage random sampling and quota sampling ensuring representativeness according to geography, age, sex and educational level. The first stage of sampling will be of geographical areas, here defined as municipalities within each acute care hospital catchment area. Norway's four regional health authorities include Northern, Central, Western and South-Eastern, with more than half the population residing in the South-Eastern health region. The catchment areas served by the 54 acute care hospitals cover all Norwegian residents (see [figure 1](#F1){ref-type="fig"}). They vary considerably in the number of residents that they serve, from 15 000 to 500 000 residents. One acute care hospital will be randomly selected from each health region with the exception of the South-Eastern region, where three will be randomly selected to account for the disproportionate number of people residing in this region. Hospital catchment areas within each region will be sampled with proportional allocation, ensuring equal probability proportionate to the number of people residing in each area within the region. {#F1} Within each sampled geographical area, the possible locations for data collection will constitute the sample frame for the second stage of sampling ([table 1](#T1){ref-type="table"}). Locations will include public places (eg, public libraries, town halls), workplaces, recreational organisations (eg, sports clubs) and healthcare providers (eg, hospitals, rehabilitation institutions). The bodies concerned must be willing to grant the study permission for data collection and cooperate with provision of a suitable space for completion of the interviews. The locations will act as clusters of possible respondents, stratified into groups based on the characteristics of target respondents, for example, age and educational level. Stratification will increase homogeneity per cluster and ensure the representation of specific groups less likely to participate including those with poorer health, lower socioeconomic status, or faced with time constraints, including those with young children or in full-time employment.[@R51] Locations within each group in the sample frame will be randomly selected. The number of locations selected within each sample frame will be based on the size of the area and quotas. Response rates, recruitment and data quality will be assessed for the different location strata and compared across catchment areas. ###### Locations for recruitment of participants, by age group and health status Healthy Reduced health ------------------------------------------ ------------------------------------------- ------------------------------------------------ ---------------- ------------------------ Places of higher education Workplaces Eldery homes Public library Hospitals Child daycare facilities/Primary schools Recreational organisations (sports teams) Recreational organisations (choirs/orchestras) Town hall Rehabilitation centres Social welfare\* Social welfare\* Community volunteer centres Health centres Adult education\* \*Locations chosen to increase participation of those with lower socioeconomic status. Within each catchment area and at the respondent level, quota sampling will be applied according to age, sex and level of education (see [tables 2 and 3](#T2 T3){ref-type="table"}). The total quota will first be allocated to each region proportionate to the number of people residing in each region. For the three regions with one sampled hospital catchment area, the quota for each of these hospital catchment areas will correspond to the regional quota. In the South-Eastern region, the regional quota is further allocated to each hospital catchment area proportionate to the number of people residing in each of these areas. The quota is then allocated to groups according to gender, age group (young adults: age 18--34, middle-aged adults: age 35--64, elderly: age 65+) and level of education (lower education, no higher than high school education, higher education-bachelor, masters or PhD) equivalent to the distribution of these attributes in the respective regions. The quotas for each group are calculated using data available from <http://microdata.no> (see [table 4](#T4){ref-type="table"}), a national platform in Norway giving researchers direct access to national registries for which Statistics Norway has processing authority, such as the Norwegian National Registry, National Education Database, labour market data, register for Personal Tax Payers and FD-Trygd (event history database).[@R52] ###### Example sampling of hospital catchment areas and quotas per catchment area Region Population in region Catchment area Population in catchment area Quota per catchment area --------------- ---------------------- ---------------- ------------------------------ -------------------------- Northern 381 907 Hospital 1 130 000 140 Central 560 690 Hospital 2 60 000 205 Western 843 899 Hospital 3 330 000 309 South-Eastern 2 299 890 Hospital 4 500 000 448 South-Eastern '' Hospital 5 160 000 143 South-Eastern '' Hospital 6 280 000 251 ###### Example of quotas within a sampled catchment area based on the composition of sex, age and educational level in the general population of the respective region (source: official statistics for 2017 generated from microdata.no) Sex Highest attained educational level Age groups Total quota per sex and educational level --------------------------- ------------------------------------ ------------ ------------------------------------------- ---- ---- ---- ---- ----- ---- Male Primary or secondary 8 9 8 9 9 7 5 56 Tertiary 1 3 3 3 3 2 1 16 Female Primary or secondary 7 6 5 7 8 8 7 47 Tertiary 1 5 5 5 3 2 1 22 Total quota per age group 17 22 22 25 22 18 14 140 Example given sampling scenario and catchment area for Hospital 1 in [table 2](#T2){ref-type="table"}. ###### Reference data for the calculation of quotas, data for 2018 (<http://microdata.no>, statistics Norway, data accessed: 12 March 2019) Region Sex Highest attained educational level Age group ---------------------- ---------------------- ------------------------------------ ----------- --------- --------- --------- --------- --------- -------- South-Eastern region Male Primary or secondary 117 220 130 448 133 470 143 252 119 278 94 473 62 167 Tertiary 13 603 72 661 77 273 66 785 51 553 40 368 19 650 Female Primary or secondary 100 571 94 904 99 033 120 226 114 228 107 739 103 859 Tertiary 24 196 104 395 101 833 79 908 55 565 34 126 17 126 Western region Male Primary or secondary 48 863 54 616 52 141 54 172 44 925 34 032 23 977 Tertiary 5129 26 041 27 176 21 446 17 552 12 302 5291 Female Primary or secondary 40 743 35 932 34 778 42 701 40 672 36 866 38 127 Tertiary 9928 39 550 36 796 27 107 18 494 9777 4750 Central region Male Primary or secondary 32 425 33 771 32 095 36 110 32 525 26 289 18 441 Tertiary 3674 15 730 15 703 13 497 11 291 8664 3521 Female Primary or secondary 26 707 21 526 21 130 28 292 29 998 28 218 28 275 Tertiary 6456 23 177 22 577 18 320 12 267 6980 3024 Northern region Male Primary or secondary 22 976 23 320 21 793 25 812 23 582 20 282 13 464 Tertiary 1736 7895 8724 9427 7450 5273 1845 Female Primary or secondary 18 357 15 382 14 562 19 478 20 589 20 492 19 767 Tertiary 3470 13 212 14 402 13 721 8872 4300 1707 The study will largely rely on recruitment of potential participants by contact persons at each sampled location. Contact persons will assist in identifying and recruiting potential respondents to the study. Prior to data collection, contact persons will receive information and materials for publication in local newspapers and social media designed to enhance participation. In addition to the recruitment of respondents through locations, potential respondents will be able to contact the project group for more information about the study and enquire about participation. Potential respondents will be informed of a gift incentive. Cash has been found to be more effective than other incentives for increasing response rates and following the interview, respondents will receive a cash card equivalent to €30. Data collection will take place from November 2019 to June 2020. Depending on the final sampling, and with an estimate of a minimum of four interviews per interviewer per day, a minimum of 55--80 working days are required for data collection. The recruitment strategy will be piloted in the catchment area sampled closest to Oslo. Necessary adjustments will follow before data collection proceeds in the rest of the country. Interviewer training {#s2-2} -------------------- Interviewers with Masters education level or equivalent will receive training in accordance with EuroQol Foundation guidelines and recommendations, with initial training prior to, and revised training after, the first phase of data collection.[@R53] Based on existing studies and recommendations from EQ (Elly Stolk, personal communication), eight to twelve interviewers are required. Quality control (QC) reports will help monitor progress and data quality throughout.[@R54] The reports will include assessments of protocol compliance, face validity of data collected and value distributions per interviewer. QC reports have been found to further the homogeneity of interviewer performance and reduce protocol violations and the number of inconsistent responses.[@R54] Interviewers not meeting predefined standards will be flagged, recommended for retraining and ultimately excluded. Evaluation of the data collected and interviewer performance will be regularly discussed with interviewers in face-to-face group meetings throughout data collection, and with EQ contact persons. EuroQol valuation technology {#s2-3} ---------------------------- EQ-VT was developed to meet the challenges involved with valuation of the 5 L version of the EQ-5D, with emphasis on improving data quality and cross-country comparability.[@R49] The standard protocol includes digital representation of visual aids to assist the respondent throughout the interview (see [figures 2 and 3](#F2 F3){ref-type="fig"}). The study will use the portable version of the software, EQ-PVT, which for the respondent has the same functionality and for the most part resembles the standard EQ-VT software package. {#F2} {#F3} The interview will start with administration of the EQ-5D-5L questionnaire, including the Visual Analogue Scale (VAS), followed by background questions for age, sex and experience with serious illness. Next, cTTO is administered, beginning with an explanation of the task demonstrated with 'the wheelchair example' including an introduction to the 'worse than dead' part of the task. This is followed by practice tasks for three states described with the EQ-5D-5L descriptive system, selected to reflect a mild, moderate and severe health state, familiarising the respondent further with the cTTO, the concept of health states worse than being dead and the use of lead-time in the cTTO for the valuation of such states. Lastly, respondents are administered their current health state as a cTTO task, allowing for the comparison of how respondents value their own health state with both cTTO and VAS. Respondents are randomised to 1 of 10 TTO blocks of EQ-5D-5L health states, each consisting of 10 health states, one of which is always the worst state (level 5 on each dimension, state 55555), and one among the 5 mildest states (11112, 11121, 11211, 12111 and 21111), for a total of 86 unique EQ-5D-5L health states for direct valuation.[@R49] Respondents get the opportunity to review their responses in a feedback module (see [figure 4](#F4){ref-type="fig"}), where individual task responses can be removed. On completion of the TTO tasks, respondents are randomised to 1 of 28 state pair blocks for discrete choices, each block consisting of seven state pairs. In both the TTO and DCE parts of the interview, the order of presentation is randomised. The randomised TTO and DCE tasks do not explicitly include a valuation of the respondents own health state, however, respondents can by chance be presented their own health state as a choice, in which case the task will be completed as normal. {#F4} The interview ends with further background questions specific to this study relating to variables known to be associated with valuations of health states including caregiver status, educational level and marital status.[@R56] The influence of such variables will be assessed for the final value set. Analysis {#s2-4} -------- The demographic characteristics and health status, that is, EQ-5D-5L profile, of respondents will be assessed and compared with national data. Parallel to this study, the NIPH has initiated data collection for a postal survey assessing the health status of the Norwegian population using the EQ-5D-5L, allowing for comparison of the health status of study populations. Health state values for EQ-5D-5L will be estimated through statistical modelling of the survey data. The EQ-5D protocols are not prescriptive with regard to modelling and approaches will depend on the characteristics of the data obtained.[@R50] Following previous research, different models will be assessed including either the cTTO data, or combining the cTTO and DCE data in a hybrid model, and the results compared for adequacy with those for existing national value sets.[@R14] Modelling of values for the national value set will exclude valuations from respondents recruited from locations specifically for the collection of experience-based values and the valuations of respondents' own health state. Subgroup analysis will identify variables contributing to health state valuation in the Norwegian population. Values for health states defined as respondents' 'own health today' will be compared with values estimated for the same health states by the general population. In addition, all experienced-based valuations by those with serious illness and/or less than perfect health will be compared with valuations based on the total general population sample and, given sufficient data, those without experience of serious illness and/or with perfect health today. To assess experience-based valuations, and explore both the wider and more narrow concepts of experience-based valuations,[@R48] three potential profiles will be assessed: (1) respondents' valuation of own health state, (2) valuations given by respondents recruited from locations specifically chosen to target those with poorer health, that is, health services, (3) valuations given by respondents who have indicated that they have experience with serious illness. Patient and public involvement {#s2-5} ------------------------------ Patients and members of the public were not invited to comment on the study design or contribute to the writing or editing of this document for readability or accuracy. Strengths and limitations {#s2-6} ------------------------- This is the first Norwegian valuation study with both cTTO and DCE undertaken on a scale large enough to meet the most recent EQ-5D protocol. The study intends to complete 1300--1500 face-to-face computer-assisted interviews across a country with a relatively dispersed population of citizens and large geographical distances between them. Data collection involves a small number of interviewers working over an 8-month period. Both the duration and magnitude of the tasks involved make the interview demanding. It is important that data collection is cost-effective, which includes considerations of data quality, representativeness and total number of valuations. Given the strategy of sampling locations and organisations rather than individuals, the assessment of its effectiveness in terms of number and representativeness of respondents will be important following the initial data collection period. Poor recruitment and data collection in remote geographical locations will be costly. The number and characteristics of respondents per location will be monitored throughout data collection. Adaptive sampling will allow for inclusion of additional locations where response rates are low and quotas are not met. Additional locations will be chosen at random from the predefined frame of possible locations within the selected geographical area. Due to the need for extensive training, interview experience and understanding of the task, only 8--12 interviewers will be included. This will give more control over the data collection and the quality of the data collected. However, this and potential costs saved in terms of interviewer recruitment, training and travel costs, must be balanced against the increased impact of any loss of interviewers through illness or resignation during data collection. Norway has a harsh winter climate and apart from the Southern and Eastern region, where the interviewers are based, the interviews will primarily take place outside the winter months to reduce the risk of travel delays and interviewer illness. The NIPH, which is conducting the research, has several experienced interviewers familiar with the study who will be able to complete training and contribute to data collection if needed. The main justification for the strategy of sampling stratified locations and the use of quotas on the respondent level is to ensure representativeness of the final sample according to geographical region, age, sex and educational level. An additional quota will be used to recruit those with less than perfect health, through locations such as hospitals and rehabilitation centres. Locations will also be selected to directly seek out others who are typically harder to reach and are less likely to participate in research studies, such as those with different ethnic backgrounds or with young children. Studies have found that some attributes, such as marital and caregiver status/having young children, may influence the respondents response to the task, such as their willingness to trade time in the TTO task, despite showing similar preferences for given health states when using other types of valuation tasks.[@R56] Hence, it is important that respondents with such attributes are included in the study and locations such as day care facilities for young children and primary schools will be selected to facilitate this. Questions relating to these attributes will be included in the background questions closing the interview, and as such will allow for subgroup analysis of the effect of these attributes on the valuation of health in the Norwegian sample. The derivation of values based on experienced health states is a recent development in the field of health state valuation.[@R48] In recent years, there have been major developments in the field of standardised protocols for health state valuation, including EQ-VT. Such standardisation is a long way off for experienced health state valuation and, as was the case for hypothetical health state valuation up until the last decade, there is considerable variation in the choice of methods.[@R60] In Norway and other countries, the feasibility of collecting such data is still in its infancy, including choice of sampling strategies, recruitment and how to minimise respondent burden. This study builds on existing methodology in the form of EQ-VT protocol, to assess the feasibility of recruiting potential respondents (including from healthcare settings) for experience-based health state valuation, respondent burden in the form of completed interviews and data quality. The study design is constrained by the EQ-VT protocol, but the results of the study will inform the development of more appropriate methodology in the future. Furthermore, the design will allow the comparison of results with those for hypothetical health state valuation. Ethics and dissemination {#s3} ======================== The study was reviewed by The Regional Health Authority Research Ethics Committee and found to be outside of the scope of the ethics committee thus not in need of ethical approval. All study participants will give informed consent. The final scoring algorithm will contribute to the quality and relevance of the results of EQ-5D applications in Norway, and it is highly likely that, when available, the EQ-5D-5L with a Norwegian scoring algorithm will be the recommended instrument of choice for future economic evaluations undertaken in Norway by the pharmaceutical industry and other important users. Application of the same instrument and scoring across the health services and industry will further enhance decision-making relating to scarce healthcare resources. Moreover, scores based on Norwegian preferences will further enhance the appropriateness of the EQ-5D in clinical and health services research and quality indicators work, including the national quality registers. The study results will be published in peer-review scientific journals, presented at appropriate forums, including national and international conferences, and scoring algorithms made publicly available for R, Stata and other widely used statistical software. Presentations will be given to users of the research, including research centres that widely use the EQ-5D in clinical, health services and health economics research in Norway. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript **Contributors:** AG conceived the study and secured funding. TMH, YH and AG designed the study. LAA, KR and KS commented and recommended revisions. TMH and AG drafted and revised the manuscript. YH, LAA, KR and KS have read and approved the final version. All authors agree to be accountable for all aspects of the work. **Funding:** The study is funded by the Norwegian Research Council (grant number: 262673) with additional support from the EuroQol Foundation (EQ Project 20190280) and Norwegian Institute of Public Health. **Map disclaimer:** The depiction of boundaries on this map does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. This map is provided without any warranty of any kind, either express or implied. **Competing interests:** None declared. **Patient consent for publication:** Not required. **Provenance and peer review:** Not commissioned; externally peer reviewed.

Introduction ============ The annual economic cost of chronic pain in the United States is estimated to be \$560--635 billion, with arthritis accounting for approximately \$190 billion (1/3rd) of that total ([@ref-17]). The same report recommended various strategies to help address this problem, and these included the development of more and better animal models of chronic pain. Our report describes the initial evaluation of a naturally occurring animal model of arthritis as a model of arthritis-induced sleep disturbance. One-third to half of those people with painful arthritis suffer from sleep disturbances ([@ref-20]; [@ref-22]; [@ref-31]; [@ref-32]; [@ref-28]) often associated with restlessness ([@ref-13]). Pain is thought to mediate a substantial amount of the relationship between arthritis and sleep problems ([@ref-22]; [@ref-32]). Resulting sleep deprivation has been shown to have hyperalgesic effects, worsening the overall pain state ([@ref-21]; [@ref-9]). A recent focus group study highlighted the importance of night pain in human osteoarthritis (OA), and the relative lack of studies on night pain ([@ref-32]). In dogs, OA is a common condition affecting a large percentage of the population. The disease process of canine OA of the hip is considered to be very similar to human OA ([@ref-4]) and therefore a potentially useful model---a spontaneous disease model ([@ref-8]). Additionally, such a model has the added benefit that these companion dogs share the same environment as humans, making the model more relevant. Recently, the use of activity monitors as an objective measurement of a dog's spontaneous activity has been described and validated ([@ref-6]; [@ref-3]; [@ref-18]). Day-time activity, as measured by an accelerometer, increases when non-steroidal anti-inflammatory (NSAID) analgesia is administered ([@ref-3]; [@ref-30]) or when therapeutic diets are introduced ([@ref-23]). Our clinical veterinary experience is that oftentimes, owners of dogs with OA report that their pets suffer nighttime restlessness (unpublished observations); clinical observations suggest this dissipates when analgesics (such as NSAIDs) are administered. This nighttime restlessness may be a manifestation of pain-associated sleep disturbance as seen in human OA patients, and so could present a good pre-clinical model of this complex state associated with OA. If day-time activity can be measured using accelerometry, and is measurably increased when analgesics are administered to dogs with OA-pain, it is possible that any nighttime restlessness associated with pain may be measureable in the same population. The aim of our study was to perform preliminary investigations into whether naturally occurring canine OA is associated with nighttime restlessness as assessed using accelerometry and a clinical metrology instrument, and whether this might be a potential model of OA-associated sleep disturbance in humans. We hypothesized that NSAID-responsive nighttime restlessness occurs in dogs suffering from painful OA. This study used objective accelerometry (AM) and a novel clinical metrology instrument (Sleep and Night Time Restlessness Evaluation Score; SNoRE) as primary outcome measures to assess the nighttime activity changes with administration of an NSAID, and to understand what demographic factors influenced nighttime activity. Materials and Methods ===================== Design ------ Data for the current study was collected in two parts. Part A used previously unreported data from a masked, parallel, placebo-controlled clinical study evaluating dose titration of NSAIDs in dogs with naturally occurring OA pain which has been described already ([@ref-30]). Part B used data gathered from a different cohort of dogs from a masked, placebo-controlled crossover study evaluating the effect of an NSAID versus placebo on nighttime activity in dogs with naturally occurring painful OA. The Institutional Animal Care and Use Committee (IACUC) approved both studies (IACUC \#08-077-O, \#07-188-O), and in all cases owners signed a written consent form following a detailed explanation of the study protocol. Study populations ----------------- Sixty and nineteen client-owned dogs with OA-associated pain and mobility impairment were included in Part A and Part B, respectively. Dogs of any breed, age, sex or weight were recruited. Group size in Part A was based on power calculations determined from the expected decrease in efficacy with NSAID dose titration, and published previously ([@ref-30]). Study B was designed as a pilot study focused on evaluating nighttime restlessness. Power analysis for this was based on the known change in pain following NSAID administration, detected using validated owner assessment tools. The expected difference between placebo and active treatment was 30%, and the SD 40%, giving us an estimation of 16 dogs being required in the crossover study for an 80% power for owners detecting the improvement due to an NSAID. ### Inclusion criteria To be eligible for either study, dogs were required to have impaired mobility (according to their owners), and to be considered by the investigators as clinically appropriate candidates for pain relief using NSAIDs. They were required to have not received oral or parenteral steroids or injectable polysulphated glycosaminoglycans within the last 4 weeks, and owners were required to agree to stop administering NSAIDs 2 weeks prior to the start of the study. The screening evaluation included a physical, neurological and orthopedic examination and a complete blood count (CBC), serum biochemical analysis and urinalysis. Exclusion criteria included the presence of suspected or demonstrated systemic or local disease other than OA. Dogs were also excluded if they were suffering from recent joint instability such as cranial cruciate ligament rupture, or had undergone joint surgery within the previous 12 months. Dogs known to have intolerance to NSAIDs were excluded from the study. Other exclusion criteria were confirmed or suspected neurological, cardiac or endocrine disease. Results of all laboratory testing must have been either within the reference range values or considered clinically nonsignificant. If alanine aminotransferase or alkaline phosphatase were twice the high end of the reference range, then pre- and post- prandial bile acid tests were performed. If the postprandial bile acid value was within the reference, the dog was considered an acceptable candidate for the study. Digital radiographs of all clinically abnormal (painful) appendicular joints were taken under sedation.[^1^](#peerj-772-fn1){ref-type="fn"}[^1] Dogs with no detectable systemic disease, and with at least one appendicular joint where manipulation elicited an aversive response and whose radiographs showed the presence of OA, were included. Dogs were designated as being either predominantly 'forelimb' or 'hind limb' impaired. Owners were instructed not to change the management of dogs for the period of the studies. Study protocol -------------- ### Part A As previously reported ([@ref-30]), the study lasted for a total of 16 weeks. The present report concerns data collected over the nighttime periods of the first four weeks. On day 0, all dogs started the study but were not administered any analgesics for the subsequent 14 days. On day 14, all dogs received the full FDA-approved dose of NSAID (meloxicam, 0.2 mg/kg meloxicam orally on day 14, followed by 0.1 mg/kg orally, every evening, for the next 2 weeks). On day 0, the owners completed 3 clinical metrology instruments: the Client Specific Outcome Measures (CSOM) ([@ref-11]), Helsinki Chronic Pain Index (HCPI) ([@ref-7]), and the Canine Brief Pain Inventory (CBPI) ([@ref-2]). Accelerometry data was collected continuously over the 4-week period using a collar-mounted accelerometer,[^2^](#peerj-772-fn2){ref-type="fn"}[^2] as previously described ([@ref-30]). When the owner returned on day 14 and day 28 to collect/return the study medication, the accelerometer data were downloaded to a personal computer via a telemetric reader. ### Part B This masked, placebo-controlled crossover study was conducted over a 5-week period. Inclusion criteria were as described above, and NSAID (meloxicam, dose as above) or placebo was administered for the first two weeks, followed by a 1 week wash-out period, followed by another two weeks of NSAID or placebo. Each dog received both NSAID and placebo in a random order. Randomization was based on a 3-block design, randomized in groups of 2 within each block. Randomization was blocked based on high, medium or low impairment, using the CSOM score (Low impairment group = scores of 1--14; intermediate = 15--29; high = 30--44). The placebo was visually identical to the regular meloxicam solution and prepared by the NCSU pharmacy.[^3^](#peerj-772-fn3){ref-type="fn"}[^3] On day 0, owners completed 3 clinical metrology instruments: the CSOM, HCPI, and the CBPI. In addition, they completed the Sleep and Night Time Restlessness Evaluation Score (SNoRE) on days 14, 21, 28 and 35. Owners kept a daily diary of the times the household went to bed, and the time the household got up the following morning. These data were used to define the true nighttime period. Accelerometry data was collected as continuously over the 4-week period, as described above. On day 35, accelerometer data was downloaded to a personal computer as described above. Owners completed the clinical metrology instruments via a phone call on days 14, 21 and 28. Outcome measures ---------------- Primary outcome measures, in both Part A and Part B, were activity monitor (accelerometer) counts over the nighttime periods. In addition, in Part B, the SNoRE was used as a subjective outcome measurement. The CSOM, HCPI, and the CBPI (total score, pain score and activity score) collected on day 0 were used to evaluate how the degree of impairment and pain (measured at the start of the study) influenced nighttime activity. ### Accelerometry (AM) Spontaneous activity of dogs was measured with accelerometers as previously described ([@ref-6]; [@ref-30]). The epoch length was set to 1-minute. In Part A, data from the time period 12 am--5 am during 7 nights of the second week of the study (nights 7--13) and data from the time period 12 am--5 am during 7 nights of the fourth week of the study (nights 21--27) was used in the analysis. The period 12 am--5 am was chosen as these were the average times that a separate cohort of owners (Part B) went to bed and rose the following morning, less 1-hour at the start and the end. In Part B, accelerometer data from the individual nighttime periods for each day (as recorded by owners) were used, but the first hour of the nighttime, and the hour prior to waking were discarded in order to focus on the true nighttime period. This was performed in order to focus on the true nighttime period, and these periods varied from night to night. Only complete hours of accelerometer data were used. ### Sleep and Night Time Restlessness Evaluation Score (SNoRE) SNoRE ([Appendix S1](#supp-1){ref-type="supplementary-material"}) is a 6-item owner-administered clinical metrology instrument that was designed for this study and has not undergone any validation previously. The questionnaire was developed based on some of the most common observations of dog behaviors perceived as painful from the owners over a 10-year period in our clinical practice. It was designed to assess the quality of sleep of dogs over a 7-day time period. Statistical analysis -------------------- Data from Part A study was used to discover variables that were associated positively or negatively with nighttime activity, and to test for differences in nighttime activity between the baseline period (nights 7--13) and the treatment period (nights 21--27). Factors that were significant in Part A study were evaluated in Part B study for replication, and the effect of placebo versus NSAID on nighttime activity were also assessed in Part B. To evaluate potential factors that were associated with nighttime activity, average activity per nighttime hour were calculated for each dog, and clinical and demographic variables were tested for association using appropriate tests of association. First, differences in activity between the baseline (nights 7--13) and NSAID treatment (nights 21--27) were tested using a repeated measures analysis of variance (ANOVA), with a term for time (by hour), and the baseline or treatment category. Based on negative results in this comparison, activity was averaged for each dog, and those averages were used in the subsequent analysis. Then, to test for differences according to demographic variables, repeated measures generalized linear models (linear regression) with appropriate degrees of freedom were used to test for associations with variables and activity levels. Finally, to test for multivariable associations, forward step-wise variable selection using a Bayesian Information Criteria (BIC) selection criterion was used within the repeated measure regression to build a multivariable model and to automatically perform variable selection. BIC was chosen as the selection criteria given its model consistency properties ([@ref-15]), and to control for concerns with over-fitting with the forward stepwise approach, a Bonferroni correction for the number of hypotheses tested was used for the multivariable analyses. The following variables were tested for univariate analysis and were entered as potential predictive variables in the stepwise modeling: Sex, Age, Weight, Predominantly Fore or Hind limbs affected, CSOM score at day 0, CBPI pain score at day 0, CBPI function score at day 0, CBPI total score at day 0, HCPI total score at day 0, Total pain score at day 0, and day of the week. To control for multiple comparisons in the regression modeling process, a Bonferroni correction was used to maintain a family-wise error rate of 0.05, and such that a *p*-values less than 0.05/10 = 0.005 were considered statistically significant (0.05/10 was used as only 10 variables were found to be significant in univariate analysis, and only significant variables were used in the secondary analysis). Because the overall goals of the study were exploratory, the univariate analyses were considered at both nominal levels (not corrected for multiple comparisons) and using a Bonferroni correction, though we recognize that more conservative controls could be considered. After variables were selected as significant in Part A study, data from Part B were used to replicate these associations (to help reduce any false positives from the multiple testing in Part A). Only significant variables were tested for association with activity using appropriate tests of hypothesis, as performed in Part A. Differences in activity between placebo and NSAID administration were tested using matched-pairs *t*-tests, and again, based on null results, the two activity scores for each hour per dog were averaged prior to subsequent analysis. Variables with *p*-values less than 0.05 were considered statistically significant. SNoRE total values were reasonably normally distributed, and total scores from day 0 (baseline), day 14 and day 35 were compared (baseline and NSAID; baseline and placebo; placebo and NSAID) using a paired *t*-test, with a critical *p*-value of 0.016 to adjust for multiple comparisons. Additionally, the same procedure was repeated for each of the 6 individual questions on the SNoRE. The correlation between SNoRE values and CBPI pain scores measured at the three time points was calculated. All analyses were performed in Stata v11 ([www.stata.com](www.stata.com)) and R software ([www.cran.org](www.cran.org)), and also JMP (JMP, SAS, Cary, NC, USA). Results ======= In Part A, dogs were a mean of 9.29 years old (SD ± 2.83); 30.0 kg (SD ± 10.4) in bodyweight, and consisted of 11 spayed females and 8 neutered males. The mean total CBPI score (addition of all questions) of dogs in Part A was 46 (SD ± 18). In Part B, dogs were a mean of 9.96 years old (SD ± 2.42); 32.0 kg (SD ± 5.6) in bodyweight, and consisted of 11 spayed females and 8 neutered males. The mean total CBPI score (addition of all questions) of dogs in Part B was 38 (SD ± 14). The degree of impairment (as measured by total CBPI scores) was not different between the groups (*P* = 0.31). In Part A of the study there was a nominal increase in nighttime activity but no significant difference between nighttime activity during the baseline period (mean 4,642; SD ± 2,052) and the treatment period (mean 5,414; SD ± 2,020) (*p* \< 0.5976, uncorrected), with baseline minus NSAID treatment mean hourly activity being −772 ± 2,931. There was also no significant association between treatment and activity in Part B (*p* \< 0.2645, uncorrected). Values for the change in activity (per hour) calculated for placebo minus NSAID treatment (mean 121; SD ± 963) show a slight, but non-significant decrease in hourly nighttime activity in Part B. The results of the variable selection in Part A indicated one variable (affected limb) that was statistically significant using a Bonferroni adjusted *p*-value, affected limb (*p* \< 0.001, uncorrected). This variable was tested for association in Part B, and was also statistically significant (*p* \< 0.001). In both studies, the average nighttime activity was higher for dogs with an affected forelimb than hind limb ([Fig. 1](#fig-1){ref-type="fig"}). However, using data from Part A, comparison of CBPI scores indicated that the forelimb OA dogs (*n* = 12) had significantly lower total CBPI scores (less impaired) than the hindlimb OA dogs (*n* = 48) (*p* = 0.014). Additionally, the univariate analyses indicated some nominal associations (not statistically significant after multiple testing corrections) in both parts of the study. In Part A, weight (*p* \< 0.0013), CBPI pain score (*p* \< 0.0301), and total CSOM score (*p* \< 0.0302) were nominally significant. In Part B, CBPI pain score (*p* \< 0.001) and CSOM score (*p* \< 0.0374) were nominally significant. Each of these variables was negatively correlated with activity, such that lower weight, lower CBPI pain score, and lower CSOM score were associated with higher activity. For weight, the correlation was *r* = −0.44 (moderate) in Part A. For CBPI, *r* = −0.24 (modest) in Part A and *r* = −0.42 (moderate) in Part B. For CSOM, the correlation was *r* = −0.41 (moderate) for Part A, and *r* = −0.38 (moderate) in Part B. {#fig-1} The SNoRE instrument detected a positive improvement due to the NSAID (*p* = 0.001), and detected a difference between the NSAID and placebo (*p* = 0.049). Questions 2 and 3 appeared to be the best at detecting the positive effects of the NSAID on the quality of sleep ([Table 1](#table-1){ref-type="table"}). There was a significant (*p* \< 0.001) moderate (*r* = 0.47) correlation between the SNoRE and the CBPI pain score. 10.7717/peerj.772/table-1 ###### Table of change in SNoRE scores. Mean change in scores between treatments and baseline for the SNoRE clinical metrology instrument. Negative changes indicate improvement in sleep. Critical *p*-value was adjusted to *p* = 0.016 within each question (or the total score) to reflect multiple comparisons within each question.  NSAID--Baseline Placebo--Baseline NSAID--Placebo ------------ ----------------- ------------------- ---------------- -------- ------- ------------ Question 1 −1.05 0.0286 −0.37 0.4209 −3.58 0.2435 Question 2 −1.21 **0.0149** −0.26 0.5433 −0.94 0.0462 Question 3 −1.37 **0.0030** −0.21 0.5311 −1.15 **0.0029** Question 4 −1.26 0.0175 −0.21 0.7255 −1.05 0.1036 Question 5 −0.89 **0.0072** −0.63 0.0688 −0.26 0.3835 Question 6 0.21 0.4647 0.42 0.3664 −0.21 0.7112 Total −6.00 **0.0012** −1.26 0.4227 −4.73 0.0496 Discussion ========== Overall, we did not find evidence of nighttime restlessness, as measured by accelerometry, in dogs with naturally occurring OA. The SNoRE clinical metrology instrument did detect a subjective improvement in sleep with the provision of analgesia, indicating that quality of sleep is disturbed in naturally occurring OA in dogs due to pain. The SNoRE deserves further evaluation in a larger cohort of dogs, as do other measures of sleep disturbance. One major criticism of this study is that we did not include an age-matched cohort of normal (non-OA) dogs to determine if nighttime activity differed between normal and OA dogs, and further work should evaluate this. Accelerometry has been widely used in human studies as an objective assessment of sleep disturbances ([@ref-24]; [@ref-33]). However, relatively little work has been done in human OA patients. One study found that accelerometry did not distinguish between sleep efficiency in older persons with and without chronic pain, but sleep diaries did ([@ref-16]). Interestingly, that study appears to reflect the results of our study where the quality of sleep was improved with NSAID treatment (pain alleviation), but accelerometry did not change. Another study suggested human patients with OA have higher levels of nocturnal body motility, but this was not proven ([@ref-13]). Another study, that included accelerometry as a measure of nighttime restlessness, showed no effect of yoga on sleep disturbance when accelerometry data was evaluated, despite the fact that participants reported significantly fewer nights with insomnia following an 8-week yoga program ([@ref-27]). However, the study was relatively small (13 participants; accelerometry data available in 11), and was not masked. Regardless, collectively, it appears that sleep may be disturbed without any resulting measurable increase in movement as measured by simple actigraphy. Simultaneous use of multiple accelerometers and integration of the data to measure sleep position may be more useful ([@ref-33]), but as yet untried in OA patients. Indeed, overall the assumption that sleep disturbance from OA-pain is associated with increased movement has not been proven. In a rodent model of OA (iodoacetate), sleep patterns, as measured electrophysiologically, were altered ([@ref-25]; [@ref-26]); in adjuvant-induced arthritis, fragmented sleep patterns occurred following the induction of arthritis ([@ref-10]); and in a model of gout arthritis, sleep-wake patterns were altered ([@ref-5]). However, none of these studies measured body movement or activity induced by arthritis in rodents. Further work is required to determine if accelerometry is an appropriate measure of sleep disturbance in any species. In our study, the level of pain may not have been severe enough to elicit sleep disturbance. Although the severity of impairment in the present study reflects dogs that are recruited to OA-pain studies, further work should be directed at determining if a particular level of owner-noted impairment, or other phenotypic characteristics appear to be associated with the owner-noted sleep impairment. However, in a recent focus group study ([@ref-32]), night pain was found to occur amongst people with varying levels of current pain severity ([@ref-32]). Additionally, estimates of sleep disturbance in human OA patients suggest between 30 and 80% of OA patients suffer sleep disturbances ([@ref-20]; [@ref-31]; [@ref-22]; [@ref-32]; [@ref-28]), indicating a broad cross-section of the OA population can be affected. Despite 30 to 80% of OA human patients suffering sleep disturbances ([@ref-20]; [@ref-22]; [@ref-31]; [@ref-32]; [@ref-28]), a recent study showed no disturbances in slow-wave sleep in a rodent model of bilateral osteoarthritic pain ([@ref-14]), suggesting this naturally occurring model in dogs should be further evaluated for translational research utility. Our results showed that dogs that were classified as having predominately forelimb impairment had significantly higher activity counts when compared to dogs with predominately hind limb impairment; however, the forelimb OA dogs were less impaired (as determined by owner scores) than the hindlimb OA dogs. Future studies should stratify for fore/hind limb involvement, or focus on one or the other phenotype. The subjective clinical metrology instrument showed improvement in the quality of sleep related to the NSAID, compared to baseline, and over placebo. There was no difference between placebo and baseline. Despite this robust responsiveness validity, however, with no change in accelerometry the criterion validity of the instrument cannot be inferred. The improvement in scores following NSAID treatment in this masked, placebo-controlled study, indicated responsiveness validity, suggest that this subjective instrument should be further tested and evaluated. It is unclear what aspects of sleep detected by the instrument were altered by the NSAID but, looking at the questions that appeared to be most responsive, we get some indication. The most responsive questions appeared to be questions asked about twitching and dreaming, followed by those asking about shifting position and vocalizing. The least-responsive questions appeared to be those asking about moving and pacing. When the responsiveness of the individual questions are viewed like this, the lack of effects on accelerometry-measured activity seem more inline with the SNoRE results. Further research could use objective video ([@ref-6]; [@ref-12]) and acoustic analysis of sleeping behavior to assess the effects of analgesic intervention. This would help further characterize the OA-dog as a model of human sleep disturbance, and would help understand if the improvements seen in the SNoRE reflected real changes in sleep quality. It was reassuring that the correlation between the SNoRE and the CBPI pain score (a validated measure of pain and pain relief in dogs with naturally occurring OA ([@ref-2]; [@ref-29]) was positive, significant and moderate. This suggests the SNoRE is measuring a similar construct, but likely different aspects. Future research should evaluate test-retest stability (reliability) of the SNoRE, and also evaluate criterion validity through the use of videography. There are excellent induced large animal models of OA in dogs and horses ([@ref-1]), and rodent models of chronic pain for the evaluation of sleep disturbance ([@ref-14]), and these have great utility. However, as previously suggested ([@ref-19]) the use of such spontaneous disease preclinical models, where pain was not artificially induced and has a time course more similar to the human condition, and particularly where pain is spontaneous, could also be helpful in preclinical testing of analgesics and understanding the underlying mechanisms. Supplemental Information ======================== 10.7717/peerj.772/supp-1 ###### SNoRE questionnaire Sleep and Night Time Restlessness Evaluation Score (SNoRE) clinical metrology instrument used in Part B of this study. ###### Click here for additional data file. 10.7717/peerj.772/supp-2 ###### Study A demographic and subjective data ###### Click here for additional data file. 10.7717/peerj.772/supp-3 ###### Study A activity data dog 01 ###### Click here for additional data file. 10.7717/peerj.772/supp-4 ###### Study A activity data dog 02 ###### Click here for additional data file. 10.7717/peerj.772/supp-5 ###### Study A activity data dog 03 ###### Click here for additional data file. 10.7717/peerj.772/supp-6 ###### Study A activity data dog 04 ###### Click here for additional data file. 10.7717/peerj.772/supp-7 ###### Study A activity data dog 05 ###### Click here for additional data file. 10.7717/peerj.772/supp-8 ###### Study A activity data dog 06 ###### Click here for additional data file. 10.7717/peerj.772/supp-9 ###### Study A activity data dog 07 ###### Click here for additional data file. 10.7717/peerj.772/supp-10 ###### Study A activity data dog 08 ###### Click here for additional data file. 10.7717/peerj.772/supp-11 ###### Study A activity data dog 09 ###### Click here for additional data file. 10.7717/peerj.772/supp-12 ###### Study A activity data dog 10 ###### Click here for additional data file. 10.7717/peerj.772/supp-13 ###### Study A activity data dog 11 ###### Click here for additional data file. 10.7717/peerj.772/supp-14 ###### Study A activity data dog 12 ###### Click here for additional data file. 10.7717/peerj.772/supp-15 ###### Study A activity data dog 13 ###### Click here for additional data file. 10.7717/peerj.772/supp-16 ###### Study A activity data dog 14 ###### Click here for additional data file. 10.7717/peerj.772/supp-17 ###### Study A activity data dog 15 ###### Click here for additional data file. 10.7717/peerj.772/supp-18 ###### Study A activity data dog 16 ###### Click here for additional data file. 10.7717/peerj.772/supp-19 ###### Study A activity data dog 17 ###### Click here for additional data file. 10.7717/peerj.772/supp-20 ###### Study A activity data dog 19 ###### Click here for additional data file. 10.7717/peerj.772/supp-21 ###### Study A activity data dog 20 ###### Click here for additional data file. 10.7717/peerj.772/supp-22 ###### Study A activity data dog 22 ###### Click here for additional data file. 10.7717/peerj.772/supp-23 ###### Study A activity data dog 23 ###### Click here for additional data file. 10.7717/peerj.772/supp-24 ###### Study A activity data dog 24 ###### Click here for additional data file. 10.7717/peerj.772/supp-25 ###### Study A activity data dog 25 ###### Click here for additional data file. 10.7717/peerj.772/supp-26 ###### Study A activity data dog 26 ###### Click here for additional data file. 10.7717/peerj.772/supp-27 ###### Study A activity data dog 27 ###### Click here for additional data file. 10.7717/peerj.772/supp-28 ###### Study A activity data dog 28 ###### Click here for additional data file. 10.7717/peerj.772/supp-29 ###### Study A activity data dog 29 ###### Click here for additional data file. 10.7717/peerj.772/supp-30 ###### Study A activity data dog 30 ###### Click here for additional data file. 10.7717/peerj.772/supp-31 ###### Study A activity data dog 31 ###### Click here for additional data file. 10.7717/peerj.772/supp-32 ###### Study A activity data dog 32 ###### Click here for additional data file. 10.7717/peerj.772/supp-33 ###### Study A activity data dog 33 ###### Click here for additional data file. 10.7717/peerj.772/supp-34 ###### Study A activity data dog 34 ###### Click here for additional data file. 10.7717/peerj.772/supp-35 ###### Study A activity data dog 35 ###### Click here for additional data file. 10.7717/peerj.772/supp-36 ###### Study A activity data dog 36 ###### Click here for additional data file. 10.7717/peerj.772/supp-37 ###### Study A activity data dog 37 ###### Click here for additional data file. 10.7717/peerj.772/supp-38 ###### Study A activity data dog 38 ###### Click here for additional data file. 10.7717/peerj.772/supp-39 ###### Study A activity data dog 39 ###### Click here for additional data file. 10.7717/peerj.772/supp-40 ###### Study A activity data dog 40 ###### Click here for additional data file. 10.7717/peerj.772/supp-41 ###### Study A activity data dog 41 ###### Click here for additional data file. 10.7717/peerj.772/supp-42 ###### Study A activity data dog 42 ###### Click here for additional data file. 10.7717/peerj.772/supp-43 ###### Study A activity data dog 43 ###### Click here for additional data file. 10.7717/peerj.772/supp-44 ###### Study A activity data dog 44 ###### Click here for additional data file. 10.7717/peerj.772/supp-45 ###### Study A activity data dog 45 ###### Click here for additional data file. 10.7717/peerj.772/supp-46 ###### Study A activity data dog 46 ###### Click here for additional data file. 10.7717/peerj.772/supp-47 ###### Study A activity data dog 48 ###### Click here for additional data file. 10.7717/peerj.772/supp-48 ###### Study A activity data dog 49 ###### Click here for additional data file. 10.7717/peerj.772/supp-49 ###### Study A activity data dog 50 ###### Click here for additional data file. 10.7717/peerj.772/supp-50 ###### Study A activity data dog 51 ###### Click here for additional data file. 10.7717/peerj.772/supp-51 ###### Study A activity data dog 52 ###### Click here for additional data file. 10.7717/peerj.772/supp-52 ###### Study A activity data dog 53 ###### Click here for additional data file. 10.7717/peerj.772/supp-53 ###### Study A activity data dog 54 ###### Click here for additional data file. 10.7717/peerj.772/supp-54 ###### Study A activity data dog 55 ###### Click here for additional data file. 10.7717/peerj.772/supp-55 ###### Study A activity data dog 56 ###### Click here for additional data file. 10.7717/peerj.772/supp-56 ###### Study A activity data dog 58 ###### Click here for additional data file. 10.7717/peerj.772/supp-57 ###### Study A activity data dog 60 ###### Click here for additional data file. 10.7717/peerj.772/supp-58 ###### Study A activity data dog 62 ###### Click here for additional data file. 10.7717/peerj.772/supp-59 ###### Study A activity data dog 63 ###### Click here for additional data file. 10.7717/peerj.772/supp-60 ###### Study A activity data dog 64 ###### Click here for additional data file. 10.7717/peerj.772/supp-61 ###### Study A activity data dog 65 ###### Click here for additional data file. 10.7717/peerj.772/supp-62 ###### Study B subjective data ###### Click here for additional data file. 10.7717/peerj.772/supp-63 ###### Study B activity data dog 01 ###### Click here for additional data file. 10.7717/peerj.772/supp-64 ###### Study B activity data dog 02 ###### Click here for additional data file. 10.7717/peerj.772/supp-65 ###### Study B activity data dog 03 ###### Click here for additional data file. 10.7717/peerj.772/supp-66 ###### Study B activity data dog 04 ###### Click here for additional data file. 10.7717/peerj.772/supp-67 ###### Study B activity data dog 05 ###### Click here for additional data file. 10.7717/peerj.772/supp-68 ###### Study B activity data dog 06 ###### Click here for additional data file. 10.7717/peerj.772/supp-69 ###### Study B activity data dog 07 ###### Click here for additional data file. 10.7717/peerj.772/supp-70 ###### Study B activity data dog 09 ###### Click here for additional data file. 10.7717/peerj.772/supp-71 ###### Study B activity data dog 12 ###### Click here for additional data file. 10.7717/peerj.772/supp-72 ###### Study B activity data dog 14 ###### Click here for additional data file. 10.7717/peerj.772/supp-73 ###### Study B activity data dog 15 ###### Click here for additional data file. 10.7717/peerj.772/supp-74 ###### Study B activity data dog 16 ###### Click here for additional data file. 10.7717/peerj.772/supp-75 ###### Study B activity data dog 18 ###### Click here for additional data file. 10.7717/peerj.772/supp-76 ###### Study B activity data dog 19 ###### Click here for additional data file. 10.7717/peerj.772/supp-77 ###### Study B activity data dog 20 ###### Click here for additional data file. The authors would like to acknowledge the assistance of Dr. Ben Wernham, Dr. Brian Trumpatori and Mr. Jon Hash in the collection of data, and Tonya Lee for manuscript editing assistance. Additional Information and Declarations ======================================= BDXL has received honoraria for continuing education presentations from Boehringer Ingelheim, and has received research support from Boehringer Ingelheim. David Knazovicky, Andrea Tomas and Alison Motsinger-Reif report no conflicts of interest. B Duncan X. Lascelles is an Academic Editor for PeerJ. [David Knazovicky](#author-1){ref-type="contrib"} and [Andrea Tomas](#author-2){ref-type="contrib"} performed the experiments, wrote the paper, reviewed drafts of the paper. [Alison Motsinger-Reif](#author-3){ref-type="contrib"} analyzed the data, contributed reagents/materials/analysis tools, wrote the paper, prepared figures and/or tables, reviewed drafts of the paper. [B. Duncan X. Lascelles](#author-4){ref-type="contrib"} conceived and designed the experiments, performed the experiments, analyzed the data, contributed reagents/materials/analysis tools, wrote the paper, reviewed drafts of the paper. The following information was supplied relating to ethical approvals (i.e., approving body and any reference numbers): The Institutional Animal Care and Use Committee (IACUC) approved both studies (IACUC \#08-077-O, \#07-188-O). [^1]: Canon Medical CXDI-50G Sensor, Eklin Medical Systems, Santa Clara, CA. [^2]: Actical Activity Monitor, Philips Respironics Co, Bend, OR. [^3]: Methylcellulose (Ora Plus^®^) opacifier, and coloring (McCormick Food Colors Yellow and Blue).

1. Introduction {#sec1} =============== Rheumatoid arthritis (RA) is a common, chronic, systemic, and inflammatory autoimmune disorder that primarily affects the small diarthrodial joints of the hands and feet, affecting approximately 1% of the world\'s population \[[@B1]--[@B3]\]. The main characteristics of this disease are synovium hyperplasia, lymphocyte infiltration, and the abnormal proliferation of fibroblast-like synoviocytes (FLS) that can lead to the destruction of bone and cartilage and eventual disability \[[@B4]\]. Abnormal T cell immunity plays a critical role in the development of RA. Numerous factors that are involved in alternative T cell activation have been characterized, including the activation of inflammatory cells and expression of various cytokines. Inflammatory mediators, such as interleukin-6 (IL-6), interleukin-1 (IL-1), and tumor necrosis factor-*α* (TNF-*α*), are abundant in synovial tissues and fluid from patients with RA, and the overexpression of these cytokines promotes chronic inflammation and joint destruction. Many of the inflammatory mediators involved in the pathology of RA are regulated by nuclear factor kappa B (NF-*κ*B) transcription factors \[[@B5], [@B6]\]. Abnormal NF-*κ*B activation occurs during many pathological conditions including allergic and autoinflammatory diseases and malignancies \[[@B7]\]. Recently, A20, a negative regulator of NF-*κ*B, was identified as a key regulator for inflammation signaling and may be involved in RA pathogenesis \[[@B8]\]. A20 has been reported to be ubiquitin-editing enzyme with several functions. A20 is also known as tumor necrosis factor-*α*- (TNF-*α*-) induced protein 3 (TNFAIP3), which was first discovered in 1990 by Dixit and colleagues as a cytokine-induced gene in human umbilical vein endothelial cells \[[@B9], [@B10]\]. Subsequent studies demonstrated that A20 overexpression inhibits NF-*κ*B activation in response to different stimuli \[[@B11]--[@B13]\]. The cloning and characterization of the A20 promoter revealed two NF-*κ*B DNA binding elements, which are recognition sequences for NF-*κ*B transcription factors. It was also found that multiple NF-*κ*B activating stimuli induce A20 expression via NF-*κ*B sites in the A20 promoter \[[@B14]\]. Therefore, A20 has been demonstrated to downregulate its own expression, and it has been proposed that A20 participates in a negative feedback loop to attenuate TNF-*α*-induced inflammatory responses. A20 overexpression was subsequently demonstrated to block the NF-*κ*B activation mediated by TNF-*α*, IL-1, LPS, phorbol esters, and hydrogen peroxide in different cell types \[[@B11], [@B12], [@B15]--[@B18]\]. This inhibition is most likely due to the inhibition of NF-*κ*B activation in endothelial cells in response to proinflammatory stimuli, and an antiproliferative effect on smooth muscle cells has been observed upon A20 overexpression*in vitro*. All of these findings suggest that A20 attenuates the activity of proximal signaling complexes at proinflammatory receptors \[[@B19]--[@B21]\]. A20 is regulated by the CARMA1-Bcl-10-MALT1 (mucosa-associated lymphoid tissue lymphoma translocation gene 1) upstream signaling pathway complex, which bridges T cell antigen receptor (TCR) signaling with the canonical I*κ*B kinase (IKK)/NF-*κ*B pathway \[[@B20], [@B22]--[@B25]\]. TCR stimulation induces the recruitment of A20 and the Bcl-10 adaptor protein into the MALT1 complex, leading to MALT1-mediated A20 processing. Similarly, API2-MALT1 expression results in A20 cleavage. MALT1 cleaves A20 at arginine 439 and impairs its NF-*κ*B inhibitory function. Therefore, A20 was identified as a MALT1 substrate, emphasizing the importance of the MALT1 proteolytic activity in "fine-tuning" T cell antigen receptor signaling \[[@B26]\]. A20 dysfunction by deletion or mutation was identified in numerous lymphocytic malignancies \[[@B27]\]. Recently, polymorphisms in the A20 region were reported in autoimmune diseases such as systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), Crohn\'s disease, and psoriasis. Single nucleotide polymorphisms in the A20 region, including rs13192841, rs2230926, and rs6922466, have been independently associated with increased susceptibility for SLE \[[@B28], [@B29]\], and this finding provides a critical link between A20 and the etiology of SLE. More recently, it was shown that A20 deficiency in myeloid cells triggers erosive polyarthritis, resembling RA in a myeloid-specific, A20-deficient mice model \[[@B30]\]. There are three strongly associated genetic variants, including rs6920220, rs6927127, and rs6933404, which result in an A20 functional decrease in RA \[[@B8]\]. The etiology of RA remains to be understood; however, A20 deficiency may be a pivotal regulator for inflammation in RA. To characterize the role of A20 in RA, we analyzed the expression level of A20, NF-*κ*B, and the A20 regulatory factor MALT1 in samples from Chinese patients with RA in this study. 2. Materials and Methods {#sec2} ======================== 2.1. Samples {#sec2.1} ------------ This study included 16 patients with untreated RA (age: 26--71 years) and 20 healthy individuals (age: 22--70 years) who served as controls. The diagnosis of RA was based on the American College of Rheumatology criteria and expert opinion (1987 ACR criteria). All patients with RA were assessed for clinical disease activity by a trained rheumatologist using the disease activity score (DAS) \[[@B28]\]. The most recent erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), and rheumatoid factor (RF) were collected ([Table 1](#tab1){ref-type="table"}) \[[@B31]\]. All of the procedures were conducted according to the guidelines of the Medical Ethics Committee of the Health Bureau of Guangdong Province of China. Peripheral blood samples were collected by heparin anticoagulation, and peripheral blood mononuclear cells (PBMCs) were isolated using the Ficoll-Hypaque gradient centrifugation method. RNA extraction and cDNA synthesis were performed according to the manufacturer\'s instructions. 2.2. Quantitative Real-Time RT-PCR (qRT-PCR) {#sec2.2} -------------------------------------------- The sequences of the primers for MALT1, A20, and NF-*κ*B gene amplification are listed in [Table 2](#tab2){ref-type="table"}. According to the structure of the MALT1 gene, there are two variants, that is, MALT1-V1 and MALT1-V2, and the latter contains a 33 bp deletion located between exons 6 and 8. To amplify the two MALT1 transcript variants, the MALT-V1-for and MALT-V1-rev primer pair was designed for MALT1-V1 amplification to cover the region that is missing in MALT1-V2, and the MALT1-for and MALT1-rev primer pair was designed to amplify the conserved region, which is contained by both variants. The expression level of the A20, MALT1, MALT1-V1, NF-*κ*B, and *β*2-microglobulin (*β*2M) genes was determined by SYBR Green I real-time PCR. Briefly, PCR in a 20 *μ*L total volume was performed with approximately 1 *μ*L of cDNA, 0.5 *μ*M of each primer pair, 9 *μ*L of 2.5 × Real Master Mix (Tiangen Biotech (Beijing) Co., Ltd., Beijing, China), and 9 *μ*L of dH~2~O. After initial denaturation at 95°C for 15 minutes, 45 cycles of the following procedure were performed: 30 seconds at 95°C and 40 seconds at 60°C for the *β*2M, MALT1-V1, MALT1, A20, and NF-*κ*B genes. The plate was read immediately after the 60°C step using an MJ Research DNA Engine Opticon 2 PCR cycler (Bio-Rad, Hercules, CA, USA) \[[@B32]\]. The relative amount of the genes of interest and *β*2M reference gene was measured in two independent assays. The specific, amplified PCR products were analyzed by melting curve analysis. The data are presented as the relative expression of the genes of interest compared with the internal control gene as determined by the 2(^−ΔCT^) method \[[@B33]\]. In addition, to analyze the MALT1-V1 expression characteristics, we calculated the MALT1-V1 expression ratio as MALT1-V1/MALT1 × 100%. 2.3. Statistical Analysis {#sec2.3} ------------------------- Two independent-samples Wilcoxon tests were performed to compare the median expression level for each gene between patients with RA and the control group. Pearson correlation and linear regression analyses were used to determine the association between different genes in different groups. A *P* \< 0.05 was considered statistically significant \[[@B34]\]. 3. Results and Discussions {#sec3} ========================== Abnormal T cell activation is a common feature of RA \[[@B35]\], and the upregulation of some positive regulating factors such as TNF-*α*, IL-6, and IL-2 was identified during the initiation of RA. In contrast, the downregulation of negative regulatory factors has the same effect on initiation of RA \[[@B16]\]. For example, T cell activation leads to the downregulation of A20 expression in mature thymocytes and peripheral T cells \[[@B9], [@B10]\]. Recently, abnormal A20 expression was described in patients with RA \[[@B8]\], and decreased A20 results in increased NF-*κ*B expression and enhanced inflammation \[[@B25]\]. In this study, we analyzed the expression of A20 in 16 patients with RA in the active phase, and a significantly lower expression of A20 (median: 6.530) was found compared with those in the healthy group (median: 44.614, *P* \< 0.001) ([Figure 1(a)](#fig1){ref-type="fig"}). These results are similar to findings of different reports examining mouse models or patients with RA \[[@B8], [@B36]\]. A20-deficient mice develop severe multiorgan inflammation. Moreover, it is well accepted that the activation of NF-*κ*B-dependent gene expression plays a key role in the development of RA \[[@B30]\]; thus, decreased A20 expression may be a key reason for NF-*κ*B overexpression in RA. Our study also demonstrated that NF-*κ*B overexpression could be detected in patients with RA (median: 0.798) in comparison with healthy controls (median: 0.605, *P* = 0.042) ([Figure 1(b)](#fig1){ref-type="fig"}), indicating that decreased A20 resulting in NF-*κ*B overexpression is also a common feature for Chinese patients with RA. Lower A20 expression is associated with polymorphisms in the A20 genomic locus \[[@B29]\]; however, whether there is any dysregulation in A20 by upstream pathway factors is unknown. MALT1 is an upstream A20 pathway factor that cleaves A20 at arginine 439 and impairs its NF-*κ*B inhibitory function. To characterize the relationship between MALT1 and A20 expression, we also examined the expression level of MALT1. Interestingly, the MALT1 expression level is downregulated in patients with RA (median: 0.541) compared with those in the healthy group (median: 1.638, *P* \< 0.001) ([Figure 2(a)](#fig2){ref-type="fig"}). This result appears to be inconsistent with the lower A20 and higher NF-*κ*B expression level results because MALT1 is also a positive regulatory factor of NF-*κ*B. It is known that there are two MALT1 variants, MALT1-V1 and MATL1-V2, according to data in GenBank, and our previous study has found that these MALT1 variants could be identified by RT-PCR and sequencing (data not shown). In addition, we analyzed the expression level of the different variants, and similar results were found including the fact that a significantly lower MALT1-V1 expression level was detected in patients with RA (median: 0.062) compared with healthy controls (median: 0.140, *P* \< 0.001) ([Figure 2(b)](#fig2){ref-type="fig"}). Because we could not directly amplify MALT1-V2, which contains a 33 bp deletion, the expression level of MALT1-V2 could only be indirectly calculated by the relative expression of MALT1-V1/total MALT1 \[[@B37]\], and there are no significant differences in the ratio of MALT1-V1/total MALT1 between patients with RA and healthy controls (13.43 ± 7.98% versus 11.76 ± 6.66%), implying that the MALT1-V2 expression level was also downregulated in RA. Overall, either MALI1-V1 or MALT1-V2 was decreased in RA, unlike the finding in T cells from acute myeloid leukemia (AML), in which we found that the MALT1-V1 expression level was significantly higher in T cells from AML patients compared with healthy controls, while the MALT1-V2 expression level was downregulated \[[@B37]\]; this may indicate different expression pattern of these two MALT1 variants in RA. Little is known about the functional difference between the variants, and whether this is a feedback response from the expression pattern of A20 and NF-*κ*B in RA is unknown because the impact of A20 cleavage by MALT1 on its capacity to regulate NF-*κ*B has only been partially elucidated \[[@B24]\]. Further investigation is needed to characterize the upstream pathway regulators of A20 in addition to MALT1. The role of MALT1 in the development of inflammation is largely unknown for RA and other autoimmune diseases, and only one study has reported that patients who had MALT-type lymphomas may also suffer from rheumatoid arthritis due to MALT1 dysfunction and continuous NF-*κ*B activation \[[@B38]\]. A study has reported that MALT1 rearrangement in gastric MALT lymphoma is frequently associated with Sjögren\'s syndrome \[[@B39]\]. More recently, Brüstle A and coworkers indicated that MALT1 is a central cell intrinsic factor that determines the experimental autoimmune encephalitogenic potential of inflammatory Th17 cells*in vivo* \[[@B40]\]. It appears that MALT1 may play a role in the development of autoimmune diseases, and it may interfere with specific T cell subsets. Thus, our finding of lower MALT1 expression may imply a loss of the control of T cell activation in some T cell subsets in RA, which remains an open question. Further studies are needed to investigate the pathways upstream of MALT1 and TCR-CD3 signaling in different T cell subsets in RA. In general, A20 is cleaved by MALT1; thus, the expression level of both genes should be negatively correlated with the expression pattern of A20 and MALT1 \[[@B24]\]. However, we found a positive correlation between MALT1 and A20 (*r* = 0.520, *P* = 0.039) ([Figure 3(a)](#fig3){ref-type="fig"}) and MALT1-V1 and A20 (*r* = 0.511, *P* = 0.043) ([Figure 3(b)](#fig3){ref-type="fig"}) in RA, and a tendency towards a negative correlation was found between MALT1 and NF-*κ*B (*r* = −0.098, *P* = 0.718), MALT1-V1 and NF-*κ*B (*r* = −0.204, *P* = 0.448), and A20 and NF-*κ*B (*r* = −0.264, *P* = 0.322), indicating that the MALT1-A20-NF-*κ*B expression pattern may be more complex in RA. In conclusion, we characterized, for the first time, the alternative expression pattern of MALT1, A20, and NF-*κ*B in RA, which may be related to abnormal T cell activation. Lacking A20 and MALT1 dysfunction are common characteristics of Chinese patients with RA, and our results provide new inflammation targets to consider for RA treatment. Moreover, further investigation is needed to follow up on patients with different MALT1-A20-NF-*κ*B expression patterns and their association with cancer development. This study was supported by grants from the National Natural Science Foundation of China (no. 91129720) and a Collaborated Grant for HK-Macao-TW of the Ministry of Science and Technology (2012DFH30060). Conflict of Interests ===================== The authors declare that there is no conflict of interests regarding the publication of this paper. Authors\' Contribution ====================== Yangqiu Li contributed to the concept development and study design. Xu Wang, Lihua Zhu, Ziwei Liao, and Fan Zhang performed the real-time PCR. Ling Xu, Yan Xu, Shaohua Chen, and Lijian Yang performed the PBMC isolation, RNA extraction, and cDNA synthesis. Lihua Zhu and Yi Zhou were responsible for the collection of clinical data. Yangqiu Li, Xu Wang, Lihua Zhu, and Ziwei Liao coordinated the study and helped draft the paper. All authors read and approved the final paper. Xu Wang, Lihua Zhu, and Ziwei Liao contributed equally to this paper. {#fig1} {#fig2} {#fig3} ###### Characteristics of RA samples. Patient number Gender Age Disease duration (mo) RF (IU/mL) ESR (mm/h) CRP (mg/L) DAS28 scores CCP status ---------------- -------- ----- ----------------------- ------------ ------------ ------------ -------------- ------------ 1 F 17 48 252.00 73 38.90 6.00 \+ 2 F 56 24 1940.00 111 7.40 6.78 \+ 3 F 51 2 17.50 77 37.70 7.09 \+ 4 F 43 60 74.30 37 5.34 5.53 \+ 5 F 62 72 38.30 69 17.00 6.88 \+ 6 F 32 120 19.90 85 9.06 7.51 ND 7 F 60 6 150.00 90 103.00 6.09 ND 8 F 26 9 31.00 32 1.84 4.31 \+ 9 F 53 6 368.00 89 31.90 7.41 − 10 F 54 9 9.19 64 33.25 7.39 − 11 F 45 12 102.00 82 68.98 6.26 \+ 12 F 53 240 65.30 76 27.08 7.91 \+ 13 F 71 12 58.90 110 71.35 6.79 \+ 14 F 63 12 153.00 32 4.84 5.23 \+ 15 F 33 84 299.00 42 18.70 5.61 \+ 16 F 60 6 10.10 41 0.57 6.63 − Note: mo: months; F: female; +: positive; −: negative; ND: no detection. ###### List of primer information. Primer Sequences Accession number PCR products -------------- ------------------------------ ------------------ -------------- A20-for 5′-CTGGGACCATGGCACAACTC-3′ NM_006290 182 bp A20-rev 5′-CGGAAGGTTCCATGGGATTC-3′ MALT1-V1-for 5′-AAGCCCTATTCCTCACTACCAG-3′ NM_006785.2 195 bp MALT1-V1-rev 5′-CACTCCACTGCCTCATCTGTTC-3′ MALT1-for 5′-TCTTGGCTGGACAGTTTGTGA-3′ NM_006785.2 230 bp MALT1-rev 5′-GCTCTCTGGGATGTCGCAA-3′ NF-*κ*B-for 5′-CCACAAGACAGAAGCTGAAG-3′ NM_003998 149 bp NF-*κ*B-rev 5′-AGATACTATCTGTAAGTGAACC-3′ *β*2M-for 5′-TACACTGAATTCACCCCCAC-3′ J00105 145 bp *β*2M-rev 5′-CATCCAATCCAAATGCGGCA-3′ [^1]: Academic Editor: Qifa Liu

{.183} {.184}

Introduction ============ The cerebellum controls motor coordination, cognitive function, and emotion. Its unique and well organized neuro-architecture is comprised of numerous layers each containing specific neuron-types that process incoming excitatory information from the brain stem and spinal cord. Mossy fibers, one of the two major excitatory inputs into the cerebellar cortex, synapse onto cerebellar granule cells (GCs) whose axons form parallel fibers that transfer information to Purkinje cells, the main output of the cerebellar cortex. GCs function as the gateway for information into the cerebellar cortex as their excitability is controlled by activation of synaptic GABA~A~ receptors (GABA~A~Rs) via phasic GABA release from a specialized interneuron, the Golgi cell. In addition, spillover of GABA released from Golgi cells tonically activates extrasynaptic GABA~A~Rs that help regulate the excitability of GCs. In cerebellar GCs, tonic currents are mediated by extrasynaptic GABA~A~Rs containing α6βδ subunits. These extrasynaptic GABA~A~Rs have a have high affinity for GABA, and are easily activated by ambient levels of neurotransmitter (Brickley et al., [@B8]; Hamann et al., [@B14]) as glial GABA transporters are unable to fully eliminate GABA from Golgi cell-to-GC synapses (Attwell et al., [@B3]). In addition, a glomerulus encases these synapses and slows diffusion, thus facilitating the activation of extrasynaptic receptors by GABA (Wall and Usowicz, [@B26]; Rossi et al., [@B23]). Numerous studies have characterized tonic currents in the cerebellum and have shown that these are sensitive to GABA~A~ antagonists (Brickley et al., [@B7]; Carta et al., [@B10]; Bright et al., [@B9]). They are also mediated in part by action potential-dependent mechanisms, as tetrodotoxin (TTX) can significantly reduce both synaptic GABA events driven by spontaneous firing of Golgi cells and the tonic current mediated by extrasynaptic GABA~A~Rs (Kaneda et al., [@B16]; Brickley et al., [@B7]; Wall and Usowicz, [@B26]; Carta et al., [@B10]). Studies examining the effects of acute ethanol on tonic currents in GCs have revealed that ethanol potentiates these currents, at least in part, by increasing Golgi cell firing via slight inhibition of the Na^+^/K^+^ ATPase (Carta et al., [@B10]; Botta et al., [@B4]). The facilitation of the tonic current by ethanol was abolished by TTX (Carta et al., [@B10]), consistent with the hypothesis that the effects of ethanol on the tonic current are predominantly mediated through changes in Golgi cell excitability. It has also been proposed that ethanol increases the tonic current via direct potentiation of extrasynaptic GABA~A~Rs (Hanchar et al., [@B15]), but these findings are controversial (Botta et al., [@B5],[@B6]). Alternative mechanisms responsible for the actions of ethanol on tonic GABAergic currents in GCs have yet to be investigated. Bestrophin channels have recently been suggested as a mechanism providing the predominant source of GABA for tonic currents in GCs. These are Ca^2+^-activated anion-channels that have been linked to human eye diseases \[reviewed in (Marmorstein et al., [@B19])\], but have also been shown to exist in the brain (Petrukhin et al., [@B21]). Specifically, bestrophin1 (Best1) channels are abundantly expressed in hippocampal astrocytes where they conduct anions and glutamate (Park et al., [@B20]). A recent study reported that cerebellar astrocytes and Bergmann glia express Best1 channels that were found to be permeable to GABA (Lee et al., [@B18]). Importantly, this study showed that GABA release from Best1 contributes up to 50--75% of the tonic current in GCs as three different Cl^−^-channel blockers, 5-nitro-2-(3-phenylpropylamino)benzoic acid (NPPB) with the largest effect, 4′-diisothiocyanatostilbene-2,2′-disulfonic acid disodium salt hydrate (DIDS), and niflumic acid (NFA) significantly attenuated tonic GABAergic currents by blocking Best1. In addition, shRNA-mediated down regulation of Best1 eliminated the effect of NPPB on the tonic current. Lee et al. ([@B18]) also demonstrated that these agents blocked Cl^−^ and GABA conductances in recombinant Best1 channels, but did not act on recombinant GABA receptors, further suggesting that the observed inhibition of the tonic current by these antagonists was solely mediated by blockade of GABA release through glial Best1 channels. Based on these findings, it can be concluded that extrasynaptic GABA~A~Rs in GCs are activated by three pools of GABA: (1) GABA released by glial cells via Best1; (2) spillover of GABA synaptically released from Golgi cells; and (3) ambient GABA levels. As mentioned above, we previously showed that potentiation of tonic currents by ethanol in GCs is not observed under conditions where the Golgi cell-dependent component of the tonic current is blocked (Carta et al., [@B10]). According to the findings of Lee et al. ([@B18]) the majority of the Golgi cell-independent component of the tonic current is mediated by Best1 channel-mediated GABA release from glia cells. We therefore hypothesized that the Best1-dependent component of the tonic current is insensitive to acute ethanol exposure. We tested this hypothesis using patch-clamp electrophysiological techniques on human recombinant Best1 channels expressed in human embryonic kidney (HEK)-293 cells and acute cerebellar slices from rodents. Materials and Methods ===================== Studies with recombinant Best1 ------------------------------ Unless indicated, all chemicals were purchased from Sigma Chemical Company (St. Louis, MO, USA). The wild-type human Best1 cDNA was kindly provided by Dr. J. Nathans (Baltimore, MD, USA). Site-directed mutagenesis was performed using the Quik-Change mutagenesis kit (Invitrogen, Carlsbad, CA, USA) and mutants were confirmed by DNA sequencing. HEK-293 cells were obtained from American Type Culture Collection (Manassas, VA, USA) and were maintained in serum-supplemented Dulbecco's Modified Eagle Media in a humidified incubator supplied with 5% CO~2~ (Xu and Woodward, [@B27]). For recordings, cells were plated onto poly ornithine coated 35 mm dishes and transfected with plasmids encoding Best1, Best1 (W93C), or enhanced green fluorescent protein (eGFP) using Lipofectamine 2000 (Invitrogen Inc., Carlsbad, CA, USA) according to the manufacturer's recommendation. The Best1 cDNA contained an Internal Ribosome Entry Site--eGFP domain that allowed for detection of transfected cells. Following transfection, cells were maintained in the incubator for 24--72 h prior to use. Dishes containing transfected cells were mounted on the stage of an Olympus IX50 inverted microscope and perfused with extracellular recording solution at 1--2 ml/min. The recording solution contained (in mM): 135 NaCl, 5.4 KCl, 1.8 CaCl~2~, 5 HEPES, 10 glucose (pH adjusted to 7.4 and osmolarity adjusted to 310--325 mOsm with sucrose). Patch pipettes (2--5 MΩ) were pulled from borosilicate glass (1.5 mm × 0.86 mm) and filled with internal solution containing (in mM): 133 CsCl, 4 MgCl~2~, 3.5 CaCl~2~, 5 EGTA, 10 HEPES, 3.1 Na-ATP, and 0.42 Na-GTP (pH adjusted to 7.1 with CsOH). Transfected cells were identified by eGFP fluorescence and whole-cell voltage-clamp recordings were carried out at room temperature using an Axopatch 200B amplifier (Molecular Devices, Union City, CA, USA). Cells were held at −70 mV to monitor seal breakthrough and then stepped to 0 mV. Series resistance was monitored over the course of the experiment and cells with unstable holding currents or significant changes in series resistance were not used for analysis. Best1 currents were evoked using a ramp protocol in which cells were held at 0 mV and then jumped to −100 mV followed by a 1.3 s ramp to 100 mV and a return step to 0 mV. Control ramps obtained in normal recording solution were interleaved with those in which the cells were exposed to 80 mM ethanol delivered using a Warner Faststep multi-barrel perfusion system (Hamden, CT, USA). In another set of experiments, Best1-expressing HEK-293 cells were subject to the previously mentioned protocol using 40 mM ethanol or NPPB (100 μM). Data were filtered at 1--2 kHz and acquired at 5 kHz using an Instrutech ITC-16 digital interface (Instrutech Corp., Port Washington, NY, USA) controlled by AxographX software (Axograph, Sydney, Australia). Data were analyzed offline using AxographX software (Axograph, Sydney, NSW, Australia). Statistical analyses were done with Prism (GraphPad, San Diego, CA, USA) using linear regressions followed by an unpaired *t*-test to compare slopes. A *P* \< 0.05 was considered statistically significant. Cerebellar slice electrophysiology ---------------------------------- We utilized male Sprague-Dawley rats at postnatal day (P) 23--30 from Harlan (Indianapolis, IN, USA) and C57/B6 mice (P28--33) from Charles River (Wilmington, MA, USA). Animals were group-housed and received food and water *ad libitum* until the day of the experiment. All animal procedures were approved by the UNM-Health Sciences Center Institutional Animal Care and Use Committee and conformed to National Institutes of Health Guidelines. Animals were sacrificed by rapid decapitation under deep anesthesia with ketamine (250 mg/kg i.p.). For most experiments, brains were quickly removed and submerged for 2 min in cold sucrose artificial cerebral spinal fluid (aCSF) containing (in mM): 220 sucrose, 2 KCl, 1.25 NaH~2~PO~4~, 26 NaHCO~3~, 12 MgSO~4~, 10 glucose, 0.2 CaCl~2~, and 0.43 ketamine, pre-equilibrated with 95% O~2~/5% CO~2.~ The vermis of the cerebellum was sliced in the sucrose aCSF at 200 μm using a vibrating tissue slicer (Leica Microsystems, Bannockburn, IL, USA). Immediately following this procedure, slices were transferred to a chamber containing normal aCSF and allowed to recover for 40 min at 35--36°C. This normal aCSF contained (in mM): 126 NaCl, 2 KCl, 1.25 NaH~2~PO~4~, 26 NaHCO~3~, 10 glucose, 1 MgSO~4~, 2 CaCl~2~, and 0.4 ascorbic acid and was continuously bubbled with 95% O~2~/5% CO~2.~ When indicated, we used the procedures described by Lee et al. ([@B18]); please refer to Table [1](#T1){ref-type="table"} for more details. ###### **Methodological differences in slice preparation and electrophysiological recording conditions**. Valenzuela lab methods Lee et al. ([@B18]) methods ---------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- P22--28 Sprague-Dawley rat and/or P28--P30 C57/B6 mouse P28 or \>8 weeks old C57/B6 mouse Sucrose cutting solution (in mM): 220 sucrose, 2 KCl, 1.25 NaH~2~PO~4~, 26 NaHCO~3~, 12 MgSO~4~, 10 glucose, 0.2 CaCl~2~, and 0.43 ketamine Sucrose cutting solution (in mM): 250 sucrose, 2.5 KCl, 1.25 NaH~2~PO~4~, 26 NaHCO~3~, 4 MgCl~2~, 10 glucose, 0.1 CaCl~2~, 3 myo-inositol, 2 sodium pyruvate, 0.5 ascorbic acid, and 1 kynurenic acid, pH 7.4 aCSF (in mM): 126 NaCl, 2 KCl, 1.25 NaH~2~PO~4~, 26 NaHCO~3~, 10 glucose, 1 MgSO~4~, 2 CaCl~2~, 0.4 ascorbic acid aCSF (in mM): 126 NaCl, 2.5 KCl, 1 NaH~2~PO~4~, 24 NaHCO~3~, 10 glucose, 2 MgCl~2~. 2.5 CaCl~2~, pH 7.4 Internal solution (in mM): 135 KCl, 10 HEPES, 2 MgCl~2~, 0.5 EGTA, 5 Mg-ATP, 1 Na-GTP, and 1 QX314-(Br), pH 7.25 adjusted with KOH (280--290 mOsm) Internal solution (in mM): 135 CsCl, 10 HEPES, 4 NaCl, 0.5 CaCl~2~, 5 EGTA, 2 Mg-ATP, 0.5 Na~2~-GTP, 10 QX-314, pH adjusted to 7.2 with CsOH (278--285 mOsmol) Incubation protocol: 40 min at 32--33°C then at least 20--30 min at room temp Incubation protocol: room temperature for at least 1 h prior to recording Holding potential: −70 mV Holding potential: −70 mV Pipette resistance: 3--5 MΩ Pipette resistance: 10--12 MΩ Whole-cell patch-clamp techniques were used to record tonic currents and spontaneous activity. Recordings were performed in a chamber perfused with aCSF at a rate of 2--3 ml/min and maintained at 32--33°C. Neurons were visualized using infrared-differential interference contrast microscopy and recordings were performed with an Axopatch 200B amplifier. GCs were identified on the basis of their location in the GC layer, morphology (small and round sized), and capacitance = ∼2--5 pF. Patch pipettes (tip resistance = 3--5 MΩ) were filled with an internal solution containing (in mM): 135 KCl, 10 HEPES, 2 MgCl~2~, 0.5 EGTA, 5 Mg-ATP, 1 Na-GTP, and 1 *N*-(2,6-Dimethylphenylcarbamoylmethyl)triethylammonium) bromide (QX314-Br), pH 7.25, osmolarity 280--290 mOsm. The holding potential was −70 mV. Inclusion criteria for analysis was that access resistance did not change \>20% throughout the duration of the experiment. GABAergic synaptic transmission was isolated by blocking AMPA and NMDA receptors using kynurenic acid (1 mM) and DL-APV (50 μM), respectively. During application of glutamate antagonists, neurons were allowed to equilibrate (∼5 min) prior to beginning an experiment. Data were acquired in gap-free mode at 10 kHz and filtered at 2 kHz. Data were analyzed with Clampfit-10 (Molecular Devices) and Mini Analysis 6.0.3 (Synaptosoft, Decatur, GA, USA). As previously shown, the tonic current amplitude and noise were calculated by fitting a Gaussian distribution to an all-point histogram for every minute of the recording, constraining the fit to eliminate a contribution of spontaneous synaptic events (Botta et al., [@B6]). Tonic current amplitude and noise were defined as the mean current and the standard deviation, respectively, recorded in the absence minus that recorded in the presence of gabazine (10 μM). Data were initially analyzed with the D'Agostino and Pearson omnibus normality tests. If data followed a normal distribution, these were analyzed using parametric tests. If this was not the case, then non-parametric tests were used. Pooled data were statistically analyzed with Prism and are presented as mean ± SEM. A *P* \< 0.05 was considered to be statistically significant. Ethanol was purchased from Aaper Alcohol and Chemical Company (Shelbyville, KY, USA). QX-314 Br and gabazine were obtained from Tocris (Ellisville, MO, USA or Bristol, UK), TTX was obtained from Calbiochem (Darmstadt, Germany) while all other chemicals were obtained from Sigma Chemical Company (St. Louis, MO, USA). Results ======= Effect of ethanol on recombinant Best1 channels ----------------------------------------------- We initially investigated whether recombinant Best1 channels were sensitive to acute ethanol exposure. These channels were transiently transfected into HEK-293 cells and activated using a ramp protocol. Figure [1](#F1){ref-type="fig"} shows a current density--voltage relationship in two different sets of Best1-expressing HEK-293 cells demonstrating that the recombinant Best1 channels easily pass current in and out of the cell with a reversal potential of 0 mV under our recording conditions (*n* = 9 for both experiments). Using a high Cl^−^ and Ca^2+^-based internal solution, these channels do not show any voltage dependent activation, consistent with what has been previously shown (Park et al., [@B20]). We found that application of 80 mM ethanol (*n* = 8) did not alter current density at any given membrane voltage potential, suggesting no effect of this ethanol concentration on Best1 conductance (Figure [1](#F1){ref-type="fig"}A, slope: Best1 = 0.1742 ± 0.0053; Best1 + 80 mM; EtOH = 0.1634 ± 0.0035). To confirm that currents were mediated by Best1 channels, we transfected cells with Best1 (W93C), in which a mutation in the pore significantly reduces Ca^2+^-activated anion-channel currents (Park et al., [@B20]). As expected, currents were largely eliminated in cells expressing Best1 (W93C) channels (*n* = 3) and this was significantly different from Best1 (Figure [1](#F1){ref-type="fig"}A, slope: Best1 (W93C) = 0.0425 ± 0.0009; *P* \< 0.0001 vs. Best1). Using another set of Best1-transfected HEK-293 cells, we also assessed a lower concentration of ethanol and found that application of 40 mM ethanol (*n* = 7) also did not alter current density at any of the membrane potentials tested (Figure [1](#F1){ref-type="fig"}B, slope: Best1 = 0.3421 ± 0.0129; Best1 + 40 mM; EtOH = 0.3267 ± 0.0117). We also tested the putative Best1 antagonist, NPPB, on Best1 functionality and found that the currents were significantly attenuated (Figure [1](#F1){ref-type="fig"}B; slope: Best1 + NPPB = 0.1423 ± 0.0027, *n* = 8; *P* \< 0.0001; vs. Best1) similar to currents measured in non-transfected cells (Figure [1](#F1){ref-type="fig"}B; slope: non-transfected = 0.0887 ± 0.0041, *n* = 6; *P* \< 0.0001 vs. Best1). Together, these data suggest that Best1 is insensitive to acute ethanol exposure and that NPPB blocks conductances in recombinant Best1 channels. These findings are in agreement with a previous report indicating that NPPB blocked recombinant Best1 channels (Park et al., [@B20]). ![**Ethanol does not change Best1 currents in HEK-293 cells**. Non-transfected HEK-293 cells and HEK-293 cells transfected with Best1 or Best1 (W93C) were voltage-clamped at 0 mV and then stepped to −100 mV. A 1.3 s voltage ramp from −100 to +100 mV was then run before returning cells to 0 mV \[**(A)**, inset\]. Ramps were repeated in the presence of **(A)** 80 mM ethanol (EtOH), **(B)** 40 mM EtOH, or NPPB (100 μM). Data are shown as mean current density expressed as pA/pF (±SEM; *n* = 9 for Best1 (top and bottom graphs), *n* = 8 for Best1 + 80 mM Ethanol, *n* = 3 for Best1(W93C), *n* = 7 for Best1 + 40 mM EtOH, *n* = 8 for Best1 + NPPB, *n* = 6 for non-transfected).](fnins-05-00148-g001){#F1} Effect of NPPB on GABAergic currents in GCs ------------------------------------------- NPPB was shown to decrease tonic GABAergic currents in GCs in cerebellar slices from mice (Lee et al., [@B18]). We first tested whether this effect could be reproduced in rat GCs. In contrast to the findings of Lee et al. ([@B18]) and in agreement with a previous report (Rossi et al., [@B23]), we found that NPPB increases the tonic current amplitude, with no effect on the tonic current noise (Figure [2](#F2){ref-type="fig"}A1). To test if the Best1 contribution to the tonic current was species specific, we performed the same experiment in slices from C57/B6 mice and again found that NPPB facilitates the tonic current amplitude with no effect on the noise under our recording conditions (Figure [2](#F2){ref-type="fig"}A2). Given that there were several methodological differences between our study and the Lee et al. ([@B18]) study, we repeated the experiments using conditions that more closely matched those described in that study (Table [1](#T1){ref-type="table"}). Under these conditions, we also found robust potentiation of the tonic current amplitude by NPPB with no effect on the tonic current noise (Figure [2](#F2){ref-type="fig"}A3). Given that we found similar results under all of the conditions, we combined the data obtained with rat and mouse slices and this yielded an NPPB-induced increase in tonic current amplitude of 21.20 ± 5.59 pA (*P* \< 0.05 by Wilcoxon Signed Rank Test compared to 0; Figure [2](#F2){ref-type="fig"}B1) or 232.2 ± 118.6% (*P* \< 0.05 by Wilcoxon Signed Rank Test compared to 0; n = 7) and there was a non-significant change in tonic current noise of 0.14 ± 0.3 pA or 21.2 ± 15.9%; *n* = 7 (Figure [2](#F2){ref-type="fig"}B2). These experiments clearly demonstrate that NPPB is not a selective Cl^−^-channel/Best1 antagonist, as it potentiates tonic GABAergic currents in rodent GCs. ![**NPPB potentiates tonic currents in the absence of TTX**. Sample traces of tonic currents with application of NPPB (100 μM) followed by gabazine (10 μM) in **(A1)** rats, **(A2)** mice, or **(A3)** mice\* (using the methods from Lee et al., [@B18]; see Table [1](#T1){ref-type="table"}). Effect of NPPB on **(B1)** tonic current amplitude and **(B2)** noise of individual cells (open triangles-rat, open squares-mice, dark circles-mice using the methods of Lee et al. ([@B18]). NPPB significantly increased the tonic current amplitude while having no effect on the tonic current noise. \**P* \< 0.05, paired *t*-test; *n* = 7.](fnins-05-00148-g002){#F2} We also examined whether NPPB altered sIPSCs in GCs, as it has been previously suggested that NPPB may act to increase GABA release (Rossi et al., [@B23]). Analysis of sIPSCs (Figure [3](#F3){ref-type="fig"}) from rat and mice slices (combined, *n* = 7) revealed no significant effect of NPPB (shown as % change from baseline) on frequency (3.09 ± 32.97%) or amplitude (11.30 ± 10.91%). In contrast, NPPB significantly increased the total current charge (change from control in sIPSC area = 98.93 ± 33.56%, *P* \< 0.05, one sample *t*-test vs. 0) and decay time (change from control = 81.78 ± 26.86%, *P* \< 0.05, Wilcoxon Signed Rank Test vs. 0). {#F3} Effect of other Best1 channel antagonists on tonic currents in rat GCs in presence of TTX ----------------------------------------------------------------------------------------- To further characterize the effect of Best1 antagonists on tonic GABAergic currents in rat slices, we blocked the Golgi cell-dependent component of these currents using TTX. This agent significantly decreased the tonic current (change in tonic current amplitude = −28.66 ± 10.38%; change in tonic current noise = −27.23 ± 8.89%; *n* = 20; *P* \< 0.05 by Wilcoxon Signed Rank Test vs. 0), in addition to reducing sIPSCs (change in sIPSC frequency = −82.71 ± 4.22%; *P* \< 0.0001 by Wilcoxon Signed Rank Test vs. 0). In the presence of TTX, NPPB significantly potentiated tonic current amplitude \[Figure [4](#F4){ref-type="fig"}A, tonic current amplitude, TTX = 8.51 ± 2.48 pA, TTX + NPPB = 27.28 ± 4.11 pA (*P* \< 0.05, paired *t*-test)\] with a non-significant effect on tonic current noise \[TTX = 2.41 ± 0.33 pA, TTX + NPPB = 4.52 ± 0.93 pA (*P* \> 0.05, paired *t*-test; *n* = 5)\]. We next tested the effect of two chemically distinct Cl^−^-channel blockers (DIDS and NFA) that were previously shown to inhibit Best1 channels and reduce the tonic current in GCs (Lee et al., [@B18]). In contrast to NPPB and the results of Lee et al. ([@B18]), DIDS did not significantly modulate tonic currents \[Figure [4](#F4){ref-type="fig"}B, tonic current amplitude, TTX = 12.41 ± 2.60 pA, TTX + DIDS = 11.29 ± 2.66 pA (*P* \> 0.05 paired *t*-test); tonic current noise, TTX = 2.36 ± 0.43 pA, TTX + DIDS = 2.65 ± 0.42 pA (*P* \> 0.05 paired *t*-test); *n* = 5\]. On average, we found that NFA had no effect on the tonic current amplitude, but significantly decreased the tonic current noise \[Figure [4](#F4){ref-type="fig"}C, tonic current amplitude, TTX = 14.30 ± 3.32 pA, TTX + NFA = 12.18 ± 2.17 pA (*P* \> 0.05 paired *t*-test); tonic current noise, TTX = 2.66 ± 0.44 pA, TTX + NFA = 1.89 ± 0.23 pA (*P* \< 0.05 paired *t*-test); *n* = 24\]. ![**Effect of different Best1 antagonists on tonic currents in presence of TTX in rat slices**. **(A1,B1,C1)** Sample traces showing the effect of Best1 channel antagonists on tonic currents in the presence of TTX. NPPB significantly increased the tonic current amplitude by 387.1 ± 180.6% **(A2)**, but had no effect on the tonic current noise \[101.7 ± 52.02%; **(A3)**; *n* = 5\]. DIDS had no effect on the tonic current amplitude \[3.79 ± 23.31%; **(B2)**; *n* = 5\] or noise \[15.02 ± 7.89%; **(B3)**\]. NFA had no effect on the tonic current amplitude \[−37.27 ± 36.87%; **(C2)**; *n* = 24\], but significantly decreased the tonic current noise by −18.34 ± 8.08% **(C3)**. \**P* \< 0.05, paired *t*-test.](fnins-05-00148-g004){#F4} We also determined whether the Best1 antagonists altered the properties of spontaneous synaptic events recorded in the presence of TTX (i.e., mIPSCs). Similar to our findings on sIPSCs, NPPB (Figure [5](#F5){ref-type="fig"}A) had no significant effect on mIPSC frequency (change from control = 18.62 ± 39.92%) or amplitude (change from control = 20.32 ± 22.44%). Although NPPB did not significantly increase the area of mIPSCs (change from control = 55.96 ± 23.53%) it did significantly increase the decay time (Figure [5](#F5){ref-type="fig"}A, change from control = 50.83 ± 17.73%, *P* \< 0.05, one sample *t*-test vs. 0). DIDS (Figure [5](#F5){ref-type="fig"}B) did not significantly alter mIPSC frequency (change from control = −17.25 ± 13.49%), amplitude (change from control = −15.26 ± 7.92%), area (change from control = 10.97 ± 25.55%), or decay time (change from control = 18.56 ± 20.51%). Interestingly, NFA (Figure [5](#F5){ref-type="fig"}C) significantly decreased the frequency, amplitude, and area (change from control = −32.35 ± 1.93, −16.94 ± 3.73, and −21.74 ± 9.23%, respectively, *P* \< 0.05 one sample *t*-test vs. 0) with no effect on the decay time (change from control = −0.02 ± 11.59%). ![**Effect of Best1 antagonists on mIPSCs**. Sample traces of mIPSCs during baseline and after application of **(A1)** NPPB, **(B1)** DIDS, and **(C1)** NFA. Exemplar traces of averaged mIPSCs illustrating the effect of NPPB **(A2)**, DIDS **(B2)**, and NFA **(C2)**. Summary graphs illustrating that NPPB significantly increased mIPSC decay time only \[**(A3)**; *n* = 3\], DIDS had no effect on mIPSCs \[**(B3)**; *n* = 3\], and NFA significantly decreased frequency, amplitude, and area of mIPSCs \[**(C3)**; *n* = 3\]. \**P* \< 0.05, one sample *t*-test compared to 0.](fnins-05-00148-g005){#F5} To determine the effect of the Best1 antagonists on resting conductances in GCs, we tested their actions in the presence of the GABA~A~R antagonist gabazine. NPPB minimally, but significantly, increased the residual current (not shown; current amplitude, gabazine (10 μM) = 1.92 ± 2.17 pA, gabazine + NPPB = 4.47 ± 2.65 pA, *P* \< 0.05 paired *t*-test; *n* = 3). DIDS did not significantly affect the residual current amplitude (not shown; current amplitude, gabazine = −3.20 ± 2.87 pA, gabazine + DIDS = 1.10 ± 0.70; *n* = 3, *P* \> 0.05 paired *t*-test). Likewise, NFA did not alter the residual current (not shown; current amplitude, gabazine = 1.88 ± 3.35 pA, gabazine + NFA = 0.02 ± 3.33 pA; *n* = 3, *P* \> 0.05 by paired *t*-test). Discussion ========== Understanding the mechanism by which ethanol modulates the tonic GABA~A~R current in GCs is important given the role of the cerebellar GC layer on spatio-temporal encoding (see review by D'Angelo and De Zeeuw, [@B13]). In the current study, we tested whether ethanol's effects on the tonic current were, at least in part, mediated by Best1, an anion-channel found on glia that has recently been shown to mediate release of GABA from these cells and robustly contribute to the tonic current in cerebellar slices (Lee et al., [@B18]). We initially assessed the effect of acute ethanol on recombinant Best1 channels expressed in HEK-293 cells and found that it had no significant effect on currents carried by these channels. Unexpectedly, in cerebellar brain slices we were unable to detect theBest1-mediated contribution to the tonic current with several anion-channel blockers (regardless of the recording condition or the animal species) precluding assessment of the effect of ethanol on native Best1 function in this preparation. Based on these findings, we conclude that the acute effects of ethanol on the tonic current in cerebellar slices are independent of any action on Best1. Acute ethanol does not modulate recombinant Best1 channels ---------------------------------------------------------- We successfully expressed recombinant Best1 channels in HEK-293 cells that could be blocked with NPPB. As expected, cells transfected with a functionally inactivated Best1 channel (Best1 W93C) had very small conductances over the membrane potentials tested, similar to non-transfected cells. We found that acute exposure to 40 and 80 mM ethanol had no effect on the functional conductance of recombinant Best1 at increasing or decreasing membrane potentials, supporting the hypothesis that these channels are insensitive to acute ethanol exposure. Moreover, these data support our previous findings indicating that the facilitatory effect of ethanol on tonic GABAergic currents in GCs is due to an increase in Golgi cell firing, with direct potentiation of extrasynaptic GABA~A~Rs perhaps contributing to the mechanism of action of ethanol under some experimental conditions (Carta et al., [@B10]; Hanchar et al., [@B15]; Botta et al., [@B4]), but see (Botta et al., [@B5]). It is worth noting that although the experiments with recombinant Best1 suggest no effect of ethanol on this channel, it is still possible that it may have an effect on Best1 in a native system. Future studies in which Best1-mediated currents are directly recorded from glial cells, similar to the studies done by Lee et al. ([@B18]) should be pursued to address this issue. It is also possible that ethanol could affect native Best channels expressed in other cell types such as epithelial cells, underlying certain actions of ethanol in these tissues. In fact, it has been shown that ethanol can inhibit Ca^2+^-activated Cl^−^-channels (Sanna et al., [@B24]; Clayton and Woodward, [@B12]), which Best1 channels are (Kunzelmann et al., [@B17]), suggesting that some Best channel sub-types may show some sensitivity to ethanol. Additional studies should investigate this possibility as well as the chronic actions of ethanol on these channels. NPPB potentiates tonic and phasic GABA~A~ receptor-mediated currents in GCs. ---------------------------------------------------------------------------- We attempted to block Best1-mediated tonic GABAergic currents using NPPB, an anion-channel blocker that was shown to inhibit the tonic current in GCs (Lee et al., [@B18]). Unexpectedly, we consistently found that this agent induces significant potentiation of the tonic current under a variety of experimental conditions, despite the fact that NPPB produced the expected block of recombinant Best1 channel conductances. We thought that differences in animal species or slice preparation/recording methodology could explain the differences between our results and those of Lee et al. ([@B18]). However, we ruled out some of these factors by measuring GABAergic currents in slices from both rats and mice, and also by closely following the methodology used by Lee et al. ([@B18]). NPPB-mediated potentiation of tonic GABAergic currents in GCs was reported in a previous paper, where it was suggested that NPPB could produce this effect by increasing GABA release from Golgi cells (Rossi et al., [@B23]). Our sIPSC and mIPSC recordings suggest that NPPB does not increase action potential-dependent or -independent GABA release, as the frequency of these events was not significantly affected by NPPB. Moreover, NPPB increased tonic GABAergic currents to a similar extent in the absence and presence of TTX, suggesting that it does not potentiate these currents by increasing spontaneous Golgi cell firing. This finding is consistent with direct potentiation of extrasynaptic GABA~A~Rs by NPPB. Our experiments indicate that NPPB also modulates synaptic GABA~A~Rs, as the area and decay times of sIPSCs were significantly increased. The Lee et al. ([@B18]) study reported NPPB did not affect the function of GABA~C~ receptors expressed in HEK-293 T cells, but the effect of this agent on GABA~A~Rs was not evaluated. In addition, that study also reported that NPPB slightly increased currents in GCs induced by local application of GABA, but the authors interpreted this finding as a consequence of blockade of the tonic current noise, leading to enhancement of synaptic events. The results from the current study argue against this mechanism, suggesting that NPPB directly potentiates GABA~A~Rs expressed in GCs. Clearly, this agent lacks selectivity for Best1 channels and should not be used to characterize the role of these channels on GABAergic transmission. Different anion--channel blockers failed to reduce tonic GABAergic currents. ---------------------------------------------------------------------------- Tonic GABAergic currents are, in part, mediated by action potential-dependent GABA release from Golgi cells, while Best1 channels have been suggested to mediate the remaining portion of the tonic current (Lee et al., [@B18]). However, when we eliminated the Golgi cell component using TTX and isolated the (presumed) Best1-mediated component, we were still unable to significantly reduce tonic currents with any of the Best1 antagonists used in the study of Lee et al. ([@B18]). As mentioned above, NPPB potentiated the tonic current to the same extent as in experiments done without TTX. In addition, analysis of mIPSCs showed that NPPB significantly lengthened the decay time, providing further evidence for a post-synaptic effect on GABA~A~Rs. DIDS and NFA had no significant effect on the Golgi-independent component of the tonic current, although NFA reduced the tonic current in some cells. DIDS did not alter mIPSCs, while NFA significantly decreased frequency, amplitude, and area of mIPSCs, suggesting a pre- and post-synaptic effect of NFA on phasic GABA transmission, with the post-synaptic effect being consistent with NFA's affinity for GABA~A~ receptors (Sinkkonen et al., [@B25]). Although we were unable to replicate the findings of Lee et al. ([@B18]) using anion-channel blockers in a slice preparation, it should be noted that these investigators demonstrated a Best1-mediated component in the GC tonic current by knocking down Best1 using shRNA techniques. A potential explanation for the discrepancies between our study and the work of Lee et al. ([@B18]) is that Best1 channels do mediate GABA release from astrocytes, but only under certain conditions; for example, when slice oxygenation is not optimal. Indeed, an increase in tonic GABAergic currents was demonstrated in the frontal cortex of an animal model of stroke (Clarkson et al., [@B11]), and ambient GABA levels were shown to increase following ischemic-like insults in the hippocampus (Allen and Attwell, [@B1]; Allen et al., [@B2]; Ransom et al., [@B22]). Consistent with this possibility, our studies revealed an average gabazine-sensitive tonic current of 17.6 ± 2.5 pA (*n* = 29 cells), in contrast to the 35.7 ± 0.1 pA tonic current amplitude previously reported (Lee et al., [@B18]). Moreover, the tonic current sample traces included in the paper of Lee et al. ([@B18]) do not show the presence of sIPSCs, indicating that Golgi cells were not active under their recording conditions, perhaps due to compromised slice health. Clearly, studies should further investigate the possibility that Best1 channel function depends on the metabolic status of GCs in the acute slice preparation, as well as the intact cerebellum. Conclusion ========== In the current study, we sought to investigate the effect of ethanol on Best1 channels and the contribution of these channels to the ethanol-induced potentiation of tonic GABAergic inhibition in cerebellar GCs. In a recombinant system, we found that acute ethanol did not alter NPPB-sensitive Best1 channel conductances. Using three chemically distinct antagonists of Best1 channels, we were unable to detect a Best1-dependent component in the tonic currents in a slice preparation. The results of this study are consistent with the model that the ethanol-induced potentiation of the tonic GABAergic current in cerebellar GCs is, at least in part, a consequence of an increase in Golgi cell firing and GABA spill-over. 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: A. Leslie Morrow, University of North Carolina School of Medicine, USA [^2]: Reviewed by: Jason B. Wu, Cedars-Sinai Medical Center, USA; Thomas Louis Kash, University of North Carolina Chapel Hill, USA [^3]: This article was submitted to Frontiers in Neuropharmacology, a specialty of Frontiers in Neuroscience.

INTRODUCTION ============ Smoking is one of the modifiable risk factors for many chronic diseases, such as cardiovascular disease (CVD), cancer, chronic obstructive lung disease, asthma, and diabetes. However, the adverse effects of smoking on diabetes have been generally under recognized. In the guidelines from the Korean Diabetes Association, smoking cessation is recommended as one of the most important steps in preventing the cardiovascular complications of diabetes \[[@B1]\]. Many studies have shown that the adverse effects of smoking on diabetes mellitus are not only diabetic macrovascular complications but the causal nature of its association with diabetes and the progression of diabetic microvascular complications has yet to be explored. Although smoking is known to decrease body weight, it is associated with central obesity \[[@B2]\]. Smoking also increases inflammation and oxidative stress \[[@B3]\], to directly damage β-cell function \[[@B4]\] and to impair endothelial function \[[@B5]\]. The prevalence of smoking in Korean men is near 50%, which is the highest smoking rate in the Western Pacific region. In addition to obesity, the high prevalence of smoking is one of the major health problems for Korea\'s public health. This review is about the various smoking effects on diabetes mellitus, diabetic complications, and diabetic incidence. Understanding the hazardous effects of smoking on diabetes mellitus may lead to more emphasis on smoking prevention and smoking cessation as important strategies in the management of diabetes mellitus. SMOKING AND DIABETES INCIDENCE ============================== There is much evidence that smoking increases the risk of diabetes. Several cohort studies in Korea have reported that smoking was associated with an increased risk for the development of diabetes. Cho et al. \[[@B6]\] followed 4,041 men for 4 years in rural and urban settings in Korea, and found that past and current smokers had a significantly increased risk for type 2 diabetes, and the risk increased with the number of cigarettes smoked. Another study reported a 14-year-long prospective cohort study, in which the risk of diabetes among men and women who smoked 20 cigarettes or more per day was 1.55 (95% confidence interval \[CI\], 1.51 to 1.60) compared to those who never smoked \[[@B7]\]. A Japanese study reported similar results of a positive correlation between cigarette consumption and risk for diabetes \[[@B8]\]. The health professionals\' follow-up study demonstrated that the risk for diabetes among men who smoked ≥25 cigarettes per day was 1.94 (95% CI, 1.25 to 3.03) \[[@B9]\]. Another British study showed the risk for diabetes in smoking men was around 1.7, after adjusting for confounding factors, such as age, body mass index, physical activity, alcohol intake, social class, and antihypertensive treatment \[[@B10]\]. There have been few studies on the effect of smoking on the risk of diabetes in women as generally the prevalence of smoking is lower in women than men. However, the results from the Nurses\' Health Study in the United States (114,247 women, 1,227,589 person-years follow-up) showed that the risk for diabetes in smokers was 1.42 after adjustment for other risk factors \[[@B11]\]. The same cohort was followed for 16 years, and a new analysis was performed. The predictable risk factors for diabetes were overweight and obesity, as in men, low physical activity, a poor diet, current smoking, and abstinence from alcohol were all independently associated with the risk for diabetes. The adjusted risk for diabetes in smokers was 1.4 compared with non-smokers \[[@B12]\]. THE EFFECT OF SMOKING ON INSULIN ACTION ======================================= The exact mechanism for why smoking increases the risk of diabetes and deteriorates glucose homeostasis has not been fully elucidated, but the available evidence shows that smoking increases insulin resistance. In healthy young men, acute smoking showed an increased insulin resistance \[[@B13]\]. Smokers had a significantly increased homeostatic model assessment insulin resistance index an hour after smoking \[[@B14]\]. The smoking reduced insulin mediated glucose uptake by 10% to 40% in men who smoked compared with non-smoking men \[[@B15],[@B16]\]. In type 2 diabetic subjects, insulin and C-peptide responses to oral glucose load were significantly higher in smokers than non-smokers and the insulin resistance, as determined by the euglycemic clamp technique, was positively correlated in a dose dependent manner \[[@B17]\]. Thus smoking induced insulin resistance in patients with type 2 diabetes, as well as healthy subjects. In addition to increased insulin resistance, smoking also showed dyslipidemia prone to atherosclerosis. Smokers had higher fasting triglycerides and lower high density lipoprotein cholesterol levels, and an increased proportion of small dense low density lipoprotein particles. Fibrinogen levels and plasminogen activator inhibitor 1 activity were also elevated in smokers \[[@B18]\]. In terms of glucose homeostasis, smoking has a negative effect on glucose control. In a population-based prospective study, cigarette smoking was positively associated in a dose dependent manner with elevated HbA1c after adjustment for possible confounding by dietary variables \[[@B19]\]. This finding was also reported in patients with diabetes in Sweden; smoking type 1 and type 2 patients had a higher mean HbA1c but a lower mean body mass index than non-smokers \[[@B20]\]. SMOKING AND DIABETIC MICROVASCULAR COMPLICATIONS ================================================ The smoking effects on microvascular diabetes complications vary across reports. Generally, several studies have shown that smoking has an adverse effect on diabetic nephropathy, but the influence of smoking independently with glucose control, on retinopathy and neuropathy are unclear. SMOKING AND NEPHROPATHY ======================= Several studies have demonstrated that smoking promotes diabetic microalbuminuria and exacerbates diabetic nephropathy. In the study by Biesenbach et al. \[[@B21]\], a 13-year follow-up study, the progression of nephropathy was clearly increased in smokers. The authors showed that smoking was a risk factor for diabetic kidney disease, independent of age, sex, and duration of diabetes and HbA1c levels. In prospective studies by Chuahirun and Wesson \[[@B22]\] and Chuahirun et al. \[[@B23]\], the adverse effects on diabetic neprhropathy in type 2 patients were confirmed, even in optimal hypertensive patients. SMOKING AND RETINOPATHY ======================= The association of smoking and diabetic retinopathy has not been clear. It was reported that retinopathy has been associated with glycemic control and not smoking state \[[@B24]\]. Some studies have reported no association with smoking and retinopathy in type 2 diabetes \[[@B24],[@B25]\]. The United Kingdom Prospective Diabetic (UKPD) study to determine risk factors related to the incidence and progression of diabetic retinopathy followed patients over 6 years from diagnosis. The development of retinopathy was associated with glycemia and higher blood pressure, but not smoking \[[@B26]\]. Thus in type 2 patients, the effects of smoking on diabetic retinopathy has not been as clear as with nephropathy. SMOKING AND NEUROPATHY ====================== There are few studies about smoking and diabetic neuropathy. Smoking may affect diabetic neuropathy differently according to the type of diabetes \[[@B27]\]. In type 2 diabetic patients, smoking was not a risk factor in the presence of polyneuropathy or sensory neuropathy as diagnosed by symptom and sign \[[@B27],[@B28]\]. It was reported that there was no relationship between current or previous levels of smoking and the severity and duration of chronic painful neuropathy \[[@B29]\]. But in the study by Tamer et al. \[[@B30]\], while smoking was not associated with neuropathic complaints, using electromyography-supported neuropathy examination there were significant relationships with smoking, as well as HbA1c. Therefore, more studies are needed to evaluate the association between smoking and neuropathy. SMOKING AND MACROVASCULAR COMPLICATIONS ======================================= Smoking has been shown to be a significant risk factor for all-cause mortality, and for mortality due to CVD and coronary heart disease (CHD) in diabetics. Smokers die on average 8 to 10 years younger than non-smokers, as age is entered into most multi-regression analysis. SMOKING AND CHD =============== Smoking is a major risk factor for CVD in non-diabetic subjects, as well as diabetic subjects. In an 8-year prospective study, smoking was significantly associated with an increased risk for CHD in diabetic patients \[[@B31]\]. The UKPD study clearly showed that smoking was a significant and independent risk factor for CHD in type 2 diabetic patients \[[@B32]\]. In the Nurses\' Health Study, in women with type 2 diabetes, it was demonstrated that cigarette smoking was associated in a dose-dependent manner with an increased mortality and CHD. Compared with never-smokers, the relative risks for CHD were 1.66 for current smokers of 1 to 14 cigarette per day, and 2.68 for current smokers of 15 or more cigarettes per day \[[@B33],[@B34]\]. Recently, a meta-analysis in the Asia-Pacific region, in men with diabetes, the hazard ratio comparing current smokers with non-smokers was 1.42 for CHD. In Asia, where there are high rates of smoking, and a rapidly increasing prevalence of diabetes, the author concluded that cigarette cessation strategies there were huge benefits in terms of reducing the burden of CVD in men with diabetes \[[@B35]\]. SMOKING AND STROKE ================== Smoking also increases the risk of stroke in patients with diabetes, but may not be as strong as CHD. In the UKPD study, mathematical models were developed to estimate the risk of stroke, and the variables were smoking, duration of diabetes, age, sex, systolic blood pressure, total cholesterol to high density lipoprotein cholesterol ratio, and presence of arterial fibrillation \[[@B36]\]. In a study using the general practice research database in the United Kingdom, smoking was an additional risk factor for stroke in type 2 diabetic patients \[[@B37]\]. Another 4-year prospective study, also showed that smoking and HbA1c were predictors of stroke among the type 2 diabetic patients without a history of a previous stroke \[[@B38]\]. The relative risk of smoking for stroke has not been as high as that for CHD. In the Nurses\' Health Study, in smokers who smoked 1 to 14 cigarette per day, the risk was significant for CHD but not for stroke. In those who smoked 15 cigarettes or more per day, the relative risk for CHD and stroke were 2.68 and 1.84, respectively \[[@B33]\]. Similar trends were shown in a Swedish study, in which the relative risk of smoking was higher in myocardial infarction (2.33) than for stroke (1.12) in 30 to 59 year-old patients \[[@B39]\]. CONCLUSIONS =========== There have been many studies showing that smoking has harmful effects on patients with diabetes. Smoking increases diabetic incidence and aggravates glucose homeostasis and chronic diabetic complications. In microvascular complications, the onset and progression of diabetic nephropathy is highly associated with smoking. In macrovascular complications, smoking is associated with a 2 to 3 times higher incidence of CHD and mortality. However, smoking prevention and smoking cessation may not be emphasized enough in diabetic clinics. Thus, educating patients on the importance of not smoking and engaging in smoking cessation programs are important strategies for the management of diabetes. No potential conflict of interest relevant to this article was reported.

1. Introduction {#sec1} =============== The effect of nitrate supplementation on human health has been evaluated. Nitrate can be used as a substrate for the synthesis of nitric oxide (NO) from the nitrate-nitrite-NO pathway \[[@B1], [@B2]\]. Recently, our group and others have shown that nitrite can be used as an index of training intensity and load \[[@B3]\] in the reduction of blood pressure \[[@B4], [@B5]\], in the resting metabolic rate \[[@B6]\], and in the modulation of mitochondrial function \[[@B7]\]. In addition, nitrate supplementation decreases the oxygen cost of submaximal exercise and increases high-intensity exercise tolerance \[[@B8]\]. The increase in NO bioavailability acts on the relaxation of the vascular wall \[[@B9]\], as well as on the balance between antioxidants and prooxidant agents, improving vascular function \[[@B9]\]. These findings suggest improvement in vascular health, in addition to enhancing physical performance from increased NO bioavailability. Besides NO formation, nitrate supplementation can act as an antioxidant potential agent, which suppresses free radical formation \[[@B10]\]. Oxidative stress is characterized as an imbalance between the increase in reactive oxygen species (ROS) and the ability of biological systems to decrease reactive intermediates \[[@B11]\]. Regarding its antioxidant role, NO can eliminate the oxidants produced by the Fenton reaction, reduce equivalents provided by superoxide (O~2~^--^), or prevent the reaction of peroxide \[[@B12]\]. Thus, sodium nitrate and other nitrate donors, such as beetroots, have been shown to suppress radical formation and to be scavengers of potentially damaging reactive oxygen and nitrogen species, suggesting nitrate may also exhibit antioxidant effects \[[@B10], [@B12]\]. Regarding nitrate supplementation and physical exercise, evidence indicates that an improvement in cardiovascular health and physical performance is plausible but inconsistent \[[@B13]\]. For example, the duration of the supplementation period and maintenance of a normal diet during the supplementation period can enhance exercise tolerance and performance \[[@B8], [@B14], [@B15]\] or not improve performance \[[@B16], [@B17]\], but this is by no means a universal finding. Likewise, there is little evidence on the influence of nitrate supplementation on oxidative stress in physical exercise. Carriker et al. \[[@B18]\] showed that acute dietary nitrate supplementation does not attenuate oxidative stress during submaximal exercise. However, no study showed the effect of chronic nitrate supplementation on oxidative stress after exercise. Therefore, considering that nitrite can increase dilatation of blood vessels and decrease free radical formation, in this study, we investigated the effect of sodium nitrate supplementation on oxidative stress markers and blood pressure responses after moderate-intensity aerobic exercise performance in physically active males. 2. Material and Methods {#sec2} ======================= 2.1. Volunteers {#sec2.1} --------------- Fourteen healthy young men (22 ± 3 years of age, 69.3 ± 3 kg body mass, 23 ± 1 kg/m^2^ BMI, 56.9 ± 1.5 kg lean mass, 14.8 ± 1.5 kg fat mass, and 243 ± 11 W maximum workload) that were nonsmokers and not using any kind of supplementation were recruited. The volunteers were physically active but not highly trained in sports. An incremental test was performed to evaluate the maximal workload and to prescribe the main test intensity. The experimental protocol was approved by the local Institutional Review Board and is in accordance with the Declaration of Helsinki. 2.2. Design {#sec2.2} ----------- This study was conducted to investigate the acute and chronic effects of a 5-day supplementation of sodium nitrate or placebo in a crossover design and in random order for each volunteer. All volunteers were instructed to maintain their normal exercise and diet routines throughout the experimental period but to avoid alcohol, caffeine, and foods rich in nitrate 24 h prior to the tests. [Figure 1](#fig1){ref-type="fig"} shows the study design. During the first visit, body composition was measured by bioelectrical impedance (BIA balance-pole, Tanita BC558) and the volunteers were familiarized with the cycle ergometer (Cycle Ergometer: Bike Mechanics Braking CEFISE, Campinas, SP). A maximal incremental test \[[@B19]\] was performed to identify the exercise intensity. The test consisted of applying 35-watt increments every 2 min and a fixed rotation of 70 rpm. The maximal workload was estimated according to the equation described by Stegmann and colleagues \[[@B20]\]. During the following 4 visits, both acute (AC) response (T1 and T3) and the response after 5 days with nitrate supplementation (FD) (T2 and T4) in moderate exercise performance were investigated. This intensity was chosen to allow the participant to perform at least 30 min on the bike, and it is well established to be moderate intensity that could promote both postexercise hypotension and enough oxidative stress \[[@B19]\]. The volunteers performed 30 min of cycle ergometer exercise with 50% maximum workload from 8:00--10:00 am. Nitrate supplementation (NaNO~3~ 10 mg kg^−1^ body weight, or placebo, sodium chloride in identical capsule) was carried out in both acute supplementation (AC: T1 and T3) and after 5-day supplementation (FD: T2 and T4). The washout period between T2 and T3 tests was 7 days in random order with a crossover design. During the exercise tests, saliva samples were collected and heart rate (HR, Polar RS800CX) and blood pressure (BP, OMRON HEM 7200) were monitored at rest (0′); 1 h after supplementation (60′); immediately after the exercise test (90′); and 15 (105′), 30 (120′), 45 (135′), and 60 (150′) min after the test. 2.3. Saliva Collection {#sec2.3} ---------------------- Saliva was collected in collection vials with no exogenous stimulation using the guidelines proposed by Granger et al. \[[@B21]\]. The subjects were instructed to avoid the ingestion of foods with high nitrite levels, such as green vegetables, beets, strawberries, grape, and tea, and the use of mouthwash during the entire study. They were also instructed to accumulate saliva, and the spit was deposited for 1 min and consequently centrifuged. The supernatant was stored at --80°C until the assay. 2.4. Nitrite (NO~2~^--^) {#sec2.4} ------------------------ Nitric oxide bioavailability was estimated by the formation of nitrite (NO~2~^−^) in saliva using the Griess reaction \[[@B22]\]. Equal volumes of saliva and Griess reagent (1% sulfanilamide and 0.1% N-(1-naphthyl)ethylenediamine dihydrochloride in 2.5% phosphoric acid) were incubated. Nitrite content was calculated based on a standard curve of sodium nitrite (NaNO~2~). The concentration of total protein in each sample was determined using the Bradford method \[[@B23]\]. The analyses were performed using a VersaMax microplate reader (Molecular Devices, Menlo Park, CA, USA). 2.5. Alpha-Amylase {#sec2.5} ------------------ During salivary alpha-amylase (sAA) analysis, 10 *μ*g of total protein from each sample was loaded onto 5--22% SDS-PAGE \[[@B24]\]. Proteins were transferred onto nitrocellulose membranes (0.45 *μ*m) for 2 h at 100 mA at 4°C. Membranes were blocked for 4 h at 4°C in blocking buffer and incubated overnight at 4°C with a homemade affinity purified polyclonal rabbit anti-human sAA antibody, as reported in Santos et al. (2011). Membranes were subsequently incubated with secondary antibodies for 3 h, and alpha-amylase was detected using ECL reagents. Densitometry analyses of the spots were performed using ImageJ (US NIH, Bethesda, Maryland, USA). 2.6. Salivary Lactate {#sec2.6} --------------------- Salivary lactate was analysed by the electroenzymatic method using a biochemical analyser YSI 2300 Stat Plus (Yellow Springs, Ohio, USA). 2.7. Salivary Uric Acid {#sec2.7} ----------------------- Salivary uric acid concentrations were measured with a kit supplied by Labtest (Labtest, Brazil). Uricase was used in the assay to transform uric acid into allantoin and hydrogen peroxide. In the presence of peroxidase, hydrogen peroxide reacts with 4-aminoantipyrine and DHBS, forming the chromogen antipirilquinonimine. 2.8. Total Antioxidant Capacity (FRAP) {#sec2.8} -------------------------------------- Total antioxidant capacity was evaluated by assessing the reduction of iron in its ferric state (Fe^3+^) to its ferrous state (Fe^2+^) at low pH. Total antioxidant capacity was calculated based on a standard curve constructed using 6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid (Trolox). 2.9. Thiobarbituric Acid (TBARS) {#sec2.9} -------------------------------- Lipid peroxidation was assayed by measuring TBARS products using the method described by Walls et al. (1976) and adapted for a microplate format. Trichloroacetic acid (50% *w*/*v*) was added to the saliva supernatant, followed by incubation and centrifugation. Subsequently, the supernatant was removed and 0.75% thiobarbituric acid and 0.75% in 0.1 M HC1 were added. The samples were heated at 90--95°C for 20 min and centrifuged. The standard curve was prepared with malonaldehyde bis(dimethyl acetal) hydrolysed with 6 M HC1. 2.10. Superoxide Dismutase (SOD) {#sec2.10} -------------------------------- Superoxide dismutase (SOD) activity was kinetically evaluated according to the inhibition of the reaction of superoxide radical with pyrogallol. The SOD activity was determined by measuring the oxidized pyrogallol formation rate. 2.11. Statistical Analyses {#sec2.11} -------------------------- Data was tested for normality using the Shapiro--Wilk test prior to the analyses. All values at each sampling time were averaged and compared between sessions using one-way ANOVA, followed by Tukey\'s posttest for parametric values and the Kruskal--Wallis test, followed by Dunn\'s test for nonparametric values. The area under the curve (AUC) was used as temporal analysis of the variables. For all analyses, the significance level was set at *P* \< 0.05. Results are shown as means ± standard error mean (SEM). 3. Results {#sec3} ========== The habitual physical activity program and dietary intake among all subjects was similar during the experimental period. As expected, no side effects were reported, and salivary nitrite concentration increased after 60 min of nitrate supplementation. This increase remained 150 min postsupplementation (Figures [2(a)](#fig2){ref-type="fig"} and [2(c)](#fig2){ref-type="fig"}). In addition, 5 days of nitrate supplementation increased nitrite salivary in NO, as compared to the PL group in preexercise. No difference in nitrite concentration was observed postexercise (60′--150′) compared to preexercise (0′) in the NO group ([Figure 2(c)](#fig2){ref-type="fig"}). AUC in postexercise was higher in NO than in PL in both the acute and 5-day supplementations (Figures [2(b)](#fig2){ref-type="fig"} and [2(d)](#fig2){ref-type="fig"}). Systolic and diastolic blood pressure profiles are shown in [Figure 3](#fig3){ref-type="fig"}. The AUC of systolic blood pressure in postexercise decreased (*P* \< 0.05) only after 5 days of nitrate supplementation in NO compared to the PL group ([Figure 3(f)](#fig3){ref-type="fig"}) but did not change after acute supplementation (AC). No differences were observed in diastolic blood pressure after acute or 5-day supplementations (Figures [3(g)](#fig3){ref-type="fig"}--[3(h)](#fig3){ref-type="fig"}). It was also found that salivary NO^2−^ was higher (*P* \< 0.05) in the baseline at time 0 after FD (data not shown in [Figure 3(c)](#fig3){ref-type="fig"}). Salivary alpha-amylase content increased after exercise in both acute and 5-day supplementations as compared to baseline (0′). No difference was observed in the sAA content after 30 min of cycle ergometer exercise in both the NO and the placebo groups (Figures [4(a)](#fig4){ref-type="fig"} and [4(b)](#fig4){ref-type="fig"}). In addition, salivary lactate increased after 30 min (90′ times) of cycle ergometer exercise, returning to baseline after 15′ of rest ([Table 1](#tab1){ref-type="table"}). Regarding the antioxidant potential of nitrate supplementation, the AUC of salivary uric acid concentration increased after 5 days of supplementation in NO compared to the PL group ([Figure 5(d)](#fig5){ref-type="fig"}). In addition, 5 days of nitrate supplementation increased the AUC of salivary uric acid in NO compared to the PL group preexercise ([Figure 5(e)](#fig5){ref-type="fig"}). When the total antioxidant capacity was analysed, 5 days of nitrate supplementation increased the AUC in NO compared to the PL group after 30 min of exercise ([Figure 6(d)](#fig6){ref-type="fig"}). In addition, 5 days of nitrate supplementation decreased the AUC in NO compared to the PL group after exercise ([Figure 7(d)](#fig7){ref-type="fig"}). In the same way, 5 days of nitrate supplementation decreased the AUC of SOD activity in NO compared to the PL group ([Figure 8(d)](#fig8){ref-type="fig"}). 4. Discussion {#sec4} ============= The present study investigated the effect of nitrate supplementation on blood pressure responses and antioxidant status after 30 min of aerobic exercise on cycle ergometer performance in trained subjects. The major results showed that 5 days of nitrate supplementation decreased systolic blood pressure and improved antioxidant response after aerobic exercise performance, as compared to acute or placebo supplementation. As expected, nitrate supplementation increased the salivary nitrite concentration in both acute and 5-day supplementations. The daily nitrate dose used in this study corresponds to the amount normally found in 150 to 250 g of nitrate-rich vegetables, such as spinach, beetroot, or lettuce \[[@B25]\]. The intensity of exercise can be evaluated from an increase in salivary lactate \[[@B26]\], salivary alpha-amylase concentration, and salivary nitrite \[[@B3], [@B27]\]. Here, 50% maximum workload during 30 min of cycle ergometer exercise could increase both the salivary lactate and sAA content. However, compared to other experimental designs, the increase in salivary lactate and sAA found in our results was lower than that in exercises with 75% maximum workload \[[@B3]\]. Therefore, we consider that the 30 min of cycle ergometer exercise was of moderate intensity. In addition, our experimental group was composed of trained subjects, and after the first visit in our lab, these volunteers continued with the training load until returning to the lab. Nitrate supplementation may reduce blood pressure after exercise performance \[[@B4], [@B5], [@B28]\]. These findings corroborate with the results of this study, which showed a decrease in systolic blood pressure AUC after exercise performance. The increase in salivary nitrite is related to the nitrate-nitrite-NO pathway \[[@B29]\], which can improve vascular health \[[@B30]\]. This blood pressure-lowering effect can be attributed to NO synthesis \[[@B31]\], which plays an important role in vascular health. This finding indicates that 5 days of nitrate supplementation can improve blood pressure responses mediated by aerobic exercise performance. This is important data because most studies have shown only blood pressure responses after nitrate supplementation alone, or after exercise training, but not after 90 min of exercise performance, which is well-described in the literature as an important tool to control blood pressure and prevent peaks of blood pressure and cardiovascular events \[[@B5], [@B15]\]. Regarding the antioxidant role of nitrate supplementation, the results showed an increase in uric acid in the NO group after 5 days of nitrate supplementation. Uric acid is the most important antioxidant molecule in saliva \[[@B32], [@B33]\]. In the extracellular environment, urate can scavenge hydroxyl radicals, singlet oxygen, and peroxynitrite, especially when combined with ascorbic acid or thiols \[[@B34]\]. Similar to the increased salivary uric acid, these results showed an increase in total antioxidant capacity (FRAP) only after 5 days of nitrate supplementation. Approximately 60% of total antioxidant capacity measured with the FRAP test was from uric acid \[[@B35]\], indicating uric acid is an antioxidant agent. In addition, FRAP reflects the ability to keep a balance between oxidants and antioxidants or even the ability to repair damage caused when the production of oxidants is found in greater proportions \[[@B36]\]. Here, lipid peroxidation (TBARS) and SOD activity decreased after 30 min of cycle ergometer exercise only in the group that received 5 days of nitrate supplementation. One possible explanation to improve lipid peroxidation is the increase in NO \[[@B37]\] from the nitrate-nitrite-NO pathway \[[@B1], [@B2]\]. Nitric oxide acts as a potent inhibitor of the lipid peroxidation chain reaction by scavenging propagatory lipid peroxyl radicals \[[@B12], [@B38], [@B39]\]. In addition, in the acute group, the nitrate supplementation did not attenuate oxidative stress. These results corroborate an earlier study, which showed that acute dietary nitrate supplementation did not attenuate oxidative stress \[[@B18]\]. Therefore, chronic nitrate supplementation can be an alternative method for improving oxidative stress after physical exercise. As limitations of this study, we considered the absence of measures of vascular function and only measurements of antioxidant biomarkers in saliva. Also, we showed that total antioxidant capacity data is in consonance with uric acid levels, the main antioxidant in saliva, although only the FRAP method was used to evaluate the total antioxidant capacity. In addition, biomarker analysis in salivary fluid has practical appeal, as reported in many studies revealing systemic biomarkers of health status, and those specifically related to oxidative stress and antioxidant status are still needed. 5. Conclusion {#sec5} ============= Taken together, this study showed that 5 days of sodium nitrate supplementation reduced systolic blood pressure and improved antioxidant responses after moderate-intensity aerobic exercise in healthy and physically active men. We appreciate the significant contribution of Dr. Olga Lucia Bocanegra and Silvio Santos Soares in this study. The authors gratefully acknowledge the financial support of Foundation for Research Support of the Minas Gerais State (FAPEMIG, APQ-04655-10). EFM and ABJ received postgraduate fellowships from Coordination for the Improvement of Higher Education Personnel (CAPES), LGP received postdoctoral fellowships from National Postdoctoral Program (PNPD/CAPES, 1743115), and FSE is a grant recipient of the National Council for Scientific and Technological Development (CNPq, 308965/2015-9). Data Availability ================= The results\' data used to support the findings of this study are restricted by the local ethics board Comitê de Ética em Pesquisa em Seres Humanos from Federal University of Uberlandia in order to protect patient privacy. Data are available from Foued Salmen Espindola, Federal University of Uberlandia, Institute of Genetics and Biochemistry (INGEB), Rua Acre, S/N, Bloco 2E, Sala 237, Campus Umuruama, CEP 38400-902, Uberlandia, MG, Brazil (e-mail: <foued@ufu.br>, for researchers who meet the criteria for access to confidential data). Disclosure ========== With their contribution, some data of this study were presented as an abstract in the 23rd Congress of the International Union for Biochemistry and Molecular Biology/44th 66 Annual Meeting of the Brazilian Society for Biochemistry and Molecular Biology, Foz do Iguaçu, PR, Brazil, August 24th to 28th, 2015. Some results of this study were already presented and published in the book of abstract of the 23rd Congress of the International Union for Biochemistry and Molecular Biology/44th Annual Meeting of the Brazilian Society for Biochemistry and Molecular Biology, Foz do Iguaçu, PR, Brazil, August 24th to 28th, 2015. Conflicts of Interest ===================== The authors declare that there is no conflict of interest regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} ###### Salivary lactate concentration (*μ*m/L) in rest and post 30 min of cycle ergometer exercise. ------------- ------------- ---------------- ------------- ------------- ------------- ------ NO group 0′ 60′ 90′ 105′ 120′ 150′ 0.34 ± 0.18 0.39 ± 0.11 0.69 ± 0.29^∗^ 0.53 ± 0.24 0.38 ± 0.18 0.43 ± 0.18 PL group 0′ 60′ 90′ 105′ 120′ 150′ 0.34 ± 0.23 0.37 ± 0.24 0.66 ± 0.31^∗^ 0.32 ± 0.24 0.40 ± 0.23 0.32 ± 0.20 ------------- ------------- ---------------- ------------- ------------- ------------- ------ ^∗^Statistical difference from rest (*P* \< 0.05). [^1]: Academic Editor: David Nieman

1. Introduction {#sec1} =============== Cervical cancer is the third most common cancer of women worldwide, after breast and colon cancer \[[@B1]\]. The World Health Organization (WHO) estimates that approximately 528,000 women are diagnosed with cervical cancer each year and that the disease results in approximately 266,000 deaths annually. Known risk factors for cervical cancer include human papillomavirus (HPV) infection, promiscuous intercourse, sexually transmitted disease infection, long-term hormonal contraceptive use, and smoking \[[@B2]\]. Of these risk factors, HPV infection is the predominant cause of cervical cancer. Thus, effective HPV screening is essential to facilitate accurate and rapid precancer diagnosis and is currently used along with cytological and histological examination worldwide \[[@B3]--[@B6]\]. Current "gold-standard" methods for precancer diagnosis are cytological and histological examinations. For cytological examination, exfoliated cervical cells are collected by swabbing the cervix (as part of a "Pap-smear" test), before being placed onto a slide to be inspected for abnormalities. In case of histological examination, it is diagnosed via a microscopic examination of a stained tissue biopsy \[[@B7]\]. Both of these precancer diagnosis methods are affected by sensitivity of test itself and examiners\' subjectivity. Recently, a molecular method via the identification of HPV nucleic-acid sequences was developed for use in conjunction with standard cytological and histological examination techniques, commonly \[[@B8]\]. Currently, commonly used diagnostic markers include the HPV-related proteins L1, E6, and E7. Of these, L1 is a major viral capsid protein that is produced in the cytoplasm, before being translocated into the nucleus of intermediate and superficial squamous epithelial cells, as previously visualized using immunochemical staining. E6 and E7 are primary HPV oncoproteins with numerous cellular targets including p53, and the retinoblastoma tumor suppression protein (pRB). E6 inhibits p53 to prevent apoptosis, whereas E7 is the primary transforming protein, and inhibits pRB to regulate cell-cycle arrest \[[@B9], [@B10]\]. In the previous study, we assessed the efficacy of cervical cancer diagnosis via screening for the mRNA expression of commonly used HPV markers *L1*, *E6*, and *E7*, along with the additional cancer markers human telomerase reverse transcriptase (*hTERT*) and *Ki67.* hTERT represents the catalytic subunit of telomerase. Telomeres are highly specialized structures that are located at chromosome ends and are known to be essential for genome stability \[[@B11]\]. In fact, telomere dysfunction and telomerase activation have been previously implicated in human cancer progression \[[@B12]\]. The expression level of *hTERT* is known to be the rate-limiting factor for human telomerase activity, and as such, likely a more sensitive indicator of telomerase function and activity than the expression levels of other telomerase subunits that are constitutively expressed in both normal and cancer cells \[[@B13]\]. On the other hand, Ki67 is a nuclear antigen expressed during all active phases of the cell cycle (i.e., G1, S, G2, and M) except G0, and thus, its expression level can be used to determine the cell proliferation status and to predict tumor development \[[@B14]\]. Screening of these diagnostic markers may also be of use in assessing the progression of cervical cancer past the midstage, as demonstrated by a previously conducted prospective study of their expression in formalin-fixed paraffin-embedded (FFPE) clinical histological samples \[[@B15]\]. Cytological samples actually used in clinical screening test were also conducted. However, severe precancerous lesion samples were not enough to conduct statistical analysis. Especially it takes a long time to collect high-grade squamous intraepithelial lesion (HSIL) samples which are the most severe precancerous lesion of cancer. Therefore, in the present study, microscope slides were attempted as samples. They are sealed with Canada balsam in a vacuum state which induce longer storage period relatively. And they could be collected quickly and easily through documented clinical information. In the present study, HPV and cancer markers mentioned above were analyzed with 110 HSIL and 50 normal microscope slides. 2. Materials and Methods {#sec2} ======================== 2.1. Clinical Samples {#sec2.1} --------------------- A total of 110 and 50 slides with exfoliated cervical-cell samples mounted with Canada balsam (Merck, Frankfurter, Germany) were retrospectively obtained from patients diagnosed to HSIL and normal, respectively, between 2000 and 2004, from the Department of Pathology, Yonsei University Wonju Severance Christian Hospital, Wonju, Republic of Korea. Strictly speaking, normal means negative for intraepithelial lesions or malignancy (NILM) in this context. To reduce interpretive diagnostic error, we only utilized HSIL specimen confirmed with cervical intraepithelial neoplasia grade 2 or worse (CIN2+). All subjects provided written informed consent for their participation in the study, which was approved by the Institutional Ethics Committee at Yonsei University Wonju College of Medicine (approval no. CR315052). 2.2. Histological and Cytological Diagnosis {#sec2.2} ------------------------------------------- Clinical diagnosis was predominantly determined cytologically using the 2001 Bethesda System terminology; however, cases with available tissue biopsies were also histologically reviewed. 2.3. Slide Preparation and Total RNA Extraction {#sec2.3} ----------------------------------------------- Slides with exfoliated cervical cells (microscope slides) were used for total RNA extraction. Slides were initially placed in coplin jars with xylene (Duksan, Ansan, Republic of Korea) for 4 days to remove their cover clips (which were mounted with Canada balsam). They were then dried (5 min) and placed into a six-well culture plate (SPL Life Sciences Co., Pocheon, Republic of Korea). The Isol-RNA Lysis Reagent (1 mL; 5 Prime, Austin, TX) was added onto each slide, and a 1 mL exfoliated cervical-cell sample was then collected from each slide via scraping (twice) with an autoclaved slide glass. Each collected exfoliated cervical-cell sample was transferred to an RNase-free 1.7 mL tube, lysed via vortexing/repeated pipetting, and allowed to incubate in the reagent (room temperature, 5 min). After the addition of 200 *μ*L of chloroform, the tube was shaken vigorously, incubated (room temperature, 3 min), and then centrifuged (12,000 g, 15 min). The resultant aqueous layer was transferred to a new tube and mixed with an equal volume of isopropanol by inverting the tube. The mixture was incubated (25°C, 10 min) and then centrifuged (12,000 g, 10 min) before the resulting supernatant was removed, and 1 mL of 75% ethanol was added to the remaining pellet. After mixing via tube inversion, the mixture was centrifuged (7,500 g, 5 min), and the supernatant subsequently removed. The remaining RNA pellet was dried and eluted in 25 *μ*L of diethylpyrocarbonate- (DEPC-) treated water (Intron Biotechnology, Seoul, Republic of Korea). The purity and concentration of the extracted total RNA was determined by measuring its absorbance at 260 and 280 nm using an Infinite 200 plate reader (Tecan, Salzburg, Austria). The isolated total RNA was finally stored at −70°C until use. Note that all preparation and handling of total RNA were performed in a laminar flow hood, under RNase-free conditions. 2.4. cDNA Synthesis {#sec2.4} ------------------- Complementary DNA (cDNA) was synthesized using an M-MLV Reverse Transcriptase kit (Invitrogen, Carlsbad, CA, USA) and random hexamers (Invitrogen), according to the manufacturer\'s recommendations. Briefly, 10 *μ*L of total RNA was added to a master mix containing 1 *μ*L of 10 mM dNTP mix (containing 10 mM each of dATP, dGTP, dCTP, and dTTP at a neutral pH), 0.25 *μ*g of random hexamers, and 1 *μ*L of DEPC-treated water in PCR tubes. The reaction mixture was incubated (65°C, 5 min) and then quickly chilled on ice. A mixture of 4 *μ*L of 5× First-Strand Buffer, 2 *μ*L of 0.1 M dithiothreitol (DTT), and 1 *μ*L of M-MLV reverse transcriptase (RT) was added to the reaction mixture, and the cDNA synthesis reaction then performed via cycling at 25°C for 10 min, 37°C for 50 min, and 70°C for 15 min. 2.5. HPV Genotyping Using PCR-REBA {#sec2.5} ---------------------------------- A REBA HPV-ID® PCR-REBA test (YD Diagnostic, Yongin, Republic of Korea), in which a "nested PCR" method was used to amplify target regions between MY11-MY9 and GP5-GP6 using two primer pairs, was used for HPV genotyping. PCR was performed using a 20 *μ*L reaction mixture (Genetbio, Daejeon, Republic of Korea) consisting of 2× master mix, 1× primer mixture, 3 *μ*L of sample DNA, and sterile deionized water (DW). This mixture was subjected to PCR cycling conditions comprising 94°C for 5 min (predenaturation), followed by 15 cycles of 94°C for 30 s (denaturation) and 55°C for 30 s (annealing), 45 cycles of 94°C for 30 s (denaturation) and 52°C for 30 s (annealing), and a final cycle of 72°C for 7 min (strand synthesis). The amplified biotin-labeled PCR products were then denatured (25°C, 5 min) in denaturation solution, diluted in 970 *μ*L of hybridization solution, applied to the REBA membrane strip in the blotting tray, and hybridized (50°C, 30 min) to the desired probes. The membrane strips were then washed twice with 1 mL of washing solution (50°C, 10 min, with gentle shaking), before being incubated (25°C, 30 min) with a streptavidin-alkaline phosphatase (AP) conjugate (Roche Diagnostics, Mannheim, Germany) diluted (1 : 2 000) in a conjugate diluent solution (CDS). After two final washes with 1 mL CDS (room temperature, 1 min), colorimetric hybridization signals were visualized via incubation with an NBT/BCIP solution (1 : 50 dilution, Roche Diagnostics) for sufficient time to detect the enzymatic conversion of the NBT/BCIP substrate to its colored form. The resulting band patterns were then read and interpreted. 2.6. Multiplex Quantitative Reverse-Transcriptase (RT-Q) PCR Assay {#sec2.6} ------------------------------------------------------------------ HPV *E6/E7*, *hTERT*, and *Ki67* mRNA expression in cervical specimens was assessed via a multiplex RT-qPCR TaqMan assay that was performed using the CervicGen HPV *E6/E7* and *hTERT-Ki67* mRNA RT-qDx assay kits (Optipharm, Osong, Republic of Korea) and the CFX-96 real-time PCR system (Bio-Rad, Hercules, CA, USA) for thermal cycling and fluorescence detection. Real-time PCR amplification of HPV *E6/E7* mRNA was performed in a reaction mix containing 10 *μ*L of 2× Thunderbird probe qPCR mix (Toyobo, Osaka, Japan), 5 *μ*L each of primer and TaqMan probe mixture, and 5 *μ*L of template cDNA. Real-time PCR amplification of *hTERT* and *Ki67* mRNA was performed in a reaction mix containing 10 *μ*L of 2× Thunderbird probe qPCR mix (Toyobo, Osaka, Japan), 3 *μ*L of primer and TaqMan probe mixture, 2 *μ*L of template cDNA, and sufficient DW to produce a final volume of 20 *μ*L. Positive and negative controls were included throughout the procedure, and likewise, no-template controls (containing sterile DW instead of template DNA) were amplified with each PCR. The utilized PCR cycling conditions comprised 95°C for 3 min, followed by 41 cycles of 95°C for 3 s, and 55°C for 30 s. Each mRNA expression level was quantified by determining the "cycle threshold" (CT), which is the number of PCR cycles required for the fluorescence to exceed a value significantly higher than the background fluorescence. To avoid the generation of false negative results due to mRNA degradation, the expression level of glyceraldehyde-3-phosphate dehydrogenase (*GAPDH*) was used as an internal control. The six samples of HSIL group and four samples of normal group did not show *GAPDH* value; therefore, they were excluded from the following experiments. In other words, researcher performed experiments with 104 HSIL and 46 normal samples. Target gene mRNA expression levels relative to *GAPDH* were automatically calculated according to the comparative C~t~ method, using CFX Manager v1.6 (Bio-Rad) or Genex (Bio-Rad) Software. Gene expression was assessed using the comparative C~t~ (ΔΔC~t~) method, in which mRNA expression levels are represented relative to the expression level of the reference gene. *hTERT* and *Ki67* expression levels in analyzed slide samples from patients without HSIL were considered to indicate the "baseline" expression level for each gene. 2.7. Statistical Analyses {#sec2.7} ------------------------- Statistical analyses were performed using GraphPad Prism software version 5.02 (GraphPad, La Jolla, CA, USA). Student\'s *t*-test, 95% confidence interval (CI), and ROC curve were used to assess the statistical significance of generated data. Cohen\'s kappa coefficient which measures agreement between two raters for qualitative items is also used. 3. Results {#sec3} ========== 3.1. Histological and Cytological Diagnosis of Clinical Specimens {#sec3.1} ----------------------------------------------------------------- All analyzed slide samples were confirmed to exhibit the precancerous condition, cervical intraepithelial neoplasia (CIN) III. Furthermore, histological and cytological methods were used to confirm a diagnosis of HSIL among the 110 relevant slides. Patients with HSIL (range 10-79 years-of-age) were found to be predominantly aged between 30 and 39 (36/104 patients, 34.62%) or 40 and 49 (35/110 patients, 33.65%) years. A lesser number of HSIL patients were aged between 20 and 29 (11/104 patients, 10.58%) or 50 and 59 (12/104 patients, 11.54%) years, and very few were aged less than 20 or greater than 59 years ([Table 1](#tab1){ref-type="table"}). 3.2. REBA Analysis of the HPV Infection Status of Analyzed HSIL Clinical Specimens {#sec3.2} ---------------------------------------------------------------------------------- Of the 104 slides with exfoliated HSIL cervical-cell samples, 83 (79.81%) were found to be infected with at least a single HPV genotype, including 26 (25%) that were infected with multiple (i.e., more than two) HPV genotypes. Among these 83 cases, 56 (53.85%) were determined to be infected with a high-risk (HR) HPV genotype, while a single case (0.96%) was shown to be infected with a low-risk (LR) HPV genotype ([Table 2](#tab2){ref-type="table"}). Among the 26 cases found to be infected with multiple HPV genotypes, 21 (20.19%) were shown to be infected with HR-HPV genotypes, five (4.81%) were determined to be infected with both HR- and LR-HPV genotypes, and no cases were found to be infected with LR-HPV genotypes only. 3.3. HPV Genotype Distribution in the Analyzed Clinical Samples {#sec3.3} --------------------------------------------------------------- As shown in the constructed cumulative graph ([Figure 1](#fig1){ref-type="fig"}), the most frequently detected HPV genotype in the HPV-positive exfoliated cervical-cell samples was HPV *16* (cumulative proportion 40.22%), followed by HPV *52* (55.43%), *58* (66.30%), *31* (77.17%), *18* (84.78%), *33* (90.22%), *35* (94.57%), *66* (97.83%), *84* (98.91%), and *45* (100%). 3.4. *hTERT* and *Ki67* mRNA Expression Levels as Determined by RT-qPCR {#sec3.4} ----------------------------------------------------------------------- We conducted RT-qPCR analyses of *hTERT* and *Ki67* expression levels in the 104 HSIL-diagnosed exfoliated cervical-cell samples. After excluding samples with a no *GAPDH* expression level, we determined that 98 (94.23%) and 49 (47.12%) of the 104 remaining HSIL samples were positive for *hTERT* and *Ki67* mRNA expression, respectively ([Figure 2](#fig2){ref-type="fig"}). 3.5. Coincident Expression of *hTERT* or *Ki67* with HPV *E6/E7* mRNA in the Analyzed HSIL Clinical Samples {#sec3.5} ----------------------------------------------------------------------------------------------------------- The results of the conducted RT-qPCR analyses showed that of the 104 (nonexcluded) HSIL samples, 80 (76.92%) were positive for both HPV *E6/E7* and *hTERT*, three (2.88%) were positive for HPV *E6/E7* only, 18 (17.31%) were positive for *hTERT* only, and three (2.88%) were positive for HPV *E6/E7*, but negative for *hTERT* ([Table 3](#tab3){ref-type="table"}). Conversely, 38 (36.54%) of the 104 samples were shown to be positive for both HPV *E6/E7* and *Ki67* expression, 45 (43.27%) were positive for HPV *E6/E7* expression only, 11 (10.58%) were positive for *Ki67* expression only, and ten (9.62%) were positive for HPV *E6/E7*, but negative for *hTERT* expression ([Table 4](#tab4){ref-type="table"}). 3.6. Combination of HPV *E6/E7*, *hTERT*, and *Ki67* mRNA Expression {#sec3.6} -------------------------------------------------------------------- The positivity rates were 90.38% (94/104) for a combination of *HPV E6/E7* and *Ki67* mRNA expressions, 96.15% (100/104) for *hTERT* and *Ki67* mRNA expression, 97.12 (101/104) for *HPV E6/E7* and *hTERT* mRNA expressions, and 98.08% (102/104) for the HPV *E6/E7*, *hTERT*, and *Ki67* mRNA expressions ([Figure 3](#fig3){ref-type="fig"}). 4. Discussion {#sec4} ============= The previous study was conducted with FFPE clinical tissue samples \[[@B16]\]. It was appropriate to determine the availability of the biomarker by tissue samples because tissue samples collected precisely only cancerous region through microscopy and IHC test. We also confirmed that hTERT and Ki67 mRNA expression could be complementary biomarkers in diagnosing cervical lesions with histological samples. In the present study, microscope slides with exfoliated cervical cell samples were collected from subjects diagnosed as either healthy or with HSIL and screened for HPV genotypes, assessing the usefulness of *hTERT* and *Ki67* expression as diagnostic markers of cervical cancer. Exfoliated cervical cells before being placed onto a slide are currently used for screening test specimen because they accompanied with less invasive and less labor-intensive procedure than other tests \[[@B17], [@B18]\]. They are exposed to the air and need refrigeration condition, so their storage period is limited. Above all, precancerous lesion samples are relatively infrequent compared to normal samples, and there is no idea how much time needed to collect enough samples for statistical analysis. In fact, HSIL samples, which are the most severe precancerous lesion, were collected over three years in Korea and China, but there were about 50 samples \[[@B16], [@B19]\]. In contrast, microscope slides are sealed with Canada balsam in a vacuum state, so their storage period is longer relatively. In fact, there was no difference in the housekeeping gene expression between samples by 5 years. And the slides were always prepared for cytology test as routine screening test. These factors enable to collect large number of HSIL samples over 100 easily and immediately. Thus, the slides are adequate specimen for retrospective cohort study. There were several retrospective studies with the slides; however, they are primarily limited to inspect staining of appearance so far \[[@B20]--[@B23]\]. For the first time, molecular tests were performed with microscope slides in this study. To verify nucleic acid degradation and variation, the assays were performed with endogenous control genes during the experiment. The results showed that HR-HPV infection is more closely associated with cervical cancer progression than LR-HPV both in the context of a single or of multiple HPV infections ([Table 1](#tab1){ref-type="table"}). In addition, the most commonly detected HPV genotype among the HPV-positive HSIL specimens was HPV *16,* and notably, detection rate of HPV *18* of slides is lower than that of histological samples. Cervical cancer oncogenesis is initiated and mediated via the upregulation of the HPV oncoproteins E6 and E7, such that the overexpression of *E6/E7* mRNA transcripts has been shown to be associated with a significantly increased risk of both precancerous group (CIN) and of cervical cancer \[[@B24]\]. The hypothesis that *E6/E7* expression levels may be specific and effective predictors of cervical cancer risk was supported by the results of the present study, which showed the sensitivity of the utilized *E6/E7* mRNA RT-qPCR assay to 79.81% in the 104 analyzed exfoliated cervical-cell samples. It indicate cervical cancer occurrence could be affected other factors not only HPV. Therefore, *hTERT* and *Ki67* confirmed by tissue samples \[[@B16]\] were also applied. The sensitivity of *hTERT* mRNA RT-qPCR screening of the 104 clinical samples was 94.23%, whereas that of *Ki67* screening was only 47.12%. Interestingly, a previous study demonstrated *hTERT* mRNA expression to be higher in cytological than in tissue samples from high-grade cervical lesions, while conversely, *Ki67* mRNA expression was found to be higher in tissue than cytological samples from the high-grade cervical lesions \[[@B20]\]. The mechanism underlying this observed discrepancy between marker genes expression in cytological versus histological samples remains to be elucidated. While *hTERT* and *Ki67* mRNA expression was only detected in 94.23% and 47.12% of the analyzed cytology samples, respectively, combined screening for HPV *E6/E7* and *Ki67*, HPV *hTERT* and *Ki67*, and HPV *E6/E7* and *hTERT* mRNA expression identified 90.38%, 96.15%, and 97.12% of samples, respectively ([Figure 3](#fig3){ref-type="fig"}). Furthermore, coincident screening for HPV *E6/E7*, *hTERT*, and *Ki67* mRNA expression resulted in an RT-qPCR assay sensitivity of 98.08%, suggesting this as a promising combination of markers for the diagnosis of HSIL. The present study demonstrates the validity of using the non-HPV markers and of analyzing microscope slides for the first time to identify novel diagnostic precancer and cancer biomarkers. Further study is required to assess the suitability of their use as diagnostic markers for low-grade squamous intraepithelial lesions (LSIL) and/or as indicators of the progression of cervical lesions. This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF), funded by the Ministry of Science, ICT, and Future Planning (2015R1A2A2A04004455). This work was also supported (in part) by the Yonsei University Research Fund of 2018. Data Availability ================= The datasets generated and analyzed during the current study are not publicly available due to patent application but are available from the corresponding author on reasonable request. Ethical Approval ================ All subjects provided written informed consent for their participation in the study, which was approved by the Institutional Ethics Committee at Yonsei University Wonju College of Medicine (approval no. CR315052). Consent ======= This is to state that I give full permission for the publication. Conflicts of Interest ===================== The authors declare that they have no competing interests, and all authors should confirm its accuracy. Authors\' Contributions ======================= Jemberu Taye collected the patient clinical data and specimens. Kwangmin Yu performed the experiment. Geehyuk Kim analyzed and interpreted the patient data. All authors read and approved the final manuscript. {#fig1} {#fig2} {#fig3} ###### Cytological diagnosis of clinical specimens with respect to patient age. Patient age (years) Cytological diagnosis *N* (%) --------------------- ------------------------------- ------------ ------------ 10-19 1 (2.17) 1 (0.96) 2 (1.33) 20-29 3 (6.52) 11 (10.58) 18 (9.33) 30-39 17 (36.96) 36 (34.62) 55 (35.33) 40-49 14 (30.43) 35 (33.65) 51 (32.67) 50-59 8 (17.39) 12 (11.54) 22 (13.33) 60-69 2 (4.35) 8 (7.69) 10 (6.67) 70-79 1 (2.17) 1 (0.96) 2 (1.33) Total 46 (100) 104 (100) 150 (100) HSIL: high-grade squamous intraepithelial lesion. ###### Distribution of analyzed HSIL specimens with respect to HPV infection type (as assessed via REBA). ----------------------------------------------------------------------------------- HPV infection type *N* (%) -------------------------- ---------------------- ---------------- ---------------- HPV-positive HSIL\ Single HPV infection HR-HPV 56/104 (53.85) *N* = 83/104\ (79.81) *N* = 57/104 (54.81) LR-HPV 1/104 (0.96) Multiple HPV infections\ HR-HPV 21/104 (20.19) *N* = 26/104 (25) LR-HPV 0/104 (0) HR- & LR-HPV infections 5/104 (4.81) HPV-negative HSIL 21/104 (20.19) ----------------------------------------------------------------------------------- HPV: human papillomavirus; REBA: reverse-blot hybridization assay; HSIL: high-grade squamous intraepithelial lesion; HR-HPV: high-risk HPV; LR-HPV: low-risk HPV. ###### RT-qPCR analysis of HPV *E6/E7* and *hTERT* mRNA expression in HSIL versus normal clinical specimens. Cytological diagnosis HPV *E6/E7*-positive cases HPV *E6/E7*-negative cases ----------------------- ---------------------------- ---------------------------- ---------------- --------------- HSIL 80/104 (76.92) 3/104 (2.88) 18/104 (17.31) 3/104 (2.88) Normal 0/46 (0) 2/46 (4.35) 0/46 (0) 44/46 (95.65) HPV: human papillomavirus; RT-qPCR: quantitative reverse-transcriptase polymerase chain reaction; HSIL: high-grade squamous intraepithelial lesion. ###### RT-qPCR analysis of HPV *E6/E7* and *Ki67* mRNA expression in HSIL versus normal clinical specimens. Cytological diagnosis HPV *E6/E7*-positive cases HPV *E6/E7*-negative cases ----------------------- ---------------------------- ---------------------------- ---------------- --------------- HSIL 38/104 (36.54) 45/104 (43.27) 11/104 (10.58) 10/104 (9.62) Normal 0/46 (0) 2/46 (4.35) 1/46 (2.17) 43/46 (93.48) HPV: human papillomavirus; RT-qPCR: quantitative reverse-transcriptase polymerase chain reaction; HSIL: high-grade squamous intraepithelial lesion. [^1]: Academic Editor: Madhyastha Harishkumar

Introduction {#H1-1-ZOI190360} ============ Pancreatic cancer is among the deadliest malignant neoplasms in the United States. It is the fourth leading cause of cancer-related deaths, with 56 770 new cases estimated for 2019.^[@zoi190360r1]^ For early-stage pancreatic cancer, surgery offers patients the best chance of cure, but outcomes remain poor. Despite primary surgery with curative intent, 73% of patients may develop local failure.^[@zoi190360r2]^ Multiple clinical trials previously investigated the clinical outcomes of various adjuvant therapy regimens.^[@zoi190360r3],[@zoi190360r4],[@zoi190360r5],[@zoi190360r6],[@zoi190360r7],[@zoi190360r8],[@zoi190360r9],[@zoi190360r10]^ A 2008 study sponsored by the Radiation Therapy Oncology Group (RTOG)^[@zoi190360r6]^ suggested the combination of chemotherapy and chemoradiation for select patients, while trials in 2017^[@zoi190360r7]^ and 2018^[@zoi190360r9]^ established the current adjuvant chemotherapy regimens as the standard of care. A current trial, RTOG 0848 (ClinicalTrials.gov identifier: [NCT01013649](https://clinicaltrials.gov/ct2/show/NCT01013649)) is examining the role of sequential chemoradiation following adjuvant chemotherapy. Delays in the initiation of adjuvant treatment or total duration of treatment time have been associated with worse survival outcomes, seen in a wide range of cancers including colorectal,^[@zoi190360r11]^ head and neck,^[@zoi190360r12]^ cervical,^[@zoi190360r13]^ and breast^[@zoi190360r14]^ cancer. While the benefit of adjuvant therapy to patients with resected pancreatic cancer is accepted, its optimal timing after surgery remains under investigation. Prior studies suggested no benefit to early initiation of adjuvant therapy but only compared survival between arbitrarily assigned time periods.^[@zoi190360r15],[@zoi190360r16],[@zoi190360r17],[@zoi190360r18],[@zoi190360r19]^ This National Cancer Database (NCDB) study compares the outcomes of patients who received adjuvant chemotherapy or chemoradiation at various time intervals, which were defined by a Cox model with restricted cubic splines (RCS). Methods {#H1-2-ZOI190360} ======= Patient Population {#H2-1-ZOI190360} ------------------ Patients with pancreatic adenocarcinoma diagnosed from 2004 to 2015 were identified using the NCDB registry. The NCDB is a national cancer database that collects information on approximately 70% of new cancer diagnoses in the United States.^[@zoi190360r20]^ The NCDB provides access to deidentified data sets from Commission on Cancer--accredited programs through online application. Informed consent for this study has been waived because NCDB data are deidentified. The Roswell Park Comprehensive Cancer Center institutional review board exempted our study from institutional review board review. This report follows the Strengthening the Reporting of Observational Studies in Epidemiology ([STROBE](http://www.equator-network.org/reporting-guidelines/strobe/)) reporting guideline. Our patient selection criteria are detailed in [Figure 1](#zoi190360f1){ref-type="fig"}. Our initial query resulted in patients with stage I to II, clinical T1-3N0-1M0 pancreatic adenocarcinoma who had been treated with curative-intent resection alone or resection followed by adjuvant chemotherapy or chemoradiation. American Joint Committee on Cancer sixth and seventh editions were used to identify stage I to II disease from 2004 to 2015. {#zoi190360f1} Whipple and Whipple-variant surgical procedures were previously described.^[@zoi190360r21]^ Adjuvant therapy was defined as adjuvant chemotherapy alone or chemotherapy with radiation. Those with adjuvant chemotherapy more than 180 days before adjuvant radiation therapy or with adjuvant radiation therapy more than 30 days before adjuvant chemotherapy were excluded.^[@zoi190360r22]^ Patients with adjuvant therapy more than 360 days after resection were also excluded. Surgical margin was divided into either negative (ie, R0) or positive (ie, R1, R2, or positive margin not otherwise specified). Patients were stratified by age (ie, \<67 years or ≥67 years), tumor size (ie, \<3.1 cm or ≥3.1 cm), time from diagnosis to surgery (ie, \<17 days or ≥17 days), and duration of postoperative inpatient admission (ie, \<1 week or ≥1 week) based on median values. The cutoff value of 98 U/mL for cancer antigen (CA) 19-9 levels was predetermined in the NCDB. All missing values were coded as unknown. To address the immortal time bias, patients with postdiagnosis survival of less than 3 months were excluded as a conditional landmark.^[@zoi190360r23]^ Pertinent prognostic variables, such as performance status, type and duration of chemotherapy, toxicity, and tumor recurrence outcomes, are unavailable in the NCDB. The primary end point was overall survival (OS), defined as time from diagnosis to the last follow-up or death. Statistical Analysis {#H2-2-ZOI190360} -------------------- To model the association of time to adjuvant therapy with survival, a multivariable Cox model with RCS was used, as previously shown in other disease sites.^[@zoi190360r24],[@zoi190360r25],[@zoi190360r26]^ Restricted cubic splines is a smooth, piecewise polynomial function that evaluates the association of a variable with an outcome without assuming any association a priori.^[@zoi190360r27],[@zoi190360r28]^ The Cox model with RCS was constructed using 4 knots at the 5th, 35th, 65th, and 95th percentiles, and the 4 knots were determined based on the lowest Akaike information criterion.^[@zoi190360r27],[@zoi190360r29]^ The model was adjusted for facility type, age, sex, race/ethnicity, insurance, income, residential setting, Charlson/Deyo comorbidity score (CDS), year of diagnosis, primary tumor location within pancreas, tumor grade, tumor size, pathologic T and N stages, CA 19-9 level, surgery type, surgical margin, chemotherapy, radiation therapy, unplanned readmission within 30 days after surgery, duration of inpatient stay after surgery, and time from diagnosis to surgery. The model-derived thresholds were determined based on those at which the hazard ratio (HR) was the smallest with the 95% CIs below an HR of 1.00. These thresholds were used to divide the adjuvant therapy cohort into early, reference, and late interval groups, with the reference interval group having the lowest mortality. Kaplan-Meier method and log-rank tests were used for OS analysis. Fisher exact test and Mann-Whitney *U* test were used to compare categorical and continuous variables, respectively. Logistic regression univariable analysis and multivariable analysis (MVA) were used to identify potential variables that predicted the delayed use of adjuvant therapy. Cox proportional hazard univariable analysis and MVA were used to identify variables that predicted survival. The MVA was initially built based on all statistically significant variables from the univariable analysis and was finalized using a backward stepwise elimination. Possible interactions between the treatment and other variables were evaluated using Cox MVA by adding interaction terms.^[@zoi190360r30]^ When the interaction terms were statistically significant, the final Cox MVA model was reanalyzed for each subgroup of variables, and a corresponding forest plot was performed.^[@zoi190360r30]^ To address selection bias, propensity score matching was performed. The reference interval cohort was matched with early and late cohorts separately to examine the association of the timing of adjuvant therapy with survival outcomes. Matched pairs were established by matching baseline characteristics, including facility type, age, CDS, year of diagnosis, tumor grade, tumor size, pathologic T and N stages, CA 19-9 level, surgery type, surgical margin, chemotherapy, radiation therapy dose, time from diagnosis to surgery, duration of postoperative inpatient admission, unplanned readmission within 30 days after surgery, and any additional variable that was statistically significant in Cox MVA. Matching was performed based on nearest neighbor method in a 1:1 ratio without any replacement, with a caliper distance of 0.1 of the standard deviation of the logit of the propensity score.^[@zoi190360r31]^ The standardized difference of each variable was less than 0.1, suggesting the adequate match.^[@zoi190360r32]^ To examine the survival outcome of adjuvant therapy initiated more than 12 weeks after surgery compared with surgery alone, the Cox MVA, Kaplan-Meier method, and propensity score matching were repeated among such cohorts. All analyses were performed using R software version 3.5.1 (R Project for Statistical Computing). All *P* values were 2-sided, and *P* values less than .05 were considered statistically significant. Results {#H1-3-ZOI190360} ======= Patient Characteristics {#H2-3-ZOI190360} ----------------------- A total of 7548 patients (3770 men \[49.9%\]; median \[interquartile range (IQR)\] age, 67 \[59-74\] years) with resected clinical stage I to II, T1-3N0-1M0 pancreatic adenocarcinoma met inclusion criteria. Of those, 5453 patients (72.2%) were treated with adjuvant therapy, and 2095 (27.8%) were not ([Figure 1](#zoi190360f1){ref-type="fig"}). Among patients who received adjuvant therapy, 2329 of 5453 patients (42.7%) received it at 40 to 60 days after surgery (eFigure 1 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). Median (IQR) follow-up was 38.6 (24.4-62.0) months (eTable 1 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). Adjuvant Therapy Timing {#H2-4-ZOI190360} ----------------------- The Cox model with RCS showed the interval with the lowest mortality to be from 28 to 59 days after surgery ([Figure 2](#zoi190360f2){ref-type="fig"}). Based on these model-determined thresholds, the adjuvant therapy cohort was stratified into 3 intervals: early (n = 269, \<28 days), reference (n = 3048, 28-59 days), and late (n = 2136, \>59 days) interval cohorts. Early and late cohorts received adjuvant therapy before and after the reference interval, respectively. {#zoi190360f2} On Cox MVA for adjuvant therapy cohorts (eTable 2 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}), early and late adjuvant therapy cohorts were associated with worse mortality (early: HR, 1.17; 95% CI, 1.02-1.35; *P* = .03; late: HR, 1.09; 95% CI, 1.02-1.17; *P* = .008). After Cox MVA, treatment interactions favored the reference interval for OS compared with the early cohort in patients with worse comorbidities and with unplanned readmission within 30 days after surgery (eFigure 2 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). Compared with the late cohort, the reference interval was favored for OS in patients with smaller tumors (eFigure 3 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). No treatment interactions were observed in other variables, including age, CDS, pathologic T stage, pathologic N stage, surgical margin, postoperative inpatient admission, unplanned readmission within 30 days after surgery, and time from diagnosis to surgery. On logistic MVA for initiating adjuvant therapies more than 59 days postoperatively, patients with primary tumor at pancreatic body and tail, multiagent chemotherapy, and radiation therapy were less likely to receive delayed adjuvant therapy (eTable 3 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). Patients with older age, black race, lower income, postoperative inpatient admission longer than a week, and unplanned readmission within 30 days after surgery were more likely to have delayed initiation of adjuvant therapy. A total of 268 and 2042 propensity-matched pairs were constructed using early and late cohorts in comparison with the reference interval cohort. All variables were well balanced among these cohorts (eTable 4 in the [Supplement](#note-ZOI190360-1-s){ref-type="supplementary-material"}). The overall median (IQR) follow-up was 45.6 (25.7-71.6) months for the early cohort and 39.3 (24.9-69.2) months for the reference cohort. The median (IQR) OS was 20.6 (12.2-37.8) months for the early cohort and 23.1 (14.8-38.3) months for the reference cohort ([Figure 3](#zoi190360f3){ref-type="fig"}). Overall survival at 2 years was 45.1% (95% CI, 39.5%-51.6%) and 52.5% (95% CI, 46.7%-59.0%) for the early and reference cohorts, respectively (*P* = .02). The overall median (IQR) follow-up was 38.9 (24.8-58.8) months for the late cohort and 36.9 (24.1-58.3) months for the reference cohort. The median (IQR) OS was 20.4 (12.8-34.7) months for the late cohort and 22.4 (13.6-36.3) months for the reference cohort ([Figure 4](#zoi190360f4){ref-type="fig"}). Overall survival at 2 years was 45.4% (95% CI, 43.3%-47.7%) and 51.3% (95% CI, 49.1%-53.6%) for the late and reference cohorts, respectively (*P* = .01). {#zoi190360f3} {#zoi190360f4} Adjuvant Therapy More Than 12 Weeks After Surgery vs Surgery Alone {#H2-5-ZOI190360} ------------------------------------------------------------------ On Cox MVA (data available on request), adjuvant therapy more than 12 weeks after surgery (HR, 0.75; 95% CI, 0.66-0.85; *P* \< .001) was associated with improved survival compared with surgery alone. After Cox MVA, treatment interactions were observed in a node-positive disease subgroup, favoring the adjuvant therapy for OS compared with surgery alone (data available on request). No treatment interactions were seen in other variables, including age, CDS, tumor size, pathologic T stages, surgical margin, duration of postoperative inpatient admission, unplanned readmission within 30 days after surgery, and time from diagnosis to surgery. A total of 655 propensity-matched pairs were constructed for adjuvant therapy initiated more than 12 weeks after surgery vs surgery alone. All variables were well balanced between these cohorts (data available on request). The overall median (IQR) follow-up was 41.9 (26.2-64.5) months for the adjuvant therapy cohort and 33.6 (21.5-56.3) months for the surgery alone cohort. The median OS was 21.1 (13.1-35.9) months for the adjuvant therapy cohort and 15.5 (7.4-29.3) months for the surgery alone cohort. Overall survival at 2 years was 47.2% (95% CI, 43.5%-51.3%) and 38.0% (95% CI, 34.4%-42.0%) for the adjuvant therapy and surgery alone cohorts, respectively (*P* \< .001). Discussion {#H1-4-ZOI190360} ========== To our knowledge, this is the first study using a multi-institutional national registry to evaluate the timing of adjuvant therapy in resected pancreatic adenocarcinoma across specific time intervals with the reference interval defined by a Cox model with RCS. The results of our study differ from previous reports that found no survival difference when adjuvant therapy was delayed.^[@zoi190360r15],[@zoi190360r16],[@zoi190360r17],[@zoi190360r18],[@zoi190360r19]^ By examining survival of patients initiating adjuvant therapy before and after set time periods, such as 8 or 12 weeks, these studies may have missed the optimal window for identifying survival benefits. Another factor that may account for this difference is our study included both adjuvant chemotherapy and chemoradiation, while several prior reports focused on adjuvant chemotherapy alone.^[@zoi190360r15],[@zoi190360r16],[@zoi190360r17],[@zoi190360r18]^ After adjusting for radiation therapy on Cox MVA and propensity-matched analysis in our study, the survival benefit of the reference interval remained statistically significant. It is important to note, however, that the European Study Group for Pancreatic Cancer^[@zoi190360r10]^ demonstrated a worse OS in patients treated with adjuvant radiation. Although interpretation of these results was limited by a lack of quality assurance of radiation plans, the increase in toxic effects and shorter time to recurrence with adjuvant radiation highlight the need for careful patient selection. In addition, since the publication of the RTOG study in 2008, 3 weeks of chemotherapy followed by chemoradiation and subsequent chemotherapy were recommended as an option for adjuvant therapy regimen.^[@zoi190360r6]^ Among patients with adjuvant chemoradiation in our study, the median (IQR) duration from the initiation of chemotherapy to the initiation of radiation was 31 (0-70) days. Because our study includes patients diagnosed starting in 2004, the heterogeneity of chemoradiation regimens may in part reflect the patterns of care in the United States. As a result, a subset of patients in our study may not have received the standard of care in accordance with national guidelines. In both MVA and propensity-matched pairs, our results suggest that patients who received adjuvant therapy more than 12 weeks after surgery had improved OS compared with patients who had surgery alone, supporting findings reported in previous studies.^[@zoi190360r15],[@zoi190360r16],[@zoi190360r17]^ Palacio et al^[@zoi190360r17]^ found this survival benefit to persist with a delay in the initiation of adjuvant chemotherapy for as long as 6 months. It is not clear from our NCDB analysis if delayed adjuvant therapy is palliative intent. In our study, patients with more comorbidities and unplanned postoperative readmission favored the reference interval compared with the early cohort. Such patients may have had more postoperative complications and likely needed additional time to recover rather than starting adjuvant therapy as early as possible.^[@zoi190360r33]^ Our logistic MVA similarly showed those with older age, prolonged postoperative inpatient admission, and unplanned postoperative readmission were more likely to have delayed adjuvant therapy. In our interaction term analysis, the reference interval was favored compared with the late cohort among patients with smaller tumor size. In addition, adjuvant therapy given more than 12 weeks after surgery was favored compared with surgery alone among patients with lymph node metastases. A 2018 study^[@zoi190360r34]^ described a subgroup of small pancreatic tumors with lymph node metastases that may have more aggressive cancer biology, which may benefit from either receiving adjuvant therapy promptly or avoiding surgery alone. In our Cox MVA results, being diagnosed in more recent years was associated with improved survival. This is likely in part owing to recent improvements in adjuvant chemotherapy regimens, such as FOLFIRINOX (folinic acid, fluorouracil, irinotecan, and oxaliplatin) and the combination of gemcitabine and capecitabine, with improved survival as shown in clinical trials.^[@zoi190360r7],[@zoi190360r9]^ Our study also showed that older age, lower income, rural residential setting, elevated CA 19-9 levels, and positive surgical margin were associated with worse mortality, and being African American was associated with delayed adjuvant therapy. These results are consistent with prior studies that found that socioeconomic status plays a role in treatment outcomes despite similar tumor characteristics.^[@zoi190360r35],[@zoi190360r36],[@zoi190360r37],[@zoi190360r38],[@zoi190360r39]^ Limitations {#H2-6-ZOI190360} ----------- This study has limitations. As a national registry--based study, it is limited by missing patient information, documentation error, and its retrospective nature. Detailed information regarding chemotherapy and chemoradiation regimens are unavailable in the NCDB. Importantly, the reason for early or late initiation of adjuvant therapy was unknown. Those who received adjuvant therapy late may have had postoperative complications or poor performance status, delaying the initiation of treatment. Toxicity profile and performance status are not included in the NCDB. To minimize selection bias, we performed propensity score matching for baseline characteristics, including unplanned postoperative readmissions and duration of postoperative inpatient admission, as proxy measures for postoperative performance status and complications.^[@zoi190360r40],[@zoi190360r41],[@zoi190360r42]^ While those with unplanned readmissions and prolonged postoperative inpatient admission were more likely to receive delayed adjuvant therapy based on our logistic MVA, they were not statistically significant for worse mortality in Cox MVA. Despite these limitations, the NCDB provides data on large numbers of patients with pancreatic cancer not otherwise available through single-institutional studies. Conclusions {#H1-5-ZOI190360} =========== This study used a multi-institutional national registry to evaluate the timing of adjuvant therapy in resected pancreatic adenocarcinoma across specific time intervals. To our knowledge, it is the first study to suggest that patients who commence adjuvant therapy within 28 to 59 days after primary surgical resection of pancreatic adenocarcinoma have improved survival outcomes compared with those who waited for more than 59 days. However, patients who recover slowly from surgery may still benefit from delayed adjuvant therapy initiated more than 12 weeks after surgery. Further studies examining the optimal treatment strategy following surgery in this challenging population are warranted. ###### **eFigure 1.** Histogram for Frequency of Patients Based on Days After Surgery **eFigure 2.** Forest Plot for Subgroup Analysis Between Early vs Reference Interval Cohorts **eFigure 3.** Forest Plot for Subgroup Analysis Between Late vs Reference Interval Cohorts **eTable 1.** Baseline Characteristics **eTable 2.**Cox MVA for Adjuvant Therapy Cohorts **eTable 3.** Logistic MVA for Initiating Adjuvant Therapies More Than 59 Days After Surgery **eTable 4.** Baseline Characteristics of Matched Cohorts for Adjuvant Therapy ###### Click here for additional data file.

Aux éditeurs du Journal Panafricain de Médecine {#S0001} =============================================== Dans un pays à faible revenu comme Madagascar, l'échographie est un moyen pas coûteux, non invasif, largement disponible et simple en imagerie rénale. Cette exploration tient une place non négligeable pour chercher diverses anomalies en pratique clinique. Mais peu d'études étaient effectuées à Madagascar concernant les dimensions rénales \[[@CIT0001]--[@CIT0004]\], alors que la connaissance des valeurs habituelles chez les malgaches aide à une décision médicale particulière, entre autre la prescription pour la biopsie rénale ou l\'initiation d\'un traitement spécifique. Les objectifs de ce travail étant de recueillir les dimensions des reins à partir des dossiers d\'un service clinique, de rapporter la corrélation entre ces dimensions et les paramètres démographique et anthropométrique des patients malgaches. Notre étude, rétrospective, réalisée au service de Néphrologie de l\'hôpital Joseph Raseta de Befelatanana, Antananarivo, contient 200 patients non diabétiques, sans maladie rénale chronique, recrutés entre janvier 2010 et décembre 2013. Les patients inclus avaient une échographie sans anomalie rénale ni des voies urinaires, un débit de filtration glomérulaire estimé normal. Le débit de filtration glomérulaire était estimé par la formule MDRD simplifiée. L'échographie était effectuée à Antananarivo. La créatininémie était dosée dans différents laboratoires à Antananarivo. Les dimensions rénales, les paramètres démographique et anthropométrique étaient recueillis dans les dossiers d\'hospitalisation des patients inclus. L'étude des corrélations entre les dimensions rénales et le genre, l′âge, la taille, le poids était effectué dans ce travail. Les données étaient exprimées sous forme de moyenne avec ecart-type. Les données de base étaient collectées en utilisant le logiciel Excel (Microsoft). Le test statistique était fait en utilisant le logiciel R. L\'analyse est significative quand p\< 0,05. Sur ces 4 ans d'étude, 200 patients étaient colligés, avec 95 hommes (47,5%) et 105 femmes (52,5%) donnant un sex-ratio de 0,9. L'âge médian de notre population d'étude était de 45 ans (17 à 91 ans). Le [Tableau 1](#T0001){ref-type="table"} rapporte les caractéristiques démographique et biométrique des patients inclus. Le rein droit avait une dimension légèrement plus petite que le rein gauche, soit sur la longueur (p\<0,00001), ou sur la largeur (p = 0,03) mais la différence n'était pas significative sur l'épaisseur (p = 0,06). Pour le rein droit, la longueur rénale avait une relation positive avec l'âge (p = 0,0016), la largeur avec le poids du sujet (p = 0,032). Pour le rein gauche, la longueur rénale avait une relation positive avec l'âge (p = 0,04), elle avait aussi un lien significatif avec le poids du sujet (p = 0,03499). L'épaisseur des reins n\'avait pas de relation significative avec la biométrie du sujet. ###### Caractéristiques des patients inclus --------------------------------------------------------------------------------------- Paramètres Moyen\ Moyen\ Moyen\ ± écart-type ± écart-type ± écart-type ------------------------------------------ -------------- -------------- -------------- Age (ans) 45 ±16 49 ±16 42 ±15 Poids moyen (kg) 52 ± 10 54 ± 9 50 ± 10 Taille moyenne (cm) 160±8 165±7 156±7 Surface corporelle (m^2^) 1,53 ±0,15 1,58 ±0,14 1,5 ±0,15 Indice de masse corporelle (kg/m^2^) 20± 3,8 19± 3 21± 4 Créatininémie (µmol/L) 87±19,6 \- \- Débit de filtration glomérulaire (ml/mn) 95± 37,2 \- \- Rein droit longueur (mm) 98±10 98±8 99±10 largeur (mm) 42±7 43±7 42±7 épaisseur (mm) 30± 16 32± 17 28± 15 Rein gauche longueur (mm) 99±10 100±10 99±10 largeur (mm) 45±6 45±8 45±7 épaisseur (mm) 31± 16 33± 18 30± 14 --------------------------------------------------------------------------------------- Notre étude rapporte les valeurs biométriques des reins d\'un adulte malgache, issues des résultats d\'un service clinique en se basant sur une exploration échographique, éléments essentiels pour aider un clinicien à prescrire un acte diagnostique ou thérapeutique: réaliser une biopsie rénale ou argumenter la sévérité d\'une insuffisance rénale. Ces décisions se basent surtout sur la longueur rénale qui est un bon indicateur de la dimension rénale en pratique clinique \[[@CIT0005], [@CIT0006]\]. Pour Ahmad et al, l'étude effectuée chez des malgaches adultes rapportait les dimensions suivant: le rein droit mesurait 91,16 mm de hauteur, 48,18 mm de largeur et 37,04 mm d′épaisseur. Le rein gauche faisait 90,6 mm de hauteur, 37,04 mm de largeur, 40,4 mm d′épaisseur. L′âge moyen pour la série d\'Ahmad et al \[[@CIT0001]\] était plus jeune avec un médian à 36,83 ans (18 à 77 ans) par rapport aux patients dans notre série, expliquant la tendance plus inférieure des différentes mesures dans leur série. L'âge a une corrélation positive avec la dimension du rein, mais qui s\'inverse ensuite après 60 ans \[[@CIT0007]\]. Dans notre série et dans les séries antérieures \[[@CIT0001], [@CIT0007]\], le rein gauche avait une dimension plus grande que le rein droit; cette différence entre les 2 reins serait due au fait que le foie est plus grand que la rate et le rein droit a moins d\'espace pour son développement \[[@CIT0007]\]; l\'artère rénale gauche est moins courte avec comme conséquence un flux de sang plus important pour le rein gauche \[[@CIT0007]\]; enfin, la taille plus petite des patients limiterait le développement longitudinal des 2 reins \[[@CIT0007]\]. Dans notre série, aucune corrélation entre le genre et les dimensions rénales était constatée statistiquement, même si la série de Glodny et al rapportait que le rein de l\'homme est plus grand que ce de la femme, par l\'influence de l\'hormone sexuelle sur cette corrélation \[[@CIT0008]\]. Dans notre résultat et celui d\'une étude antérieure chez les tananariviens adultes \[[@CIT0001]\], aucune relation n'était constatée entre la taille du patient et les dimensions des reins. Les limites de notre étude sont non négligeables, les résultats des échographies étaient recueillis rétrospectivement à partir des dossiers des patients d\'un service clinique. Ces échographies étaient issues de différents centres d'échographie d\'Antananarivo, faites par différents opérateurs avec différents appareils d'échographie, les résultats étant opérateur-dépendants; même si la mesure de la longueur rénale varie moins par rapport à la mesure du volume rénal \[[@CIT0005], [@CIT0006]\]. Conclusion {#S0002} ========== Les données issues de notre étude vont servir de bases en se rajoutant à celles obtenues antérieurement chez la population malgache, en attendant les études sur de large échantillon et les méta-analyses. Conflits d\'intérêts {#S0003} ==================== Les auteurs ne déclarent aucun conflit d\'intérêts. Contributions des auteurs {#S0004} ========================= Benja Ramilitiana et Mihary Dodo ont contribué aux collectes des données et à la rédaction du manuscrit; Henintsoa Nirina Rakotoarimanga a contribué à la rédaction du manuscrit; Rado Lalao Randriamboavonjy a réalisé l′analyse statistique; Willy Franck Randriamarotia a supervisé l′ensemble du travail et a contribué à la correction finale du manuscrit. Tous les auteurs ont lu et approuvé la version finale du manuscrit.

Hydrogen has long been considered as a clean, abundant and efficient energy carrier[@b1][@b2]. Developing appropriate storage media is of the importance for practical application of hydrogen energy. As an earth-abundant element, boron is widely applied for hydrogen storage with its chemical hydrides and nanostructural forms[@b3]. Boron-based chemical hydrogen storage materials such as borohydrides (e.g., LiBH~4~ and NaBH~4~) are promising compounds because of their high hydrogen capacities[@b4][@b5][@b6]. However, due to kinetic and/or thermodynamic limitations, the chemical hydrides suffer from poor reversibility[@b7], there are still difficulties in practical application of borohydrides[@b8]. An efficient solution is to find suitable all-boron nanostructures as replacement. Since the bulk boron cannot be found in nature, the design and synthesis of bulk boron allotropes still keeps challenging to theoretical and experimental chemists. It attracts more interest on all-boron fullerenes after the theoretical prediction of B~80~ fullerene[@b9], which is a hollow cage-like cluster resembling the C~60~. It is revealed that all of the boron allotropes are based on different arrangements of B~12~ icosahedrons[@b9][@b10]. After that, various types of boron fullerene nanostructures were proposed and simulated by theoretical calculations, such as B~32 + 8n~ (from B~32~ to B~80~)[@b11], B~32 + 6n~ (from B~32~ to B~56~)[@b12], 80 n^2^ boron fullerenes series (from B~80~ to B~2000~)[@b13], B~100~[@b14], etc. Boron fullerenes are seen as efficient hydrogen storage media since they are light-weight and have the capability to bind with metal adatoms. Combined with the fact that isolated transition metal (TM) has the ability to bind a certain number of hydrogens in molecular form, theoretical simulations on hydrogen adsorption by metal-adsorbed boron fullerenes were reported[@b15][@b16][@b17]. By density functional theory (DFT) calculations, Li *et al.*[@b16] declaimed that Ca-coated B~80~ fullerene can store up to 8.2 wt% H~2~ with an adsorption energy of 0.12-0.40 eV/H~2~. Before that, Zhou *et al.*[@b17] reported the hydrogen adsorption on alkli-metal (Na, K) doped B~80~. They found that B~80~Na~12~ and B~80~K~12~ show fairly low adsorption energies (0.07 eV/H~2~ and 0.09 eV/H~2~), indicating that alkli-metal is unsuitable for hydrogen storage. So far, all the theoretical investigations are based on the "proposed" boron fullerenes. Their applications in hydrogen storage may be unfeasible due to the uncertainty of the adsorbents. Recently, an all-boron fullerene-like cage cluster B~40~^−^ was produced and observed[@b18]. Its neutral counterpart B~40~ exhibits the fullerene-like cage (*D*~*2d*~ symmetry) and is calculated to be the most stable structure among the B~40~ allotropes. The relevant theoretical simulation indicates that B~40~ fullerene is thermally stable at temperature as high as 1000 K[@b18]. This is the first experimental evidence for the existence of all-boron fullerene. For the hydrogen storage materials, transition metal (TM) atoms are important components due to their strong attraction to hydrogen molecules[@b19][@b20][@b21][@b22]. Among the TMs, titanium (Ti) is regarded as an ideal binding metal in nanomaterials since it takes great advantages in hydrogen storage, which has been concluded[@b16]. Because of the outstanding performance in hydrogen storage, Ti-decorated nanostructures have been widely reported[@b19][@b23][@b24][@b25][@b26][@b27][@b28][@b29][@b30][@b31][@b32]. However, previous computational researches on hydrogen storage of B~80~[@b16][@b17] indicated that Ca is the appropriate adsorbate for boron fullerene due to the stable adsorption and high storage capacity. So which kind of metal atom would be the best adsorbate for B~40~ as hydrogen storage material? Here we perform DFT calculations on the binding capability of different metal atoms (Ca and TM: Sc, Ti, V, Fe, Co, Ni, Cu) decorated B~40~ fullerene. The simulations on hydrogen storage by metal-decorated B~40~ fullerene are also carried out. Results and Discussion ====================== The surface of B~40~ fullerene contains 48 boron triangles, embedded by 4 heptagonal rings and 2 hexagonal rings. The hexagons are planar while the heptagons are non-planar. We placed metal atoms on different sites of surface of B~40~ and calculated the binding energy (E~bind~) followingwhere *n* is the number of metal adatom coated on B~40~. *E*~*M*~, *E*~*B40*~ and *E*~*nM\@B40*~ stand for the total energies of metal adatom, B~40~ and the metal-coated B~40~ complex, respectively. We first calculated the binding energies of single metal atom on different binding sites of B~40~, including the centers of hexagon and heptagon, as well as the B-B bridges around hexagon (B1) and heptagon (B2). We take 8 different metal adatoms (Sc, Ti, V, Fe, Co, Ni, Cu, and Ca) for comparison. As shown in [Fig. 1](#f1){ref-type="fig"}, the centers of hexagon and heptagon are confirmed as the energy-favorable sites due to the significantly higher E~bind~ than sites B1 and B2. Ca atoms even cannot stably bind to the B-B bridges. To avoid the metal adatoms forming cluster on surface of B~40~[@b23][@b33], the metal species should meet the requirement that the binding energies are higher than their corresponding crystalline cohesive energies (E~coh~)[@b19][@b34]. [Figure 1](#f1){ref-type="fig"} indicates that Sc, Ti and Ni show higher binding energies with B~40~ than their cohesive energies, both on the centers of hexagon and heptagon. Thus Sc, Ti and Ni could be used as good adsorbates to decorate B~40~. The average binding energies of 1-6 metal adatoms (Sc, Ti, and Ni) on different facets of B~40~ are listed in [Table 1](#t1){ref-type="table"}.When there are more than 4 Sc atoms, the Sc-coated B~40~ complexes will distort and the cause instability of the fullerene-like substrate. Oppositely, the introduction of more Ti and Ni atoms will not affect the geometric structure of B~40~ significantly. When all of the hexagonal and heptagonal facets are coated by Ti or Ni atoms, the Ti~6~B~40~ or Ni~6~B~40~ complexes keep stable and provide high E~bind~. It is worth noting that due to the differences in valence electron configuration, Ni and Ti show significantly different bonding structures and binding energies with the different facets of B~40~. By comparing the binding energies with equal number of metal atom in [Table 1](#t1){ref-type="table"}, it can be concluded that Ti is more energy-favorable on hexagon, while Ni is more energy-favorable on heptagon. To reveal the bonding rules, we performed Mayor bond order[@b35] analysis on single Ti- and Ni-decorated B~40~ fullerene, as shown in [Fig. 2](#f2){ref-type="fig"}. Different binding conformations on hexagonal ring and heptagonal ring are named M\@hexagon and M\@heptagon, respectively. The bonding structures reveal that Ni covalently bonds with all the surrounding boron atoms, but Ti only forms 4 and 3 stable covalent bonds (with bond order value larger than 0.5) when binds to hexagon and heptagon respectively. Considering their valence electron configurations (Ti: 3d^2^4s^2^, Ni: 3d^8^4s^2^), the rich valence electrons determine that Ni can form as much as 7 weak Ni-B covalent bonds, while Ti only forms up to 4 Ti-B covalent bonds due to its 4 valence electrons. Ti-B average bond length (\~2.17 Å) is longer than Ni-B average bond length (\~2.0 Å), which explains why the Ti-coated hexagon expands in [Fig. 2(a)](#f2){ref-type="fig"} compared with Ni\@hexagon in [Fig. 2(c)](#f2){ref-type="fig"}. However, the Ti-coated heptagon changes slightly, mostly due to its non-planar arrangement of boron atoms. Ti\@hexagon shows higher stability than Ti\@heptagon since there are more Ti-B covalent bonds. Similarly, Ni\@heptagon is more stable than Ni\@hexagon because of the 7 covalent bonds. This is the reason why Ti is more energy-favorable on hexagon while Ni is more energy-favorable on heptagon. Another important finding is that the average binding energy is related with the number of metal adatoms on different facets. That is, for Ti-decorated B~40~ fullerene, the average binding energy increases as the number of Ti atoms on heptagon increases, and decreases as the number of Ti atoms on hexagon decreases. Differently, for Ni-decorated B~40~ fullerene, the average binding energy decreases as the number of Ni atoms increases for both binding sites. It can be inferred that there exists attractive interaction between the decorated Ti atoms, while the interaction between the coated Ni atoms is repulsive. In summary, when all the hexagonal and heptagonal rings are embedded by metal atoms, the binding of Ti will be stronger than Ni, and also the strongest among the chosen metal species. The stable binding of Ti on B~40~ leads to promising applications of the Ti-decorated B~40~ fullerene. Here we consider it as a suitable candidate for hydrogen storage. According to the well-known 18-electron rule[@b19][@b36], the maximum number of adsorbed hydrogen molecules (N~max~) is limited by the valence electrons that participating in covalent bonds. For metal-decorated B~40~ fullerenes we design here, the 18-electron rule can be specified aswhere n~v~(M) represents the valence electron number of the metal element, n~v~(B~40~) represents the electrons contributed by B~40~, which is 4 for Ti\@hexagon and 3 for Ti\@heptagon. The N~max~ is calculated to be 5 and 5.5 for Ti\@hexagon and Ti\@heptagon, which demonstrates that the single Ti-decorated B~40~ can store up to 5 and 6 H~2~ molecules when Ti atom binds to hexagon and heptagon, respectively. However, the N~max~ is calculated to be 1 for Ni-decorated B~40~ fullerene. Obviously the Ni-derad B~40~ fullerene is inefficient as hydrogen storage medium. We use average adsorption energy (E~ads~) to evaluate the adsorption capability of the Ti-decorated B~40~ fullerene. We also define consecutive adsorption energy (ΔE) as the energy gained by successive additions of H~2~ molecules to evaluate the reversibility for storage of H~2~ molecules. They are calculated based on the following formulasandwhere n stands for the number of adsorbed H~2~ molecules. *E*~*Ti\@B40*~ and *E*~*H2*~ are the total energies of Ti-decorated B~40~ and H~2~ molecule. *E*~*Ti\@B40 + nH2*~ and *E*~*Ti\@B40 + (n-1)H2*~ are the total energies of Ti-decorated B40 with n and (n-1) H~2~ molecules, respectively. For efficient hydrogen storage at ambient conditions, the ideal adsorption energy should be in the range of 0.16-0.42 eV/H~2~[@b37][@b38] to realize reversible adsorption and desorption. This energy range leads to intermediate between physisorption and chemisorptions[@b16]. The calculated E~ads~ and ΔE are summarized in [Table 2](#t2){ref-type="table"}. With all of the ΔE larger than 0.2 eV/H~2~, our simulations confirm that the maximum adsorption numbers of H~2~ molecules can reach 5 for Ti\@hexagon and 6 for Ti\@heptagon, respectively. For H~2~ adsorption on Ti\@hexagon, the first H~2~ molecule exhibits significantly larger adsorption energy than the following H~2~ molecules. Addition of the second to fifth H~2~ molecule gains energies within 0.2-0.3 eV per H~2~, and they are adsorbed around the first H~2~, as shown in [Fig. 3(a)](#f3){ref-type="fig"}. Our analysis on the Ti-H~2~ distance reveals that for the Ti\@hexagon, the first added H~2~ molecule keeps a close distance to the Ti atom (1.950 \~ 1.975 Å in [Table 2](#t2){ref-type="table"}). Particularly, affected by the 4 surrounding H~2~ molecules, the 1^st^ H~2~ molecule of 5 H~2~ molecules adsorbed Ti\@hexagon will be closer to the Ti atom. Moreover, as shown in [Fig. 3(b)](#f3){ref-type="fig"}, the first H~2~ molecule always shares the highest occupied molecular orbital (HOMO) with the adsorbent, indicating the strong chemical adsorption between the first H~2~ molecule and Ti\@hexagon. The case of adsorption on Ti\@heptagon is different. As we can see in [Table 2](#t2){ref-type="table"}, the first and second H~2~ molecules both show strong binding to the Ti atom. This can be attributed to the extra 3d electron of Ti, which doesn't participate in forming covalent Ti-B bond. [Figure 3(d)](#f3){ref-type="fig"} indicates that the Ti 3d orbital overlaps with the H 1s orbital when there is one or two H~2~ molecules adsorbed. With the addition of third H~2~ molecule, the overlapping between Ti and H~2~ is interrupted. From the addition of 3^rd^ to 6^th^ H~2~ molecule, the HOMOs only distribute on surface of Ti\@heptagon, indicating the weakening of the H~2~-Ti interaction. On the other hand, distances between Ti and the first two H~2~ molecules become significantly larger with the addition of 3^rd^ to 6^th^ H~2~ molecules, which is consistent with the HOMO analysis. Addition of the third to sixth H~2~ molecules gains consecutive adsorption energies within 0.3--0.4 eV per H~2~, which also meets the requirement for reversible uptake and release of H~2~ molecules. As displayed in [Fig. 3](#f3){ref-type="fig"}, it should be pointed out that either the geometric structures or the distribution of HOMOs of the adsorption substrate (Ti-decorated B~40~ fullerene) keep stable and are little changed with the increasing of adsorbed H~2~ molecules, revealing the high stability of Ti-decorated B~40~ fullerene. The geometric and electronic structure of the substrate is little affected by the addition of H~2~ molecules, which is important for the realization of reversible hydrogen storage. To check if the first adsorbed H~2~ molecule will dissociate into two hydrogen ions on centered Ti atom and form dihydride complex, as mentioned in similar work[@b19][@b22][@b24][@b25][@b27], we also modeled the dihydride contained complexes (B~40~TiH~2~) as initial configurations and performed full geometry optimization. Our simulation results (as displayed in [Figure S1](#S1){ref-type="supplementary-material"}) show that the dihydride complex is less stable than our determined local minimum (about 1.10 eV higher in total energy). Meanwhile, singlet state should be considered as the ground state for dihydrogen adsorbed Ti\@B40 complexes due to the higher stability. The dihydrogen molecule with a slight elongation of H-H is determined as the local minimum for adsorption of the first H~2~ molecule on Ti-decorated B~40~. To look insight of the influence of B~40~ in adsorbing hydrogen, we checked all the possible adsorption sites of undecorated B~40~ for H~2~ adsorption. Calculation results show that the B~40~ fullerene itself is unsuitable for H~2~ adsorption with E~ads~ ranges from 0.15 eV to 0.20 eV (as listed in [Table S1](#S1){ref-type="supplementary-material"}). All of the distances from the adsorbed H~2~ to B~40~ surface are larger than 2.8 Å, indicating the nature of weak physisorption. To see whether the H~2~ molecule will transfer from Ti to B~40~ when adsorbs to Ti-decorated B~40~, the possibility of H~2~ adsorption onto B~40~ in Ti-decorated B~40~ (Ti~6~B~40~) is also checked. Our simulations elucidate that comparing with the H~2~ adsorption on Ti atoms, the H~2~ adsorption on B~40~ is rather weaker with E~ads~ around 0.2 eV ([Table S1](#S1){ref-type="supplementary-material"}). Adsorption energies of H~2~ on B~40~ in Ti~6~\@B~40~ complex enhance slightly compared with the undecorated one (for the same adsorption site), indicating that the decoration of Ti atoms won't improve the adsorption performance of B~40~ for H~2~ much. For our modeled Ti~6~B~40~ complexes, the Ti atoms exhibit high attraction for hydrogen molecules due to the high localization of FMO on them, as shown in [Figure S2](#S1){ref-type="supplementary-material"}. This localization won't be significantly affected by the increasing H~2~ molecules, making the transfer of H~2~ molecule to B~40~ difficult to happen. Based on the calculation results of hydrogen adsorption on single Ti-decorated B~40~, we constructed and optimized the adsorption configuration of H~2~ molecules on Ti~6~B~40~ complex. As shown in [Fig. 4](#f4){ref-type="fig"} (the atomic coordinates of the optimized Ti~6~B~40~ and Ti~6~B~40~ with 34 H~2~ molecules adsorbed are listed in [Table S2](#S1){ref-type="supplementary-material"} and [S3](#S1){ref-type="supplementary-material"}), up to 34 H~2~ molecules are adsorbed around the Ti adatoms \[named Ti~6~B~40~(H~2~)~34~\]. Our calculated gravimetric density of hydrogen stored by Ti~6~B~40~ can reach 8.7 wt%, with an average adsorption energy of 0.37 eV/H~2~. As we have mentioned above, the first H~2~ molecule on Ti\@hexagon and the first two H~2~ molecules on Ti\@heptagon have stronger binding with the Ti atoms than the following H~2~ molecules. We measured the average distance between the H~2~ molecules and the corresponding Ti atoms for Ti~6~B~40~(H~2~)~34~. For H~2~ adsorption on hexagon-embedded Ti atoms, the average distance of the 1^st^ H~2~ molecules to Ti atom is 1.952 Å, almost the same distance with the occasion of 5 H~2~ molecules adsorbed Ti\@hexagon. However, for H~2~ adsorption on heptagon-embedded Ti atoms, the average distances of the 1^st^ and 2^nd^ H~2~ molecules to Ti atom are 2.052 Å and 2.358 Å, respectively. The values are significantly larger than the occasion of 6 H~2~ molecules adsorbed Ti\@heptagon, indicating the repulsive interaction from H~2~ molecules on other facets. Analysis on H~2~-Ti distance demonstrates that the increase of H~2~ molecule mainly affects the hydrogen adsorption on heptagon-embedded Ti atoms, which is the origin of reduction of the average H~2~ adsorption energy. Evaluating from our calculation results on successive addition of H~2~ molecules, among the 34 adsorbed H~2~ molecules on Ti~6~B~40~, 24 of them have moderate adsorption energies within the range of 0.2-0.4 eV/H~2~, corresponding to a reversible storage capacity of 6.1 wt%. It is notable that the bonding type and geometric structure of the B~40~Ti~6~ complex is also little affected by the adsorption of H~2~ molecules. The favorable consecutive adsorption energy assures the reversible storage of hydrogen molecules under ambient conditions. B~40~ is a newly discovered boron nanostructure and also the first experimentally observed all-boron fullerene. Here we performed computational investigations on hydrogen storage capacity of Ti-decorated B~40~ fullerene. Comparative calculations reveal that, among the chosen metal species, Ti exhibits the strongest binding on surface of B~40~. Ti-decorated B~40~ fullerene exhibits strong adsorption and high capacity for H~2~ molecules. Single Ti decorated B~40~ fullerene can store up to 5 and 6 H~2~ molecules for Ti\@hexagon and Ti\@heptagon, respectively. All of the adsorption happens on Ti atom, and B~40~ shows weak capability in adsorbing H~2~ molecules. This leads to a maximum storage capacity of 34 H~2~ molecules for Ti~6~B~40~ complex with an average adsorption energy of 0.37 eV/H~2~, corresponding to a gravimetric density of 8.7 wt%. The consecutive adsorption energy of H~2~ molecules within the range of 0.2--0.4 eV/H~2~ assures the reversible storage of 6.1 wt% under ambient conditions. Our computational investigations confirm that the Ti-decorated B~40~ fullerene is favorable for hydrogen adsorption, which makes it promising as a new hydrogen storage material. Methods ======= Density functional theory (DFT) calculations are carried out by DMol~3~ program[@b39][@b40]. The generalized gradient approximation (GGA) functional by Perdew and Wang (PW91)[@b41], along with a double numerical basis set including p-polarization function (DNP), is applied for the geometry optimization and property calculations. Dispersion-corrected DFT (DFT-D)[@b42][@b43][@b44] scheme put forward by Ortmann, Bechstedt, and Schmidt (OBS)[@b45] is used to describe the van der Waals (vdW) interaction. DFT semi-core pseudo-potentials (DSPPs)[@b46] are employed to efficiently treat with the core electron of TM element after Ca. Self-consistent-field (SCF) convergence tolerance is set to 1 × 10^−6^ Ha. The convergence threshold values are specified as 1 × 10^−5^ Ha for energies, 2 × 10^−3^ Ha/Å for gradient, and 5 × 10^−3^ Å for displacement, respectively. Reliability of PW91/DNP level in treating metal-boron system has been proven by Zhou *et al.*[@b17] in calculating the binding of alkli-metal (AM) on B~80~ fullerene as well as the hydrogen storage capacity of B~80~-AM complexes. The incorporation of DFT-D scheme further improves the accuracy in evaluating weak interactions. Author Contributions ==================== Y.L. developed the main idea and supervised the project. H.D. performed all the calculation work and analyzed the results. H.D., T.H., S.L. and Y.L. wrote the paper. Additional Information ====================== **How to cite this article**: Dong, H. *et al.* New Ti-decorated B~40~ fullerene as a promising hydrogen storage material. *Sci. Rep.* **5**, 09952; doi: 10.1038/srep09952 (2015). Supplementary Material {#S1} ====================== ###### Supplementary Information The work is supported by the National Basic Research Program of China (973 Program, Grant No. 2012CB932400), the National Natural Science Foundation of China (Grant No. 91233115, 21273158 and 91227201), a Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). This is also a project supported by the Fund for Innovative Research Teams of Jiangsu Higher Education Institutions, Jiangsu Key Laboratory for Carbon-Based Functional Materials and Devices, Collaborative Innovation Center of Suzhou Nano Science and Technology. {#f1} {#f2} {#f3} {#f4} ###### The average binding energies (E~bind~) of 1-6 metal adatoms (Sc, Ti, and Ni) on different facets of B~40~, the cohesive energies (E~coh~) of the metal are shown as comparison[@b47] Metal adatom Sc Ti Ni ---------------------- ------ ------ ------ E~coh~/eV 3.90 4.85 4.44 E~bind_hex1~/eV 5.17 5.30 5.38 E~bind_hept1~/eV 5.04 5.09 6.14 E~bind_hex2~/eV 5.22 5.27 5.36 E~bind_hept2~/eV 5.24 5.29 6.07 E~bind_hex2hept2~/eV \_ 5.58 5.69 E~bind_hept4~/eV \_ 5.88 6.06 E~bind_6~/eV \_ 5.83 5.81 The subscripts "hex" and "hept" indicate that the metal adatoms are adsorbed to the hexagonal and heptagonal facets of B~40~, while the Arabic number indicates the number of metal adatoms that coated on the corresponding facet. E~bind_6~ means that all the 6 facets are decorated by the metal adatoms. ###### Calculated average adsorption energies (E~ads~) and consecutive adsorption energies (ΔE) by the successive addition of H~2~ molecules to Ti\@hexagon and Ti\@heptagon, as well as the distance between Ti atom and the first (Ti-1^st^ H~2~) or second (Ti-2^nd^ H~2~) added H~2~ molecule. N(H~2~) Ti\@hexagon Ti\@heptagon --------- ------------- -------------- ------- ------ ------ ------- ------- 1 0.82 0.82 1.953 0.74 0.74 1.928 -- 2 0.51 0.20 1.962 0.78 0.82 1.898 1.926 3 0.43 0.26 1.957 0.65 0.39 1.932 2.110 4 0.40 0.31 1.975 0.57 0.30 1.942 2.116 5 0.36 0.22 1.950 0.52 0.32 1.956 2.111 6       0.49 0.39 1.955 2.113

Introduction ============ Nitric oxide (NO) is an important vasodilator in the cardiovascular system. The synthesis of NO is catalyzed by the enzyme family called nitric oxide synthases (NOS); neuronal (nNOS, NOS1), inducible (iNOS, NOS2), and endothelial synthases (eNOS, NOS3).^[@R1]^ Of the NOS family, the most important NOS isoform in the context of basal release of vascular NO is endothelial NOS (eNOS). The endothelial release of NO is nonetheless reduced in diabetes and hypertension leading to endothelial dysfunction.^[@R2]^ Expression of iNOS can be induced in a wide range of cells and tissues especially in inflammatory conditions by cytokines and other agents, leading to production of high amounts of NO until the enzyme is degraded.^[@R3]^ There have been suggestions that hypertension could have an inflammatory background^[@R4]^ and there is also evidence that the amount of NO^[@R5]^ and expression/activity of iNOS is increased in hypertensive patients.^[@R6]^ However, the lack of hits in genome-wide association studies (GWAS) for inflammation-associated genes in hypertension,^[@R7],[@R8]^ prompted us to study whether genetic variants of the iNOS gene could be involved in hypertension at younger age. Two common functional polymorphisms have been described for the iNOS gene and both variations lead to increased NO production. iNOS variant rs2779249 (--1026 C/A) is located in the promoter region of the gene and it has been shown that nucleotide change from C to A increases iNOS promoter transcriptional activity to fivefold leading to higher NO production. The other variant rs2297518 (2087 G/A), located in exon 6, causes an amino acid substitution from serine to leucine which increases iNOS activity (alters iNOS protein function) and conferees higher NO production based on the A-allele.^[@R9]^ The purpose of this study was therefore to investigate whether iNOS genetic variants are associated with hypertension in a Finnish population by analyzing cohorts from the Tampere adult population cardiovascular risk study (TAMRISK).^[@R10]^ METHODS ======= Subjects -------- The data for the TAMRISK study were collected from periodic health examinations (PHEs) done for 50-year-old men and women living in Tampere, Finland. TAMRISK data include information of risk factors for hypertension: blood pressure (BP), weight, parental history of cardiovascular diseases, lipid values and smoking, diabetes and exercise habits. Buccal swabs for DNA extraction and a permissions form to use PHE data were collected by mail separately of the physical examination. The DNA samples were collected during years 2006 to 2010. Informed consent was obtained from all participants. A detailed description of the design and data collection as well as protocol of baseline measurements of the study is described elsewhere.^[@R10],[@R11]^ The study protocol was approved by The Ethics Committees of the Tampere University Hospital and the City of Tampere. Cases (n = 320) in this study were the subjects who had hypertension and/or CAD at the age of 50 years (as diagnosed by a physician by normal healthcare procedures) and for each case, at least 1 normotensive control (n = 439) with the same sex and similar smoking habits, were chosen from a PHE cohort (n = 6000). Smoking status was evaluated based on self-reporting. Of the same individuals, we also analyzed the subpopulation of men and women who had available PHE data at the age of 35, 40, 45, and 50 years. Registration of BP was made at the examination visit (mm of mercury) using calibrated mercury sphygmomanometer. Genotyping ---------- Genomic DNA was extracted from buccal swabs using a commercial kit (Qiagen, Inc., Valencia, CA). The samples were transferred into 96-well plates and the 2 single nucleotide polymorphisms (SNPs) for iNOS were genotyped at the LGC genomics using Competitive Allele Specific PCR (KASP) technique (LGC genomics, Hertz, UK). Statistical Analyses -------------------- One-way ANOVA for continuous variables and Chi-square test for categorical variables were applied for the comparison of cases and controls. If the distribution was skewed, the analysis was performed using transformed values to approximately normalize the distribution. Associations of the 2 genotyped SNPs for hypertension with risk factors were analyzed by logistic regression analysis. The analysis of variance (ANOVA) for repeated measures was used to assess the differences in mean BP: s between genotypes at the age of 35, 40, 45, and 50 years. *P* values less than 0.05 were considered significant. Hardy--Weinberg equilibrium of the genotypes was calculated using OEGE (online encyclopedia calculator for genetic epidemiology studies).^[@R12]^ Analyses were carried out using SPSS 20.0 for Windows (SPSS, Inc., Chicago, IL). RESULTS ======= The clinical characteristics of the middle-aged (50 ± 0 years) study subjects are listed in Table [1](#T1){ref-type="table"}. In addition a subgroup population (n = 417) had clinical measurements also at the age of 35, 40, and 45 years. In the whole study population the frequencies of the rs2779249 variants were 0.553 for CC (n = 444), 0.372 for CA (n = 299), and 0.075 (n = 60) for AA. Frequencies of the rs2297518 variants were 0.702 for GG (n = 567), 0.265 for GA (n = 214), and 0.032 for AA (n = 26). The measured genotype frequencies were not significantly different from the expectations of Hardy--Weinberg equilibrium (χ^2^ = 0.96 for rs2779249 and χ^2^ = 1.09 for rs2297518). ###### Characteristics of Study Groups at the Age of 50 Years  At the age of 50 years, the SNP rs2779249 (C/A) associated significantly with hypertension (*P* = 0.009); specifically, subjects carrying the A-allele had higher risk of hypertension compared to those carrying the CC genotype (*P* = 0.016, OR = 1.43; 95% CI: 1.07--1.91). Among smokers, the risk for hypertension increased to 2-fold with the A-allele carriers (*P* = 0.019, 95% CI: 1.12--3.56). In addition, we noticed a significant association among the A-allele carriers with hypertension already at the age of 35 years, which continued the whole follow-up time (age groups 35, 40, 45, and 50 years) (Table [2](#T2){ref-type="table"}). Also BP readings according to genetic variants confirmed these observations and showed that those who carried the A-allele had higher levels of both systolic (*P* = 0.047) and diastolic (*P* = 0.048) BP during the 15-year follow-up period (Fig. [1](#F1){ref-type="fig"}). ###### Association of iNOS With Hypertension at Different Ages  {#F1} No significant associations between rs2297518 (G/A) variants alone and hypertension were found (Table [2](#T2){ref-type="table"}). In order to study a possible haplotype effect on hypertension, different haplotypes from rs2779249 (C/A) and rs2297518 (G/A) were generated. Three most common haplotypes were CC--GG (n = 389, 48.8%), CA--GG (n = 153, 19.2%), and CA--GA (n = 137, 17.2%). All other haplotype combinations represented less than 6% each in the study population. In the most common haplotype group H1 (CC--GG), 145 out of 389 (37.3 %) of the carriers had hypertension at the age of 50 years, whereas 66 cases out of 137 (48.2%) in the H3-group (CA--GA) had this diagnosis. In BMI-adjusted logistic regression, the risk of hypertension was 2 times higher in the H3-haplotype group compared with the H1-group (OR 2.01, 95% CI: 1.29--3.12, *P* = 0.002) (Table [3](#T3){ref-type="table"}). ###### Haplotype Frequencies for iNOS Genetic Variants and Their Estimated Association With Hypertension  DISCUSSION ========== In the present study, we have amplified 2 functional polymorphic sites of the iNOS gene and studied their association with hypertension in a series of Finnish men and women living in the Tampere region (TAMRISK study). We found that mutation in the promoter region (rs2779249 (C/A)) of the iNOS gene associates with increased risk of hypertension. Moreover, we also found that those having the allele A had higher risk for hypertension already at the age of 35 years. A haplotype effect on hypertension was found with the above rs2779248 and the other iNOS SNP rs2297518 (G/A), a polymorphism which leads to differences in activity of iNOS protein and further NO production.^[@R9]^ Thus our results not only further suggest that iNOS gene has a functional role in hypertension, but also imply that allelic variants of this gene may affect early manifestation of the disease. The reason for increased prevalence of hypertension among A-allele carriers of the iNOS gene is unknown. The present polymorphism rs2779249 may affect transcription of iNOS as it is in the regulatory region of the gene. In fact, Fu et al^[@R13]^ have found significant differences in the promoter activity of the present alleles as the A-allele was associated with 5-fold increases in iNOS transcriptional activity compared with the C-allele. Lyamina et al showed that young men with high normal BP had a higher NO production than those with normal or optimum BP. In addition, the higher the BP, the more pronounced was NO overproduction. They also suggest that the inflammatory mediators may induce NO overproduction.^[@R5]^ Increased concentration of NO is converted to peroxynitrite and superoxide in the prooxidant environment characteristic to essential hypertension. In addition, iNOS upregulates araginase activity, which limits NO production through eNOS. Animal model experiments provide evidence that iNOS participates in the regulation of renal function and BP^[@R14]^ and it has been suggested that iNOS could be involved in the early rise in BP.^[@R15]^ The sympathetic nervous system plays a fundamental role in modulating cardiovascular functions. Catecholamine release leads to the activation of 3 β-adrenergic receptor subtypes (β1, β2, and β3-ARs), which regulate vasomotor tone. Endothelium might also control or facilitate β-AR effects of the vessels, since β-AR activation stimulates eNOS activity and could increase release of endothelial NO. In hypertension, the regulatory importance of NO is underlined by the observation that reduction of eNOS activity plays a pivotal role in BP control.^[@R16]^ BP control by β-AR signaling is heavily regulated by G-protein-coupled receptor kinase (GRK2)-mediated desensitization.^[@R17]^ Permanently high catecholamine levels could lead to over activation of β-ARs, increasing eNOS activity and expression. Age-related increase in circulating catecholamines and decreased β-AR responsiveness is seen in the elderly.^[@R18]^ Hyposensitivity to catecholamines is also seen during heart failure, and enhanced levels of catecholamines lead to chronic stimulation of cardiac β-ARs.^[@R19]^ The third adrenergic receptor (β3-AR), which is suggested to act as a brake in sympathetic overstimulation in high catecholamine concentrations antagonizing β1- and β2-AR activity has in some conditions been shown to increase NO production through iNOS by an unknown mechanism.^[@R20]^ This suggests that in aging and heart failure, there is an important relationship between adrenergic system and iNOS. In experimental heart failure in the rat, increased NO may contribute to decreased basal heart rate.^[@R21]^ In fact, acute NOS inhibition led to an instantaneous recovery of the inotropic response to catecholamines in transgenic iNOS overexpressing mice.^[@R22]^ Taken together, although iNOS has been previously implicated to be associated with heart failure by the adrenergic system, our findings suggest that it may have a role also in hypertension, such as that shown for eNOS. There are only few publications concerning the role of iNOS genetic variants with hypertension and results have been conflicting apparently due to ethnic and environmental differences between populations. In opposite to our findings, Fu et al described the association of CC genotype of the iNOS rs2779249 polymorphism with hypertension in hypertensive Chinese families. They also found that the activity of a reporter gene construct containing the CC-genotype was approximately 5-fold lower than that with the A-allele.^[@R13]^ Oliveira-Paula and coworkers found no significant associations between hypertension and rs2779429 polymorphism in their study. Instead, they reported that rs2297518 (G/A) polymorphism affects susceptibility to hypertension. Moreover, haplotype analysis containing pentanucleotide repeat polymorphism in the promoter region of the iNOS gene, rs2779429 (C/A) and rs2297518 (G/A) showed that S-C-A haplotype associated with hypertension and with responsiveness to antihypertensive therapy.^[@R23]^ In another study, Glenn et al^[@R24]^ did not find any association of 2 different repeat promoter variants of iNOS gene with hypertension in older hypertensives. It is known that cigarette smoking results in increased levels of iNOS^[@R25]^ and there are some reports showing joint effect with genotypes of the iNOS gene and smoking.^[@R26],[@R27]^ One inclusion criterion in our study population was that every case had at least one control with similar smoking habits. Markedly, when smokers were analyzed in their own group, association of A-allele of the rs2779249 polymorphic site with hypertension was even stronger. Strengths of the study are that we have analyzed a study population with similar age and genetic background. It has been suggested that genetic determination in young individuals is much stronger than in old individuals, an observation that might be explained by increasing participation of nongenetic factors in disease etiology with aging.^[@R28]^ Limitations of the study were that a possible difference in BP values at the age of 45 years and after might be difficult to establish, since practically all of the subjects who had hypertension were on BP medication by this age. In addition, each registration of BP values used in the repeated measures analysis was made at one examination visit only. Furthermore, the study subjects are from a restricted genetic pool (Finnish Caucasian), and the findings might not be extrapolated to different genetic populations. In conclusion, our results support that A-allele of the iNOS gene (rs2779249) as well as haplotype AA (of variants rs2779249 (C/A) and rs2297518 (G/A)) contribute to increased risk of hypertension in the Finnish population. In addition, allelic variants of the gene affect the prevalence of the disease already at early middle-age. The authors would like to thank all the participants of the TAMRISK study. Abbreviations: ANOVA = analysis of variance, AR = adrenergic receptor, BMI = body mass index, BP = blood pressure, eNOS = endothelial nitric oxide synthase, GWAS = genome-wide association study, iNOS = inducible nitric oxide synthase, PHE = periodic health examination, SNP = single nucleotide polymorphism, TAMRISK = Tampere adult population cardiovascular risk study. This study was supported by grants from Competitive research funding of the Pirkanmaa Hospital District. The funding body did not play a role in the study design, collection, analysis, and interpretation of data, in the writing of the manuscript, or the decision to submit the manuscript for publication. The authors have no conflicts of interest to disclose.

Background {#Sec1} ========== European societies are undergoing rapid demographic changes. Thus, for contemporary societies, health promotion for older people (HP4OP) is becoming a highly relevant issue. The literature confirms that the older population has rarely been a target of significant dedicated health promotion programmes. The study presented here mainly focuses on the questions of who, how and in which way provides HP4OP activities in European countries. Currently, the lack of a systematic, comparative cross-country institutional analysis of that topic is evident. Moreover, a dedicated methodological approach to the topic does not seem to be being performed \[[@CR1]\]. This paper addresses the research gap outlined above. The basic assumption here is that the institutional environment and organisation constitute crucial components of the implementation of any policy in this field, especially in the context of capacity building for a greater sustainability of health promotion. The paper is based on the two intertwining research goals. Firstly, it explores the issue of which institutions/organisations perform HP4OP activities in selected European countries by detailing which sectors are involved, what kind of roles the institutions perform, and how they organise such activities. Secondly, it develops an institutional approach to HP4OP and provides the analytical tools for further research in the area by developing dedicated definitions and classifications. For the purpose of this study, i.e. an institutional approach to HP4OP, we define the key terms below. The definitions are a result of the research performed and are explained thoroughly in the results section. *Health promotion* is the process of improving population health status by enabling the individuals to leave healthy within a community and through government interventions, i.e. to increase the control over their health and its determinants. It encompasses prevention, education, and advocacy. Health promotion should respect human autonomy, be sustainable, evidence-based and adjusted to the specificity of the target group and local context. It should also be performed as a concerted action by various entities belonging to all sectors and on all levels of governance in order to effectively engage the available resource. Health promotion *activities* (the processes of health promotion as defined above) can take various forms. The most general is an *intervention* in the current state of affairs aimed at a behavioural, social, environmental or policy change. A health promotion *programme* is a set of organised activities strictly designed with a health promotion objective in mind, which are limited in terms of their scope, organisation, engaged actors and duration. Finally, we define an institution as "an interlocking double-structure of persons-as-role-holders or office-bearers and the like, and of social practices involving both expressive and practical aims and outcomes"\[[@CR2]\]. The above definitions are applied in our analysis to compare the provision of HP4OP activities in selected European countries. The analysis constitutes a part of the research carried out in the EU project ProHealth65+, which engages ten EU member states. These countries are also included in our analysis. Methods {#Sec2} ======= The objectives described in this paper were addressed with the mutual use of two complementary methods: (a) a literature review of the scientific and grey literature; and (b) questionnaire surveys with selected expert respondents from the ten project countries. The literature review took into account the theoretical and interdisciplinary nature of the study. The expert respondents, selected by the project's collaborating partners, filled in a custom designed questionnaire on the institutional aspects of HP4OP in their countries. Below, we first present a general overview of the study approach, and details about the two methods applied. The countries included in the study allow obtaining a broader overview of the problem as they represent different parts of Europe, varying in welfare and healthcare models. The countries include: Bulgaria, the Czech Republic, Germany, Greece, Hungary, Italy, Lithuania, the Netherlands, Poland and Portugal. The project countries represent different institutional regimes of the welfare state and can be grouped in three main models corresponding to particular regions: (a) continental European countries, (b) southern European countries and (c) CEE countries. The health status of the older populations varies from country to country. The differences in the health status of older populations are also of a great significance. The study was based on collecting information about HP4OP activities and was carried out in 2015 in the analysed countries. This served as a based to systemise the knowledge on HP4OP practices in these EU countries and to indicate sources for examples of good practices. Based on the analysis of the country experts' opinions and the information given, preceded and supported by an extended literature review, the following sectors were chosen for comprehensive analysis: health, social assistance, government (central), regional/local authorities, voluntary/NGO, education, sport, and media. These sectors were indicated as dominating the sphere of health promotion activities in general. This paper concentrates on a description of the complex institutional picture of health promotion activities focused on older people in the analysed European countries, taking into account the above mentioned sectorial pre-conditions. Literature review {#Sec3} ----------------- The initial knowledge concerning institutions acting in HP4OP in selected countries was acquired as a result of a literature review. The main sources of information were scientific papers and grey literature as well as other materials: government websites, strategic documents, programmes and projects, guidelines and other publicly sources that were accessible. Notable sources were reports developed by other institutions \[[@CR3]--[@CR21]\]. Previous literature reviews were profiled for theoretical background as well as for specificity to selected countries. A literature review on countries' institutional specificity was performed for English-language papers on HP4OP published between 2000 and 2015 in the project countries. The database selection was limited to PubMed and healthPROelderly database due to the scope of the research. This selection, provided a much needed comparison between scientific and grey literature sources. Search terms were selected in order to retrieve as many relevant sources as possible. Three sets of search terms were distinguished due to sector's specificity: see Fig. [1](#Fig1){ref-type="fig"}. Two independent researchers performed the source selection in two stages: by abstract and full text screening. For detailed information on the literature search flow see Fig. [1](#Fig1){ref-type="fig"}. Publications not explicitly mentioning health promotion, focused on treating diseases, not explicitly addressing the target group of the older people, or with a purely medical-care focus were excluded. Publications concerning the screening or clinical evaluation of projects, as well as studies on ageing populations -- observational studies such as EPOSA \[[@CR20]\] -- were included in order to identify institutions that study, monitor, evaluate and produce evidence-based knowledge for HP4OP \[[@CR21]\]. Programmes aimed at the general population, oriented towards diseases, for which old age is a risk factor \[[@CR22]\], were also included.Fig. 1Literature Review Process It was assumed, however, that some relevant institutional information was present in sources that could not be identified by the search engines mentioned above. Data provided by reviews on countries' institutional specificity mostly required further specifications. For this purpose, a follow-up narrative review of other available sources (including the grey literature) was performed for the proper identification of the institutions mentioned (their character, status, sector classification, etc.). The goal of the reviews was limited to the identification of institutions involved in health promotion initiatives and their roles; thus a quality check of the publications reviewed was not necessary. A supplementary review was also performed using:- journal databases: Health Policy Journal, Public Health, Journal of Public Health, and Health Systems in Transition \[[@CR23]--[@CR29]\];- web search engines (e.g. Google) for websites, grey literature, and other non-indexed sources;- websites of institutions dealing with issues related to the topic (promoters, providers, stakeholders, public agencies, NGOs, etc.);- websites of responsible authorities (by selected states -- after initial review) and the HealthPROelderly Project database \[[@CR13]\]. Detailed information on the number of relevant publications identified during the above literature search phases is shown in Fig. [1](#Fig1){ref-type="fig"}. The publications were analysed following the method of content analysis. Specifically, the content of the publications was reviewed to extract information related to the research goals specified at the outset of the paper. Questionnaire survey {#Sec4} -------------------- The first methodological question in the survey concerned the identification of institutions and organisations involved in HP4OP, in order to select those who are the most active in this field. Another important question concerned the research structure, determined by the involvement of many different project countries with a diversity of health promotion providers which had to be taken into account. The areas to be analysed were indicated as follows: the programmes and interventions undertaken by different institutions within the given health care and public health system as well as other subjects situated outside of these (usually public/state) areas of responsibility. Attention was also paid to preventive measures, which led to the inclusion of occupational health promotion issues and issues related to the situation of older workers (occupational health usually relates to persons younger than 65). Another issue that appeared important for completing the picture was the media involvement in health promotion (TV, radio, press, and new media -- Internet), which might play a significant role, especially in health marketing. The role played by education and sport institutions also needed to be taken into account. The institutional approach was further enhanced, in light of competencies concerning health promotion activities, by the basic differentiation of research into three main stages: (A) health policy concept, creation, standards and plans (using necessary instruments, procedures and supporting incentives of different kinds); (B) policy/strategy introduction into the system (formal -- in the form of adopting legislation or executive regulations, and declaratory -- in the form of different policy documents, strategic plans, guidelines, standards and programmes); (C) implementation of policy, practical application/introduction and monitoring, control and surveillance, and the final process of assessment and evaluation. The analysis presented here was performed in the following two major steps: (I) identification of sectors most engaged in HP4OP in a group of EU countries, and (II) acknowledgment of country specific sectors and institutions supported by the survey. A dedicated survey instrument -- a questionnaire with a range of specific questions -- was prepared in order to supplement and confirm the information for the country-specific perspective regarding the engagement of selected sectors in HP4OP activities obtained from the literature review. Bearing in mind that the information and data to be collected, were supposed to relate to a broader scope of health promotion, two aspects were addressed to the respondents: health promotion in general and HP4OP specifically. This meant that most of the questions had two sections for answers: relating to health promotion in general, and HP4OP specifically. The expert respondents, selected by the project's collaborating partners, were also asked to fill in the questionnaire concerning HP4OP institutional aspects. The pilot version of the questionnaire was consulted and pre-tested with project partners in Germany, the Netherlands and Italy. After receiving some minor feedback, its fine-tuned, final version was distributed among the expert respondents indicated by the project partners in the ten European countries. The questionnaire was sent by mail. The questionnaire itself and all communication with experts were conducted in English. All countries' responses have been received and reviewed. The applied version of the Questionnaire may be found on the project website: [http://pro-health65plus.eu](http://pro-health65plus.eu/). No language problems have been noticed. The whole process took about four months. The final version of the questionnaire was composed of ten main questions (see Table [1](#Tab1){ref-type="table"}) with empty fields for answers, mostly divided into 2 sections (health promotion and HP4OP), as mentioned above. In order to ensure clarity for respondents and to provide better structure concerning the responses, the questions were located within a table in a word-processor file. The list of questions, as they were posed in the questionnaire, is presented in Table [1](#Tab1){ref-type="table"}.Table 1The English wording of the questions included in the questionnaireQuestions*1) Indicate the 2--3 most important sectors in your country where there are institutions/organisations providing health promotion functions generally, as well as Health Promotion specifically addressed towards older people.2) Identify the 2--3 most important, particular institutions/organisations affecting health promotion functions generally and addressed at older people in a given country in sectors indicated previously (in section (1)).3) Are there any regulations formally obligating health promotion activity in your country (e.g., Public Health Act, Health Promotion Act)? If yes, indicate the names of those acts, year(s) when they were passed and links. If there are such regulations that only apply to particular sector(s), then please indicate below the name of the sector and the relevant names of acts, and year(s) when they were passed and links.4) Is there in force, a country-wide, general, official, long-term National health programme (Strategy, Plan...)? If so, indicate its name, time scope and year when it was passed; if NOT -- please indicate other relevant sectorial programmes/strategies/plans.5) If such an official act or strategy/plan exists -- does it encompass issues of Health Promotion addressed towards the older people? (Refers to HP4OP column only).6) If there is/are regulation(s) as above (in section (4)) -- do they enumerate/indicate any specific groups of people towards whom Health Promotion should be addressed?7) If such a programme(s) exist(s) -- what is its/their practical transmission (real causative influence) on the functioning of the organisations carrying out (or obliged to carry out) Health Promotion activities addressed towards older people (refers to HP4OP column only).8) How are health promotion activities funded in your country? Who is the main public funder of those Health Promotion activities? Are there any non-public financial resources spent on them?9) Are there evidence-based knowledge approaches used in the planning and shaping of the main Health Promotion programmes? Which kind of expertise is used? Who provides it?10) Which are the most relevant national documents (reports, surveys, analysis, papers, etc.) which may be helpful for supplementing/enlarging the above information concerning the country?* The first part of the survey (questions 1 and 2 in Table [1](#Tab1){ref-type="table"}) aimed at identifying the sectors and particular organisations most active in health promotion in general and HP4OP in particular. The respondents had eight possibilities listed there: seven sectors -- health, social, government (central), regional/local authorities, voluntary/NGO, education & sport and media, as well as an empty space to indicate other sector(s) if so required. Questions 3 through 7 (see Table [1](#Tab1){ref-type="table"}) were focused on the existence of formal regulations and official documents such as legislation, strategies and national plans -- relating to both health promotion in general and HP4OP specifically in a given country. This was done in order to get information concerning the presumed diversity of the formal approach to this issue. The last question in this section was only related to HP4OP and concerned respondents' opinions about the real causative influence of the organisations oriented towards older people. Again, as a consequence of the supposed variety in different countries, question number 8 referred to the main sources of public and private financing of health promotion and HP4OP activities. Thus, respondents were asked whether there was an evidence-based approach to planning and shaping health promotion programmes -- and if yes, how it was provided and by whom. Finally, there was a request to indicate the sources of data for the most relevant national documents. This was expected to help in completing or increasing the above information concerning a particular country. Respondents were also provided with a closing section for any additional comments or supplementary information. The main results obtained from this survey, as well as the final conclusions -- which give equal weight to the analysis of the literature review -- were analysed in view of the research goals of the paper. Results {#Sec5} ======= Literature review results: institutional approach to HP4OP {#Sec6} ---------------------------------------------------------- The literature review provides an analytical insight into the organisational aspect of HP4OP. First of all, the great differences between countries should be indicated here, with regard to the institutional dimension, the structure and nature of the institutions involved, the scope of competencies, the scale of activities, the size of the institutions themselves, and, of course, the financial resources available (especially in respect to health promotion). The institutional definition of health promotion is commonly based on two compounds: the first one focuses on the different activities undertaken in the field of health promotion (the functional approach -- see Table [2](#Tab2){ref-type="table"}), whereas the second concentrates on the different kinds of institutions engaged in the problem (the sector typology approach -- see Table [3](#Tab3){ref-type="table"}). Regarding the main activities in question, further division into two categories may be suggested accordingly to the general and specific addressees, i.e., the population as a whole and the older population. In relation to the second category and based on the WHO documents, there are different kinds of health promotion activities specifically dedicated to the older generation, such as walking for health, healthy nutrition, smoking cessation and mental health protection - stress avoidance, emotional intellectual skill promotion, development and intellectual activity maintenance. Apart from those, other important activities for older persons include: the prevention of sexually transmitted diseases, health instructions by the education system, health in workplace protection as well as occupational diseases prevention, health care within health care and social care - units (healthy senior homes/nursing homes), prevention of infections connected to living conditions in public institutions, the presence of healthy life understood as a value in every community, community relationship support and social integration, information on different health risks and health oriented behaviours, explanation of understanding health determinants and healthy life style in media promotion.Table 2Roles performed by institutions for HP4OPSPOFER roleDescription of functions performed by institution for a HP4OP programme:(S) SettingThe given institution constitutes a health promotion setting.(P) PromoterThe institution (its personnel) implements the programme as street-level promoters, educators, informers or advocates.(O) OrganiserThe institution is responsible for organisation of a given intervention by initiating, providing administrative support, coordinating actions, managing, etc.(F) FinancingThe institution provides funding (entirely or partly) for the given intervention.(E) Expertise & evaluationThe institution guarantees the proper evidence-based quality of health promotion intervention by providing: guidelines, knowledge, advisement, training, collecting and sharing experiences, but also by evaluating results, etc.(R) Regulation, monitoring & controlThe institution provides legal regulations, monitoring and control: through supervision, registration or by issuing obligatory approval.Table 3Who undertakes activities in the sphere of health promotion for older people? The institutional analysis scope and sketched results*SectorMAIN Institutions with health promotion functions indicated for the project purposesStreet level health promoters- professionals, health care and public health specialistsPlace of settingTarget groupsParticular actions (examples) -- where, how*Health (health care sector understood as involved mainly in diagnostics, treatment, prophylactic processes)GP/Primary care, Organisations, InsurersGPs, Nurses, Public health professionals, PhysiotherapistsHealth centres/units Patient's homesOlder PatientsWithin service delivery -- oriented on health conservation, improvement, postponing of worsening health condition, promotion of expected life style (improving health -- diet/physical activity recommendation)Occupational therapists, Dieticians, Exercise counsellorsPharmacists, Opticians/optometrists, Speech and language therapistsEducation/Education offices/institutionsTeachers, pedagogy specialists, Sport trainersSchools, other educational institutions, Sport clubs, Sport centresPopulation by ageEducational programme realisation, sport/physical activity support and organisationSportSports organisations/clubs/associationsSocial AssistanceSocial ServicesSocial workers, therapists, officialsDifferent settings (depending on the particular activity)Vulnerable older peopleAccompanying social service delivery, direct contact with professionals (advocacy for life style/habit change, personal support)Environmental nursesGovernmentalNational public health agencies/organs/bodiesPublic health professionals, EpidemiologistsDifferent settings (depending on the particular activity)PopulationProgrammes, research, policy/strategyRegional/Local AuthoritiesRegional/local public health departmentsPublic health professionals, Teachers, Play workers, Community workers, Social workers, Environmental health officersDifferent settings (depending on the particular activity)Population by ageStrategies and policies at the local level, activities undertaken by particular professionals at the local level (local government. initiatives)EnterpriseHealth and safety at workplace services (inspectorates), Trade unions and workers organisations, Employers organisationsOccupational medicine specialistsCompanies/workplaceOlder employeesRegular worker check-ups, diagnostics and other services performed by occupational medical services and professionals, Programmes/trainings organised at the workplaceInspectorsOccupational medicine unitsActivists/educatorsNGO/VoluntarySocial and civic organisations -- NGOsNGO activists, Public health professionals, Trade union safety representatives, Pressure groupsDifferent settings (depending on the particular NGO and particular activity)Groups of the older populationActions of different kinds addressed to the older population in need in different settings (determined by the NGO type and mission)MediaMedia organisationsJournalists - Health correspondentsDifferent media (press, audio and TV programmes, internet)The population generally and seniors particularlyMedia designing/participating/supporting programmes/actions supporting health promotion for older persons.(Source: own elaboration) The institutional and organisational aspect of health promotion is an important part of the wider perspective of public health functions. It is essential in functions such as: the creation of standards, evidence collecting, evaluation and monitoring, financing, communication and cooperation, as well as leadership \[[@CR4], [@CR30]--[@CR32]\]. Definitions adequate to the goal of the study are elaborated for "health promotion" and "intervention", as well as "institution" itself. This in turn, leads to the development of the analytical framework for the institutional approach to HP4OP. The definitions of health promotion include normative aspects present in international declarations and policy documents but also various (traditional or modern) approaches to health promotion \[[@CR31], [@CR33]\]. These definitions have evolved and transformed over decades to accommodate new health challenges \[[@CR34]\]. The following proposition was developed within a definitional framework suggested by Jill Maben and Clark \[[@CR35]\] as a concise presentation of relevant aspects. It was rearranged and amended with reference to newer publications and international policy documents. According to the literature review, *health promotion* -- as a core function of public health -- is the process of improving people's health status by enabling them individually, and also within a community and through the political system \[[@CR36]\] -- to increase control over their health and its determinants \[[@CR37]\]. Health promotion is a unifying concept -- an umbrella term \[[@CR38]\] -- encompassing various activities (prevention, education, and advocacy) that should: respect human autonomy \[[@CR31], [@CR33]\], be sustainable, evidence-based, and be adjusted to the specificity of the target group and local context \[[@CR36]\]. It should be performed as a concerted action by various entities from all sectors and on all levels of governance, thus effectively engaging all available resources \[[@CR36], [@CR39]\]. A successful health promotion intervention therefore requires the following key functions that are provided by institutions: setting and promoting (core functions) as well as organising, funding, providing expertise and evaluation, and regulation (see Table [2](#Tab2){ref-type="table"}). This definition is devised to encompass the widest possible range of activities actually performed under the term "health promotion" (descriptive aspect), and also recommended activities (normative aspect) in order to be operationalised and serve as an analytical tool -- hence the "umbrella approach" \[[@CR38]\]. It focuses not on activities but rather on issues of management, organisation, actors' involvement, functioning, etc. Thus, there is a strong incorporation of policy-oriented definitions \[[@CR36], [@CR39]\] that contain a set of key organisational functions or roles that institutions can perform in HP4OP. As seen in the literature review, health promotion *activities* (the processes of health promotion as defined above) can take various forms. The most general is *intervention*, i.e., "interference" in the current state of affairs -- an intrusion aimed at change (behavioural, social, environmental or policy change). In this case, health promotion is not clearly distinct from other activities. However, the most organised and clearly distinguishable form is the health promotion *programme* -- a set of organised activities strictly designed with the health promotion objective in mind, which are limited in terms of their scope, organisation, engaged actors and duration. For the purpose of this study, the choice was made to define an *institution* as "an interlocking double-structure of persons-as-role-holders or office-bearers and the like, and of social practices involving both expressive and practical aims and outcomes" \[[@CR2]\]. It should be noted that in the health promotion literature, the term "institution" is usually restricted to organisations providing social or other types of care on a daily basis, including those that are based on involuntary commitment, like psychiatric wards or prisons: "penitentiary institutions" \[[@CR40]\]. The latter are considered as one of the possible settings for health promotion. Furthermore, the term "institution" occurs in concepts like "institutional setting" as opposed to workplace setting, community setting or health facility setting \[[@CR41]\]. A definition of an institution as an organisation \[[@CR42]\] need to be adopted in the analysis of HP4OP for the identification of institutions that perform crucial roles towards the realisation of health promotion activities for older people. The focus should be on formal organisations -- those with defined objectives; but without the exclusion of the informal ones, since they can also be involved, even in formalised programmes. Not only are homes and local communities frequently a setting for health promotion, but also informal networks and groups that are involved in organisation and promotion. Overall, health promotion is a core part of the work of many different institutions, inside and outside the health sector, which operate in different political, economic, social and legal circumstances (see Table [3](#Tab3){ref-type="table"}). Literature review results: systemic context {#Sec7} ------------------------------------------- An important result of the literature review was the overview of the political and historical context of HP4OP in Europe. The traditions of different socio-economic approaches (models of welfare-states) and institutional structures of political systems shape the level of commitment to the most extensive approaches to health promotion (health advocacy). These factors determine which institutions (belonging to which sectors and levels of governance) are involved in health promotion. In this context, scholars classify welfare states as "social democratic" (Nordic countries), "liberal" (e.g., UK), "Latin" (e.g., Italy, and Greece), and "conservative" (e.g. France, Germany, and Belgium), which is based on the modified Esping-Andersen classic typology of social policy models \[[@CR43], [@CR44]\]. However, post-Soviet countries also have their own set of experiences rooted in the Semashko model of health-care system \[[@CR45], [@CR46]\]. In some countries, the main responsibility for health promotion is attached to the central government; in others to the territorial self-government. The involvement of various sectors depends on the approach. For instance, "conservative" and "liberal" states often utilise the participation of non-governmental and even for-profit private institutions whereas "social-democratic" Scandinavian countries prefer the involvement of decentralised public authorities \[[@CR43]\]. Also, in Central and Eastern Europe and the countries of the former Soviet Bloc, health promotion was, and still is, neglected \[[@CR45]\]. This remains a problem even despite a strong focus on preventive medicine (secondary prevention, often ineffective screening \[[@CR46]\]) and the significance of the SANEPID health promotion services (primary prevention) \[[@CR47], [@CR48]\]. It is perhaps due to the association of "public health" with the central public authorities that creates barriers to inter-sectorial action. Significant disparities can be observed here, not only in respect to economic development but also regarding political ideologies that have been determining the health promotion strategies for decades. This impact has lasted long after the systemic transitions \[[@CR49]\]. Central and Eastern European countries -- when compared to the "old" EU countries -- have underdeveloped and fragmented health promotion. Consequently, little attention is paid to non-communicable diseases and mental health \[[@CR47]\]. The problem of disparities concerns not only activities but also appropriate expertise and research on health promotion \[[@CR49]\]. The literature review also suggests that currently, governing political parties shape the approach to health promotion. Economically liberal and conservative governments tend to appreciate individualistic approaches that focus on health education and individual lifestyle choices while opposing health advocacy that could actually make healthy choices easier. This policy orientation drives those countries to a stronger emphasis on the role of NGO and commercial sectors but on contractual and non-obligatory terms. Such a tendency is more common than international declarations would suggest \[[@CR50]\]. While declaratory commitment to health promotion is strongest in social-democratic (Nordic countries) and "liberal" (e.g. UK) states, the latter are less likely to put policies into practice \[[@CR49]\]. Identification of health promotion leadership at a national level is problematic due to significant local differences \[[@CR51]\]. An important issue related to the legal basis in respect to the research undertaken, is the creation of a health promotion glossary defining crucial terms and frames in this area \[[@CR52]\]. Such glossary is necessary to provide terminology that should be used in legal documents and can have a supranational harmonising impact, stimulating the new approach to public health \[[@CR53]\]. In terms of legislative actions, laws on health, public health and health promotion originally served to define and justify public health measures and the scope of the responsibility of the state in this area. In this context and regarding the supra-national level, very general international acts have to be mentioned with particular attention to the WHO international legal initiatives: The Ottawa Charter and The Bangkok Charter \[[@CR36], [@CR39]\], as well as the 2009 WHO's *Milestones in health promotion* \[[@CR54]\]. In the 21^st^ century, another influential element of legislation can be identified, namely the cross-sectorial impact on health, in accordance with the "Health in All Policies" idea, based on the World Health Declaration. Recently, in 2015 at the 67th World Health Assembly (Resolution WHA 67.12), WHO requested the WHO Director-General to prepare, for the consideration of the 68th World Health Assembly, a "Framework for Country Action Across Sectors for Health and Health Equity", which could be used for different purposes, with regard to the Health in All Policies document. Accordingly, a draft framework was developed in three rounds of informal consultations, which is expected to culminate successfully in a form of a new international regulation \[[@CR55]\]. Regarding internal legislation however, only a minority of European countries (mainly in northern Europe) have established laws that relate specifically to health promotion. As an example, the Finnish legislation should be highlighted \[[@CR56]\]. This act emphasises, among other things, that the "*Health During Working Life* and *Health in Old Age policies, including the psychophysical demands of work, are health-promoting and appropriate for workers of different ages and preconditions for promoting the health of older people and for reducing the health differences created by reducing prejudices and attitudes contributing to age discrimination"*. With regard to older workers in particular, the fact that they have increased in number and gained importance only in recent years, explains that existing promotion programmes specifically targeted at this group of people are not sufficient \[[@CR57]\]. Literature review results: overview of activities and institutions {#Sec8} ------------------------------------------------------------------ Following the literature review on country specificity, the emerging picture of the selected countries' institutional structures for HP4OP is quite incomplete and heterogeneous. The scientific literature review showed that a disproportionally large amount of institutional information is available on the Netherlands and Germany. At the same time very little is provided on Hungary, Portugal, Bulgaria and Lithuania. Often, most notably in the scientific literature, a precise identification of the number of institutions involved was not possible due to the vagueness of information. In some cases, only a group of institutions in the same category was recorded. Also, a lack of sources concerning the issue in some countries does not always signify a lack of such activities. This also results from a shortage of substantial evaluation of HP4OP in given countries, or a lack of publicly available report on such evaluations (at least in English). The literature review indicates an abundance of literature devoted to the characteristics, efficiency and performance of various HP4OP programmes. Scientific papers usually focus on the content of health promotion rather than on its institutional arrangements. Interestingly, the grey literature and project web databases provided much more substantial information on the issue, whereas scientific literature usually delivered a mostly sketchy and incomplete picture of institutional arrangements since it was not dedicated to institutional analysis. Eventually, the literature review shows that the grey literature and dedicated health promotion databases, as well as ongoing pilot surveys and interviews, are much more effective data-sources on institutions involved in health promotion for the older people in the analysed countries. As expected, the literature review points out seven main sectors potentially involved in health promotion, and specifically health promotion addressed to the population aged 65+ (except for the workplace sector): (1) health sector, (2) social sector (3) central and local government, (4) workplace/enterprises/employers, (5) NGOs & voluntary organisations, (6) sport & education, and (7) media. A great degree of inter-sectorial cooperation is identified. Frequently programmes engage more than one institution and often, these institutions belong to different sectors. Questionnaire survey results {#Sec9} ---------------------------- Experts' opinions -- supplemented by the literature review results -- helped to identify the leading sectors, i.e. those sectors that are (practically and formally) more engaged in or responsible for HP4OP than others in each of the analysed countries. This reveals quite a complex picture of the diversity of sectors and organisations/institutions active in this field in the countries. The results of the questionnaire survey reveal a few quite interesting details. First, the presumption of a diversity of sectors and categories of organisations most engaged in HP4OP in the analysed countries is confirmed. Next, the sectors that are most frequently indicated as being principally active in HP4OP in the majority of cases are:Regional/local authorities (in 9 countries): Germany, Italy, the Netherlands, Poland, Bulgaria, Greece, Hungary, Portugal and LithuaniaHealth sector (in 8 countries): Germany, Italy, the Netherlands, Poland, Bulgaria, Greece, Portugal and Lithuania.Voluntary/NGO organizations (in 5 countries): Poland, the Czech Republic, Greece, Portugal and Hungary. A general overview of those results are presented in Table [4](#Tab4){ref-type="table"}. This table shows a summary of results: the most active sectors in each country are marked as \"Most important\" while sectors in given countries that are still active in HP4OP, but of less importance, are marked as \"Important\". As shown in the table, the sectors indicated as less relevant are: enterprise (2 countries: Italy and the Netherlands) media (3 countries: the Netherlands, Bulgaria and Lithuania) and education & sport (5 countries: Germany, Italy, the Netherlands, the Czech Republic and Lithuania). In all these cases, there are only weaker oriented (secondary) HP4OP activity compared to those indicated as most relevant. Interestingly, the social assistance sector is only indicated once as the most active in HP4OP (in the Czech Republic), plus 4 times in a secondary position (in the Netherlands, Bulgaria, Greece and Lithuania).Table 4Outline of the most important sectors for HP4OP in selected EU countries1.2.3.4.5.6.7.8.SECTOR→HealthEducation & SportSocial AssistanceGovernmentRegional / LocalEnterpriseVoluntary / NGOMedia*Sectors chosen for further analysis - per country*\ *(number of sectors chosen)*Country:↓*I*BGMost importantImportantMost importantImportantImportantHealth, Government, Regional/Local\ (3)*II*CZImportantImportantMost importantImportantImportantMost importantSocial Assistance, Government, Voluntary/NGO\ (3)*III*DMost importantImportantImportantMost importantHealth, Education & Sport,\ Government\ (3)*IV*GRMost importantImportantImportantMost importantMost importantHealth, Government, Voluntary/NGO\ (3)*V*HImportantMost importantMost importantGovernment, Regional/Local, Voluntary/NGO\ (3)*VI*IMost importantImportantImportantMost importantImportantImportantHealth, Government, Regional/Local\ (3)*VII*LTMost importantImportantImportantMost importantImportantImportantHealth, Government, Voluntary/NGO\ (3)*VIII*NLMost importantImportantImportantMost importantMost importantImportantImportantImportantHealth, Education & Sport, Government\ (3)*IX*PMost importantMost importantMost importantMost importantHealth, Government, Regional/Local\ (3)*X*PLMost importantImportantMost importantMost importantHealth, Government, Regional/Local, Voluntary/NGO\ (4)*Number of counties where a given sector has been chosen for analysis*821105-5- The table also shows the level of diversity of sectors engaged in HP4OP in each of these countries, according to the survey. For example, the Netherlands, the Czech Republic, Lithuania and Italy seem to have these activities spread over a range of different sectors (as many as 6--8), while in other countries they are "concentrated" in fewer sectors, e.g. in Hungary (2 sectors) or Poland (3 sectors). Only in the case of Germany, one additional sector is indicated under the category of "o*thers*": namely, National Cooperation Networks (National Health Targets, Equity in Health, Healthy Cities Network). No other sector is added to the list in any of the analysed countries, which may indicate that the range of sectors chosen for HP4OP analysis is comprehensive. There is a range of national regulations concerning HP4OP issues, such as the national strategy of healthy ageing, for example in Italy, Bulgaria and Poland. In other countries, such documents are under construction (Greece). In yet other countries, there are already nationwide long-term national health programmes (strategies, plans), including references to HP4OP, for example in Germany and Lithuania. However, there are a few countries where there are no such dedicated official documents, such as the Czech Republic and Hungary. Opinions about the implementation of the contents of such documents in practice differ from country to country, but in several of them, for example in Bulgaria, Germany, Greece, Italy and Poland, the effects are reported to be partial and the results limited. A low level of financing, the lack of staff and inefficient coordination is used to explain this. In the Netherlands, the situation seems to be better due to the government commitment and subsidies dedicated to different aspects of HP4OP. The funding of HP4OP activities is attributed to both central and regional governments, and is supplied through taxation or by national health insurance/funds in the majority of the countries (Bulgaria, Greece, Hungary, Italy, Lithuania, the Netherlands and Poland). NGOs and different foundations are reported as being involved in HP4OP financing as well, for example in Bulgaria, Greece and the Netherlands. In some countries, these organisations were indicated as beneficiaries of public funds and/or EU grants (the Czech Republic, Greece, Lithuania, and the Netherlands). Germany has strictly defined funders of health promotion in general: statutory health insurance, public budgets, statutory accident insurance, employers and private households. The application of the concept of *evidence-based knowledge* is rather rarely reported regarding the evaluation of health promotion in general and HP4OP programmes in the investigated countries. Only in a few cases, there are dedicated national organisations (agencies) responsible for setting standards and assessing such programmes (in Germany, Lithuania and, to some extent, in the Netherlands). This survey -- especially its final question -- indicated several additional sources of information on the survey topic. The expert respondents however point out that although many organisations even issue their internal evaluation documents in English, a number of these sources, especially on the local level, are mostly in the original, native languages and are not translated. Combined results on organizations involved in HP4OP and their analysis (review and survey) {#Sec10} ------------------------------------------------------------------------------------------ The survey results are consistent with the presumption that the European countries approach HP4OP differently. Also the involvement of different sectors and a broad range of institutions and providers has been confirmed. Such a multi-faceted picture corresponds to the countries' diversity in the organisation of health systems as well as other national systems (regulatory, administrative, economic, social, etc.). HP4OP is also executed at diverse levels of government and administration (central, regional and local) with a varying degree of involvement \[[@CR1], [@CR3], [@CR11], [@CR16], [@CR17], [@CR23]--[@CR29], [@CR58], [@CR59]\]. It should be noted that the questionnaire results described above do not have stand-alone value. They may be treated as an important, but only partial contribution. Together with the literature review, the results create the preliminary picture of the institutional approach to HP4OP in the analysecd countries. It seems that such a combined approach advances knowledge and can set the direction for further, in-depth research on this vital issue of health promotion for older people in European countries. Thus, based on the literature review and questionnaire survey results, a set of organisations, most involved in HP4OP have been identified in the countries of interest. This comprises the following sectors per country:Bulgaria-Health, Government, Regional/Local AuthoritiesCzech Rep.-Social Assistance, Government, Voluntary/NGOGermany-Health, Education and Sport, GovernmentGreece-Health, Government, Voluntary/NGOHungary-Government, Regional/Local authorities, Voluntary/NGOItaly-Health, Government, Regional/Local AuthoritiesLithuania-Health, Government, Voluntary/NGONetherlands-Health, Education and Sport, GovernmentPortugal-Health, Government, Regional/Local Authorities,Poland-Health, Government, Regional/Local Authorities, Voluntary/NGO It should be mentioned however that the sectors listed for each country do not fully overlap with the survey results. This is an aggregate result of not only a confrontation with the literature review, but also a consequence of certain methodological reasons. Above all, the decision was made to include the analysis of the central government sector for all countries. In the case of Germany and the Netherlands, the education and sports sectors replaces the local self-government sector. In case of Lithuania the NGO's sector is selected as relevant as well. Media and enterprise sector are not defined as the most important factor in HP4OP in any country. The presented institutional approach provides guidance for further analyses of the HP4OP institutions with the sectors listed above. Within the health promotion definition (see above), a set of key institutional functions -- required for successful HP4OP intervention -- can be identified: setting (S), promoting (P) (core functions) as well as organising (O), funding (F), providing expertise (E) and regulating (R) with evaluation and control (see Table [2](#Tab2){ref-type="table"}). These sets of functions together form a framework (SPOFER) that can serve future studies in two ways: firstly as a classification of roles that institutions can perform for any given HP4OP programme and -- secondly -- as a checklist of key roles required for any HP4OP programmes. Discussion {#Sec11} ========== The analysis presented here, concerning the institutions providing HP4OP in the selected European countries, undertaken with the use of a literature review and survey among experts, provides results relevant to both future research and policy. We find that the scientific literature provides a sketchy and incomplete picture of HP4OP. In contrast, the grey literature is a much more substantial source on the institutional dimension of HP4OP. We also find that there are significant HP4OP organisational differences in the group of analysed countries, linked to systemic differences: political, welfare state models, etc. Inter-sectorial cooperation is common in all countries as also confirmed in the survey. The need of a cross-sectorial approach is thus recognized. The original institutional approach applied in the present study has provided a clearer and more systematised picture of health promotion activities in terms of the institutions involved and roles they play. The literature review performed together with the survey among the group of experts from the ten European countries has provided a still limited, but already indicative picture of the sectors and institutions involved in HP4OP in these countries. Seven main sectors are potentially involved in health promotion: (1) health sector, (2) social sector (3) central and local government, (4) workplace/enterprises/employers, (5) NGOs & voluntary organisations, (6) sport & education, (7) media. The most engaged sectors in HP4OP in the examined countries have proven to be: health care, regional/local authorities and NGO's. The importance of this study concerns not only a wide range of results from the studied countries but also a comparative perspective on the institutional arrangement of HP4OP in Europe. This, at least partially, fills an existing gap in respective knowledge as outlined at the outset of the paper. The results confirm the main assumptions for a cross-sectorial approach to public health and HP4OP. Moreover, the role of different institutions and inter-institutional arrangements in HP4OP was stressed as fundamental in the study results. The final selection of sectors per country outlined at the end of the results section can be used for further in-depth analysis to better understand the health promotion activities in the selected countries. In view of the increasing trend of aging workforces, it is necessary to update information about policies and programmes implemented in the field of workplace health promotion in European countries. Taking into account the influence of the workplace during the final years of professional activity, and the obviously growing role of the media in the process of providing information and promotion of a healthy lifestyle, it is important not to leave out these two sectors in further research. Also, the central government (represented usually by the Ministry of Health, the Ministry of Labour and Social Policy, the Ministry of Sport, and the National Institutes of Public Health \[or the given countries' equivalents\]) is important to be studied in each of the countries. The reason is the key role in public health that is usually attributed to this stakeholder as well as the expectation to obtain overarching, complementary guidelines for in-depth country specific research. Consequently, the next relevant stage of research is considered to be a comprehensive analysis of the most important sectors that are active in HP4OP in each of the countries, as listed above. Also, it is assumed that central governments -- as supervisory and regulatory institutions -- can provide a representative overview of the situation in each country. Through those institutions, we expect a further snowball effect in data collection. The engagement of experts from the investigated countries turned out to be exceptionally helpful as a complementary source of knowledge to the results of the literature review. Their opinions revealed aspects which enable properly fine-tuning actions in the next steps of research. Moreover, elaboration of a relatively complete and reliable picture of HP4OP in a given country without the support of local expert knowledge concerning the specificity of the country's situation and solutions, would not have been possible. The limits concerning research results, however, still have to be accounted for, namely the native language of some sources, the absence of data concerning specifically HP4OP in the literature and the overall problem of the non-existence of a common model for HP4OP activities. Certain study limitations grow out of the specificity and interdisciplinary characters of the issue. Most of those were indicated in the results and method sections. The authors believe that the evidence presented here forms quite a solid foundation for further in-depth analyses, which are planned as the next steps of the Pro-Health 65+ project. Conclusions {#Sec12} =========== The institutional approach applied in the present study has provided a clear and systematised picture of health promotion activities and institutions, specifically in the HP4OP sphere. This approach to HP4OP also enables the identification of existing knowledge gaps in this area that can be addressed in further research. The results show the necessity of adopting a cross-sectorial approach to explore the role of different institutions and inter-institutional arrangements in HP4OP. The engagement of country experts to supplement the literature review concerning the HP4OP in individual countries proved to be a well-reasoned approach. It can provide guidance and points of reference for furture in-depth analysis. According to this study, the sectors most engaged in HP4OP are: health care, regional/local authorities and NGO's/voluntary organizations, all of them also being strongly related to each other regarding health promotion activities. Health programmes are often implemented with the participation of all three sectors as they are responsible for different aspects and cooperate in respect to similar or common tasks. Interestingly, the role of the social assistance sector has been emphasised in a few of the analysed countries, as being most important for HP4OP, and in a majority of cases, as being of secondary importance. According to the results of this study, the workplace is not considered the most relevant institution in HP4OP interventions. Nevertheless, the scientific and grey literature review indicates the essential problem: at the workplace, age discrimination should be eliminated through adequate policies, the promotion of lifelong learning, the adaptation of work demands and the environment and the promotion of health and wellbeing. These should serve to ensure a longer working life and higher employment rates amongst older workers. Furthermore, it is difficult to unequivocally determine the role of the media in HP4OP. The literature review clarified the issue to some extent, namely the aggregated scientific evidence showing global or focused media interventions in the analysed countries does not exist. Neither did experts indicate media as an important factor for HP4OP in any of the analysed countries. Bearing in mind the specificity of HP4OP and the group of beneficiaries as main addressees of TV and radio programmes, further studies are necessary in this context. A dedicated research tool and the support of local experts would be necessary, as well as deeper desk research. There are not many examples of separate HP4OP actions within the sport sector, although a rising awareness of the importance of this area may be observed. Still, a few very good examples of sport projects dedicated to supporting and enhancing older people's health in some countries may be indicated. At the stage of research reported here, it is assumed that HP4OP is rather "spread out" amongst sport and educational health promotion activities that are undertaken by both governmental and non-governmental bodies. Again, this point is to be more deeply examined in future studies. Abbreviations {#Sec13} ============= GP, general practitioner, family doctor; HP4OP, health promotion for older people; NGO, non-governmental organisation; SANEPID, sanitary & Epidemiological surveillance (a body originating from the health systems of Soviet Bloc countries). This publication arises from the project Pro-Health 65+ which has received funding from the European Union, in the framework of the Health Programme (2008--2013). The content of this publication represents the views of the authors and it is their sole responsibility; it can in no way be taken to reflect the views of the European Commission and/or the Executive Agency for Health and Consumers or any other body of the European Union. The European Commission and/or the Executive Agency do(es) not accept responsibility for any use that may be made of the information it contains. For more information about Pro-Health 65+, see: [http://www.pro-health65plus.eu](http://www.pro-health65plus.eu/). Publication co-financed from funds for science in the years 2015--2017 allocated for implementation of an international co-financed project. The authors express special gratitude to all experts who supplied information about health promotion and HP4OP for each country. The work of student-volunteers who helped prepare the literature review is also appreciated. Declarations {#FPar1} ============ This article has been published as part of BMC Health Services Research Volume 16 Supplement 5, 2016: Economic and institutional perspectives on health promotion activities for older persons. The full contents of the supplement are available online at <http://bmchealthservres.biomedcentral.com/articles/supplements/volume-16-supplement-5>. Funding {#FPar2} ======= This publication arises from the project Pro-Health 65+ which has received funding from the European Union, in the framework of the Health Programme (2008--2013). The content of this publication represents the views of the authors and it is their sole responsibility; it can in no way be taken to reflect the views of the European Commission and/or the Executive Agency for Health and Consumers or any other body of the European Union. The European Commission and/or the Executive Agency do(es) not accept responsibility for any use that may be made of the information it contains. Publication co-financed from funds for science in the years 2015--2017 allocated for implementation of an international co-financed project. Availability of data and materials {#FPar3} ================================== The tools used, data and information on which the analysis has been done and conclusions of the manuscript are to be found in the ProHealth65+ Project materials on the web site of the project [http://pro-health65plus.eu](http://pro-health65plus.eu/) and/or in CHAFEA documents relating to this Project. <http://ec.europa.eu/chafea/projects/database.html?prjno=20131210> Authors' contributions {#FPar4} ====================== SS contributed to the development of the study design, elaborated the questionnaire, collected and analysed the data, drafted, improved and guided the improvement of the manuscript, approved the final version and agreed to be accountable for his contribution. IKB contributed to the development of the study design, was responsible for the systemic context of the literature review, reviewed and commented on the drafts and final version of the paper, approved the final version and agreed to be accountable for her contribution. AM contributed to the development of the study design, reviewed and commented on the drafts and final version of the paper, approved the final version and agreed to be accountable for her contribution. MZJ contributed to the development of the study design, carried out the literature search and analysis, reviewed and commented on the drafts and final version of the paper, approved the final version and agreed to be accountable for his contribution. AD contributed to the development of the study design, approved the final version and agreed to be accountable for her contribution. NM carried out the literature search, commented and approved the final version and agreed to be accountable for his contribution. AP carried out the literature search, commented and approved the final version and agreed to be accountable for his contribution. MR contributed to the literature review, approved the final version and agreed to be accountable for his contribution. AS contributed to the literature review, approved the final version and agreed to be accountable for her contribution. SG contributed to the development of the study design, contributed in development of the manuscript, commented on the drafts, approved the final version and agreed to be accountable for her contribution. Authors' information {#FPar5} ==================== SS PhD, Health Policy and Management Department, Institute of Public Health, Jagiellonian University Medical College, Kraków, Poland IKB PhD, Health Policy and Management Department, Faculty of Health Science, Jagiellonian University Medical College, Kraków, Poland AM PhD, Health Policy and Management Department, Institute of Public Health, Jagiellonian University Medical College, Kraków, Poland MZJ Research and teaching assistant, Health Policy and Management Department, Institute of Public Health, Faculty of Health Science, Jagiellonian University Medical College, Kraków, Poland. NM Aggregate Professor of Occupational Health at the Institute of Public Health, Università Cattolica del Sacro Cuore, Rome, Italy. AP PhD, Institute of Public Health, Università Cattolica del Sacro Cuore, Rome, Italy. AD PhD, Department of Health Policy and Management, Institute of Public Health Faculty of Health Sciences, Jagiellonian University Medical College, Kraków, Poland. MR PhD, Health Policy and Management Department, Institute of Public Health, Jagiellonian University Medical College, Kraków, Poland. AS Senior lecturer, Health Policy and Management Department, Institute of Public Health, Faculty of Health Science, Jagiellonian University Medical College, Kraków, Poland. SG Professor of Economy, Head of the Health Economics and Social Security Department, Institute of Public Health, Faculty of Health Science, Jagiellonian University Medical Collage, Kraków, Poland. Competing interests {#FPar6} =================== The authors declare that they have no competing interest. Consent for publication {#FPar7} ======================= Not applicable. The paper does not contain any individual person's data in any form (including individual details, images or videos) nor presentations of case reports. Ethics approval and consent to participate {#FPar8} ========================================== Not applicable. The paper does not report on or involve the use of any animal or human data or tissue. The respondents to the questionnaire have been participants of the Project ProHealth65+ in the framework of which this research has been completed and they took part in it of their own free will. Ethical approval is not required for this type of questionnaire. Endnotes {#FPar9} ======== None.

{#sp1 .271} {#sp2 .272} {#sp3 .273} {#sp4 .274}

Introduction {#Sec1} ============ Accommodation in the human eye is a dynamic process that originates from the contraction of ciliary muscle and results in the reshaping of the crystalline lens. According to the classic theory by Helmholtz^[@CR1]^, the accommodation occurs as the crystalline lens thickens and moves forward toward the anterior surface. In this process, the cornea changes only slightly^[@CR2]^. However, whether or not the posterior segment of the eye, including the vitreous chamber and the retina, changes during accommodation remains controversial. Changes of ocular axial length (AL) with accommodation are still a matter of controversy. Some studies found a small but significant transient increase in AL accompanying accommodation in youth and elderly people^[@CR3]--[@CR12]^, but others reported the AL remained constant during accommodation^[@CR13],[@CR14]^. Several techniques have been used to measure ocular AL during accommodation. With A-scan ultrasound, which requires contact of the probe on the corneal surface, the AL was reported to increase^[@CR3]^ or remain constant^[@CR13],[@CR14]^ during accommodation. In these studies, the effect of placing the ultrasound probe on the cornea could have influenced the results. The IOLMaster, which uses an average ocular refractive index that converts optical distances into geometric distances, has also been used to evaluate axial elongation during accommodation^[@CR5],[@CR7],[@CR8]^. As an improvement, the Lenstar measures each compartmental distance^[@CR6],[@CR9],[@CR10],[@CR12]^. Similarly, we^[@CR11],[@CR15],[@CR16]^ and others^[@CR17]^ previously developed ultra-long scan depth optical coherence tomography (UL-OCT) instruments that can measure the dimensions of each component of the whole eye individually during accommodation. However, changes in the shape of the crystalline lens during accommodation may result in changes of the effective refractive index. In turn, this could lead to a change of optical path distance and subsequent deviation of the measurement from the true change in AL^[@CR18]--[@CR21]^. That was the main limitation of the previous studies using IOLMaster and Lenstar^[@CR4]--[@CR10],[@CR12]^. Some research reported that the refractive index increases by +0.001 to +0.0015 per diopter (D) during accommodation^[@CR20],[@CR22]--[@CR24]^, while other studies found no change in the lens gradient refractive index^[@CR18],[@CR19]^. Recent studies have attempted to overcome this potential error by using the ocular biometry data of individual subjects to calculate the error in measured AL^[@CR6],[@CR9],[@CR10]^ or measuring AL immediately following accommodation or after accommodation ceased^[@CR8]^. Because the exact change in refractive index of the crystalline lens during accommodation is not known, the values of AL during accommodation in those studies still may not be the true value at that time. This cross-sectional and population-based study was performed using UL-OCT to investigate changes in axial biometry of pseudophakic eyes during pilocarpine-induced accommodation. Changes in axial ocular biometry, including the central corneal thickness (CCT), anterior chamber depth (ACD), intraocular lens thickness (IOLT), and vitreous length (VL) were analyzed. To avoid the measurement artifacts potentially caused by changes in the refractive index of the natural crystalline lens^[@CR20],[@CR22]--[@CR24]^, we enrolled patients with implanted intraocular lenses (IOLs), which have a constant refractive index. Considering that the ciliary muscle in aging human eyes still has the capability of contraction^[@CR25]--[@CR27]^, we used UL--OCT to evaluate the change in ocular AL during pilocarpine-induced accommodation in patients with implanted IOLs. Methods {#Sec2} ======= Study population and Ethics Statement {#Sec3} ------------------------------------- The study was approved by the Wenzhou Medical University review board, and written informed consent was provided by each subject. All patients were treated in accordance with the tenets of the Declaration of Helsinki. The right eyes of 25 patients (9 males and 16 females) were included in the study. Their ages ranged from 49 to 84 years (mean ± standard deviation: 68.3 ± 8.4 years). All subjects were recruited at the Affiliated Eye Hospital of Wenzhou Medical University, Wenzhou, Zhejiang, China. Cataracts were diagnosed in both eyes for each subject, and at least one eye underwent phacoemulsification surgery with implantation of an IOL (Akreos MI 60 aspheric micro-incision intraocular lens, Bausch & Lomb, Rochester, NY, USA). The UL-OCT measurements were performed 1 to 3 months after the operation. Exclusion criteria included post-operative best corrected visual acuity less than 0.5, inflammation, previous ocular surgery except cataract extraction and IOL implantation, contact lens wear, or other current ocular or systemic diseases. Instruments {#Sec4} ----------- As reported previously^[@CR11],[@CR15],[@CR16]^, we used a custom-built UL-OCT instrument to image the whole eye from cornea to retina through the pupil. Briefly, the light source had a center wavelength of 840 nm with a bandwidth of 50 nm. The spectrometer included a collimating lens with a focal length of 50 mm, an 1,800 line/mm transmission grating, an enlargement lens with a focal length of 240 mm, and a line array complementary metal-oxide-semiconductor camera (Basler Sprint spL4096-140k; Basler AG, Ahrensburg, Germany). The scan depth was 11.964 mm, while the axial resolution was 7 µm in air. To extend the effective scan depth, a switchable reference arm with 3 mirrors at different distances was utilized to sequentially acquire 3 frames. The two frames from the first and second reference mirrors were overlapped to obtain the entire anterior segment through the pupil, while the third mirror was for imaging the retina. The optical path distance between the first and the third mirror was 33.160 mm in air. Thus the whole eye image could be obtained by reconstruction of the 3 frames. A vision target consisting of a white Snellen letter E with a logMAR visual acuity size of 0.4 was displayed by a light-emitting diode on a black background. The target was mounted onto the sample arm and was congruent to the beam from the UL-OCT light source by a hot mirror. During image acquisition, the patients were asked to fix on the target with the measured eye while the other eye was covered. A cross-hair live view at both the horizontal and vertical meridians was applied to monitor that the measured eye was positioned correctly. This approach ensured that all scans were at the same sagittal section and that each measurement was conducted on the visual axis of the eye. Study Design and Examination {#Sec5} ---------------------------- The scan speed of UL-OCT was set as 10,000 A-lines/s and each frame consisted of 1,024 A-lines. Fifteen frames were obtained at baseline in approximate 1.5 seconds for each subject. The subjects were asked to take a short rest (approximate 30 s) before the measurement was repeated. Then accommodation was induced by 2 drops of 0.5% pilocarpine hydrochloride separated by a 5-minute interval. Thirty minutes after the first drop, the images were acquired again with the same protocol. All measurements were performed by the same researcher (DH). Data Analysis {#Sec6} ------------- Automated custom software was used to segment the full-range ocular images as described in our previous papers^[@CR11],[@CR16]^. At first, every 3 frames were reconstructed to obtain one whole eye image. After removing the saturated A-lines resulting from the specular reflection on the corneal apex, longitudinal reflectivity profiles were obtained from the 20 central A-lines of each image. The structural boundary of each compartment was detected from the corneal anterior surface to the retinal pigment epithelium. Optical path lengths through the central cornea, anterior chamber, IOL, and vitreous were calculated. The geometric length of each compartment was determined by dividing the optical path length by the refractive index of each structure (refractive indices of the cornea, aqueous humor, IOL and vitreous were 1.387^[@CR28]^, 1.342^[@CR29]^, 1.460^[@CR30]^, and 1.341^[@CR29]^ at 840 nm wavelength, respectively). The AL of the whole eye along the visual axis was determined as the sum of CCT, ACD, IOLT, and VL. The results from five images were averaged as a single measurement for further analysis. Descriptive statistics were determined as means ± standard deviations. Bland-Altman plots were used to determine interclass correlation coefficients (ICCs) as a measure of agreement between the repeated measurements of each compartment. Moreover, the coefficients of repeatability (CoRs) were determined by dividing the two standard deviations of the differences between the repeated measurements by the average of the two repeated measurements^[@CR31]^. After determining that the data for each variable was normally distributed, two-tailed paired t-tests were used for statistical comparisons of lengths between the baseline and accommodative states, and P \< 0.05 was considered statistically significant. Validation for the measurement of the axial elongation using UL-OCT {#Sec7} ------------------------------------------------------------------- To validate the measurements of the axial elongation using UL-OCT, we used a modified physical model eye (OEMI-7; Ocular Instruments, Inc., Bellevue, WA, USA). It consisted of a poly-methyl methacrylate (PMMA) water-cell model that simulated the cornea and the lens (Fig. [1](#Fig1){ref-type="fig"}). The model eye was separated into two portions. The anterior portion, including the cornea and lens, was mounted within a water tank that was secured to a micrometer stage. To elongate the model eye, the stage was advanced along the ocular axis, extending the anterior portion while the posterior portion remained fixed in position. The model axis was elongated from 0 to 200 µm in steps of 25 µm, and UL-OCT images along the optical axis were captured after each step. To calculate changes in compartment thickness of the model eye, we used 1.485 as the refractive index of PMMA and 1.326 for water at a wavelength of 840 nm.Figure 1Photograph (**a**) and schematic diagram (**b**) of the physical model eye. A PMMA shell and lens simulated the cornea and crystalline lens. The anterior portion (cornea and lens) was inserted into a water tank, which was mounted on a micrometer stage that was moved along the ocular axis. The posterior portion was fixed on the platform by a holder. Accuracy of the axial elongation measurements by UL-OCT {#Sec8} ------------------------------------------------------- Using the eye model, we measured the differences between the pre-set elongation values and the UL-OCT-determined values. The mean difference was 4 ± 3 μm (range, −1 to 9 μm), and a Bland-Altman plot (Fig. [2](#Fig2){ref-type="fig"}) showed that the differences were within the 95% confidential interval.Figure 2Bland-Altman plots of the differences between the UL-OCT-measured and pre-set elongation provided by the micrometer stage. Values on the vertical axis correspond to the differences between the measurements and setting values, and values on the horizontal axis correspond to the setting values. Solid and dashed lines indicate the mean difference and 95% confidence intervals, respectively. Results {#Sec9} ======= Changes in ocular axial biometry during pilocarpine-induced accommodation {#Sec10} ------------------------------------------------------------------------- After pilocarpine-induced accommodation, the ACD increased from the baseline of 4.24 ± 0.31 mm to 4.33 ± 0.31 mm, an increase of +0.08 ± 0.09 mm (Table [1](#Tab1){ref-type="table"}, Fig. [3](#Fig3){ref-type="fig"}, P \< 0.01, paired t-test). The VL decreased from 17.85 ± 0.94 mm to 17.81 ± 0.95 mm, a decrease of −0.04 ± 0.09 (Table [1](#Tab1){ref-type="table"}, Fig. [3](#Fig3){ref-type="fig"}, P \< 0.05, paired t-test). CCT and IOLT remained constant during accommodation (P \> 0.05). Finally, the AL increased from 23.47 ± 0.93 mm at baseline to 23.51 ± 0.94 mm after accommodation. The overall increase was + 0.04 ± 0.04 mm (Table [1](#Tab1){ref-type="table"}, Fig. [3](#Fig3){ref-type="fig"}, P \< 0.01, paired t-test).Table 1Axial biometry of pseudophakic eyes as measured by ultra-long scan depth OCT.CCT^\*^ACD^\*^IOLT^\*^VL^\*^AL^\*^RestingV10.55 ± 0.034.25 ± 0.310.82 ± 0.0317.85 ± 0.9323.47 ± 0.93V20.55 ± 0.034.24 ± 0.310.83 ± 0.0317.85 ± 0.9423.47 ± 0.93mean0.55 ± 0.034.24 ± 0.310.82 ± 0.0317.85 ± 0.9423.47 ± 0.93Dif0.00 ± 0.010.01 ± 0.04−0.01 ± 0.01−0.00 ± 0.040.00 ± 0.02CoR2.11%1.71%3.54%0.46%0.16%ICC0.9910.9970.9281.0001.000AccommodationV10.55 ± 0.034.33 ± 0.310.82 ± 0.0217.81 ± 0.9523.51 ± 0.93V20.55 ± 0.034.33 ± 0.320.82 ± 0.0217.81 ± 0.9523.51 ± 0.94mean0.55 ± 0.034.33 ± 0.310.82 ± 0.0217.81 ± 0.9523.51 ± 0.94Dif−0.00 ± 0.01−0.00 ± 0.04−0.00 ± 0.020.00 ± 0.03−0.00 ± 0.03CoR2.81%1.88%4.22%0.36%0.23%ICC0.9850.9960.8411.0001.000DIF+0.00 ± 0.01 +0.08 ± 0.09−0.00 ± 0.02−0.04 ± 0.09 +0.04 ± 0.04P value\*\*0.39**\<0.01**0.44**\<0.05\<0.01**^\*^All measurements were in millimeters; CCT, central corneal thickness; ACD, anterior chamber depth; IOLT, thickness of the intraocular lens; VL, vitreous length; AL, axial length.\*\*Comparisons between resting and accommodative states by two-tailed paired t-tests.V1 and V2: 1st and 2nd measurements; Dif: difference between the V1 and V2; CoR: coefficient of repeatability; ICC: interclass correlation coefficients; DIF: Difference in values between the resting and accommodative states.Figure 3Changes in axial biometry of pseudophakic eyes during pilocarpine-induced accommodation. CCT, central corneal thickness; ACD, anterior chamber depth; IOLT, intraocular lens thickness; VL, vitreous length; AL, axial length; \*P \< 0.05, comparisons between resting and accommodative states by two-tailed paired t-tests. Bar = 1 standard error. Repeatability of ocular axial biometry measurements by UL-OCT {#Sec11} ------------------------------------------------------------- Images of the whole eye from the cornea, through the pupil, to the retina were successfully obtained by UL-OCT at both baseline and accommodative states (Fig. [4](#Fig4){ref-type="fig"}). Bland-Altman plots of the differences between the repeated measurements for each parameter were constructed (Fig. [5](#Fig5){ref-type="fig"}). At baseline in the resting state, the CoRs of the parameters ranged from 0.16% to 3.54%, and the ICCs ranged from 0.928 to 1.000 (Table [1](#Tab1){ref-type="table"}). After accommodation was induced by pilocarpine, the CoRs ranged from 0.23% to 4.22%, and the ICCs ranged from 0.841 to 1.000 (Table [1](#Tab1){ref-type="table"}).Figure 4Whole eye from the cornea to the retina through pupil of a 49-year-old subject before and after pilocarpine-induced accommodation. The images were acquired by ultra-long scan depth OCT. The left side was obtained at the resting state, while the right side was taken during accommodation. In this subject, the IOL exhibited a backward movement of 0.086 mm, while the axial length increased from 24.034 to 24.069 mm. CCT, central corneal thickness; ACD, anterior chamber depth; IOLT, intraocular lens thickness; VL, vitreous length; AL, axial length. Bar = 1 mm.Figure 5Bland-Altman plots of the difference between the two repeated measurements of whole eye axial biometry in 25 eyes before (blue, Baseline) and after (red, ACC) pilocarpine-induced accommodation. Values on the vertical axis correspond to the difference between the two measurements, and values on the horizontal axis correspond to the average of the two measurements. Solid and dashed lines indicate the mean differences and 95% confidence intervals, respectively. Blue and red values were the mean difference ± standard deviation of the repeated measurements at the baseline and the accommodative states, respectively. CCT, central corneal thickness; ACD, anterior chamber depth; IOLT, intraocular lens thickness; VL, vitreous length; AL, axial length. Discussion {#Sec12} ========== In this study, we used UL-OCT to demonstrate that the ocular axial biometry undergoes significant change during pilocarpine-induced accommodation in pseudophakic eyes. Sheppard & Davies^[@CR12],[@CR26]^ reported the morphologic changes in the aging ciliary muscle (aged from 19--70 years), but those changes appeared not to affect the contractile ability of the muscle during accommodation. Strenk *et al*.^[@CR32]^ also found that the accommodation-related change in ciliary muscle ring was undiminished by age (22--91 years) or IOL implantation. Therefore, the ciliary muscle seems to retain the ability for contraction throughout life, even though the accommodative amplitude decreases with age. The work of Park *et al*.^[@CR33]^ showed that the centripetal movement of human ciliary body increases significantly after cataract extraction. Thus, even though a recent study by Croft *et al*.^[@CR34]^ suggested the accommodative ciliary muscle contraction declined with age in phakic eyes, the results might be different in pseudophakic eyes. Pilocarpine causes contraction of the ciliary muscle, and Koeppl *et al*.^[@CR35]^ demonstrated that it acts as a "superstimulus" in presbyopic phakic eyes, resulting in an "unphysiological" response. The authors hypothesized that pilocarpine stimulation may produce a "physiological" accommodation in both young and presbyopic subjects. However the stimulus-driven accommodation is sub-maximal in presbyopes and may be insufficient to cause detectable axial elongation. Therefore, we used pilocarpine, even though it could act "unphysiologically" but elicit a superstimulus effect, to ensure adequate contraction of the ciliary muscle. The high accuracy of the UL-OCT instrument was validated by measuring the axial elongation of a physical eye model that was stretched by mechanical tension. The mean absolute discrepancy between the UL-OCT length measurements and the micrometer setting values was 4 µm. With the UL-OCT instrument used in our study, the CoRs and ICCs for axial biometry of the pseudophakic eyes are consistent with the excellent repeatability of this instrument that we previously reported^[@CR11],[@CR16]^. This high level of accuracy and repeatability assures that the eye elongation was associated with the pilocarpine-induced accommodation rather than any measurement error. The ICCs and CORs of the IOLT were relatively worse than for other parameters. The explanation may be associated with the probable occurrence of IOL tilt that could cause the measurements to diverge slightly from the central axis of the IOL. Previous studies have investigated the change in AL with accommodation (Table [2](#Tab2){ref-type="table"})^[@CR3]--[@CR6],[@CR9]--[@CR12]^. For instance, Woodman *et al*.^[@CR9]^ found a mean increase in the AL of +0.020 mm in phakic eyes for a 4.0 D accommodation. Mallen *et al*.^[@CR5]^ reported a +0.014 mm increase with 2.0 D of accommodation, and +0.037 mm increase with 6.0 D of accomodation. Using a similar UL-OCT system in our previous study^[@CR11]^, we demonstrated axial elongation of +0.026 mm during 6.0 D of accommodation. Most of the previous studies used the near target as the accommodative stimulus^[@CR3],[@CR5],[@CR6],[@CR9]--[@CR12]^. Here, we induced accommodation by pharmacological stimulation of the ciliary muscle in pseudophakic eyes of older subjects (49--84 years of age) and measured an increase in AL of +0.04 mm. Lister *et al*.^[@CR36]^ found that the pilocarpine instillation significant increased lens thickness for both young and older groups, indicating that pharmacological stimulation influenced both young and older subjects. The exact level of accommodation due to the pharmacologically induced ciliary muscle contraction stimulated was unknown, and the biometrics of the older and pseudophakic eyes might not be similar to the younger phakic eyes, as reported in Lister's study^[@CR36]^. Though not directly comparable, this value was slightly larger than that reported by the other studies.Table 2Summary of previous studies investigating the changes in axial length during accommodation.StudyTechniqueSample sizeAgeAccommodationIncrease in axial length^[@CR48]^Current studyUL-OCT2549--84Pilocarpine induced0.04 ± 0.04Shum *et al*.^[@CR3]^Ultrasonic biometer10618--223.0 D0.06 ± 0.01Drexler *et al*.^[@CR4]^Custom PCIEM: 11 M: 1221--30Near pointEM: 0.013 ± 0.003M: 0.005 ± 0.002Mallen *et al*.^[@CR5]^IOLMasterEM: 30 M: 30EM: 21.4 ± 2.0 M: 21.5 ± 2.12.0 D\ 4.0 D\ 6.0 DEM: 0.014 ± 0.019M: 0.019 ± 0.020EM: 0.026 ± 0.021M: 0.037 ± 0.026EM: 0.037 ± 0.027M: 0.058 ± 0.037Read *et al*.^[@CR6]^LenstarEM: 19 M: 2118--333.0 D\ 6.0 DEM: 0.013 ± 0.013M: 0.011 ± 0.012EM: 0.025 ± 0.015M: 0.023 ± 0.023Woodman *et al*.^[@CR9]^LenstarEM: 22 M: 3718--304.0 DEM: 0.006 ± 0.022M: 0.022 ± 0.034Ghosh *et al*.^[@CR10]^LenstarEM: 10 M: 1018--302.5 D0.007 ± 0.008Laughten *et al*.^[@CR12]^LenstarEM: 10 M: 1034--414.5 DEM: 0.002 ± 0.006\*M: 0.005 ± 0.004\*Zhong *et al*.^[@CR11]^UL-OCT2119--356.0 D0.026 ± 0.013PCI: Partial coherence interferometry; UL-OCT: Ultra-long scan depth optical coherence tomography; EM: Emmetropes; M: Myopes; D: diopters.\*Change per diopter of accommodation. We enrolled IOL-implanted patients because the refractive index of each component of their eyes was constant. Thus the absence of refractive index changes in each component of the eye during accommodation ensured that it could not have influenced the measurement of axial elongation. In contrast, Mallen *et al*.^[@CR5]^ used a commercial IOLMaster device for AL measurements that did not include individual optical correction for each compartment. Thus, the measured changes could have been overestimated due to the potential error associated with lens thickening during accommodation^[@CR21]^. The possible alteration of the crystalline lens refractive index may also cause another measurement error^[@CR20],[@CR22],[@CR24]^. The effects of accommodation on the refractive index of the lens are contentious. Using a ray trace method, Gullstrand deduced that the reshaping of the crystalline lens during accommodation causes the fibers inside the lens to slide on each other and change the equivalent refractive index of the lens from 1.413 to 1.424^[@CR24]^. Based on *in vivo* examination, Kasthurirangan *et al*.^[@CR20]^ found a significant increase in the size of lens central plateau region during accommodation with constant refractive index. This resulted in an increase of the equvilant refractive index of the lens. However, Hermans *et al*.^[@CR18]^ suggested the equivalent refractive index of the human lens might remain unchanged with accommodation, and Jones *et al*.^[@CR19]^ found a slight, but not significant, decrease in central refractive index with accommodation. Those potential changes in the refractive index of the crystalline lens were not taken into account in previous studies^[@CR4]--[@CR12]^. Misestimating the refractive index during the accommodative state will result in a deviation of the measurement from the lens true thickness. This will then result in misestimatation of the AL and the axial elongation during accommodation. This measurement artifact was avoided in our subjects because the IOLs had a constant refractive index. To explain the elongation of the eye during accommodation, some researchers proposed that the force of ciliary muscle contraction produces an inward pull on the choroid and sclera adjacent to the ciliary muscle^[@CR4],[@CR5]^. The force of the contraction would decrease the circumference of the globe at the equator, forcing the rearward movement of the posterior portion of the globe to maintain a constant ocular volume. Coudrillier *et al*.^[@CR37]^ found that the human posterior sclera stiffens with age, which could limit the scleral stretching under the stress. This suggests that the aging human eyes enrolled in the present study may respond differently to elongation forces than eyes in young people. The elongation of AL found in the current study was greater than previous studies that enrolled younger subjects. The pilocarpine we used could have acted as a superstimulus for ciliary muscle contraction, which may have caused a larger increase of AL than that during physiological accommodation. Another explanation for the anatomical change may be a thinning of the choroid during accommodation. Using the Lenstar, Woodman *et al*.^[@CR9]^ observed a decrease in choroidal thickness that had a weak but significant negative correlation with the changes in AL, although the magnitude of thinning in choroid was only 38% of the increase in AL. Using OCT, Woodman *et al*.^[@CR38]^ confirmed the decrease of choroidal thickness with accommodation. As with the sclera, the elasticity of the choroid also decreases with age^[@CR39],[@CR40]^. However, compared to sclera, the choroid is a much more elastic tissue that is comprised of blood vessels, melanocytes, fibroblasts, resident immunocompetent cells, and supporting collagenous and elastic connective tissue. Thus the thinning of the choroid is unlikely to be solely responsible for eye elongation during accommodation; however, it may contribute to the stretching of the globe, especially in elderly subjects with stiffer scleras. In addition to eye elongation, we found a significant increase in the ACD during the pilocarpine-induced accommodation. The increase in ACD results from movement of the IOL in the posterior direction, as previously reported by others^[@CR41],[@CR42]^. However in phakic eyes, the crystalline lens moves forward during accommodation and therefore causes a decrease in the ACD^[@CR11],[@CR43]--[@CR47]^. There were limitations in the current study. First, we used pilocarpine to induce contraction of the ciliary muscle in older subjects with pseudophakic eyes. Typically, the accommodative ability in phakic individuals of this age group is lower than that of young subjects. For this reason, we chose to induce accommodation pharmacologically to ensure a strong contraction of the ciliary muscle. While this mode of accommodation might not be precisely like natural accommodation during near work, the utilization of pilocarpine was justifid to achieve a strong ciliary muscle contraction in the older subjects. Second, the results may have been affected by the different times of measurements after date of surgery. A previous study^[@CR31]^ showed that the reduction of ciliary ring diameter during accommodation due to the contraction of ciliary muscle was 0.11 mm at one month after phacoemulsification surgery. The changes in diameter slightly increased to 0.14 mm at two months and to 0.18 mm at 12 months after surgery. These changes were stable at all of the succeeding follow-up examinations, indicating that the capability of ciliary muscle was restored after surgery and maintained for at least for one year. For our study, period between phacoemulsification surgery and measurement of the ocular axial parameters by UL-OCT was 1--3 months. Thus, the influence of the time after surgery for our subjects was probably small. The third limitation arose because the telecentric optical design of the scanning probe could not obtain detailed images of the retina. Thus, the retina appeared as a straight line rather than having the normal curvature. For that reason, we asked the subjects to fixate on the target that was adjacent to the UL-OCT beam to ensure that the measuring beam was aimed at the visual axis. In conclusion, we used UL-OCT to demonstrate that the ocular AL increased with pilocarpine-induced accommodation in pseudophakic eyes of older subjects. Because the IOL has a constant refractive index, it is clear that the axial elongation was due to accommodation rather than a measurement artifact associated with a change in the refractive index of the natural crystalline lens. In addition to the elongation of the AL during accommodation, there was a backward movement of the IOL. These findings provide a more precise understanding of changes in ocular biometric properties with accommodation. This new information may help to establish a link between near work and longer-term axial elongation of the eye. Yilei Shao and Qiuruo Jiang contributed equally to this work. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank the patients, their families and all the members of the Affiliated Eye Hospital of Wenzhou Medical University, Wenzhou, Zhejiang, China, for their helpful support. Grant/financial support: This study was supported by research grants from the National Natural Science Foundation of China (Grant No. 81170869 and 81570880 to Lu; Grant No. 81400441 to Shen; Grant No. 81200672 to Chen), National Major Equipment Program of China (2012YQ12008004 to Lu), Natural Science Foundation of Zhejiang Province (Grant No. LQ16H120007 to Shao; Grant No. LY16H120007 to Yuan; Grant No. LY14H180007 to Zhu). The sponsors and funding organization had no role in the design or conduct of this research. We thank Dr. Britt Bromberg of Xenofile Editing ([www.XenofileEditing.com](http://www.XenofileEditing.com)) for providing editing services for this article. Design of the study (Y.S., M.S. and F.L.); Conduct of the study, data collection, analysis and interpretation (Y.S., Q.J., D.H., L.Z., M.S., S.H., L.L., Y.Y., Q.C., D.Z., J.W., and F.L.). Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

1. Introduction =============== Data from genome-wide transcriptional analysis in humans have shown that the amount of protein coding transcripts accounts for approximately 2% of the entire genome, while the non-coding RNAs (ncRNAs) correspond to around 98% of all the genomic output ([@b1-ijo-46-01-0017],[@b2-ijo-46-01-0017]). Interestingly, it has been reported that the proportion of non-coding regions in the genome increases according to the complexity of the organism, suggesting a important role for these sequences in physiology and development of the organisms ([@b3-ijo-46-01-0017],[@b4-ijo-46-01-0017]). For this reason, much attention has been given to the studies on these non-protein-coding RNAs in many fields, especially in cancer, leading to new hypothesis about cancer biology ([@b5-ijo-46-01-0017]). Additionally, the identification of the circulating microRNAs (miRNAs) in bodily fluids makes them potential non-invasive biomarkers for cancer diagnosis and prognosis. The comprehension of the mechanisms involved in the interactions between tumor cells and the surrounding environment is relevant for tumor biology elucidation and for the improvement of innovative and more efficient therapy approaches ([@b6-ijo-46-01-0017]). The role of extracellular vesicles in cell-to-cell communication in cancer has been the focus of several studies. MiRNAs are one of the most studied exosomal cargos due to their potential role in tumor diagnosis, prognosis and therapy. In this review, we summarize the role of ncRNAs in cancer, focusing on miRNAs. Additionally, we focus on the role of exosomes in intercellular communication and their potential use in providing diagnostic opportunities, unraveling new therapeutic targets and predicting therapeutic responses. 2. World of non-coding RNAs =========================== The ncRNAs can be divided in two main groups, according to their sizes: long non-coding RNAs (lncRNAs), which are greater than 200 nucleotides and small non-coding RNAs with no more than 200 nucleotides ([@b7-ijo-46-01-0017],[@b8-ijo-46-01-0017]). These two main categories also show some subgroups, based on the structure and biological function of the transcripts, as long intergenic ncRNAs, pseudogenes, enhancer RNAs, transcribed ultra-conserved region, repeated-associated ncRNAs and antisense RNA in the lncRNAs group. In the small ncRNAs, miRNAs, tiny transcription initiation RNAs, small interfering RNAs, promoter-associated short RNAs, antisense termini associates short RNAs and retrotransposon-derived RNAs have been reported in the literature ([@b5-ijo-46-01-0017],[@b8-ijo-46-01-0017]). The miRNAs are the most widely described ncRNA in the literature, since the first small ncRNA lin-4 was described in *C. elegans* more than 20 years ago ([@b9-ijo-46-01-0017],[@b10-ijo-46-01-0017]). The synthesis of these evolutionarily conserved endogenous short single-stranded RNAs (18--20 nucleotides in length) begins in the nucleus, when the transcription of miRNA-coding genes generate long primary transcript (pri-miRNA) with stem-loop, which will be detached by the RNase III Drosha/Pasha/DGCR8 complex, and then producing a 70-nucleotide precursor (pre-miRNA). After being transported to the cytoplasm by the protein Exportin-5 (XPO5), the pre-miRNA will be converted in mature miRNA by the action of the Dicer and binding to Argonaute 2 (Ago2) to form the RNA-induced silencing complex (RISC) ([@b11-ijo-46-01-0017],[@b12-ijo-46-01-0017]). Overall, miRNAs regulate gene expression post-transcriptionally, most commonly through the binding to a specific sequence at the 3′-untranslated region (3′-UTR) of a target protein-coding mRNA, causing a translational repression or cleavage of the target transcripts ([@b13-ijo-46-01-0017]). Thus, miRNAs have a relevant role in many pathological and physiological processes, such as cell proliferation, differentiation, development and apoptosis, acting as oncogenes or tumor suppressors, depending on which genes they regulate ([@b14-ijo-46-01-0017]). The involvement of miRNA genes in cancer was first described in 2002, when the authors reported that two miRNAs (miR-15a and miR-16-1) are mapped at 13q14, a chromosomal region frequently deleted in B-cell chronic lymphocytic leukemia (B-CLL) and that both genes are down-regulated in a high proportion of the cases ([@b15-ijo-46-01-0017]). Since then, the number of studies on miRNAs and cancer has been increasing considerably, adding novel insights into the role of the miRNAs in human tumor such as in hematological malignancies ([@b16-ijo-46-01-0017]--[@b19-ijo-46-01-0017]), colorectal ([@b20-ijo-46-01-0017]--[@b23-ijo-46-01-0017]), breast ([@b24-ijo-46-01-0017]--[@b28-ijo-46-01-0017]), head and neck ([@b29-ijo-46-01-0017]--[@b32-ijo-46-01-0017]) and gastric cancer ([@b33-ijo-46-01-0017]--[@b36-ijo-46-01-0017]). The lncRNAs are transcripts longer than 200 nucleotides, a cutoff based on RNA purification protocols that exclude small RNAs rather than for its functional role ([@b37-ijo-46-01-0017]). The lncRNAs were first described in a study involving large-scale sequencing and annotation of full-length cDNA libraries in mouse ([@b38-ijo-46-01-0017]), and the number of reports about characterization and functions of the lncRNAs has been constantly increasing in the literature ([@b39-ijo-46-01-0017]). The lncRNAs have many features in common with mRNAs, as transcription by RNA polymerase II, polyadenilation and splicing mechanism. This category of ncRNAs composes a heterogeneous group, which makes the lncRNAs classification difficult ([@b40-ijo-46-01-0017]). Most commonly, the lncRNAs can be classified as sense or antisense, divergent or convergent and intronic or intergenic, depending on their position relative to the neighboring protein-coding genes ([@b7-ijo-46-01-0017],[@b41-ijo-46-01-0017]). Due to lncRNA structure heterogeneity, it is also difficult to assign a specific function to this group and still requires further studies. Evidences suggest that lncRNAs act mainly in regulation of protein-coding genes transcription, but in more complex ways than the miRNAs ([@b42-ijo-46-01-0017]). lncRNAs can repress the transcription of target genes involving epigenetic modifications like chromatin remodeling, since some lncRNAs have been reported to interact with many chromatin modifiers ([@b43-ijo-46-01-0017]). Additionally, lncRNAs can either play a role as putative gene enhancers or decoy RNAs in transcriptional control ([@b41-ijo-46-01-0017]). Some lncRNAs (antisense ncRNA) also play a role in post-transcriptional regulation by interfering with the RNA splicing mechanism ([@b44-ijo-46-01-0017]). Due to their roles in the functions mentioned above, the lncRNAs have been related to many human cancers, contributing to tumor development and progress ([@b40-ijo-46-01-0017]). Many lncRNAs have been mapped at cancer risk loci in the human genome, such as PTCSC3 (14q13.3) in thyroid cancer ([@b45-ijo-46-01-0017],[@b46-ijo-46-01-0017]), PCA3 (9q21--22) in prostate cancer ([@b47-ijo-46-01-0017],[@b48-ijo-46-01-0017]), ANRIL (9p21) in prostate and breast cancers, leukemia and melanoma ([@b49-ijo-46-01-0017]--[@b52-ijo-46-01-0017]), MALAT1 (11q13) in liver, colorectal, prostate, bladder and lung cancers ([@b53-ijo-46-01-0017]--[@b56-ijo-46-01-0017]). The role of ncRNAs in many human tumor types has been exhaustingly studied in the past few years and its relevance in mechanisms involved in cancer development and progress is unquestionable. Additionally, the discovery of stable miRNAs in bodily fluids introduced new insights in the ncRNAs comprehension and can represent a new diagnostic approach using less invasive methods ([@b57-ijo-46-01-0017]). The use of circulating miRNAs as tumor biomarkers has many advantages since these transcripts are conserved across species, shows tissue or disease-specific expression and their levels can be quantified by various methods, as microarray profiling, northern blot analysis, *in situ* hybridization, high-throughput sequencing and qRT-PCR, which is the most used method due to its sensitivity, specificity and reliability ([@b58-ijo-46-01-0017]--[@b61-ijo-46-01-0017]). The first evidence of the presence of miRNAs in serum was reported by Lawrie *et al* (2008), who showed the higher serum levels of miR-21 in large B-cell lymphoma ([@b62-ijo-46-01-0017]). Since then, many studies have reported the presence of different circulating miRNAs in various tumor types, such as in colorectum ([@b63-ijo-46-01-0017],[@b64-ijo-46-01-0017]), esophagus ([@b65-ijo-46-01-0017],[@b66-ijo-46-01-0017]), breast ([@b67-ijo-46-01-0017],[@b68-ijo-46-01-0017]), stomach ([@b69-ijo-46-01-0017],[@b70-ijo-46-01-0017]) and ovary ([@b71-ijo-46-01-0017],[@b72-ijo-46-01-0017]). Also, the circulating miRNAs have been reported in many other fluids such as plasma, urine, saliva and cerebrospinal fluid ([@b58-ijo-46-01-0017],[@b73-ijo-46-01-0017]). Considering this discovery, it is possible that other types of ncRNAs similar to lncRNAs can also be identified in bodily fluids ([@b40-ijo-46-01-0017]). 3. World of extracellular vesicles ================================== The intercellular communication can occur by a direct cell-to-cell contact including the adhesion junctions, or by the releasing of soluble signaling molecules or by the exchange of cellular fragments as extracellular vesicle (EV) ([@b74-ijo-46-01-0017],[@b75-ijo-46-01-0017]). The EVs are bilayered membrane vesicles secreted by all cell types, and released in the interstitial space or into circulating bodily fluids, where they can travel long distances until they are up taken by receptor cells ([@b76-ijo-46-01-0017]). Different terminology is used to describe EVs based on their morphology and diameter. Exosomes, microvesicles, ectosomes, microparticles and others, are classified based on their size, shape and membrane surface composition ([@b77-ijo-46-01-0017]). The most accepted classification in the literature shows two major groups of EVs, based on their mechanism of biogenesis and sizes: exosomes and microvesicles (or ectosomes). Additionally, apoptotic bodies have been considered by some as a third category of EVs ([@b78-ijo-46-01-0017]--[@b80-ijo-46-01-0017]). In this review, we will focus on the exosomes and microvesicles. Exosomes are 40--140 nm diameter bilayered-membrane vesicles of endocytic origin, with a cup-shaped morphology, showing densities ranging between 1.13--1.19 g/ml ([@b81-ijo-46-01-0017]). The exosomes are originated by the inward budding of clathrin-coated domains in the plasma membrane, generating the multivesicular bodies (MVBs) containing intraluminal vesicles (ILVs) in the late endosome. The formation of ILVs occurs during the endosome maturation, when specific cytosolic proteins are incorporated into these vesicles inside de MVBs. These initial steps occur under control of the ESCRT (endosomal sorting complex required for transport) machinery. Later, the MVBs fuse with lysosomes for degradation or with the cell membrane releasing the exosomes to the extracellular space, process regulated by the RAB family ([@b76-ijo-46-01-0017],[@b82-ijo-46-01-0017]--[@b84-ijo-46-01-0017]). Microvesicles (or ectosomes) are larger than exosomes, with size ranging between 100 and 1,000 nm in diameter and heterogeneous in morphology. Differently from the exosomes, microvesicles (MVs) are originated from the plasma membrane through direct outward budding into the extracellular space. During this process, the newly originated vesicle captures the donor cellular cytosolic content and the receptors on the plasma membrane ([Fig. 1](#f1-ijo-46-01-0017){ref-type="fig"}). The regulation of MVs biogenesis is intracellular calcium-dependent and it is the result of the activation of cell surface receptors, phospholipid redistribution and cytoskeletal protein contraction ([@b84-ijo-46-01-0017],[@b85-ijo-46-01-0017]). The apoptotic bodies (ABs) are membrane vesicles, heterogeneous in shape, showing sizes ranging between 50--500 nm in diameter. The ABs are released from the outward protrusion of the plasma membrane during the late phase of cell death by apoptosis and are featured by the presence of organelles inside the vesicles ([@b85-ijo-46-01-0017],[@b86-ijo-46-01-0017]). The EV cargo specificity ------------------------ The interaction between the EVs and the target cells can occur by different mechanisms, as direct interaction of the surface proteins of the EVs with the receptors on the target cells, triggering the activation of the intracellular pathways. EVs can also be engulfed by the target cells through membrane fusion or by endocytosis/phagocytosis, with transfer and release of their cargo. Transcripts as mRNAs and miRNAs contained inside de EVs can be transferred to the target cells and be functional ([@b6-ijo-46-01-0017]). The EVs carry specific contents (cargo) as membrane receptors, ligands, proteins, nucleic acids and infectious agents, depending on the cell of origin and how they were originated from the donor cell ([@b75-ijo-46-01-0017]). There is no consensus regarding the specific content of different EVs, but it seems that MVs are characterized by the presence of cell-surface proteins from the donor cells such as receptors and adhesion proteins. In turn, exosomes have been found to be characterized by proteins associated to their endosomal origin and MVBs formation ([@b84-ijo-46-01-0017]). Some specific markers have been associated to exosomes as tetraspanins (CD9, CD63, CD81 and CD82), major histocompatibility complex class I and II, LAMP1 and LAMP2, flotilins, annexins, TSG101 and heat shock proteins ([@b83-ijo-46-01-0017],[@b87-ijo-46-01-0017],[@b88-ijo-46-01-0017]). The protein content of MVs seems to be more heterogeneous, containing cell membrane markers, phosphatidylserine (PS) residues, integrins, selectins and CD40 ligands ([@b84-ijo-46-01-0017]), and high levels of cholesterol and signaling complexes known as lipid rafts ([@b76-ijo-46-01-0017]). Most importantly, it has been shown that the population of exosomes secreted by cancer cells contains a representation of the entire genome of the cell of origin, providing exciting opportunities of using exosomes as a liquid biopsy ([@b89-ijo-46-01-0017]). Tumor-derived exosomes ---------------------- The exosomes are by far the most extensively studied due to their characteristics (as presence in bodily fluids and expression of specific markers) that can contribute not only to intercellular communication but also to their potential role in diagnosis ([@b82-ijo-46-01-0017]). The release of exosomes can be a response to different cellular stress conditions common in cancer, such as hypoxia, acidic pH, heat shock and oxidative stress, resulting in alterations of the tumor microenvironment and distal organs activating angiogenesis and promoting migration and leading to metastasis ([@b90-ijo-46-01-0017]--[@b92-ijo-46-01-0017]). An important step before considering using the exosomal content for study or diagnosis purposes is a reliable exosome isolation method, to insure the quality of the results. Exosomes can be collected from fluids or cell supernatant by a series of sequential centrifugations to remove larger cellular debris and filtration through 100--220 nm filters to exclude larger EVs, including MVs and apoptotic bodies. Then, the exosomes are pelleted by ultracentrifugation and/or suspension in a sucrose gradient for the completely remove of protein contamination ([@b77-ijo-46-01-0017],[@b93-ijo-46-01-0017],[@b94-ijo-46-01-0017]). The exosomes isolation can also be performed using specific filters, immune isolation by magnetic beads or microfluidic separation ([@b95-ijo-46-01-0017]). Recently, some commercial isolation kits are available based on polymer-based precipitation and on the magnetic bead isolation. Then, some additional procedures are necessary to confirm the purity of the isolated exosomes. One of them is to verify the size and shape of the exosomes by electron microscopy analysis. Vesicles diameter and morphology can also be assessed by specific instruments that can visualize, characterize and measure small vesicles. Another important factor that must be evaluated is the protein content, that can be assessed by flow cytometry and western blot analysis for markers as CD63, CD81, Tsg101 and flotilin ([@b77-ijo-46-01-0017],[@b96-ijo-46-01-0017]). 4. The fusion of the two worlds =============================== Exosomal miRNAs --------------- In the bodily fluids, the miRNAs have been reported to play a role at intercellular communication, and can act at short and long distant sites in a hormone-like behavior ([@b14-ijo-46-01-0017],[@b87-ijo-46-01-0017]). The transport of circulating miRNAs can be carried by protein transporters or by exosomes. It is known that serum contains ribonucleases, suggesting that the circulating miRNAs are protected from the RNase action within extravesicles. The miRNA recruitment to the exosomes depends on the attachment of RNA-induced silencing complexes (RISCs) to the ESCRT components. However, the release of exosomal miRNAs is under control of a ceramide-dependent machinery, as reported by Kosaka *et al* ([@b73-ijo-46-01-0017]). These authors showed that the inhibition of neutral sphingomyelinase 2 (a regulator of the ceramide biosynthesis) resulted in lower levels of miRNA secretion ([@b73-ijo-46-01-0017]). The first evidence of the existence of miRNAs in exosomes was reported by Valadi *et al*, showing that these vesicles contain both mRNA and miRNAs, which can be transferable to another cell, where the transcripts can be functional ([@b97-ijo-46-01-0017]). Since then, the number of studies regarding the identification of exosomal miRNAs in cancer has been increasing in the literature. A summary of some studies in the literature in this field is in [Table I](#tI-ijo-46-01-0017){ref-type="table"}. Most of these reports are based on *in vitro* studies involving a variety of cancer cell lines, identifying the exosomal miRNA content as in breast cancer ([@b98-ijo-46-01-0017],[@b99-ijo-46-01-0017]), leukemia ([@b100-ijo-46-01-0017]), melanoma ([@b101-ijo-46-01-0017],[@b102-ijo-46-01-0017]), prostate cancer ([@b103-ijo-46-01-0017]), ovarian ([@b104-ijo-46-01-0017]) and gastric cancer ([@b105-ijo-46-01-0017]). The transcripts content of the exosomes usually can differ from that in the donor cells, and the exosomal miRNA profile of tumor cells can also differ from that released by normal controls ([@b106-ijo-46-01-0017]). Studies have reported evidence of the intercellular transfer of the exosomal content between different cells. Chiba *et al* showed that exosomes derived from colorectal cancer cells can be transferred to hepatoma and lung cancer cells ([@b107-ijo-46-01-0017]). In addition, some of these reports have demonstrated that the transferred exosomal content can be functional in the receptor cells. Yang *et al* reported the presence of a specific miRNA for IL-4-activated macrophage - miR-233, in the co-cultured breast cancer cells and it can enhance the invasiveness potential of the receptor cells ([@b108-ijo-46-01-0017]). The transfer of the leukemia cell line-derived exosomal miR-92a to the endothelial cells affected the endothelial cell migration and tube formation in the receptor cells ([@b109-ijo-46-01-0017]). Kosaka *et al* showed that exosomal miR-143 derived from a normal prostate cell line act as a tumor suppressor by inhibiting the growth in the target prostate cancer cells ([@b110-ijo-46-01-0017]). Similar results were reported by Roccaro *et al* in a study demonstrating that the exosomal miR-15 from the normal bone marrow mesenchymal stromal cells causes a tumor suppressor effect when transferred to tumor cells, where this miRNA is downregulated ([@b111-ijo-46-01-0017]). In a recent study, Valencia *et al* demonstrated that the exosomal miR-192 derived from lung adenocarcinoma cell lines repressed the angiogenic activity in the co-cultured endothelial cells by the inhibition of the proangiogenic factors ([@b112-ijo-46-01-0017]). More recently, Zhou *et al* showed that the transfer of exosomal miR-105 to non-metastatic breast cancer cells induces metastasis and vascular permeability by targeting the cellular tight junction protein ZO-1 ([@b113-ijo-46-01-0017]). The intercellular transfer of the exosome cargo can also affect the resistance or sensitivity of cancer cells to therapy. The transfer of the exosomes derived from chemoresistant breast cancer cell lines can also spread the resistance potential to receptor chemosensitive cell lines, possibly due to the action of the exosomal content as miR-100, miR-222 and miR-30a ([@b114-ijo-46-01-0017]). Similarly, in lung cancer, Xiao *et al* reported that after miR-21- and miR-133-enriched exosome transfer from the chemoresistant tumor cells, the chemosensitive target cells acquire resistance to the drug exposure ([@b115-ijo-46-01-0017]). It is well known that hypoxia is an important factor that triggers angiogenesis and metastasis formation and evidence has been presented for the involvement of the exosome in this mechanism. King *et al* found an increased concentration of released exosomes and higher expression of exosomal miR-210 secreted by breast cancer cells under hypoxic conditions when compared to normoxic cells ([@b116-ijo-46-01-0017]). In leukemia, the miR-210 can also be found in a subset of miRNAs upregulated in exosomes released by tumor cells under hypoxic conditions ([@b117-ijo-46-01-0017]). The exosomal miRNA expression profiling from serum and plasma samples has also been assessed in glioblastoma, where the expression of 11 miRNAs known to be upregulated was slightly lower in exosomes than in the donor cells but still reflecting the tumor profile ([@b118-ijo-46-01-0017]). When the serum samples from ovarian cancer patients were evaluated, a distinct exosomal miRNA profile was identified from that of benign disease ([@b119-ijo-46-01-0017]). Another study reporting a potential use of exosomal miRNAs as diagnostic markers showed a higher expression of miR-21 released from esophageal cancer serum samples when compared to non-tumoral samples, and it correlated with advanced tumor stages, lymph node involvement and metastasis ([@b120-ijo-46-01-0017]). Exosomal let-7a, miR-1229, miR-1246, miR-150, miR-21, miR-223 and miR-23a from colorectal tumor samples and cancer cell lines are more highly expressed than those from healthy controls samples and normal colon cell lines, and the expression levels of these miRNAs are significantly decreased in exosomes samples collected after tumor resection, indicating the cancer status ([@b121-ijo-46-01-0017]). The exosomal miRNA profiling from plasma samples was assessed to develop a diagnostic screening method for lung adenocarcinoma. The expression pattern of 12 specific upregulated miRNAs (miR-17-3p, miR-21, miR-106a, miR-146, miR-155, miR-191, miR-192, miR-203, miR-205, miR-210, miR-212, miR-214) in tumor samples was similar in the tumor plasma-derived exosomes and distinct from the control samples, indicating exosomal miRNAs could be relevant as a screening method for this tumor ([@b122-ijo-46-01-0017]). In another report, the exosomal miRNAs miR-378a, miR-379, miR-139-5p and miR-200-5p were identified as possible markers to distinguish tumor from normal samples in lung adenocarcinoma ([@b123-ijo-46-01-0017]). In addition, the miRNAs miR-151-5p, miR-30a-3p, miR-200b-5p, miR-629, miR-100 and miR-154-3p are possible markers to discriminate lung adenocarcinoma from granulomas ([@b123-ijo-46-01-0017]). More recently, Rodríguez *et al* evaluated the exosomes derived from bronchoalveolar lavage (BAL) and plasma samples from lung cancer, in which the exosomal miRNA content derived from tumor plasma samples is more elevated than in the BAL, suggesting that the a higher concentration of exosomal miRNAs are released in the plasma than in the bronchoalveolar fluid ([@b124-ijo-46-01-0017]). The use of exosome as a diagnostic tool has also been evaluated in other fluids than plasma and serum samples. Recently, a report identified the exosomal miRNAs in bile from cholangiocarcinoma patients and a potential diagnostic panel that includes miR-486-3p, miR-16, miR-1274b, miR-484 and miR-191 as predictive markers ([@b125-ijo-46-01-0017]). In a study involving cervicovaginal lavage fluids, the miR-21 and miR-146a were highly expressed in fluids from cervical cancer samples when compared to those from HPV(+) and HPV(−) normal samples ([@b126-ijo-46-01-0017]). Exosomal lncRNAs ---------------- The study of lncRNAs is a relatively new field on cancer research, and many questions about their expression and functions remain unclear, like the presence of these ncRNAs in the bodily fluids. Since the use of circulating miRNAs in diagnostic screening methods and therapeutics have been intensively evaluated in many tumor types, the presence of released lncRNAs in bodily fluids, specially within extravesicles as exosomes, could be a source of novel potential biomarkers for diagnosis, prognosis and therapeutics purposes. Of our knowledge, there are only few data about this particular aspect of the lncRNAs ([Table II](#tII-ijo-46-01-0017){ref-type="table"}). The ucRNA (ultranconserved lncRNA) TUC399 was identified in exosomes derived from hepatocellular cancer cell lines, and the intercellular transfer of exosomal TUC399 can contribute to tumor growth and progression ([@b127-ijo-46-01-0017]). More recently, the same group demonstrated that the expression of lncRNAs linc-RoR (long intergenic non-coding RNA, regulator of reprogramming) in hepatocellular cancer is responsive to hypoxic conditions and the transfer of exosomal linc-RoR can modulate the intercellular response to hypoxia ([@b128-ijo-46-01-0017]). At present, reports regarding the circulating lncRNAs in bodily fluids are also scant in the literature. In a study evaluating the expression of the lncRNAs H19, HOTAIR and MALAT1 in gastric cancer plasma samples, Arita *et al* (2013) showed that only H19 is higher expressed in tumor samples than in the controls, and reported significanly decreased expression levels in post-operative tumor samples, indicating that the release of lncRNAs into the plasma can reflect the disease status ([@b129-ijo-46-01-0017]). In another report, Ren *et al* identified the MALAT1-derived mini-RNA (MD-miniRNA) as potential novel plasma biomarker in prostate cancer ([@b130-ijo-46-01-0017]). Some reports have demonstrated the use of lncRNA PCA3 as a specific and reliable marker detectable in urine samples from patients of prostate cancer, instead of the standardized use of the prostate-specific antigen (PSA). The evidence that highly upregulated in liver cancer (HULC) lncRNA expression is significantly higher in plasma tumor samples than in the healthy controls indicates the use of this lncRNA as potential circulating biomarker for diagnosis in hepatocellular cancer ([@b131-ijo-46-01-0017],[@b132-ijo-46-01-0017]). The evaluation of lncRNAs expression in plasma samples from leukemia and multiple myeloma showed that TUG1, MALAT1, HOTAIR and GAS5 are more highly expressed in leukemia than in the control samples, and only lincRNA-p21 is upregulated in multiple myeloma ([@b133-ijo-46-01-0017]). However, some limitations in using exosomal ncRNAs in diagnostics have been pointed out by many authors. The specificity and sensitivity of exosomal tumor marker detection in bodily fluids is still challenging. For example, serum and plasma-derived extravesicles as exosomes can be released by other than tumor cells, such as different blood cell types, affecting the purity of the tumor-derived exosome samples. In addition, the release of these exosomes depends on the age of the patient, infection or inflammation status of the disease, possibly introducing a bias in comparison analysis if not appropriately normalized to these conditions. Another issue that must be considered is the need of a standardized protocol for collecting and handling of the samples, as well as the exosomes isolation method ([@b95-ijo-46-01-0017],[@b106-ijo-46-01-0017]). Relevance in therapy -------------------- Since the miRNA is able to target multiple genes and signaling networks simultaneously, acting like oncogenes or tumor suppressor factors, it makes them a suitable tool for therapeutics interventions. A well and highly specific design is necessary for a successful result and to prevent undesirable targets. However, one of the principal limitations of this approach is the nuclease activity, causing the degradation before the miRNAs can achieve the targets. The use of vesicles for the delivery of exogenous therapeutic molecules to the targets has been intensively considered as a new promising therapeutic intervention. As mentioned before, the exosomes have the ability to transfer functional proteins and transcripts as perfect non-immunogenic carriers of therapeutic agents to target cells, making them suitable as therapeutic tool ([@b134-ijo-46-01-0017]). Considering that the exosome content can act as modulator of the microenvironment, facilitating tumor growth and metastasis, the blockade of the production, release and uptake by receptor cells could reverse the influence of the increased levels of exosomes in tumor progression ([@b95-ijo-46-01-0017]). Based on this, focusing on the inhibition of key components of the extravesicle production and release, such as the members of the ESCRT machinery, could be a useful strategy for therapy ([@b6-ijo-46-01-0017]). A third possible direction is represented by the drug or gene delivery by extravesicles to the target sites. Considering the elucidation of the intercellular transfer by exosomes, many reports have demonstrated the use of the extravesicles as small RNA carriers ([@b135-ijo-46-01-0017],[@b136-ijo-46-01-0017]). Intercellular transfer by exosomes can be used as miRNA carriers to restore miRNA expression in the target cells, where they play a therapeutic role as tumor suppressor factors. The targeted delivery of miRNAs by exosomes was demonstrated in a study in breast cancer cells expressing high levels of EGFR. This was achieved by the engineering of the donor cells to modify the surface of the exosomes to express the transmembrane domain of the PDGFR fused to the GE11 peptide. Then, the modified exosomes can deliver the let-7a miRNA after intravenous injection to EGFR-expressing xenograft breast cancer tissue in immunodeficient mice ([@b137-ijo-46-01-0017]). The ability of the miRNAs to target multiple genes can be a limitation for the specificity of this method as a selective approach for targeted therapy. The use of synthetic siRNA has been exploited as a more selective therapeutic tool. In an interesting study reported by Alvarez-Erviti *et al*, dendritic cells expressing a specific protein of the exosomal membrane Lamp2b fused to a neuron-penetrating RVG peptide were isolated from mice, and the exosomes derived from these cells were loaded with exogenous siRNA to GAPDH by electroporation. Subsequently, these RVG-targeted exosomes were intravenously injected, which delivered GAPDH siRNA to specific cells in the brain, leading to a selective gene knockdown ([@b138-ijo-46-01-0017]). Similarly, another report showed the delivery of siRNA into monocytes and lymphocytes by exosomes as gene delivery vector, reflecting a selective gene silencing of MAPK1 ([@b139-ijo-46-01-0017]). 5. Concluding remarks ===================== The discovery of the intercellular communication by the extravesicles has opened a new field for tumor biology. Exosomes can be found in the bodily fluids in a variety of tumor types and many reports have proved that the exosomal content as proteins, mRNA, miRNA and DNA can reflect the disease status, making them suitable for biomarkers for non-invasive diagnostic and prognosis purposes. With the advance of the engineering that permits the manipulation of the exosomal content and surface markers, many studies have been focusing on the development of therapeutic approaches in various tumor types, involving more specific delivery to the target tumor cells with more selective and efficient results. However, despite the efforts focusing on the study of the extracellular vesicles, specially exosomes, there are many aspects of the their biology that still need to be elucidated so that it would improve the advantages of the use of this promising approach in tumors. G.A.C. is The Alan M. Gewirtz Leukemia & Lymphoma Society Scholar. Study in G.A.C.'s laboratory is supported in part by the NIH/NCI grants 1UH2TR00943-01 and 1 R01 CA182905-01, Developmental Research Awards in Prostate Cancer, Multiple Myeloma, Leukemia (P50 CA100632) and Head and Neck (P50 CA097007) SPOREs, a SINF MDACC\_ DKFZ grant in CLL, a SINF grant in colon cancer, a Kidney Cancer Pilot Project, the Duncan Family Institutional Seed Funds, The Blanton-Davis Ovarian Cancer - 2013 Sprint for Life Research Award, the Laura and John Arnold Foundation, the RGK Foundation and the Estate of C. G. Johnson, Jr and by the CLL Global Research Foundation. {#f1-ijo-46-01-0017} ###### Summary of the studies reporting the identification of exosomal miRNAs in cancer. Tumor Sample Exosome extraction miRNA Refs. ----------------------------------- -------------------------------------------------------- -------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ -------------------------- Breast cancer Cell line SC, 0.22 μm filtering and UC miR-233 ([@b108-ijo-46-01-0017]) Breast cancer Cell line SC and UC or ExoQuick (System Biosciences) miR-210 ([@b116-ijo-46-01-0017]) Breast cancer Cell line SC, 0.22 μm filtering and UC miR-100, miR-17, miR-222, miR-342-3p, miR-451, miR-30a ([@b114-ijo-46-01-0017]) Breast cancer Cell line SC, 0.22 μm filtering, UC and sucrose gradient miR-198, miR-26a, miR-34a, miR-49a, let-7a, miR-328, miR-130a, miR-149, miR-602 and miR-92b ([@b99-ijo-46-01-0017]) Breast cancer Serum samples, tumor samples, cell line, animal models SC, UC miR-105 ([@b113-ijo-46-01-0017]) Cervical cancer Cervicovaginal lavage fluid SC and UC miR-21 and miR-146a ([@b126-ijo-46-01-0017]) Cholangiocarcinoma (biliary tree) Bile sample SC and 0.22 μm filtering miR-222, miR-126, miR-486-3p, miR-484, miR-19a, miR-19b, miR-16, miR-191, miR-31, miR-1274b, miR-618, miR-486-3p, miR-16, miR-1274b, miR-484, miR-191 ([@b125-ijo-46-01-0017]) Colorectal cancer Cell line SC and 0.22 μm filtering miR-21, miR-192, miR-221 ([@b107-ijo-46-01-0017]) Colorectal cancer Serum samples SC, 0.22 μm filtering and UC let-7a, miR-1229, miR-1246, miR-150, miR-21, miR-223 and miR-23 ([@b121-ijo-46-01-0017]) Esophageal cancer Serum samples SC, 0.45 μm filtering and ExoQuick (System Biosciences) miR-21 ([@b120-ijo-46-01-0017]) Gastric cancer Cell line SC, 0.1 μm filtering and UC let-7 family (a, b, c, d, e, f, g, i) ([@b105-ijo-46-01-0017]) Glioblastoma Tumor samples, serum samples SC, 0.22 μm filtering and UC let-7a, miR-15b, mR-16, miR-19b, miR-21, miR-26a, miR-27a, miR-92, miR-93, miR-320, miR-20 ([@b118-ijo-46-01-0017]) Glioblastoma Tumor samples, cell lines, animal models SC, 0.22 μm filtering, and UC miR-1 ([@b140-ijo-46-01-0017]) Leukemia Cell line ExoQuick (System Biosciences) miR-19a, miR-146-5p, miR-454, miR-18b, miR-574-3p, miR-21, miR-431, miR-345, miR-210, miR-197, miR-20a, miR-24, miR-19b, miR-130b, miR-106b, miR-224, miR-210, miR-652, miR-379, miR-185 ([@b117-ijo-46-01-0017]) Leukemia Cell line SC, 0.22 μm filtering and ExoQuick (System Biosciences) miR-17--92 cluster, miR-24, miR-222 ([@b109-ijo-46-01-0017]) Leukemia Cell line SC, 0.22 μm filtering and UC miR-1908 and miR-298 ([@b100-ijo-46-01-0017]) Lung adenocarcinoma Plasma samples Size exclusion by chromatography, magnetic beads (EpCAM) miR-17-3p, miR-21, miR-106a, miR-146, miR-155, miR-191, miR-192, miR-203, miR-205, miR-210, miR-212, miR-214 ([@b122-ijo-46-01-0017]) Lung adenocarcinoma Plasma samples ExoQuick (System Biosciences) miR-378a, miR-379, miR-139-5p, miR-200-5p, miR-151-5p, miR-30a-3p, miR-200b-5p, miR-629, miR-100 and miR-154-3p ([@b123-ijo-46-01-0017]) Lung adenocarcinoma Cell line SC, 0.22 μm filtering, and UC miR-192 ([@b112-ijo-46-01-0017]) Lung cancer Cell line UC and ExoQuick (System Biosciences) miR-21, miR-98, miR-133b, miR-138, miR-181a, miR-200c ([@b115-ijo-46-01-0017]) Lung cancer Plasma sample, bronchoalveolar lavage fluid SC, 0.22 μm filtering, and UC miR-222, miR-126, miR-144, miR-302a, miR-302c ([@b124-ijo-46-01-0017]) Melanoma Cell line SC, 0.22 μm filtering and UC; SC and ExoQuick (System Biosciences) miR-181b, miR-181a, miR-4802-3p, miR-23b, miR22, miR-107, miR-103a, miR-9, miR-338-3p ([@b101-ijo-46-01-0017]) Melanona and colon carcinoma Cell line SC, 0.22 μm filtering, prominin-1 based immuno-magnetic selection miR-216b, miR-889, miR-4307, miR-4272, miR-203, miR-4289, miR-3149, miR-203, miR-3145, miR-1911, miR-513a-3p, miR-3916, miR-886-3p, miR-1182, miR-3613-5p, let-7i, miR-3132, miR-3914, miR-3618, miR-1307, miR-3614-3p, miR-519c-3p, miR-3160, miR-3153, miR-4278, miR-3646, miR-3926, miR-515-5p, miR-3169, miR-590-3p, miR-525-5p, miR-548g, miR-365, miR-525-3p, miR-320d ([@b102-ijo-46-01-0017]) Multiple myeloma Cell line 0.22 μm filtering, UC and ExoQuick (System Biosciences) miR-125q-3p, miR-128, miR-15a, miR-185, miR-192, miR-212, miR-324-3p, miR-331-5p, miR-345, miR-422a, miR-429, miR-511, miR-576-3p, miR-618, miR-9, miR-1271, miR-139, miR-148, miR-151-3p, miR-15b, miR-19b-1, miR-21, miR-34b, miR-378, miR-589, miR-592, miR-625, miR-93 ([@b111-ijo-46-01-0017]) Ovarian cancer Serum samples, cell line Magnetic activated cell sorting (MACS) - EpCAM miR-21, miR-141, miR-200a, miR-200c, miR-200b, miR-203, miR-205, miR-214 ([@b119-ijo-46-01-0017]) Ovarian cancer Cell line SC, UC and sucrose gradient let-7 ([@b104-ijo-46-01-0017]) Prostate cancer Cell line SC, 0.22 μm filtering, and UC miR-143 ([@b110-ijo-46-01-0017]) Protaste cancer Cell line SC and UC miR-1280, miR-720 and miR-1260b ([@b103-ijo-46-01-0017]) SC, sequential centrifugation; UC, ultracentrifugation; MV, microvesicle. ###### Summary of the reports of the circulating lncRNAs in cancer. Tumor Sample Extravesicles isolation Long ncRNA Refs. ------------------------------- -------------------------------- ------------------------------------ ----------------------------------------- -------------------------- Gastric cancer Plasma samples, cell lines NA H19, HOTAIR, MALAT1 ([@b129-ijo-46-01-0017]) Hepatocellular cancer Cell line UC and density gradient separation TUC399 ([@b127-ijo-46-01-0017]) Hepatocellular cancer Tissue samples, plasma samples NA HULC ([@b132-ijo-46-01-0017]) Hepatocellular cancer Cell line, animal model SC and UC linc-RoR ([@b128-ijo-46-01-0017]) Leukemia and multiple myeloma Plasma samples NA TUG1, MALAT1, HOTAIR, lincRNA-p21, GAS5 ([@b133-ijo-46-01-0017]) Prostate cancer Tissue samples, plasma samples NA MALAT-1 and PCA3 ([@b130-ijo-46-01-0017]) SC, sequential centrifugation; UC, ultracentrifugation; NA, not applied.

INTRODUCTION ============ Spinocerebellar ataxias (SCA) represents a large and complex group of heterogeneous autosomal dominant degenerative diseases characterized by progressive degeneration of the cerebellum and its afferent and efferent connections. Other nervous system structures are typically affected, including the basal ganglia, brainstem nuclei, pyramidal tracts, the posterior column and anterior horn of the spinal cord, and the peripheral nerves ([@b1-cln_67p443]-[@b6-cln_67p443]). SCA are clinically characterized by the presence of cerebellar gait and limb ataxia (with dysmetria, dysdiadochokinesia, intention tremor, dysarthria, and nystagmus), which may be accompanied by extracerebellar signs, such as ophthalmoplegia, pyramidal signs, movement disorders (including parkinsonism, dystonia, myoclonus, and chorea), dementia, epilepsy, visual disorders (including pigmentary retinopathy), and peripheral neuropathy ([@b1-cln_67p443]-[@b6-cln_67p443]). SCA have a prevalence ranging from 1 to 5 cases per 100,000 individuals ([@b7-cln_67p443],[@b8-cln_67p443]), and disease onset typically occurs between 30 and 50 years of age, although cases developing before the age of 20 and after the age of 60 have also been described ([@b2-cln_67p443]-[@b10-cln_67p443]). The degenerative neuropathological process has been studied in depth in transgenic mice and Drosophila models ([@b2-cln_67p443]-[@b4-cln_67p443],[@b3-cln_67p443],[@b8-cln_67p443]). Neuroimaging, particularly magnetic resonance imaging, typically reveals cerebellar atrophy with or without brainstem involvement (olivopontocerebellar atrophy) ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b9-cln_67p443],[@b10-cln_67p443]). Initially defined as autosomal dominant cerebellar ataxias, SCA was subsequently classified by Harding into the following four basic types: type 1, which is characterized by cerebellar ataxia (CA) with optic atrophy, ophthalmoplegia, dementia, amyotrophy, and extrapyramidal signs; type 2, involving retinal degeneration and accompanied by ophthalmoplegia and extrapyramidal signs; type 3, which is considered a "pure" type of CA; and type 4, which may present as deafness and myoclonia in addition to CA ([@b5-cln_67p443]). With recent advances in molecular genetics, several SCA genetic loci and genes have been identified on different chromosomes, and these findings have enabled the application of an improved classification system based on clinical as well as genetic data ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443]),. Thirty-two different types of SCA have been identified to date, and these are designated SCA1 to SCA36. Dentatorubral-pallidoluysian atrophy (DRPLA) has also been included in this group of disorders. The particular gene responsible for each type of disease has been identified for SCA types 1-3, 5-8, 10-15, 17, 27, 31, and DRPLA. The remaining types (SCA 4, 18-23, 25, 26, 28-30, 32, 33-35, and 36) have been defined by linkage studies, as the associated genes and mutations have not yet been identified ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b9-cln_67p443],[@b10-cln_67p443]),. Finally, it should be mentioned that SCA types 9 and 24 remain undefined, and these two types have been reserved for disorders yet to be described in the literature. Additionally, SCA16 appears to be identical to SCA15, and SCAs 29 and 15, as well as SCAs 22 and 19, may represent different allelic forms of the same gene ([@b2-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b11-cln_67p443],[@b12-cln_67p443]). SCA3 is the most common form of the disease worldwide, whereas the prevalence of types 1, 2, 6, 7, and 8 is varied depending upon the ethnic background of the population ([@b1-cln_67p443],[@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b9-cln_67p443]-[@b12-cln_67p443]). The objective of this study was to evaluate the genotype-phenotype correlations in 104 Brazilian families with SCA. METHODS ======= Patients -------- We studied 150 patients from 104 families with SCA, which had been extracted from a large Brazilian series of 190 SCA families (382 clinically affected family members and 296 patients evaluated by molecular genetic testing) and had received positive test results at the Hospital de Clínicas of the Federal University of Paraná in Curitiba, Brazil between 1989 and 2009. The inclusion criteria included the following: 1) a progressive clinical phenotype in which ataxia was the prominent symptom; and 2) a positive familial history compatible with autosomal dominant inheritance. All of the patients were evaluated by at least one neurologist (HT) using a standard protocol that assessed gender, age of onset, duration of the disease, mean CAG or ATTCT polynucleotide expansion, clinical manifestations (such as CA, ocular movement disorders, visual loss, movement disorders, pyramidal signs, peripheral nerve signs, cognitive dysfunction, or epilepsy), neuroimaging findings (brain CT and/or MRI), and any additional findings. All of the patients underwent routine laboratory tests, including CSF studies, electroencephalography (EEG), nerve conduction velocity with electromyography (NCV/EMG), and neuropsychological tests (NPT) in selected cases. Signed informed consents were obtained following a protocol approved by the Institutional Ethics Committee of the Federal University of Paraná. Genetic Analysis ---------------- Molecular diagnostic testing for SCA types 1, 2, 3, 6, 7, and DRPLA was performed at the Centre for Research in Neuroscience at the Montreal General Hospital Research Institute of McGill University in Montreal, Quebec, Canada between 1994 and 1998 (Prof. Guy Roleau, Dr. Izabel Silveira, and Dr. Iscia Lopes-Cendes) and subsequently (from 1998 to 2000) at the Medical Genetic Department of UNICAMP (Prof. Iscia Lopes-Cendes). Molecular genetic tests were performed from 2003 to the present at the Molecular Biology Laboratory (Neurology Service, Hospital de Clínicas, Federal University of Paraná: Prof. Lineu C. Werneck) and Genetika Laboratory (Prof. Salmo Raskin) in Curitiba/PR (SCA 1, 2, 3, 6, 7, 8, 12, and 17). Molecular diagnostic tests for SCA10 were performed at the Baylor DNA Diagnostic Laboratory (Baylor College of Medicine, Houston, Texas, USA) (Prof. T. Ashisawa\'s group) and subsequently at the Galveston Department of Neurology (University of Texas Medical Branch) (Prof. T. Ashizawa). Peripheral blood was collected from patients and relatives, and genomic DNA was isolated from peripheral blood leukocytes using the standard technique (Sambrock et al. 1989, Cold Spring Harbor Laboratory Press, NY, USA) ([@b13-cln_67p443]). Expanded triplet repeats in the genes responsible for SCA1, SCA2, SCA3, SCA6, SCA7, SCA8, SCA12, SCA17, and DRPLA were amplified using the primer pairs Rep-1/Rep-2 (Orr et al. 1993), F-1/F-2 (Sanpei et al. 1996; Pulst et al. 1996), MJD25/MJD52 (Kawaguchi et al. 1994), S-5-F1/S-5-R1 (Zhunchenko et al. 1997), 4U1024/4U716 (David et al. 1997), SCA8-F3/SCA8-R4 (Koob et al. 1999), PPP2R2B-A/PPP2R2B-B (Holmes et al. 1999), TBP-F/TBP-R (Nakamura et al. 2001) and CTG-B37 (699)/(840) (Koide et al. 1994), respectively ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b9-cln_67p443],[@b10-cln_67p443],[@b11-cln_67p443],[@b12-cln_67p443]). The analysis of the ATTCT repeat region within the SCA10 gene was performed by polymerase chain reaction (PCR) amplification using the primers attct-L (5\'-AGAAAACAGATGGCAGAATGA-3\') and attct-R (5\'-GCCTGGGCAACATAGAGAGA-3\'), as previously described ([@b11-cln_67p443],[@b14-cln_67p443]). Patient DNA samples that showed a single, normal SCA10 allele by PCR underwent Southern blot analysis to assess large expansion ([@b14-cln_67p443]). The identification of normal and expanded allele size was based on previous reports, which had defined the following conditions: for the SCA1 gene, affected alleles contain 40-82 CAG repeats (normal alleles demonstrate 6-39 CAG repeats); for SCA2, affected alleles contain 33-64 repeats (normal alleles have 14-31); in SCA3, affected alleles contain 54-86 repeats (normal alleles have 12-42); in SCA6, affected alleles contain 19-30 repeats (normal alleles have 4-18); in SCA7, affected alleles contain 37-200 repeats (normal alleles have 4-27); in SCA8, affected alleles contain 107-127 repeats (normal alleles have 16-91); in SCA12, affected alleles contain 55-78 repeats (normal alleles have 7-32); in SCA17, affected alleles contain 47-63 repeats (normal alleles have 25-44); and in DRPLA, affected alleles contain 49-79 CAG repeats (normal alleles have 6-36) ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b9-cln_67p443],[@b10-cln_67p443],[@b11-cln_67p443],[@b12-cln_67p443]). In SCA10, the pentanucleotide ATTCT repeat in the SCA10 gene was evaluated in reference to pathological expansion consisting of 800-4,500 repeats (normal alleles have 10-21 repeats) ([@b14-cln_67p443]). Statistical Analysis -------------------- The statistical analysis of the results was performed using basic descriptive statistics and the correlation coefficient (r), Student\'s t-test, chi-square test, and Yates\' correction. The level of statistical significance was set to *p*-values \<0.05. RESULTS ======= Mutations were identified in 66.3% of the families and were most frequently identified for SCA3 (72.46%). SCA10 represented the second most common type of SCA (11.60%), followed by SCA2 (7.25%), SCA7 (4.34%), SCA1 (2.90%), and SCA6 (1.45%). Of the affected patients examined, the male to female ratio was 1:1.2, and the age at onset ranged from 22 to 60 years with a mean onset at 35 years ±6.2 years of age. Also, the average length of time with disease for these patients was 13.6±10.7 years (variance of 1-32 years). [Table 1](#t1-cln_67p443){ref-type="table"} shows a brief summary of the clinical data and the neuroimaging findings for SCA types 1,2,3,6,7,10 as well as overall SCA, excluding SCA10. This table also includes the demographic, clinical, and molecular data for each form of SCA. In comparison to patients with SCA10, those with SCA3 presented more frequently with appendicular ataxia, ophthalmoplegia, diplopia, facial fasciculations, eyelid retraction (bulging eyes), severe hyperreflexia, severe hypo/areflexia, and muscular fasciculations. Significant differences (*p*\<0.05) were also observed between these two SCA types regarding the presence of Babinski\'s sign, spasticity, amyotrophy, and cognitive dysfunction. In contrast, ocular dysmetria was significantly more frequently seen in patients with SCA10. In addition, when comparing patients with SCA10 to those with all other forms of SCA (excluding SCA3), this specific subtype presented more frequently with nystagmus and ocular dysmetria, whereas ophthalmoplegia/paresis, diplopia, slow saccadic movements, and hyporeflexia were observed less commonly in patients with SCA10 than other forms of SCA (SCA 1, 2, and 7). Slow saccadic eye movements associated with hypo/areflexia are typically suggestive of SCA2. As only one case of SCA6 was detected in our patient sample, no comparisons between this subtype and the other SCA subtypes could be made. The overall results are presented in [Table 2](#t2-cln_67p443){ref-type="table"}. The neurological signs most commonly found in patients with SCA3, SCA10, and other forms of SCA (SCA types 1, 2, 6, and 7) are summarized in [Table 3](#t3-cln_67p443){ref-type="table"}. DISCUSSION ========== In this sample of Brazilian patients with SCA, we observed that SCA3, which is also known as Machado-Joseph disease, was the most commonly identified subtype. In fact, SCA3 is also the most commonly encountered SCA worldwide, despite certain regional variations (SCA1 in Italy, SCA2 in India and Cuba, and SCA7 in Sweden, Finland, and South Africa) ([@b2-cln_67p443],[@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443]),. In Brazil, several previously published studies of patients with SCA have identified SCA3 as occurring at the highest frequency ([@b9-cln_67p443],[@b10-cln_67p443],[@b12-cln_67p443],[@b15-cln_67p443]). Surprisingly, in our patient series, SCA10 was the second most frequently encountered SCA, accounting for approximately 12% of the cases. As the causative genetic mutation for this subtype was discovered fairly recently, SCA10 was not investigated prior to the year 2000 and was therefore was not reported in the previously published SCA patient series. Several other investigators ([@b6-cln_67p443],[@b11-cln_67p443],[@b16-cln_67p443]) evaluated the occurrence of SCA10 in European, Asian, and North American patients, although no cases were identified. SCA10 was originally described in Mexican patients and subsequently in Brazilian patients, and SCA10 represents the second most common type of SCA after SCA2 (in Mexico) and SCA3 (in Brazil) ([@b6-cln_67p443],[@b11-cln_67p443]),. Moreover, cases of SCA10 have been detected in other Latin American countries, including Argentina and Venezuela ([@b19-cln_67p443]). The other forms of SCA found in this patient series consisted of SCA2, SCA7, SCA1, and SCA6. No cases of DRPLA were found. We initially compared the two most commonly identified forms of SCA and found that several neurological signs could be significantly correlated with either SCA3 or SCA10. In general, clinical experience has demonstrated that SCA3 presents great phenotypic heterogeneity with large inter-familial variation. However, the neurological signs identified in the current study (including ophthalmoplegia, diplopia, facial fasciculations, eyelid retraction (bulging eyes), spasticity, severe hyperreflexia, Babinski\'s sign, severe hypo/areflexia, amyotrophy, and muscular fasciculations) reflect a more common phenotype among patients with SCA3. The genotype-phenotype correlation in SCA3 has been the focus of several studies, and many of these published results are identical to those reported in the present study ([@b20-cln_67p443],[@b21-cln_67p443]). Since the original description of SCA3 by Coutinho and Andrade in 1978, which was followed by additional studies from the same group, most investigators have emphasized the presence of a characteristic phenotype consisting of ophthalmoplegia, bulging eyes, facial fasciculations, pyramidal signs, and peripheral neuropathy, as well as dystonia in younger patients ([@b20-cln_67p443]-[@b22-cln_67p443]). More recently, the observation of variant phenotypes for SCA3 has led to the description of the following sub-phenotypes: subtype I (with a predominance of extrapyramidal signs and dystonia); subtype II (with CA and pyramidal tract signs); subtype III (with CA and peripheral neuropathy signs); subtype 4 (with parkinsonism); and subtype 5 (resembling spastic paraplegia) ([@b21-cln_67p443]). In contrast, the study by Schöls et al. ([@b23-cln_67p443]) describing German patients did not report the typical signs described for Portuguese SCA3 cases, such as bulging eyes, dystonia, and rigidity. In addition, Klockgether et al. as well as other groups have suggested that significant overlap exists between different SCAs, which impedes the utility of purely clinical-based diagnoses ([@b4-cln_67p443],[@b24-cln_67p443]). Finally, other authors have attempted to describe oculomotor phenotypes suggestive of several forms of SCA, although these finding have no significant impact in regards to daily clinical practice ([@b25-cln_67p443]). With respect to movement disorders in our patient series, we observed that 15.3% of patients presented signs of dystonia and parkinsonism (nine patients), and the majority of these symptoms were presented by patients with SCA3. Similar results have also been reported by Schols et al. and Garcia Ruiz et al. ([@b26-cln_67p443],[@b27-cln_67p443]). The majority of patients with SCA10 in the current study presented a phenotype of "pure" CA with occasional slight pyramidal tract signs (hyperreflexia and discrete spasticity). This presentation without epilepsy or peripheral neuropathy differs from the phenotype described in Mexican patients, who were shown to present with epilepsy in 72.2% of cases and peripheral nephropathy in 66% of cases ([@b28-cln_67p443]-[@b32-cln_67p443]). These observations have led to debates regarding the reason for these different phenotypic manifestations in Mexican and Brazilian SCA10 patients. The first explanation for these differences correlated the observed phenotype with the expansion of the ATTCT pentanucleotide repeat size (greater on average in Mexicans) but was refuted in 2004 ([@b11-cln_67p443]). Later, Matsuura et al. ([@b33-cln_67p443]) showed that the presence of a complex interruption pattern composed of two different repeat interruptions, ATTTTCT and ATATTCT, could explain the presence of a phenotype consisting of epilepsy associated with CA. More recently, Teive at al. evaluated the frequency of epilepsy in a group of 80 patients from 10 Brazilian families with SCA10 and identified the presence of epilepsy in only 3.75% of the cases, whereas case reports from Argentina and Venezuela have presented a phenotype resembling those of Mexican origin ([@b18-cln_67p443]). Another interesting clinical aspect of patients with SCA10 is that they generally do not present ophthalmoplegia/paresis, although they do commonly present nystagmus in lateral views and ocular dysmetria ([@b11-cln_67p443],[@b19-cln_67p443],[@b34-cln_67p443]). Finally, we also evaluated the genotype-phenotype correlation between patients with SCA10 and all other types of SCA (excluding SCA3). These results demonstrated that ophthalmoplegia/paresis and diplopia were the most frequently observed signs in patients with other types of SCA, including SCA 1, 2, and 7. However, the presence of slow saccadic ocular movements and hypo/areflexia were more common in patients with SCA2. These findings have been repeatedly reported in the literature since the initial description of SCA2 ([@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b35-cln_67p443]). Also, we should highlight that the association between slow saccadic eye movements and signs of peripheral neuropathy are highly suggestive, at least from a clinical standpoint, of SCA2 ([@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b35-cln_67p443]). The same is also true for the association between CA and vision loss due to retinopathy in the case of SCA7 ([@b3-cln_67p443],[@b4-cln_67p443],[@b6-cln_67p443],[@b35-cln_67p443]). In our clinical setting, neither patients with SCA1 nor SCA6 presented a characteristic phenotype, which was likely due to the small number of cases identified. SCA represents an extensive and complex group of autosomal dominant neurodegenerative diseases, and to date, SCA3, SCA1, SCA2, SCA6 and SCA7 are the most frequently identified types of SCA worldwide. In this series of Brazilian SCA patients, SCA3 was the most commonly identified type, followed by SCA10. Upon comparisons between SCA3, SCA10 and other SCAs, patients with each of these conditions revealed several clinical features that may be useful for the clinical classification of an affected patient or family. In SCA3 patients, the phenotype was highly pleomorphic, although CA, ophthalmoplegia, diplopia, eyelid retraction, facial fasciculation, pyramidal signs and peripheral neuropathy were the most frequently observed signs of disease. In SCA10 patients, the phenotype was rather peculiar and consisted of pure CA, saccadic eye movement, and ocular dysmetria. For the other SCAs, SCA2 and SCA7 exhibited highly suggestive phenotypes, presenting as CA with slow saccadic ocular movements and profound areflexia in SCA2 and cerebellar ataxia and visual loss in SCA7. The phenotypes of patients with SCA1, SCA6, and SCA10 were commonly nonspecific (e.g., pure cerebellar ataxia). ACKNOWLEDGMENTS =============== The authors thank Prof. Guy Rouleau, Dr. Izabel Silveira (Centre for Research in Neuroscience, The Montreal General Hospital Research Institute, McGill University, Montreal, Quebec, Canada), Aline Freund, PhD (Molecular Biology Laboratory, Neurology Service, Hospital de Clínicas, Federal University of Paraná), Dr. Ben Roa, Dr. Ping Fang (Baylor DNA Diagnostic Laboratory, Baylor College of Medicine, Houston, Texas, USA), Dr. Rui Gao (University of Texas Medical Branch, Galveston, Texas, USA), Jilin Liu (University of Florida, Gainesville, Florida, USA), and Ms. Misti C. White (Baylor DNA Diagnostic Laboratory, Baylor College of Medicine, Houston, Texas, USA), for their collaboration in performing the molecular genetic exams of SCA patients. The authors also thank Prof. Rosana H. Scola (Electroneuromyography Unit, Neurology Service, Hospital de Clínicas, Federal University of Paraná) and Dr. Alexandre L. Longo, Dr. Carla C. Moro, Dr. Norberto Cabral (Clínica Neurológica de Joinville, Santa Catarina), Prof. Ylmar Correa Neto (Internal Medicine Department, Federal University of Santa Catarina), Prof. Paulo N. D. Sá, and Prof. Paulo C. T. Bittencourt (Neurology Service, Federal University of Santa Catarina) for their help with this study. Support by NIH NS041547 (TA). **Conflicts of interests** Drs. Arruda, Munhoz, Raskin, Lopes-Cendes, Werneck, and Ashizawa have nothing to report. Dr. Teive has received speaking honoraria from Allergan, Boeringher-Ingelheim, Ipsen, Roche, and Novartis. ###### SCA Types 1, 2, 3, 6, 7, 10 - Clinical Data/Neuroimaging. Clinical Data/Neuroimaging X SCA type: SCA1 SCA2 SCA3 SCA6 SCA7 SCA10 SCATotal ------------------------------------------ -------- --------- ---------- -------- -------- ---------- ---------- Expansion CAG/ATTCT (mean) 48.5 45.66 70.27 24 64.8 1820 50.64 Patient number 4 12 101 1 5 27 123 Gender     Male 4 7 41 0 3 14 55     Female 0 5 60 1 2 13 68 Age of onset (mean) 33.75 34.17 34.87 43.0 28.4 35.00 34.83 Disease duration (mean) 7.65 6.16 8.65 9.0 7.80 13.60 7.85 Cerebellar Dysfunction     Gait Ataxia 4(100) 12(100) 100(99) 1(100) 5(100) 27(100) 122(100)     Appendicular Ataxia 1(25) 11(92) 92(9) 1(100) 1(20) 19(70.3) 106     Dysarthria 4(100) 12(100) 96(95) 1(100) 5(100) 27(100) 118 Ocular Movement Disorders     Nystagmus 4(100) 7(58) 98(97) 1(100) 3(60) 27(100) 112     Ocular Dysmetria 2(50) 2(16) 14(13.8) 0 1(20) 20(74) 19     Slow Saccadic Movement 0 12(100) 6(5.9) 0 1(20) 0 19     Ophthalmoplegia/paresis 4(100) 6(50) 90(89) 0 5(100) 0 105     Diplopia 2(50) 4(32) 61(60.4) 0 4(80) 0 71     Facial Fasciculations 1(25) 1(8) 42(41.6) 0 1(20) 0 45     Lid Retraction 2(50) 4(32) 88(87) 0 2((40) 3(11.1) 96 Visual Loss 0 0 0 0 5(100) 0 0 Movement Disorders     Parkinsonism 0 1(8) 8(7.9) 0 0 0 9     Dystonia 0 1(8) 12(11.9) 0 1(20) 0 14     Myoclonus 0 0 0 0 0 0 0     Chorea 0 0 0 0 0 0 0 Pyramidal Dysfunction     Babinski\'s Sign 2(50) 0 20(19.8) 0 1(20) 0 23     Hyperreflexia 4(100) 0 69(68.3) 0 3(60) 518.5) 76     Spasticity 4(100) 0 32(31.7) 0 1(20) 3(11.1) 37 Peripheral Nerve Abnormalities     Hypo/areflexia 0 12(100) 22(21.8) 0 0 0 34     Amyotrophy 0 2(16) 17(16.8) 0 0 0 19     Paresis 0 0 0 0 0 0 0     Fasciculations 0 2(16) 22(21.8) 0 0 0 24     Exteroceptive Sensation Dysfunction 0 4(32) 12(11.9) 0 0 0 16     Proprioceptive Sensation Dysfunction 0 2(16) 5(4.9) 0 0 0 7 Cognitive Dysfunction 0 3(25) 0 0 1(20) 2(7.4) 4 Epilepsy 0 0 0 0 0 0 0 Neuroimaging: Cerebellar Atrophy: 4(100) 12(100) 96(95) 1(100) 5(100) 27(100) 118 NI:Cerebellar + BS Atrophy 0 1(8) 2(1.9) 0 0 1(3.7) 3 Others: Dysphagia 1(25) 2(16) 44(43.5) 0 1(20) 6(22.2) 48 SCA =  Spinocerebellar Ataxia; NI =  Neuroimaging; BS =  Brainstem; SCA Total =  All SCAs, without SCA10. (%) ###### A comparative statistical analysis between the clinical data and neuroimaging results of patients with SCA3 x SCA10 and SCA10 x other SCAs. SCA3 X SCA10, SCA10 X Other SCAs (X^2^, p) SCA10 SCA3 X^2^ *p*-value SCA10 OthersSCAs X^2^ *p*-value -------------------------------------------- ------- ------ -------- ------------ ------- ------------ -------- ------------ Patient Number 27 101 \- \- 27 22 \- \- Cerebellar Dysfunction Gait Ataxia 27 100 0.27 0.6037 27 22 \- \- Appendicular Ataxia 19 92 7.94 **0.0048** 19 14 0.25 0.6171 Dysarthria 27 96 1.39 0.2383 27 22 \- \- Ocular Movement Disorders Nystagmus 27 98 0.82 0.3648 27 14 11.73 **0.0006** Ocular Dysmetria 20 14 39.60 **0.0000** 20 5 12.79 **0.0003** Slow Saccadic Movement 0 6 1.68 0.1946 0 13 21.72 **0.0000** Ophthalmoplegia/paresis 0 90 81.04 **0.0000** 0 15 26.53 **0.0000** Diplopia 0 61 31.15 **0.0000** 0 10 15.42 **0.0001** Lid Retraction 3 88 59.91 **0.0000** 3 8 Y 3.11 0.0779 Facial Fasciculations 0 42 16.71 **0.0000** 0 3 Y 1.91 0.1672 Visual Loss 0 0 \- \- 0 5 Y4.58 **0.032** Movement Disorders Parkinsonism 0 8 2.28 0.1310 0 1 1.25 0.2631 Dystonia 0 12 3.54 0.0599 0 2 2.56 0.1097 Myoclonus 0 0 \- \- 0 0 \- \- Chorea 0 0 \- \- 0 0 \- \- Pyramidal Dysfunction Babinski\'s Sign 0 20 6.34 **0.0118** 0 3 Y 1.91 0.1672 Hyperreflexia 5 69 21.66 **0.0000** 5 7 1.16 0.2816 Spasticity 3 32 4.54 **0.0332** 3 5 1.20 0.2739 Peripheral Nerve Abnormalities Hypo/areflexia 0 22 7.10 **0.0077** 0 12 19.50 **0.0000** Amyotrophy 0 17 5.24 **0.0221** 0 2 2.56 0.1097 Paresis 0 0 \- \- 0 0 \- \- Fasciculations 0 22 7.10 **0.0077** 0 2 2.56 0.1097 Exteroceptive Sensation Dysfunction 0 12 3.54 0.0599 0 4 Y 3.20 0.0739 Proprioceptive Sensation Dysfunction 0 5 1.39 0.2383 0 2 2.56 0.1097 Cognitive Dysfunction 2 0 Y 3.55 0.0557 2 4 1.31 0.2525 Epilepsy 0 0 \- \- 0 0 \- \- NI: Cerebellar Atrophy 27 96 1.39 0.2383 27 22 \- \- NI: Cerebellar + BS Atrophy 1 2 0.28 0.5990 1 1 0.02 0.8822 Others: Dysphagia 6 44 Y 3.23 0.0723 6 4 0.12 0.7271 X^2^  =  Chi-Square Test; Y =  Yates\'s Correction; NI = Neuroimaging; BS = Brainstem; P = Probability (*p*\<0.05). ###### The neurological signs most commonly identified in patients with SCA10, SCA3, or other types SCA. SCA10 SCA3 Other SCAs ------------------------- ------- ------ ------------ Appendicular Ataxia \- \+ \- Nystagmus \+ \- \+ Ocular Dysmetria \+ \- \- Slow Saccadic Movement \- \- \+ Ophthalmoplegia/paresis \- \+ \+ Diplopia \- \+ \+ Lid Retraction \- \+ \- Facial Fasciculations \- \+ \- Babinski\'s Sign \- \+ \- Hyperreflexia \- \+ \- Spasticity \- \+ \- Hypo/Areflexia \- \+ \+ Amyotrophy \- \+ \- Fasciculations \- \+ \- [^1]: Teive HA conceived and designed the study and was also responsible for the acquisition, analysis and interpretation of data, manuscript drafting, critical revision of the manuscript for important intellectual content and study supervision. Werneck LC conceived and designed the study and was also responsible for the analysis and interpretation of data, critical revision of the manuscript for important intellectual content, study supervision and administrative, technical, and material support. Arruda WO was responsible for the data acquisition, analysis and interpretation, manuscript drafting and critical revision of the manuscript for important intellectual content. Munhoz RP was responsible for the data acquisition, analysis and interpretation, critical revision of the manuscript for important intellectual content, study supervision, and administrative, technical and material support. Raskin S was responsible for the data analysis and interpretation, critical revision of the manuscript for important intellectual content, and administrative, technical, and material support. Lopes-Cendes I was responsible for the data analysis and interpretation, critical revision of the manuscript for important intellectual content, study supervision, administrative, technical, and material support. Ashizawa T was responsible for the critical revision of the manuscript for important intellectual content, study supervision;, and contributed with administrative, technical and material support.

1. Introduction =============== This paper reports the development of a Standard Reference Material (SRM) which characterizes the zero-dispersion wavelength (*λ*~0~) and the dispersion slope (*S*~0~) at *λ*~0~ of single-mode optical fibers. We have documented a system which measures the dispersion of both dispersion-unshifted (*λ*~0~ near 1.3 μm) and dispersion-shifted fibers (*λ*~0~ near 1.55 μm). While the principal system uses the frequency-domain phase shift technique \[[@b1-j23mec]\], differential phase shift \[[@b2-j23mec]\] and four-wave mixing techniques \[[@b3-j23mec]\] have also been investigated. The fiber SRMs have their *λ*~0~ value measured with a combined expanded uncertainty (coverage factor *k* = 2, thus a 2 standard deviation estimate) of 0 ± 0.060 nm. Dispersion slope was also studied, but with a more limited scope. The slope *S*~o~ is determined with an expanded uncertainty (coverage factor *k* = 2) of ± 0.008 ps/nm^2^. Current high bit-rate telecommunication systems, both terrestrial and transoceanic, require precise information about the zero-dispersion wavelength of the installed fiber. Operating the system at a wavelength within a few nanometers of *λ*~0~ enables the use of bit rates up to and exceeding 10 Gbit/s. Knowledge of the system's operating wavelength with respect to *λ*~0~ is also crucial for the avoidance of detrimental nonlinear effects, such as four wave mixing \[[@b4-j23mec], [@b5-j23mec]\]. A standard reference fiber is useful to manufacturers because of the difficulty involved in accurately measuring *λ*~0~ and *S*~0~. A National Institute of Standards and Technology sponsored interlaboratory comparison between members of the Telecommunications Industry Association (TIA), reported individual measurements of high precision, but large systematic error \[[@b6-j23mec]\]. This is a situation where SRM calibration artifacts can be especially useful. 2. Frequency-Domain Phase Shift System ====================================== We have constructed a frequency-domain phase shift system designed to minimize systematic errors in measurements of *λ*~0~. Original work using this technique was performed with light-emitting diode (LED) sources \[[@b7-j23mec]\], later systems utilizing laser diodes were developed for measuring optical fibers in both the 1.3 μm and 1.55 μm regions \[[@b8-j23mec], [@b9-j23mec]\]. [Figure 1](#f1-j23mec){ref-type="fig"} shows a schematic diagram of the NIST system. Our system differs from typical implementations of the frequency-domain phase shift technique in the following ways: Higher modulation frequency (1.9 GHz) to achieve a large phase angle shift per unit group delay.Laser sources for increased optical power necessary to obtain sufficient signal-to-noise ratios for 0.1° phase angle resolution (electrical).Interferometric monitoring of the wavelength.Chirp free external modulation of the source.Temperature stabilization of the fiber sample. The continuous wave (CW) output of an external cavity tunable laser diode, with a linewidth under 2 GHz and an optical power of 1 mW, is connected to an external intensity modulator by a polarization-maintaining fiber. A grating-tuned erbium fiber laser has also been used, as an alternative source, in the 1.55 μm region. A 1.9 GHz electrical signal from a temperature-stabilized crystal oscillator is used to drive the integrated-optic Mach-Zehnder modulator. The modulated light is passed through a fiber coupler, with 20 % of the optical power being monitored by a commercial interferometric wavemeter. The remaining 80 % passes through the fiber under test. The test fiber is placed in a temperature controlled chamber whose temperature is stabilized at 23 °C (slightly above room temperature). The temperature is determined with a calibrated quartz thermometer, which has an expanded uncertainty (*k* = 2) of less than 0.1 °C. Temperature gradients within the chamber limit our fiber temperature measurement to a standard uncertainty (*k* = 1) of ± 0.15 °C. Precise control of the fiber's temperature is necessary because the value of *λ*~0~ is temperature dependent (+ 0.030 nm/°C for dispersion-shifted fiber). The optical signal is then detected and amplified by a low noise amplifier. After the RF signal passes through a narrow bandpass filter (3 dB width of 10 MHz), it is read by the vector voltmeter (phase detector). The reference port of the vector voltmeter receives an RF signal directly from the crystal oscillator which is used as a phase reference. Because we are interested in measuring dispersion only in the vicinity of *λ*~0~, we can use a high modulation frequency without problems due to modulo-2*p* phase uncertainty. The vector voltmeter has a phase sensitivity of 0.1°, which at our modulation frequency corresponds to a temporal resolution of 0.15 ps. The linearity of the vector voltmeter's phase response is verified by a mechanically calibrated variable air gap which is placed in the optical beam path. After a fiber measurement, the specimen is removed, the system's short fiber jumpers are connected and the measurement repeated; any residual dispersion in the system is subtracted out. A dispersion measurement is performed by recording the phase from the vector voltmeter while tuning the laser over the wavelength region of interest. Each data point consists of a wavelength value and its corresponding phase. A change in group delay δ*τ*, normalized with respect to length, is related to the change in phase δ*ϕ* by the relation $$\delta\tau = \frac{\delta\phi}{2\pi f_{m}L},$$where *f*~m~ is the modulation frequency and *L* is the fiber length. To calculate the dispersion of a fiber specimen, we must first fit the group delay data to a theoretical curve and then differentiate the best fit group delay curve with respect to wavelength. A sample fit to group delay data is shown in [Fig. 2](#f2-j23mec){ref-type="fig"}. The dispersion coefficient *D*, defined as $$D = \frac{d\tau}{d\lambda},$$has the unit ps/(nm · km) and goes to zero at the so called "zero-dispersion" wavelength. Using the calculated dispersion values we can also obtain the dispersion slope (*S*), $$S = \frac{dD}{d\lambda}.$$ The parameter *S*~o~ is defined as the dispersion slope evaluated at *λ*~0~. For dispersion-unshifted fibers, chromatic dispersion is dominated by material dispersion. Group delay, in this case, is well described by the Sellmeier equation. Following the recommendations of Fiber Optic Test Procedure (FOTP) 169 \[[@b10-j23mec]\], we fit the group delay data to a three-term Sellmeier equation $$\tau(\lambda) = a\lambda^{2} + b\lambda^{- 2} + c.$$ The unknown variables *a*, *b*, and *c* are solved for by the method of least squares. The group delay in a dispersion-shifted fiber has a different functional form, due to a larger waveguide contribution to dispersion. In this case, again following FOTP-169, we fit the group delay data to a quadratic equation $$\tau(\lambda) = a\lambda^{2} + b\lambda + c.$$ There are many types of equations for fitting group delay data including: the 5-term Sellmeier, various polynomials and fits with terms involving natural logarithms. Each functional form best describes the group delay for a different class of fibers. Comparisons of these various fitting equations can be found in the literature \[[@b11-j23mec], [@b12-j23mec]\]. 2.1 Determination of Group Delay -------------------------------- From [Eq. (1)](#fd1-j23mec){ref-type="disp-formula"} above, we can see that the uncertainty with which we can determine group delay is related to the uncertainty with which we know the modulation frequency. We measured the modulation frequency *f*~m~, which was derived from an oven stabilized crystal oscillator, to be 1920.007 MHz ± 0.002 MHz (For the rest of this paper, we will assume a coverage factor of *k* = 2 unless explicitly stated otherwise.) using a frequency counter which was calibrated in terms of NIST primary frequency standards. We measured the short term frequency stability (minutes) to be approximately a few times 10^−9^ *f*~m~. The crystal oscillator's specified aging rate is a few times 10^−7^ *f*~m~ per month. We observed the oscillator's output over a 7 month time span and noted shifts of no more than 3 times 10^−7^ *f*~m~. Both the short term and long term frequency behavior of the crystal oscillator are more than adequate for our uncertainty requirements. The short term frequency stability is the important parameter when measuring *λ*~0~, while the long term stability of *f*~m~ is relevant when determining *S*~0~. Absolute group delay does not need to be known to determine *λ*~0~; therefore the modulation frequency only needs to be stable for the duration of the measurement, which is a few minutes. To specify the dispersion slope *S*~0~, we need to measure absolute relative group delay. The vector voltmeter measures the phase angle between the RF reference signal from the crystal oscillator and the RF signal from the optical detector in the test arm. The vector voltmeter makes phase measurements which are largely independent of the input RF power level. However, the phase angle measured by the vector voltmeter still retains a weak dependence on the input power level, see [Fig. 3](#f3-j23mec){ref-type="fig"}. During a typical measurement the input power from the reference arm is held constant at level −7.6 dB below a reference level of 1 mW. The laser's output power is not perfectly constant across the scan however; the output power varying by approximately 1 dB across the 30 nm to 40 nm wavelength scan. The variation in received RF power from the measurement arm is therefore ≈ 1 dB. These levels of power fluctuation do not have a significant effect on the phase measurements. Two effects concerning the intensity modulation were viewed as potential sources of systematic error: the depth of modulation and drift in the modulator bias point. Both act to change the power distribution in the modulation sidebands. If either *λ*~0~ or *S*~0~ are affected by this, then there is a systematic error. The external modulator can have its bias point drift as a function of temperature; indeed, this is not uncommon with this type of integrated-optic modulator. This means that during a measurement, the modulator may not be centered on the most linear bias point, thereby introducing more power into higher order modulation sidebands. Our investigations indicated that neither the depth of modulation nor drift in the bias point caused an appreciable change in *λ*~0~ or *S*~0~. [Figure 4](#f4-j23mec){ref-type="fig"} shows a schematic diagram of the variable air gap used to ascertain the linearity and uncertainty of the vector voltmeter's phase response. A linear phase response is especially important for measurements of dispersion slope. The 1.9 GHz modulated optical signal is collimated in air by a lens. After traversing approximately 25 cm in air, the light was coupled back into a single-mode fiber by another collimating lens. The two collimating lenses had identical numerical apertures and are designed for optimal coupling efficiency at 1550 nm. Additionally, they are designed for low back reflection, with return losses of 45 dB. One of the collimators is mounted on a precise *X*-*Y*-*Z* translation stage. The other fiber collimator is mounted on a linear translation stage which is used to vary the length of the air gap. The stage is driven by a linear actuator with a 1 μm stepsize. To calibrate our system with respect to delay, we measure the change in the air gap distance. This is done by mounting a mirror on the back of the translation stage and measuring its position with a commercial interferometer. The details are beyond the scope of this paper, but the entire system is carefully aligned to minimize any errors due to cosine error or Abbe offset. The least count of the two frequency interferometer is less than 2 nm. The distance measured by this interferometer is traceable to a known helium-neon laser transition. The vector voltmeter was calibrated by changing the air gap over 80 mm in 2 mm steps and accurately recording its position and the corresponding RF phase measured by the vector voltmeter. In vacuum, the modulation frequency *f*~m~ = 1.920 007 GHz has a wavelength of 0.156 141 3 m. By changing the air gap by this amount, we would expect to see a phase change on the vector voltmeter of exactly 360°. If we assume the group index for air in our laboratory is 1.000 236, the modulation wavelength in air equals 0.156 104 5 m. We experimentally measured the change in the air gap needed to induce a phase shift of 360° and compared this value to the theoretically predicted wavelength to obtain a quantitative estimate of the system's linearity and accuracy. Phase angle (*ϕ*) as a function of air gap distance was measured many times and our values systematically differed from theory by 2 times 10^−4^*ϕ*. We attribute this difference in the phase angle to imperfections in the vector voltmeter itself. This discrepancy however, is unimportant for our purposes. The random spread in our measurements of *S*~0~ is approximately 4 times 10^−3^ *S*~0~, and so the nonlinearity in the vector voltmeter is not a limiting factor. Laser modal noise can also contribute to phase noise. If the laser hops between different longitudinal cavity modes, each mode will propagate at a different velocity through the fiber. If the laser operates in multiple modes simultaneously, then the group arrival time will be a composite of modal arrival times. The severity of the problem depends on the length of fiber, the amount of fiber dispersion, and the magnitude of the frequency hops. During a measurement, the system is most sensitive to modal noise when the laser is at the end of a scan, approximately 15 nm from *λ*~0~. Assuming a typical dispersion slope of 0.070 ps/(nm^2^·km), this yields a dispersion value of *D* = 10.5 ps/nm for a 10 km fiber. Since the system resolution is 0.15 ps, it would take a mode hop of 0.014 nm (1.75 Ghz at 1550 nm) to generate a large enough change in group delay to be resolvable. The tunable laser diode we are using at 1550 nm has an operating spectral width of less than 50 MHz; laser mode hops are therefore not a problem. Optical reflections in the measurement system will cause a systematic phase error. An optical reflection will arrive out of phase with respect to the signal and add to it vectorially. We intentionally introduced optical reflections to monitor the effect on dispersion measurements. Our experimental observations agreed well with theoretical predictions. If the return losses from connectors in the system are greater than 25 dB, there should be no significant error in the measurements. 2.2 Determination of Wavelength ------------------------------- A commercial wavemeter is used to determine the vacuum wavelength of the tunable source. The wavemeter has an internal vacuum chamber and can measure *λ* with an uncertainty of 1 part in 10^6^ when the chamber is evacuated. Without evacuation there is an error due to the dispersion of the refractive index of air. The wavemeter operates by comparing fringe counts from an internal reference laser against those produced by the unknown laser. To verify the accuracy of the wavemeter, we measured the known vacuum wavelength of a 1.523 49 μm He-Ne laser before and after every set of measurements. No systematic offset was observed between the theoretical vacuum wavelength of the He-Ne laser line and the value measured by the wavemeter \[[@b13-j23mec]\]. All of our reported values for *λ*~0~ refer to the wavelength that would be measured in a vacuum. 2.3 Errors Caused by Fitting of Group Delay Data ------------------------------------------------ We use the FOTP recommended fitting [Eqs. (4)](#fd4-j23mec){ref-type="disp-formula"} and [(5)](#fd5-j23mec){ref-type="disp-formula"} when fitting group delay data for dispersion-un-shifted and shifted fibers respectively. Given these fitting functions, we are interested in the dependence of *λ*~0~ on the exact manner in which the fits are implemented. Two effects were investigated: first, how *λ*~0~ changes as we change the width of the wavelength interval used in the fit; second, the sensitivity of *λ*~0~ to the centering of the group delay data about *λ*~0~. Ideally, *λ*~0~ should not depend upon either of these parameters. We investigated the dependence of *λ*~0~ on these parameters for different fiber samples: our evaluations of uncertainties are based on the largest or least favorable results observed. [Figure 5](#f5-j23mec){ref-type="fig"} shows the dependence of *λ*~0~ upon the group delay scan width. Once the wavelength range exceeds approximately 18 nm, *λ*~0~ appears to asymptotically approach a constant value. We believe smaller wavelength scans do not yield large enough changes in group delay to be immune from noise. However, fitting group delay over a very large spectral range to achieve a more global fit may well shift the value of *λ*~0~. How well a scan of group delay is centered about *λ*~0~ appears to have a weak effect upon the measured value of *λ*~0~. We have estimated this effect, by taking wide scans of group delay versus wavelength and then utilizing different 20 nm segments. [Figure 6](#f6-j23mec){ref-type="fig"} shows *λ*~0~ as the symmetry is varied. Our results indicate that the general shape of the dependence seems to be repeatable from fiber to fiber, but that the magnitude of the effect varies. During an actual measurement, we estimated that scans can be centered to within 1 nm to 2 nm of *λ*~0~; therefore, we conservatively estimate the uncertainty in *λ*~0~ due to the above mentioned effects to be ± 0.007 nm (*k* = 2). 2.4 Effect of Chirp on *λ*~0~ ----------------------------- Chirp is the instantaneous variation of optical frequency with time. Together with chromatic dispersion, chirp can effect how a pulse broadens and distorts as it propagates through an optical fiber \[[@b14-j23mec], [@b15-j23mec]\]. Depending upon whether a system is operating in the anomalous or normal dispersion region, a chirped pulse can be additionally broadened or compressed by dispersion. This makes it difficult to distinguish how much of the pulse distortion is due to chirp and how much is due to chromatic dispersion alone; ambiguous dispersion measurements can be the result. Most directly modulated semiconductor lasers have a chirped output \[[@b16-j23mec]\]. Changes in the injected carrier concentration result in changes in the index of refraction within the active region of the laser. The output light is shifted towards the blue or red depending upon whether the carrier concentration is temporarily above or below its equilibrium value. We avoid this problem by taking the CW output of the tunable laser and intensity modulating it with a Mach-Zehnder LiNbO~3~ integrated-optical modulator. The chirp induced by this type of modulator should essentially be zero if the propagation constants in the two arms of the modulator are equal \[[@b17-j23mec]\]. We experimentally verified that there was no influence from residual chirp by observing the dependence of *λ*~0~ upon the modulation depth. Residual chirp could result from a difference in the mode propagation constant in each of the two interferometer arms or by a slight path difference or asymmetry between the two arms. By experimentally changing the depth of modulation we would expect to increase or decrease the contribution of chirp. No effect was observed, and we therefore assume any residual chirp in the system is negligible. 2.5 Uncertainty analysis ------------------------ We present our uncertainty analysis in the ISO-recommended format \[[@b18-j23mec]\]. Uncertainties were categorized as Type A, those whose distribution could be based on statistical analysis of repeated observations, or Type B, those based on scientific judgment, whose magnitude and distribution could only be estimated. Together, these uncertainties in *λ*~0~ are presented in [Table 1](#t1-j23mec){ref-type="table"}. The type A uncertainties due to random noise and long term drift were statistically determined from fiber control charts. The expanded type A uncertainty (coverage factor *k* = 2) was estimated to be ± 0.035 nm \[[@b19-j23mec]\]. An additional type A uncertainty due to residual dispersion in the measurement system was added in quadrature. We estimated, assuming a normal distribution, the 2*σ* Gaussian widths for the type B uncertainties and added them in quadrature to the type A uncertainties. The uncertainty assigned to wavelength accounts for dispersion in the refractive index of air between the reference wavelength used in the wavemeter (633 nm) and the measurement wavelength (1550 nm region). The combined expanded uncertainty for measurements of *λ*~0~ was ± 0.060 nm, with a coverage factor of *k* = 2. 3. Differential Phase Shift System ================================== The second system used in our laboratory, shown in [Fig. 7](#f7-j23mec){ref-type="fig"}, is a variation of the differential phase shift technique \[[@b20-j23mec], [@b21-j23mec]\]. The CW output of a grating tuned laser diode or erbium-doped fiber laser (EDFL) is intensity modulated at 4 GHz. The wavelength is also dithered from 1 nm to 4 nm at a low frequency (100 Hz) by rotating a diffraction grating which is mounted on a galvanometric scanner. After passing through a low frequency intensity stabilizer, the modulated laser light is sent to the fiber under test. The optical signal is detected, filtered, and transmitted to the RF port of a quadrature mixer. The quadrature mixer operates as a phase detector and provides signals related to the phase difference between the RF and local oscillator (LO) ports. The in-phase and quadrature signals are divided to give an output proportional to tan *θ*, where *θ* is the phase angle between the RF and LO ports. By taking this ratio, the phase measurement is less sensitive to amplitude variations of the input signals. As the laser source is linearly scanned across the wavelength region of interest (typically a few nm window about *λ*~0~), the lock-in amplifier, locked to the 100 Hz dither frequency, gives an output proportional to the derivative with respect to *λ* of tan\[*θ* (*λ*)\]. Therefore, the output near *λ*~0~, with the proper setting of the variable phase offset, is directly proportional to the dispersion coefficient without the need for curve fitting. The entire measurement is completed in a few tens of seconds. This fast measurement time makes the system less sensitive to thermal drift of the fiber sample. This, and the elimination of curve fitting, are the principal advantages of the differential phase shift technique. At *λ*~0~, the detected output signal goes to zero; on either side of *λ*~0~, it undergoes a change in sign, as can be seen in [Fig. 8](#f8-j23mec){ref-type="fig"}. We average multiple runs over the wavelength region about *λ*~0~. Typically, *λ*~0~ can be determined with type A expanded uncertainties approaching 0 ± 0.1 nm (*k* = 2). To obtain accurate results it is necessary to use the linear response region of the quadrature mixer, to account for residual dispersion within the system, and to ensure adequate source intensity stabilization. A measurement of the laser wavelength is made with an interferometric wavemeter (see Sec. 2.2). In our implementation of the differential phase shift technique, accurate measurements of *λ*~0~ are more technically challenging than those of the phase shift technique described in Sec. 2. In particular, the laser intensity must remain stable over the scan range and during wavelength dither. With our semiconductor laser diode, external cavity modes resulted in mode competition and mode hopping, making this difficult, especially far from the diode gain peak. With fiber laser sources, we have encountered intensity fluctuations from etalon effects in the cavity and complications due to *Q*-switching as the grating is dithered. These noise sources were minimized with external stabilization, intensity-insensitive detection, and modifications to the laser sources themselves. Nevertheless, the remaining noise from these sources placed a lower limit on the type A uncertainties obtainable with this system. 4. Measurement System Based on a Four-Wave Mixing Technique =========================================================== We also measured *λ*~0~ using a nonlinear four-wave mixing (FWM) technique \[[@b3-j23mec]\]. Four-wave mixing in an optical fiber is a nonlinear parametric process, where the optical fiber acts as a passive media in which the multiphoton interaction occurs \[[@b22-j23mec], [@b23-j23mec]\]. Conservation of energy and momentum dictate that the nonlinear process occurs efficiently only near *λ*~0~ in single-mode fibers. We are interested in the "partially degenerate" case of four-wave mixing, where two of the photons have the same frequency. There are different regimes in which FWM can occur. With a strong pump laser, bound electrons are driven hard enough to elicit a nonlinear response. In this case, nonlinear effects contribute to the phase matching condition. In the case of a weak pump laser, nonlinear effects do not contribute appreciably to phase matching. The later case is the one which we utilize. In either case, the probe laser acts as a seed which stimulates generation of the FWM signal, see [Fig. 9](#f9-j23mec){ref-type="fig"}. Two pump photons combine to create two new photons, one at the probe wavelength and the other at the FWM wavelength. Energy conservation dictates that the FWM and probe signals appear symmetrically spaced about the pump wavelength. Momentum conservation dictates that the FWM process will be most efficient when the pump wavelength is at *λ*~0~. In order for momentum to be conserved, the mode propagation constants must be matched. If the phase matching condition is not precisely met, there will be a large reduction in the efficiency of the FWM process. Inoue and Toba have derived the following expression for the phase mismatch Δ*β* as a function of pump offset from *λ*~0~ \[[@b24-j23mec], [@b25-j23mec]\], $$\Delta\beta = \frac{2\pi\lambda^{4}}{c^{2}}S_{0}\left( {f_{1} - f_{0}} \right)\left( {f_{1} - f_{3}} \right)^{2},$$where *S*~0~ is the dispersion slope at *λ*~0~, *c* is the speed of light in vacuum, *f*~1~, *f*~0~, and *f*~3~ are the frequencies of the pump, zero-dispersion wavelength, and probe respectively. It is important to note that the optimum efficiency occurs when the pump wavelength is equal to *λ*~0~. The efficiency with which a FWM signal is generated has a functional form which has the approximate shape of a sinc function (Fourier transform of a rectangle) centered about *λ*~0~. [Figure 10](#f10-j23mec){ref-type="fig"} gives both theoretical and experimental FWM efficiency as a function of pump wavelength. In this case, the theoretical curve from Inoue \[[@b25-j23mec]\], and the experimental data both have a mid-scan pump-probe spacing of 7 nm and a dispersion slope of 0.070 ps/(nm^2^·km). The width of the FWM efficiency peak is related to a number of parameters including the separation between the probe wavelength and *λ*~0~ and the dispersion slope at *λ*~0~ \[[@b24-j23mec], [@b25-j23mec]\]. The width of the FWM efficiency curve places a practical limit on the type A uncertainty with which *λ*~0~ can be determined. By tuning the pump wavelength and observing the magnitude of the FWM signal, *λ*~0~ can be identified as the wavelength where the FWM signal is produced with maximum efficiency. A schematic of the experiment is illustrated in [Fig. 11](#f11-j23mec){ref-type="fig"}. The outputs of two tunable fiber lasers, one the pump and the other the probe, are passed through optical bandpass filters and polarization control paddles before being transmitted to the fiber under test. The pump laser has a CW power of +10 dB (in reference to 1 mW) after being amplified by an erbium-doped fiber amplifier (EDFA), whereas the probe laser has a CW power of −3 dB (in reference to 1 mW). The bandpass filters reduce the background amplified spontaneous emission (ASE) from the EDFA and fiber lasers. The FWM signal can be partially obscured by background ASE so it is desirable to filter the ASE to the maximum extent possible. To further reduce noise, the probe laser is mechanically chopped so the FWM signal can be synchronously detected by a lock-in amplifier at the chopping frequency. Polarization paddles are used to align the pump and probe polarization states, thereby maximizing the FWM signal. The procedure for determining *λ*~0~ is as follows. The probe laser wavelength is selected to be within approximately 10 nm of *λ*~0~ (approximate *a priori* knowledge of *λ*~0~ is necessary). With the probe wavelength held constant, the pump laser is scanned in wavelength, and the output of the test fiber is monitored on an optical spectrum analyzer. When the pump is within a few nanometers of *λ*~0~ a FWM signal is observed. There is a range of pump wavelengths that yield an observable FWM signal, but the FWM signal is maximized when the pump wavelength is equal to *λ*~0~. During a measurement we vary the probe wavelength and look for the optimum pump wavelength for each probe setting. The measured value for *λ*~0~ should be insensitive to the probe wavelength chosen. The type A uncertainty (*k* = 2) for measuring *λ*~0~ with this system is ± 0.5 nm. The FWM technique is sensitive to different segments of the fiber having different *λ*~0~ values. Each fiber segment will interact with the pump and probe differently to generate its own FWM signal. Therefore fibers in which *λ*~0~ varies as a function of length will have complex output spectrums, due to the superposition of different FWM signals. Meaningful information about a fiber's *λ*~0~ value can be difficult or impossible to extract from the resulting complex patterns \[[@b26-j23mec], [@b27-j23mec]\]. When the fiber sample has a nominally uniform value for *λ*~0~, however, then meaningful comparisons can be made with other measurement techniques. Fibers drawn from a single preform (a large glass cylinder from which the fiber is created) should exhibit the best uniformity. We measure the maximum FWM signal (i.e., *λ*~0~) for a range of different pump polarizations; in each case polarization paddles are used on the probe laser to optimize the field overlap. In fibers with low polarization mode dispersion, polarization does not affect the value obtained for *λ*~0~. However, in other fibers, an average over polarization is taken to avoid ambiguous results. 5. Comparisons Between Measurement Methods ========================================== The three measurement systems discussed in this paper have been compared. We consider our frequency-domain phase shift system to be the most accurate because of its stability, low type A uncertainty, and the extensive documentation compiled for it. The differential phase shift system is not designed to measure dispersion, but rather *λ*~0~ directly. It was difficult to account for the system's residual dispersion and other possible systematic errors (see Sec. 3). [Table 2](#t2-j23mec){ref-type="table"} presents comparisons between the frequency-domain and differential phase shift systems for long fibers. The average difference between the two systems is −0.07 nm, which is within the expected uncertainties of the two systems. [Table 3](#t3-j23mec){ref-type="table"} presents comparisons between the FWM and the frequency-domain phase shift system. The average discrepancy between the two systems is −0.16 nm. The differences are both positive and negative, with the largest differences occurring for the shorter fibers. A number of potential systematic errors were investigated, including the effect of *λ*~0~ nonuniformity and fiber birefringence. The FWM system is a fundamentally different method for measuring *λ*~0~ in the fibers. For this reason, the agreement lends support to the uncertainty claims of the frequency-domain system. 6. Fiber Properties Effecting *λ*~0~ ==================================== The effect of the environment is of great importance if meaningful comparisons between independent laboratories or measurement systems are to be made. Parameters whose effect on dispersion were investigated either experimentally or through the literature include: temperature, strain, and pressure. The long-term behavior of *λ*~0~ and dispersion slope were ascertained by repeated measurements over many months. Both fibers that remained in the laboratory and packaged fibers shipped around the country were periodically measured. The plotted results of these measurements were used to create "control" charts which help to monitor the long-term stability of the measurement system and the fiber samples. 6.1 Environmental Effects ------------------------- The largest environmental factor affecting *λ*~0~ is the temperature of the fiber. In dispersion-unshifted fiber samples, the temperature dependence of *λ*~0~ is + 0.025 nm/°C, while for dispersion-shifted fibers the dependence is + 0.030 nm/°C \[[@b28-j23mec], [@b29-j23mec]\]. There have been no reports of a temperature hysteresis effect on *λ*~0~, where the temperature was cycled over a large range (−60 °C to +250 °C) \[[@b28-j23mec]\]. We have verified the temperature dependence of *λ*~0~ experimentally by ramping the temperature control chamber between +10 °C and +35 °C and monitoring *λ*~0~. [Figure 12](#f12-j23mec){ref-type="fig"} shows the data and our experimentally determined temperature dependence of + 0.028 nm/°C for a dispersion-shifted fiber. The dispersion slope S~0~ has no significant temperature dependence. Before measuring fiber specimens we typically allow the fiber at least 24 hours to come to thermal equilibrium in the temperature controlled chamber. The thermal dependence of group delay is approximately 180 ps/(°C ·km) \[[@b30-j23mec]\]. This strong temperature dependence can lead to problems since group delay is used to measure dispersion. Changes in group delay due to temperature fluctuations can exceed the change caused by dispersion. Fortunately, thermal time constants are typically long, and if measurements of group delay are made quickly, this problem can be minimized. However, the extreme sensitivity of our system to changes in group delay makes small temperature changes significant. Data collected during a measurement may be skewed if the group delay is drifting with time (due to temperature changes). Indeed, for a 10 km fiber sample, the temperature change required to change the group delay by 0.15 ps (the resolution of our system) is only 0.1 mK. Problems caused by thermal drift are reduced in a number of ways. First, the fiber sample is allowed to come to thermal equilibrium in a temperature controlled chamber. Second, the measurements are performed quickly, which does not allow for large temperature excursions in the sample or room. Finally, we take group delay data in two directions, first with increasing wavelength, then with decreasing wavelength. These averaged data pairs reduce errors caused by a linear drift in group delay. We also measured *λ*~0~ in test fibers as a function of winding tension. This could be an issue, since with time the fiber may relax on the spool. We wound both dispersion-shifted and unshifted fibers at tensions varying from 0.2 N to 1.0 N (the equivalent gravitational force exerted by 20 g to 100 g masses). We observed no effect on *λ*~0~ within our measurement repeatibility. The effect of longitudinal strain upon *λ*~0~ has also been investigated by others \[[@b31-j23mec]\]. Their findings for strain dependence, δ*λ*~0~/δ*S* = 0.015 nm/N, are consistent with our observations. Pressure also has a small effect on *λ*~0~. Typically, this is of significance only in transoceanic undersea systems. The pressure dependence of *λ*~0~ is 0.0076 nm/MPa and can be observed in simulated transoceanic environments \[[@b31-j23mec]\]. At oceanic depths of 8 km, extreme pressures of 82.7 MPa (12 000 psi) can be encountered. However, pressure does not have a significant effect upon SRM performance. 6.2 Control Chart and Longterm Stability of Fiber Samples --------------------------------------------------------- To determine how *λ*~0~ and the dispersion slope S~0~ are affected by fiber aging and/or relaxation, we performed repeated measurements over an 8 month time span. [Figure 13](#f13-j23mec){ref-type="fig"} shows measured values for *λ*~0~ on a 12 km dispersion-shifted "control" fiber. We did not observe any statistically significant variations in *λ*~0~ as a function of time. Also, to determine fiber robustness, we kept control charts on fibers shipped to various laboratories for measurement intercomparisons. These specimens were exposed to vibration and large temperature fluctuations during shipping; statistically significant fluctuations in *λ*~0~ were not observed. The stability of the control fibers helps ensure the measurement system is operating correctly and no sudden systematic errors have appeared. 6.3 Polarization Mode Dispersion -------------------------------- Polarization mode dispersion (*PMD*) in single-mode fibers is caused by the breakdown of the polarization-state degeneracy in the fundamental guided mode \[[@b32-j23mec]\]. Internal and external stresses on the fiber core and a lack of circular symmetry all lead to birefringence in the fiber and therefore to different mode propagation times for different polarizations transmitted through the fiber. The input polarizations (they need not be linear) that propagate the fastest and slowest through the fiber are known as the fast and slow principle states \[[@b33-j23mec]\]. The difference in propagation time between the fast and slow principal states is known as differential group delay (*DGD*). *PMD* is the expected value of *DGD* within a given wavelength interval. Any other input polarization will result in a mixture between principle states with a mean arrival time somewhere in between. Since *PMD* is dependent upon the birefringence and mode-coupling properties of the optical fiber, *PMD* is time variant and best described as a statistical process. *PMD* has the unit ps/km^1/2^, where the square root dependence on length originates from the random mode coupling present in a weakly birefringent fiber (nonpolarization maintaining fiber) \[[@b34-j23mec]\]. We have measured *PMD* values between 150 fs/√10km^1/2^ and $450\ \text{fs/}\sqrt{10\ \text{km}^{1/2}}$ for several packaged fiber samples. These measurements were performed with a narrow linewidth tunable laser and a commercial polarimeter using the Jones matrix eigenanalysis method \[[@b35-j23mec]\]. The *PMD* magnitude seemed to be stable over several hours due to the stable fiber packaging and the small amount of mode-coupling. When the initially selected state of polarization was varied, using the phase shift method of Sec. 2, we could observe differences in the mode propagation time for the two principal states. The presence of first-order PMD does not directly effect dispersion measurements; however, the wavelength dependence of *PMD* or so called "second-order" *PMD* does. Second-order *PMD* is simply the wavelength derivative of first-order *PMD* \[[@b36-j23mec], [@b37-j23mec], [@b38-j23mec]\]. It has a net dispersive effect that is identical in form to chromatic dispersion. Total dispersion can be written $$D_{\text{total}}{(\omega_{0})}_{\pm} = D_{\text{chromatic}}(\omega_{0}) \pm \frac{\Delta\tau^{\prime}(\omega_{0})}{2},$$where Δτ*t*′ is the instantaneous wavelength derivative of first-order *PMD*, $$\Delta\tau^{\prime} = \frac{d\Delta\tau}{d\lambda}.$$ It is important to note that as the wavelength interval over which *PMD* is averaged goes to zero, *PMD* becomes equivalent to *DGD*. In the vicinity of *λ*~0~, when chromatic dispersion approaches zero, the contribution of second-order *PMD*, can be significant and thereby cause a shift in *λ*~0~. To verify the theory expressed by [Eq. (7)](#fd7-j23mec){ref-type="disp-formula"}, we measured a high *PMD* fiber produced by winding 10 km of typical dispersion-shifted fiber onto a 1.9 cm radius spool. Winding the fiber onto a small spool induced a significant amount of bend birefringence. Additionally, the fiber winding induced mode-coupling so that the *DGD* was a function of wavelength. [Figure 14](#f14-j23mec){ref-type="fig"} shows three measurements of the fiber's DGD as a function of wavelength. The measurements were taken over a 2 day period. Second-order *PMD* was measured at five wavelengths, designated by the letters A through E: see [Fig. 14](#f14-j23mec){ref-type="fig"}. At these wavelengths, chromatic dispersion was measured along the principal states. Polarization control paddles were used to find the principal states and then the change in group delay over a 2 nm wavelength interval was measured for each principle state. In this manner the change in chromatic dispersion between the two principal states could be determined. According to theory, the observed difference in dispersion along the two principal states should be equal to that predicted by [Eq. (7)](#fd7-j23mec){ref-type="disp-formula"}. Comparisons between the experimentally observed changes in dispersion and those predicted by theory are presented in [Table 4](#t4-j23mec){ref-type="table"}. The correlation between direct observations of second-order *PMD* is fairly good. One of the difficulties of the experiment is finding the principal states accurately with the polarization paddles. In our packaged fiber samples, the wavelength dependence of *DGD* is quite weak; as shown in [Fig. 15](#f15-j23mec){ref-type="fig"}. We measured DGD in three 10 km fiber samples over a period of days. The largest amount of second-order *PMD* we observed was 0.011 ps/nm. If we use this value in conjunction with the typical fiber slope value (for dispersion-shifted fiber) of 0.070 ps/(nm^2^ · km), a possible shift in *λ*~0~ of 0.018 nm occurs between the two principal states. Since we will be producing many fiber SRM samples with potentially larger *DGD* values, a conservative estimate for this type B expanded uncertainty of ± 0.035 nm is adopted (coverage factor of *k* = 2). 7. Discussion ============= We have developed a frequency-domain phase shift system capable of measuring *λ*~0~, in 10 km packaged SRM fibers, with an expanded uncertainty of 0.060 nm. Most of this uncertainty originates in the measurement system itself, but we have estimated an expanded type B uncertainty in the fiber samples of ± 0.035 nm due to "second-order" polarization mode dispersion. In the near future, SRM fibers with characterized *λ*~0~ and *S*~0~ values will be commercially available from NIST \[[@b39-j23mec]\]. Comparisons with two other measurement systems have yielded reasonable agreement. The authors would like to acknowledge the many contributions of their NIST colleagues. We would especially like to thank Paul Hale for his help with high frequency electronics and frequency measurements, Paul Williams for many helpful discussions about polarization mode dispersion, Matt Young for his general optics wisdom and advice on uncertainty analysis, and Jack Wang for his help with statistical methods. **About the authors:** Steven Mechels is a Physicist with the Optoelectronics Division of the Electronics and Electrical Engineering Laboratory in Boulder, Colorado and has received the U.S. Department of Commerce Gold Medal for work on fiber optic geometry measurements. John Schlager is also a Physicist with the Optoelectronics Division. Earlier, he was a guest researcher with the NTT laboratories in Japan. Douglas Franzen is a Group Leader in the Optoelectronics Division and has published many papers in the field of fiber optics. He has received the U.S. Department of Commerce Gold Medal for work promoting fiber optic standardization. The National Institute of Standards and Technology is an agency of the Technology Administration, U.S. Department of Commerce. {#f1-j23mec} {#f2-j23mec} {#f3-j23mec} {#f4-j23mec} {#f5-j23mec} {#f6-j23mec} {#f7-j23mec} {#f8-j23mec} {#f9-j23mec} {#f10-j23mec} {#f11-j23mec} {#f12-j23mec} {#f13-j23mec} {#f14-j23mec} {#f15-j23mec} ###### Measurement uncertainties in *λ*~0~ Expanded uncertainty (2*σ*) (nm) ---------------------------------------------- ---------------------------------- Type A uncertainties Random noise (including long term stability) 0.035 Correction for residual system dispersion 0.007 Type B uncertainties 2nd-order PMD 0.040 Curve fitting 0.007 Wavelength 0.007 Chirp negligible Short-term frequency stability negligible Temperature 0.025 Expanded uncertainty, 2*σ* 0.060 ###### Comparison of the differential phase shift and the frequency-domain phase shift techniques. Δ*λ*~0~ = *λ*~0\ differential~−*λ*~0\ Freq.-Domain~ Fiber *λ*~0\ Differential~ (nm) *λ*~0\ Freq.-Domain~ (nm) Δ*λ*~0~ (nm) -------------- --------------------------- --------------------------- -------------- C2 (10 km) 1548.7 ± 0.1 1548.86 −0.16 J (10 km) 1549.2 ± 0.1 1549.21 −0.01 F (12 km) 1552.7 ± 0.1 1552.70   0.00 J C2 (20 km) 1549.0 ± 0.1 1549.09 −0.09 ###### Comparison of the four wave mixing and frequency-domain phase shift techniques. Δ*λ*~0~ = *λ*~0\ FWM~−*λ*~0\ Freq.-Domain~ Fiber *λ*~0\ Differential~ (nm) *λ*~0\ Freq.-Domain~ (nm) Δ*λ*~0~ (nm) ------------------ --------------------------- --------------------------- -------------- C2 (20 km) 1548.94 1549.17 −0.23 C2-10A (10 km) 1548.96 1548.86 +0.10 C2-10B (10 km) 1549.14 1549.44 −0.30 C2-5A (5 km) 1549.12 1549.04 +0.08 C2-5B (5 km) 1549.57 1549.87 −0.30 C2-2.5A (2.5 km) 1549.12 1548.83 +0.29 C2-2.5B (2.5 km) 1549.14 1549.22 −0.08 C2-2.5C (2.5 km) 1549.33 1549.56 −0.23 C2-2.5D (2.5 km) 1549.44 1549.94 −0.50 C2-1A (1.25 km) 1549.37 1549.82 −0.45 ###### A comparison between experiment and theory for the difference in dispersion along the two principal states of a high *PMD* fiber. Δ*D* = *D*~fast-axis~−*D*~slow-axis~ Point Wavelength (nm) Δ*D*, Theory (ps/nm) Δ*D*, Experiment (ps/nm) ------- ----------------- ---------------------- -------------------------- A 1504 0.104 0.100 B 1527 0.300 0.248 C 1541 0 0.054 D 1549 0.270 0.278 E 1557 0 0.083

Thermoelectric conversion is a solid-state and environmentally friendly energy conversion technology with broad applications including solid-state cooling, energy harvesting, and waste heat recovery[@b1]. Flexible thermoelectric devices are especially attractive for waste heat recovery along contoured surfaces and for energy harvesting applications to power sensors, biomedical devices, and wearable electronics -- an area under exponential growth[@b2]. The efficiency of thermoelectric materials is determined by the figure of merit ZT defined as ZT = α^2^σ*T*/κ, where α, σ, κ and *T* are the Seebeck coefficient, electrical conductivity, thermal conductivity, and absolute temperature respectively[@b3][@b4]. Nanostructured thermoelectric materials have been widely studied in recent years and proven to have unique and superior thermoelectric performance compared to their bulk counterparts due to the ability to tailor electron and phonon transport and effectively increase ZT[@b5][@b6][@b7]. Despite significant ZT improvements in nanostructured materials[@b8][@b9][@b10], the lack of scalable and low-cost manufacturing processes remains a major challenge to the wide use of these materials[@b11]. In addition, major progress in ZT enhancement through nanostructuring has historically been achieved in mechanically rigid materials, while flexible thermoelectric materials are still relatively unexplored and have fairly low ZT[@b2]. Among all the methods to fabricate thermoelectric materials, wet deposition of nanocrystal-based colloidal inks using screen printing, inkjet printing, direct writing, or other layer-by-layer methods hold many advantages due to the ability to directly convert nanocrystal inks into micro/macroscale functional materials and devices with great scalability, flexibility, and cost effectiveness[@b12][@b13]. Using inkjet or disperser printing, several research groups have achieved ZT of \~0.3 in thermoelectric films printed on flexible substrates[@b14][@b15]. Screen printing has also been explored as a more efficient way to fabricate thermoelectric devices[@b16][@b17][@b18][@b19]. Despite the above proof-of-concept demonstrations, flexible thermoelectric films fabricated by printing methods continue to exhibit fairly low ZT in the 0.1--0.3 range, significantly lower than their rigid bulk counterparts fabricated using conventional approaches such as hot press or spark plasma sintering[@b2]. There are many challenges in printing efficient and flexible thermoelectric materials using nanocrystals, including scalable synthesis of high-performance nanocrystals, nanocrystal surface oxidation during printing processes, and poor density and electrical conductivity of the printed films[@b20]. Here, we report a study of flexible thermoelectric films by screen printing colloidal inks composed of bismuth telluride based nanoplates fabricated using a scalable microwave-stimulated wet chemical approach[@b21] (shown in [Fig. 1](#f1){ref-type="fig"}). The peak ZT of our flexible films reaches 0.43 at 175 °C due to a combination of high power factor and low thermal conductivity, which is among the highest ZT reported for flexible thermoelectric materials fabricated by printing. The films demonstrate superior flexibility with negligible changes in electric resistance with 150 bending cycles. In addition to the unprecedented high ZT and flexibility, another significant advantage of our work is the use of thioglycolic acid (TGA) as a surface capping agent to inhibit nanocrystal oxidation[@b21], thus enabling large-scale manufacturing at ambient conditions. Methods ======= Nanocrystal ink synthesis ------------------------- Our doped and functionalized pnictogen chalcogenide nanocrystals of Bi~2~Te~2.8~Se~0.2~ were synthesized using a microwave stimulated wet-chemical synthesis method based on inexpensive organic solvents and metal salts described earlier[@b21]. In this method, the reaction between molecularly ligated chalcogen and pnictogen complexes was activated by microwave stimulation with the presence of thioglycolic acid (TGA), which serves as a shape-directing, oxide-inhibiting and sulfur-dopant delivery agent. The resulting precipitate is cleaned and dried in ambient conditions to obtain powders consisting of single-crystal nanoplates of 5- to 20-nm-thickness with bounding edge dimensions ranging between 50 to 500 nm. The dried nanocrystals with the TGA capping agent were mixed with solvent and binder to produce viscous and thixotropic inks for screen printing. The optimized ink contains 58 wt.% Bi~2~Se~2.8~Se~0.2~ nanopowders, 39 wt.% Solvent (α-Terpineol, from Sigma-Aldrich), 2 wt.% Binder (Disperbyk-110, from BYK U.S.A. Inc.), 1 wt.% Glass Frits (from Artglass Supplies, 325 mesh). The ink is thoroughly mixed using a planetary centrifugal mixer for 20 minutes followed by a vortex mixer for 10 minutes to get a uniform mixture. Flexible film fabrication ------------------------- As-prepared ink is screen printed on flexible polyimide substrates. The thermoelectric films of various thicknesses in the range of 10--100 μm were obtained by controlling the screen mesh size and the number of repeated print passes. The printed films were first dried in air at 200 °C on a hot plate to remove the solvent and binder, followed by a cold compaction using a hydraulic press to consolidate the films. The film was finally sintered at 430 °C for 45 minutes in vacuum in order to remove the TGA surfactant and further improve the film density. The sintering temperatures are kept below the melting point of the polyimide substrate, though better thermoelectric properties could be obtained at higher sintering temperatures. Thermoelectric property measurement ----------------------------------- The temperature-dependent in-plane electrical conductivity and Seebeck coefficient of the film sample were measured simultaneously using a commercial Linseis Seebeck and resistivity instrument. The above two properties of the same sample were also measured using a home-built testing system, and the two sets of measurement results are within 2%. In order to measure the thermal conductivity of the sample, a freestanding film of about 100 μm thickness was prepared under the same conditions as those for preparing thinner films on substrate. The temperature-dependent cross-plane thermal diffusivity of the freestanding film was measured using a laser flash instrument. The cross-plane thermal conductivity was then determined using the sample density measured using Archimedes method and the specific heat measured using a DSC instrument. The in-plane thermal conductivity of the freestanding film was measured directly using a steady-state method in vacuum, which is within 5% of the cross-plane thermal conductivity, indicating the sample is isotropic. Details about the thermal conductivity measurement are included in the [supplementary information](#S1){ref-type="supplementary-material"}. The carrier concentration and mobility were measured using Hall measurement conducted on the Physical Property Measurement System (PPMS) with 4 wire connection and the magnetic field sweeping from −1 T to 1 T. Thermoelectric device fabrication and testing --------------------------------------------- Five 10 mm × 2 mm × 0.01 mm n-type Bi~2~Te~2.8~Se~0.2~ elements were printed onto a flexible polyimide substrate with 4 mm spacing. Thin copper foils were soldered to the five elements in order to connect them electrically in series. A custom test bed was built using two commercial Peltier modules, one operating as a heater and the other as a cooler. The hot side and cold side of the TE device were thermally grounded to the two Peltier modules to create a temperature gradient. Two 75 *μ*m diameter k-type thermocouples were mounted to the hot and cold sides of the device to measure the temperature difference. The device was connected electrically in series with a shunt resistor and a variable resistor for impedance matching at each measurement temperature. The open circuit voltage, load voltage, current, internal resistance, and power output from the device were measured at each hot-side temperature while the cold-side temperature is maintained constant. Results and discussion ====================== [Figure 2(a)](#f2){ref-type="fig"} shows scanning electron microscope (SEM) images of the Bi~2~Te~2.8~Se~0.2~ nanocrystals, indicating the plate-like structures of several tens of nanometers thickness. [Figure 2(b)](#f2){ref-type="fig"} shows a cross-section SEM image of a flexible Bi~2~Te~2.8~Se~0.2~ film of about 10 μm thickness fabricated by screen printing. The films have about 85% relative density, and contain nanoscale pores primarily due to incomplete sintering of the nanocrystals and the removal of the additives in the ink. Temperature-dependent thermoelectric properties were obtained on a flexible film of 10 μm thickness printed using the nanocrystal ink and a reference pellet sample of 500 μm thickness made by the pure nanocrystal powders using cold compaction and sintering under the same conditions. The relative densities of the film and the pellet are 85% and 90% respectively. As shown in [Fig. 3(a)](#f3){ref-type="fig"}, the room-temperature electrical conductivity of the film is about 53% lower than the pellet. [Figure 3(b)](#f3){ref-type="fig"} shows the Seebeck coefficients of the two samples are within 10% for the entire measurement temperature, indicating approximately the same carrier concentrations for both the film and the pellet. Indeed, the Hall measurement validated that the carrier concentrations of the film and the pellet are within 10% (1.56 × 10^19^ cm^−3^ versus 1.42 × 10^19^  cm^−3^), whereas the film mobility is about 56% lower than the pellet mobility (127 cm^2^V^−1^s^−1^ versus 290 cm^2^V^−1^s^−1^) due to increased electron scattering by impurities and porosity present in the printed films. The room-temperature lattice thermal conductivity κ~L~ of the film and the pellet is estimated to be 0.41 Wm^−1^K^−1^ and 0.66 Wm^−1^K^−1^, respectively, using the equation κ~L~ = κ − σTL, where κ is the total thermal conductivity, σ is the electrical conductivity, T is the absolute temperature, and L is the Lorenz number determined from our previous work[@b22]. The κ~L~ of these samples is significantly lower than their bulk counterpart attributed to the nanoscale grains and porosities originated from the nanocrystals[@b22]. Furthermore, the κ~L~ of the film is lower than the pellet mainly due to additional phonon defects scattering caused by the addition of glass particles as well as a small contribution from slightly higher porosities. As shown in [Fig. 3d](#f3){ref-type="fig"}, the film demonstrates a peak ZT of 0.43 at 175 °C, which is only 20% lower than control pellet despite 53% lower electrical conductivity. The significantly reduced thermal conductivity largely compensates the electrical conductivity losses and contributes to the high ZT in the printed films. In comparison, [Table 1](#t1){ref-type="table"} summarizes the peak ZT (or room-temperature ZT if peak ZT is not available) of several representative n-type flexible thermoelectric materials, summarizing the highest reported ZT thus far in each category. The peak ZT of our flexible film is significantly higher than the previously reported bismuth telluride materials fabricated by printing, and is also among the highest reported value in all the reported n-type flexible thermoelectric films. In order to test the flexibility, the room-temperature electrical resistance of the printed films was tested using Van der Pauw method as a function of bending cycles on two cylinders of 7 mm radius and 5 mm radius respectively. Electrical resistance is chosen here to evaluate film flexibility because it is very sensitive to any cracks that may develop during bending test. After 150 bending cycles, the electrical resistances of the film show 1.4% increase for the 7 mm bending radius and 4.5% increase for the 5 mm bending radius respectively, indicating superior bending flexibility (shown in [Fig. 4](#f4){ref-type="fig"}). A thermoelectric generator device consisting of 5 n-type elements (shown in [Fig. 5(a)](#f5){ref-type="fig"} inset) was fabricated in order to validate the performance of the flexible films. [Figure 5](#f5){ref-type="fig"} shows the experimental and simulation results of the device tested at different temperature differences (ΔT) when the hot side temperature was varying from 40--80 °C and the cold side was maintained at 20 °C. The open circuit voltage, device voltage during operation, and power output increase as the ΔT increases. As shown in [Fig. 5(a,c)](#f5){ref-type="fig"}, the maximum open circuit voltage and power density reach 41 mV and 4.1 mW/cm^2^ with 60 °C ΔT. The power density was determined based on the total cross-sectional area (2 mm × 0.01 mm × 5) of the five thermoelectric elements. The experimental results are within 10% of the finite element simulation results based on the thermoelectric properties shown in [Fig. 3](#f3){ref-type="fig"}, which further verified the measured film properties. [Figure 5(b,d)](#f5){ref-type="fig"} shows the device operating voltage and power output as a function of electrical current tested by varying the external load resistances. The maximum power of 6.1 μW is obtained with 60 °C ΔT when the external load resistance matches with the internal resistance of the device. Conclusions =========== Flexible thermoelectric films were screen printed at ambient conditions using nanocrystals synthesized by a highly scalable microwave assisted wet chemical method. The films show an unprecedented peak ZT of 0.43 at 175 °C and superior flexibility with negligible changes of electrical conductivity after 150 bending cycles. The flexible thermoelectric device fabricated using the printed n-type thermoelectric elements produces a high power density of 4.1 mW/cm^2^ with a small temperature difference of 60 °C, opening up lots of applications for low-temperature energy harvesting. The performance of the printed thermoelectric films and devices can be further improved by increasing the electrical conductivity through optimization of the ink formulation and refinement of the sintering process. Additional Information ====================== **How to cite this article**: Varghese, T. *et al*. High-performance and flexible thermoelectric films by screen printing solution-processed nanoplate crystals. *Sci. Rep.* **6**, 33135; doi: 10.1038/srep33135 (2016). Supplementary Material {#S1} ====================== ###### Supplementary Information This work is funded by the US Department of Energy, Office of Nuclear Energy, under Award number DE-NE0008255. C.H. acknowledges financial support from US NSF fellowship support. N.K. acknowledges financial support from US DOE NEUP fellowship support. The authors thank Luke Schoensee for graphical support on this work. **Author Contributions** T.V. fabricated the flexible films. C.H. and N.K. carried out thermoelectric property measurements. J.R. fabricated and tested the thermoelectric device. P.G. carried out simulation of the TEG device. C.H. performed SEM characterizations. D.E. and R.J.M. contributed to the ideas and discussions of this work. Y.Z. is the principle investigator who devised and supervised this work. All the authors contributed to the writings of the manuscript. {#f1} {#f2} {#f3} {#f4} {#f5} ###### The thermoelectric performance comparison between our work and previous reported n-type flexible thermoelectric films. Materials details Power factor (mWm^−1^K^−2^) Peak/room T^\*^ ZT Ref. Fabrication methods --------------------- ----------------------------- -------------------- -------- ------------------------------- Bi~2~Te~2.8~Se~0.2~ 0.56 0.43 (Ours) Screen printing Bi~2~Te~3~ 1.33 0.35^\*^ [@b16] Screen printing Bi~2~Te~3~ + Epoxy 0.28 0.31^\*^ [@b15] Dispenser printing CNT 0.15 N.A. [@b23] Drop casting WS~2~ 0.007 N.A. [@b24] Vacuum filtration TiS~2~-Polymer 0.45 0.28 [@b25] Electrochemical intercalation CNT-PEDOT-TDAE 1.05 \~0.5^\*^ [@b26] Spraying and spin coating

Introduction {#s1} ============ The discovery of calcitonin (CT), a hormone that is released in hypercalcemia and lowers the serum calcium, was first made by Copp *et al*. as a result of perfusing isolated thyroid-parathyroid gland preparations in the anesthetized dog \[[@SFV011C1], [@SFV011C2]\]. Initially, the source of CT was mistakenly thought to be the parathyroid gland, but a thyroid origin was subsequently established \[[@SFV011C3], [@SFV011C4]\]. Pearse showed that the origin of CT was the C-cells of the thyroid gland \[[@SFV011C5]\]. Potts *et al*. determined the amino acid sequence of human and salmon CT, which led to the synthesis and commercial use of the more potent salmon CT \[[@SFV011C6], [@SFV011C7]\]. The hormone was first named thyrocalcitonin, but subsequently has been called CT. Since its discovery more than 50 years ago, little progress has been made in understanding its pathophysiologic role in humans, in part because a deficiency or excess of CT does not result in abnormalities except for diarrhea in a few patients with medullary thyroid carcinoma. In this review, we will discuss the role of CT in maintaining the serum calcium and regulating 1,25 (OH)2 vitamin D (1,25D) production in pregnancy and lactation and also consider potential bone effects and links to the gastrointestinal hormone gastrin. When appropriate a comparison between CT secretion and function and that of parathyroid hormone (PTH), the major hormone protecting against hypocalcemia, will be highlighted. CT is a 32 amino acid hormone secreted by the C-cells of the thyroid gland. In species in which the structure of CT has been determined, common features include a 1--7 amino terminal disulfide bridge with cysteine at positions 1 and 7 and proline at the carboxy-terminal \[[@SFV011C6]\]. Divergence is seen in the interior 10--27 amino acids. Non-mammalian CTs such as salmon have the greatest potency. Salmon CT, which differs from human CT in 16 amino acids, has been used for treatment of hypercalcemia. CT is primarily metabolized by the kidney \[[@SFV011C8]--[@SFV011C10]\]. The gene for CT is on the short arm of chromosome 11 and encodes CT and calcitonin gene-related peptide (CGRP). CT is present in large amounts only in the C-cells of the thyroid while CGRP, a potent vasodilator, is present not only in the thyroid, but also in central and peripheral nervous tissue. CT is present in ocean fish which live in a high calcium environment with the need to expel calcium. CT is thus older than parathyroid hormone (PTH) which was first recognized in early land-dwelling animals when conservation rather than expulsion of calcium became important. Besides modifying increases in serum calcium and increasing 1,25D production, several potential roles for CT have been suggested and important observations about CT have been made. In azotemic and non-azotemic animal models, CT has been shown to decrease the magnitude of hypercalcemia during calcium loading \[[@SFV011C1]--[@SFV011C3], [@SFV011C11]\]. Studies in the 1980s showed that CT increased renal production of 1,25D \[[@SFV011C12], [@SFV011C13]\]. In contrast to other stimuli of 1,25D production such as PTH and hypophosphatemia in which 1,25D production occurs in the convoluted proximal tubule, stimulation of 1,25D by CT occurs in the straight proximal tubule \[[@SFV011C12]\]. Similar to the reciprocal relationships between both PTH and 1,25D and FGF23 and 1,25D, CT stimulates 1,25D secretion while 1,25D suppresses CT secretion \[[@SFV011C14], [@SFV011C15]\]. But during pregnancy and lactation, both 1,25D and CT levels are increased \[[@SFV011C16]--[@SFV011C18]\]. The effectiveness of CT in the treatment of hypercalcemia is attributed to reducing osteoclast activity \[[@SFV011C19]\]. But there is also a subsequent escape from the calcium-lowering effect of CT \[[@SFV011C20]\]. Besides decreasing osteoclast activity, CT has been suggested to facilitate deposition of calcium and phosphorus in bone especially in the post-prandial state \[[@SFV011C21]\]. Whether CT is stimulated by the ingestion of food through gastrin stimulation and consequently affects bone has been a subject of interest \[[@SFV011C21], [@SFV011C22]\]. However, no differences in bone mass has been shown when CT is absent in congenital hypothyroidism provided thyroid hormone is replaced \[[@SFV011C23], [@SFV011C24]\]. Gender and age differences in CT have been shown with women having lower values than men and with decreasing CT values with age in some, but not all studies \[[@SFV011C9], [@SFV011C25]\]. CT screening has been shown to be a useful tool for the diagnosis of medullary thyroid carcinoma \[[@SFV011C26]\]. However, despite several attractive and plausible hypotheses, CT has not been shown to have an important physiologic role in humans more than 50 years after its discovery. Calcitonin and acute hypercalcemia {#s2} ================================== Studies during the past 50 years have shown that CT modifies the development of hypercalcemia. In 1960, a calcium infusion in thyroparathyroidectomized dogs was shown to result in a greater magnitude of hypercalcemia and a longer time to return to baseline values than in normal dogs \[[@SFV011C27]\]. Subsequently, Hirsch *et al.* showed that the calcemic response to parathyroid extract was greater in thyroparathyroidectomized rats than in parathyroidectomized or sham-operated rats demonstrating the thyroid origin of CT \[[@SFV011C3]\]. In rats with hypercalcemia from NH~4~Cl-induced metabolic acidosis, CT administration decreased the serum calcium concentration \[[@SFV011C28]\]. As shown in Figure [1](#SFV011F1){ref-type="fig"}, the absence of CT in both azotemic and non-azotemic rats enhanced the calcemic response to PTH during a 24 and 48 h infusion of PTH \[[@SFV011C11]\]. In the CT receptor knockout mouse, 1,25D-induced hypercalcemia was greater than in the control mouse \[[@SFV011C29]\]. In a mouse study consisting of PTH knockout, PTH and calcium-sensing receptor (CaSR) double knockouts, and wildtype mice, it was shown that both CaSR-mediated CT secretion and enhanced renal calcium excretion were important for preventing the development of hypercalcemia while inhibition of PTH secretion was not required for a robust defense against hypercalcemia \[[@SFV011C30]\]. In summary, strong evidence exists that CT is an important modifier of the hypercalcemic effect of acute calcium loading. Fig. 1.The effect of a PTH infusion on the serum calcium concentration was greater in the absence of calcitonin (CT) in rats both with (**A**) chronic renal failure (CRF) and (**B**) normal renal function (NRF). Solid circle is (CT−)CRF; open circle is (CT+)CRF; solid square is (CT−)NRF; and open square is (CT+)NRF \[[@SFV011C11]\]. Reprinted with permission from Kidney International. Similar to PTH, CT secretion is modified by the CaSR. But while activation of the CaSR suppresses PTH secretion, CT secretion is stimulated. The mechanisms for the opposite effects downstream from the CaSR are still poorly understood. As with PTH, CT gene transcription is suppressed by 1,25D \[[@SFV011C14]\]. Both the CT and PTH receptors belong to the class II subclass of G-protein-coupled receptors \[[@SFV011C31]\]. The CT response to hypercalcemia is a sigmoidal curve opposite in direction to the sigmoidal curve of the PTH response to hypocalcemia \[[@SFV011C32]--[@SFV011C34]\]. In one animal study, a rapid induction of hypercalcemia resulted in a greater CT response than a slow induction of hypercalcemia of similar magnitude \[[@SFV011C35]\]. Also, as PTH secretion is suppressed by the induction of hypercalcemia, CT secretion in both control and CKD patients is suppressed by the induction of hypocalcemia \[[@SFV011C34]\]. In a study of normal and CKD subjects, a maximal CT response was seen after an increase in ionized calcium of 0.4 mM \[[@SFV011C34]\]. Many more studies of the PTH response to hypocalcemia have been performed than of the CT response to hypercalcemia. Interestingly, while every normal and azotemic subject in studies of PTH secretion has shown a sigmoidal response to hypocalcemia, studies in normal and azotemic humans and animals have shown that some patients and animals fail to increase CT secretion in response to hypercalcemia induced by a calcium infusion \[[@SFV011C34], [@SFV011C36]\]. In a study in cats, a strong correlation was found between the number of CT-positive cells in the thyroid gland and the plasma CT concentration induced by hypercalcemia \[[@SFV011C36]\]. Finally, both in azotemic patients and rats, baseline CT values have been shown to be increased compared with normal \[[@SFV011C32], [@SFV011C34], [@SFV011C37]\] and also to stimulate more during the induction of hypercalcemia \[[@SFV011C34]\]. Adaptation of calcitonin secretion to the existing serum calcium concentration {#s3} ============================================================================== Similar to PTH secretion, CT secretion also appears to adapt to the ambient or existing serum calcium concentration \[[@SFV011C38]\]. In 1973 Deftos *et al*. studied the CT response to a calcium infusion in hypocalcemic patients with pseudohypoparathyroidism, idiopathic hypoparathyroidism and hypocalcemia from other causes, mostly malabsorption \[[@SFV011C23]\]. CT values were shown to increase before hypercalcemia developed. In a subsequent study in normal young male subjects, oral calcium loading that increased serum calcium to higher, but still normal values resulted in increases in serum CT that correlated with the increase in serum calcium \[[@SFV011C22]\]. In a study in normal, parathyroidectomized, and azotemic rats with the latter divided by a serum calcium less or greater than 8.5 mg/dL, it was shown that the two hypocalcemic groups, parathyroidectomized and azotemic rats, increased CT secretion in response to a PTH infusion before hypercalcemia was observed \[[@SFV011C32]\] (Figure [2](#SFV011F2){ref-type="fig"}). Finally, Messa *et al*. showed in normal subjects and CKD patients, a tight correlation between the set points for PTH and CT secretion with the latter greater than the former \[[@SFV011C34]\]. The correlation between the set points for PTH and CT secretion suggests that the PTH response to hypocalcemia and the CT response to hypercalcemia move together. Fig. 2.The sigmoidal relationship between serum calcium and calcitonin is shown in the four groups of rats: normal (N)---open circle; parathyroidectomized (PTX)---solid circle; renal failure with baseline serum calcium \<8.5 mg/dL (RF~a~)---solid square; and renal failure with baseline serum calcium \>8.5 mg/dL (RF~b~)---open square. When hypocalcemia was present (PTX and RF~a~), the calcitonin response to an increase in serum calcium began before hypercalcemia developed shifting the set point for calcitonin secretion to the left \[[@SFV011C32]\]. Reprinted with permission from Kidney International. The adapting of PTH and CT secretion to the existing serum calcium concentration is an interesting phenomenon that may reflect a broader concept of physiologic adaptation. For example, adaptation to cold and hot weather as well as oxygen adaptation to high altitude in Himalayan mountain climbers is well known. However, adaptation of hormonal effects and secretion is less well appreciated. In type 1 diabetic patients, an unawareness of hypoglycemia is associated with prolonged insulin therapy and frequent episodes of hypoglycemia \[[@SFV011C39]\]. In a study of patients before and after removal of insulinomas (baseline serum glucose, 50 ± 7 mg/dL), symptomatology as well as the epinephrine, norepinephrine, glucagon, growth hormone, and cortisol response to a hypoglycemic clamp was greatly blunted \[[@SFV011C39]\]. After resection of the insulinoma, these same responses became similar to those in normal subjects. Calcitonin secretion and chronic hypercalcemia: depletion of calcitonin versus adaptation {#s4} ========================================================================================= Besides adaptation to hypercalcemia, other causes for changes in CT secretion are also possible. Raue and associates have shown that chronic hypercalcemia resulted in a decrease in CT content of the thyroid gland and a diminished CT response to acute calcium stimulation while basal serum CT levels remained unchanged \[[@SFV011C40]\]. Also, it was shown that 1,25D-induced hypercalcemia failed to stimulate CT secretion \[[@SFV011C40]\] which may be related to the presence of 1,25D receptors on C cells \[[@SFV011C41]\] and a 1,25D decrease in CT gene transcription \[[@SFV011C14], [@SFV011C15], [@SFV011C42]\]. Interestingly, unlike hypocalcemia which stimulates PTH mRNA, calcium-induced hypercalcemia failed to affect CT mRNA \[[@SFV011C15], [@SFV011C42]\]. In an *in vitro* study with rat C cells, repetitive calcium stimulation led to a decline in CT release, but 2 h after reversing the calcium concentration in the media to basal values, the CT response to a calcium stimulus was restored \[[@SFV011C43]\]. In the cat, the CT response to hypercalcemia correlated with the number of CT-positive cells in the thyroid \[[@SFV011C36]\]. Also, it was shown that several cats failed to increase CT in response to hypercalcemia. Chronic hypocalcemia induced by parathyroidectomy in rats has resulted in increased thyroidal CT content after 50 days and longer, but interestingly not at 32 days \[[@SFV011C44], [@SFV011C45]\]. Such a finding could explain an enhanced CT response to calcium loading in chronic hypocalcemia \[[@SFV011C23]\]. However, in the study by Torres *et al*., the enhanced CT response to a calcium increase in hypocalcemic rats was observed in the absence of prolonged hypocalcemia \[[@SFV011C32]\]. Several studies have evaluated CT secretion in primary hyperparathyroidism which is characterized by chronic hypercalcemia. In one study, a gender difference was observed. Males, but not females had elevated baseline CT values and a further increase in CT values during a calcium infusion \[[@SFV011C46]\]. In a subsequent study, men and women with primary hyperparathyroidism had normal serum CT levels and the CT response to a calcium infusion was indistinguishable between men and women with primary hyperparathyroidism and normal men and women \[[@SFV011C47]\]. In another study performed in post-menopausal women with primary hyperparathyroidism, all the patients had normal serum CT levels and a blunted response to a calcium stimulus as compared with normal women \[[@SFV011C48]\]. In a study in horses, the rapid induction of hypercalcemia increased CT values by 6-fold, but CT values returned to baseline values before hypercalcemia resolved \[[@SFV011C49]\]. Thus, the question remains whether the lack of a CT response to hypercalcemia in primary hyperparathyroidism is due to a depletion of CT stores or a reset of the set point for CT secretion. If the hormonal response of CT to hypercalcemia is indeed similar to that of the PTH response to hypocalcemia, CT secretion might have similar characteristics as PTH secretion. When dogs were subjected to hypocalcemia, repetitive cycles of the induction of and recovery from hypocalcemia done without pause, whether for 30 or 60 min, produced the same PTH response on the second as the first cycle (Figure [3](#SFV011F3){ref-type="fig"}A and B) \[[@SFV011C38]\]. Also, the PTH value for the same serum calcium concentration was greater during the induction of than the recovery from hypocalcemia (Figure [3](#SFV011F3){ref-type="fig"}C) and greater during the recovery from than the induction of hypercalcemia (not shown) \[[@SFV011C38]\]. This phenomenon, known as hysteresis, is not unique to PTH secretion, but is seen with other physiologic phenomenon \[[@SFV011C38]\]. But for PTH secretion, hysteresis may be important in preventing an overcorrection during the restoration of a normal serum calcium concentration \[[@SFV011C38]\]. Another interesting observation has been that an episodic versus a linear induction of hypocalcemia of the same magnitude over the same time period resulted in differences in PTH secretion \[[@SFV011C50]\]. Finally, it was shown that metabolic acidosis stimulates and metabolic alkalosis inhibits PTH secretion in the rat and dog \[[@SFV011C51]--[@SFV011C54]\]. Because the CaSR is sensitive to pH changes, it would not be surprising if CT secretion also responded to changes in pH. Moreover, because calcium suppresses PTH secretion and stimulates CT secretion, the effects of acidosis and alkalosis may be reversed from that of PTH with alkalosis stimulating and acidosis suppressing CT secretion. Fig. 3.The PTH response to the sequential induction of and recovery from hypocalcemia is shown for 30 min (**A**) and 60 min (**B**) cycles in the dog. PTH values were similar during the first and second cycles both in the 30 and 60 min groups. (**C**) The lower PTH value for the same serum calcium concentration during the recovery from hypocalcemia than during the induction of hypocalcemia is shown and is known as hysteresis. Results of PTH hysteresis from dogs in the 60 min cycle were similar (data not shown). Data are the mean ± SE \[[@SFV011C38]\]. Reprinted with permission from the Clinical Journal of the American Society of Nephrology. Calcitonin and gastrin {#s5} ====================== Besides being stimulated by calcium, CT secretion has been shown to be stimulated by gastrointestinal hormones such as gastrin. An infusion of pentagastrin, a synthetic peptide with gastrin-like effects, has been used to evaluate CT stimulation \[[@SFV011C55]\]. In one study of oral calcium loading, the increase in serum CT correlated with the increase in serum calcium, but not that of gastrin \[[@SFV011C22]\]. In a recent study, the peak CT response to a pentagastrin infusion in normal women and men was 26 and 38 pg/mL, respectively, while the response to a calcium infusion was 90 and 131 pg/mL, respectively \[[@SFV011C56]\]. Moreover, the CT response to pentagastrin was absent in 12 of 25 women and 4 of 25 men, while the CT response to calcium, was absent only in 2 of 18 women and none of the men. Because gastrin and other gastrointestinal hormones potentially stimulate CT secretion, interest developed in the concept that postprandial CT secretion acted to facilitate calcium deposition in bone and potentiate bone mass \[[@SFV011C21]\]. However, in studies in children with congenital hypothyroidism and in thyroidectomized patients, bone loss has been attributed to inadequate thyroid replacement and not to the absence of CT \[[@SFV011C24], [@SFV011C57]\]. Also, skeletal changes have not been observed in patients with elevated CT values in medullary thyroid carcinoma \[[@SFV011C31]\]. Potentially, there may be different challenges among species in defending against hypercalcemia. CT is an older hormone phylogenetically than PTH dating back to fish in the ocean with a need to protect against hypercalcemia while PTH developed in land-dwelling animals to protect against hypocalcemia. As such, this may be an explanation why salmon CT is more potent than CT from mammals \[[@SFV011C7], [@SFV011C58]\]. Not often discussed is that the need to protect against hypercalcemia may be different among land-dwelling animals. Humans and domesticated animals eat meals on a regular basis throughout the day. Conversely, carnivorous animals in the wild eat large meals consisting of a high percent of their body mass at irregular intervals. Besides a large protein load, the ingestion of calcium and phosphate is also great. In Burmese pythons fed meals of mice or rats once every 2 weeks with a combined mass of the meal averaging 25% of the snake\'s body mass, the minimum rate of oxygen consumption increased 17-fold 24 h after feeding \[[@SFV011C59]\]. Serum calcium did not change, but serum phosphorus increased from 6.7 to 11.4 mg/dL 3 days after eating. Serum bicarbonate increased from 11 mM fasting to 20 mM and remained significantly elevated for 6 days. Blood pH which was 7.37 fasting increased to 7.49 24 h after feeding before decreasing to 7.19 5 days after feeding. Whether similar physiologic adaptations occur in infrequently feeding carnivorous mammals in the wild and whether CT secretion plays a role and is possibly stimulated by gastrin secretion or the calcium load is an interesting question. Also, the possibility exists that post-prandial alkalemia could be an additional stimulus for CT secretion. Calcitonin and gender/age {#s6} ========================= Gender and age differences in serum CT values have been reported. Deftos and associates reported that in normal adult subjects, basal CT values were similar between men and women, but basal CT values decreased with age in both groups and the response to a calcium or pentagastrin challenge was less in women than in men \[[@SFV011C25], [@SFV011C60]\]. A subsequent study by another group showed lower basal and stimulated CT values in women than in men, but no decrement in CT values with age \[[@SFV011C9]\]. The CT difference between women and men has been ascribed to a lower CT secretion rate in women \[[@SFV011C9]\]. In post-menopausal women, estrogen administration may increase CT levels \[[@SFV011C61]\]. Finally, infants and children may have higher serum CT levels than adults \[[@SFV011C62], [@SFV011C63]\]. Calcitonin and pregnancy/lactation {#s7} ================================== Pregnancy and lactation are unique because of the demand for transfer of maternal calcium to the fetus/infant. Serum CT and 1,25D values are increased during pregnancy and lactation \[[@SFV011C16], [@SFV011C17], [@SFV011C64]\]. Interestingly, CT has been shown to stimulate 1,25D production in the kidney in the proximal straight tubule while PTH and phosphorus stimulate 1,25D production in the proximal convoluted tubule \[[@SFV011C12], [@SFV011C13]\]. Clinical studies have shown that women lose ∼5--10% of trabecular bone mineral content during 6 months of lactation and fully regain it after weaning \[[@SFV011C18]\]. Similar changes are seen in mice and rats during lactation and weaning \[[@SFV011C18]\]. In a study in calcitonin/calcitonin gene-related-peptide-α-null mice (Ctcgrp), the decrease in bone mineral content was greater and the return to baseline bone mineral content took longer in Ctcgrp-null mice than in wildtype mice \[[@SFV011C64]\]. Administration of salmon CT in the null mice prevented the differences between null and wild-type mice. While CT is known to inhibit osteoclast activity \[[@SFV011C19], [@SFV011C65]\], the identification of CT receptors on osteocytes suggests that CT may also modify osteocyte function \[[@SFV011C31]\]. In summary, CT may have an important functional role in pregnancy and lactation. Calcitonin and bone {#s8} =================== Since the discovery of CT in the early 1960s, much interest has been focused on whether CT has an important role in normal bone health. It has been postulated that CT acts postprandially to use phosphate to store calcium in bone \[[@SFV011C21], [@SFV011C66]\]. But except perhaps for pregnancy and lactation, no specific CT deficiency or excess bone syndromes have been identified \[[@SFV011C66]\]. CT has been used to treat post-menopausal osteoporosis \[[@SFV011C67]\], but its benefit has not been dramatic. At least in ovariectomized and in normal beagle dogs in which bone histomorphometry and physical bone strength were evaluated, CT administration decreased osteoblast function, bone formation and physical bone strength \[[@SFV011C68], [@SFV011C69]\]. However, in a knockout mouse model in which the CT gene was deleted, no developmental defects were observed and procreation was normal \[[@SFV011C70]\]. Moreover, bone findings included an increase in trabecular bone volume and bone formation at 1 and 3 months of age while bone resorption was not changed. Baseline serum calcium values were unaffected but the administration of PTH resulted in greater hypercalcemia and urine deoxypyridinoline crosslinks excretion than in wild-type mice. Interestingly, the CT female knockout mouse appeared to be protected against ovariectomy-induced bone loss. While the effect of CT administration to treat hypercalcemia has been attributed to its inhibition of osteoclasts \[[@SFV011C19]\], an escape of osteoclasts from the effects of prolonged CT administration is well known \[[@SFV011C20], [@SFV011C71]\]. This result could possibly explain the failure to observe increased bone resorption in the absence of CT. The enhanced bone formation observed in the absence of CT is perhaps more difficult to explain. But CT has been shown to modify cell cultures of osteoblasts and osteocytes \[[@SFV011C72]\]. One intriguing hypothesis based on the finding of CT receptors on osteocytes, is that CT could potentially modify osteocyte products such as FGF23 or perhaps more importantly, sclerostin, a known regulator of bone formation \[[@SFV011C31]\]. Calcitonin and renal failure {#s9} ============================ The kidney is the primary site for metabolism of CT \[[@SFV011C8], [@SFV011C10]\]. As might be expected with decreased metabolism of CT in renal failure, CT values are increased in azotemic humans and animals \[[@SFV011C32], [@SFV011C34], [@SFV011C73], [@SFV011C74]\]. As in normal humans and animals, a sigmoidal CT response is seen during the induction of hypercalcemia in azotemic humans and animals \[[@SFV011C32]--[@SFV011C34]\]. The CT response to hypercalcemia has been shown to protect against the development of acute hypercalcemia in azotemic rats \[[@SFV011C11]\]. Also, a normal stimulatory response to pentagastrin has been reported in patients with chronic renal failure \[[@SFV011C74]\]. In a bone histomorphometric study, no correlation was found between bone activity and CT levels while correlations were observed with PTH \[[@SFV011C75]\]. Finally, several studies from many years ago showed that treatment with CT failed to improve renal osteodystrophy \[[@SFV011C76]--[@SFV011C78]\]. In summary, \>50 years have transpired since the discovery of calcitonin. Strong evidence exists that the calcitonin response to an increase in the serum calcium concentration protects against the development of acute hypercalcemia. Calcitonin levels are increased during pregnancy and lactation and may play an important role in increasing 1,25D, which separately or in concert with calcitonin preserve and restore maternal bone mass during transfer of calcium to the fetus/infant. Unlike other hormones, no specific developmental or metabolic abnormalities have been associated with a deficiency or excess of calcitonin except for diarrhea in a few patients with medullary thyroid carcinoma. Unlike other hormones except for gender-specific hormones such as estrogen and androgens, distinct differences in the characteristics of calcitonin secretion appear to be present between males and females. Like the PTH response to hypocalcemia, the calcitonin response to hypercalcemia is a sigmoidal curve. Unlike PTH for which virtually all humans and animals respond to hypocalcemia with a PTH response, a number of humans and animals fail to produce a calcitonin response to hypercalcemia. Unlike PTH which only seems to respond to calcium perhaps modified by pH, calcitonin, besides responding to calcium, also responds to gastrointestinal hormones of which gastrin has been the best studied. The postprandial increase in gastrin and its association with calcitonin secretion has led to the hypothesis that calcitonin may have an important postprandial role in facilitating calcium and phosphate deposition in bone. Despite interest in this concept, no definite proof has been established. However, renewed interest in calcitonin may develop because of potential effects of calcitonin on the osteocyte, but bone abnormalities resulting from a deficiency or excess have not been shown so that any effect must be subtle or counterbalanced by other factors. Recent studies in the calcitonin gene knockout mouse model, have suggested that the absence of calcitonin increased bone mass through enhanced bone formation without having an effect on bone resorption which might have been predicted based on previous known effects of calcitonin on the osteoclast. Finally, whether increased calcitonin levels have any role in chronic kidney disease remains unknown. A summary of the known or possible effects or responses of calcitonin is shown in the Table [1](#SFV011TB1){ref-type="table"}. In conclusion, calcitonin, which is no longer even included as a chapter in the most recent edition of the ASBMR Primer on the Metabolic Bone Diseases and Disorders of Mineral Metabolism \[[@SFV011C79]\], is a hormone preserved during evolution from ocean to land-based animals that continues to be an enigma more than 50 years after its discovery. Table 1.Known or possible effects or responses of calcitoninA.Known effects or responses of calcitoninNo specific developmental or metabolic abnormalities from a deficiency or excess except for diarrhea in a few patients with medullary thyroid carcinomaProtects against the development of hypercalcemiaSigmoidal response to the development of hypercalcemiaSecretion adapts to ambient serum calcium concentrationStimulated by calcium infusion and less predictably by pentagastrin infusionImmediate, but short-term inhibition of osteoclast activityReciprocal relationship with 1,25 (OH)2 vitamin D (1,25D)---increased 1,25D production through 1-alpha hydroxylase in the straight proximal tubule and suppression of calcitonin mRNA by 1,25DIncreases during pregnancy and lactationGender differences with higher baseline values in males than in females; values may also decrease with ageElevated in chronic renal failure, but no proven effectB.Possible, but unproven effects of calcitoninImportant factor during pregnancy and lactation in stimulating 1,25D thus preventing maternal bone loss and transferring calcium to the fetus/infant in conjunction with 1,25DEnhanced post-prandial deposition of calcium and phosphate in bone from stimulation of calcitonin by gastrointestinal hormones such as gastrinModifier of osteocyte function and productsModifier of bone formation based on results in a calcitonin-gene knockout mouse model Conflict of interest statement {#s10} ============================== None of the authors have any conflicts of interest. The content of this review has not been published previously.

1. Introduction {#sec1-ijerph-17-04365} =============== Academics and governments have been paying increased attention to the balance between economic development and environmental protection \[[@B1-ijerph-17-04365],[@B2-ijerph-17-04365]\]. China is not an exception to this trend. In the past several decades, China has made tremendous economic achievements. However, environmental deterioration has become another serious problem due to the accelerated development of the economy \[[@B3-ijerph-17-04365],[@B4-ijerph-17-04365]\]. To address environmental deterioration, a series of environmental regulations (ERs) have been enacted by the Chinese government in recent years \[[@B5-ijerph-17-04365],[@B6-ijerph-17-04365]\]. The central government of China had implemented 30 state laws and 1400 industrial environmental standards by 2013 \[[@B7-ijerph-17-04365]\]. By 2017, local governments had implemented 480 local laws and 314 local rules and regulations related to environmental governance. However, from a theoretical perspective, it remains unclear whether ERs can address environmental issues effectively or not and whether ERs enable a win-win strategy between economic growth and environmental protection \[[@B8-ijerph-17-04365]\]. Traditional wisdom believes that environmental protection and economic development are contradictory \[[@B9-ijerph-17-04365]\]. In contrast to traditional wisdom, the Porter Hypothesis claims that properly designed ERs can encourage firms to innovate both their products and production process. Technological innovation (TI) can help offset the compliance costs, enhance the utilization efficiency of resources and achieve competitive advantages \[[@B10-ijerph-17-04365],[@B11-ijerph-17-04365]\]. Under this circumstance, the win-win situation can be achieved. Hence, determining whether ERs can stimulate TI or not is the key for enabling the win-win situation between economic development and environmental protection \[[@B12-ijerph-17-04365],[@B13-ijerph-17-04365]\]. The impacts of ERs on TI have been widely studied for several decades \[[@B14-ijerph-17-04365],[@B15-ijerph-17-04365]\], and diverse resulting relationships (e.g., positive, negative, neutral and nonlinear) have been found \[[@B14-ijerph-17-04365],[@B16-ijerph-17-04365],[@B17-ijerph-17-04365],[@B18-ijerph-17-04365]\]. In this paper, we also analyze the impacts of ERs on TI in the Chinese context, but there are several differences compared to prior studies. First, a comprehensive literature survey shows that most of previous studies treat TI as a single production process \[[@B19-ijerph-17-04365],[@B20-ijerph-17-04365]\]. In reality, the whole process of TI can be more complex. According to the innovation value chain, TI can be divided into two related subprocesses, namely technology investment and technology transformation \[[@B15-ijerph-17-04365],[@B19-ijerph-17-04365]\]. To match the reality more closely, this study thus decomposes TI into technology investment and technology transformation. Decomposing TI can help open the black box and see the inefficient subprocess of TI. Second, when investigating the impacts of ERs on TI, the majority of extant studies generally employ a single indicator, such as research and development (R&D) investment and the number of patents, to capture TI \[[@B19-ijerph-17-04365]\]. However, only using one indicator to capture TI may cause bias, and TI can be better captured with multiple indicators \[[@B21-ijerph-17-04365]\]. Hence, we use multiple indicators to measure TI in this study. Besides, we show a novel application of the weighted slack-based measure (SBM) Network data envelopment analysis (DEA) model (hereafter called Network SBM DEA model) to evaluate the efficiency of TI. The Network SBM DEA model is a nonradial method that is suitable for inputs and outputs that do not change proportionally. The Network SBM DEA model can also take the importance of a subprocess into consideration by setting different weights to different subprocesses exogenously \[[@B22-ijerph-17-04365]\]. Hence, the Network SBM DEA model has some advantages in measuring the efficiency, but it is seldom employed to measure the overall efficiency of TI along with the efficiencies of its subprocesses in the existing studies. By employing the Network SBM DEA model, we not only evaluate the overall efficiency of TI (TIE) but also evaluate its subprocesses' efficiencies, including the efficiency of technology investment (TVE) and the efficiency of technology transformation (TTE). In sum, our innovative application of the Network SBM DEA model to measure the overall efficiency of TI along with the efficiencies of its subprocesses is a significant contribution. Third, as highlighted earlier, the influence of ERs on TI has been widely examined for several decades. However, the existing studies fail to examine whether or not the impacts of ERs on different subprocesses of TI are heterogeneous \[[@B19-ijerph-17-04365]\]. As we divide TI into technology investment and technology transformation, apart from investigating the linear and nonlinear impacts of ERs on TI, we also examine whether the impacts of ERs on the decomposed components of TI are heterogeneous or not. In this regard, this study can further open the black box and provide a new perspective for understanding the effects of ERs on TI. Overall, the purposes of this study are to (i) estimate the overall efficiency of TI along with the efficiencies of two subprocesses of TI (TVE and TTE), (ii) explore both the linear and nonlinear ERs--TI links and (iii) explore whether the impacts of ERs on different subprocesses of TI are heterogeneous or not. By doing so, several contributions have been made. First, we open the black box by decomposing the whole process of TI into two different subprocesses rather than treating it as a single process. This can help identify the inefficient subprocess of TI. Second, we use multiple indicators instead of a single indicator to capture TI, and we employ a novel application of a Network SBM DEA model to measure the overall efficiency of TI and the efficiencies of its subprocesses. Third, we further open the black box by exploring whether the impacts of ERs on different subprocesses of TI are heterogeneous or not, which can provide a new perspective for understanding the link between ERs and TI. Finally, the novel examination of the impacts of ERs on TIE, TVE and TTE in the Chinese context is another contribution of this paper. 2. Literature Review {#sec2-ijerph-17-04365} ==================== The literature review section is divided into four sections. The first section focuses on Chinese context on regulations and innovation. [Section 2.2](#sec2dot2-ijerph-17-04365){ref-type="sec"} focuses on the decomposition of TI, and [Section 2.3](#sec2dot3-ijerph-17-04365){ref-type="sec"} focuses on the link between ERs and TI. The last section focuses on the application of DEA and Network DEA models. 2.1. A Brief Overview of Regulations and Innovation in the Chinese Context {#sec2dot1-ijerph-17-04365} -------------------------------------------------------------------------- China is a vast country with a number of environmental regulations and innovation regimes. To resolve the environmental problems caused by economic development, the Chinese government has made great efforts in environmental governance of industrial pollution. The government has established a sound environmental governance system with multiple actors implementing multiple environmental policy instruments that target polluting industries (or firms) \[[@B7-ijerph-17-04365]\]. Besides, the government has tightened its ERs in recent years. Some examples include the promulgation of the Air Pollution Prevention and Control Action Plan in 2013, the newly implemented policy of Environmental Protection Admonishing Talk in 2014, the amendment of Environmental Law in 2015 and the newly revised Air Pollution Prevention and Control Law in 2015 \[[@B23-ijerph-17-04365]\]. Overall, the tightened ERs can force firms to invest more in innovation so as to further improve environmental quality. TI plays a crucial role in the coordinated development between economic development and environmental protection \[[@B12-ijerph-17-04365],[@B13-ijerph-17-04365]\]. Hence, it is important for firms to improve their innovation capability and efficiency. To encourage firms to pay more attention to TI, the Chinese government has enacted a series of policies such as the National Medium- and Long-Term Plan for the Development of Science and Technology (2006--2020) in 2006, the Innovation-driven Development Strategy in 2012 and the Made in China 2025 government-led initiative in 2016 \[[@B24-ijerph-17-04365]\]. Through TI, traditional industries can be upgraded with advanced applicable technologies. This can not only help reduce energy consumption and adverse impacts on the environment, but also change the traditional development model of excessive resource consumption and environmental pollution. 2.2. Decomposing Technological Innovation into Different Subprocesses {#sec2dot2-ijerph-17-04365} --------------------------------------------------------------------- TI is usually treated as a single production process in the existing literature \[[@B19-ijerph-17-04365],[@B20-ijerph-17-04365]\]. Under this circumstance, the whole process of TI is regarded as a black box. However, the real TI production process can be more complex. According to innovation value chain, firms usually transform the knowledge they obtain into different technologies and products from which they generate revenue \[[@B25-ijerph-17-04365]\]. In this sense, TI can be decomposed into two parts. The first part relates to the activities that firms conduct to seek inputs for innovation \[[@B26-ijerph-17-04365]\]. The second part relates to the transformation of knowledge acquired by the firms into innovation outputs, which is also called the process of knowledge transformation \[[@B27-ijerph-17-04365]\]. Accordingly, Chen et al. \[[@B25-ijerph-17-04365]\] decompose innovation activities into the R&D process and the commercialization process. The R&D process applies the obtained knowledge to innovation, whereas the commercialization process introduces innovation into the market. Li et al. \[[@B19-ijerph-17-04365]\] decompose TI into technology investment and technology transformation. Zhu et al. \[[@B15-ijerph-17-04365]\] argue that TI can be decomposed into two stages: knowledge innovation and product innovation. Extant literature suggests the need for decomposing TI into different subprocesses from the perspective of the innovation value chain. 2.3. The Link between Environmental Regulations and Technological Innovation {#sec2dot3-ijerph-17-04365} ---------------------------------------------------------------------------- ### 2.3.1. The Linear Link between Environmental Regulations and Technological Innovation {#sec2dot3dot1-ijerph-17-04365} #### Positive, Negative and Neutral Links The direct linear link between ERs and TI has been a constant focus among scholars, especially in the early literature. Most conclusions support the Porter Hypothesis and suggest that ERs can stimulate TI. A positive link between ERs and TI has been found in different countries, including India \[[@B28-ijerph-17-04365],[@B29-ijerph-17-04365]\], Chile \[[@B30-ijerph-17-04365]\], Germany \[[@B31-ijerph-17-04365]\], Japan \[[@B32-ijerph-17-04365],[@B33-ijerph-17-04365],[@B34-ijerph-17-04365]\] and China \[[@B35-ijerph-17-04365]\]. Apart from TI, scholars also argue that ERs can also stimulate environmental innovation \[[@B36-ijerph-17-04365],[@B37-ijerph-17-04365],[@B38-ijerph-17-04365]\]. The positive influence of ERs on eco-innovation has been found in the US \[[@B39-ijerph-17-04365]\], Germany \[[@B40-ijerph-17-04365]\] and China \[[@B41-ijerph-17-04365],[@B42-ijerph-17-04365],[@B43-ijerph-17-04365],[@B44-ijerph-17-04365]\]. Besides, Johnstone et al. \[[@B45-ijerph-17-04365]\] use unbalanced panel data obtained from 77 countries during 2001--2007 and conclude that ERs have positive impacts on environment-related innovation. Calel and Dechezleprêtre \[[@B46-ijerph-17-04365]\] find that the European Union Emissions Trading System contributes to low-carbon patenting and does not have a crowding-out effect on patents in regulated firms. In addition to the positive link, some scholars also prove that the link can be negative, as the compliance costs of ERs can crowd out the expenditure on TI \[[@B21-ijerph-17-04365],[@B47-ijerph-17-04365]\]. Hence, the Porter Hypothesis is not supported. Brunnermeier and Cohen \[[@B48-ijerph-17-04365]\] confirm that the current laws and regulatory regimes do not provide incentives for firms to carry out TI in US manufacturing industries. Similarly, Chintrakarn \[[@B49-ijerph-17-04365]\] also argues that ERs have positive and significant impacts on technical inefficiency for the US manufacturing sector. For the neutral link, Smith and Crotty \[[@B50-ijerph-17-04365]\] confirm that the environmental policy called EU End of Life Vehicles Directive has limited influence on product innovation. Li et al. \[[@B19-ijerph-17-04365]\] find that ERs have no impact on the efficiency of TI. Yi et al. \[[@B51-ijerph-17-04365]\] show that the instruments of ERs employed in China do not favor green innovation. #### Uncertain Link Some scholars also look beneath the surface and investigate some other factors that may affect the focal link. The results provide evidence for a more complex and uncertain relationship between ERs and TI, as the focal link can be affected by the organization's structural flexibility and production process flexibility \[[@B52-ijerph-17-04365]\], the levels of firms' innovation \[[@B53-ijerph-17-04365]\], and the extent of market uncertainty \[[@B54-ijerph-17-04365]\]. In addition, some scholars also distinguish ERs and innovation into different types and further demonstrate that the relationships between different kinds of ERs and TI \[[@B8-ijerph-17-04365],[@B55-ijerph-17-04365],[@B56-ijerph-17-04365]\] and the links between ERs and different types of innovation \[[@B57-ijerph-17-04365],[@B58-ijerph-17-04365]\] are heterogeneous. Hence, both the types of ERs and the types of TI can affect the focal link. ### 2.3.2. The Nonlinear Link between Environmental Regulations and Technological Innovation {#sec2dot3dot2-ijerph-17-04365} Since there have been no consistent conclusions, scholars have further investigated the potential nonlinear link. The ever-increasing stringency of ERs can increase the compliance costs \[[@B59-ijerph-17-04365],[@B60-ijerph-17-04365]\]. According to Restraint Theory, firms generally have limited resources and the ever-increasing compliance costs can crowd out firm's R&D investment \[[@B3-ijerph-17-04365],[@B61-ijerph-17-04365]\]. Therefore, ERs may inhibit TI in the short run. Nevertheless, the ever-increasing intensity of ERs may also provide incentives for firms to invest more in innovation in the long run. The increasing investments in innovation may generate extra profits so that the compliance costs may be partially or completely offset. Meanwhile, firms' innovation capabilities can improve as well \[[@B62-ijerph-17-04365]\]. Thus, with the increasing of ER stringency, the effects of ERs on TI can shift from negative to positive from the perspective of dynamic development \[[@B63-ijerph-17-04365]\]. In summary, the link between ERs and TI can be U-shaped. Yuan et al. \[[@B64-ijerph-17-04365]\] show that the impact of ERs on TI is U-shaped in the subsample consisting of medium eco-efficiency sectors. According to the Porter Hypothesis, well-designed ERs can stimulate TI, which can help in achieving a win-win strategy between economic development and environmental protection \[[@B10-ijerph-17-04365],[@B11-ijerph-17-04365]\]. However, drawing on the theory of the 'too-much-of-a-good-thing' (TMGT) effect, there exists a maximum value for the link between two constructs, after which the increase in the beneficial antecedents can result in the decrease in the outcomes. The reason for this is that the additional costs can exceed the generated benefits when the beneficial antecedents surpass a certain level. Hence, the link between two constructs is inverted U-shaped \[[@B65-ijerph-17-04365],[@B66-ijerph-17-04365]\]. As such, the link between ERs and TI is also likely to be inverted U-shaped. There exists an optimum intensity of ERs, after which the increase in ER can lead to the decrease in TI. Perino and Requate \[[@B67-ijerph-17-04365]\] prove the nonlinear impact of policy stringency on broad technology adoption. Liu and Gong \[[@B62-ijerph-17-04365]\] also demonstrate that the link between ERs and green innovation capacity is inverted U-shaped. In sum, the link between ERs and TI can be either U-shaped or inverted U-shaped on the basis of different theories, which proves the possibility of the nonlinear link. 2.4. Applications of DEA and Network DEA to Capture Technological Innovation When Investigating the Impacts of Environmental Regulations on Technological Innovation {#sec2dot4-ijerph-17-04365} -------------------------------------------------------------------------------------------------------------------------------------------------------------------- DEA is a well-known methodology for estimating the relative efficiency of decision making units (DMUs) via a series of inputs and outputs \[[@B68-ijerph-17-04365],[@B69-ijerph-17-04365]\]. Generally, the objective of DEA would be to identify the DMU that has maximized outputs consuming minimal inputs. ### 2.4.1. Applications of DEA when Investigating the Impacts of Environmental Regulations on Technological Innovation {#sec2dot4dot1-ijerph-17-04365} In the existing literature, TI is usually measured with a single indicator, like R&D investment or the number of patents. However, TI can be better captured with multiple indicators \[[@B21-ijerph-17-04365],[@B70-ijerph-17-04365]\]. Hence, the DEA model is a suitable choice. Scholars also developed some new methodologies after the traditional DEA was proposed. When scholars examine the effects of ERs on TI, they also use the newly developed methodologies to evaluate the efficiency of TI. Feng et al. \[[@B16-ijerph-17-04365]\] measure the efficiency of green innovation via super-slack-based measure and demonstrate that the influence of ERs on the efficiency of green innovation is significantly negative. Zhu et al. \[[@B15-ijerph-17-04365]\] construct a game cross-efficiency model to measure technological innovation efficiency and indicate that the effect of mandatory regulation on technological innovation efficiency turns out to be insignificant, whereas voluntary regulation has positive effects at the provincial level. Deng et al. \[[@B20-ijerph-17-04365]\] evaluate regional innovation performance via super-efficiency DEA model and show that stringent ERs can help to enhance regional innovation performance. Although these studies have used the evaluation results via different DEA methodologies as proxies for TI, TI is still treated as a single process. ### 2.4.2. Applications of Network DEA When Investigating the Impacts of Environmental Regulations on Technological Innovation {#sec2dot4dot2-ijerph-17-04365} As mentioned above, TI should be decomposed into different subprocesses. Hence, DEA is not suitable, as it treats a DMU as a single production process that transforms inputs into outputs \[[@B71-ijerph-17-04365]\]. Instead, Network DEA is more suitable, as it considers a DMU as a network of interrelated processes. Decomposing the production process into different interrelated processes can help open the black box and provide deeper insights into the sources of the inefficiency \[[@B71-ijerph-17-04365]\]. However, only a few scholars have adopted Network DEA to evaluate the efficiency of different subprocesses \[[@B19-ijerph-17-04365],[@B25-ijerph-17-04365]\], and no scholar has further explored the heterogeneous impacts of ERs on different subprocesses of TI. Overall, numerous studies have been conducted and no consensus has yet been reached. A systematic review of the literature indicates that decomposing TI into different subprocesses is more suitable, even though most scholars treat TI as a single process \[[@B15-ijerph-17-04365],[@B16-ijerph-17-04365],[@B20-ijerph-17-04365]\]. According to the literature on the link between ERs and TI, most scholars adopt a single indicator to capture TI \[[@B21-ijerph-17-04365]\]. Although some scholars also use the evaluation results via relevant DEA methodologies as proxies for TI, TI is still treated as a single process in most studies. Furthermore, the existing literature has neglected the heterogeneous impacts of ERs on the decomposed components of TI. Therefore, this study intends to enrich the investigation on the influence of ERs on TI in these aspects. [Figure 1](#ijerph-17-04365-f001){ref-type="fig"} presents the conceptual framework of our study. 3. Research Methods, Data and Analysis {#sec3-ijerph-17-04365} ====================================== We employed multiple research methods to conduct our analyses. For decomposing TI, we used a Network SBM DEA model. This is described in [Section 3.1.1](#sec3dot1dot1-ijerph-17-04365){ref-type="sec"}. We used a Tobit regression model to explore the link between ERs and TI. This model is described in more detail in [Section 3.2.4](#sec3dot2dot4-ijerph-17-04365){ref-type="sec"}. 3.1. Stage 1: Opening the Black Box---Decomposing the Efficiency of Technological Innovation Using Network DEA {#sec3dot1-ijerph-17-04365} -------------------------------------------------------------------------------------------------------------- Following Li et al. \[[@B19-ijerph-17-04365]\] and Chen et al. \[[@B25-ijerph-17-04365]\], we also divided TI into two related subprocesses according to the innovation value chain: technology investment and technology transformation. The objective of technology investment is to promote the progress of basic science and technologies, whereas the objective of technology transformation is to achieve and improve the commercial application of certain technologies produced in the former subprocess \[[@B25-ijerph-17-04365]\]. [Figure 2](#ijerph-17-04365-f002){ref-type="fig"} shows the decomposition of TI. The next section discusses the model, while the details of inputs, intermediates and final outputs for TI are included in [Section 3.1.2](#sec3dot1dot2-ijerph-17-04365){ref-type="sec"}. ### 3.1.1. Network SBM DEA Model {#sec3dot1dot1-ijerph-17-04365} To operationalize the decomposition outlined in [Figure 2](#ijerph-17-04365-f002){ref-type="fig"}, we adopted the weighted slack-based measure Network data envelopment analysis model (Network SBM DEA model) to measure the efficiencies of TI and its decomposed components \[[@B22-ijerph-17-04365],[@B72-ijerph-17-04365]\]. The Network SBM DEA model has some merits in measuring efficiency, but it is seldom adopted to measure the efficiency of TI. By employing the Network SBM DEA model, the overall efficiency of TI (TIE) along with the efficiencies of two subprocesses (TVE and TTE) were measured in this study. More specifically, we utilized the input-oriented SBM under the variable returns-to-scale assumption with the fixed link case for evaluating the efficiencies of TI and its decomposed components. Besides, the weights of different subprocesses were needed during the calculation process of the overall efficiency. Both technology investment and technology transformation are equally important. Hence, the weights for technology investment and technology transformation were both set as 0.5. ### 3.1.2. Sample, Indicators and Data Sources {#sec3dot1dot2-ijerph-17-04365} Our first objective was to estimate TIE, TVE and TTE of Chinese two-digit industrial sectors. The first two digits of the code represents the industrial sector to which the business belongs. From the total of 41 two-digit Chinese industrial sectors we included the 36 sectors described in [Table 1](#ijerph-17-04365-t001){ref-type="table"} and collected data from between 2005 and 2015. Some sectors were not included because of missing data. More detailed information can be found in [Table 1](#ijerph-17-04365-t001){ref-type="table"}. Following Chen et al. \[[@B25-ijerph-17-04365]\], Li et al. \[[@B19-ijerph-17-04365]\] and Zhu et al. \[[@B15-ijerph-17-04365]\], several indicators were adopted to evaluate TIE, TVE and TTE. For the measurement of TVE, the inputs included full-time equivalent of R&D personnel, R&D intramural expenditure and expenditure for technology acquisition and purchase. The output was the number of patent applications, which was also the input of technology transformation. For the measurement of TTE, the new inputs included expenditure for developing new products and expenditure for technology assimilation and renovation, whereas the final outputs were sales from new products and industrial sales output value. The data needed for this study were collected from *Statistics on Science and Technology Activities of Industrial Enterprises* and *China Industry Statistical Yearbook*. [Table 2](#ijerph-17-04365-t002){ref-type="table"} shows the details of the indicators and the data sources of the indicators. ### 3.1.3. Evaluation Results and Analysis {#sec3dot1dot3-ijerph-17-04365} We employed MaxDEA 8 Ultra to evaluate the efficiency. While we do not show the annual efficiency score of each sector in this article, they are available upon request. The evaluation results of all sectors' annual average TIE, TVE and TTE scores are displayed in [Figure 3](#ijerph-17-04365-f003){ref-type="fig"}. From the holistic perspective, TIE, TVE and TTE increased slightly, although the efficiency score of each type fluctuated dramatically during the research period. The value of decomposition is clearly visible in [Figure 3](#ijerph-17-04365-f003){ref-type="fig"}; while TTE seems to follow the same trend as TIE, the trend of TVE is somewhat different. First, TVE consistently recorded a high value, meaning that Chinese firms are very efficient in generating patents. Second, TVE showed a decreasing trend during 2006--2019 and an increasing trend in the subsequent two years, although TIE and TTE showed mixed trends during the same periods. Thus, opening the black box has revealed new insights into Chinese TI. Chinese industrial sectors currently place a heavy emphasis on technology investment rather than technology transformation. There is much room for the improvement of TTE. 3.2. Stage 2: Exploring the Impacts of ERs on TI {#sec3dot2-ijerph-17-04365} ------------------------------------------------ To further explore the impacts of ERs, we continued to use the sector-level data over the period of 2005 to 2015. ### 3.2.1. Dependent Variable: Technological Innovation {#sec3dot2dot1-ijerph-17-04365} We carried out three different analyses to understand the impacts of ERs on (i) TI (without decomposition), for which the dependent variable is TIE; (ii) TVE; and (iii) TTE. All dependent variables were measured by the Network SBM DEA model described in [Section 3.1.1](#sec3dot1dot1-ijerph-17-04365){ref-type="sec"}. ### 3.2.2. Independent Variable: Environmental Regulations {#sec3dot2dot2-ijerph-17-04365} Scholars have adopted many different methods to measure ERs. The costs and expenditures of pollution abatement or pollutant discharge fees are frequently used methods \[[@B6-ijerph-17-04365],[@B57-ijerph-17-04365],[@B58-ijerph-17-04365],[@B73-ijerph-17-04365]\]. Pollutant emissions are also used by scholars to represent ERs \[[@B74-ijerph-17-04365]\]. There are some other proxies employed by scholars, such as the number of newly implemented laws, regulations and rules \[[@B5-ijerph-17-04365],[@B6-ijerph-17-04365]\]; the number of complaint letters on pollution and environment-related problems \[[@B6-ijerph-17-04365],[@B7-ijerph-17-04365]\]; questionnaire survey results \[[@B75-ijerph-17-04365]\]; and the number of inspections \[[@B76-ijerph-17-04365]\]. Following Ye and Wang \[[@B8-ijerph-17-04365]\], Yuan et al. \[[@B64-ijerph-17-04365]\], Yuan and Xiang \[[@B21-ijerph-17-04365]\] and Liu and Xie \[[@B77-ijerph-17-04365]\], we used the operating costs of pollution treatment facilities, including industrial waste gas and industrial wastewater, to represent the stringency of ERs. The operating costs of industrial waste solid treatment were excluded due to the lack of data. Taking the influence of different scales into account, the operating costs of pollution treatment facilities were further divided by the industrial sales output value of each sector. The data required for ERs were collected from *China Industry Statistical Yearbook* and *China Statistical Yearbook on Environment* \[[@B78-ijerph-17-04365]\]. ### 3.2.3. Control Variables {#sec3dot2dot3-ijerph-17-04365} We also incorporated a couple of control variables to ensure the quality of the empirical results according to extant literature. [Table 3](#ijerph-17-04365-t003){ref-type="table"} describes the detailed information regarding the control variables. The required data for control variables were collected from *China Industry Statistical Yearbook.* ### 3.2.4. Model Selection {#sec3dot2dot4-ijerph-17-04365} Given that the range of the efficiency score was from 0 to 1, the random-effect Tobit model was used to conduct the analyses. The Tobit model can deal with a censored dependent variable \[[@B84-ijerph-17-04365]\]. To test the linear impacts of ERs, we constructed Model 1. Considering that it may take some time for ERs to exert their impacts on TI, ERs were lagged by one year \[[@B21-ijerph-17-04365],[@B74-ijerph-17-04365]\]. Moreover, we also lagged all control variables by one year to eliminate the threat of endogeneity \[[@B5-ijerph-17-04365]\]. $$TIE_{it}/TVE_{it}/TTE_{it} = C + \alpha_{1}ER_{it - 1} + \alpha_{2}Size_{it - 1} + \alpha_{3}Comp_{it - 1} + \alpha_{4}D2A_{it - 1} + \alpha_{5}FDI_{it - 1} + \alpha_{6}Export_{it - 1} + \epsilon_{it}$$ To test the nonlinear effects of ERs, we constructed Model 2. Compared with Model 1, we incorporated the quadratic term of ERs into the model. To reduce the potential threat of multi-collinearity, we centered the quadratic term before incorporating it into the models \[[@B85-ijerph-17-04365]\]. We also winsorized the continuous variables at the 1% and 99% levels to account for the bias caused by the outliers. $$TIE_{it}/TVE_{it}/TTE_{it} = C + \beta_{1}ER_{it - 1} + \beta_{2}ER_{it - 1}^{2} + \beta_{3}Size_{it - 1} + \beta_{4}Comp_{it - 1} + \beta_{5}D2A_{it - 1} + \beta_{6}FDI_{it - 1} + \beta_{7}Export_{it - 1} + \epsilon_{it}$$ The other estimations of this study were conducted in the STATA 14.0 software. [Table 4](#ijerph-17-04365-t004){ref-type="table"} shows the descriptive statistics of all variables. Because of missing data, the number of observations for ERs and foreign direct investment (FDI) were 393 and 391, respectively. The mean and median values of ERs were 33.04 and 18.36, showing a significant difference. The results indicated that regulation pressures that most firms face do not arrive at the average level. In addition, the mean values of TVE were the highest when compared with TIE and TTE. [Table 5](#ijerph-17-04365-t005){ref-type="table"} presents the correlation coefficients. The correlation coefficient between ERs and TIE was negative but not significant (β = −0.010, *p* = n.s.), whereas the correlation coefficient between ERs and TTE was also insignificant but positive (β = −0.026, *p* = n.s.). Conversely, ERs were significantly correlated with TVE (β = −0.103, *p* \< 0.05). Although TIE, TVE and TTE were all significantly correlated with each other, they were dependent variables in separate models. We also examined the values of variance inflation factors (VIF) to check for problems with multi-collinearity. For all the regressions below, the maximum value of VIF was 3.87. Generally, if the values of VIF are under 10, multi-collinearity can be avoided \[[@B86-ijerph-17-04365]\]. Hence, multi-collinearity was not a problem in our estimation. 4. Empirical Results and Discussion {#sec4-ijerph-17-04365} =================================== 4.1. Regression Results {#sec4dot1-ijerph-17-04365} ----------------------- [Table 6](#ijerph-17-04365-t006){ref-type="table"} shows the results of the influence of ERs on TI. The linear impacts of ERs on TIE are statistically insignificant (α = −0.0007, *p* = n.s.). However, the nonlinear impacts of ERs on TIE are significant, as the coefficient of (ER~st−1~)^2^ is significantly positive (β = 0.00002, *p* \< 0.1). Together, the results prove that the impacts of ERs on TIE are nonlinear rather than linear. In terms of the influence of ERs on the decomposed components of TI, the empirical results are different. Similar to the results of the influence of ERs on TIE, the impacts of ERs on TVE are also nonlinear. However, for the linear impacts of ERs on TTE, the regression results show a neutral type (α = −0.0007, *p* = n.s.), and the nonlinear regression results also present a neutral type (β = −0.0024, *p* \< 0.1.; β = 0.00001, *p* = n.s.). It can be concluded that the effects of ERs on TTE are not statistically significant. Taken together, the results support that ERs have heterogeneous impacts on different subprocesses of TI. 4.2. Robustness Tests {#sec4dot2-ijerph-17-04365} --------------------- In our model specification, ERs are lagged by one year. However, lagging ERs only one year may not account for the endogeneity sufficiently. To further address the reversed causality, following Leiter et al. \[[@B87-ijerph-17-04365]\], we incorporate two-year lagged ERs into the models and conduct the analyses again. [Table 7](#ijerph-17-04365-t007){ref-type="table"} shows the relevant results. Except for the impacts of ERs on TTE, which show a U-shaped type, the rest of the results are similar with the results we have obtained for one-year lagged ERs. Besides, to address the endogeneity problem systematically, as suggested by Newey \[[@B88-ijerph-17-04365]\], we employ instrumental variable tobit to estimate the models again. We use the lagged ERs as the instrument \[[@B89-ijerph-17-04365]\]. The results are also shown in [Table 7](#ijerph-17-04365-t007){ref-type="table"}; they also confirm our previous findings. 4.3. Discussion {#sec4dot3-ijerph-17-04365} --------------- Our results prove that ERs do not have linear impacts on TIE, TVE and TTE and the impacts of ERs on TIE are further from linear effects. The findings are consistent with some prior studies. Yuan et al. \[[@B64-ijerph-17-04365]\] show that the link between ERs and TI is U-shaped in the subsample consisting of medium eco-efficiency sectors. The nonlinear impacts of ERs on TIE prove the existence of a turning point. Since the quadratic term of ERs is positive, the negative impacts can be diminished with the increasing of the stringency of ERs. After surpassing the threshold, ERs can start to play a positive role in TI. In present-day China, the overall intensity of ERs is still not great enough to play a positive role, as the coefficient of ER~st−1~ is significantly negative (β = −0.0030, *p* \< 0.05). Hence, the stringency of ERs should be improved continuously so that ERs can be beneficial to TI. In addition, we open the black box by providing interesting results showing that ERs have heterogeneous impacts on different subprocesses of TI. The results strongly prove the need for the decomposition of TI and provide a novel perspective for understanding the effects of ERs on TI. The impacts of ERs on TVE and TTE changing from nonlinear to neutral suggests that the current intensity of ERs has different effects on technology investment and technology transformation. Under the same intensity of ERs, ERs can exert significant impacts on technology investment but exert insignificant impacts on technology transformation. Hence, to exert ERs' significant and positive effects on technology transformation rather than only on technology investment, the stringency of ERs should be increased. Only when the intensity of ERs is strong enough can ERs have significant impacts on both technology investment and technology transformation. We also present a new finding showing that technology investment can be more sensitive than technology transformation as a response to ERs. A change in the increasing of the intensity of ERs may be able to exert their positive impacts on technology investment but not on technology transformation. Why should the intensity of ERs should be increased so that ERs can exert their significant impacts on technology transformation compared with technology investment? The reasons may lie in that technology investment and technology transformation represent different environmental strategies that firms adopt. When firms mainly focus on technology investment, they may respond to ERs sensitively but reactively. An increase in the intensity of ERs may quickly result in greater expenditure on technology investment \[[@B90-ijerph-17-04365]\]. Conversely, when firms focus more on technology transformation, they may start to adopt more proactive environmental strategies as a response to ERs. The aim of technology transformation is to generate revenue through the commercial application of certain technologies produced in the subprocess of technology investment \[[@B21-ijerph-17-04365],[@B27-ijerph-17-04365],[@B91-ijerph-17-04365]\]. Hence, when firms engage more in technology transformation and aim to achieve the commercial success of TI, they need to consider the demands of consumers and thus respond to ERs more proactively. As mentioned earlier, sectors attach more emphasis on technology investment rather than technology transformation in present-day China (see [Figure 3](#ijerph-17-04365-f003){ref-type="fig"}). Hence, how should firms be encouraged to adopt proactive environmental strategies rather than reactive environmental strategies? The key also lies in the intensity of ERs. With the increasing of the stringency of ERs, the environmental strategies that firms adopt may shift from reactive to proactive \[[@B82-ijerph-17-04365],[@B92-ijerph-17-04365]\]. Hence, the stringency of ERs should be increased so that ERs can exert a positive effect on technology transformation. Our findings can help to deepen the understanding of the Porter Hypothesis. When firms engage more in technology transformation and not only in technology investment, the Porter Hypothesis can be better supported. When firms give priority to technology transformation, they start to adopt proactive rather than reactive environmental strategies so that TI can fully exert its positive effect on performance. When ERs are able to promote technology transformation, the win-win situations that benefit both the environment and the economy can be achieved more easily. ### Policy Implications To address environmental problems, the Chinese government has promulgated and enacted a series of policies aimed to improve the stringency of ERs and encourage firms to invest more in TI. To fully achieve the win-win strategy between economic development and environmental protection, drawing on our findings, we can propose some valuable implications for policy-makers. On one hand, the government needs to improve the intensity of ERs until the intensity can surpass the turning point so that ERs can start to exert positive effects on TI. On the other hand, under a certain intensity of ERs, ERs may exert their positive impacts on technology investment but not on technology transformation. Hence, the stringency of ERs should be increased so that ERs can exert positive impacts on technology transformation. In this way, ERs can exert positive impacts on two different components of TI. All in all, when the government improves the intensity of ERs, ERs can not only exert their positive impacts on all sectors, but also exert their positive impacts on different subprocesses of TI. 5. Conclusions {#sec5-ijerph-17-04365} ============== Determining whether ERs can stimulate TI or not is the key for enabling a win-win strategy benefiting both economic development and environmental protection. In this study, we innovatively employ a Network SBM DEA model to estimate not only TIE but also TVE and TTE via dividing TI into two different subprocesses (technology investment and technology transformation). We further examine the linear and nonlinear links between ERs and TIE and examine whether ERs have varying impacts on TVE and TTE in order to help open the black box and see the heterogeneous impacts of ERs on different subprocesses of TI. The data needed for the research questions come from different Chinese statistical yearbooks, and 36 sectors from 2005 to 2015 are included in our analyses. The evaluation results show that there is an overall increase in TIE, TVE and TTE. The importance of decomposing TI has been revealed, as it has been shown that the patterns of changes in TVE and TTE are different in China during the research period. The results further suggest that the impacts of ERs on TIE are nonlinear rather than linear. Besides, ERs have varying impacts on TVE and TTE. The impacts of ERs on TVE are nonlinear, whereas the impacts of ERs on TTE are statistically insignificant. On the basis of our results, we conclude that the intensity of ERs should be increased so that ERs can exert positive effects on technology transformation. This study enriches the existing studies in three aspects. First, we divide TI into two different subprocesses, rather than treating it as a single process, in order to open the black box and identify the inefficient subprocess of TI. Second, the majority of previous studies use a single indicator to represent TI and then investigate the impacts of ERs on TI. Instead, we use multiple indicators and innovatively use the evaluation results via the Network SBM DEA model to capture TI. Third, our study enhances the prior studies by analyzing the heterogeneous impacts of ERs on different subprocesses of TI in order to further open the black box and provide a novel perspective for understanding the effect of ERs on TI. Finally, the novel examination of the impacts of ERs on TIE, TVE and TTE in Chinese context is another contribution of this paper. Despite of the contributions mentioned above, this study has some limitations as well. We believe future research can overcome these limitations. First, for the measurement of ERs, we could not include the operating costs of industrial waste solid treatment due to the unavailability of data. Future studies can consider these costs when available. Second, some scholars have proved that different types of ERs can exert varying influence on TI \[[@B15-ijerph-17-04365],[@B55-ijerph-17-04365]\]. However, the data for different types of ERs on the sector level are not available. Future studies can investigate whether or not different types of ERs can exert varying influence on TI on the sector level when available. Third, future studies can research the moderation of industrial heterogeneity on the impacts of ERs on TI. Fourth, future studies can take regional heterogeneity into consideration and investigate whether the impacts of ERs on TI vary across different regions or not. Lastly, our study is conducted under the research setting of China. Given that China is the biggest developing country, this may limit the generalizability of the research findings \[[@B75-ijerph-17-04365]\]. Hence, scholars can collect data from other developed countries and verify the conclusions obtained in our study in their future research. Software, M.L.; writing---original draft, M.L. and R.L.; writing---review and editing, M.L., R.L. and R.R.; supervision, J.Z. and R.R. All authors have read and approved the submitted version of the manuscript. This research was funded by Philosophy and Social Science Foundation of Heilongjiang Province, grant number 18JYD389 and grant number 19GLC16. The authors declare no conflict of interest. {#ijerph-17-04365-f001} {#ijerph-17-04365-f002} {#ijerph-17-04365-f003} ijerph-17-04365-t001_Table 1 ###### Industrial sectors included in this study. No. Sectors No. Sectors ----- ------------------------------------------------------------------------------------ ----- ------------------------------------------------------------------------- 1 Mining and Washing of Coal 19 Manufacture of Raw Chemical Materials and Chemical Products 2 Extraction of Petroleum and Natural Gas 20 Manufacture of Medicines 3 Mining and Processing of Ferrous Metal Ores 21 Manufacture of Chemical Fibers 4 Mining and Processing of Nonferrous Metal Ores 22 Manufacture of Rubber and Plastic 5 Mining and Processing of Nonmetal Ores 23 Manufacture of Nonmetallic Mineral Products 6 Processing of Food from Agricultural Products 24 Smelting and Pressing of Ferrous Metals 7 Manufacture of Foods 25 Smelting and Pressing of Nonferrous Metals 8 Manufacture of Alcohol, Beverage and Refined Tea 26 Manufacture of Metal Products 9 Manufacture of Tobacco 27 Manufacture of General-Purpose Machinery 10 Manufacture of Textile 28 Manufacture of Special-Purpose Machinery 11 Manufacture of Textile and Apparel 29 Manufacture of Transportation Equipment 12 Manufacture of Leather, Fur, Feather and Related Products and Footwear 30 Manufacture of Electrical Machinery and Equipment 13 Processing of Timber, Manufacture of Wood, Bamboo, Rattan, Palm and Straw Products 31 Manufacture of Computers, Communications and Other Electronic Equipment 14 Manufacture of Furniture 32 Manufacture of Measuring Instruments 15 Manufacture of Paper and Paper Products 33 Other Manufacturing 16 Printing, Reproduction of Recording Media 34 Production and Supply of Electric Power and Heat Power 17 Manufacture of Articles for Culture, Education and Sport Activity 35 Production and Supply of Gas 18 Processing of Petroleum, Coking and Processing of Nuclear Fuel 36 Production and Supply of Water ijerph-17-04365-t002_Table 2 ###### Indicators for evaluating the efficiency of Technological Innovation and data sources. Technological Innovation Indicators Unit Sources of Data -------------------------------------------------------- ------------------------------- ----------------------------------------- ------------------ --------------------------------------------------------------------------- Technology investment Inputs Full-time equivalent of R&D personnel Man-year Statistics on Science and Technology Activities of Industrial Enterprises R&D intramural expenditure 10 thousand yuan Expenditure for technology acquisition and purchase 10 thousand yuan Intermediates Number of patent applications item Technology transformation New inputs Expenditure for developing new products 10 thousand yuan Expenditure for technology assimilation and renovation 10 thousand yuan Final outputs New product sales 10 thousand yuan Industrial sales output value 100 million yuan China Industry Statistical Yearbook ijerph-17-04365-t003_Table 3 ###### The details of control variables. Variable Name Measurement References --------------------------- -------- --------------------------------------------------------------------- --------------------------------------------------- Industrial size Size Natural logarithm of total assets \[[@B79-ijerph-17-04365]\] Industrial competition Comp Natural logarithm of the number of firms \[[@B80-ijerph-17-04365],[@B81-ijerph-17-04365]\] Debt ratio D2A Ratio of debt to assets \[[@B82-ijerph-17-04365]\] Foreign direct investment FDI Ratio of foreign direct investment to industrial sales output value \[[@B5-ijerph-17-04365]\] Export Export Ratio of export value to industrial sales output value \[[@B83-ijerph-17-04365]\] ijerph-17-04365-t004_Table 4 ###### Descriptive statistics of all variables. Variable *n* Mean S.D. Median Min Max ----------------------------------------------- ----- ------- ------- -------- ------- ------- Technological innovation efficiency (TIE) 396 0.615 0.347 0.503 0 1 Technological investment efficiency (TVE) 396 0.763 0.237 0.769 0.191 1 Technological transformation efficiency (TTE) 396 0.667 0.317 0.633 0.010 1 Environmental regulations (ERs) 393 33.04 40.32 18.36 0.711 196.5 Industrial size (Size) 396 8.723 1.216 8.759 6.273 11.14 Industrial competition (Comp) 396 6.763 1.080 6.896 3.970 8.447 Debt ratio (D2A) 396 0.544 0.080 0.560 0.239 0.681 Foreign direct investment (FDI) 391 0.249 0.166 0.246 0.001 0.814 Export 396 0.160 0.179 0.090 0 0.729 ijerph-17-04365-t005_Table 5 ###### Correlation coefficients of all variables. Variable TIE TVE TTE ERs Size Comp D2A FDI Export ---------- --------------- --------------- --------------- --------------- --------------- -------------- -------------- -------------- -------- TIE 1 TVE 0.836 \*\*\* 1 TTE 0.984 \*\*\* 0.731 \*\*\* 1 ERs −0.010 −0.103 \*\* 0.026 1 Size −0.114 \*\* −0.128 \*\* −0.112 \*\* 0.119 \*\* 1 Comp −0.276 \*\*\* −0.219 \*\*\* −0.277 \*\*\* −0.035 0.587 \*\*\* 1 D2A −0.142 \*\*\* −0.166 \*\*\* −0.121 \*\* 0.179 \*\*\* 0.354 \*\*\* 0.465 \*\*\* 1 FDI 0.060 0.142 \*\*\* 0.030 −0.341 \*\*\* −0.120 \*\* 0.279 \*\*\* 0.203 \*\*\* 1 Export 0.101 \*\* 0.186 \*\*\* 0.072 −0.346 \*\*\* −0.264 \*\*\* 0.244 \*\*\* 0.115 \*\* 0.807 \*\*\* 1 \*\* *p* \< 0.05, \*\*\* *p* \< 0.01. TIE, technological innovation efficiency; TVE, technology investment efficiency; TTE, technology transformation efficiency; ERs, environmental regulations; Size, industrial size; Comp, industrial competition; D2A, debt ratio; FDI, foreign direct investment. ijerph-17-04365-t006_Table 6 ###### Regression results. Variable TIE TVE TTE --------------- ---------------- ---------------- --------------- ---------------- ---------------- ---------------- ERs~t−1~ −0.0007 −0.0030 \*\* −0.0004 −0.0031 \*\*\* −0.0007 −0.0024 \* (−0.86) (−2.08) (−0.70) (−2.91) (−0.92) (−1.81) (ERs~t−1~)^2^ 0.00002 \* 0.00002 \*\*\* 0.00001 (1.93) (3.06) (1.58) Size~t−1~ 0.1636 \*\* 0.1571 \*\* 0.0806 0.0726 0.1597 \*\* 0.1552 \*\* (2.27) (2.18) (1.54) (1.41) (2.39) (2.32) Comp~t−1~ −0.2682 \*\*\* −0.2865 \*\*\* −0.1435 \*\* −0.1629 \*\* −0.2650 \*\*\* −0.2794 \*\*\* (−2.89) (−3.07) (−2.16) (−2.48) (−3.08) (−3.21) D2A~t−1~ −0.5593 −0.4056 −0.5669 −0.3809 −0.4855 −0.3660 (−0.87) (−0.63) (−1.19) (−0.81) (−0.82) (−0.61) FDI~t−1~ −0.1754 −0.2112 −0.1146 −0.1584 −0.2114 −0.2384 (−0.28) (−0.34) (−0.26) (−0.36) (−0.37) (−0.42) Export~t−1~ 0.3679 0.3680 0.1588 0.1596 0.4309 0.4293 (0.69) (0.70) (0.41) (0.42) (0.88) (0.88) \_cons 1.5151 \*\*\* 1.6638 \*\*\* 1.4950 \*\*\* 1.6525 \*\*\* 1.5264 \*\*\* 1.6378 \*\*\* (2.88) (3.14) (3.92) (4.37) (3.15) (3.34) N 353 353 353 353 353 353 Chi^2^ 9.6474 13.3171 6.6694 16.1014 10.9326 13.2726 Z-statistics in parentheses; \* *p* \< 0.1, \*\* *p* \< 0.05, \*\*\* *p* \< 0.01. TIE, technological innovation efficiency; TVE, technology investment efficiency; TTE, technology transformation efficiency; ERs, environmental regulations; Size, industrial size; Comp, industrial competition; D2A, debt ratio; FDI, foreign direct investment. ijerph-17-04365-t007_Table 7 ###### Regression results of robustness tests. Variable ERs Lagged Two Years Variable Instrumental Variable Tobit Regression --------------- ---------------------- ---------------- ---------------------------------------- --------------- ---------------- ---------------- ---------------- ERs~t−2~ −0.0042 \*\*\* −0.0035 \*\*\* −0.0036 \*\*\* ERs~t−1~ −0.0039 −0.0054 \*\*\* −0.0024 (−2.89) (−3.36) (−2.64) (−1.43) (−2.89) (−0.99) (ERs~t−2~)^2^ 0.00002 \*\* 0.00002 \*\*\* 0.00002 \* (ERs~t−1~)^2^ 0.0001\* 0.0001 \*\*\* 0.00003 (2.10) (2.94) (1.77) (1.66) (2.72) (1.09) Size~t−1~ 0.1492 \* 0.0798 0.1467 \*\* Size~t−1~ 0.1735 \*\*\* 0.1125 \*\*\* 0.1524 \*\*\* (1.91) (1.45) (2.02) (3.71) (3.51) (3.59) Comp~t−1~ −0.2723 \*\*\* −0.1534 \*\* −0.2679 \*\*\* Comp~t−1~ −0.3383 \*\*\* −0.2135 \*\*\* −0.3051 \*\*\* (−2.86) (−2.31) (−3.00) (−6.33) (−5.87) (−6.27) D2A~t−1~ −0.6245 −0.4578 −0.5789 D2A~t−1~ −0.2529 −0.3228 −0.1316 (−0.91) (−0.92) (−0.90) (−0.45) (−0.84) (−0.26) FDI~t−1~ −0.2467 −0.0915 −0.2809 FDI~t−1~ −0.2657 −0.1173 −0.2747 (−0.38) (−0.20) (−0.47) (−0.81) (−0.52) (−0.92) Export~t−1~ 0.3815 0.1493 0.4371 Export~t−1~ 1.1529 \*\*\* 0.7569 \*\*\* 1.0389 \*\*\* (0.69) (0.37) (0.85) (3.31) (3.16) (3.28) \_cons 1.7956 \*\*\* 1.5646 \*\*\* 1.7920 \*\*\* \_cons 1.6297 \*\*\* 1.4919 \*\*\* 1.5618 \*\*\* (3.21) (3.98) (3.45) (5.31) (7.06) (5.60) N 318 318 318 N 317 317 317 Chi^2^ 17.7257 18.0344 17.2371 Chi^2^ 48.4098 53.5856 46.2037 Z-statistics in parentheses; \* *p* \< 0.1, \*\* *p* \< 0.05, \*\*\* *p* \< 0.01. TIE, technological innovation efficiency; TVE, technology investment efficiency; TTE, technology transformation efficiency; ERs, environmental regulations; Size, industrial size; Comp, industrial competition; D2A, debt ratio; FDI, foreign direct investment.

Published: June 26, 2020 Introduction {#sec1} ============ The midbody (MB) is the narrow bridge that connects the two nascent daughter cells resulting from animal cell division. MB cleavage results in the physical separation of the cells, through a process known as abscission, and in the formation of an MB remnant (MBR) ([@bib18], [@bib29]). Increasing evidence indicates that, instead of being an abscission by-product, the MBR assumes important roles in development and differentiation ([@bib25], [@bib43], [@bib17], [@bib35]). In polarized renal epithelial cells, the MBR licenses the centrosome to assemble the primary cilium, which is a solitary plasma membrane protrusion involved in the regulation of multiple developmental signaling pathways ([@bib6]), and defines the location of the apical membrane during lumen formation ([@bib27]). The MB is continuous with the plasma membrane and consists of an electron-dense central region called the Flemming body (FB) ([@bib7]), which comprises anti-parallel microtubule bundles. Flanking the FB, the MB has two arms, containing parallel microtubule bundles, vesicles, and protein factors that bridge the two daughter cells. In principle, when severing occurs on both arms, the MBR becomes extracellular and can remain free in the extracellular milieu, stay attached to the surface of one of the daughter cells or of a neighboring cell, or be eliminated. However, severing on just one arm should lead to the MBR to be inherited by the cell on the opposite side, although this has not been well documented experimentally ([@bib40]). Given the importance of the MBR ([@bib36], [@bib12]), it seems inevitable that its fate must be tightly regulated ([@bib34], [@bib14]). However, despite the enormous effort expended on trying to understand the mechanism of the first cleavage of the MB, which marks the end of cell division, little attention has been paid to the inheritance of the connected MBRs and, thus, to the regulation of the cut of the membrane of the other MB arm. In this study, using ultra-high-resolution scanning electron microscopy (SEM), we demonstrate the existence of physical continuity between the MBR membrane and the plasma membrane of Madin-Darby canine kidney (MDCK) cells and show that only one side of the MB is cleaved in most cases. We find that, once abscission is completed, the charged multivesicular body protein (CHMP) 4C subunit of the endosomal sorting complex required for transport (ESCRT) complex delays the cleavage of the membrane of the other arm, allowing the MBR to remain on the cell surface as an organelle physically connected to the rest of the plasma membrane. The connection enables the MBR to license the centrosome for primary cilium assembly and might be also important in other processes involving the MBR. Results {#sec2} ======= MBRs of MDCK Cells Are Connected to the Plasma Membrane by a Membranous Extension {#sec2.1} --------------------------------------------------------------------------------- Epithelial MDCK cells constitute a paradigm of polarized epithelial cell ([@bib37]). Given the important role of the MBR in MDCK cells, we chose this cell line as a model cell system to study whether there was continuity between the MBR and the rest of the cell. Unlike tumor-derived cell lines ([@bib25], [@bib17]), MDCK cells have a single MBR at most ([@bib6]). Quantitative analysis indicates that \>95% of MBRs are on the apical surface ([Figures 1](#fig1){ref-type="fig"}A and 1B). Mitotic kinesin-like protein 1 (MKLP1), which localizes to the FB, is a key component of the centraspindlin complex, responsible for antiparallel microtubule bundling during anaphase, MB stability, and ESCRT recruitment during cytokinesis ([@bib30], [@bib48]). Super-resolution structured illumination microscopy showed that MBRs are formed by the FB, which was visualized by staining endogenous MKLP1, flanked by two small microtubule pools ([Figure 1](#fig1){ref-type="fig"}C). It is of note that, unlike previous stages of the abscission process ([Figure 1](#fig1){ref-type="fig"}D), the MBR did not show large microtubule bundles flanking the FB ([Figure 1](#fig1){ref-type="fig"}C).Figure 1The MBR Is on the Surface of MDCK Cells(A and B) The localization of MBRs was analyzed by confocal microscopy. Data are summarized as the mean ± SD of the percentage of surface and intracellular MBRs from three independent experiments (n = 285--296 cells) (A). (B) Representative examples of an MBR on the cell surface (left panels) or inside the cell (right panels). XY (top panels) and XZ projections (bottom panels) are shown. The arrowheads indicate the MBR.(C) An MBR as seen by structured illumination microscopy in cells stained for tyrosinated α-tubulin and the FB marker MKLP1. Dashed lines indicate cell and nuclear contours. The enlargement of the boxed region shows the characteristic ring-like structure of the FB flanked by microtubules, as seen in both XY and XZ views. The arrowhead indicates the absence of microtubules in the region adjacent to cell body.(D) Structured illumination microscopy view of two sister cells connected by an MB (left panel) and enlargement of the boxed region containing the intercellular bridge before and immediately after abscission (right panels). The dashed line delineates the cell contour. Abscission requires both the MB membrane and microtubules to be severed. Loss of tubulin staining on one side of the FB, coupled with the retraction of the structure, is a reliable indicator of the first membrane cleavage event. On the other side of the FB, however, additional techniques should be used to ascertain the integrity of the remaining membranous MB arm. SEM is a powerful tool for examining cell-surface topography. The most recent generation microscopes equipped with field emission tips and very-low-voltage (VLV) operation capabilities (incidence electron beam energy E~0~ ≤ 1 keV) allow direct, high-resolution imaging of cells on glass substrates without the need for metal coating ([@bib47]). To investigate the existence of a membranous stalk connecting the MBR membrane and the plasma membrane, we used correlative light microscopy and VLV SEM (CLEM) in cultures of cells stably expressing GFP-tubulin ([Figures S1](#mmc1){ref-type="supplementary-material"}A and S1B). Light microscopy, on the one hand, allowed selection of MBR candidate structures by the strong labeling of the FB with GFP-tubulin, discarding native MBs or MB-derived structures that still maintain microtubule bundles flanking the FB. Only the structures previously selected that had typical morphology under the SEM---a bulge flanked by two cones ([@bib12], [@bib13])---and size (1--2 μm long) were considered MBRs. Inspection of the candidate structures by VLV SEM identified unambiguously 87 *bona fide* MBRs of 117 structures analyzed. As revealed by CLEM, MBRs consist of a central "core" region, which corresponds to the bulge observed by transmission electron microscopy that contains the FB ([@bib7]), flanked by two opposed conical structures ([Figures 2](#fig2){ref-type="fig"}A and 2B). In top-view images, some of the MBRs examined have an evident membranous connection, emerging from one of the cones, with the plasma membrane ([Figure 2](#fig2){ref-type="fig"}A) that is absent from other MBRs ([Figure 2](#fig2){ref-type="fig"}B). After acquisition of a top-view image, the sample stage was tilted through 45° and rotated ([Figure S1](#mmc1){ref-type="supplementary-material"}C), making it possible to observe the MBR from different angles to examine the existence of a connection ([Figures 2](#fig2){ref-type="fig"}A and 2B, middle and right panels, and [Video S1](#mmc2){ref-type="supplementary-material"}). MBRs lacking microtubular connections with the cell body and showing membrane continuity between the tip of one of the cones and the plasma membrane were classified as connected. We reasoned that the connection should restrict MBR movement in live cells in such a way that the MBR could move, defining a funnel-shaped volume whose narrowest end coincides with the connection point ([Figure 2](#fig2){ref-type="fig"}C). To confirm the existence of the connection, we carried out time-lapse analysis of MBR movement and observed that this was the case ([Figures 2](#fig2){ref-type="fig"}D and 2E, and [Video S2](#mmc3){ref-type="supplementary-material"}). In summary, the two independent experimental approaches used support the existence of a physical connection between some MBRs and the plasma membrane.Figure 2Analysis of MBRs by Correlative Light and Scanning Electron Microscopy(A and B) Images of a connected (A) and a non-connected MBR (B) on the plasma membrane as observed by SEM in top (left panels) and side views (middle panels). Numbers indicate the angle of rotation of the sample stage. The arrowhead shows the connection point with the plasma membrane, and the arrows indicate the tip of the non-connected cones. The boxed regions were enlarged to show the tip of the cones in greater detail (right panels). The conical structures at the sides of the FB of these MBRs are of similar length and, therefore, they are shown as representative of symmetrical connected (A) and non-connected MBRs (B).(C--E) Graphical representation in top and side views of the predicted confinement volume in which MBR movement is restricted (C). (D) Kymograph showing a 3D reconstruction of the movement of an MBR, as visualized with GFP-tubulin, over time in a live cell. (E) Top and side views of the funnel-shaped confinement volume from the same MBR. A single cell was analyzed in (D and E).See also [Figure S1](#mmc1){ref-type="supplementary-material"} and [Video S1. Side Views of MBRs by SEM, Related to Figures 2A and 2B](#mmc2){ref-type="supplementary-material"}, [Video S2. MBR Movement on the Apical Surface, Related to Figures 2D and 2E](#mmc3){ref-type="supplementary-material"}. Video S1. Side Views of MBRs by SEM, Related to Figures 2A and 2BComplete rotation of a connected (left) and non-connected MBR (right). Numbers indicate the angle of rotation of the sample stage. Video S2. MBR Movement on the Apical Surface, Related to Figures 2D and 2E3D analysis of the movement of an MBR in a live cell expressing GFP-tubulin. The Membranous Connection Extends from the Tip of the Largest MBR Cone {#sec2.2} ---------------------------------------------------------------------- The MBRs identified in our analysis were quantified and classified according to the existence of a membranous connection with the plasma membrane, the symmetry between the two cones, and the size of the cone from which the connection arises ([Figure 3](#fig3){ref-type="fig"}A). Top-view SEM images showed a clear connection with the plasma membrane in 45/87 of the MBRs, whereas no discernible connection was found in 17/87 MBRs ([Figure 3](#fig3){ref-type="fig"}B). The remaining 25/87 MBRs were classified as "unclear" because their arrangement on the cell surface precludes the visualization of the possible connection in top-view images ([Figures 3](#fig3){ref-type="fig"}B and [S2](#mmc1){ref-type="supplementary-material"}A). The number of "informative" (45 + 17) top-view images of MBRs was considered sufficient to make further analysis of unclear cases unnecessary. The inclination angle formed by the long axis of the MBR and the cell surface observed for the unclear cases was more similar to that of the clearly connected MBRs than to those of the non-connected ones ([Figures S2](#mmc1){ref-type="supplementary-material"}B and S2C), suggesting the presence of a connection in most of the unclear cases. This observation implies that the observed fraction of connected MBRs with respect to the total "informative" cases (45/62, or 72.6%) is likely an underestimate of the genuine fraction of connected MBRs.Figure 3Most MBRs Remain Connected to the Plasma Membrane(A) Representative examples of MBR morphologies others than those shown in [Figure 2](#fig2){ref-type="fig"}. Arrowheads indicate connection points.(B) Sankey diagram showing the results of our MBR morphology analysis. Large and small sized numbers indicate the population size of each class and of the subclasses, respectively. Only the structures with strong labeling of the FB with GFP-tubulin that lacked microtubular connections with the cell body on both flanks and that had typical morphology and size were considered MBRs.(C) Quantification of FB width (n = 87) and total MBR length of connected (n = 38) and non-connected (n = 17) structures.(D) Length of the two sides flanking the FB in connected MBRs (n = 38). Black bars indicate median values.See also [Figure S2](#mmc1){ref-type="supplementary-material"} A morphological feature of MBRs is the apparent degeneration of one of the cones. One cone tends to have a defined form and size, whereas the other is frequently shorter and rounder, giving rise to an asymmetrical MBR ([Figure 3](#fig3){ref-type="fig"}A). It is of note that the connection arose from the larger cone in most (19/22, or 86.4%) of the connected MBRs with asymmetrical cones ([Figure 3](#fig3){ref-type="fig"}B). To characterize the MBR, we measured its dimensions using top-view SEM images. Connected MBRs were longer and more variable in length than the non-connected ones, the connected side being longer than the opposite one ([Figures 3](#fig3){ref-type="fig"}C and 3D). Independent measurements of the length based on MBR movement yielded similar values, supporting the validity of this approach ([Figures S2](#mmc1){ref-type="supplementary-material"}D--S2F). In conclusion, the analyses presented so far indicate that MBRs display a number of prevalent structural features, the most common one being the presence of a membranous stalk presumably derived from the uncleaved arm of the bridge, which most often coincides with the largest cone, physically connecting the MBR membrane to the plasma membrane. The ESCRT Machinery Concentrates at the Connection between the MBR and the Plasma Membrane {#sec2.3} ------------------------------------------------------------------------------------------ The final steps of the abscission process are carried out by the ESCRT machinery ([@bib11], [@bib32], [@bib41]), which progressively accumulates into rings at both sides of the FB ([@bib16]). ESCRT-III assembles spiral polymers whose diameter decreases as they grow away from the FB, constricting the MB to the limit allowed by the microtubules inside. After microtubule clearance, the ESCRT polymer remodels generating a second ESCRT pool that is positioned at the future cleavage site ([@bib15], [@bib22]). To investigate the involvement of ESCRT-III proteins in the cleavage of the membrane of the other MB arm, we expressed GFP-fused forms of the human ESCRT proteins CHMP4B (GFP-L-CHMP4B) and CHMP4C (GFP-L-CHMP4C) and analyzed their localization before and after the end of the abscission process. These proteins, in which GFP is separated from CHMP4C and CHMP4B by a 25-nm-long flexible linker, were previously shown to have the expected localization at the MB, and their expression did not delay MB abscission time ([@bib45], [@bib38]). Both proteins first accumulated in ring-like structures on both sides of the FB and then polymerized toward the abscission site, resulting in the appearance of cone-shaped staining in one of the MB arms. Once microtubules were cleared from this arm, membrane cleavage and, consequently, daughter cell separation occurred. After abscission, CHMP4B, CHMP4C, and microtubules followed essentially the same sequence of events on the other side of the FB, generating an MBR ([Figures 4](#fig4){ref-type="fig"}A, 4B, and [S3](#mmc1){ref-type="supplementary-material"}A). The same was observed in a panel of endogenous ESCRT proteins ([Figure S3](#mmc1){ref-type="supplementary-material"}B) and was confirmed for CHMP4B by videomicroscopic analysis of cells co-expressing GFP-L-CHMP4B and Cherry-tubulin ([Figure S3](#mmc1){ref-type="supplementary-material"}C and [Video S3](#mmc4){ref-type="supplementary-material"}). All MBRs contained ESCRT proteins ([Figure S3](#mmc1){ref-type="supplementary-material"}D), but the pattern of distribution was not the same in all MBRs. The MBRs that presented a similar pattern on both sides of the FB, mainly with staining only on the FB rims, were classified as "even" MBRs, whereas those that, in addition to the FB rims, had a second ESCRT pool in only one side of the FB were categorized as "uneven" MBRs. The second ESCRT pool in the uneven MBRs adopted the form of a cone, filament, or dot ([Figures 4](#fig4){ref-type="fig"}A, 4B, [S3](#mmc1){ref-type="supplementary-material"}C, and S3E, and [Video 3](#mmc4){ref-type="supplementary-material"}). Quantitative analysis revealed that most MBRs display uneven ESCRT distribution ([Figure 4](#fig4){ref-type="fig"}C). It is of particular note that the MBR side with the extra ESCRT pool coincides with that having the membranous stalk. This pool is present in a region of the connection proximal to the plasma membrane, as determined by CLEM of cells stably expressing Cherry-tubulin and GFP-L-CHMP4C or GFP-L-CHMP4B ([Figures 5](#fig5){ref-type="fig"}A--5E, and [Videos S4](#mmc5){ref-type="supplementary-material"} and [S5](#mmc6){ref-type="supplementary-material"}). Supporting this localization, time-lapse analysis of MBR movement showed that the pool remained immobile, as may be seen in the projected kymograph, whereas the distal pool, which corresponds to the FB rims, moved drawing a circle around it ([Figure 5](#fig5){ref-type="fig"}F and [Video S6](#mmc1){ref-type="supplementary-material"}).Figure 4The ESCRT Machinery Shows an Uneven Distribution along Most MBRs(A and B) Distribution of GFP-L-CHMP4B (GFP-L-4B) (A) and GFP-L-CHMP4C (GFP-L-4C) (B) at the MBR. XY and XZ views of MBRs with uneven (top panels) and even (bottom panels) distribution of these markers. The arrow and the arrowheads in (A) and (B) indicate the FB and the MBR tips, respectively.(C) Histogram showing the percentage of MBRs with uneven and even distribution for GFP-L-CHMP4B, GFP-L-CHMP4C, and a panel of endogenous ESCRT markers.Data are summarized as the mean ± SD from three independent experiments (n = 29--93). See also [Figure S3](#mmc1){ref-type="supplementary-material"} and [Video S3](#mmc4){ref-type="supplementary-material"}.Figure 5CHMP4C and CHMP4B Are Present at the Membranous Connection(A--D) CLEM images showing the presence of GFP-L-CHMP4C (A and B) and GFP-L-CHMP4B (C and D) at the connection of the MBR with the plasma membrane. (A and C) top-view images of connected MBRs acquired by SEM (top) and confocal microscopy (bottom). (B and D) Side-view SEM images (left) and matching confocal images obtained by 3D reconstruction (right).(E) Quantification of GFP-L-4B and GFP-L-4C distribution in connected MBRs as observed by CLEM (n = 16 and 18, respectively).(F) Tracks of GFP-L-CHMP4C and Cherry-tubulin movement of an MBR in a live cell. (i) GFP-L-CHMP4C and Cherry-tubulin distribution in an MBR; (ii) image of the distribution GFP-L-CHMP4C using the indicated depth-color scale; (iii and iv) 3D reconstruction of the movement followed by the MBR over a 3-min period.See also [Video S4. CLEM Images Showing the Presence of GFP-L-CHMP4C in the Connection of the MBR with the Plasma Membrane, Related to Figure 5B](#mmc5){ref-type="supplementary-material"}, [Video S5. CLEM Images Showing the Presence of GFP-L-CHMP4B in the Connection of the MBR with the Plasma Membrane. Related to Figure 5D](#mmc6){ref-type="supplementary-material"}, [Video S6. GFP-L-CHMP4C and Cherry-Tubulin Distribution in a Moving MBR, Related to Figure 5F](#mmc7){ref-type="supplementary-material"}. Video S3. Dynamics of GFP-L-CHMP4B during and after Abscission, Related to Figure 4CGFP-L-CHMP4B and Cherry-tubulin was tracked in cells during cytokinesis. Video S4. CLEM Images Showing the Presence of GFP-L-CHMP4C in the Connection of the MBR with the Plasma Membrane, Related to Figure 5BSide view SEM images (left) and matching confocal images obtained by 3D reconstruction (right). The angle of rotation of the sample stage is indicated. Video S5. CLEM Images Showing the Presence of GFP-L-CHMP4B in the Connection of the MBR with the Plasma Membrane. Related to Figure 5DSide view SEM images (left) and matching confocal images obtained by 3D reconstruction (right). The angle of rotation of the sample stage is indicated. Video S6. GFP-L-CHMP4C and Cherry-Tubulin Distribution in a Moving MBR, Related to Figure 5F(Left) GFP-L-CHMP4C and Cherry-tubulin fluorescence. (Right) The GFP-L-CHMP4C signal was pseudocolored using the indicated depth-color scale. In summary, ESCRT proteins localize to the membranous stalk that connects the MBR to the plasma membrane and have a similar distribution to that found in pre-abscission stages right before the MB arm is first cleaved ([@bib22]). Since the presence of an ESCRT pool distant from that surrounding the FB has been associated with the last stage of membrane cleavage ([@bib22]), we proceeded to analyze how the cleavage of the connection is prevented. CHMP4C Depletion Reduces the Percentage of Cells with an MBR and Impairs Primary Ciliogenesis {#sec2.4} --------------------------------------------------------------------------------------------- The abscission checkpoint delays abscission by regulating the ESCRT machinery in the case of mitotic problems, such as persisting chromatin within the bridge, incomplete nuclear pore reformation defects, or tension in the bridge produced by opposite pulling forces from the daughter cells ([@bib1], [@bib26], [@bib8]). The activation of the abscission checkpoint retards abscission by promoting the phosphorylation of the ESCRT-III subunit CHMP4C by the kinase Aurora B, Ser210 being the major phospho-acceptor residue ([@bib10], [@bib9]). To investigate the involvement of this mechanism in the regulation of the second cleavage of the MB, we used specific siRNA (siCHMP4C) to knock down (KD) CHMP4C expression ([Figures 6](#fig6){ref-type="fig"}A and 6B). This siRNA targets the sequence in dog CHMP4C equivalent to that of the siRNA previously used for human CHMP4C in HeLa cells ([@bib10]). As a control, we observed that CHMP4C KD accelerated abscission ([Figure 6](#fig6){ref-type="fig"}C) without affecting the number of dividing cells ([Figure S4](#mmc1){ref-type="supplementary-material"}A), as has been noted in other cell lines ([@bib10], [@bib38], [@bib8]). The acceleration of abscission is statistically significant but modest. We interpret that this effect is probably to defective activation of the abscission checkpoint and speculate that the frequency of events that engage the abscission checkpoint, such as high-membrane tension and persistent chromatin bridges, may be reduced in MDCK cells. It is of note that the percentage of cells with an MBR was much lower in CHMP4C KD cells than in the control cells ([Figure 6](#fig6){ref-type="fig"}D). Videomicroscopic analysis showed MBR shedding in 13 of 26 CHMP4C KD cells examined, in contrast to control cells in which it was observed in only 3 of the 22 cells examined ([Video S7](#mmc8){ref-type="supplementary-material"}), implying that the reduction in the percentage of cells with an MBR is a consequence of de-repressed MBR cleavage. As a control, we observed that CHMP2A KD did not affect the percentage of cells with an MBR ([Figures 6](#fig6){ref-type="fig"}D, [S4](#mmc1){ref-type="supplementary-material"}B, and S4C), suggesting that there is a specific effect of the regulatory CHMP4C subunit, but not of any ESCRT component, in the regulation of MBR inheritance. The effect of CHMP2A depletion is consistent with previous observations showing that it dramatically delays but does not inhibit abscission ([@bib50]) indicating that CHM2A-depleted cells eventually divide and form MBRs.Figure 6CHMP4C Is Required for MBR Inheritance(A) Representative immunoblot showing the effect of siCHMP4C on endogenous CHMP4C levels.(B) Quantification of CHMP4C KD. The histogram represents the levels of CHMP4C in siCHMP4C-transfected cells relative to that of cells transfected with control siNT 48 h after the first siRNA transfection.(C) The time between the formation of the MB and abscission was measured in control siNT (gray points) and siRNA-mediated CHMP4C KD cells (red points) (n = 27--159 in control cells; n = 8--95 in CHMP4C KD cells). Black bars represent median values.(D) The percentage of MBRs ± SD in CHMP4C KD cells (red bars) expressing or not the indicated exogenous CHMP4C proteins and of CHMP2A KD cells (blue bar) was calculated relative to that of control cells (siNT, gray bars) 48 h after the first siRNA transfection (n = 2,714--3,447 cells for control and n = 800--1,463 for KD cells). Three independent experiments were performed. Probabilities are those associated unpaired two-tailed Student\'s t test (B and D) and Mann-Whitney test (C).See also [Figure S4](#mmc1){ref-type="supplementary-material"} and [Video S7](#mmc8){ref-type="supplementary-material"}. Video S7. CHMP4C KD Produces Shedding of the MBR. Related to Figure 6Control and CHMP4C KD cells expressing GFP-tubulin were recorded before and after abscission. The CHMP4C mutants S210A and A232T, which is a CHMP4C allele associated with increased susceptibility to cancer, are unable to replace endogenous CHMP4C in the regulation of the first cut of the bridge membrane ([@bib10], [@bib38]). The effect of CHMP4C knockdown on the second cut was rescued by the exogenous expression of siCHMP4C-resistant forms of GFP fusions of wild-type but not of the S210A and A232T CHMP4C mutants ([Figures 6](#fig6){ref-type="fig"}D and [S4](#mmc1){ref-type="supplementary-material"}D--S4F). Note that the exogenous wild-type and mutant CHMP4C proteins were expressed at similar levels and, therefore, the failure of the mutants to restore normal values was not due to poor expression. The percentage of MBRs positive for the mutants and their even or uneven distribution within the MBR were similar to those of the wild-type CHMP4C protein ([Figures S4](#mmc1){ref-type="supplementary-material"}G and S4H). The total number of cells per field in the case of the CHMP4C mutants was similar to that expressing exogenous wild-type CHMP4C ([Figure S4](#mmc1){ref-type="supplementary-material"}I). As a control, we observed that the number of cells connected by an MB decreased in siCHMP4C-treated cells and that this effect was corrected by the intact protein but not by the CHMP4C mutants ([Figure S4](#mmc1){ref-type="supplementary-material"}J). The results illustrated in [Figure 6](#fig6){ref-type="fig"}C for the second cut of the membrane of the bridge are consistent with those reported for CHMP4C in the control of the first cut by the abscission checkpoint mechanism ([@bib10], [@bib9]). They suggest that CHMP4C has a similar role in the second cut. Most mammalian cells have a primary cilium that projects from their surface as a single appendage ([@bib46]). The primary cilium orchestrates important signaling pathways involved in development and cell proliferation, differentiation, survival, and migration ([@bib28], [@bib21]). Primary cilium formation proceeds by two distinct pathways, depending on the position of the centrosome in the cell ([@bib44]). In cells with centrosome near the nucleus, as in NIH 3T3 fibroblasts, ciliogenesis starts intracellularly and finishes at the plasma membrane, whereas when the centrosome is close to the plasma membrane, as in polarized epithelial MDCK cells, the process takes place entirely at the plasma membrane. The first route is referred to as the intracellular or "classic" pathway; the second route is known as the alternative pathway ([@bib2], [@bib31], [@bib5]). Since we described previously that the MBR plays an important role in preparing the centrosome for primary ciliogenesis in MDCK cells ([@bib6]), we examined the effect of CHMP4C KD on this process. We noted a dramatic drop in the percentage of ciliated cells ([Figures 7](#fig7){ref-type="fig"}A and 7B), which is consistent with the loss of MBRs in CHMP4C-deficient cells ([Figure 6](#fig6){ref-type="fig"}D). Confirming the requirement for CHMP4C, the exogenous expression of siCHMP4C-resistant wild-type CHMP4C rescued primary cilium formation ([Figures 7](#fig7){ref-type="fig"}A and 7B). The results in [Figure 7](#fig7){ref-type="fig"} are in agreement with a previous report showing that the physical removal of the MBR greatly reduces primary ciliogenesis ([@bib6]) and further highlights the importance of the MBR in this process by providing a genetic evidence of the requirement for MBR in primary cilium formation by polarized epithelial cells.Figure 7CHMP4C Is Required for Primary Cilium Formation(A and B) Effect of CHMP4C KD on the frequency of ciliated cells. The percentage of primary cilia in cells expressing shCHMP4C and GFP in the absence or presence of exogenous human CHMP4C was expressed relative to that of control cells, which expressed only GFP, 72 h after shRNA transfection (n = 77--88 for control; n = 70--161 for CHMP4C KD cells). The mean ± SD from three independent experiments is shown (A). Probabilities are those associated with unpaired two-tailed Student\'s t test. (B) Representative fields of the cells analyzed in (A). Acetylated tubulin staining was used to visualize the primary cilium.See also [Figure S4](#mmc1){ref-type="supplementary-material"}. Discussion {#sec3} ========== Although the midbody was first described more than 125 years ago, the discovery of its role in abscission is relatively recent, and even more so is the evidence of important post-mitotic roles for the MBR ([@bib12]). Accumulation of MBRs has been associated with increased cell reprogramming efficiency of stem cells and *in vitro* tumorigenicity of cancer cells ([@bib25], [@bib17], [@bib35]). In polarized epithelial cells, the MBR meets the centrosome at the center of the apical membrane and enables the centrosome for primary cilium formation ([@bib6]). Using CLEM, we identified a membranous stalk in polarized epithelial MDCK cells that physically connects the MBR membrane and the plasma membranes of most MBR-containing cells. The stalk is derived from the uncleaved arm of the bridge and contains ESCRT machinery, including the regulatory subunit CHMP4C. CHMP4C silencing causes the loss of the MBR and, consistent with its role in primary cilium formation, a dramatic reduction in the percentage of ciliated cells. These results indicate that an MBR physically connected to the plasma membrane by a membranous stalk, whose integrity is regulated by CHMP4C, is the form of MBR used by MDCK cells to license primary ciliogenesis. We first identified candidate MBR structures from the presence of GFP-tubulin in the MB core and its absence from the two MB arms. The selected structures were analyzed in a state-of-the-art, VLV SEM using samples that were prepared by a gentle procedure ([@bib24]) omitting conductive coating. This equipment revealed the subnanometric topography of MBRs, which enabled structures without the typical MBR morphology to be discounted. Using this approach, we visualized a membranous stalk between the MBR and the plasma membrane in a large proportion of MBRs. However, such a connection was not observed in a previous CLEM study ([@bib13]) that combined phase-contrast microscopy to identify MBR candidates, sample preparation by standard procedures, and analysis under conventional SEM equipment ([@bib19]). This morphological discrepancy might be due to the different cell lines analyzed by SEM---HeLa cells in [@bib13] and MDCK cells in ours---or to the distinct protocols for sample preparation and the different type of SEM equipment used. In addition to detecting the connection, our CLEM analysis revealed that one of the MBR cones is larger than the other, likely because the shorter one results from the degeneration of the cone on the side where abscission occurs. Consistent with this possibility, we observed that the connecting stalk most often arises from the largest cone of the MBR. The degeneration of the connected cone is probably prevented by the presence of both conical ESCRT polymers ([@bib42]), which could act as a scaffold for the structure, and the microtubules filling the cone. The majority of MBRs were connected (72.6%), but we detected a smaller fraction (27.4%) that were non-connected. At least part of the latter fraction may be a consequence of artifacts arising when the samples were prepared, but it is plausible that some MBRs exist on the cell surface in a non-connected form. The treatment of a variety of cell lines with EDTA reduces the number of MBRs on the cell surface ([@bib13]). This observation was interpreted as meaning that MBRs are not physically connected to the plasma membrane but, instead, adhered to the cell surface through an interaction with an unknown Ca^2+^/Mg^2+^ receptor. In the case of MDCK cells, the number of MBRs decreased upon EDTA treatment by approximately 50%. It is plausible that this treatment affects only those non-connected MBRs, the number of which might be dependent on the cell growth conditions ([@bib6]). Another possibility is that Ca^2+^/Mg^2+^ chelation alters the stalk of the connected MBRs, somehow affecting its integrity, leading to MBR shedding. The loss of tubulin staining on one side of the FB together with the retraction of the structure indicates the first cut of the membrane of intercellular bridge ([@bib16], [@bib26]). However, it is important to note that the loss of tubulin staining on the other side should not be assumed to represent the second cut of the bridge membrane, since CLEM revealed the existence of a membranous connection with the plasma membrane in MBRs with no flanking microtubules. In addition, the use of phase-contrast microscopy is not adequate for distinguishing MBRs that are attached to the plasma membrane from those that are physically connected by a membranous stalk because the connection is very small ([@bib13]). Therefore, our work is consistent with previous studies ([@bib16], [@bib26]) but differs in that we have investigated whether the second cut of the bridge membrane takes place and have used CLEM as a highly reliable technical approach to do precisely so. We observed that most MBRs contained ESCRT polymers only on the side corresponding to the largest cone, similar to those present just before the first cleavage of the MB membrane. We mapped the ESCRT pool at the membranous connection between the MBR and the plasma membrane by CLEM and confirmed the localization by analyzing the MBR motion. This location of ESCRT proteins is consistent with the presence of helical filaments in the uncleaved MB arm, as observed by soft X-ray cryotomography ([@bib42]). This pool contains CHMP4C, which is a crucial component of the checkpoint mechanism that delays abscission when mitotic problems occur. In those cases, the knockdown of CHMP4C accelerates abscission and only the expression of wild-type CHMP4C but not of the CHMP4C S210A or A232T mutants can substitute the endogenous protein to delay membrane cleavage. Since the number of cells with an MBR was greatly diminished in CHMP4C KD cells and the effect was corrected by expression of intact CHMP4C but not by CHMP4C mutants, we propose that, similar to its role in the abscission checkpoint ([@bib10], [@bib9]), CHMP4C allows MBRs to remain connected to the plasma membrane by delaying the cleavage of the connection. Although CHMP4C regulates the abscission checkpoint only when mitotic problems arise during cytokinesis, it appears that CHMP4C acts constitutively on the second cut of the bridge to preserve the MBR connected to the plasma membrane. We currently do not understand how this asymmetric regulation of CHMP4C in one or other of the cleavage sites takes place. It is plausible that Aurora B or Alix, or other mechanisms, for instance, selective actin clearance by actin oxidation, could differentially influence the local activity of CHMP4C ([@bib20], [@bib3]). In addition, since CHMP4C activity is known to be regulated by phosphorylation, with at least three phosphorylation sites in its C-terminal region that are important for CHMP4C regulation ([@bib9], [@bib10]), we speculate that the existence of different phosphorylated forms of CHMP4C ([@bib49]) could be responsible for the selective regulation of the first and second cut. The model of primary ciliogenesis based on MBR-mediated centrosome licensing that we proposed implies that only cells with an MBR can assemble a cilium ([@bib6]). Therefore, MBR retention is important at the cell population level to guarantee a high percentage of MBR-bearing cells and at the single cell level to permit the centrosome to form a primary cilium. The existence of the physical connection in MDCK cells might facilitate, at the cell population level, the retention of the MBR to be transmitted to the next generation of cells, as seen by videomicroscopic analysis ([@bib6]), and, at the single cell level, the directional movement of the MBR to meet the centrosome by direct anchoring to the cytoskeleton. In addition, the continuity of the MBR with the rest of the plasma membrane might enable the transfer of materials from the MBR to the centrosome in order to assemble the primary cilium. There are conflicting results about the requirement of ESCRT proteins for primary ciliogenesis in cells such as NIH 3T3 cells that, unlike MDCK cells, use the intracellular route of ciliogenesis. Primary cilium formation appears to be independent of ESCRT proteins in one of the studies ([@bib33]) but not in the other ([@bib23]). We cannot discount the contribution of indirect effects, because the ESCRT-III machinery participates in numerous cellular functions, but, despite this, we propose that a functional consequence of the loss of the connection caused by CHMP4C silencing is the impairment of primary ciliogenesis in MDCK cells and that, therefore, the connection is required for the alternative route of primary ciliogenesis. In conclusion, our study reveals that the majority of MBRs inherited in MDCK cells are physically connected to the plasma membrane through a membranous stalk derived from the uncleaved arm of the cytokinetic bridge. The ESCRT subunit CHMP4C controls the integrity of this arm to ensure the continuity between the MBR membrane and the plasma membrane and, in this way, the MBR facilitates primary cilium formation. Limitations of the Study {#sec3.1} ------------------------ The MDCK cell line is considered to be a *bona fide* representative of polarized epithelial cells ([@bib37]). However, since we have not confirmed our results in primary polarized epithelial cells, we cannot rule out that our findings are totally valid for normal epithelial cells. A second limitation is that, although we used a gentle protocol of sample preparation for CLEM, some of the connections between the MBR and the plasma membrane might have broken during the procedure, leading to the underestimation of the percentage of connected MBRs. Another limitation concerns the dynamics of ESCRT III CHMP proteins. These dynamics have been reconstructed from images obtained from fixed cells at different times of cytokinesis ([Figures 4](#fig4){ref-type="fig"}A, 4B, [S3](#mmc1){ref-type="supplementary-material"}A, S3B, and S3E). These include a number of exogenous ([Figures 4](#fig4){ref-type="fig"}A, [S3](#mmc1){ref-type="supplementary-material"}A, and S3E and endogenous ([Figure S4](#mmc1){ref-type="supplementary-material"}B) CHMP proteins. The proposed dynamics were confirmed by videomicroscopy only in the case of exogenous CHMP4B ([Figure S3](#mmc1){ref-type="supplementary-material"}C and [Video 3](#mmc4){ref-type="supplementary-material"}). Finally, owing to technical constraints---the time required to efficiently knockdown CHMP4C expression and the time needed to have a tight monolayer with a high percentage of ciliated cells---the experiment in [Figure 7](#fig7){ref-type="fig"} was carried out under suboptimal conditions for primary cilium formation. Resource Availability {#sec3.2} --------------------- ### Lead Contact {#sec3.2.1} Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Miguel A. Alonso (<maalonsocbm.csic.es>). ### Materials Availability {#sec3.2.2} All unique/stable reagents generated in this study are available from the Lead Contact with a completed Material Transfer Agreement. ### Data and Code Availability {#sec3.2.3} All data are available in the main text or the [Supplemental Information](#appsec1){ref-type="fn"}. Methods {#sec4} ======= All methods can be found in the accompanying [Transparent Methods supplemental file](#mmc1){ref-type="supplementary-material"}. Supplemental Information {#appsec2} ======================== Document S1. Transparent Methods and Figures S1--S4 The expert technical advice of the Optical and Confocal Microscopy Facility of CBMSO is gratefully acknowledged. We thank Dr. Phil Mason for revising the English language of the manuscript, Laura Fernandez-Martín for invaluable technical help, and David Esteban-Mendoza for his contribution to optimization of the SEM protocol. This work was supported by a grant (PGC2018-095643-B-I00) to M.A.A. from the Spanish Ministerio de Ciencia e Innovación (MICIN), 10.13039/501100011033Agencia Estatal de Investigación, y Fondo Europeo de Desarrollo Regional, European Union (MICIN/AEI/FEDER, UE), and by 10.13039/100010269Wellcome Trust funding (WT102871MA) to J.M.-S. We also acknowledge the Micro and Nanofabrication Laboratory of the Instituto de Micro y Nanotecnología (MiNa), which is funded by the Comunidad de Madrid (S2018/NMT-4291 TEC2SPACE), MICIN (project CSIC13-4E-1794), and EU (FEDER, FSE), for invaluable help on SEM. A contract (FPU14/00295) and a short-term fellowship from EMBO to J.C.-A. are also acknowledged. Author Contributions {#sec6} ==================== J.C.-A. performed most of the experiments and designed some of the experiments; M.U.G. and A.S.P. collaborated in the SEM analysis with equal contribution; L.N.V. and J.B.A.S collaborated in some of the experiments done with ESCRT proteins; D.G.M. helped on the calculation of the MBR size; L.L-de-H., A.R.-R, L.R., M.B.-R., and J.F-B. participated in some of the experiments; I.C. gave advice, discussed the results in the manuscript, and supervised the work; J.M.-S. provided materials, supervised part of the work done with ESCRT proteins, and hosted J.C.-A. during a short-term stay; M.A.A. designed and supervised the work and wrote the manuscript with input from all authors, especially from J.C.-A. Declaration of Interests {#sec7} ======================== The authors declare no competing interests. Supplemental Information can be found online at <https://doi.org/10.1016/j.isci.2020.101244>. [^1]: These authors contributed equally [^2]: Lead Contact

Introduction {#cesec10} ============ Developmental venous anomaly (DVA), previously known as cerebral venous angioma, is a vascular malformation thought to be a benign embryologic variant. Its incidence has been reported as 2.5% at autopsy. At CT and MRI, the diagnosis of DVA is suggested with visualization of a draining vein. Although usually incidentally discovered on enhanced CT or MRI of the brain, a DVA can present with headache, focal neurological deficits or bleeding. However, a nonhemorrhagic brain infarction is a rare complication, with only about 10 previous cases reported in the literature \[[@bib1]\]. We describe a patient who presented with focal neurological deficits and paresthesia due to an infarct associated with a developmental venous anomaly and a thrombosed draining vein. Case Report {#cesec20} =========== A 22-year-old right-handed Caucasian woman awoke the morning of admission with slurred speech, numbness and weakness of the left side of her face, left arm and leg. She had neither headache nor visual disturbance. She was in good health prior to this event and had no past medical history. Her family history was unremarkable. Her only medication was an oral contraceptive. The symptoms lasted for about two hours and then quickly resolved. On examination, she was afebrile with normal vital signs. She had no signs of trauma and was normocephalic. She had equal and reactive pupils and a normal cranial nerve exam. Her speech was normal and there was no facial weakness at the time of examination. She had weakness of the left upper extremity with a weak left hand grasp and decreased biceps and triceps strength. Her left leg had slight decrease in strength, but she was able to keep it up against resistance. She was able to ambulate without a limp and could toe walk and heel walk. Routine laboratory tests, including hematology and coagulation studies, were normal. Antithrombin III levels were normal. Lupus anticoagulant and cardiolipin antibodies were absent. She had no factor V Leiden mutation or prothrombin gene mutation. A CT of the brain before and after contrast administration showed an area of decreased attenuation within the genu and posterior limb of the right internal capsule, and a prominent enhancing vessel within the affected region of the brain. These findings represent a DVA with venous thrombosis and infarction ([Fig. 1A](#fig1a){ref-type="fig"}). A subsequent brain MRI showed an acute infarction with abnormal signal on the T2 ([Fig. 1B](#fig1b){ref-type="fig"}), FLAIR, diffusion ([Fig. 1C](#fig1c){ref-type="fig"}), and gadolinium enhanced T1 sequences within this same area. A curvilinear, prominent vessel coursed through the infarction, representing the thrombosed DVA ([Fig. 1D](#fig1d){ref-type="fig"}). Intracranial MR angiography was unremarkable, showing no aneurysm or stenosis of the cerebral arteries or vertebrobasilar system. She was discharged to home after being admitted overnight for observation. Discussion {#cesec30} ========== Developmental venous anomalies, previously termed venous angiomas, represent the most common cerebral vascular anomaly. The term "developmental venous anomaly" was coined by Lasjaunias et al. \[[@bib2]\] after suggesting that venous angiomas are actually embryologic variants of venous drainage instead of true vascular anomalies. These lesions are thought to represent an arrest of venous development after arterial development is nearly complete or a thrombosis of the developing venous drainage in a specific region. The process results in the retention of primitive, embryologic medullary veins that drain into a single, large draining vein, the DVA \[[@bib3]\]. The region drained by a DVA has no other normal venous drainage. DVAs lack an arterial component and have unaffected intervening neural parenchyma, which distinguishes these lesions from arteriovenous malformations \[[@bib4]\]. About one third of these lesions are located in the cerebellum and in the brainstem; the remaining are supratentorial. DVAs have a characteristic appearance on angiography, CT and MRI. In the late venous phase of angiography, several radially arranged dilated veins converge in a central large caliber central vein to form a caput medusae image \[[@bib1]\]. On contrast-enhanced CT, the draining vein of a venous angioma appears as an enhancing linear or curvilinear structure, typically coursing from the deep white matter to a cortical vein, a deep vein, or to a dural sinus \[[@bib3]\]. DVAs demonstrate low signal intensity on T1-weighted images and high signal intensity on T2-weighted images. Following gadolinium administration there is marked enhancement of the main collecting vein and peripheral radial veins \[[@bib1]\]. Venous angiomas are often asymptomatic and therefore are often found incidentally. The most frequent signs include headache, seizure, focal neurological deficits and dizziness. Thrombosis of the draining vein of the malformation is an extremely rare complication leading to nonhemorrhagic venous infarction. To our knowledge, only ten previous cases of a nonhemorrhagic brain infarction due to a DVA have been reported in the literature \[[@bib1]\]. Hammoud et al. \[[@bib5]\] believe that the same predisposing factors for dural sinus thrombosis in the central nervous system apply for the thrombotic event associated with a DVA, including oral contraceptive use and hereditary hypercoaguable states such as protein C or S deficiency, antithrombin deficiency, or factor V Leiden mutation. Our patient had a negative hypercoagulability screen, but used oral contraceptives, which may have contributed to her thrombotic event involving the DVA. Developmental venous anomalies are usually thought of as incidental findings on brain imaging. However, infarctions are a known complication of developmental venous anomalies. Non-hemorrhagic infarction is a potential complication of a DVA that should be considered in the differential diagnosis of a bland infarct, where a thrombosed draining vein may be identified. Published: December 22, 2007 {#fig1a} {#fig1b} {#fig1c} {#fig1d} [^1]: Brian J. Parker, M.D. is in the Department of Radiology, St. Joseph Mercy Oakland Hospital, Pontiac, Michigan 48341, USA. [^2]: Brian J. Sabb, M.D. is in the Department of Radiology, University of Michigan, Ann Arbor, MI, USA.

The review by Ferrara et al. ([@B8]) provides a summary of a fascinating method of researching sleep, which is gaining more and more importance and momentum in the literature. Intra-cranial-EEG recordings provide us with the unique opportunity to delve more deeply into the human brain instead of just "scratching on the surface." The classical Rechtschaffen and Kales classification of sleep has been a reliable and faithful way to view sleep for more than 40 years. However, criticisms have arisen (Himanen and Hasan, [@B10]) and it is perhaps time to advance to a new age and not see sleep as a whole-brain state anymore. For example, during sleepwalking a dissociation of activity in different brain regions can be observed, with strongly enhanced activity in the cerebellum and cingulate cortex while large areas of frontal and parietal association cortices remain deactivated (Bassetti et al., [@B4]). Another example is lucid dreaming, in which prefrontal and medial parietal regions become activated during otherwise normal REM sleep (Voss et al., [@B17]; Dresler et al., [@B6]). However, as Ferrara et al. nicely summarized in their review, intra-cranial-EEG recordings suggest that also normal sleep is a rather local phenomenon. Slow oscillations (Nir et al., [@B12]) as well as spindles (Andrillon et al., [@B2]) seem to occur more locally and not globally. Further, different brain areas, e.g., hippocampus and cortex seem to be able to display features of different sleep-states -- REM and NREM -- at the same time (Moroni et al., [@B11]). The review also highlights the evidence regarding sleep features important for memory consolidation. They present cumulative data suggesting that perhaps during hippocampal sharp wave ripples (hSWR) a replay of the memory trace occurs, which is then "transferred" into the cortex via sleep spindles and perhaps slow oscillations. At this point it is important to note that for one most evidence regarding this theory is of correlative nature and therefore cannot conclude causality. Up until now only two studies in rodents could show more than correlational evidence by actively disrupting SWR and therefore causing decreased memory consolidation (Girardeau et al., [@B9]; Ego-Stengel and Wilson, [@B7]). A more recent study in humans showed enhanced memory performance by stimulation of the entorhinal region during memory encoding (Suthana et al., [@B16]). Perhaps future studies could combine these two approaches. Another important caveat is that replay coinciding with hSWR is seen in wake-rest as well as sleep (Axmacher et al., [@B3]). In Axmachers study the highest incidence of ripples occurred during rest and the ripples during rest actually correlated with memory performance. However, studies in rodents reported that cortical replay of memory related neural-patterns only occurred during sleep-SWR (Peyrache et al., [@B14]). Perhaps hippocampal replay and thus consolidation occurs during wake and sleep-SWR, but this replay can only be coordinated with cortical replay and thus cortical consolidation during sleep, which provides the required network properties. Connectivity analyses in human fMRI studies could show general decreased cortical--cortical connectivity during slow wave sleep, but increased cortical--cortical connectivity during stage 2 light sleep (Spoormaker et al., [@B15]). And even more compelling, the same group could show increased hippocampal-cortical connectivity during sleep spindles (Andrade et al., [@B1]). Surprisingly a very recent study in rodents presented that during sleep spindles the cortex seems to be functionally deafferented from its hippocampal inputs (Peyrache et al., [@B13]). Further, Dang-Vu et al. ([@B5]) provided evidence that humans who generate more sleep spindles exhibit a higher tolerance for noise during sleep and they suggest that the function of sleep spindles is to enable consolidation by blocking interfering processes. Perhaps the increased connectivity found by Andrade et al. was due to the close temporal coupling of sleep spindles and hSWR and not because of the properties of the spindles themselves. Hopefully future studies investigated sleep via intra-cranial-EEG recordings in humans will provide a better understanding of these issues. However, we should never forget that these recordings are always made in patients with pharmacologically resistant epilepsy undergoing evaluation for surgery and not healthy subjects. To arrive at the point of needing such an evaluation the patients have suffered under sever forms of epilepsy for many years. This makes it likely that some damage to the neural circuits has occurred and perhaps compensatory mechanisms have been activated. Further, the actual cause of the epilepsy can also impact the findings. In summary, studies investigating memory and sleep via intra-cranial-EEG recording show much promise to provide us with invaluable insights into physiological processes and will perhaps carry us into the next generation of sleep research perceiving sleep as a more complex and local instead of a global phenomenon. [^1]: ^†^Current address: Lisa Genzel, Centre for Cognitive and Neural Systems, University of Edinburgh, 1 George Square, EH8 9JZ Edinburgh, UK. [^2]: This article was submitted to Frontiers in Sleep and Chronobiology, a specialty of Frontiers in Neurology.

Background {#Sec1} ========== The morphological diversity of the phylum Mollusca is represented by eight recent clades, including the well-known gastropods, bivalves, and cephalopods. This bodyplan plasticity renders mollusks an ideal target for evolutionary and developmental studies. Although numerous aspects of intra-molluscan relationships are still controversially discussed, recent phylogenomic analyses show a basal dichotomy comprising the monophyletic Aculifera (including the eight-shelled Polyplacophora and the shell-less, spicule-bearing aplacophoran clades Neomeniomorpha or Solenogastres and Chaetodermomorpha or Caudofoveata) and the primarily single-shelled Conchifera (Monoplacophora, Gastropoda, Cephalopoda, Scaphopoda, Bivalvia) \[[@CR1]--[@CR3]\]. The vermiform Neomeniomorpha represents one of the least investigated molluscan taxa, although recent comparative studies have demonstrated its importance to reconstruct ancestral aculiferan traits \[[@CR4], [@CR5]\]. The adult neomeniomorph nervous system reflects the specific molluscan condition of an esophageal nerve ring, which includes a cerebral ganglion as well as a paired pedal ganglion. Two lateral (visceral) nerve cords emanate from the cerebral ganglion, while the pedal ganglia give rise to a pair of ventral (pedal) nerve cords. Together, these four longitudinal nerve cords form the molluscan tetraneural nervous system \[[@CR6]--[@CR10]\] (see also Fig. [1](#Fig1){ref-type="fig"}). During early stages of neomeniomorph neurogenesis the subsidence of epidermal cells results in two lateral depressions of the anterior larval episphere (often termed "ectodermal cerebral depressions"), which give rise to the anlagen of the cerebral ganglion \[[@CR11], [@CR12]\]. Before the tetraneural condition is established, two longitudinal neurite bundles emerge simultaneously from both the anterior and the posterior pole and subsequently fuse in the region of the prototroch, the locomotive ciliary band of the free-swimming larva \[[@CR13]\] (Fig. [1](#Fig1){ref-type="fig"}a). This formation of an intermediate stage with a single pair of nerve cords has recently been interpreted as a putative ancestral feature of spiralian neurogenesis \[[@CR13]\].Fig. 1Summary of neomeniomorph neurogenesis and adult neuroanatomy. **a** Schematic representation of neomeniomorph neurogenesis from the freshly hatched test cell larva until the juvenile stage. Anterior faces up. Neural structures are drawn in yellow. Reconstruction based on Redl et al. (2014). **b** Confocal scan (maximum intensity projection, color depth coding) of the adult CNS, anterior region. Anterior to the right. Anti-serotonin (5-HT) staining. **c** Same confocal scan (maximum intensity projection) as in (**b**), but with additional nucleic acid staining (DAPI) to reveal nuclei of the cerebral ganglia. **d** Schematic representation of the posterior and anterior major elements of the adult nervous system of *Wirenia argentea* based on (**b**) and (**c**) as well as Todt et al. (2008). Abbreviations: vestibular (atrial) sense organ (vso); basal ganglion (bg); buccal commissure (buc); buccal ganglion (bug); cerebral ganglion (cg); dorsoterminal sense organ (dts); frontal ganglia (fg); innervation of vestibular (atrial) sense organ (iv); innervation of pedal pit (ip); lateral ganglion (lg); lateroventral commissure (lvc); mouth (m); pedal ganglion (pg); pedal pit (pp); posterior ganglion (pog); ventral commissure (vc); ventral neural plexus (vnp); nerves of pedal pit (arrowhead); lateral (visceral) nerve cord (arrow); ventral (pedal) nerve cord (double arrowhead); mouth opening (asterisk). Scale bars: 100 μm Adult neomeniomorphs lack eyes but do exhibit several sensory organs, including a vestibular (atrial) sense organ and a dorsoterminal sense organ that is often considered a homolog of the osphradium of other molluscan clades \[[@CR6], [@CR14]--[@CR16]\] (Fig. [1b, c, d](#Fig1){ref-type="fig"}). Additionally, some neomeniomorphs exhibit a pedal commissural sac that was suggested to serve as a geosensory organ \[[@CR17]\]. Similar to many other spiralians, neomeniomorph larvae exhibit an apical organ with a ciliary tuft \[[@CR13], [@CR18]\]. Interestingly, and in contrast to the aplacophoran clades, representatives of the Polyplacophora exhibit ventrally positioned posttrochal larval eyes, which are lost some time after metamorphosis. The highly conserved paired box (Pax) genes encode for a family of metazoan transcription factors, which are crucial for cell fate specification and tissue differentiation and are known to be involved in the development of excretory organs, myogenesis, neurogenesis, biomineralisation processes (skeletogenesis), and the development of visual and geosensory systems in numerous bilaterians (e.g., \[[@CR19]--[@CR21]\]). Although neomeniomorph larvae lack a number of sensory organs such as eyes and statocysts, they possess Pax genes whose orthologs have a conserved function in neurogenesis and sensory organ development. This poses the question as to where these genes are expressed during the ontogeny of neomeniomorphs. *Pax2/5/8* is known to be involved in the development of excretory systems of certain annelids, onychophorans, vertebrates, as well as in the formation of auditory/geosensory systems and in establishing the midbrain/hindbrain boundary of vertebrates as well as the deutocerebrum/tritocerebrum boundary of ecdysozoans \[[@CR22]--[@CR24]\]. The developmental expression and putative function of *Pax2/5/8* has also been studied in a few lophotrochozoan taxa \[[@CR25]--[@CR28]\]. In mollusks, *Pax2/5/8* is expressed during development of multimodal sensory systems of gastropods \[[@CR25]\], polyplacophorans, and cephalopods \[[@CR27]\]. Furthermore, *Pax2/5/8* is expressed in the mantle of gastropods, polyplacophorans, bivalves, and cephalopods, which is known to be rich in sensory structures \[[@CR25], [@CR27]\]. In the polychaete annelid *Platynereis dumerilii Pax2/5/8* expression was found during development of photoreceptor cells of regenerating segments \[[@CR29]\], while in the leech *Helobdella austinensis Pax2/5/8* expression is confined to the neuroectoderm of the developing ventral nerve cords and to the developing nephridia \[[@CR26]\]. A recent study of *Pax6* and *Pax2/5/8* in two brachiopods suggested that these genes have a putative role as regulators of the segment polarity gene *engrailed*, although further details remain vague \[[@CR28]\]. *Pax6* is generally involved in the development of the bilaterian central nervous system (CNS) and is a key player in eye gene regulatory networks of most bilaterians, including cephalopod and polyplacophoran mollusks \[[@CR30]--[@CR32]\]. *Paxβ* represents the lophotrochozoan-specific ortholog of the recently discovered Paxα/β subfamily and has hitherto been exclusively studied in the leech *Helobdella austinensis* \[[@CR33]--[@CR35]\]. Late embryos of *H. austinensis* show broad, segmentally restricted mesodermal expression as well as expression in the CNS and eyes during organogenesis \[[@CR33]\]. So far, the expression pattern of *Paxα* (the deuterostome and ecdysozoan *Paxβ* ortholog) is exclusively known from the onychophoran *Euperipatoides rowelli* \[[@CR35]\]. In order to further assess putative ancestral versus novel roles of these important developmental regulators, we here provide the first detailed study of selected Pax family genes in a hitherto largely neglected but evolutionarily highly important molluscan clade, the Neomeniomorpha. Methods {#Sec2} ======= Animal cultures {#Sec3} --------------- Adult and developmental stages of the neomeniomorph *Wirenia argentea* Odhner, 1921 were collected, maintained, and reared from January to May 2012, November 2012 to February 2013, and from November to December 2013, respectively, following Redl et al. \[[@CR13]\] with the following modifications during the last season: The sediment sample was kept in 20μm-filtered and UV-sterilized sea water with a salinity of 35‰ (FSSW) precooled to 4°C. Every four days the adult specimens were transferred to clean plastic jars with fresh FSSW, which increased egg laying productivity. The freshly laid eggs were transferred into clean plastic jars with FSSW and kept under the same conditions as the adults. After hatching *Wirenia* develops via the so-called pericalymma or test cell larva. Herein, age of the larvae is given in days post hatching (dph). We fixed four morphologically distinguishable stages: freshly hatched test cell larva (0--1 dph), early test cell larva (6--7 dph), mid-stage test cell larva (10--11 dph), and late test cell larva (14--15 dph), whereby each stage encompasses a developmental time range of approximately 24 h. Immunochemistry {#Sec4} --------------- Relaxation, fixation, storage as well as all immunocytochemical procedures followed standard protocols as described in detail in Redl et al. \[[@CR13]\] and Scherholz et al. \[[@CR5]\], respectively. RNA extraction and fixation of animals {#Sec5} -------------------------------------- All molecular biological procedures, ranging from RNA extraction until the end of the *in situ* hybridization protocol, were conducted using RNase-free (or diethylpyrocarbonate (DEPC)-treated) water. Total RNA was extracted from larvae using the Qiagen RNeasy Mini Kit (Qiagen, Venlo, The Netherlands) with the QIAshredder homogenizer (Qiagen). Prior to RNA extraction the larval material was either shock frozen on dry ice or conserved in RNAlater. Additional RNA was extracted from adult specimens. In order to separate animal tissue from the gut content and to avoid contamination, all adult specimens were starved for at least two weeks prior to RNA extraction. Adult RNA was extracted using TRI reagent (Sigma-Aldrich) according to the manufacturer's instructions with the optional centrifugation step after homogenization and the following modifications: The animals were shock frozen with dry ice immediately before homogenization and RNA precipitation was performed with a 1:1 mixture of isopropanol and a high salt precipitation solution containing 0.8mol/l trisodium citrate dihydrate and 1.2mol/l sodium chloride. RNA pellets were dissolved in water and stored at −80°C. Advanced larval stages were relaxed prior to fixation for 20 to 30 min at 4°C by adding a 3.2% magnesium chloride solution. Developmental stages were fixed for in situ hybridization with 4% paraformaldehyde (PFA) in 0.1M MOPS buffer (with 0.5M/l NaCl, 2mM/l MgSO~4~, and 1mM/l EGTA added) for 45 min at room temperature (RT). Fixed larvae were stepped into precooled (4°C) or prechilled (−20°C) 75% EtOH, washed three times for a total period of 15--30 min in precooled (4°C) or for 45 min to 1.5 h in prechilled (−20°C) 75% EtOH, and stored in fresh 75% EtOH at −20 °C. RNAseq and transcriptome assembly {#Sec6} --------------------------------- Larval and adult RNA samples were pooled and sequenced by Illumina technology (Eurofins, Ebersberg, Germany). Sequencing and bioinformatic processing of the resulting paired-end libraries as well as the transcriptome assembly were performed as described in Redl et al. \[[@CR36]\]. Gene identification and orthology assignment {#Sec7} -------------------------------------------- We identified candidate genes in our transcriptome database by reciprocal BLAST \[[@CR37]\] searches using bilaterian Pax gene sequences from NCBI GenBank as queries. Nucleotide sequences of the best-fitting contigs were translated into amino acid sequences using Geneious, Version 6.1.6 (Biomatters, Auckland, New Zealand). Gene orthology was determined by phylogenetic reconstruction. FASTA-formatted files were generated with the inferred amino acid sequences for cloned genes and representative homologs from other bilaterian taxa (see Table [1](#Tab1){ref-type="table"} for accession numbers). Sequence alignment was performed with the online version of MAFFT (<http://www.ebi.ac.uk/Tools/msa/mafft/>; \[[@CR38]\]) with the following modifications to the standard setting: MAXITERATE → 100 (long run); PERFORM FFTS → localpair. The resulting alignment was checked and manually edited using BioEdit \[[@CR39]\] to remove non-conserved regions. The phylogenetic analysis was carried out with the Bayesian phylogenetic program MrBayes v.3.2.6 \[[@CR40]\]. A specified evolutionary model was determined using Akaike Information Criterion (AIC) as implemented in ProtTest 3 \[[@CR41]\]. The following parameters were employed in MrBayes: Jones-Taylor-Thornton model of amino-acid substitution \[[@CR42]\]; six rates categories for the gamma distribution; 30,000,000 generations; sample frequency 1,000. After the removal of 25% of the sampled trees as burn-in, the final phylogenetic tree was created and subsequently edited with FigTree v1.4.2 (<http://tree.bio.ed.ac.uk/software/figtree/>; \[[@CR43]\]). The resulting illustration was modified with Adobe Illustrator CC 2015 (Adobe Systems, San José, California, USA).Table 1GenBank accession numbers of genes used in the phylogenetic analysisSpeciesPhylumGeneGenBank accession number\**Wirenia argentea*Mollusca*Paxβ*\[KY488206\]\**Acanthochitona crinita*Mollusca*Paxβ*\[KY488203\]*Aplysia californica*Mollusca*Paxβ*\[DAA12510\]*Lottia gigantea*Mollusca*Paxβ*\[DAA12512\]*Capitella teleta*Annelida*Paxβ*\[DAA12511\]*Helobdella austinensis*Annelida*Paxβ2*\[ABQ45871\]*Schmidtea mediterranea*Platyhelminthes*Paxβ2*\[DAA12514\]*Euperipatoides rowelli*Onychophora*Paxα*\[AJG44471\]*Tribolium castaneum*Arthropoda*Pox-neuro*\[EFA07416\]*Drosophila melanogaster*Arthropoda*Pox-neuro*\[AAA28832\]*Saccoglossus kowalevskii*Hemichordata*Pox-neuro*\[NP_001158393\]\**Wirenia argentea*Mollusca*Pax2/5/8*\[KY488205\]*Acanthochitona crinita*Mollusca*Pax2/5/8*\[ALM30867\]*Platynereis dumerilii*Annelida*Pax2/5/8*\[AGC12568\]*Crassostrea gigas*Mollusca*Pax2A*\[EKC36239\]*Saccoglossus kowalevskii*Hemichordata*Pax2/5/8*\[ADB22664\]*Euperipatoides rowelli*Onychophora*Pax2/5/8*\[AJG44467\]*Gallus gallus*Chordata*Pax2*\[NP_990124\]*Branchiostoma belcheri*Chordata*Pax2*\[ABK54277\]*Tribolium castaneum*Arthropoda*Pax2/5/8*\[EFA01334\]*Ciona intestinalis*Chordata*Pax2/5/8*\[NP_001027652\]*Capitella teleta*Annelida*Pax3/7*\[ABC68267\]*Branchiostoma belcheri*Chordata*Pax3/7*\[ABK54280\]*Octopus bimaculoides*Mollusca*Pax3/7*\[ACR19857\]*Sepia officinalis*Mollusca*Pax3/7*\[AHG12548\]*Crassostrea gigas*Mollusca*Pax7*\[EKC41820\]*Ciona intestinalis*Chordata*Pax1/9*\[BAA74829\]*Terebratalia transversa*Brachiopoda*Pax1/9*\[AJV21320\]*Ptychodera flava*Hemichordata*Pax1/9*\[BAA78380\]*Branchiostoma belcheri*Chordata*Pax1/9*\[ABK54274\]\**Wirenia argentea*Mollusca*Pax6*\[KY488204\]\**Acanthochitona crinita*Mollusca*Pax6*\[KY488202\]*Doryteuthis opalescens*Mollusca*Pax6*\[AAB40616\]*Euprymna scolopes*Mollusca*Pax6*\[AF513712\]*Idiosepius paradoxus*Mollusca*Pax6*\[BAM74253\]*Terebratalia transversa*Brachiopoda*Pax6*\[ADZ24784\]*Lineus sanguineus*Nemertea*Pax6*\[CAA64847\]*Platynereis dumerilii*Annelida*Pax6*\[CAJ40659\]*Saccoglossus kowalevskii*Hemichordata*Pax6*\[NP_001158383\]Pax sequences of Neomeniomorpha and Polyplacophora retrieved from our transcriptomic data are labeled by asterisk Gene cloning and probe synthesis {#Sec8} -------------------------------- Oligonucleotide primers were designed from contigs using Geneious, Version 6.1.6 (Biomatters, Auckland, New Zealand). Thus, the defined gene fragments include the entire sequences of the conserved Paired domain and its flanking 5′ and 3′ ends in case of *War-Pax2/5/8* and *War-Paxβ*, whereas the gene fragment of *War-Pax6* comprises a partial sequence of the conserved prd-class homeodomain (and the flanking 3′ end of this conserved region), which is entirely lacking in the *Paxβ* paralog and partially lacking in the *Pax2/5/8* paralog. Specific primers and fragment lengths of probes are available in Table [2](#Tab2){ref-type="table"}. Primers were synthesized by Invitrogen, Life Technologies. The nucleotide sequences as well as the amino acid sequences of the amplified fragments have been deposited at NCBI GenBank (see Table [1](#Tab1){ref-type="table"} for accession numbers). PCR amplification was performed on a cDNA library synthesized from combined mixed-stage embryonic and adult total RNA. cDNA was synthesized using a 1^st^ Strand cDNA Synthesis Kit for RT-PCR (AMV) (Roche, Basel, Switzerland). Amplified fragments were cloned into pGEM-T Easy Vector System I (Promega, Madison, WI, USA) and the plasmids were used to transform *E. coli* JM109 Competent Cells (Promega). Plasmids were extracted from minipreps using the QIAprep Spin Miniprep Kit (Qiagen) and sequenced by Microsynth (Vienna, Austria). Obtained sequences were compared to known Pax sequences from NCBI GenBank and to the original contigs of the *Wirenia argentea* transcriptome using the BLAST program and Geneious, Version 6.1.6. The linear template for probe synthesis was generated via standard PCR using GoTaq Flexi DNA Polymerase reagents and M13 forward and reverse primers (FFW 5′-GTTTTCCCAGTCACGACGTT-3′, annealing temperature: 60°C; REV 5′-GACCATGATTACGCCAAGCTA-3′, annealing temperature: 60°C). Amplified products were purified using the GeneJET PCR Purification Kit (Thermo Fisher Scientific, Waltham, MA, USA) and subsequently used as templates for anti-sense RNA probe syntheses. Synthesis reaction was performed using a DIG RNA Labeling Kit (SP6/T7) (Roche Life Science). 2μl of 100mM dithiotreitol (DTT) were added to the transcription reaction and, after the reaction, template DNA was removed by incubation with DNase I, RNase free (Roche Life Science). Precipitation of RNA was done at −80°C. Protector RNase Inhibitor (Roche Life Science) was added after dissolving the RNA pellet in water. RNA probes for in situ hybridization were stored at −80°C.Table 2Primers used for PCRGeneFragment length (in bases)DirectionPrimer sequence*War-Pax2/5/8*426ForwardCAGATTTTCTGTCGGACTTCCTCTGATGTCReverseGACCCTAACGGTCACGGAGGAGTAAAC*War-Pax6*709ForwardGAATTTGAGAGGACACATTATCCAGACReverseGACATAATATATCCGGATCTCCATATTCG*War-Paxβ*667ForwardGCATGGAAAGGTGAGAGTAGATGATTCReverseGTAACAGATTCTAACGTGATCAACCAC Whole-mount in situ hybridization {#Sec9} --------------------------------- Samples stored in 75% EtOH were stepped into 4% PFA in 1x Roti-Stock phosphate buffered saline (pH = 7.4; PBS; Carl Roth) with 0.05M/l EGTA (PPE) and decalcified in PPE for 1h at RT. Subsequently, samples were washed six times for 5 min each in PBS with 0.1% Tween 20 (PBT; Carl Roth) at RT and then heated to 37°C in a water bath during the last washing step. Proteinase K treatment was done with a solution of 10μg/ml Proteinase K (Roche Life Science) in PBT for 10 min at 37°C without agitation. The specimens were then washed twice for 5 min each at RT in PBT, twice for 5 min each in 1% triethanolamine (TEA) in PBT, four times for 5 min each in an instantly made mixture of 0.3% acetic anhydride and 1% TEA in PBT, and again twice for 5 min each in PBT. Afterwards, the samples were postfixed in 4% PFA in PBS for 45 min at RT and washed five times for 5 min each at RT in PBT. Samples were subsequently stepped into the hybridization buffer (HB) consisting of 50% formamide with 0.075M/l trisodium citrate, 0.75M/l sodium chloride, 5mM/l EDTA, 50μg/ml heparin sodium salt (Sigma-Aldrich), 1x Denhardt's Solution (Carl Roth), 100μg/ml RNA from torula yeast, Type VI (Sigma-Aldrich), and 5% dextran sulfate sodium salt from *Leuconostoc* spp. (Sigma-Aldrich). The samples were then transferred into fresh HB, heated in a water bath to 56°C (in case of the *Pax6* probe) or 60°C (in case of *Pax2/5/8* and *Paxβ*), respectively, and prehybridized at these specific temperatures for 15--20 h. All RNA probes were diluted in HB to a final concentration of 1--2μg/ml, denatured for 10 min at 85°C, and applied to the samples. Hybridization was conducted for 24--26 h at the specific temperatures of 56°C and 60°C, respectively. Then, the samples were kept at hybridization temperature and washed three times for 20 min each in pre-warmed 50% formamide with 0.06M/l trisodium citrate, 0.6M/l sodium chloride, and 0.1% Tween 20, twice for 20 min each in 50% formamide with 0.03M/l trisodium citrate, 0.3M/l sodium chloride, and 0.1% Tween 20, and three times for 15 min each in 50% formamide with 0.015M/l trisodium citrate, 0.15M/l sodium chloride, and 0.1% Tween 20. The samples were then placed at RT to cool down. Afterwards, they were washed thrice for 20 min each at RT in 0.015M trisodium citrate solution with 0.15M/l sodium chloride and 0.1% Tween 20. Subsequently, the samples were washed thrice for 5 min each at RT in 0.1M maleic acid buffer (MAB) with 0.15M/l sodium chloride and 0.1% Tween 20 (pH = 7.5). Blocking of unspecific binding sites was done for 3 h at RT with a 2% solution of Blocking Reagent (Roche Life Science) in MAB. Anti-Digoxigenin (DIG)-AP, Fab fragments (Roche Life Science, Ref. 11093274910) were applied in a 1:2,500 or 1:5,000 dilution in 2% block solution for 13--16 h at 4°C. After incubation, the samples were rinsed eight times for 20 min each at RT in PBT, twice for 5 min each without agitation in 0.1M Tris buffer with 0.1M/l sodium chloride (pH = 9.5; AP buffer) and 0.1% Tween 20, and twice for 10 min each without agitation in AP buffer with 50mM/l magnesium chloride and 0.1% Tween 20. Finally, all samples were transferred into staining buffer (AP buffer with 50mM/l magnesium chloride, 7.5% polyvinyl alcohol, and 20μl/ml NBT/BCIP stock solution (Roche Life Science)). Color reaction took place at 4°C for 3--8 h (depending on probe, probe concentration, and DIG concentration) and was stopped by washing twice in 0.1M glycine in PBT (pH = 2.2) followed by additional two washes in PBT, 5 min each. Stained specimens were then fixed for 12--24 h at 4°C in 4% PFA in PBS, subsequently rinsed four times for at least 5 min each in PBT, and finally stored in PBT at 4°C. Mounting and clearing {#Sec10} --------------------- Larval stages were stepped into deionized water and washed four times for 5 min each in deionized water. Larvae were then stepped into EtOH and washed thrice for 5 min each in 100% EtOH. Next, the larvae were transferred into a 1:1 mixture of benzyl benzoate and benzyl alcohol, and subsequently mounted in this medium on microscope slides. Approximately 250 specimens were processed and investigated in total and 81 (25 with *Pax2/5/8* expression, 31 with *Pax6* expression, and 25 with *Paxβ* expression) were scanned with a confocal microscope. Microscopy, 3D rendering, and image processing {#Sec11} ---------------------------------------------- Specimens were analyzed and light micrographs were taken on an Olympus BX53 microscope equipped with an Olympus DP73 camera and the software cellSens Standard, Version 1.11 (Olympus Corporation, Shinjuku, Tokyo, Japan). Confocal laser scanning microscopy was conducted using a Leica DMI6000 CFS microscope equipped with a Leica TCS SP5 II scanning system (Leica Microsystems, Wetzlar, Germany) and software LAS AF, Version 2.6.0 or 2.6.3. The autofluorescent signal was scanned in fluorescence mode using a 405 nm laser and gene expression signal was scanned in reflection mode using a 633nm laser (see \[[@CR44]--[@CR47]\]). The obtained confocal image stacks were processed and used to prepare 3D reconstructions of the gene expression signal as well as 3D renderings of larval tissue with Imaris x64, Version 7.3.1 (Bitplane AG, Zurich, Switzerland). The same software was used to create video files of larval stages including gene expression patterns. Generated images were finally processed with Adobe Photoshop CS6 Extended, Version 13.0.1 x64, and Adobe Photoshop CC 2015 (Adobe Systems). Figures and schematic drawings were generated with Adobe Illustrator CS5, Version 15.0.0 and Adobe Illustrator CC 2015, Version 1.0 (Adobe Systems). Results {#Sec12} ======= Larval morphology of the neomeniomorph *Wirenia argentea* {#Sec13} --------------------------------------------------------- *Wirenia argentea* develops via a lecithotrophic trochophore-like larva, the so-called pericalymma or test cell larva. The bell-shaped freshly hatched test cell larva features an anterior episphere with an apical tuft and a posterior hyposhere with posterior invagination, the so-called "pseudo-blastopore" \[[@CR12]\] (Fig. [2](#Fig2){ref-type="fig"}a). The episphere is separated from the hyposphere by the ciliary prototroch formed by one row of trochoblasts (Fig. [2](#Fig2){ref-type="fig"}a). All examined larval stages of *Wirenia* reveal a characteristic cell arrangement of the covering test cells (Fig. [3](#Fig3){ref-type="fig"}; see also Additional file 1). In freshly hatched test cell larvae the posterior hyposphere is composed of two rows of test cells with two additional posteriormost and laterally positioned cells (Fig. [2](#Fig2){ref-type="fig"}a, b) \[[@CR12]\]. The anterior episphere contains two rows of five test cells each. Both rows are arranged such that the median Z-axes of the upper cells align with the cell-cell boundary of the row below (Fig. [3a, b, c, f](#Fig3){ref-type="fig"}). Additionally, the episphere features six knob-like structures (most likely either small cells or cell protuberances), two dorsolaterally, two laterally, and two ventrally located, which are bilaterally arranged (Fig. [3](#Fig3){ref-type="fig"}f). The epispheric arrangement of bilateral knob-like structures and the two rows of test cells with opposing configuration always correlates with the ventral mouth opening of the advanced larval stages. This, in combination with the two posteriormost and laterally positioned test cells of the freshly hatched test cell larvae, allows for determining the dorso-ventral axis in larval stages that still lack significant gross morphological features such as the ventral mouth opening.Fig. 2Morphology of a freshly hatched *W. argentea* larva. Dorsal (d)--ventral (v), left (l)--right (r), and anterior (a)--posterior (p) axes indicate the orientation. **a** Optical section of an autofluorescence scan of a freshly hatched test cell larva with anterior facing up. Dashed line indicates the prototroch, which separates the episphere (ep) from the hyposphere (hy). **b** 3D volume rendering based on an autofluorescence scan. Posterior view on the larval hyposphere and the invagination of the pseudo-blastopore of the same specimen as in (**a**). Abbreviations: cells of the apical organ (ao); episphere (ep); hyposphere (hy); lateral depression (ld); nucleus (n); pseudo-blastopore (pb); trochoblast (tb); test cell (tc); the two most posteriorly and laterally positioned test cells (arrowhead). Scale bars: 50 μm Fig. 3Line drawings of gross morphology of the early *W. argentea* larva. Schematic drawing of an early test cell larva (6-7 days post hatching) based on autofluorescence confocal scans. Anterior faces up in (**a**)-(**e**) and towards the viewer in (**f**). **a** Ventral view. **b** Lateral view from the right. **c** Dorsal view. **d** Horizontal section. **e** Sagittal section. **f** Anterior view of the episphere showing lateral depressions. Note the characteristic cell arrangement of two rows of test cells in the hyposphere, two rows of test cells with opposing configuration in the episphere (*grey/dark grey*), and six bilaterally arranged cells or cell protuberances (*orange*) in the episphere. Dorsal (d)--ventral (v) and left (l)--right (r) axes indicate the orientation. Abbreviations: apical tuft (at); lumen of foregut (fg); posterior knob-like structure (ks); lateral depression (ld); peri-imaginal space (pis); prototroch (pt); trochoblast (tb); test cells (tc); epidermis of the trunk (te); telotroch (tt) **Additional file 1:** Video of an autofluorescence scan and subsequent 3D rendering of an early test cell larva (6--7 days post hatching) of *Wirenia argentea* (Neomeniomorpha). (AVI 17546 kb) Larval development of *Wirenia* is characterized by the outgrowth of a posterior trunk from the base of the pseudo-blastopore, resulting in a mushroom-like appearance of later stages. The first external indication of this outgrowing trunk is visible in the early test cell larva, in which the major part of the developing trunk is still masked by the outer apical cap (Fig. [3](#Fig3){ref-type="fig"}). This mode of formation results in an epidermal fold enclosing the "peri-imaginal space" ("Peri-Imaginalraum" *sensu* Salvini-Plawen \[[@CR48]\]), which is lined by the epidermal cell layer of the outgrowing trunk on both sides (Fig. [3](#Fig3){ref-type="fig"}d, e). The early test cell larva already exhibits a ventral mouth opening and the developing foregut but lacks the two posteriormost test cells of the freshly hatched test cell larva. The typical mushroom-like appearance of later stages is recognizable in the mid-stage test cell larva, where the trunk is elongated. Late test cell larvae are already in the settlement phase and the apical cap and the covering test cells start to degenerate. The length of this phase varies individually and can last for several days \[[@CR12]\]. At this stage, the posterior trunk is already thickened, elongated, and dorsally and laterally covered by spicules. The ventral trunk exhibits a ventrolongitudinal ciliary band, which marks the developing creeping sole. Gene orthologs and phylogenetic analysis {#Sec14} ---------------------------------------- Seven bilaterian Pax groups or subfamilies have been identified by the presence or absence of highly conserved structural domains \[[@CR49]\]. All Pax genes exhibit a conserved N-terminal Paired domain, whereas the originally existing octapeptide was lost in the paralogs *Pax4/6/10*, *Paxα/β*, and *Pax-eyg* but is still present in the paralogs *Pox-neuro*, *Pax1/9*, *Pax2/5/8*, and *Pax3/7* (Fig. [4](#Fig4){ref-type="fig"}) \[[@CR50]\]. Another Pax gene-specific element is a so-called prd-class homeodomain, which is present in the paralogs *Pax-eyg*, *Pax3/7*, and *Pax4/6/10* as well as partially present in *Pax2/5/8* but absent in all other Pax paralogs (Fig. [4](#Fig4){ref-type="fig"}).Fig. 4Alignment of *W. argentea* Pax genes and bilaterian orthologs. Alignment of predicted amino acid sequences of various bilaterian Pax genes including *War-Pax2/5/8, War-Pax6, War-Paxβ, Acr-Pax6* and *Acr-Paxβ* (labeled with asterisk). The conserved N-terminal PAIRED domain (*red*), and, if present, the octapeptide (*green*) and the (partial) prd-class homeodomain (*blue*) are shown, while the C-terminal transactivation domain is omitted. GenBank accession numbers of all encoding genes used are listed in Table [1](#Tab1){ref-type="table"} Our multiple sequence alignment includes bilaterian orthologs of all Pax subfamilies and comprised, if present, the conserved N-terminal Paired domain, the octapeptide, the prd-class homeodomain, and other conserved C-terminal domains (Fig. [4](#Fig4){ref-type="fig"}). Although our transcriptomic data revealed only partial sequences of *War-Pax6* and *Acr-Pax6*, BLAST searches against bilaterian Pax orthologs and the presence of diagnostic conserved domains (i.e., prd-class and transactivation domain) identified the two aforementioned molluscan sequences as true Pax genes (Fig. [4](#Fig4){ref-type="fig"}). The phylogenetic analysis demonstrates that all translated neomeniomorph and polyplacophoran Pax amino acid sequences (War-Pax2/5/8, War-Pax6, War-Paxβ, Acr- Pax6, Acr-Paxβ) cluster with their corresponding bilaterian orthologs (Fig. [5](#Fig5){ref-type="fig"}).Fig. 5Phylogenetic reconstruction of the Pax gene family. The consensus tree was inferred through Bayesian phylogenetic analysis with MrBayes discarding 25% of samples as burn-in. The branch support values are posterior probability values of Bayesian likelihood. Pax protein families are labeled in different colors. Orthologous sequences retrieved from our transcriptomes are displayed in white letters. Note that our sequences cluster with other appropriate bilaterian orthologs *Pax2/5/8* expression {#Sec15} --------------------- The free-swimming, freshly hatched test cell larvae express *War-Pax2/5/8* in three pretrochal domains, two of them bilaterally orientated and a third one located medioventrally (Fig. [6](#Fig6){ref-type="fig"}a-d; see also Additional file 2). The two bilateral domains are each expressed in a single ectodermal cell, which lies adjacent and anterior to the trochoblasts (Fig. [6b](#Fig6){ref-type="fig"}) at the base of the lateral depressions. The medioventral domain likewise lies adjacent and anterior to the trochoblasts (Fig. [6c](#Fig6){ref-type="fig"}) but at the base of a ventral depression. Additional expression of *War-Pax2/5/8* is present in ectodermal cells lining the posterior pseudo-blastopore, although the expression is weaker in the ventral ectodermal cells of the invagination (Fig. [6b, e](#Fig6){ref-type="fig"}).Fig. 6Expression of *War-Pax2/5/8.* Anterior faces up in all panels except for the anterior views in (**d**) and (**i**). Dorsal (d)--ventral (v), left (l)--right (r), and anterior (a)--posterior (p) axes indicate the orientation. First, second, and third column as well as (**d**) and (**i**) are 3D-rendered images based on autofluorescence signal (turquoise) and reflection signal of *War-Pax2/5/8* expression staining (yellow). The ventral half of the larvae in the first and second column was omitted in order to visualize the location of the ventral gene expression signal. The lateral left larval hemisphere in images of the third column was omitted in order to enable the lateral view of the gene expression signal. Clipping plane projections of the respective panels are indicated by dotted white lines. (**n**), (**s**), and fourth column are light micrographs. Location of the prototroch is indicated by dashed black lines and the location of the mouth opening is indicated by an asterisk. Numbers mark distinguishable expression domains. **a** Ventral view. **b** Ventral view with most ventrally located expression omitted. **c** Lateral left view. **d** Anterior view. **e** Ventral view. **f** Ventral view. **g** Ventral view with most ventrally located expression omitted. **h** Lateral left view. **i** Anterior view. **j** Ventral view. Note the strong expression of *War-Pax2/5/8* in the epidermal cell layer of the outgrowing trunk. **k** Ventral view. **l** Ventral view with most ventrally located expression omitted. **m** Lateral left view. **n** Ventral view. **o** Lateral left view. **p** Ventral view. **q** Ventral view with most ventrally located expression omitted. Note the expression in the epidermal cell layer of the trunk, in cells of the developing CNS (\#1, \#6), and in the region where the protonephridia develop (\#4). **r** Lateral left view. **s** Ventral view. **t** Lateral left view. Abbreviation: foregut (fg). Scale bar: 50 μm **Additional file 2:** *Pax2/5/8* expression in a freshly hatched test cell larva. (AVI 25082 kb) Early test cell larvae exhibit two paired bilaterally orientated pretrochal expression domains of *War-Pax2/5/8* (Fig. [6](#Fig6){ref-type="fig"}h, i, j), which are herein referred to as "groups" (see also Additional file 3). Comparing the position of the neuropil and the corresponding nuclei of the developing cerebral ganglia (Fig. [7](#Fig7){ref-type="fig"}) with the position of both paired expression domains (Fig. [6](#Fig6){ref-type="fig"}j) shows that the anterior expression domains are located in cells or compartments of the anlagen of the cerebral ganglia. Several *War-Pax2/5/8*-expressing cells lie close to the region of the cerebral commissure (\#1 in Fig. [6](#Fig6){ref-type="fig"}i, j), whereas other *War-Pax2/5/8*-expressing cells are located more laterally and adjacent to the single row of trochoblasts (\#2 in Fig. [6](#Fig6){ref-type="fig"} h, i, j). Furthermore, *War-Pax2/5/8* is expressed in the epidermal cell layer of the outgrowing trunk, which also lines the peri-imaginal space (Fig. [6](#Fig6){ref-type="fig"}f, g, h, j). This epidermal expression has a ventral gap where the future creeping sole develops. A further bilateral group of *War-Pax2/5/8*-expressing cells is located ventrally, at the posterior pole of the trunk (\#3 in Fig. [6](#Fig6){ref-type="fig"}f, g, j).Fig. 7Mid-stage test cell larva. Confocal scan of a larva stained for F-actin (*red*) and DAPI (cell nuclei, *blue*). Anterior faces up. Note the staining of the F-actin-rich neuropil (np) underlying the (nuclei of the) cerebral ganglia. The dashed line indicates the prototroch. Abbreviations: nuclei of the developing cerebral ganglia (cg); developing foregut (fg); neuropil of the cerebral ganglia and cerebral commissure (np); peri-imaginal space (pis); spicule-secreting epidermal cells of the outgrowing trunk (sp); location of mouth opening (asterisk). Scale bar: 50 μm **Additional file 3:** *Pax2/5/8* expression in an early test cell larva. (AVI 29128 kb) Mid-stage test cell larvae express *War-Pax2/5/8* (Additional file 4) in cells of the anlagen of the cerebral ganglia. This cerebral expression can be subdivided in two pretrochal groups of *War-Pax2/5/8-*expressing cells. The most anterior group lies close to the cerebral commissure and adjacent and anterior to the epithelial cells of the developing foregut (\#1 in Fig. [6k, l, n, o](#Fig6){ref-type="fig"}). The second group of pretrochal *War-Pax2/5/8-*expressing cells is located more posteriorly and lies adjacent to the trochoblasts (\#2 in Fig. [6](#Fig6){ref-type="fig"}k-o). A third bilateral group of *War-Pax2/5/8-*expressing cells is located posttrochally and ventrally but slightly anterior to the invagination of the peri-imaginal space and the outgrowing trunk (\#4 in Fig. [6](#Fig6){ref-type="fig"}k-o). Two further bilateral expression domains are located on both sides of the ventral mouth opening, which lies at the base of the invagination of the peri-imaginal space (\#5 in Fig. [6](#Fig6){ref-type="fig"}o). Scattered *War-Pax2/5/8* expression is also detected in cells on both sides of and adjacent to the developing creeping sole, i.e., the region where the paired ventral nerve cords develop (\#6 in Fig. [6l, n, o](#Fig6){ref-type="fig"}). As in early test cell larvae, mid-stage test cell larvae still express *War-Pax2/5/8* in the epidermal cell layer of the outgrowing trunk, which also lines the peri-imaginal space (Fig. [6](#Fig6){ref-type="fig"}k-o). A ventral gap of this epidermal expression is still present in cells of the developing creeping sole. As in early test cell larvae, a bilateral group of *War-Pax2/5/8*-expressing cells is located ventrally at the posterior pole of the trunk (\#3 in Fig. [6](#Fig6){ref-type="fig"}l, n, o). **Additional file 4:** *Pax2/5/8* expression in a mid-stage test cell larva. (AVI 25505 kb) The pretrochal expression of *War-Pax2/5/8* in late test cell larvae (Additional file 5) is restricted to a marginal expression in the cerebral ganglia, with a more distinct expression close to the cerebral commissure (\#1 in Fig. [6p, q, s](#Fig6){ref-type="fig"}). Similar to the expression pattern in mid-stage test cell larvae another bilateral expression domain of *War-Pax2/5/8* is located posttrochally and rather ventrally of the mid-horizontal section plane, and anterior to the invagination of the peri-imaginal space and the outgrowing trunk (\#4 in Fig. [6](#Fig6){ref-type="fig"}p-s). As in mid-stage test cell larvae, *War-Pax2/5/8* is expressed in cells on both sides of and adjacent to the developing creeping sole, i.e., in the region where the paired ventral nerve cords are located (\#6 in Fig. [6](#Fig6){ref-type="fig"}q, s). Furthermore, *War-Pax2/5/8* is still expressed in the epidermal cell layer of the outgrowing trunk (Fig. [6](#Fig6){ref-type="fig"}p-t). Thus, the ectodermal expression of *War-Pax2/5/8* in the epidermal layer of the outgrowing trunk is present throughout all developmental stages investigated (Fig. [6f, k, p](#Fig6){ref-type="fig"}). The same applies to the characteristic ventral gap of expression in the region of the developing creeping sole. **Additional file 5:** *Pax2/5/8* expression in a late test cell larva. (AVI 29202 kb) *Pax6* expression {#Sec16} ----------------- The freshly hatched test cell larvae express *War-Pax6* (Additional file 6) in a single pretrochal domain and in two posttrochal domains. The pretrochal expression of *War-Pax6* is assigned to a single medioventral cell, which lies adjacent and slightly anterior to the trochoblasts, whereas the posttrochal expression of *War-Pax6* is confined to bilaterally orientated ectodermal cells at the ventral base of the pseudo-blastopore (Fig. [8](#Fig8){ref-type="fig"}a-d).Fig. 8Expression of *War-Pax6.* Anterior faces up in all panels. Dorsal (d)--ventral (v) left (l)--right (r), and anterior (a)--posterior (p) axes indicate the orientation. First and second column are 3D-rendered images based on autofluorescence signal (*turquoise*) and reflection signal of *War-Pax6* expression staining (*yellow*). The ventral half of the larvae in the first column was omitted in order to visualize the location of the ventral gene expression signal. The lateral left larval hemisphere in images of the second column was omitted in order to enable a lateral view of the gene expression signal. Clipping plane projections of the respective panels are indicated by dotted white lines. Third and fourth column are light micrographs. Location of the prototroch is indicated by dashed black lines and the mouth opening by an asterisk. Numbers mark distinguishable expression domains. **a** Ventral view. **b** Lateral left view. **c** Ventral view. **d** Lateral left view. **e** Ventral view. Note the first expression signal in cells of the developing cerebral ganglia (\#1). **f** Lateral left view. **g** Ventral view. i) Ventral view of ventral expression domains (\#2). ii) Ventral view of dorsal expression domains (\#3 - \#6). **h** Lateral left view. **i** Ventral view. Note the expression in the neuroectoderm of the developing ventral nerve cords (\#2´ and \#2´´). **j** Lateral left view. **k** Ventral view. **l** Lateral left view. **m** Ventral view. Note *War-Pax6* expression in cells of the developing CNS, in particular in the cerebral ganglia (\#1) and the ventral nerve cords (\#2´ and \#2´´), as well as the expression in cells flanking the mouth opening at the base of the invagination of the peri-imaginal space (\#7). **n** Lateral left view. **o** Ventral view. **p** Ventral view. Note the decreasing expression in the CNS of the late larva. Abbreviation: foregut (fg). Scale bar: 50 μm **Additional file 6:** *Pax6* expression in a freshly hatched test cell larva. (AVI 16204 kb) Early test cell larvae exhibit strong expression of *War-Pax6* (Additional file 7) in cells associated with the developing cerebral ganglia (\#1 in Fig. [8](#Fig8){ref-type="fig"}e-h). Furthermore, strong ectodermal expression is present in two ventral stripes of epidermal cells, which line the peri-imaginal space on both sides and therefore show a hook-like appearance (\#2 in Fig. [8e, f, g i, h](#Fig8){ref-type="fig"}). The outgrowing trunk exhibits three dorsal bilateral expression domains, each most likely composed of a pair of single *War-Pax6-*expressing epidermal cell (\#3, \#4 and \#5 in Fig. [8f, g ii, h](#Fig8){ref-type="fig"}). A further paired bilateral expression domain (\#6) is formed by a single cell belonging to the outer dorsolateral epidermal lining of the peri-imaginal space (Fig. [8f, g ii, h](#Fig8){ref-type="fig"}). **Additional file 7:** *Pax6* expression in an early test cell larva. (AVI 20713 kb) No dorsal expression of *War-Pax6* was observed in mid-stage test cell larvae. As in early test cell larvae, *War-Pax6* is strongly expressed in the anlagen of the cerebral ganglia (\#1 in Fig. [8](#Fig8){ref-type="fig"}i-l; Additional file 8). Likewise, the hook-like expression domains of *War-Pax6* in the epidermal lining of the peri-imaginal space are constantly present in early and mid-stage test cell larvae (\#2´ in Fig. [8](#Fig8){ref-type="fig"}i-k). Two longitudinal expression domains of *War-Pax6* (\#2´´) are separated by a small gap and are situated in direct posterior extension of the hook-like domains of \#2´ (Fig. [8](#Fig8){ref-type="fig"}i-l). *War-Pax6-*expressing cells of \#2´´ are epidermal cells flanking the epidermal cells of the developing creeping sole (Fig. [8](#Fig8){ref-type="fig"}i, k). **Additional file 8:** *Pax6* expression in a mid-stage test cell larva. (AVI 20261 kb) The expression of *War-Pax6* is generally weaker in late test cell larvae, but is still present in the cerebral ganglia (\#1 in Fig. [8](#Fig8){ref-type="fig"}m-p; Additional file 9). The epidermal layer with spicule-secreting cells is now restricted to the elongated trunk, which causes the longitudinal appearance of the formerly hook-like expression domain of \#2´ in this late stage (Fig. [8m, n, p](#Fig8){ref-type="fig"}). *War-Pax6-*expressing cells of \#2´´ are also present in late test cell larvae and, together with the cells of \#2´, form two longitudinal stripes of weak expression on both sides of the developing creeping sole, i.e., in the area where the ventral nerve cords develop (Fig. [8](#Fig8){ref-type="fig"}m, n). Separated from \#2´, additional *War-Pax6-*expressing cells flank the ventral mouth opening at the base of the invagination of the peri-imaginal space (\#7 in Fig. [8m, n, o](#Fig8){ref-type="fig"}). **Additional file 9:** *Pax6* expression in a late test cell larva. (AVI 19856 kb) *Paxβ* expression {#Sec17} ----------------- The expression pattern of *War-Paxβ* is largely consistent with the corresponding expression of *War-Pax6*. Freshly hatched test cell larvae express *War-Paxβ* in a single pretrochal domain in the same medioventral region adjacent to the ventral trochoblasts that also shows *War-Pax6* expression; however, the *War-Paxβ* expression domain is wider and encompasses more than one single cell (Fig. [9](#Fig9){ref-type="fig"}a-d, Additional file 10). Besides the pretrochal expression, freshly hatched test cell larvae also exhibit posttrochal expression in epidermal cells located ventrally of the pseudo-blastopore (Fig. [9](#Fig9){ref-type="fig"}a-d).Fig. 9Expression of *War-Paxβ.* Anterior faces up in all panels. Dorsal (d)--ventral (v) left (l)--right (r), and anterior (a)--posterior (p) axes indicate the orientation. First and second column are 3D rendered images based on autofluorescence (turquoise) and reflection signal of *War-Paxβ* expression staining (*yellow*). The ventral half of the larvae in the first column was omitted in order to visualize the location of the ventral gene expression signal. The lateral left larval hemisphere in images of the second column was omitted in order to enable a lateral view of the gene expression signal. Clipping plane projections of the respective panels are indicated by dotted white lines. Third and fourth column are light micrographs. Location of the prototroch is indicated by dashed black lines and the location of the mouth opening is indicated by an asterisk. Numbers mark distinguishable expression domains. **a** Ventral view. **b** Lateral left view. **c** Ventral view. **d** Lateral left view. **e** Ventral view. Note the first expression signal in cells of the developing cerebral ganglia (\#1). **f** Lateral left view. **g** Ventral view. **h** Lateral left view. **I** Ventral view. Note the expression in the neuroectoderm of the developing ventral nerve cords (\#2) and cerebral ganglia (\#1). **j** Lateral left view. **k** Ventral view. **l** Lateral left view. **m** Ventral view. Note *War-Paxβ* expression in cells of the cerebral ganglia (\#1) and the faint expression in the region of the developing ventral nerve cords (\#2) as well as the expression in cells flanking the mouth opening at the base of the invagination of the peri-imaginal space (\#4). **n** Lateral left view. **o** Ventral view of the most ventral part of a late larva. **p** Ventral view. Note the weak *War-Paxβ* expression in the CNS (\#1, \#2). Abbreviation: foregut (fg). Scale bar: 50 μm **Additional file 10:** *Paxβ* expression in a freshly hatched test cell larva. (AVI 30190 kb) As is the case for *War-Pax6,* early test cell larvae express *War-Paxβ* (Additional file 11) in cells of the developing cerebral ganglia (\#1) as well as in two ventral stripes of epidermal cells, which line the peri-imaginal space on both sides and therefore show a hook-like appearance (\#2 in Fig. [9](#Fig9){ref-type="fig"}e-h). Furthermore, *War-Paxβ* is expressed in two rather dorsally located spots at the posterior pole of the outgrowing trunk (\#3 in Fig. [9](#Fig9){ref-type="fig"}f). Each spot is most likely formed by a single *War-Paxβ-*expressing cell. **Additional file 11:** *Paxβ* expression in an early test cell larva. (AVI 18566 kb) In mid-stage test cell larvae *War-Paxβ-*expressing cell groups are located in similar expression domains as described for the early test cell larvae and thus closely resemble the *War-Pax6* expression pattern of mid-stage test cell larvae (Additional file 12). Strong *War-Paxβ* expression is present in the anlagen of the cerebral ganglia (\#1) and in two ventral stripes of ectodermal cells adjacent to the cells of the developing creeping sole (\#2 in Fig. [9](#Fig9){ref-type="fig"}i-l). As is the case for the *War-Pax6* expression, the stripes of *War-Paxβ-*expressing ectodermal cells of group 2 also line the fold of the peri-imaginal space, which causes a hook-like appearance at the anterior pole of the expression domain. **Additional file 12:** *Paxβ* expression in a mid-stage test cell larva. (AVI 16661 kb) Expression of *War-Paxβ* in late test cell larvae (Additional file 1) is comparatively low. The most intense expression is still present in the developing cerebral ganglia (\#1 in Fig. [9](#Fig9){ref-type="fig"}m, n, p), which is congruent with the expression of *War-Pax6* in late test cell larvae. A further analogy to the expression pattern of *War-Pax6* is the location of two spots of *War-Paxβ-*expressing cells flanking the mouth opening at the base of the invagination of the peri-imaginal space (\#4 in Fig. [9m, n, o](#Fig9){ref-type="fig"}). The ventral stripe-like *War-Paxβ* expression on both sides adjacent to the creeping sole (\#2; similar to the younger stages) is hardly detectable in late test cell larvae and, if visible, is mostly limited to a faint expression in the ventral part of the trunk (\#2 in Fig. [9](#Fig9){ref-type="fig"}m, n, p). **Additional file 13:** *Paxβ* expression in a late test cell larva. (AVI 11934 kb) Discussion {#Sec18} ========== Putative ancestral and co-opted roles of *Pax2/5/8* in neomeniomorph mollusks {#Sec19} ----------------------------------------------------------------------------- *Pax2/5/8* is an ortholog of the *Drosophila Sv/Spa* and the vertebrate *Pax2*, *Pax5,* and *Pax8* genes. These play a crucial role in the development and regionalization of the highly centralized and complex brains of vertebrates and insects as well as cephalopods, where they were most likely independently recruited into similar functions \[[@CR24], [@CR27]\]. Based on its expression in sensory structures of chordates (e.g., \[[@CR23], [@CR51]\]), insects \[[@CR52]\], and mollusks \[[@CR25]\], *Pax2/5/8* was proposed to have a conserved role in the formation of structures responsible for balance and geotaxis in eumetazoans \[[@CR25]\]. For Mollusca, our data on an aplacophoran supplement existing *Pax2/5/8* expression data on polyplacophorans, gastropods, bivalves, and cephalopods \[[@CR25], [@CR27], [@CR53]\]. Although *Pax2/5/8* was found to be expressed in the adult brain of the gastropod *Haliotis* \[[@CR53]\], there is no evidence that this gene is involved in its development \[[@CR25]\]. However, developmental expression of *Pax2/5/8* in *Haliotis* was found in the statocysts, eyes, foot, and in putative chemosensory organs of the pallial chamber \[[@CR25]\]. The expression of *Pax2/5/8* in putative (precursors of) sensory cells was also confirmed for a polyplacophoran, where it is probably expressed in the developing esthetes (shell eyes) and the larval ampullary system (cf. \[[@CR54]\]). Though, it was proposed that *Pax2/5/8* expression probably predates sensory cell development in multimodal sensory systems during molluscan ontogeny \[[@CR27]\]. Although neomeniomorphs do possess distinct sensory organs, such as the atrial sense organ or the dorsoterminal sense organ, we did not find any evidence for *Pax2/5/8* expression in cells that could be unambiguously assigned to a distinct sensory structure. However, *Wirenia* exhibits *Pax2/5/8* expression in cells dedicated to the developing cerebral ganglia and the ventral nerve cords, which is in stark contrast to the other investigated molluscan representatives with a considerable simple (i.e., little concentrated) CNS, such as bivalves, gastropods, and polyplacophorans, but is in accordance with *Pax2/5/8* expression in the highly centralized brain of cephalopods. Since data for only one species per class-level taxon are currently available, more comparative analyses are required to further interpret this finding. Nevertheless, and in contrast to *War-Pax6* and *War-Paxβ, War-Pax2/5/8* is not equally expressed in the region of the developing cerebral ganglia, but instead shows increased expression in cells adjacent to the apical organ and the cerebral commissure (\#1 in Fig. [6](#Fig6){ref-type="fig"} f-t), as well as in a slightly more posterior location where the future pedal ganglia develop (\#2 in Fig. [6](#Fig6){ref-type="fig"} f-o). Although neomeniomorphs do not exhibit a highly centralized CNS, this specific *Pax2/5/8* expression could hint towards a similar function in CNS regionalization comparable to its proposed function in cephalopods (cf. \[[@CR27]\]). Aside from its putative function in patterning neural or sensory structures, *Pax2/5/8* is also expressed during development of excretory organs, such as the nephridia of onychophorans and annelids \[[@CR26], [@CR35]\] and the kidneys of vertebrates \[[@CR55], [@CR56]\]. Interestingly, *Wirenia* is the only investigated mollusk, which shows *Pax2/5/8* expression in the region where the paired protonephridia of the larva develop (\#4 in Fig. [6](#Fig6){ref-type="fig"} k-t, see also \[[@CR12]\]). Whether this hints towards a conserved function of *Pax2/5/8* in bilaterian nephrogenesis or whether this represents a coopted function of the aplacophoran lineage remains unclear until more data on various clades become available. Remarkably, *Pax2/5/8* is expressed in epidermal spicule-secreting or associated cells of neomeniomorph as well as polyplacophoran mollusks \[[@CR27]\]. Although it was not explicitly mentioned for Polyplacophora, the respective light micrographs show a distinct expression signal in the spicule-secreting perinotum (or girdle) of *Acanthochitona crinita* (cf. Fig. [6](#Fig6){ref-type="fig"} g in \[[@CR27]\]). Not much is known about spicule secretion and its molecular background in aculiferan mollusks, but a recent study on the evolution of molluscan secretomes and biomineralization processes proposed that independent co-option of gene families are important driving forces acting on molluscan biomineralization \[[@CR57]\]. Since the secretion of calcareous spicules most likely constitutes an aculiferan autapomorphy and *Pax2/5/8* appears not to be involved in molluscan shell development \[[@CR25], [@CR27]\], it seems likely that *Pax2/5/8* expression in the spicule-secreting tissues of Polyplacophora and Neomeniomorpha evolved along the line leading towards the Aculifera after their split from their conchiferan allies. Conserved *Pax6* expression in aculiferan molluscs {#Sec20} -------------------------------------------------- *Pax6* is generally expressed in developing nervous systems and in various photosensory organs, regardless of their overall morphology in numerous metazoans, which underlines its important role as regulatory gene in eye developmental networks (see, e.g., \[[@CR58]--[@CR66]\]). Prior to our study, molluscan *Pax6* expression had only been studied in developmental stages of cephalopods \[[@CR30], [@CR31], [@CR67], [@CR68]\] and a polyplacophoran \[[@CR32]\], as well as in an adult gastropod \[[@CR53]\]. The latter study showed that neural expression of *Pax6* in adults of the gastropod *Haliotis asinina* is in the cerebral and pleuropedal ganglia as well as in sensory structures such as the eyes, tentacles, and gills \[[@CR53]\]. Cephalopod developmental *Pax6* expression in the squids *Loligo opalescens* and *Euprymna scolopes* is confined to their brain lobes, eyes, and olfactory organ \[[@CR30], [@CR67]\]. The polyplacophoran *Leptochiton asellus* likewise shows *Pax6* expression in the developing nervous system as well as in the ventrally and posttrochally positioned larval eyes \[[@CR32]\]. Since the latter study largely focused on eye development, we analyzed *Pax6* expression during neurogenesis of another polyplacophoran representative, *Acanthochitona crinita* (unpublished data). Our data confirm *Pax6* expression in the region of the developing ventral nerve cords but not in the lateral nerve cords of the tetraneural nervous system of this polyplacophoran species. Interestingly, this finding is congruent with *Pax6* expression in later developmental stages of the neomeniomorph *Wirenia argentea,* which argues for a conserved expression of this gene during neurogenesis of the ventral nerve cords of protostomes (cf., e.g., data on various crustaceans, insects, onychophorans, planarians, annelids, and polyplacophorans) \[[@CR32], [@CR62], [@CR63], [@CR66], [@CR69]--[@CR71]\]. Moreover, a conserved bilaterian mode of expression of *Pax6* is also present in the larval episphere of *W. argentea*, where it correlates with the development of the cerebral ganglia. Freshly hatched test cell larvae of *Wirenia* express *Pax6* in two patches at the ventral base of the pseudo-blastopore, i.e., in the region where a pair of neurite bundles emerges from the posterior neurogenic domain (cf. \[[@CR13]\]). Cells of the apical organ, the only larval sensory organ known from neomeniomorphs, do not exhibit *Pax6* expression. The most complex *Pax6* expression pattern occurs in early test cell larvae of *Wirenia* (\#3, \#4, \#5, and \#6 in Fig. [8](#Fig8){ref-type="fig"}f, g ii, h). Since this expression pattern is neither complemented by coexpression of *Pax2/5/8* or *Paxβ* nor supported by neurotransmitter distribution of 5-HT (serotonin) and FMRF-amide, the neurogenic nature of these *Pax6* expressing cells is questionable. Remarkably, late test cell larvae of *Wirenia* exhibit two patches of *Pax6*-expressing cells at the anterior margin of the larval hyposphere and lateral to the ventral mouth opening, which is congruent with two patches of *Paxβ*-expressing cells (compare \#7 in Fig. [8](#Fig8){ref-type="fig"}m, n, o and \#4 in Fig. [9](#Fig9){ref-type="fig"}m, n, o). By comparing the polyplacophoran trochophore larva with the neomeniomorphan test cell larva we found that this specific location of the *Pax6-* and *Paxβ*-expressing cells correlates with the location of the polyplacophoran larval eyes and its associated *Pax6* expression. Although, as far as currently known, neomeniomorph larvae and adults lack distinct photosensory organs (eyes), this correlation and the proposed close relationship of Polyplacophora and Neomeniomorpha calls for further morphological investigations combined with expression studies of suitable candidate genes that should focus on putative (rudimentary?) photoreceptors in this region. The putative role of Paxα/β during bilaterian development {#Sec21} --------------------------------------------------------- The Paxα/β group is the most recently discovered Pax subfamily and has tentatively been named "Pax?" \[[@CR34]\]. Initially, sequences of the "Pax?" subfamily have been shown to be present in Ctenophora, Deuterostomia (Echinodermata, Hemichordata), Nematoda, Annelida, and Platyhelminthes \[[@CR34]\]. Another study revealed a lophotrochozoan-specific Paxβ clade \[[@CR33]\], which could also be assigned to the "Pax?" assemblage. Recently, a Paxα clade was described as supposed sister group to the lophotrochozoan-specific Paxβ clade \[[@CR35]\]. The Paxα clade comprises sequences of Panarthropoda as well as Deuterostomia, whereas the previously included ctenophore PaxA/B sequences received less support. Therefore, "Pax?" is considered a distinct bilaterian Pax subfamily termed "Paxα/β" \[[@CR35]\]. The ancestral role of the Paxα/β subfamily in bilaterians remains unknown. However, the *Paxβ* expression data of the single investigated lophotrochozoan so far, the leech *Helobdella*, reveal expression of two *Paxβ* homologs, *Hsp-Paxβ1* and *Hsp-Paxβ2*, in the developing CNS and, in case of *Paxβ1,* additionally at the site of eye formation \[[@CR33]\]. A similar situation was found in the onychophoran *Euperipatoides rowelli,* where *Ero-Paxα* is expressed in the CNS, more precisely in the lateral brain tissue associated with each eye of the adult animal \[[@CR35]\]. These findings are now complemented by our data on the expression of *War*-*Paxβ* in the region of the developing cerebral ganglia and the ventral nerve cords of the neomeniomorph mollusk *W. argentea*. Although the role of *Paxα/β* is obviously not exclusively assigned to developmental processes, the common expression in the CNS of all species hitherto investigated argues for an ancestral role in CNS specification at least in protostomes. Since data of *Paxα* expression in Deuterostomia are still lacking, the ancestral role of the Paxα/β subfamily for the entire Bilateria remains obscure. However, both investigated lophotrochozoans, *W. argentea* and *Helobdella austinensis*, show distinct *Paxβ* expression in the CNS of their anterior ("head") region as well as in their ventral nerve cords (present study, \[[@CR33]\]). The results for *W. argentea* are in accordance with the findings in another aculiferan mollusk, the polyplacophoran *Acanthochitona crinita*, where it is expressed in the lateral region of the cerebral commissure and in the ventral nerve cords (unpublished data). These congruent expression patterns in *H. austinensis*, *A. crinita*, and *W. argentea* argue for a conserved expression of *Paxβ* during neurogenesis of the ventral nerve cords of annelids and (aculiferan) mollusks. Expression of *Pax2/5/8, Pax6,* and *Paxβ* and homology of lophotrochozoan ventral nerve cords {#Sec22} ---------------------------------------------------------------------------------------------- Among other components, the molluscan nervous system is generally composed of an esophageal nerve ring (including a paired cerebral ganglion) and a paired pedal ganglion. From the former, two lateral (visceral) nerve cords emanate, while the latter gives rise to a pair of ventral (pedal) nerve cords. Together, these four longitudinal nerve cords form the tetraneural nervous system of Mollusca \[[@CR7], [@CR8]\]. The expression of *Paxβ*, *Pax6,* and *Pax2/5/8* in anterior and posterior domains of early developmental stages of *Wirenia argentea* (Fig. [10](#Fig10){ref-type="fig"}) reflects the morphogenetic data that show that neurogenesis starts with the outgrowth of a pair of neurite bundles from both the neuropil of the apical organ and a posterior neurogenic domain \[[@CR13]\]. Interestingly, none of the three genes investigated herein appears to be expressed in the developing lateral nerve cords of *W. argentea,* but are so in the developing ventral nerve cords (Fig. [10](#Fig10){ref-type="fig"}). This is congruent with *Paxβ* and *Pax6* expression in the polyplacophorans *Acanthochitona crinita* (unpublished data) and -- at least for *Pax6* expression -- *Leptochiton asellus* \[[@CR32]\]. Interestingly, annelids likewise show expression of *Pax2/5/8*, *Pax6*, and *Paxβ* in the ventral nerve cords \[[@CR26], [@CR33], [@CR62], [@CR63]\]. This shared expression domain in the paired ventral nerve cord of annelids and neomeniomorphs strongly suggests a common evolutionary origin and thus argues for homology of ventral nerve cords in the lophotrochozoan lineage, thus supporting the notion that one pair of ventral nerve cords was present in the last common lophotrochozoan ancestor (see also \[[@CR7]\]).Fig. 10Schematic summary of Pax gene expression in *Wirenia argentea* Conclusion {#Sec23} ========== Our data provide first insights into the molecular background of aplacophoran neurogenesis on gene expression level and thus add significant insight into the putative roles of the herein investigated Pax genes in molluscan development. The *Pax6* expression pattern in the aculiferan clades Neomeniomorpha and Polyplacophora largely resembles the common bilaterian expression during CNS development. The expression of *Paxβ* in the CNS of the neomeniomorph *Wirenia argentea* and the leech *Helobdella austinensis* argues for an ancestral role in lophotrochozoan neurogenesis, in particular during formation of the cerebral ganglia and the ventral nerve cords. Furthermore, we found indication for a conserved role of *Pax2/5/8* in CNS development in Bilateria, while its expression in the spicule-secreting or associated cells in both neomeniomorphs and polyplacophorans suggests a novel function of this gene in aculiferan skeletogenesis. None of the investigated Pax genes are involved in the development of the lateral (visceral) nerve cords in either Neomeniomorpha or Polyplacophora, suggesting a different molecular background of this tetraneural subset on the one hand and homology of the ventral nerve cords within Spiralia (or Lophotrochozoa) on the other. We thank Henrik Glenner (Department of Biology, University of Bergen, Norway) for providing boat time, laboratory space, and logistic support. We thank the crew of the RV Hans Brattström (University of Bergen) for assistance with collection of animals. MS thanks Martin Fritsch for providing valuable advice on confocal laser scanning microscopy and the IMARIS software. André Luiz de Oliveira gratefully acknowledges the financial support of CAPES and the Brazilian program Science without Borders (CsF -- project 6090/13-3). This study was supported by a grant of the Austrian Science Foundation (FWF) to AW (grant number P24276-B22). Availability of data and materials {#FPar1} ================================== All sequence data used in this data are available through GenBank. Authors' contributions {#FPar2} ====================== MS and AW designed the study. MS designed the *Wirenia argentea* primers, amplified and cloned the gene fragments, synthesized the probes, performed *in situ* hybridization experiments, analyzed the data, created 3D reconstructions and illustrations, and drafted the manuscript. TW contributed the Pax gene sequences for the polyplacophoran *Acanthochitona crinita*. MS, ER, and TW extracted the RNA and synthesized the cDNA. ALO performed the transcriptome assembly and Pax gene phylogeny. MS, ER, CT, and TW reared the study material in Bergen, Norway. CT coordinated and supervised research in Bergen. AW supervised the project and contributed to data interpretation and writing of the manuscript. All authors provided input and read and approved the final version of the manuscript. Competing interests {#FPar3} =================== The authors declare that they have no competing interest. Consent for publication {#FPar4} ======================= Not applicable. Ethics approval {#FPar5} =============== The entire study was carried out adhering to national and international laws. The species used in this study is neither protected nor endangered and work on it not restricted in any way.

Background ========== Meningiomas are generally slow-growing lesions that arise from intracranial and spinal meninges. They are usually perceived as benign tumours for which radical surgery is the treatment of choice \[[@B1]\]. However, they may occasionally behave aggressively in atypical or malignant meningiomas, invading the brain and/or metastasising outside the CNS, which occurs in only 0.01% of all cases \[[@B2]\]. The most common extracranial location of metastasis is the lung followed by liver, lymph nodes and bones \[[@B3],[@B4]\]. Meningiomas present ideal targets for somatostatin receptor scintigraphy (SRS) with 111In-DTPA-octreotide. However, the value of the radioreceptor therapy using radiolabeled somatostatin analog 177Lu-DOTA-octreotate is not yet well established in patients with metastasized or inoperable meningiomas \[[@B5],[@B6]\]. Here, we present a patient with metastatic anaplastic meningioma who benefited from radiopeptide targeting. Case presentation ================= A 62 year old female with intracranial anaplastic meningioma was referred to our department for a restaging with 18F-fluorodeoxyglucose (FDG)-PET/CT. The patient suffered from a protrusio bulbi of the left eye and progressive facial pain. No conventional treatment option could be offered to the patient, who had undergone multiple surgical resections and percutaneous radiation before. The fused PET/CT images (Biograph; Siemens Medical Solutions Inc) manifested multifocal accumulation in the left temporal region with local bone infiltration. Furthermore, they demonstrated multiple pulmonary metastases in the upper lobe of the left lung (Figure [1](#F1){ref-type="fig"}). In view of these findings, including the diagnosis of pulmonary metastases, the patient was referred for SRS to evaluate the option of a palliative radiopeptide therapy with 177Lu- DOTA-octreotate. SRS images showed strong uptake in the left temporal region as well as in the upper lobe of the left lung, consistent with the PET/CT findings (Figure [2](#F2){ref-type="fig"}). Due to the abundance and high affinity of somatostatin receptors (sstr), we performed radiopeptide therapy with 177Lu- DOTA-octreotate consisting of 3 cycles (cumulative dose: 691 mCi) without any serious side effects (Figure [3](#F3){ref-type="fig"}). The patient experienced a dramatic reduction of facial pain assessed by visual analogue scale (VAS) as well as a significant improvement in quality of life with a 30% increase in her performance status using karnofsky scoring 6 weeks after commencement of the treatment. Disease stabilization could also be achieved, according to functional MD Anderson criteria, evaluated 3 months after termination of radiopeptide therapy \[[@B7]\] {#F1} {#F2} {#F3} Discussion ========== The local recurrence rate of meningioma is determined by the extent of the resection, histopathological grade and biological aggressiveness of the tumor \[[@B8],[@B9]\]. Once a meningioma recurs, it is more likely to recur again later, resulting in a poor prognosis of the patient \[[@B10]\]. 18F-FDG-PET/CT has been commonly used in patients with primary tumours of central nervous system including meningioma for tumor grading, determination of the prognosis and discrimination of tumor recurrence from radiation necrosis \[[@B11],[@B12]\]. With their high sstr density and location outside the blood-brain barrier, meningiomas also present ideal targets for SRS with 111In-DTPA-octreotide which is the main imaging technique for neuro endocrine tuomors (NETs) but may be also used in other tumors expressing somatostatin receptors such as neuroblastoma, pheochromocytoma and paraganglioma \[[@B13]-[@B15]\]. This procedure is used apart from staging and monitoring the effect of treatment for selecting patients for peptide receptor radionuclide therapy (PRRT), primarily used in gastroenteropancreatic NETs with very encouraging results. The value of PRRT is not yet well established in patients with meningiomas \[[@B5]\]. Our patient experienced a dramatic symptomatic relief as well as a significant improvement in quality of life following the PRRT along with inhibition of tumor progression. Conclusions =========== The presented case may help to establish the value of PRRT in patients with the rare condition of anaplastic meningioma. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Conception of the case report: A.S., S.E., HJ.B.; Collection and assembly of data: H.A., W.W., U.H.; Literature review and interpretation of data: A.S. S.E., H.A.; Drafting of the article: A.S., HJ.B., S.E.; Critical revision of the article for important intellectual content: U.H., H.A. W.W. All authors have read and approved the final manuscript. Consent ======= Written informed consent was obtained from the patient for publication of this Case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

Following publication of this article \[[@pone.0215587.ref001]\], the following concerns were noted: - Fig 4A GAPDH panel lanes 1--7 appear highly similar; there are vertical discontinuities between lanes 3 and 4. - Fig 7B MCP1 panel appears to contain vertical discontinuities between lanes 1, 2 and lanes 2, 3, as well as a horizontal discontinuity beneath the band in lane 1. - Fig 7B MCP1 panel lane 2 appears similar to ICAM1 panel lane 1 when horizontally flipped and stretched. - Fig 7B actin panel all lanes appear to have similar dotted patterns around the bands. For Fig 7B the authors were not able to provide raw images of all full gels. The authors commented that Fig 4A lanes 1--3 and lanes 4--7 represent separate gels and a composite image was generated for the purposes of figure presentation. The underlying images provided do not resolve all concerns noted. The authors have explained that the primary data underlying other experiments except for microarray data are not available. In light of the above-mentioned image concerns and the unavailability of underlying data, the *PLOS ONE* Editors retract the article. AFP and DI did not agree with the retraction. DLP responded but did not provide a comment on the retraction. RC, IS, AC, SS, CL, FA, and EA did not respond.

Introduction {#Sec1} ============ Swallowing is a complicated physiological process involving the recruitment of several structures in a very short time to transport a bolus from the mouth to the stomach. It plays a crucial role in food transmission, nutrition intake, physical growth and even human survival. This process is subdivided into the oral phase, the pharyngeal phase and the oesophageal phase from an anatomical point of view^[@CR1]^. Any abnormality in the process (i.e., dysphagia) would not only cause dehydration, malnutrition, weight loss, aspiration and pneumonia but could also negatively impact daily activities and quality of life^[@CR2]^. Although the oral and pharyngeal phases of swallowing are presented sequentially, the physiologic reality is that both of the phases are integrally related^[@CR3],[@CR4]^. Physiologically, the transmission of the bolus through the oral and pharyngeal cavities involves a series of systematic biomechanical motor events. However, until now, the complexity of these events and the difficulty of monitoring these structural actions are undoubtedly the major reasons for our relative lack of knowledge about the motor mechanisms involved in oropharyngeal swallowing. Therefore, in modern society, it is the pursuit of the clinician to monitor oropharyngeal swallowing behaviour and evaluate the related functions conveniently and noninvasively for the timely diagnosis of any swallowing defects. During the oropharyngeal process of swallowing, the tongue participates in collecting and positioning the food bolus^[@CR5]^ and plays a critical role in the posterior propulsion of the bolus with the help of tongue pressure arising from its contact against the hard palate^[@CR6]^. In addition, the hyoid elevation towards the base of tongue facilitates the upper oesophageal sphincter (UES) opening^[@CR7]^, consequently allowing the food pass through the pharynx to the oesophagus. Obviously, optimal oropharyngeal swallow performance requires the intricate events of the tongue and the hyoid to occur in concert with each other to transport the bolus safely and efficiently. The fine-tuned relationship between the tongue and the hyoid, which are the representative anatomical structures of the oral cavity and the larynx, could not exist without their muscular connection. Various literature has documented that the suprahyoid (SH) muscles and the infrahyoid (IH) muscles are not only involved in the oral phase, contributing to fixing and elevating the tongue^[@CR8]^, but are also involved in the pharyngeal phase when the hyolaryngeal complex elevates and returns to rest^[@CR9],[@CR10]^. Accordingly, electromyography (EMG) of certain muscles could be a useful parameter for speculating the oropharyngeal function, and the temporal actions of the tongue body and the hyoid has ever been indirectly reflected by measuring the EMG of the oral and laryngeal-related muscles^[@CR11]^. Moreover, previous studies have supplied the kinetic correlation between submental muscle activity and the hyoid by concurrent application of the submental surface EMG (sEMG) and a movement transducer on the anterior neck over the larynx^[@CR12]^ or sEMG and videofluorography (VF)^[@CR13]^. Recently, researchers took advantage of the EMG bioimpedance (EMBI) system and successfully mapped characteristic functional changes in the pharynx during swallowing, specifically laryngeal elevation^[@CR14]^. This information indeed deepened our understanding of swallowing. Currently, there is still a lack of data available that thoroughly describes the biomechanical coordination of tongue pressure generation and hyoid activity in reference to the sEMG of SH and IH muscles during oropharyngeal swallowing. Moreover, although various instruments such as VF, endoscopy, CT, ultrasound, MRI, and so on have been used to monitor the swallowing behaviour in human beings in the past several decades^[@CR8],[@CR15]--[@CR18]^, certain drawbacks including radiation, expensiveness, inconvenience, and specialization of the aforementioned appliances still limit their popularization and application. The need for the development of non-invasive methods of quantification and visual evaluation of swallowing behaviour or swallowing disorders is growing within the area of odynophagia and dysphagia management. Therefore, in this investigation, we innovatively took advantage of several non-invasive appliances simultaneously to measure the tongue pressure production, hyoid movement, and SH and IH EMG during normal oropharyngeal swallowing. We intended to use the sensing system to (1) characterise the temporal pattern of the tongue, the hyoid, and the SH and IH muscles, and (2) determine how these oropharyngeal patterns of the tongue and the hyoid are related to and coordinated with the muscle activity. All of our present efforts will shed light on the biomechanical coordination during oropharyngeal swallowing. Furthermore, it could be beneficial for clinicians to simply and efficiently evaluate oropharyngeal swallowing and provide an early diagnosis of dysphagia chair-side and bed-side. Materials and Methods {#Sec2} ===================== Subjects {#Sec3} -------- Fifteen adult male subjects with an average age of 27.7 years (range = 25--32 years) without any signs of severe malocclusion, mastication or swallowing problems, neurological disease, structural disorders or other oropharyngeal problems participated in this study. As for the subject number in the current study, we used the sample size software NCSS-PASS 11.0 (NCSS LLT, Utah, USA) to calculate. A study^[@CR19]^ reported that hyoid movement lagged the onset of tongue-palate pressures in 63.2% of swallows. This phenomenon in our study is 94.7%. So the difference is 31.5%. We used the Tests for one proportion (Non-Zero Null Hypothesis) \[Differences\] and calculated the sample size as fifteen after setting the power (1-Beta) 0.90, Alpha 0.05, PB (Baseline Proportion) 0.632, d0 (Superiority Difference) 0, and d1 (Actual Difference) 0.315. Informed consent was obtained from each participant after receiving information about the experimental procedure. The study protocol was approved by the Ethics Committee of Osaka University Graduate School of Dentistry (No. H21-E32) and the experiments were performed in accordance with the Declaration of Helsinki (2008) for humans. Measuring system and procedure {#Sec4} ------------------------------ The sensing system consists of four non-invasive devices including a tongue pressure sensor sheet (Fig. [1a](#Fig1){ref-type="fig"}), surface electrodes (Fig. [1b](#Fig1){ref-type="fig"}), bend sensor (Fig. [1c](#Fig1){ref-type="fig"}) and a microphone (Fig. [1d](#Fig1){ref-type="fig"}). The tongue pressure produced in the midline and the posterior-lateral part of the hard palate was recorded by the tongue pressure sensor sheet (100 Hz, Nitta, Osaka, Japan) with a thickness of 0.1 mm and 5 measuring points (Ch.1--5). Specifically, Channel 1 to Channel 3 (Ch.1-Ch.3) were placed along the median line anteroposteriorly, and Channel 4 and Channel 5 (Ch.4 and Ch.5) were situated in the posterior--circumferential parts of the hard palate. The sensor sheet was attached to the hard palate by a sheet-type denture adhesive (Touch Correct II, Shionogi, Tokyo, Japan) after the suitable selection from three sizes according to the participant's palate form^[@CR20],[@CR21]^. We calibrated the sensor sheet by applying negative pressure using a vacuum pump through an air duct in the sensor sheet cable before measurement.Figure 1Schematic representations of the sensing system and experimental set-up. (**a**) A subject with tongue pressure sensor sheet, surface electrodes, bend sensor and microphone. (**b**) Tongue pressure sensor sheet. (**c**) Surface electrodes. (**d**) Bend sensor. (**e**) Microphone. For the EMG recording of the suprahyoid and infrahyoid muscle groups (SH EMG and IH EMG), five surface electrodes (Duo-trode, Myotronics, MA, USA) were utilized for each subject. The subject's skin was scrubbed to attach the surface electrodes. Since no side-to-side difference was found in the EMG of muscles involved in swallowing in healthy subjects^[@CR22]^, one pair of electrodes (d = 8 mm, interelectrode distance = 2 cm) was attached to the skin of the right side of the anterior belly of the digastric muscle to measure the EMG of the suprahyoid muscle group. Another pair of electrodes (d = 8 mm, interelectrode distance = 2 cm) was attached to the right side of sternohyoid muscle to measure the EMG of the infrahyoid muscle group. A reference electrode affixed to the forehead served as the ground. Signals from the EMG electrodes were band-passed filtered (100 Hz-10 kHz), amplified (BA1104, Nihon Kohden, Tokyo, Japan), full-wave rectified, and then stored on a computer through an interface (PCI-3133A, Nihon Santeku, Osaka, Japan) at a sampling rate of 10 kHz. To record the hyoid activity, the bend sensor (73.7 mm × 6.4 mm × 1.0 mm, 1000 Hz, MaP 1783BS1-056, Nihon Santeku Co. Ltd., Japan) was fixed along the midline of the frontal neck with its tip at the level of the prominence of thyroid cartilage when reaching the highest position during swallowing. It could flex physically with the laryngeal motion, and the hyoid activity could be retrieved non-invasively from the produced signal waveform^[@CR23]^. Because the symmetry of the swallowing sound could be acquired bilaterally^[@CR24]^, we placed a microphone (JM-0116, Ono-Sokki, Tokyo, Japan) over the left lateral border of trachea immediately inferior to the cricoid cartilage to detect the timing of bolus passage through the entrance of oesophagus^[@CR25]^. The subjects examined in this study were instructed to sit in an upright position with their heads supported by a headrest to avoid head retroflexion and to keep the Frankfort plane horizontal with their feet touching the floor. Then, 5 ml of water (37 °C) was given via syringe and held on the mouth floor until swallowed wholly, one time, upon verbal command. The participant was asked to relax the tongue immediately after each trial. Three repetitions were performed for each subject. The recorded tongue pressure and swallowing sound data were subsequently integrated on a personal computer through an interface board (PCD 100 A, Kyowa Electric Instruments, Tokyo, Japan). The EMG data and the obtained signal from the bend sensor were amplified and stored on the personal computer through a separate interface board (PCI-3133A, Nihon Santeku, Osaka, Japan). To ensure that all of the subjects felt comfortable and that all the devices worked properly, at least one successful practice swallow was completed before recording the experimental data. To synchronize all the data, the trigger signal to start measurement from the swallow scan was sent to the interface board (PCI-3133A, Nihon Santeku, Osaka, Japan); then, tongue pressure, EMG, hyoid motion and swallowing sound were measured at the same time. Data analysis {#Sec5} ------------- Figure [2a](#Fig2){ref-type="fig"} shows the representative raw waves of tongue pressure, EMG of suprahyoid and infrahyoid muscles, laryngeal movement and swallowing sound. The following parameters of tongue pressure on each sensor were recorded (Fig. [2b](#Fig2){ref-type="fig"}): time of tongue pressure onset (TP~on~), time of maximum tongue pressure (TPmax), time of tongue pressure offset (TPoff), peak value of tongue pressure (TPpeak) and duration of tongue pressure production (DTP). EMG bursts were full-wave rectified and smoothed (time constant, 20 ms) using the application (MaP1038A, Nihon Santeku, Osaka Japan). The onset, offset, peak value and duration of suprahyoid muscle activity (SHon, SHoff, SHpeak and DSH) and also infrahyoid muscle activity (IH~on~, IH~off~, IH~peak~ and DIH) were measured as the EMG parameters (Fig. [2c,d](#Fig2){ref-type="fig"}). The onset time of each EMG burst was the time at which it was beyond 2 standard deviations (SDs) of baseline activity, and the offset time was the time at which it was below 2 SDs^[@CR22]^. Additionally, we recorded certain time points on the laryngeal movement waveform produced by the bend sensor, i.e., T1, T2, T4, T5 and T6, to represent the onset of slight movement of the hyoid, the onset of rapid movement of the hyoid, the onset of the stationary phase of the hyoid, the offset of the stationary phase of the hyoid, and the offset of movement of the hyoid, respectively (except for T3 and T7, which were confirmed to be meaningless for hyoid activity) (Fig. [2e](#Fig2){ref-type="fig"})^[@CR23]^. Additionally, the times for T1-T5 and T2-T5 were measured because they could represent the duration from the onset of slight movement of the hyoid to the offset of the stationary phase of the hyoid and the onset of rapid movement of the hyoid to the offset of the stationary phase of the hyoid, respectively. With respect to the swallowing sound, because more than one spike was typically observed in the sound data, the spike with the greatest amplitude was chosen to be the reference time^[@CR21]^ for comparing the temporal sequence of biomechanical events during oropharyngeal swallowing.Figure 2Representative recordings of the noninvasive sensors. (**a**) Decomposed graph of waves of tongue pressure, EMG, laryngeal movement and swallowing sound. (**b**) The data analysis of TP~on~, TP~max~, TP~off~ and DTP. (**c**) The data analysis of SH~on~, SH~off~, SH~peak~ and DSH. (**d**) The data analysis of IH~on~, IH~off~, IH~peak~ and DIH. (**e**) Laryngeal signal waveform and marked time point. Statistics {#Sec6} ---------- All the data from 45 trials (15 subjects × 3 trials) were analyzed with SPSS 16.0 software (SPSS Inc., Chicago, IL, USA). A one-way analysis of variance (ANOVA) was used to compare the durations of certain physiological activities. To evaluate the sequential order of tongue pressure, muscle EMG and hyoid motion, the uniformity of variance was determined by the Kolmogorov-Smirnov test first. When uniform variance was found, significant differences were determined by repeated-measures ANOVA, and comparison testing was performed with the Bonferroni post hoc test. The interclass correlation coefficient was used to evaluate the correlations between the swallowing-related muscle activity (SH~on~, IH~on~, SH~off~ and IH~off~) and certain biomechanical events of tongue pressure (TP~on~ of Ch.1--5 and TP~off~ of Ch.1--5) and hyoid activity (T1, T2 and T5). All the data were expressed as mean ± S.D. Statistical significance was set at p \< 0.05. Results {#Sec7} ======= Duration of biomechanical events during oropharyngeal swallowing {#Sec8} ---------------------------------------------------------------- As shown in Fig. [3](#Fig3){ref-type="fig"}, the DSH, DIH, DTP for Ch. 1, Ch.2, Ch. 3, Ch.4, Ch. 5 and the times for T2-T5 and T1-T5 were 1.05 ± 0.29 s, 0.79 ± 0.31 s, 0.71 ± 0.30 s, 0.60 ± 0.26 s, 0.44 ± 0.19 s, 0.69 ± 0.28 s, 0.73 ± 0.33 s, 0.59 ± 0.21 s and 1.08 ± 0.26 s, respectively. The time is longer for DSH when compared with DIH, DTP for Ch.1, Ch.2, Ch.3, Ch.4, Ch.5 and T2-T5 (all p \< 0.001). Similar results were also found between T1-T5 and DIH, DTP for Ch.1, Ch.4, Ch.5 and T2-T5 (all p \< 0.001). The DTP for Ch.3 shows the shortest duration. The significances were confirmed between the DTP for Ch.3 and the other events (all p \< 0.005) except for DTP of Ch.2 and T2-T5 (p \> 0.05). In addition, there were no significant differences in the durations between DSH and T1-T5 (p \> 0.05), between the DTP for Ch.3 and T2-T5 or among DIH and the DTP for Ch.1, Ch.4 and Ch.5 (all p \> 0.05).Figure 3Duration of biomechanical events during oropharyngeal swallowing. DSH, duration of suprahyoid muscle activity; DIH, duration of infrahyoid muscle activity. \*p \< 0.001 v.s. DSH and T1-T5; ^\#^p \< 0.001 v.s. TP of Ch.3; ^&^p \< 0.005 v.s. DTP of Ch.3. Temporal sequence of biomechanical events during oropharyngeal swallowing {#Sec9} ------------------------------------------------------------------------- As shown in Fig. [4](#Fig4){ref-type="fig"}, the slight movement of the hyoid (T1, −0.97 ± 0.27 s) occurred first among all of the monitored biomechanical events and most closely to the subsequent SH~on~ (−0.83 ± 0.25 s) (p \> 0.05). Then, the TP~on~ of Ch.1 (−0.61 ± 0.21 s) appeared with the simultaneous appearances of IH~on~ (−0.55 ± 0.24 s), TP~on~ of Ch.5 (−0.52 ± 0.22 s) and Ch.4 (−0.51 ± 0.19 s), as well as upward movement of the hyoid (T2, −0.48 ± 0.28 s) (all p \> 0.05), then followed by the TP~on~ of Ch.2 (−0.49 ± 0.21 s) and Ch.3 (−0.39 ± 0.19 s) (p = 0.11, p \< 0.001, respectively). Though the TP~max~ (Ch.1, −0.33 ± 0.25 s; Ch.2, −0.32 ± 0.22 s; Ch.3, −0.27 ± 0.21 s; Ch.4, −0.30 ± 0.22 s; Ch.5, −0.28 ± 0.19 s) occurred before the onset of the stationary phase of the hyoid (T4, −0.08 ± 0.21 s) during swallowing, no significant differences were found among them (all p \> 0.05). In addition, the biomechanical events of the offset of the stationary phase of the hyoid (T5,0.12 ± 0.17 s), SH~off~ (0.10 ± 0.28 s), IH~off~ (0.16 ± 0.23) and TP~off~ (Ch.1, 0.10 ± 0.23 s, Ch.2, 0.12 ± 0.24 s; Ch.3, 0.06 ± 0.19 s; Ch.4, 0.18 ± 0.25 s, Ch.5,0.20 ± 0.22 s) appeared without any significant time lags (all p \> 0.05).Figure 4Temporal sequence of biomechanical events during oropharyngeal swallowing. The red line is the swallowing sound that was chosen to be the reference time. Correlation coefficient of biomechanical events during oropharyngeal swallowing {#Sec10} ------------------------------------------------------------------------------- As shown in Table [1](#Tab1){ref-type="table"}, the positive correlations between SH~on~ and T1, SH~on~ and T2, as well as IH~on~ and T2 were noted with moderate correlation coefficients (r = 0.658, p = 0.002; r = 0.616, p = 0.008; r = 0.666, p = 0.005, respectively). In addition, there were significant positive correlations between SH~off~ and TP~off~ at Chs. 1-5 and T5 with moderate correlation coefficients (r = 0.653 p = 0.002; r = 0.594, p = 0.019; r = 0.626, p = 0.008; r = 0.633, p = 0.007; r = 0.613, p = 0.010; r = 0.694, p = 0.001, respectively). This was also the case between IH~off~ and TP~off~ at Chs. 1-5 and T5 (r = 0.656, p = 0.002; r = 0.643 p = 0.003; r = 0.640, p = 0.004; r = 0.580, p = 0.026; r = 0.689, p = 0.001; r = 0.602, p = 0.010, respectively).Table 1Correlation coefficient of biomechanical events during oropharyngeal swallowing.Events of muscle EMGEventsof the tongue pressure and hyoid activity*rp*SH~on~Ch.1 TP~on~0.4720.058Ch.2 TP~on~0.4340.064Ch.3 TP~on~0.4150.070Ch.4 TP~on~0.3770.112Ch.5 TP~on~0.1920.206T10.6580.002T20.6160.008IH~on~Ch.1 TP~on~0.5430.025Ch.2 TP~on~0.3020.152Ch.3 TP~on~0.2770.175Ch.4 TP~on~0.2520.199Ch.5 TP~on~0.1980.251T10.2580.193T20.6660.005SH~off~Ch.1 TP~off~0.6530.002Ch.2 TP~off~0.5940.019Ch.3 TP~off~0.6260.008Ch.4 TP~off~0.6330.007Ch.5 TP~off~0.6130.010T50.6940.001IH~off~Ch.1 TP~off~0.6560.002Ch.2 TP~off~0.6430.003Ch.3 TP~off~0.6400.004Ch.4 TP~off~0.5800.026Ch.5 TP~off~0.6890.001T50.6020.010 Discussion {#Sec11} ========== In the present study, a pressure sensor was placed in the oral cavity to measure tongue pressure generated from the contact between the tongue and the hard palate^[@CR20]^. In addition, a bend sensor was attached to the midline skin of the neck to reflect hyoid motion with its recorded waveform^[@CR21],[@CR23]^, and the surface EMG was conducted to explore the activities of swallowing-related muscles^[@CR26]^. Furthermore, a microphone that could detect the sound when a bolus passes through the entrance of the oesophagus was used for the sound recordings^[@CR25]^. With these noninvasive appliances, we built a sensing system that successfully and synchronously measured certain essential oropharyngeal swallowing events without causing any discomfort to the subjects or disturbing their swallowing behaviour. The motor pattern and the temporal coordination of these representative oropharyngeal events were also adequately clarified. We compared the durations of biomechanical events during oropharyngeal swallowing and observed similar persistent periods of suprahyoid muscle activity and hyoid motion (T1-T5). This is consistent with previous studies that reported the movement synchronization of the suprahyoid muscle and the hyoid used EMG and VF^[@CR27]^ or EMG and CT^[@CR28]^. Therefore, the data indirectly indicate the sensitivity and the accuracy of the bend sensor to reflect the hyoid motion. Combined with the aforementioned reports, we consider the suprahyoid muscle as the main driving force of hyoid activity during swallowing. Although there was no difference between the EMG peak value of the suprahyoid muscle (0.040 ± 0.010 mV) and the infrahyoid muscle (0.037 ± 0.017 mV) in the current study, the longer duration of suprahyoid muscle activity than that of the infrahyoid muscle was observed. This phenomenon may arise from the potential and function of the two muscles during swallowing. Additionally, the durations of tongue pressure in the anterior and lateral parts corresponded well with those reported by Hori *et al*.^[@CR29]^ and Tamine *et al*.^[@CR30]^. As for the sensor signals obtained by the sensing system, we could observe certain rules. The tongue pressure signal peaked quickly, and then decreased gradually before disappearing almost simultaneously at each measured part of the hard palate. In addition, the EMG of swallowing-related muscles exhibited similar pattern with quick rise and slow descent. The sensor recorded hyoid activity produced a regular "V"-shaped waveform, with a preliminary movement at the beginning represents the onset of slight movement of the hyoid (T1), followed by rapid downward movement represents the onset of rapid movement of the hyoid (T2) with a subsequent small and obvious notch represents the onset of the stationary phase of the hyoid (T4) until reaching its peak point represents the offset of the stationary phase of the hyoid (T5). Then, the waveform reversed, quickly at first and then slowly after a turning point represents the offset of movement of the hyoid (T6). Finally, it returned to the baseline (T7)^[@CR23]^. From the view of the initial rapid motion and succeeding slow motion of the monitored organs, we could speculate that more effortful work is needed for triggering the oropharyngeal swallowing, and the later movements is mainly for maintaining swallow smoothly. With regard to the motor pattern of these representative oropharyngeal events, we noted that the slight movement of the hyoid (T1) was the first monitored biomechanical event during normal swallowing and that it even occurred a little earlier than the SH~on~, but without any significance of time lag. Because the subject in the present study needed to dip the water from the floor of the mouth to the supra-lingual location after the verbal command of swallowing^[@CR31]^, the precise activity of the tongue tip may contribute to the subtle motion of the hyoid via tongue extrinsic musculature that connects them^[@CR32]^. Taniguchi and his colleagues^[@CR4]^ have documented the suprahyoid muscle EMG burst prior to the motion of the anterior and posterior tongue. Our findings not only support this report but also confirm the fact that SH~on~ preceded the lateral tongue activity, i.e., the burst of the suprahyoid muscle was earlier than that of tongue pressure. Once the tongue pressure is produced because of the tongue-hard palate approximation, the hyoid needs to elevate superior-anteriorly to facilitate the subsequent laryngeal vestibule closure, epiglottis reversion and UES opening for the upcoming food bolus^[@CR33]--[@CR35]^. Therefore, it is reasonable to observe the simultaneous appearances of tongue pressure and upward movement of the hyoid (T2). Meanwhile, the infrahyoid muscle activity occurred at this time. Palmer *et al*. previously confirmed that little infrahyoid muscle activity occurs when each swallow starts until the hyoid moves sharply^[@CR8]^. Although the tongue pressure peak value arose before the onset of the stationary phase of the hyoid (T4) during swallowing, significant differences were not found. This suggests that, in the process of oropharyngeal swallowing, powerful tongue-hard palate contact is essential for anchoring the hyoid on the one hand, but on the other hand, the hyoid should stay at its highest position to open the pharyngeal cavity to the greatest extent accordingly in favour of receiving the bolus that is passing the fauces when the tongue pressure reaches its maximum^[@CR5]^. With the sensing system, we also observed the concurrent offset of the stationary phase of the hyoid (T5), EMG of swallowing-related muscles and tongue pressure. Physiologically, the supra- and infra- hyoid muscles participate in the movement of the tongue and the hyoid. As soon as the muscles EMG ceased, the tongue pressure disappeared, and the hyoid consequently started to retreat from the highest position in this study. From the results of the interclass correlation coefficient, we noticed that the positive correlations between the EMG burst and hyoid activities, i.e., SH~on~, correlated well with the onset of slight/rapid movement of the hyoid, and IH~on~ correlated well with the onset of rapid movement of the hyoid. Based on the structural properties of the suprahyoid muscles and their potential for moving the hyoid^[@CR28],[@CR36]^, the suprahyoid muscle provides the primary power to displace the hyoid in the anterior and superior directions. This may contribute to the high correlations between SH~on~ and T1 and T2. As for the close relationship between IH~on~ and T2, we speculated that the infrahyoid muscle may play a role in counterbalancing the suprahyoid muscle to stabilize the tongue and the hyoid during the swallowing process. However, no such case was observed for the SH~on~ or the TP~on~. This coincided with previous findings with surface electrodes and a midline disk-shaped pressure sensor showing that no significant correlation exists between the onset of the suprahyoid muscle EMG burst and the tongue tip touching the palate^[@CR4]^. Interestingly, the absent time of swallowing-related muscles EMG positively correlated with not only the offset of tongue pressure at any site but also with the offset of the stationary phase of the hyoid. All the results suggest that there must be precise coordination in the motion between the tongue and the hyoid in healthy oropharyngeal swallowing^[@CR21]^ because of the muscular connection. These interclass results suggest that we should be alert to any abnormal patterns in suprahyoid or infrahyoid muscles during swallowing because this might demonstrate dyskinesia of the tongue and the hyoid and vice versa. Previous documents have reported that gender and age exert influences on swallowing behaviour^[@CR37],[@CR38]^ and that the swallowing process would be affected by food properties and body positions^[@CR3],[@CR39]^. Therefore, some limitations exist in the present study as the subjects recruited were all males with sitting in an upright position and swallowing just water bolus. Additionally, our present findings should be confirmed in the subjects with different swallowing behavior and compared with patients with swallowing problems. These issues during swallowing will be addressed in our future study design and performance. In conclusion, the present results represent synchronous data from tongue pressure, muscle EMG and hyoid movement with non-invasive equipment that successfully exhibited the motor pattern and the temporal coordination of certain representative oropharyngeal events in healthy male subjects. The sensing system could be a good candidate for monitoring and evaluating the oropharyngeal process of swallowing and also has the potential to be useful for clinical work in dysphagia evaluation and rehabilitation. Takahiro Ono and Yongjin Chen contributed equally to this work. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The project was supported by the Grants-in-Aid from the Ministry of Education, Culture, Sports, Science and Technology of Japan (No. 24659859) and the International Scientific and Technological Cooperation and Exchange Program in Shaanxi province of China (No. 2014KW19-01). T.O., Y.C. and K.H. designed the experiments. Q.L., Y.M. and S.F. conducted the experiments and collected the data. Q.L. and K.H. analyzed the data. Q.L. and Y.C. wrote the manuscript. Y.M. assisted in conducting experiments and analyzing the data. T.O. and K.H. edited the manuscript. All authors reviewed the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

1. Introduction =============== More than 70% of breast reconstruction is done by prosthesis-based method.^\[[@R1]\]^ One of its most unwanted complications of prosthesis-based breast reconstruction is infection, which usually results in reconstruction failure.^\[[@R2]\]^ A closed suction drain in the surgical field can decrease seroma formation and possibly reduces the risk of infection.^\[[@R3],[@R4]\]^ However, prolonged drain duration may increase infection rate due to ascending infection.^\[[@R5]--[@R7]\]^ The optimal timing of drain removal after prosthesis-based breast reconstruction is still unclear. Most surgeons (87.4%) recommend drain removal when the daily drainage volume at the time of drain removal (defined as last daily drainage volume) drops below 30 mL.^\[[@R3],[@R8],[@R9]\]^ Some recommend 20 mL, or 50 mL.^\[[@R3],[@R10],[@R11]\]^ The object of this study is to identify the optimal timing of drain removal in terms of infection control. We hypothesized that drain duration plays a bigger role than last daily drainage volume in determining the timing of drain removal. 2. Materials and methods ======================== This retrospective cohort study was approved by the Institutional Review Board at the Koo Foundation Sun Yat-Sen Cancer Center, Taipei, Taiwan. The records of patients who underwent immediate prosthesis-based breast reconstruction in Koo Foundation Sun Yat-Sen Cancer Center from 1998 to 2013 were collected. Autologous breast reconstruction and delayed breast reconstruction were excluded. The goal of this study was to evaluate the influence of drain duration and last daily drainage volume on the infection rate of prosthesis-based breast reconstruction. Those who did not use any drain after breast reconstruction and those with incomplete record of drain duration or last daily drainage volume were excluded from this study. The outcome variable was infection. Infection herein was defined when the tissue expander was removed due to clinical symptoms and signs of infection, or when the seroma was positive for bacterial culture, yet the prosthesis was salvaged by antibiotic treatment. In 2-stage breast reconstruction, infection was counted only when tissue expander was still in place. In 1-stage breast reconstruction, infection was counted within 1 year of surgery. Infections sometimes might happen several years after reconstruction, which was less likely related to the drain and was not counted in this study. The main independent variables were drain duration and last daily drainage volume. Covariates in the analysis included age, history of diabetes mellitus (DM), history of smoking, body mass index (BMI), breast weight, size of tissue expander, size of drain, number of retrieved lymph node (LN), tumor size, number of metastasized lymph node, tumor stage, mastectomy type, reconstruction type, submuscular implantation, skin defect, operative time, duration of antibiotics use, chemotherapy, and radiotherapy. We do 1-stage prosthesis-based breast reconstruction when the excised breast weight is relatively small and the preserved skin is enough to close the wound without tension. Although our practice is to always preserve as much skin as possible when a breast reconstruction is considered, the term "skin sparing mastectomy" is not used. In this cohort, no patient received radical mastectomy. The mastectomy type is hence categorized into modified radical mastectomy, simple mastectomy with or without sentinel lymph node biopsy. Skin defect, in this study, is defined as skin necrosis or wound dehiscence that required surgical intervention. When the serratus anterior muscle was elevated from the ribs and the tissue expander was placed under pectoralis major muscle as well as serratus anterior muscle, it was classified as submuscular group. The rest of the tissue expanders were placed under pectoralis major muscle only. Chemotherapy referred only to the postoperative chemotherapy when the tissue expander was still in place. Neither the preoperative chemotherapy nor chemotherapy after the second stage reconstruction was counted. In the 1-stage reconstruction, only chemotherapy within 1 year of surgery was counted. Radiotherapy referred to the radiation given when the tissue expander was in place or before reconstruction. Those who had radiotherapy before reconstruction were the patients who had undergone breast conserving surgery and radiotherapy before and received a mastectomy due to local recurrence later. In the 1-stage reconstruction, radiotherapy within 1 year of surgery was counted. In the univariable analysis, *χ*^2^ test and Fisher exact test were used for categorical variables and univariable logistic regression test for continuous variable. Multicollinearity existed among the covariates where the breast weight, BMI, and size of tissue expander were highly correlated; mastectomy types were correlated with the number of retrieved LN; and the stage represented tumor size and number of metastatic LN. Therefore, multivariable logistic regression using the forward stepwise method was performed to identify the significant variables associated with infection rate. All statistical analyses were performed using IBM SPSS, version 20.0 (IBM, Armonk, NY). *P* value less than 0.05 was used for statistical significance of variables. 3. Surgical methods =================== After completion of mastectomy, axillary sentinel lymph node sampling, or radical axillary lymph node dissection, the plastic surgeon takes over for the reconstruction procedure. The skin around the surgical field is disinfected again, and a new set of sterile surgical instruments is served. The plane under pectoralis major is opened. When the serratus anterior is not elevated, we laterally advance and suture the lateral part of the pectoralis major muscle to the lateral edge of the wound so the tissue expander is shielded from the skin wound by a muscle layer. In this study cohort, no acellular dermal matrix (ADM) was used. After hemostasis is achieved, the pocket is irrigated with diluted aqua betadine solution. One gram of cephalosporin is spread onto the surgical field before the tissue expander is placed. Around 20 to 100 mL normal saline is instilled into the tissue expander. A closed suction drain is placed which exits through a separate skin incision at least 5 cm distal to the inferior edge of the pocket for a 5 cm subcutaneous tunnel to prevent or delay ascending infection. We place only 1 drain at the lateral part of the pocket. In the early period of this cohort, prophylactic antibiotics were used until the drain was removed. We gradually shortened the duration of antibiotics after embracing the Surgical Care Improvement Project.^\[[@R12]\]^ In the middle period of this cohort, postoperative prophylactic antibiotic was used for only 1 day postoperatively. In recent years, only 1 dose 30 minutes before operation was used. Almost all patients were discharged 24 hours after the operation with a drain in situ and then returned for a follow-up visit within a week. The drain was removed when the daily drainage volume reached below 30 mL in the early period of the cohort. However, this practice gradually changed to having the drain removed during the first follow-up visit around postoperative day 7 even if the last daily drainage volume was more than 100 mL. We usually began injecting the tissue expander on or around postoperative day 14, which was the second follow-up visit when the fluid accumulation in the pocket had decreased. In a few patients, expansion would be delayed to around postoperative day 21, because their fluid is still accumulating during the postoperative week 2. Postoperative rehabilitation of the shoulder is not encouraged until the fluid accumulation decreases. When symptoms and signs of infection occur, the surgical wound is usually opened to evaluate the condition inside the pocket and the fluid is sent for gram stain. If the fluid is turbid or gram stain is positive for bacteria, the tissue expander is removed in addition to adequate antibiotic treatment. 4. Results ========== From 1998 to 2013, there were 569 immediate breast reconstruction fulfilling the inclusion criteria. The minimal follow-up duration of all patients was 2 years. Infection occurred in 29 breasts. The total infection rate was 5.1%. There was only 1 patient whose tissue expander was preserved as the fluid was clear and gram stain was negative. Culture of the fluid of this patient did come back positive on a later day. She was successfully treated with antibiotics only. The mean ± standard deviation (SD) and median of drain duration were 13.6 ± 7.6 days and 12 days, ranging from 3 to 54 days. The mean ± SD and median of last daily drainage volume were 42.8 ± 22.9 and 38 mL, ranging from 1 to 170 mL. As most surgeons removed drain when the last daily drainage volume was less than 30 mL, we divided the cohort into 2 groups by the last daily drainage volume equal to or less than versus more than 30 mL. *χ*^*2*^ test showed no significant difference in the infection rate between these 2 groups (*P* = 0.32). We further looked at the last daily drainage volume of various cutoff values of 10, 20, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, and 150 mL. All of which showed no significant difference in the infection rate (*P* = 0.67, *P* = 0.57, *P* = 0.17, *P* = 0.20, *P* = 0.22, *P* = 0.14, *P* = 0.43, *P* = 0.29, *P* = 0.30, *P* = 0.27, *P* = 0.19, *P* = 0.10, *P* = 0.10, and *P* = 0.10 respectively). We divided patients into 5 groups based on the volume of last daily drainage volume at 20 mL incremental difference, that is, ≦20, ≦40, ≦60, ≦80, and \>80 mL. The results show no significant difference (*P* = 0.30, Fig. [1](#F1){ref-type="fig"}). {#F1} By contrast, when this cohort was divided into 2 groups based on drain duration, a significant statistical difference on infection rate was found between 2 groups with drain duration equal to or shorter than versus longer than 21 days (3.8% vs 14.3%, *P* = 0.001). In the later period of this cohort, we almost always removed the drain on the first follow-up visit, which was around postoperative day 7. Therefore, we divided the patients into 2 groups by drain duration of equal to or shorter than versus longer than 7 days. We did not find a significant difference of infection rate between the 2 groups (4.3% vs 5.3%, *P* = 0.82). We divided the cohort into 5 groups based on the duration of drain retention with an incremental interval of 1 week, that is, ≦1, ≦2, ≦3, ≦4, and \>4 weeks. Figure [2](#F2){ref-type="fig"} shows significant difference of infection rate in these 5 groups (*P* = 0.005). {#F2} Table [1](#T1){ref-type="table"} shows the characteristics of the patients and their diseases that were uncontrollable factors that might affect the infection rate. Between the infected and noninfected groups, the breast weight, BMI, and number of retrieved LN were significantly different, but age, tumor size, number of metastasized LN, and stage of the disease showed no significant difference. Mastectomy type showed only a borderline effect on the infection rate. This could be due to the difference in the number of axillary LN retrieved. No infection was seen in the 44 patients who underwent simple mastectomy. ###### Patients and diseases characteristics.  In this cohort, DM did not increase the infection rate. Only 3 (0.5%) patients were diabetic in this cohort which may have compounded the analysis. The mean excised breast weight in the 1-stage reconstruction group is 383 g as opposed to 424 g in the 2-staged reconstruction group (*P* = 0.33). Table [2](#T2){ref-type="table"} shows the characteristics of operation and postoperative treatments which were relatively controllable factors. In the univariable logistic regression analysis, when both the drain duration and last daily drainage volume were considered continuous variables measured by day and milliliter, respectively, the drain duration significantly influenced the infection rate (*P* = 0.001), but the last daily drainage volume did not (*P* = 0.16). Chemotherapy significantly increases the infection rate (*P* = 0.01), but radiotherapy does not increase the infection rate (*P* = 0.16). Tissue expander size significantly influenced the infection rate (*P* = 0.03). Tissue expander size is highly correlated with breast weight and BMI. Among these, breast weight is the most distinct clinical factor. One-stage or 2-stage reconstruction, submuscular implantation, skin defect, operative time, and duration of postoperative antibiotic use did not significantly influence the infection rate. ###### Operative and postoperative characteristics.  In the multivariable logistic regression analysis using the forward stepwise method, the drain duration was the first variable that was selected into the model followed by breast weight and chemotherapy. We found that the odds ratio of infection increased by 76.2% with each additional week of drain duration (*P* \< 0.001), or by 8.1% with each additional day of drain duration (*P* = 0.001, not shown in tables), after adjusting for breast weight, antibiotic duration, chemotherapy, radiotherapy, drain size, age, number of retrieved LN, number of metastasized LN, operative time, tumor size, mastectomy type, tissue expander size, reconstruction type, submuscular implantation, skin defect (Table [3](#T3){ref-type="table"}). Breast weight and chemotherapy also showed significant influence on odds ratio of infection (*P* = 0.006, *P* = 0.04 respectively). By contrast, last daily drainage volume did not affect the infection rate in multivariable analysis. (*P* = 0.28, not shown in Table [3](#T3){ref-type="table"}). When this categorical variable of last daily drainage volume (≦30 mL vs \>30 mL) was changed to continuous variable, using milliliter as a unit, there was still no significant association between last daily drainage volume and infection rate in multivariable logistic regression analysis (*P* = 0.30, not shown in Table [3](#T3){ref-type="table"}). ###### Multiple logistic regression analysis evaluating risk factors for infection rate.  5. Discussion ============= In this study, we found that drain duration was an important factor on infection rate of prosthesis-based breast reconstruction. The odds ratio of infection increases by 76.2% with each additional week of drain duration. A drain duration over 21 days is significantly more likely to cause infection. Our current practice dictates drain removal on postoperative day 7, even if the last daily drainage volume is relatively high. This cohort study showed no significant correlation between infection rate and the last daily drainage volume. Therefore, the findings support the safety guideline of our current practice of removing the drain on or around postoperative day 7 even if the daily drainage is more than 30 mL. Traditionally, draining is a routine procedure after modified radical mastectomy.^\[[@R13]\]^ In most cases, 1 or 2 drains are placed.^\[[@R3],[@R8]\]^ Because early drain removal may increase seroma formation, it is common to wait until the last daily drainage volume falls below 20, or 30 mL before removing the drain.^\[[@R14]--[@R17]\]^ The purpose of postmastectomy draining is to detect postoperative bleeding that usually stops within 48 hours. It is also used to decrease the amount of fluid accumulation. Fluid accumulation prevents the skin flaps from adhering to the underlying pectoralis major muscle and delays healing of the surgical pocket after removing breast tissue. Seroma formation is positively correlated with infection rate.^\[[@R3]\]^ In addition, a large seroma may distend the overlying skin and jeopardize the circulation of the skin flaps leading to wound dehiscence or marginal necrosis with increased risk of infection.^\[[@R18]\]^ This common practice of postmastectomy drain management has been adopted for postreconstruction care. However, such a rationale is debatable when a tissue expander is placed in the surgical field. A tissue expander occupies most of the surgical space as an intention to create the desired pocket by preventing healing around the tissue expander. A seroma around the tissue expander does increase the volume of dead space but may not significantly increase the risk of infection compared with the scenario where seroma occurs in the mastectomy wound in the absence of a tissue expander. A foreign body per se increases the susceptibility of infection.^\[[@R19]\]^ Prosthesis-based breast reconstruction carries more infection risks than mastectomy alone. Olsen et al^\[[@R20]\]^ reported a 12.4% incidence of surgical site infection following mastectomy with immediate implant reconstruction, but 4.4% following mastectomy only. In the event of bacterial contamination during the operation, the existence of a foreign body alone, namely the tissue expander, may result in infection.^\[[@R21]\]^ The benefit of prolonged drain duration is to reduce seroma, thereby reducing the risk of infection. This benefit may be subdued by the increased risk of ascending infection due to prolonged drain duration. This may explain the conflicting findings that prolonged use of drain reduces seroma formation but does not decrease the infection rate. A Cochrane review concluded that drains reduced seroma (OR = 0.46, 95% CI = 0.23--0.91) with no effect on infection.^\[[@R22]\]^ Distended skin may cause marginal necrosis and wound dehiscence. Concerning possible bacterial transfer from the skin wound into the pocket, lateral advancement of the lateral part of the pectoralis major muscle is pursued as often as possible, using the muscle layer to protect the tissue expander from the overlying skin wound. Our goal is not to cover the whole tissue expander with muscle, but only to add a muscle barrier for the tissue expander to keep out infection from the skin wound. Concerning high tension on the overlying skin, we fill the tissue expander with 20 to 100 mL intraoperatively, which is less than 20% of the full volume of the tissue expander. Greater initial tissue expander fill increased infection rate.^\[[@R23]\]^ Crosby et al reported that for every 10% increase in saline fill volume, complication risk increased 1.15 times in univariable analysis (*P* = 0.018). In their study,^\[[@R24]\]^ the mean percentage of intraoperative tissue expander saline fill volume was 68%. We routinely instruct our patients to limit their shoulder exercise until the drain is removed. Shamley et al^\[[@R25]\]^ reported a meta-analysis that showed delaying exercises significantly decreased seroma formation (OR = 0.4; 95% CI 0.2--0.5; *P* = 0.00001). After removing the drain on or around postoperative day 7, fluid accumulation may occur. We seldom aspirate the seroma. Instead, we injected the tissue expander on postoperative day 14. Hanna et al^\[[@R9]\]^ recommended early expansion of tissue expander on postoperative day 21 to reduce infectious complications. They also found an association between infection and drain duration greater than 21 days. After drain removal, we are concerned that fluid accumulation may cause leakage from the surgical wound. In our cohort, there was no fluid leakage from around the pocket to the skin. In this study, radiotherapy did not significantly increase the infection rate. Overall, there were 42 patients who received radiation. None of them had infection. This is counterintuitive. Radiation renders the tissue more vulnerable to the bacteria.^\[[@R21],[@R26]\]^ By contrast, in our study, patients who underwent chemotherapy had a higher infection rate (7.4%) than those who did not receive chemotherapy (2.6%) (AOR = 2.59, *P* = 0.04). In this study, age did not affect the risk of infection. It is a common belief that the elderly are more susceptible to infection.^\[[@R23],[@R27]\]^ However, the mean age of our cohort was 42.3 ± 7.8 years old, ranging from 19 to 69, and 99% of them were younger than 60. This relatively low average age might explain why age did not influence the risk of infection in this study. Observational study is the limitation of this study. This makes the interpretation of causality difficult. Some may argue that prolonged drain duration might have been the result rather than the cause of infection. To my knowledge, it is uncertain if the drainage volume would increase or decrease in the event of an infection. The above speculation could only be established if the drainage volume increases during infection. Assuming the daily drainage volume increased during infection, it would delay the timing of drain removal. Hence, it is likely that the drain would not be removed until the tissue expander is removed. However, in this study, it only occurred in 1 of the 29 infected patients. In the remaining 28 patients, the drain was removed before the tissue expander was. This evidence decreases the probability that infection is the cause of prolonged drain duration. ADM, as an additional foreign body, may affect the risk of seroma formation, skin necrosis, and infection.^\[[@R18],[@R28],[@R29]\]^ The object of this study is to analyze the association of infection with drain duration versus that with last drainage volume to determine the timing of drain removal. The absence of ADM in this study helped avoid a very significant confounder which might have contaminated the analysis The exact last daily drainage volume is rarely reported. Most of the articles deemed last daily drainage volume dichotomous for statistical analysis. We used various cutoff values of last daily drainage volume and still did not find any significant association with the infection rate. The finding is robust even when the last daily drainage volume was analyzed as a continuous variable, using milliliter as a unit. In conclusion, in this study, we showed evidence that the infection rate increased with longer drain duration and greater breast weight. The infection rate is not significantly related to the last daily drainage volume. We recommend that the drain is better removed no longer than 3 weeks postoperatively and can be removed as early as postoperative day 7, even when the drainage is over 30 mL in a 24-hour period. Abbreviations: ADM = acellular dermal matrix, AOR = adjusted odds ratio, BMI = body mass index, CI = confidence interval, DM = diabetes mellitus, LN = lymph node, MRM = modified radical mastectomy, SD = standard deviation, SLNB = sentinel lymph node biopsy, SM = simple mastectomy. The authors have no funding and conflicts of interest to disclose.

All relevant data is contained in the paper and supporting information files. Code used in this study is included in the Supporting Information files. 1 Introduction {#sec001} ============== Oscillatory neural activity is ubiquitous and covers a wide spatial and temporal scale from single neural cells to whole brain regions and from milliseconds to days. Neural oscillations are believed to be relevant for a wide range of brain activities from sensory information processing to consciousness \[[@pone.0174304.ref001]\]. It is believed that the phase of low frequency theta oscillations (4-8 Hz) drives the pyramidal cells and is used for information processing in the hippocampus \[[@pone.0174304.ref002]--[@pone.0174304.ref004]\]. Visual stimuli binding is believed to be related to the phase resetting of the fast frequency gamma band (30-70 Hz) \[[@pone.0174304.ref005]\]. Positive phase correlations between the theta rhythm and the amplitude of gamma oscillations were found during visual stimuli processing and learning \[[@pone.0174304.ref001], [@pone.0174304.ref006], [@pone.0174304.ref007]\] and during fear-related information processing \[[@pone.0174304.ref008], [@pone.0174304.ref009]\]. Theta rhythm resetting also drives cognitive processes \[[@pone.0174304.ref010]\]. Theoretical studies suggested that phase resetting could explain cross-frequency phase-locking of gamma rhythm within a theta cycle \[[@pone.0174304.ref011]\], which is the hallmark of successful memory retrieval \[[@pone.0174304.ref012], [@pone.0174304.ref013]\]. The phase of neural oscillations is also used to bridge a much wider frequency range from slow theta rhythms of large neural networks, such as those in the hippocampus, up to the individual fast spiking neurons used for speech decoding \[[@pone.0174304.ref014]\]. It was found that speech resets background (rest) oscillatory activity in specific frequency domains corresponding to the sampling rates optimal for phonemic and syllabic sampling \[[@pone.0174304.ref014], [@pone.0174304.ref015]\]. Phase resetting is also critical in the functioning of suprachiasmatic nucleus that produces a stable circadian oscillation by light-induced resetting of endogeneous rhythm \[[@pone.0174304.ref016], [@pone.0174304.ref017]\]. It was also shown that single sensory stimulus \[[@pone.0174304.ref018], [@pone.0174304.ref019]\] and periodic train of inputs \[[@pone.0174304.ref020], [@pone.0174304.ref021]\] induce phase resetting in electroencephalograms, which manifest as event-related evoked potentials. Most of neurobiologically inspired interval timing theories assume that neural oscillators and their relative phases could be used as internal clocks for biological rhythms \[[@pone.0174304.ref022]\]. It was experimentally and computationally found that noisy neural oscillators could produce accurate timing that also obeys scalar property, i.e. the temporal estimation error increases proportionally with the duration \[[@pone.0174304.ref023]--[@pone.0174304.ref025]\]. The attention mechanism phase resets neural oscillators and can produce either a stop or a delay in conditioned stimuli timing with intrudes such as gaps \[[@pone.0174304.ref026]\] or fear stimuli \[[@pone.0174304.ref027]\]. Recent optogentic experiments shown that the steady gamma rhythm of medial prefrontal cortex can be reset and entrained by light stimuli and modulated by amphetamines \[[@pone.0174304.ref028]\]. Delay embedding reconstruction of the phase space gave a low dimensional attractor suggesting a phase coupled model of medial prefrontal cortex that is reset by light stimuli \[[@pone.0174304.ref029]\]. Unidirectional coupling between neural oscillators, i.e. a master-slave system, suggests the simplest possible synchronization mechanism that uses phase resetting to drive a neural population to a desired phase-locked firing pattern. Phase resetting methodology has been successfully used for predicting one-to-one entrainment in networks where the receiving population always follows the driving population \[[@pone.0174304.ref030]--[@pone.0174304.ref032]\]. It was recently shown that unidirectional coupling also allows for "anticipated synchronization" \[[@pone.0174304.ref033]\] in which the receiving population anticipates the states of the driving population \[[@pone.0174304.ref034]\]. It has been analytically proven and numerically verified that time-delayed feedback can force coupled dynamical systems onto a synchronization manifold that involves the future state of the drive system, i.e. "anticipating synchronization" \[[@pone.0174304.ref033]\]. Such a result is counterintuitive since the future evolution of the drive system is anticipated by the response system despite the unidirectional coupling. It has been suggested that delayed coupling in dynamical systems separated by some distance can still promote synchronization despite the slow signal transmission and the unidirectional coupling. The first anticipating synchronization study of excitable systems was done by Ciszak et al \[[@pone.0174304.ref035]\], followed by more recent behavioral-related investigations \[[@pone.0174304.ref036], [@pone.0174304.ref037]\]. Synaptic delay and synaptic plasticity was recently extensively investigated as potential control parameters that can lead to tunable delayed and anticipating synchronization in neural networks \[[@pone.0174304.ref038]--[@pone.0174304.ref040]\]. We investigated analytically and numerically a three-neuron master-slave system with a dynamic inhibitory loop that was previously shown experimentally to exhibit anticipating synchronization \[[@pone.0174304.ref041], [@pone.0174304.ref042]\]. The three-neuron network investigated here was shown to produce both delayed synchronization, in which pre-synaptic neuron fires a spike before post-synaptic neuron, and anticipating synchronization. It was argued that the delayed synchronization is a possible mechanism for spike-timing dependent plasticity \[[@pone.0174304.ref040]\], whereas anticipating synchronization could contribute to long term depression of synaptic couplings \[[@pone.0174304.ref040]--[@pone.0174304.ref042]\]. This study focuses on deriving analytic criteria for the existence and the stability of phase-locked modes in a three-neuron network that was found to generate both delayed and anticipated synchronization. For this purpose, we used the method of phase response curve (PRC) \[[@pone.0174304.ref032], [@pone.0174304.ref043]--[@pone.0174304.ref050]\]. The novelties of this study are (1) the generalization of PRC to multiple inputs per cycle and (2) the prediction of phase-locked modes in a neural network that is no longer limited to one-to-one firing patterns. 2 Phase response curve method {#sec002} ============================= The phase response curve (PRC) method has been extensively used for predicting phase-locked modes in neural networks \[[@pone.0174304.ref051]--[@pone.0174304.ref054]\]. It assumes that the only effect of a stimulus is to reset the phase of the ongoing oscillation of a neuron. Traditionally, the PRC tabulates the transient change in the firing frequency of a neural oscillator in response to one external stimulus per cycle of oscillation. The term PRC has been used almost exclusively in regard to a **single stimulus** per cycle of neural oscillators. Recently, we suggested a generalization of the PRC that allowed us to account for the overall resetting when two or more inputs are delivered during the same cycle \[[@pone.0174304.ref055]\]. As a result, we expanded the PRC theory from the prediction of the traditional one-to-one phase-locked modes to arbitrary phase-locked firing patterns. Here we present the first quantitative application of such generalized PRC approach to a realistic neural network with a dynamic feedback loop. In the case of a single stimulus, the PRC measures the change of the free running period *P*~*i*~ of a neural oscillator to a new value *P*~1~ (see [Fig 1A](#pone.0174304.g001){ref-type="fig"}). The stimulus time *t*~*s*~ is measured from an arbitrary phase reference *φ* = 0. In our numerical simulations, the phase reference was the zero crossing of the membrane potential with a positive slope. The relative change in the duration of the current cycle, i.e. the cycle that contains the perturbation, with respect to the unperturbed duration *P*~*i*~ determines the first order PRC in response to a single stimulus (for detailed mathematical definitions see [Appendix 1](#sec011){ref-type="sec"}). As a result of the perturbation, the new firing period becomes *P*~1~ = *P*~*i*~(1 + *F*^(1)^(*φ*)), where *F*^(1)^ represents the relative shortening/lengthening of the intrinsic firing period *P*~*i*~ due to the stimulus applied at phase *φ* = *t*~*s*~/*P*~*i*~ \[[@pone.0174304.ref047], [@pone.0174304.ref048], [@pone.0174304.ref050]\]. {#pone.0174304.g001} A saddle-node bifurcation, which presents a continuous frequency versus bias current (f-I) curve that extends to arbitrarily low frequencies (see solid circles in [Fig 1B](#pone.0174304.g001){ref-type="fig"}) usually leads to a type 1 PRC that looks unimodal as in [Fig 1C](#pone.0174304.g001){ref-type="fig"} (although for counterexamples see \[[@pone.0174304.ref049], [@pone.0174304.ref056]\]). [Fig 1C](#pone.0174304.g001){ref-type="fig"} shows a typical type 1 PRC in response to a brief excitatory current perturbation that produces only phase advances (period shortening), i.e. negative resettings. A type 1 PRC looks unimodal and is often associated with a class I excitable cell, i.e. a cell that can produce stable oscillatory activity with arbitrarily low frequency \[[@pone.0174304.ref057], [@pone.0174304.ref058]\]. Usually, such excitable cells produce stable oscillations via a saddle node bifurcation on an invariant circle \[[@pone.0174304.ref059]\]. A type 2 PRC looks bimodal (see [Fig 1D](#pone.0174304.g001){ref-type="fig"}) and is often associated with a class II excitable cell \[[@pone.0174304.ref057], [@pone.0174304.ref058]\]. Class II oscillations usually emerge through a Hopf bifurcation \[[@pone.0174304.ref059]\] (see [Fig 1B](#pone.0174304.g001){ref-type="fig"}). As a side note, it was recently shown that type 1 (unimodal) PRCs do not always come from a class I excitable cell \[[@pone.0174304.ref056]\] and in fact all PRCs are bimodal with varying degrees \[[@pone.0174304.ref049], [@pone.0174304.ref050]\]. Close to the bifurcation point, accurate analytical formulas called normal forms describe the PRCs (see \[[@pone.0174304.ref057]\] and [Appendix 1](#sec011){ref-type="sec"} for mathematical details), which we used in this study to get some analytical insights into the general behavior of the three-neuron network with a dynamic loop shown in [Fig 1E](#pone.0174304.g001){ref-type="fig"}. The key assumption in generalizing the PRC method to multiple inputs per cycle was that the resetting induced by one stimulus takes effect "almost" instantaneously, i.e. before the arrival of the next stimulus \[[@pone.0174304.ref055]\]. Therefore, the effects of two stimuli applied during the same cycle are independent of each other. As a result, we used the single stimulus PRCs (*F*^(1)^) shown in [Fig 1C and 1D](#pone.0174304.g001){ref-type="fig"} to compute the phase resetting in response to two or more stimuli (see [Appendix 1](#sec011){ref-type="sec"} for the detailed mathematical derivation of *F*^(2)^ and its generalization). Briefly, the first stimulus delivered at stimulus phase *φ*~*a*~ = *t*~*sa*~/*P*~*i*~ produces a transient change in the firing period to *P*~*a*~ = *P*~*i*~(1 + *F*^(1)^(*φ*~*a*~)). The second stimulus that arrives at a stimulus phase *φ*~*b*~ = *t*~*sb*~/*P*~*a*~ \> *φ*~*a*~ further changes the firing period to *P*~*b*~ = *P*~*a*~(1 + *F*^(1)^(*φ*~*b*~)) (see [Appendix 1](#sec011){ref-type="sec"}). Combining the above effects of the two stimuli applied at phases *φ*~*a*~ and *φ*~*b*~, the new firing period *P*~*b*~ becomes *P*~*b*~ = *P*~*i*~(1 + *F*^(2)^(*φ*~*a*~, *φ*~*b*~)) (see [Appendix 1](#sec011){ref-type="sec"} for a detailed mathematical derivation). A typical two stimuli protocol (see [Fig 2A](#pone.0174304.g002){ref-type="fig"}) and the corresponding phase resetting *F*^(2)^ are shown in [Fig 2B](#pone.0174304.g002){ref-type="fig"}, where the three-dimensional surface is given by [Eq (22)](#pone.0174304.e100){ref-type="disp-formula"} in Appendix 1 and a two-dimensional contour plot also shows the contours of equal phase resetting. For this plot we used the analytical normal form of the PRC (see [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} in [Appendix 1](#sec011){ref-type="sec"}) where *P*~2*i*~ = 70 ms and the coupling strengths from neuron 1 to neuron 2 was *g*~12~ = 0.015 (excitatory) and from neuron 3 to neuron 2 was *g*~32~ = 0.002 (inhibitory) (see section 3 for a detailed description of the neural model and the synaptic couplings). {#pone.0174304.g002} 3 The neural model {#sec003} ================== In their seminal work on giant squid axon, Hodgkin and Huxley \[[@pone.0174304.ref060]--[@pone.0174304.ref064]\] experimentally identified three classes, or types, of axonal excitability: class I, where the repetitive firing is controlled by the intensity of an external stimulus; class II, where the firing frequency is almost independent on stimulus intensity; and class III, where there are no endogenous bursters regardless of stimulus intensity or duration. Our simulations were performed using a class I, single compartment, neural oscillator described by a standard conductance-based, or Hodgkin-Huxley (HH), mathematical model \[[@pone.0174304.ref064]--[@pone.0174304.ref066]\]. The rate of change of membrane potential is: $$\begin{array}{rcl} {dV/dt} & = & {- I_{Ca} - I_{K} - I_{Leak} + I_{0}} \\ & = & {- {\overline{g}}_{Ca}m\left( V \right)\left( V - E_{Ca} \right) - {\overline{g}}_{K}w\left( V - E_{K} \right) - {\overline{g}}_{Leak}\left( V - E_{Leak} \right) + I_{0},} \\ \end{array}$$ where *V* is the membrane potential, ${\overline{g}}_{ch}$ and *E*~*ch*~ are the maximum conductance and, respectively, the reversal potential for ionic channel *ch* (only calcium, potassium and leak were considered), *w* is the instantaneous probability that a potassium channel is open, and *I*~0~ is a constant bias current. Each ionic current is the product of a voltage-dependent conductance and a driving force *I*~*ch*~ = *g*~*ch*~(*V*)(*V* − *V*~*ch*~) where *g*~*ch*~(*V*) is the product of the maximum conductance for that channel and a specific voltage-dependent gating variable. Morris and Lecar (ML) mathematical model has two non-inactivating voltage-sensitive gating variables: one instantaneous, voltage-dependent, calcium activation *m*(*V*) and a delayed voltage-dependent potassium *w* given by a first order differential equation \[[@pone.0174304.ref067]\]: $$\begin{array}{r} {dw/dt = \phi\left( w_{\infty}\left( V \right) - w \right)/\tau\left( V \right),} \\ \end{array}$$ where *ϕ* is a temperature-dependent parameter, and a voltage-dependent relaxation time constant is defined by *τ*(*V*) = cosh^−1^((*V* − *V*~*w*,1/2~)/(2*V*~*w*,*slope*~)). All open-state probability functions, or steady-state gating variables *x*, have a sigmoidal form \[[@pone.0174304.ref067]\]: $$\begin{array}{r} {x\left( V \right) = \left( 1 + \tanh\left( \left( V - V_{x,1/2} \right)/V_{x,slope} \right) \right)/2,} \\ \end{array}$$ where *V*~*x*,1/2~ is the half-activation voltage and *V*~*x*,*slope*~ is the slope factor for the gating variable *x*. The ML model is widely used in computational neuroscience because it captures relevant biological processes and, at the same time, by changing only a small subset of parameters it can behave either as a type 1 or a type 2 neural oscillator. The dimensionless parameters for a type 1 ML neuron are: *V*~*m*,1/2~ = −0.01, *V*~*m*,*slope*~ = 0.15, *V*~*w*,1/2~ = 0.1, *V*~*w*,*slope*~ = 0.145, *V*~*K*~ = −0.7, *V*~*Leak*~ = −0.5, *V*~*Ca*~ = 1.0, ${\overline{g}}_{Ca} = 1.33$, ${\overline{g}}_{K} = 2.0$, ${\overline{g}}_{Leak} = 0.5$, *I*~0~ = 0.070, and *ϕ* = 0.6 (Ermentrout, 1996). The model's equations and its parameters are in dimensionless form with all voltages divided by the calcium reversal potential *V*~*Ca*0~ = 120 mV, all conductances divided by ${\overline{g}}_{Ca0} = 4$ mS/cm^2^, and all currents normalized by ${V_{Ca0}{\overline{g}}_{Ca0} = 480}\mspace{600mu}\mu\text{A}/\text{cm}^{2}$ (Ermentrout, 1996). For example, a dimensionless reversal potential for a leak current of *V*~*Leak*~ = −0.5 means *V*~*Leak*~ = −0.5*V*~*Ca*0~ = −0.5 × 120 mV = -60 mV. **The Synaptic Model.** We implemented fast chemical synapses between neurons given by a synaptic current $I_{syn} = {\overline{g}}_{syn}s\left( t \right)\left( V_{post} - E_{syn} \right)$, where ${\overline{g}}_{syn}$ is the maximum synaptic conductance, *s*(*t*) is the fraction of channels activated by neurotransmitters, *V*~*post*~ is the membrane potential of the postsynaptic neuron, and *E*~*syn*~ is the reversal potential of the synaptic coupling. We used *E*~*syn*~ = 0 for excitatory and *E*~*syn*~ = −0.6 for inhibitory coupling. The synapses activation was described by a first order kinetics *s*′ = *αT*(1 − *s*) − *βs*, where *α* = 15, *β* = 1.5, and neurotransmitter binding was described by a sigmoidal function *T*(*V*~*pre*~) = 1/(1 + *e*^(−*V*~*pre*~\ −\ 0.2)120/5)^) where *V*~*pre*~ is the membrane potential of the presynaptic neuron. We numerically computed the PRCs in open loop setup, i.e. by injecting a single synaptic input from a corresponding presynaptic neuron for neurons 2 and 3 shown in [Fig 3](#pone.0174304.g003){ref-type="fig"} for our neural network configuration. {#pone.0174304.g003} 4 The neural network model {#sec004} ========================== In order to use the PRC method (see section 2) for predicting the relative phases of neurons in a phase-locked firing pattern, we assumed a fixed firing order of the three neurons with the goal of determining if such a pattern exists and if it is stable. Based on the neural network model proposed for delayed and anticipated synchronization by Matias et al \[[@pone.0174304.ref040]--[@pone.0174304.ref042]\], we identified the following definitions for the firing period of each neuron (see [Fig 3](#pone.0174304.g003){ref-type="fig"}): $$\begin{array}{rcl} P_{1} & = & {t_{2r}\left\lbrack n - 1 \right\rbrack + t_{2sb}\left\lbrack n \right\rbrack,} \\ P_{2} & = & {t_{2sb}\left\lbrack n \right\rbrack + t_{2r}\left\lbrack n \right\rbrack,} \\ P_{3} & = & {t_{2sb}\left\lbrack n \right\rbrack - t_{2sa}\left\lbrack n \right\rbrack + t_{2r}\left\lbrack n \right\rbrack + t_{2sa}\left\lbrack n + 1 \right\rbrack,} \\ \end{array}$$ where *t*~2*r*~ is the recovery time of neuron \#2 after its last input, *t*~2*sa*~ and *t*~2*sb*~ are the corresponding stimulus times for the first and, respectively, the second input to neuron \#2, and the index of the cycle is marked with the square brackets \[...\]. The subscript index refers to the neural oscillator index according to [Fig 3](#pone.0174304.g003){ref-type="fig"}. From [Eq (4)](#pone.0174304.e012){ref-type="disp-formula"} we eliminated *t*~2*r*~\[*n* − 1\] = *P*~1*i*~ − *t*~2*sb*~\[*n*\] and substituted it into the other two equations, which led to: $$\begin{array}{rcl} P_{2} & = & {t_{2sb}\left\lbrack n \right\rbrack + P_{1i} - t_{2sb}\left\lbrack n + 1 \right\rbrack,} \\ P_{3} & = & {t_{2sb}\left\lbrack n \right\rbrack - t_{2sa}\left\lbrack n \right\rbrack + P_{1i} - t_{2sb}\left\lbrack n + 1 \right\rbrack + t_{2sa}\left\lbrack n + 1 \right\rbrack.} \\ \end{array}$$ Based on the definitions of the PRCs (see Eqs ([16](#pone.0174304.e093){ref-type="disp-formula"}) and ([22](#pone.0174304.e100){ref-type="disp-formula"}) in [Appendix 1](#sec011){ref-type="sec"}), we further expanded [Eq (5)](#pone.0174304.e013){ref-type="disp-formula"} the transiently modified firing period in terms of experimentally determined PRCs: $$\begin{array}{rcl} {P_{2i}\left( 1 + F^{(2)}\left( t_{2sa}\left\lbrack n \right\rbrack,t_{2sb}\left\lbrack n \right\rbrack \right) \right)} & = & {t_{2sb}\left\lbrack n \right\rbrack + P_{1i} - t_{2sb}\left\lbrack n + 1 \right\rbrack,} \\ {P_{3i}\left( 1 + F^{(1)}\left( t_{s3}\left\lbrack n \right\rbrack \right) \right)} & = & {t_{2sb}\left\lbrack n \right\rbrack - t_{2sa}\left\lbrack n \right\rbrack + P_{1i} - t_{2sb}\left\lbrack n + 1 \right\rbrack + t_{2sa}\left\lbrack n + 1 \right\rbrack.} \\ \end{array}$$ The above system of two recursive equations has two unknowns, i.e. *t*~2*sa*~ and *t*~2*sb*~, that describe the temporal evolution of the relative phase of neural oscillators from the firing cycle \[*n*\] to \[*n* + 1\]. 5 The existence of phase-locked modes {#sec005} ===================================== Let us assume that there is a steady state solution $\left( t_{2sa}^{*},t_{2sb}^{*} \right)$ for the recursive [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} that mimics the activity of the neural network shown in [Fig 3](#pone.0174304.g003){ref-type="fig"}, i.e. the following limits exist $\lim\limits_{n\infty}t_{2sa}\left\lbrack n \right\rbrack = t_{2sa}^{*}$ and $\lim\limits_{n\infty}t_{2sa}\left\lbrack n \right\rbrack = t_{2sa}^{*}$. By substituting the steady state, i.e. phase-locked mode, solution $\left( t_{2sa}^{*},t_{2sb}^{*} \right)$ into [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} one obtains: $$\begin{array}{rcl} {P_{2i}\left( 1 + F_{2}^{(2)}\left( t_{2sa}^{*},t_{2sb}^{*} \right) \right)} & = & {P_{1i},} \\ {P_{3i}\left( 1 + F_{3}^{(1)}\left( P_{1i} - t_{2sa}^{*} \right) \right)} & = & {P_{1i},} \\ \end{array}$$ where we used the fact that $t_{3s}^{*} = t_{2sb}^{*} - t_{2sa}^{*} + t_{2r}^{*}$ and that *t*~2*r*~\[*n* − 1\] = *P*~1*i*~ − *t*~2*sb*~\[*n*\], which led to $t_{3s}^{*} = t_{2sb}^{*} - t_{2sa}^{*} + P_{1i} - t_{2sb}^{*} = P_{1i} - t_{2sa}^{*}.$ As we notice from the second equation in [Eq (7)](#pone.0174304.e019){ref-type="disp-formula"}, the steady state value $t_{2sa}^{*}$ could be immediately determined and it only depends on *P*~1*i*~, *P*~3*i*~, and the PRC of the third neuron, which depends on the coupling strength *g*~23~. It results that the steady state value $t_{2sa}^{*}$ is given by: $$\begin{array}{r} {F_{3}^{(1)}\left( P_{1i} - t_{2sa}^{*} \right) = \frac{P_{1i}}{P_{3i}} - 1.} \\ \end{array}$$ Since the coupling from neuron \#2 to neuron \#3 is excitatory, the PRC is negative (only advances the next spike), i.e. $F_{3}^{(1)}\left( \varphi \right) < 0$. As a result of [Eq (8)](#pone.0174304.e024){ref-type="disp-formula"}, the steady states $t_{2sa}^{*}$ can only exist for *P*~1*i*~ \< *P*~3*i*~ which means that the interneuron (neuron \#3) must be slower than the pacemaker (neuron \#1) of the network. Moreover, since a type 1 PRC in response to excitatory inputs has only one negative minimum ($F_{3,min}^{(1)}$) that determines the magnitude of the strongest possible resetting, then $\frac{P_{1i}}{P_{3i}} - 1 \geq F_{3,min}^{(1)}$, which means the the interneuron intrinsic period is bounded by $P_{1i} < P_{3i} < P_{1i}/\left( 1 + F_{3,min}^{(1)} \right).$ Once we determined the steady state $t_{2sa}^{*}$ from [Eq (8)](#pone.0174304.e024){ref-type="disp-formula"}, then we plugged it into the first [Eq (7)](#pone.0174304.e019){ref-type="disp-formula"} and found $t_{2sb}^{*}.$ Using the PRC definition (see [Eq (22)](#pone.0174304.e100){ref-type="disp-formula"} from [Appendix 1](#sec011){ref-type="sec"}) we obtained: $$\begin{array}{r} {P_{2i}\alpha\left( 1 + F_{2}^{(1)}\left( \frac{t_{2sb}^{*}}{\alpha} \right) \right) = P_{1i},} \\ \end{array}$$ where $\alpha = 1 + F_{2}^{(1)}\left( t_{2sa}^{*} \right).$ The above equation can be reduced to: $$\begin{array}{r} {F_{2}^{(1)}\left( \frac{t_{2sb}^{*}}{\alpha} \right) = \frac{P_{1i}}{P_{2i}\alpha} - 1,} \\ \end{array}$$ We must emphasize that $F_{2}^{(1)}\left( t_{2sa}^{*} \right)$ and $F_{2}^{(1)}\left( t_{2sb}^{*} \right)$ are two different single stimulus PRCs for the same neuron. Here $F_{2}^{(1)}\left( t_{2sa}^{*} \right)$ is the single stimulus phase response curve of the second neuron to an input received from the third neuron, i.e. $F_{2}^{(1)}\left( t_{2sa}^{*} \right)$ is determined by *g*~32~. Similarly, $F_{2}^{(1)}\left( t_{2sb}^{*} \right)$ is the single stimulus phase response curve of the second neuron in response to an input received from the first neuron, i.e. $F_{2}^{(1)}\left( t_{2sb}^{*} \right)$ is determined by *g*~12~. 5.1 Explicit steady state solutions using normal form generic type 1 PRCs {#sec006} ------------------------------------------------------------------------- In order to get insights into the general existence criteria for steady state (phase-locked modes) derived above, we assumed that the single stimulus and the generalized PRCs are quite well approximated by the corresponding normal forms given by [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"} and, respectively, by [Eq (22)](#pone.0174304.e100){ref-type="disp-formula"} in Appendix 1. Then the steady state solution $t_{2sa}^{*}$ of [Eq (8)](#pone.0174304.e024){ref-type="disp-formula"} can be analytically written as: $$\begin{array}{r} {\cos\left( 2\pi\left( \frac{P_{1i}}{P_{3i}} - \frac{t_{2sa}^{*}}{P_{3i}} \right) \right) = 1 - \frac{1}{c_{23}P_{3i}}\left( \frac{P_{1i}}{P_{3i}} - 1 \right).} \\ \end{array}$$ By least square fitting the numerically generated PRCs for each neuron in response to a single spike from its corresponding presynaptic neuron with the theoretical formula of the normal form given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"}, we found a quantitative relationship between the abstract coupling strength coefficient *c* and the physiologically measurable maximum synaptic couplings $\overline{g}$ (see [Appendix 1](#sec011){ref-type="sec"}). Therefore, in order to simplify the mathematical notation, throughout the rest of the paper we only write, for example, *c*~23~ when referring to the coefficient of the theoretical normal form of the PRC with the understanding that it is a known function of the synaptic conductance, i.e. *c*~23~ = *c*~23~(*g*~23~). Since −1 ≤ *cos*(*x*)≤1, it results that $0 \leq \frac{1}{c_{23}\left( g_{23} \right)P_{3i}}\left( \frac{P_{1i}}{P_{3i}} - 1 \right) \leq 2$, which determines the minimum coupling strength *g*~23~ for a given ratio of the two intrinsic firing periods $\frac{P_{1i}}{P_{3i}}$ to attain a phase-locked mode pattern. Based on the above relationship, for excitatory coupling, i.e. *E*~*syn*~ = 0, the master (pacemaker) neuron \#1 (see [Fig 3](#pone.0174304.g003){ref-type="fig"}) must be faster than the interneuron \#3, i.e. *P*~1*i*~ \< *P*~3*i*~. At the same time, the coupling strength *g*~23~ must also be strong enough to reset the longer intrinsic period *P*~3*i*~ to match the shorter period of the network's pacemaker, i.e. to ensure that $P_{3i} < P_{1i}/\left( 1 + F_{3,min}^{(1)} \right)$. The above relationship allowed us to estimate that, if the coupling is very strong (*g*~23~∞), then the steady state from [Eq (11)](#pone.0174304.e042){ref-type="disp-formula"} has the solution $t_{2sa}^{*} = P_{1i} + kP_{3i}$ with *k* = 0,±1,±2,..., which is marked by vertically downward arrows in [Fig 4](#pone.0174304.g004){ref-type="fig"}. Furthermore, if the two firing periods are approximately equal (*P*~1*i*~ ≈ *P*~3*i*~), then from [Eq (11)](#pone.0174304.e042){ref-type="disp-formula"} it results that $t_{2sa}^{*} = kP_{1i}$ with *k* = 0, 1, 2, ... (see also [Fig 4](#pone.0174304.g004){ref-type="fig"}). [Fig 4](#pone.0174304.g004){ref-type="fig"} also shows that for each intrinsic period ratio *P*~3*i*~/*P*~1*i*~ there is a minimum coupling strength *g*~23~ that ensures appropriate resetting of the interneuron. For example, the minimum coupling for *P*~3*i*~/*P*~1*i*~ = 1.5 (dotted red line in [Fig 4](#pone.0174304.g004){ref-type="fig"}) is *g*~23~ = 0.024. A stronger coupling of *g*~23~ = 0.036 is necessary for a larger ratio *P*~3*i*~/*P*~1*i*~ = 2 (dashed blue line in [Fig 4](#pone.0174304.g004){ref-type="fig"})). {#pone.0174304.g004} The phase-locked modes $\left( t_{2sa}^{*},t_{2sb}^{*} \right)$ given by [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} depend on three intrinsic periods *P*~1*i*~, *P*~2*i*~, *P*~3*i*~ and three synaptic conductances *g*~12~, *g*~23~, and *g*~32~. Since the master neuron receives no input, all durations were measured relative to *P*~1*i*~. The bias current for the computational model was set such that *P*~1*i*~ = 60 ms, *P*~2*i*~ = 70 ms and *P*~3*i*~ = 80 ms (see section 3 for details and supplemental files for a computational implementation). Intuitively, the phase-locked solution $t_{2sa}^{*}$ is the stable interspike interval between neurons \#2 and \#3 (the interneuron). The other phase-locked solution $t_{2sb}^{*}$ is the stable interspike interval between neurons \#2 and \#1 (network's driver). However, this simplification only reduces the parameter space to five dimensions. In order to reduce the parameter space to four dimensions, we only show examples of phase-locked modes for a fixed inhibitory coupling with *g*~32~ = 0.002 (arb. units). For a fixed intrinsic period of the second neuron *P*~2*i*~/*P*~1*i*~ = 70/60, the parameter space further reduces to three dimensions, which allowed us to visualize the phase-locked modes. The solution $t_{2sa}^{*}$ of the second equation in [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} only depends on *P*~3*i*~/*P*~1*i*~ and the coupling strength *g*~23~ (see the green surface in [Fig 5](#pone.0174304.g005){ref-type="fig"}). {#pone.0174304.g005} However, the phase locked solution $t_{2sb}^{*}$ of the first equation in [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} depend on the additional coupling *g*~12~. Therefore, to gain insight into how *g*~12~ affects the solution $t_{2sb}^{*}$, we used the same axis *P*~3*i*~/*P*~1*i*~ and *g*~23~ as for $t_{2sa}^{*}$, but with different constant values of coupling *g*~12~ = 0.012 (red surface) and *g*~12~ = 0.05 (blue surface) in [Fig 5A](#pone.0174304.g005){ref-type="fig"}. We notice from [Fig 5A](#pone.0174304.g005){ref-type="fig"} that increasing the strength of the excitatory coupling *g*~12~ leads to an increased stimulus time $t_{2sb}^{*}$ and a wider parameter domain of the phase-locked solution. Similarly, if we hold constant the synaptic coupling *g*~12~ = 0.015 (arb. units) between the master and the slave neurons, then we could visualize the phase-locked solution for variable intrinsic period of the second neuron *P*~2*i*~/*P*~1*i*~ = 60/60 (red surface in [Fig 5B](#pone.0174304.g005){ref-type="fig"}) and *P*~2*i*~/*P*~1*i*~ = 70/60 (blue surface in [Fig 5B](#pone.0174304.g005){ref-type="fig"}). We notice that for smaller intrinsic periods *P*~2*i*~ the range of control parameters *P*~3*i*~/*P*~1*i*~ and *g*~23~ is broader. This is because for more similar firing frequencies it is easier to bring the driven neuron to the firing frequency of the driving neuron. 6 The stability of phase-locked modes {#sec007} ===================================== The possible phase-locked modes given by [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"} may not all be stable and, therefore, they may not be all experimentally observable. To determine the stability of the steady solutions $\left( t_{2sa}^{*},t_{2sb}^{*} \right)$, we assume small perturbations: $$\begin{array}{rcl} {t_{2sa}\left\lbrack n \right\rbrack} & = & {t_{2sa}^{*} + \delta t_{2sa}\left\lbrack n \right\rbrack,} \\ {t_{2sb}\left\lbrack n \right\rbrack} & = & {t_{2sb}^{*} + \delta t_{2sb}\left\lbrack n \right\rbrack,} \\ \end{array}$$ where the *n*^*th*^ cycle perturbation $\delta t_{2s}\left\lbrack n \right\rbrack < < t_{2s}^{*}$ is assumed very small for both stimuli. By substituting [Eq (12)](#pone.0174304.e066){ref-type="disp-formula"} into the existence criteria from [Eq (7)](#pone.0174304.e019){ref-type="disp-formula"} and using a Taylor series expansion one obtains: $$\begin{array}{rcl} {m_{2a}\left( 1 + F_{2}^{(1)}\left( \varphi_{2b}^{*} \right) - \varphi_{2b}^{*} \right)\delta t_{2sa}\left\lbrack n \right\rbrack + m_{2b}\delta t_{2sb}\left\lbrack n \right\rbrack} & = & {\delta t_{2sb}\left\lbrack n \right\rbrack - \delta t_{2sb}\left\lbrack n + 1 \right\rbrack,} \\ {m_{3}\left( \delta t_{2sb}\left\lbrack n \right\rbrack - \delta t_{2sa}\left\lbrack n \right\rbrack - \delta t_{2sb}\left\lbrack n + 1 \right\rbrack \right)} & = & {\delta t_{2sb}\left\lbrack n \right\rbrack - \delta t_{2sa}\left\lbrack n \right\rbrack} \\ & = & {- \delta t_{2sb}\left\lbrack n + 1 \right\rbrack + \delta t_{2sa}\left\lbrack n + 1 \right\rbrack,} \\ \end{array}$$ where $m_{2a} = \left( \frac{\partial F_{2}^{(1)}}{\partial\varphi} \right)_{\varphi_{2a}^{*}}$ is the slope of the second neuron's PRC at the phase of the first stimulus $\varphi_{2a}^{*} = \frac{t_{s2a}^{*}}{P_{2i}}$, $m_{2b} = \left( \frac{\partial F_{2}^{(1)}}{\partial\varphi} \right)_{\varphi_{2b}^{*}}$ is the slope of the second neuron's PRC at the phase of the second stimulus $\varphi_{2b}^{*} = \frac{t_{s2b}^{*}}{P_{2i}\left( 1 + F_{2}^{(1)}\left( t_{2sa}^{*} \right) \right)}$, $m_{3} = \left( \frac{\partial F_{3}^{(1)}}{\partial\varphi} \right)_{\varphi_{3}^{*}}$ is the slope of the third neuron's PRC at the phase of the stimulus $\varphi_{2}^{*} = \frac{t_{s3}^{*}}{P_{3i}}$. The stability [Eq (13)](#pone.0174304.e068){ref-type="disp-formula"} can be rewritten in a matrix form as: $$\begin{array}{r} {\begin{pmatrix} 0 & {- 1} \\ 1 & {m_{3} - 1} \\ \end{pmatrix}\begin{pmatrix} {\delta t_{s2a}} \\ {\delta t_{s2b}} \\ \end{pmatrix}_{\lbrack n + 1\rbrack} = \begin{pmatrix} {m_{2a}\left( 1 + F_{2}^{(1)}\left( \varphi_{2b}^{*} \right) - \varphi_{2b}^{*} \right)} & {m_{2b} - 1} \\ {1 - m_{3}} & {m_{3} - 1} \\ \end{pmatrix}\begin{pmatrix} {\delta t_{s2a}} \\ {\delta t_{s2b}} \\ \end{pmatrix}_{\lbrack n\rbrack},} \\ \end{array}$$ which led us to a first order recursive relationship for the perturbations: $$\begin{array}{r} {\begin{pmatrix} {\delta t_{s2a}} \\ {\delta t_{s2b}} \\ \end{pmatrix}_{\lbrack n + 1\rbrack} = \begin{pmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \\ \end{pmatrix}\begin{pmatrix} {\delta t_{s2a}} \\ {\delta t_{s2b}} \\ \end{pmatrix}_{\lbrack n\rbrack},} \\ \end{array}$$ where *a*~11~ = (1 − *m*~3~)(1 − *b*), *a*~12~ = (*m*~3~ − 1)*m*~2*b*~, *a*~21~ = −*b*, and *a*~22~ = 1 − *m*~2*b*~ with $b = m_{2a}\left( 1 + F_{2}^{(1)}\left( \varphi_{2b}^{*} \right) - m_{2b}\varphi_{2b}^{*} \right).$ The stability of the steady state is determined by the eigenvalues of [Eq (15)](#pone.0174304.e076){ref-type="disp-formula"} (see the [Appendix 2](#sec014){ref-type="sec"} for the general stability conditions in a two-dimensional recursive map). We also must keep track of the third stability condition as the original recursive system in [Eq (4)](#pone.0174304.e012){ref-type="disp-formula"} contained three variables, which were reduced to two coupled recursive equations (see [Eq (5)](#pone.0174304.e013){ref-type="disp-formula"}) by eliminating the third variable, i.e. *t*~2*r*~\[*n* − 1\] = *P*~1*i*~ − *t*~2*sb*~\[*n*\]. As a result, the steady state of the previous substitution gives $t_{2r}^{*} = P_{1i} - t_{2sb}^{*}$ and the corresponding infinitesimal perturbation is *δt*~2*r*~\[*n* − 1\] = −*δt*~2*sb*~\[*n*\]. Therefore, the stability of $t_{2r}^{*}$ solution is determined by the stability of *δt*~2*sb*~\[*n*\], which is already covered by [Eq (15)](#pone.0174304.e076){ref-type="disp-formula"} without involving additional control parameters. The general stability conditions for any first order recursion of two variables is discussed in details in the Appendix 2. Briefly, the trace *Tr*(*A*) = *a*~11~ + *a*~22~ and the determinant *Det*(*A*) = *a*~11~ *a*~22~ − *a*~12~ *a*~21~ of the recursion matrix in [Eq (15)](#pone.0174304.e076){ref-type="disp-formula"} determine the stability of each steady state obtained by solving [Eq (7)](#pone.0174304.e019){ref-type="disp-formula"}. 7 Numerical validation of the existence and the stability criteria {#sec008} ================================================================== The analytically derived criteria for the existence (see section 5) and stability (see section 6) of phase-locked modes in a master-slave network with a dynamic loop (see [Fig 3](#pone.0174304.g003){ref-type="fig"}) were only based on PRCs in response to a single stimulus. We checked our theoretical predictions based on open loop PRCs against the numerical simulations of the actual neural networks implemented according to the model presented in section 3, i.e. closed loop (fully connected neural network). The analytical normal form PRC formulas (see [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} in [Appendix 1](#sec011){ref-type="sec"}) were convenient analytical tools and even led us to some analytical results in the preceding sections. However, for the actual comparison between the multiple stimuli PRC-based phase-locked mode prediction (open loop) and the numerical simulations results of the fully coupled neural network (see [Fig 3](#pone.0174304.g003){ref-type="fig"}) we used numerically generated open lopp PRCs. The reason is that, although the analytical normal form of type 1 PRC given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} (see dashed red line in [Fig 6A](#pone.0174304.g006){ref-type="fig"}) is close to the numerically (experimentally) generated open loop PRC (see dotted blue curve in [Fig 6A](#pone.0174304.g006){ref-type="fig"}), we wanted a more accurate prediction based on the real-world PRC as it is generated in wet lab/numerical experiments. {ref-type="disp-formula"} to determined the conversion factor between the model-dependent coupling constant *c* in [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} and the synaptic coupling *g*~*syn*~. (**B**) A typical stable phase locked mode in which neuron \#2 (dashed line) receives two inputs during a single cycle: first from neuron \#3 (dotted line) at $t_{2sa}^{*}$ and then from neuron \#1 (continuous line) at $t_{2sb}^{*}$. The experimental values for the phase locked mode as measured from panel (**B**) were $t_{2sa}^{*} = 15.2$ ms and $t_{2sb}^{*} = 40.2$ ms whereas the PRC-based predictions were $t_{2sa}^{*} = 18.0$ ms and $t_{2sb}^{*} = 44.1$ ms. The network's firing period was *P* = 60 ms = *P*~1*i*~.](pone.0174304.g006){#pone.0174304.g006} We also used the least square minimization to fit actual PRCs (see dotted curve in [Fig 6A](#pone.0174304.g006){ref-type="fig"}) with the theoretical formula given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} in order to establish the conversion factor between the model-dependent coupling constant *c* in the theoretical formula given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} and the synaptic constant *g*~*syn*~ used in our numerical simulations. The Mathematica file that contains the implementation of the neural network shown in [Fig 3](#pone.0174304.g003){ref-type="fig"} based on the model equations provided in section 3 is available in supplemental files section. The synaptic couplings used for the example shown in [Fig 6B](#pone.0174304.g006){ref-type="fig"} were *g*~12~ = 0.015, *g*~32~ = 0.002, and *g*~23~ = 0.0275, which led to a phase locked mode with $t_{2sa}^{*} = 15.2$ ms and $t_{2sb}^{*} = 40.2$ ms. The PRC-based predictions were $t_{2sa}^{*} = 18.0$ ms (about 18% error) and $t_{2sb}^{*} = 44.1$ ms (about 10% error). We found that the eigenvalues of the stability matrix were *λ*~1~ = 0.489, and *λ*~2~ = 0.779, which indicated that the predicted mode was stable. Discussion {#sec009} ========== Since even for a small unidirectionally coupled three-neuron network the parameter space is six-dimensional, i.e. three intrinsic firing periods (*P*~1*i*~, *P*~2*i*~, and *P*~3*i*~), one unidirectional synaptic coupling between master-slave neurons (*g*~12~), and two coupling constants for the feedback loop (*g*~23~ and *g*~32~), we reduced it to manageable dimensions in order to visualize the phase-locked solution. Since the master neuron receives no feedback from the network, its intrinsic firing period *P*~1*i*~ was considered the reference duration, which reduces the parameter space to five dimensions. We further reduced the parameter space to four dimensions using a fixed value for the inhibitory coupling of the interneuron, i.e. *g*~32~ = 0.002 (arb. units). We numerically found the phase locked modes $\left( t_{2sa}^{*},t_{2sb}^{*} \right)$ by considering two separate cases: (1) fixed period of slave neuron \#2, of which we only show two examples with *P*~2*i*~/*P*~1*i*~ = 60/60 and *P*~2*i*~/*P*~1*i*~ = 70/60 in [Fig 5A](#pone.0174304.g005){ref-type="fig"} and (2) fixed master-slave synaptic coupling, of which we only showed two examples with *g*~12~ = 0.012 and *g*~12~ = 0.05 in [Fig 5B](#pone.0174304.g005){ref-type="fig"}. In all numerical simulations, the free parameters were the intrinsic period of the interneuron *P*~3*i*~ and the excitatory synaptic coupling to that neuron (*g*~23~). The reason is that it was previously shown that the interneuron through its intrinsic properties and its synaptic coupling can lead to either delayed or anticipating synchronization in this neural network \[[@pone.0174304.ref040], [@pone.0174304.ref042]\] and our goal was to closely match previous experimental findings using the newly developed generalized PRC method. Based on [Fig 5](#pone.0174304.g005){ref-type="fig"}, an increase in the strength of the master-slave synaptic coupling *g*~12~ leads to a larger phase difference between the two steady states $t_{2sa}^{*}$ and $t_{2sb}^{*}$. At the same time, the parameter space of the interneuron (*P*~3*i*~, *g*~22~) becomes wider. Another possibility for broadening the parameter space was to bring the intrinsic firing period of the slave neuron \#2 closer the the master neuron \#1, i.e. by reducing network heterogeneity. All out numerical simulations are in agreement with previously observed firing patterns in this type of neural network \[[@pone.0174304.ref040], [@pone.0174304.ref042]\]. Conclusions {#sec010} =========== We used a phase response curve method to predict the existence and the stability of phase-locked modes in a master-slave networks with a dynamic feedback loop. This study brings two novel solutions to phase-locked mode prediction in neural networks. First, we generalized the the phase response curve definition to include the more realistic case when neural oscillators receive more than one input per cycle. Secondly, we applied the generalized phase resetting definition to a biologically relevant neural network that has been shown to produce both delayed and anticipated synchronization. Predicting phase-locked modes in large neural networks usually requires as a first step a complexity reduction to manageable subnetworks of two neurons \[[@pone.0174304.ref068], [@pone.0174304.ref069]\] or, whenever possible, reduces the entire network to a two-population network \[[@pone.0174304.ref070]\]. Our PRC generalization to multiple inputs per cycle is a significant advance in phase resetting theory that allows investigation of large networks in which individual neurons receive multiple inputs per cycle without assuming special network connectivity. Furthermore, our generalization of phase response curve and its proof of concept application to predicting phase-locked modes existence and stability in a biologically relevant three-neuron network with a dynamic feedback loop is not limited to weak coupling nor to only one-to-one firing patterns. Indeed, the coupling strengths used were quite large such that it reset the firing period of the interneuron \#3 by 25% from 80 ms to 60 ms. Appendix 1 {#sec011} ========== Single stimulus phase response curve method {#sec012} ------------------------------------------- There are two main experimental protocols for measuring the single stimulus PRC in isolated cells: (1) single stimulus and (2) recurring (periodic) stimuli. In the case of a single stimulus protocol, a free running neural oscillator with the intrinsic period *P*~*i*~ is perturbed at a certain instant called stimulus time *t*~*s*~, which is measured from an arbitrary phase reference *φ* = 0, e.g. zero crossing of the membrane potential with a positive slope. As a result of the perturbation, the length of the current cycle that contains the stimulation (see [Fig 1A](#pone.0174304.g001){ref-type="fig"}) may be transiently shortened or lengthened to a new duration *P*~1~. The relative change in the duration of the current cycle with respect to the unperturbed duration *P*~*i*~ determines the first order PRC in response to a single and nonrecurring stimulus: $$\begin{array}{r} {F^{(1)}\left( \varphi \right) = P_{1}/P_{i} - 1,} \\ \end{array}$$ where the superscript ^(1)^ emphasizes that the resetting is due to a single input per cycle, which has been used as the "classical" definition of PRC. Based on [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"}, a negative value of the PRC means that the next spike is advanced, otherwise it is delayed. Others \[[@pone.0174304.ref058], [@pone.0174304.ref071]\] prefer to flip the sign in [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"} and associate a positive sign to a phase advance. Oftentimes, the effect of a single stimulus extends to subsequent cycles and is measured by higher order PRCs \[[@pone.0174304.ref047], [@pone.0174304.ref048], [@pone.0174304.ref050]\]. Usually, one records at least five cycles until the neural oscillatory returns back to its unperturbed oscillatory activity \[[@pone.0174304.ref031], [@pone.0174304.ref032]\]. Afterwards another single stimulus is applied at a different phase to quantify its effect on the isolated neuron (open loop experimental setup). In the case of recurring external stimuli, the interpretation of the phase resetting and its usage in phase-locked mode prediction is complicated by (1) the fact that the measured resetting compounds multiple PRC orders in a potentially nonlinear manner and (2) the activation of slow currents and/or long term potentiation (see \[[@pone.0174304.ref072]\] for examples and \[[@pone.0174304.ref032]\] for higher order PRC applications). **Normal Forms of Single Stimulus Phase Response** Curves. A saddle-node bifurcation, which presents a continuous frequency versus bias current (f-I) curve that extends to arbitrarily low frequencies (see solid circles in [Fig 1B](#pone.0174304.g001){ref-type="fig"}) usually leads to a type 1 PRC that looks unimodal as in [Fig 1C](#pone.0174304.g001){ref-type="fig"} (although for counterexamples see \[[@pone.0174304.ref049], [@pone.0174304.ref056]\]). Close to the bifurcation point, type 1 unimodal PRCs are described analytically by the following equation \[[@pone.0174304.ref057]\]: $$\begin{array}{r} {F^{(1)}\left( \varphi \right) = \frac{c_{SN}}{\omega}\left( 1 - \cos\left( 2\pi\varphi \right) \right),} \\ \end{array}$$ where *c*~*SN*~ is a constant determined by the neural model and *ω* = 2*π*/*P*~*i*~ is the intrinsic angular frequency of the oscillator. In this study, we used the simplified analytical form given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"} to get analytical insights into the general behavior of the three-neuron network with a dynamic loop shown in [Fig 3](#pone.0174304.g003){ref-type="fig"}. By least square fitting the numerically generated PRCs for each neuron in response to a single spike from its corresponding presynaptic neuron with the theoretical formula of the normal form PRC given by [Eq (17)](#pone.0174304.e094){ref-type="disp-formula"}, we found coupling strengths *c* are proportional to the maximum synaptic couplings $\overline{g}$: *c*~12~ = −6.1733*g*~12~ − 0.0003, *c*~23~ = −6.9555*g*~23~ − 0.0005, and *c*~32~ = 7.2764*g*~32~ + 0.0002. Phase resetting in response to multiple stimuli {#sec013} ----------------------------------------------- Assuming that the resetting induced by one stimulus takes effect "almost" instantaneously, i.e. before the arrival of the second stimulus, then the effects of two stimuli applied during the same cycle are independent of each other and we could use the single stimulus PRC defined by [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"} (shown in [Fig 1C and 1D](#pone.0174304.g001){ref-type="fig"}) to compute the phase resetting in response to two or more stimuli. In order to compute the phase resetting induced by the second stimulus based on [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"} we need to correctly compute its phase (see [Fig 2A](#pone.0174304.g002){ref-type="fig"}). The phase of the first stimulus that arrives at a stimulus time *t*~*sa*~ is *φ*~*a*~ = *t*~*sa*~/*P*~*i*~. The first stimulus produces an "almost" instantaneously phase resetting and changes the firing period to: $$\begin{array}{r} {P_{a} = P_{i}\left( 1 + F^{(1)}\left( \varphi_{a} \right) \right) = P_{i}\left( 1 + F^{(1)}\left( \frac{t_{a}}{P_{i}} \right) \right).} \\ \end{array}$$ When the second stimulus arrives at a stimulus time *t*~*sb*~ \> *t*~*sa*~, the neuron already has a different firing period *P*~*a*~ due to the previous stimulus. As a result, the phase of the second stimulus is *φ*~*b*~ = *t*~*sb*~/*P*~*a*~ and the new firing period due to the second stimulus is: $$\begin{array}{r} {P_{b} = P_{a}\left( 1 + F^{(1)}\left( \varphi_{b} \right) \right) = P_{a}\left( 1 + F^{(1)}\left( \frac{t_{b}}{P_{a}} \right) \right),} \\ \end{array}$$ where we used the same definition of the first order phase resetting for a single stimulus as in [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"}. By substituting [Eq (18)](#pone.0174304.e096){ref-type="disp-formula"} into [Eq (19)](#pone.0174304.e097){ref-type="disp-formula"} one obtains: $$\begin{array}{r} {P_{b} = P_{a}\left( 1 + F^{(1)}\left( \varphi_{b} \right) \right) = P_{i}\left( 1 + F^{(1)}\left( \frac{t_{a}}{P_{i}} \right) \right)\left( 1 + F^{(1)}\left( \frac{t_{b}}{P_{i}\left( 1 + F^{(1)}\left( \frac{t_{a}}{P_{i}} \right) \right)} \right) \right),} \\ \end{array}$$ which could be rewritten in a form that resembles [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"} as: $$\begin{array}{r} {P_{b} = P_{i}\left( 1 + F^{(2)}\left( \varphi_{a},\varphi_{b} \right) \right),} \\ \end{array}$$ where the superscript ^(2)^ emphasizes that the new transient period *P*~*b*~ is computed in response to two stimuli arriving at phases *φ*~*a*~ and, respectively, *φ*~*b*~ \> *φ*~*a*~ during the same cycle. By comparing the definition from [Eq (21)](#pone.0174304.e099){ref-type="disp-formula"} against the derived resetting from [Eq (20)](#pone.0174304.e098){ref-type="disp-formula"}, we found that: $$\begin{array}{r} {F^{(2)}\left( t_{sa},t_{sb} \right) = \left( 1 + F^{(1)}\left( \frac{t_{a}}{P_{i}} \right) \right)\left( 1 + F^{(1)}\left( \frac{t_{b}}{P_{i}\left( 1 + F^{(1)}\left( \frac{t_{a}}{P_{i}} \right) \right)} \right) \right) - 1,} \\ \end{array}$$ which has the advantage that can predict the phase resetting in response to two stimuli by recursively using the single stimulus PRC defined in [Eq (16)](#pone.0174304.e093){ref-type="disp-formula"}. A typical two stimuli phase response curve *F*^(2)^ is shown in [Fig 2B](#pone.0174304.g002){ref-type="fig"}. Furthermore, our novel derivation of PRC in response to two stimuli given by [Eq (22)](#pone.0174304.e100){ref-type="disp-formula"} generalizes to an arbitrary number of inputs per cycle as follows: $$\begin{array}{r} {P_{n}\left( t_{s1},t_{s2},\ldots,t_{sn} \right) = P_{i}\prod\limits_{k = 1}^{n}\left( 1 + F^{(1)}\left( \frac{t_{sk}}{P_{k - 1}} \right) \right),} \\ \end{array}$$ where *P*~0~ = *P*~*i*~ is the intrinsic firing period of the isolated neuron, *t*~*sk*~ \> *t*~*s*(*k*\ +\ 1)~, and *t*~*sk*~ \< *P*~*k*−1~ (stimulus *k* still falls inside the transiently modified period due to the previous stimulus). Appendix 2 {#sec014} ========== **Stability Conditions for Two-Dimensional Recursive Maps.** The characteristic polynomial of any first order recursive equation of two variables, such as [Eq (6)](#pone.0174304.e014){ref-type="disp-formula"}, is: $$\begin{array}{r} {P\left( \lambda \right) = \lambda^{2} - \left( Tr\left( A \right) \right)\lambda + Det\left( A \right),} \\ \end{array}$$ where Tr(A) and Det(A) are the trace and, respectively, the determinant of the recursion matrix of the perturbations (*δt*~2*sa*~, *δt*~2*sb*~), such as the [Eq (15)](#pone.0174304.e076){ref-type="disp-formula"}. The first order recursions have the following solution: $$\begin{array}{r} {\delta t\left\lbrack n \right\rbrack = C_{1}\lambda_{1}^{n} + C_{2}\lambda_{2}^{n},} \\ \end{array}$$ where *C*~1~ and *C*~2~ are some constants determined from the initial conditions and *λ*~*i*~ (with *i* = 1, 2) are the solutions of the characteristic polynomial [Eq (24)](#pone.0174304.e102){ref-type="disp-formula"}. For the perturbations to die out, all characteristic roots must be less than unit, i.e. \|*λ*~*i*~\|\<1 for both *i* = 1, 2. To ensure stability, there are two possibilities: (1) the roots of the characteristic polynomial are real and both less than the unit, or (2) the roots are complex conjugated with a magnitude less than the unit. **Real characteristic roots.** In this case, the following conditions must be met $$\begin{array}{rcl} {P\left( - 1 \right) = 1 + Tr\left( A \right) + Det\left( A \right)} & > & {0,} \\ {P\left( + 1 \right) = 1 - Tr\left( A \right) + Det\left( A \right)} & > & {0,} \\ {Det\left( A \right) - \left( Tr\left( A \right) \right)^{2}/4} & > & {0.} \\ \end{array}$$ The region where all three conditions are met is shown in [Fig 7](#pone.0174304.g007){ref-type="fig"} with crossed hashing, i.e. the region below the parabolic curve and above the two straight, tangent, lines. {#pone.0174304.g007} **Imaginary characteristic roots.** In this case, the discriminant of the characteristic polynomial is negative, i.e. −*Det*(*A*) + (*Tr*(*A*))^2^/4 \< 0. At the same time, the magnitude of each complex conjugated characteristic root is $\left| \lambda \right| = \sqrt{Det\left( A \right)} < 1$, i.e. *Det*(*A*) \< 1. As a result, the stability region in the case of complex characteristic root is above the parabolic shape shaded with horizontal lines in [Fig 7](#pone.0174304.g007){ref-type="fig"}. Supporting information {#sec015} ====================== ###### Mathematica code. The Mathematica code simulates the driven-driver neural network with adaptive feedback. It uses Morris-Lecar type 1 neurons and chemical couplings between neurons to produce a stable phase-locked firing pattern. (NB) ###### Click here for additional data file. We are grateful to the reviewers for their constructive comments that allowed us to improve the computational implementation of the neural network and the overall quality of the manuscript. This research was supported by US National Science Foundation Career Award IOS-1054914 to SAO. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceptualization:** SAO.**Data curation:** SAO.**Formal analysis:** SAO.**Funding acquisition:** SAO.**Investigation:** SAO DIA.**Methodology:** SAO DIA.**Project administration:** SAO.**Resources:** SAO.**Software:** SAO DIA.**Supervision:** SAO.**Validation:** SAO.**Visualization:** SAO.**Writing -- original draft:** SAO.**Writing -- review & editing:** SAO DIA.

Introduction {#sec1-1} ============ **What was known?** A fixed combination of an antimicrobial and retinoid is effective in the management of mild to moderate acne. Nadifloxacin, an antimicrobial and adapalene, a retinoid have frequently been used separately by dermatologists in the management of acne. A fixed combination of an antimicrobial and retinoid is effective in the management of mild to moderate acne. Nadifloxacin, an antimicrobial and adapalene, a retinoid have frequently been used separately by dermatologists in the management of acne. Acne vulgaris is a chronic inflammatory dermatosis which is notable for open and/or closed comedones (blackheads and whiteheads) and inflammatory lesions including papules, pustules, or nodules.\[[@ref1]\] It is the most common problem among both males and females, between puberty and 30 years and greatly affects the quality of life (QoL) life and psychosocial development.\[[@ref2]\] The update from global alliance to improve outcomes in acne supports the use of fixed combination of topical retinoid and topical antibiotic (Level 1 evidence) with the combination being the first choice in the treatment of mild to moderate acne.\[[@ref3]\] The India acne alliance consensus document also recommends that a combination of topical retinoid and topical antimicrobial be employed early in the treatment of mild to moderate acne.\[[@ref2]\] Nadifloxacin, a topical fluoroquinolone, is reported to have potent action against *P. acne*, *S. epidermidis* and methicillin-resistant *Staphylococcus aureus* (MRSA), with no cross-resistance with any other antibiotic or with another fluoroquinolone. Previous studies have reported that topical application of nadifloxacin cream exhibited excellent efficacy and tolerability, and did not induce resistance in *P. acnes* strains.\[[@ref4][@ref5][@ref6][@ref7]\] Adapalene, a topical retinoid is known to modulate keratinization and possesses anti-inflammatory action.\[[@ref8]\] Previous studies have reported adapalene to be safe and well tolerated as compared with other retinoids.\[[@ref9][@ref10][@ref11]\] The present study evaluated the efficacy and safety of the topical premixed formulation of nadifloxacin 1% and adapalene 0.1% in the treatment of acne vulgaris. The study protocol was approved by respective Institutional Ethics Committees and Independent Ethics Committee (Ethics R U, Mumbai, for the sites other than institutional sites) and was conducted in accordance with the Declaration of Helsinki and good clinical practice guidelines. The clinical trial registry (CTRI) number of the study is CTRI/2013/05/003616. Materials and Methods {#sec1-2} ===================== Male or female subjects (≥12 years) suffering from mild to moderate acne on the face, with mixed non-inflammatory comedones and inflammatory papulo-pustular lesions were enrolled in the present study. Subjects with known hypersensitivity to individual ingredients or any other closely related agent from quinolone or retinoid class, those suffering from severe acne requiring systemic drugs for management or any severe concomitant diseases, patients receiving any systemic antibiotics during two weeks prior to study, pregnant women and nursing mothers were excluded from the study. This was an 8 week (Sep 2010-May 2011), non-randomized, open-labeled, prospective study conducted at five centers (B J Medical college and Civil Hospital, Asarwa Ahmedabad, Mukhi Skin Clinic Nagpur, Rajiv Gandhi Medical College and Chhatrapati Shivaji Maharaj Hospital, Kalwa, Thane, M. S. Ramaiah Medical College, MSR Nagar, Bangalore, and Lokmanya Tilak Municipal General Hospital and Lokmanya Tilak Municipal Medical College Mumbai) across India. The study consisted of three follow up visits- end of week 2 (Day 14 ± 2), end of week 4 (Day 28 ± 2), and end of week 8 (Day 56 ± 2). All subjects were screened after obtaining a written informed consent for participation in the study. Eligible subjects were asked to apply 1% nadifloxacin and 0.1% adapalene gel (manufactured by Wockhardt Ltd, Aurangabad, Aug 2010, batch no: SSD (546)-03-09, expiry: July 2012) over the affected area of the face once daily in the night after washing face and before going to bed. The drugs were supplied by the investigator. The efficacy parameters were evaluated at each visit. At the end of week-8, the subjects were also investigated for laboratory parameters - complete blood count (CBC), liver profile \[Serum Glutamic Oxaloacetic Transaminase (SGOT) and Serum Glutamic Pyruvate Transaminase (SGPT)\], and renal profile \[Blood Urea Nitrogen (BUN)\] and serum creatinine. The efficacy of the 1% nadifloxacin and 0.1% adapalene gel was evaluated based on the reduction from baseline in the number of non-inflammatory lesions (comedones), reduction in the number of inflammatory lesions (papules, pustules, nodules), reduction in the total number of acne lesions, reduction in the severity of acne as per combined acne severity classification,\[[@ref12]\] and improvement in global assessment scale of acne toward normal clear skin. For this study, 30 subjects were planned to be enrolled at each center to arrive at a sample size of 100 analyzable subjects. Descriptive statistics was used for demographic variables. For total, inflammatory and non-inflammatory lesion counts, Friedman test was used for a comparison from baseline. For the comparison of acne severity scores, global assessment score from baseline to end of week 8, gamma statistics was used. Gamma statistics measures the strength of association of the cross tabulated data (baseline and end of week-8 data in this case) with variables when both are measured at the ordinal level. The laboratory parameters were analyzed using Wilcoxon\'s signed rank test. Data was analyzed using SPSS analytic software 9.2, (USA, Enterprise Guide version 4.2). Results {#sec1-3} ======= Overall, 119 subjects were enrolled (Ahmedabad-30, Nagpur-30, Bangalore-30, Thane-18, and Mumbai-11) in the present study. Of these, two subjects were lost to follow-up and one was withdrawn from the study due to "flare-up of lesions" and was considered as a treatment failure. Thus, 116 (97.5%) subjects completed the study. A total of 117 subjects (intention to treat population) were considered for efficacy analysis including the incomplete but evaluable subject who was withdrawn from the study \[[Figure 1](#F1){ref-type="fig"}\]. The baseline demographic characteristics of the subjects are presented in [Table 1](#T1){ref-type="table"}. The age of the patients ranged from 13 to 41 years; a majority were females (60%) and most of the subjects had an oily skin (79%). {#F1} ###### Demographic characteristics of subjects  A statistically significant progressive reduction in non-inflammatory lesion counts, inflammatory lesion counts, and total lesion counts were seen at the end of week 2, end of week 4 and end of week 8. Reduction and percentage change in acne count from baseline is presented in [Table 2](#T2){ref-type="table"}. ###### Percentage reduction and lesion count analysis at baseline and at end of week 2, 4 and 8  Overall, of the 117 subjects, 115 subjects (98.3%) showed improvement. Treatment failure was noted in two subjects (1.7%) as the 1% nadifloxacin and 0.1% adapalene gel did not work effectively. A progressive reduction in acne severity was also observed. Of the total four subjects with severe acne at baseline, three had severe acne at week 4 (25% reduction) and only one had severe acne at week 8 (75% reduction; *P* \< 0.001) \[[Table 2](#T2){ref-type="table"}\]. A progressive reduction in physician\'s global assessment scores of acne severity was noted (*P* \< 0.001 for within group comparison at week 8 with respect to baseline, [Figure 2](#F2){ref-type="fig"}). The laboratory parameters at the baseline and at the end of week 8 did not show any clinically significant difference. ![Global Assessment of Acne Severity at baseline and at weeks 2, 4 and 8 (\*Indicates *P* \< 0.001 for within group comparison at week 8 with respect to baseline) (Total 117 subjects \[ITT\] were evaluated)](IJD-59-385-g004){#F2} All AEs reported in the study were mild to moderate in severity. Overall, 24 AEs were reported from three centers (15 subjects). No serious adverse events (SAEs) whether expected or unexpected were reported from any participating centers \[[Table 3](#T3){ref-type="table"}\]. The most frequent AE was application site dryness. ###### AEs observed during the study  Discussion {#sec1-4} ========== The results of the present study demonstrate a significant reduction in both the inflammatory and non-inflammatory acne lesions over an 8-week period of treatment with a fixed dose combination of 1% nadifloxacin and 0.1% adapalene gel. This reduction in the total acne lesions over the entire treatment period correlates well with the simultaneous decrease in the severity of acne. Thus, by the end of 8 weeks, 74% of the patients were in the mild severity acne group with 71.5% of the patients having their score approaching to that of normal healthy skin by the end of week 8. Similarly, progressive reduction in acne severity (69.2%) and physician\'s global assessment score for acne severity (75.0%) at the end of week 8 was also noted. Acne vulgaris is a chronic inflammatory disease with a complex pathogenesis. As multifactorial effects are demanded from acne therapy, combination therapy is often recommended.\[[@ref13][@ref14]\] According to recent guidelines regarding acne therapy, oral antibiotics should not be used alone, and combination with a topical retinoid is highly recommended in cases of an inflammatory disease. Topical retinoids target the microcomedones, which are the precursor of acne lesions, and are the preferred drugs for maintenance therapy. Adapalene is a third-generation topical retinoid, which is safe, efficient, well tolerated.\[[@ref15][@ref16]\] Adapalene is a synthetic naphthoic acid derivative with retinoid activity, and it interacts with unique receptors because of its different chemical structure as compared with other retinoids. This property makes adapalene a potent modulator of cellular differentiation, keratinization, and inflammatory processes.\[[@ref17]\] Several published studies have established the efficacy and safety of nadifloxacin in the management of acne.\[[@ref18][@ref19][@ref20]\] A combination of retinoid and an antibiotic is effective in acne patients as antibiotics can penetrate up to deeper layers of skin given their property of comedonal drainage (which facilitates the penetration of other topical agent) and the thinning effect of retinoids on stratum corneum.\[[@ref2][@ref21]\] Nadifloxacin is a synthetic bactericidal fluoroquinolone with a broad spectrum antibacterial activity against aerobic Gram-positive and Gram-negative and anaerobic bacteria, including *P. acnes* and *S. epidermidis*.\[[@ref5][@ref6][@ref20]\] In recent years, antibiotic-resistant propionibacteria have been isolated with increasing frequency and carriage of resistant strains has been associated with antibiotic treatment failure.\[[@ref22]\] A systematic review identified 12 studies that investigated the presence of resistant *P. acnes* in the skin of people who had received topical or systemic antibiotics.\[[@ref23]\] Nadifloxacin, by virtue of its proven potent action and tolerability and relative newness, is a newer alternative to clindamycin for the management of acne. A 10-year surveillance study (*n* = 4274) in the UK found that resistance of *P. acnes* to erythromycin and clindamycin increased from 34.5% in 1991 to 55.5% in 2000 (peaking at 64% in 1997).\[[@ref24]\] Such high risk of antibiotic resistance necessitates the search for newer potent antibiotics that can be judiciously combined with retinoids to ensure better efficacy. Several published studies have established the efficacy and safety of nadifloxacin in the management of acne.\[[@ref18][@ref19][@ref20][@ref25]\] A comparative analysis of the efficacy and safety of the combination of antibiotic and retinoid (adapalene) with adapalene monotherapy in patients with moderate to severe acne showed that the combination therapy led to a significantly greater reduction in the number of acne lesions after 8 weeks than adapalene monotherapy. The combination therapy improved inflammatory acne earlier than adapalene along with a 40% reduction in the lesions by week 2 with the combination compared to about 19% reduction with adapalene alone.\[[@ref26]\] Similarly, the results of the present study show 98.2% reduction in the lesions by the end of week 8. In the global assessment score for acne severity, an aggravation in acne severity was observed at week 2 among the patients with score 5. This could be due to the unmasking of underlying micro-comedones caused by adapalene in the combination. Initial flare up of acne is a common observation with retinoids. In the present study, only 15 subjects reported 25 AEs, which were mild to moderate in severity. Thus, the results of the current study have demonstrated that once daily use of fixed dose combination of 1% nadifloxacin and 0.1% adapalene gel is a safe and well-tolerated topical treatment for mild to moderate acne. In conclusion, the combination of nadifloxacin (1%) and adapalene (0.1%) has been found to be effective in the treatment of mild to moderate acne with good tolerability. However, the lack of a control group and non-comparative design of study limits the utility of these results. We cannot also deny any bias in the study due to its open labeled design. These findings should be considered as preliminary findings and further research is needed in the form of well-designed, adequately powered trials with a control group to understand the efficacy of fixed combination of nadifloxacin (1%) and adapalene (0.1%) treatment in Indian subjects. **What is new?** Though these two products are being used separately in the management of acne, this is the first time that this fixed combination is formulated and evaluated in Indian patients The fixed combination of adapalene (0.1%) and nadifloxacin (1%) could be a safe and effective treatment for mild to moderate acne vulgaris. "All authors participated in analysis and interpretation of data and critical revision of the manuscript. The authors acknowledge Knowledge Isotopes (professional medical writing company, [www.knowledgeisotopes.com](htt://www.knowledgeisotopes.com)) for writing this manuscript and subsequently revising the draft by incorporating author comments." **Source of Support:** This Study was supported financially by Wockhardt Ltd, Mumbai **Conflict of Interest:** Authors BJS, TKS, RSD, RGT, VV, and JIM have received research funding for the conduct of this trial from Wockhardt Ltd, Mumbai. Authors GK, PPA are employees and are stake holders by means of salaries at Wockhardt Ltd, which sponsored the study.

Background ========== The Endogenous Cannabinoid (EC) system comprises at least two cannabinoid receptors, the CB~1~and CB~2~receptors, a series of lipophilic endogenous ligands, the endocannabinoids (ECs), and enzymes for EC biosynthesis and degradation \[[@B1]\]. Mounting evidence indicates that it is involved in the physiological modulation of many crucial functions in both the peripheral and the central nervous system \[[@B1]\]. It has been hypothesized that an alteration of EC neurotransmission could play a role in neurological, psychiatric and immunological disorders \[[@B2],[@B3]\]. In this context, it is possible that, along with other neurotransmitter systems (i.e., dopaminergic, serotonergic), anomalies of the ECs might take part in producing the clinical picture of schizophrenia. This hypothesis relies on the following considerations: 1\) Subjects with acute cannabis intoxication often display a schizophrenia-like syndrome, with hallucinations, altered judgement, false beliefs, and cognitive impairment \[[@B4]\]. In other, predisposed individuals, cannabis can precipitate a psychotic illness \[[@B5]\], although this does not necessarily indicate a causative role of ECs in mental disorders. Finally, lack of motivation, apathy and avolition are almost invariably observed in long-term cannabis users, so as to mimic the picture of chronic or residual schizophrenia. 2\) Under physiological conditions, the EC system participates in the regulations of important functions, such as rest, cognition, movement, memory, and perceptions \[[@B6]\]. Many of these functions are actually altered in the course of schizophrenia. 3\) There appears to be a substantial overlap between the areas commonly believed to be involved in the pathogenesis of schizophrenia \[[@B7]\] and those expressing the highest concentrations of EC receptors in the CNS \[[@B2]\], i.e. the limbic system (hippocampus/amygdala), nigro-striatal areas, and the prefrontal cortex, among the others. There has been evidence in the recent literature \[[@B7]-[@B9]\] that patients suffering from schizophrenia have detectable differences in their EC signalling when compared to normal controls. Leweke et al. \[[@B8]\] reported elevated levels of the endogenous cannabinoid receptor ligand anandamide \[[@B10]\] in the Cerebro-Spinal Fluid (CSF) of patients with schizophrenia. Voruganti et al. \[[@B9]\] have shown a correlation between the severity of psychotic symptoms and an increase in cannabis-induced striatal dopaminergic neurotransmission in a patient with the disorder. Finally, polymorphism of the gene encoding the CB~1~receptor, which is the cannabinoid receptor subtype mostly expressed in the brain, has been associated with increased susceptibility to hebephrenic schizophrenia \[[@B11]\], thus suggesting that a malfunctioning EC system could play a role in the etiopathogenesis of this disorder. There have been several reports that schizophrenia is accompanied by overt alterations in the immune response, as well as by changes in the function of immune blood cells, and that many of these alterations can be normalized by anti-psychotic drugs \[\[[@B12]-[@B14]\] and \[[@B15]\] for review\]. In particular, a significantly increased number of activated macrophages and lymphocytes has been detected in the CSF of schizophrenic patients during acute psychotic episodes \[[@B16],[@B17]\]. Since activated macrophages and lymphocytes release significant amounts of ECs \[[@B18]-[@B20]\], it is possible that these blood cells contribute to some extent to the previously observed elevated levels of anandamide in the CSF of patients with a diagnosis of schizophrenia \[[@B8]\]. It is worthwhile noting that some of the immune functions previously found to be altered during acute schizophrenia, such as interleukin and tumor necrosis factor-α release, are also known to be influenced by ECs acting at both CB~1~and, particularly, CB~2~cannabinoid receptors in macrophages, lymphocytes and dendritic cells \[[@B3]\]. Notwithstanding the above observations, and despite the fact that ECs can easily cross the blood brain barrier \[[@B10]\], the levels of these compounds in the blood of schizophrenic patients have never been assessed. Aims of the study ----------------- We analysed the peripheral blood of patients and normal controls, in order to detect any differences in the levels of: (i) the endogenous ligands of cannabinoid receptors (the endocannabinoids, \[[@B10]\]); (ii) the cannabinoid CB~1~and CB~2~receptors, and (iii) the fatty acid amide hydrolase (FAAH), one of the enzymes mostly involved in endocannabinoid inactivation \[[@B21]\]. We report that anandamide plasma levels are elevated in untreated patients with schizophrenia, and that the amounts of this compound as well as of CB~2~receptors and FAAH are decreased after a successful pharmacological treatment. Results ======= Psychiatric evaluation ---------------------- Table [1](#T1){ref-type="table"} displays the data related to the BPRS total score for all the patients included in the study. In addition, the patients in subgroup 1 have been rated as ranging from moderately ill to severely ill (CGI score from 4 to 6). All patients who were reassessed showed a statistically significant difference between the pre- and post-treatment scores at the BPRS and CGI, which dropped by more than 50% in all 5 cases considered (Table [1](#T1){ref-type="table"}). Patients 3, 5 and 8 in subgroup 2 scored 1 (much improved), whereas patients 6 and 7 in the same subgroup were evaluated as moderately improved (CGI score = 2). ###### BPRS and CGI scores and anandamide blood levels of the twelve patients with schizophrenia included in this study. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Patient Number** **Anandamide levels in the blood during the acute phase of schizophrenic illness (pmol/ml)** **Anandamide levels in the blood during remission of schizophrenic illness (pmol/ml)** **BPRS (and CGI) Scores Acute Remission** -------------------- ---------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------- ------------------------------------------- 1 11.7 59 (5) 2 7.3 64 (5) 3 8.4 4.4 68 (6) 25 (1) 4 8.2 48 (4) 5 6.3 6.2 60 (5) 20 (1) 6 5.3 2.2 66 (6) 29 (2) 7 7.7 4.1 56 (5) 26 (1) 8 6.7 2.5 70 (6) 29 (2) 9 7.5 55 (4) 10 9.7 57 (4) 11 8.7 56 (4) 12 6.0 52 (5) **7.79 ± 0.50**\ **3.88 ± 0.72**\ **(P = 4.5 × 10^-8^*vs.*Control)** **(P = 0.16 *vs.*Control)** -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Five of these patients were assessed before and after remission of the symptoms (indicated by the significant decrease in the global BPRS and CGI scores) induced by treatment with olanzapine. CGI score ranged from 6 (severely affected) to 1 (much improved), and are shown in parentheses. The amount of anandamide in healthy volunteers (control) was 2.58 ± 0.28 pmol/ml (mean ± SEM, n = 20). Means were compared to the control by the unpaired Student\'s t test, threshold of significance was 0.95. Anandamide, cannabinoid receptor and FAAH blood levels ------------------------------------------------------ Table [1](#T1){ref-type="table"} also shows the blood levels of anandamide in each of the twelve patients affected by schizophrenia. The mean ± SEM amounts of anandamide levels in the control volunteers and the patients with acute schizophrenia are also shown in Table [1](#T1){ref-type="table"} and in its legend. A statistically significant difference between the two groups was found, whereas the mean values for the patients in remission were not significantly different from those of the control subjects. Clinical improvement, with the reduction in the BPRS and CGI scores, was paralleled by the decrease in both the levels of peripheral ANA (assessed in those 5 patients sampled before and after pharmacological treatment, Table [1](#T1){ref-type="table"}, Fig. [1](#F1){ref-type="fig"}) and of the intensity of the bands for the mRNA transcripts of FAAH and CB~2~(assessed in those 5 patients sampled before and after pharmacological treatment, Fig. [2](#F2){ref-type="fig"}). We did not observe any consistent difference for the CB~1~mRNA transcript, which appeared to be less expressed than that of CB~2~in all the blood samples analysed (Fig. [2b](#F2){ref-type="fig"}). {#F1} {ref-type="table"}), in acute phase; lanes 2, 4, 6, 8 and 10: patients n° 3, 6, 8, 5 and 7, in remission. Panel A, B, C and D are relative to PCR analysis for β~2~-microglobulin (house-keeping), CB~1~, CB~2~and FAAH, respectively. The bands shown in the figure are from 20 (Panel A), 35 (Panel B), 30 (Panel C) and 30 (Panel D) PCR amplification cycles.](1476-511X-2-5-2){#F2} Discussion ========== Despite several suggestions that the EC system may play a role in the pathogenesis of schizophrenia \[[@B7],[@B9],[@B22]\], and the finding of elevated levels of anandamide in the CSF of patients suffering from schizophrenia \[[@B8]\], no study has addressed so far the question of whether the EC system is also altered in the serum and mononucleated cells from the blood of patients with this mental disorder. This issue is important in view of the several immunological correlates of schizophrenia, which include, among others, a shift from a Th-1-type to a Th-2-type immune response, a significant increase of activity of blood monocytes/macrophages, and a corresponding up-regulation of cytokine release \[[@B15]\]. Indeed, the ECs have a well-established immune-modulatory effect \[[@B23],[@B24],[@B20]\]. Thus, possible changes in the blood levels of these endogenous mediators could explain in part, or be a consequence of, the modified immune response observed in the course of schizophrenia. Furthermore, since the strong impact of acute schizophrenia on some immune functions can be attenuated by treatment with anti-psychotic drugs \[[@B15],[@B25]\], the effect of antipsychotic medications on the possible modifications of the EC system in the blood of these patients also needs to be assessed. Our study suggests the existence of a significant alteration in the peripheral blood levels of schizophrenic patients of both the endocannabinoid anandamide and the mRNA for the anandamide degrading enzyme, FAAH. Furthermore, we found that the elevation of anandamide levels in our sample is confined to the acute phases of disease, to then normalize after a successful pharmacological treatment. Eventually, there were no statistically significant differences in anandamide levels between treated patients and controls. However, no direct correlation between anandamide blood levels and the BPRS or CGI scores of each patient, be it before or after remission, was found. Our results suggest that an acute psychotic episode might be characterised, among other factors, by an elevation in the peripheral concentration of anandamide, thus possibly inducing a compensatory increased expression of the degrading enzyme, FAAH, in an attempt to normalise their circulating levels. It is of interest that our patients consistently showed the same pattern, with their level of circulating anandamide forming two well-identifiable and distinct clusters at around 8 and 3 pmol/ml for group 1 and 2, respectively. Anti-psychotic medications seem to have played a crucial role in these patients by thwarting clinical symptoms on the one hand, and, in parallel, by reducing the levels of endocannabinoids and FAAH mRNA on the other hand. We did not identify any patients failing to respond to their pharmacological treatment, although it would have been interesting to measure the concentration of anandamide in the absence of an improved clinical presentation. This event would have carried important information as to whether decreased EC concentration was related to clinical amelioration or rather to pharmacological treatment per se, regardless of clinical outcome. The peripheral elevation of anandamide levels might be related to the hypothesized anomaly of this signalling system in the brain of patients with schizophrenia, since these compounds can easily cross the blood brain barrier \[[@B10]\]. However, it is unlikely that changes in the peripheral amounts of ECs found in the blood can reflect to a great extent alterations occurring at the CNS level, in schizophrenia as well as other neurological and psychiatric disorders, since these compounds: (i) due to their lipophilicity act as autocrine/paracrine mediators, and no blood EC carrier protein has been identified to date; and (ii) are also produced by blood cells. More probably, the changes in anandamide blood levels found here might be a consequence, or alternatively one of the causative factors, of the previously observed immunological abnormalities in patients with schizophrenia \[[@B15]\]. For example, in view of the negative effect of leptin on endocannabinoid biosynthesis \[[@B26]\], and of the reduced levels of this hormone during acute schizophrenia \[[@B27]\], it is possible that reduced leptin causes the increased levels of blood anandamide observed in the schizophrenic patients investigated in the present study. This hypothetical mechanism would explain also why olanzapine was found here to lower EC levels back to those observed in healthy volunteers, since treatment of schizophrenia with this antipsychotic drug was shown previously to restore normal serum leptin levels \[[@B28]\]. Alternatively, since antipsychotics are also known to reduce the number and/or activity of leukocytes \[[@B15]\], it is possible that the reduced levels of anandamide following olanzapine treatment are a mere consequence of the reduced activity of lymphocytes and monocytes, which are normally capable of producing anandamide only after activation \[[@B19],[@B20],[@B29]\]. Regarding the reduction, also observed here, of cannabinoid CB~2~receptor mRNA following treatment with olanzapine, again this could be a consequence of reduced activity of blood leukocytes, since also CB~2~expression has been previously shown to be subject to regulation in activated macrophages and leukocytes \[[@B3]\]. Interestingly, also the function of G-proteins in the blood has been recently shown to be down-regulated after treatment with neuroleptic drugs \[[@B30],[@B31]\], and since CB~2~is a G-protein-coupled receptor \[[@B1],[@B10]\], it is likely that the reduction of the expression of this receptor detected after treatment with olanzapine and subsequent recovery from the symptoms of schizophrenia is also accompanied by a corresponding decrease of CB~2~receptor functionality. Clearly, blood endocannabinoids cannot be regarded yet as possible markers of acute psychotic disorders, and even less as predictors of the patients\' response to antipsychotic medication. Furthermore, we did not test the specificity of ECs alterations to schizophrenia, and it may be that other major psychoses exhibit similar biochemical aberrations, and perhaps similar stage-related fluctuations. Therefore, our data warrant further investigations with the aim of better understanding the pathological relevance of this as well as of other correlative studies. Conclusions =========== The present study carries the following interesting and potentially important implications: 1\) It lends further support to the hypothesis of an involvement of the EC system in schizophrenia, clearly demonstrating an anomaly in three separate components of this system, anomaly that, at least in our sample, appears to be associated with an acute phase of the illness. However, the small size of our sample warrants for great caution when interpreting these results. 2\) It suggests that at least part of the immunological abnormalities observed during acute schizophrenia might correlate with changes in the \"output\" of the EC system in the blood. Methods ======= Study design, subjects and psychiatric assessment and treatment --------------------------------------------------------------- A case-control study design was used. Subjects were selected from patients treated in the clinical facilities of the Department of Mental Health in Pomigliano, Naples, Italy, from January to December 2001. Patients were selected only if showing, or described in the medical records as affected by, symptoms of schizophrenia, and if wishing to take part in the study. After this first step, the recruited subjects received a diagnostic assessment, and only those meeting the DSM-IV criteria for schizophrenia \[[@B32]\] entered the study. Other necessary criteria for inclusion were: absence of previous or present neurological disorders, no abuse of cannabis in the year preceding the study, and absence of a significant learning disability (IQ\<85). All the patients included had to be in a quite controlled setting (namely, their family environment), where no abuse of cannabis and compliance to treatment could be assessed with a high degree of certainty. These strict criteria carried inevitable reflections on the number of patients included in the study. The individuals in our research (n = 12) were split in two subgroups. In subgroup 1 were those in an acute phase of their schizophrenic illness, who had not received pharmacological treatment for at least 30 days prior to evaluation, 9 males and 3 females, with a mean age of 32.9 ± 7.0 years (range, 18--45 years), of Caucasian race (n = 12). All individuals received the Diagnostic Interview for Genetic Studies (DIGS; \[[@B33]\]), a semi-structured DSM IV interview. All patients were rated also by means of the Clinical Global Impression scale (CGI; \[[@B34]\]), and the treatment outcome was evaluated with the CGI-Improvement (CGI-I). We used the Brief Psychiatric Rating Scale (BPRS; 18-item version; \[[@B35]\]) to characterize the severity of the symptoms of the illness, where 1 indicates absent and 7 severe. The second subgroup (n = 5) consisted of patients previously assigned to subgroup 1, who had achieved clinical remission since at a least one month, following a successful pharmacological treatment with olanzapine (patients 3 and 5 were on 20 mg/day, whereas subjects 6, 7 and 8 received 25 mg/day). The re-test procedure for the 5 probands was performed 92 ± 11 days after the assessment in the acute phase of the illness. Clinical remission was defined a priori as at least 50% reduction on the BPRS score and a clinical evaluation of 1 (much improved) or 2 (moderately improved) at the CGI-I score. In particular, patient Nr. 3 was investigated during the third episode of her disease. She had been treated with risperidone at the onset. We switched to olanzapine because of hyperprolactinemia. Her response to treatment was satisfactory with both medications, and she always achieved a nearly complete remission. Relapses were apparently caused by an excessive reduction or a spontaneous discontinuation of her medication regime. Patient Nr. 5 was suffering from the second episode of schizophrenia. We treated her with risperidone (first episode) and then olanzapine for the episode described here. The relapse was caused by her arbitrarily stopping her medication. Patient Nr. 6 was at his first episode of schizophrenia, which however had lasted for a prolonged time because it went untreated (more than a year). In fact, his family postponed his referral to mental health facilities because of fear of stigmatization. We started him on olanzapine 20 mg/day, but he needed 25 mg to show a satisfactory response (CGI from 6 to 2). Patient Nr. 7 was experiencing several positive symptoms since two years, with serious consequences such as loss of his job. His schizophrenia did not respond to two classical antipsychotics administered by a previous psychiatrist. The initial 20 mg of olanzapine had to be titrated up to 25 to obtain the optimal response. Finally, patient Nr. 8 was treated with olanzapine since he was included in the study. He had previously not agreed to be treated, and had been suffering from several positive symptoms for approximately seven years. He only had occasional administrations of haloperidol, which was added to his food by the patient\'s wife without him being aware of that. In retrospect, there appeared to be no response because of these treatment inconsistencies. Twenty normal controls were recruited from the medical and nursing staff at the Department of Mental Health in Pomigliano. They were in the same age range as the study patients, and were likewise comparable for level of education and gender representation. A normal IQ, absence of previous and present neurological disorders, and having not abused marijuana/exogenous cannabinoids were used as inclusion criteria for controls. The control subjects had no first-degree relative suffering from a major psychiatric disorder. All subjects gave 10 ml blood for the laboratory testing. All blood samples were taken in the morning, from non-fasting subjects, and there were no feeding-related variables that could account for any possible difference in ECs levels between groups. Biochemical analyses -------------------- ### Lipid Extraction Peripheral blood samples (10 ml) were collected by vein suction in EDTA (25 mM final concentration) and processed no later than 3 hours after blood was drawn. For the LC-MS determinations, EDTA and phenyl-methyl-sulfonyl-fluoride (PMSF) (100 μM final concentration) were added to blood samples. PMSF is an inhibitor of fatty acid amide hydrolase (FAAH), and was added in order to prevent endocannabinoid degradation. Five ml of whole blood for each sample was carefully layered over 5 ml of *HISTOPAQUE*^®^1077 solution (SIGMA) in *ACCUSPIN*^®^tubes (SIGMA) and centrifuged at 400 × g for 30 min at room temperature. After centrifugation, erythrocytes and granulocytes sedimented at the bottom of the tube. After removal of the clear plasma layer from the top of the tube, the opaque layer (about 2 ml) containing the mononuclear cells was collected, resuspended by gentle aspiration in 10 ml of phosphate buffered saline (PBS) and centrifuged at 250 × g for 10 min at room temperature. The cell pellet was resuspended, washed twice in 5 ml of PBS and stored at -80°C for RNA extraction. For quantitative determinations, the plasma and the mononucleate cell layers were collected together, the proteins precipitated by adding 3 vol. of acetone, the supernatants were collected and subjected to lipid extraction with methanol/chloroform. Enough of each solvent was added to reach a final ratio buffer/methanol/chloroform of 1:1:2 (v/v/v). Methanol containing 5 pmol of d~8~-anandamide was added as internal standard. The organic phase was then dried under nitrogen and purified by means of open bed chromatography on silica gel \[[@B36],[@B37]\]. ### Liquid chromatography-Atmospheric pressure chemical ionization-mass spectrometry (LC-APCI-MS) quantification of anandamide levels The lipid extracts were analyzed by liquid chromatography-atmospheric pressure chemical ionization-mass spectrometry (LC-APCI-MS) by using a Shimadzu HPLC apparatus (LC-10ADVP) coupled to a Shimadzu (LCMS-2010) quadrupole MS via a Shimadzu APCI interface. MS analyses were carried out either in the selected ion monitoring (SIM) mode as described previously \[[@B26]\]. The temperature of the APCI source was 400°C, the HPLC column was a Phenomenex (5 μm, 150 × 4.5 mm) reverse phase column, eluted as described \[[@B37]\]. Anandamide (retention time 14.5 min) was quantified by isotope dilution with the above-mentioned deuterated standards (same retention time and *m*/*z*= 356.3) and its amounts in pmoles normalized per ml of processed blood. Intra-assay variation using this method was 2.5 ± 0.3%, whereas inter-assay variation was 10.1 ± 2.5% (means ± SEM, n = 4). ### RNA extraction and semi-quantitative reverse transcriptase-polymerase chain reaction (RT-PCR) analyses Total RNA was extracted from mononuclear cells by Trizol^®^reagent (Life Technologies) according to the manufacturer\'s intructions. To remove contaminant DNA, 4 μg of RNA samples were DNAse-digested utilizing the DNA-free^®^(Ambion) protocol. 2 μg of total RNA were reverse transcribed in a 20 μl reaction mixture containing: 75 mM KCl, 3 mM MgCl~2~, 10 mM dithiothreitol, 1 mM dNTPs, 50 mM Tris-HCl pH 8.3, 20 units of RNAse inhibitor (Roche), 0.125 A~260~units of hexanucleotide mixture (Roche) for random-priming and 200 units of reverse transcriptase (Superscript^®^, GIBCO). The reaction mixture was incubated at room temperature for 10 minutes and at 42°C for 90 min, then the reaction was stopped by heating at 95°C for 5 min, cooled on ice, and stored to -20°C. Control samples (no-RT) were prepared by omitting reverse transcriptase in the retrotrascription mixture. DNA amplification was performed in a 50 μl PCR reaction mixture containing: 0.5--2 μl of the retro-trascription mixture, 1X PCR buffer (supplied as component of the DNA polymerase kit), 2 mM MgCl~2~, 250 μM dNTPs, 0.5 μM each of 5\' and 3\' primers and 2.5 units of *Platinum*^®^*Taq*DNA polymerase (Life Technologies). The mixtures were amplified in a PE Gene Amp PCR System 2400 thermocycler (Perkin Elmer). The primers used were: CB~1~sense primer, 5\'-CAG AAG AGC ATC ATC CAC ACG TCT G-3\'; CB~1~antisense primer 5\'-ATG CTG TTA TCC AGA GGC TGC GCA GTG C-3\'; CB~2~sense primer 5\'-TTT CCC ACT GAT CCC CAA TG-3\'; CB~2~antisense primer 5\'-AGT TGA TGA GGC ACA GCA TG-3\'; FAAH sense primer 5\'-GCC TGG GAA GTG AAC AAA GGG ACC-3\'; FAAH antisense primer 5\'-CCA CTA CGC TGT CGC ACT CCG CCG-3\'; β~2~-microglobulin sense primer 5\'-CCA GCA GAG AAT GGA AAG TC-3\'; β~2~-microglobulin antisense primer 5\'-GAT GCT GCT TAC ATG TCT CG-3\'.). The amplification profile consisted of an initial denaturation of 2 min at 95°C and 20--35 cycles of 30 sec at 95°C, annealing for 30 sec at 55°C (β~2~-microglobulin) or at 60°C (CB~1~, CB~2~and FAAH) and elongation for 1 min at 72°C. A final extension of 10 min was carried out at 72°C. The expected sizes of the amplicons were 338 bp for CB~1~; 329 bp for CB~2~; 202 for FAAH and 269 bp for β~2~-microglobulin. The β~2~-microglobulin house-keeping gene expression was used to evaluate variations in the quality and content of the mRNA and to monitor cDNA synthesis in the different preparations. Furthermore the PCR primers for β~2~-microglobulin and FAAH were selected on the basis of the sequence of the β~2~-microglobulin gene (NCBI accession number M17987) by including the intron 402--1017, and of the sequence of the FAAH human gene (NCBI accession number AF098012) by including the intron 497--722. In the presence of contaminant genomic DNA, the expected size of the amplicon would be 426 and 885 bp for FAAH and β~2~-microglobulin, respectively. 10--20 μl of PCR products were electrophoresed on 2% agarose gel (MS agarose, Roche) in TAE buffer at 4 V/cm for 4 h. Ethidium bromide (0.1 μg/ml) was included both in the gel and electrophoresis buffer and PCR products were detected by UV visualization and recorded by photo (Polapan 55, Polaroid). Evaluation of the relative expression levels was performed by analysing the amount of amplicon synthesized at different numbers of amplification cycles for a fixed quantity of cDNA in the assay. In order to obtain a quantitative evaluation of the relative amounts of the mRNA transcripts, the different PCR reactions were performed with a different number of cycles. Differences in band intensities obtained using a non-saturating number of cycles, in relation to the intensity of the housekeeping mRNA transcript (i.e. β~2~-microglobulin, which works as an \"internal standard\"), are highly indicative of differences in the levels of expression of the corresponding genes. Statistical analyses -------------------- Results of the anandamide level measurements, expressed in pmol/ml of blood and as means ± SEM, were compared by the unpaired Student\'s t test. When data from patients before and after pharmacological treatment were compared, the paired Student\'s t test was used. Authors\' contributions ======================= NDM, a senior psychiatrist, conceived this study, selected the patients and performed all clinical ratings, and was assisted by FD; VDM and LDP conceived and coordinated the study and took part in all biological assays; PO performed all molecular biology experiments; FF extracted the lipids from blood sample and measured the levels of anandamide by LC-MS. All authors have read and approved the manuscript. ###### Brief Psychiatric Rating (BPRS) Scale. Score for all the 18 items of the Brief Psychiatric Rating (BPRS) for each of the twelve patients with schizophrenia included in this study **N° Patient** **1** **2** **3** **4** **5** **6** **7** **8** **9** **10** **11** **12** **13** **14** **15** **16** **17** **18** ---------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- -------- -------- -------- -------- -------- -------- -------- -------- -------- 1 4 3 4 2 2 6 1 1 4 3 5 5 3 3 6 4 2 1 2 3 5 6 3 3 6 1 1 4 2 6 6 3 3 5 3 2 2 3 acute 5 6 4 3 3 7 2 1 3 1 7 5 4 2 6 2 1 6 3 remission 2 2 2 1 2 1 1 1 2 1 1 1 2 1 1 2 1 1 4 2 4 3 2 2 5 2 1 2 2 6 4 2 2 5 2 1 1 5 acute 2 6 2 2 1 7 1 1 2 2 6 6 2 2 7 2 1 2 5 remission 1 2 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 6 acute 3 5 5 2 2 5 4 1 3 1 5 6 6 4 6 5 1 1 6 remission 2 2 3 1 1 2 2 1 2 1 2 1 2 2 1 2 1 1 7 acute 2 3 5 2 4 5 1 1 4 1 3 4 4 3 6 6 1 1 7 remission 1 2 2 1 2 1 1 1 3 1 1 1 2 1 2 2 1 1 8 acute 4 5 5 3 3 6 2 1 2 6 6 7 4 4 7 3 1 1 8 remission 3 3 1 1 3 2 1 1 3 1 1 1 2 1 1 2 1 1 9 2 4 3 2 3 5 2 1 3 3 5 5 3 3 5 3 2 1 10 2 4 4 5 2 5 1 1 3 3 5 5 3 3 5 4 1 1 11 4 4 4 4 1 5 2 1 3 3 4 5 2 3 5 4 1 1 12 1 4 4 6 1 5 1 3 1 2 4 5 1 2 6 2 2 2 Acknowledgements ================ We thank Dr Alfredo Dama, MD, for useful advice and support, S. Guardascione and G. Capuano for assistance, and S. Piantedosi for the artwork. This study was supported by grant POP 98/5.4.2 from Regione Campania to LDP.

Introduction {#S0001} ============ Non-coding RNAs (ncRNAs) are emerging classes of non-protein coding RNA transcripts with pivotal roles in gene regulation.[@CIT0001] It has been reported that more than 98% of all human genome transcriptional output is ncRNAs and most ncRNAs are expressed in specific types of tissues and cells or under certain stress conditions or during specific developmental stages, indicating their extensive involvement in human body growth and development.[@CIT0002] Long non-coding RNAs, or lncRNAs, are a subgroup of ncRNAs longer than 200 nt.[@CIT0003] Functional characterization has revealed that lncRNAs not only participate in almost all important physiological processes, but also have critical functions in the development of diseases,[@CIT0004] such as different types of cancer.[@CIT0005],[@CIT0006] However, function of most lncRNAs is still unknown. Gastric cancer is one of the most frequently diagnosed malignancies and is also the third leading cause of cancer-related deaths worldwide.[@CIT0007] Most gastric cancer patients at early stages can be cured by surgical operations. However, more than half of patients with advanced gastric cancer will die of carcinoma recurrence even after curative gastrectomy.[@CIT0008] Therefore, more studies are needed to further characterize the molecular alterations involved in the pathogenesis of gastric cancer.[@CIT0009],[@CIT0010] Our genome-wide transcriptome identified a large number of differentially expressed lncRNAs in gastric cancer patients (data not shown). Among those differential lncRNAs, lncRNA PTCSC3 which has been reported to be downregulated in papillary thyroid carcinoma,[@CIT0011] is also downregulated in gastric cancer and is positively correlated with lncRNA Linc-pint, a characterized tumor suppressor in lymphoblastic leukemia.[@CIT0012] Our study was therefore carried out to investigate the involvement of PTCSC3 and Linc-pint in gastric cancer and explored the possible interactions between them. Materials and Methods {#S0002} ===================== Research Subjects {#S0002-S2001} ----------------- Our study included 78 patients with gastric cancer were diagnosed and treated in Fourth Hospital of Hebei Medical University from March 2010 to March 2015. Inclusion criteria: 1) gastric cancer patients who were diagnosed by pathological biopsy; 2) patients who were diagnosed for the first time and no treatment was performed; 3) patients willing to participate. Exclusion criteria: 1) patients who received treatment before admission or who were transferred from other hospitals; 2) patients who were complicated with other severe diseases, such as other types of cancer and systemic infection. The 78 patients included 43 females and 35 males, and age ranged from 29 to 66 years, with a mean age of 48.6±6.1 years. According to AJCC staging, there were 16 cases at stage I, 22 cases at stage II, 24 cases at stage III and 16 cases at stage IV. This study had been approved by Ethic Committee of Fourth Hospital of Hebei Medical University before patient admission and all participants signed informed consent. Specimen Collections and Cell Lines {#S0002-S2002} ----------------------------------- Biopsy was performed to collect cancer and paracancerous tissues from each patient. Tissues were stored in liquid nitrogen before use. SNU-1 and Hs 746T human gastric cancer cell lines were used in this study to perform in vitro experiments. Cells of these two cell lines were purchased from American Type Culture Collection (ATCC, Manassas, VA, USA). ATCC-formulated RPMI-1640 Medium (ATCC) supplemented with 10% fetal bovine serum (FBS, ATCC) was used to cultivate the cells of both cell lines at 37°C in a 5% CO~2~ incubator. Follow-Up {#S0002-S2003} --------- All patients were followed up for 5 years after admission to record their overall conditions. Patients failed to complete follow-up and patients who died of other causes during follow-up were not included in this study. Cell Transfections {#S0002-S2004} ------------------ PcDNA3.1 vector (Invitrogen) expressing PTCSC3 or Linc-pint was designed and constructed by Sangon (Shanghai, China). Linc-pint siRNA and negative control siRNA were also designed and synthesized by Sangon. Cells of SNU-1 and Hs 746T cell lines were cultivated overnight to each 70--80% confluence. All cell transfections were performed using Lipofectamine 2000 reagent (cat no. 11668-019; Invitrogen; Thermo Fisher Scientific, Inc., Waltham, MA, USA) with all operations performed in strict accordance with manufacturer's instructions. Doses of vectors and siRNAs were 10 and 45 nM, respectively. Un-transfected cells were control cells, and cell transfected with empty pcDNA3.1 vectors or negative control siRNAs were negative control cells. Cells were harvested at 24 h after transfection for subsequent experiments. Total RNA Extraction and Reverse Transcription-Quantitative Polymerase Chain Reaction (RT-qPCR) {#S0002-S2005} ----------------------------------------------------------------------------------------------- Tissues were ground in liquid nitrogen, followed by the addition of RNAzol reagent (Sigma-Aldrich, St. Louis, MO, USA) to extract total RNA. RNAzol reagent was also directly mixed with in vitro cultivated cells to extract total RNA. Total RNAs were reversely transcribed using SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific, lnc.). To detect the expression of PTCSC3 and Linc-pint, PCR reaction systems were prepared using Applied Biosystems™ PowerUp™ SYBR™ Green Master Mix (Thermo Fisher Scientific, lnc.) and all PCR reactions were performed on CFX96 Touch Deep Well™ Real-Time PCR Detection System (Bio-Rad) with 18S RNA as the endogenous control. Primers of PTCSC3 and Linc-pint as well as 18S RNA were designed and synthesized by Sangon (Shanghai, China). Expression of TCSC3 and Linc-pint was normalized to 18S RNA using 2^−ΔΔCq^ method. In vitro Cell Proliferation Assay {#S0002-S2006} --------------------------------- Cells were harvested at 24 h after transfection and in vitro cell proliferation assay was performed using Cell Counting Kit-8 (CCK-8) kit (Sigma-Aldrich) to explore the effects of TCSC3 and Linc-pint on gastric cancer cell proliferation. Briefly, cells were mixed with ATCC-formulated RPMI-1640 Medium containing 10% FBS to make single-cell suspensions. Cell density was adjusted to 4×10^4^ cell/mL and cell suspensions were transferred to a 96-well plate with 100 μL cell suspension in each well. Cells were cultivated (37°C and 5% CO~2~) and 10 μL CCK was added into each well 24, 48, 72 and 96 hrs later. After that, cells were cultivated from additional 4 hrs and 10 μL DMSO was added. OD values at 450 nm were measured to reflect cell proliferation rates. Flow Cytometry {#S0002-S2007} -------------- Cells were harvested at 24 hrs after transfection and flow cytometry was performed to explore the effects of PTCSC3 and Linc-pint on gastric cancer cell stemness. Cells were harvested by trypsinization. IgG1-PE or CD133-PE antibody (130-093-193, Meltenyi Biotec, Germany) was used to incubate with cells (5×10^5^ cells per mL) at 4°C for 20 mins. After that, cells were harvested and dissolved in PBS. FACS Aria system (BD Immunocytometry Systems, San Jose, CA, USA) was then used to detect signals and data were processed using Cell Quest software (Becton Dickinson Ltd). Statistical Analysis {#S0002-S2008} -------------------- All experiments were performed 3 times and data were recorded as mean ± standard deviation. Comparisons of expression levels of TCSC3 and Linc-pint between cancer and paracancerous tissues were performed by paired *t* test. Correlations between expression levels of PTCSC3 and Linc-pint were analyzed by Pearson's correlation coefficient. Comparisons among multiple groups were performed by one-way ANOVA followed by Tukey's test. Based on the expression levels of PTCSC3 and Linc-pint in tumor tissues, patients were divided into high (n=36) and low (n=42) PTCSC3 level groups, as well as high (n=38) and low (n=40) Linc-pint level groups according to Youden's index. Survival curves were plotted based on Kaplan--Meier method and compared by log-rank *t* test. Differences with p\<0.05 were statistically significant. Results {#S0003} ======= PTCSC3 and Linc-pint Were Both Downregulated in Cancer Tissues of Gastric Cancer Patients {#S0003-S2001} ----------------------------------------------------------------------------------------- RT-qPCR was performed to detect the expression of PTCSC3 and Linc-pint in both cancer and paracancerous tissues of 78 gastric cancer patients. Compared with cancer tissues, expression levels of PTCSC3 were significantly lower in paracancerous tissues ([Figure 1A](#F0001){ref-type="fig"}, p\<0.05). In addition, Linc-pint was also downregulated in cancer tissues than in paracancerous tissues ([Figure 1B](#F0001){ref-type="fig"}, p\<0.05).Figure 1PTCSC3 and Linc-pint were both downregulated in cancer tissues of gastric cancer patients. RT-qPCR results showed that PTCSC3 (**A**) and Linc-pint (**B**) were both downregulated in cancer tissues of gastric cancer patients compared with paracancerous tissues (\*p\<0.05). Low Levels of PTCSC3 and Linc-pint in Cancer Tissue Indicate Poor Survival {#S0003-S2002} -------------------------------------------------------------------------- Based on the expression levels of PTCSC3 and Linc-pint in tumor tissues, patients were divided into high (n=36) and low (n=42) PTCSC3 level groups, as well as high (n=38) and low (n=40) Linc-pint level groups according to Youden's index. Survival curves were plotted based on Kaplan--Meier method and compared by log-rank *t* test. As shown in [Figure 2A](#F0002){ref-type="fig"}, patients in low-level PTCSC3 group showed significantly lower overall survival rate compared with patients in high PTCSC3 group ([Figure 2A](#F0002){ref-type="fig"}). In addition, the overall survival rate of patients in low Linc-pint level group was also significantly lower than that of patients in high Linc-pint level group ([Figure 2B](#F0002){ref-type="fig"}).Figure 2Low levels of PTCSC3 and Linc-pint in cancer tissue indicate poor survival. Survival curve analysis on 5-year follow-up data showed that low levels of PTCSC3 (**A**) and Linc-pint (**B**) in cancer tissue indicate poor survival. Expression Levels of PTCSC3 and Linc-pint Were Significantly Correlated in Cancer Tissues {#S0003-S2003} ----------------------------------------------------------------------------------------- Correlations between expression levels of PTCSC3 and Linc-pint were analyzed by Pearson's correlation coefficient. It was observed that expression levels of PTCSC3 and Linc-pint were significantly and positively correlated in cancer tissues ([Figure 3A](#F0003){ref-type="fig"}). In contrast, the correlation between expression levels of PTCSC3 and Linc-pint was not significant in paracancerous tissues ([Figure 3B](#F0003){ref-type="fig"}).Figure 3Expression levels of PTCSC3 and Linc-pint were significantly correlated in cancer tissues. Pearson's correlation coefficient analysis showed that expression levels of PTCSC3 and Linc-pint were significantly correlated in cancer tissues (**A**) but not in paracancerous tissues (**B**). PTCSC3 and Linc-pint Upregulated Each Other in Gastric Cancer Cells {#S0003-S2004} ------------------------------------------------------------------- PTCSC3 and Linc-pint were overexpressed in cells of both SNU-1 and Hs 746T cell line to further investigate the interactions between PTCSC3 and Linc-pint ([Figure 4A](#F0004){ref-type="fig"}, p\<0.05). Compared with control (C) and negative control (NC), overexpression of PTCSC3 led to significantly upregulated Linc-pint expression in cells of both cell lines ([Figure 4B](#F0004){ref-type="fig"}, p\<0.05). In addition, overexpression of Linc-pint also mediated the upregulation of PTCSC3 in those cells ([Figure 4C](#F0004){ref-type="fig"}, p\<0.05).Figure 4PTCSC3 and Linc-pint upregulated each other in gastric cancer cells. Overexpression of PTCSC3 and Linc-pint was reached in cells of both SNU-1 and Hs 746T cell line at 24 h after transfection (**A**). Overexpression of PTCSC3 led to significantly upregulated Linc-pint expression in cells of both cell lines (**B**). In addition, overexpression of Linc-pint also mediated the upregulation of PTCSC3 in those cells (C) (\*p\<0.05). PTCSC3 and Linc-pint Regulated Gastric Cancer Cell Proliferation and Stemness {#S0003-S2005} ----------------------------------------------------------------------------- Compared with control (C) and negative control (NC), overexpression of PTCSC3 and Linc-pint led to significantly inhibited proliferation ([Figure 5A](#F0005){ref-type="fig"}) and decreased percentage of CD133+ cells ([Figure 5B](#F0005){ref-type="fig"}) (p\<0.05). Linc-pint siRNA played an opposite role and attenuated the effects of PTCSC3 overexpression on cancer cell proliferation and stemness (p\<0.05).Figure 5PTCSC3 and Linc-pint regulated gastric cancer cell proliferation and stemness. Overexpression of PTCSC3 and Linc-pint led to significantly inhibited proliferation (**A**) and decreased percentage of CD133+ cells (**B**). Linc-pint siRNA played an opposite role and attenuated the effects of PTCSC3 overexpression on cancer cell proliferation and stemness (\*p\<0.05). Discussion {#S0004} ========== LncRNAs are critical determinants in cancer biology, while function of most lncRNAs remains unknown. The key finding of the present study is that PTCSC3 and Linc-pint two lncRNAs play a role in tumor suppressor in gastric cancer by forming a positive feedback regulation circle. Survival of advanced gastric cancer patients is still poor.[@CIT0013] Therefore, development of novel prognostic markers is always needed to design individualized post-surgery treatment and care strategies. Some prognostic markers, such as ImmunoScore signature,[@CIT0014] have shown potentials of clinical application. Our studies identified PTCSC3 and Linc-pint as two promising prognostic biomarkers for gastric cancer. RNA expression detection as an easy disease prediction approach may be superior to most other approaches in terms of the input of time and cost. However, more clinical studies are needed to further verify the clinical values of these two lncRNAs. LncRNAs regulate gene expression at different levels, such as methylation and posttranscriptional and translational regulation.[@CIT0015],[@CIT0016] However, interactions between different lncRNAs are largely unknown. In the present study, we proved that PTCSC3 and Linc-pint may form a positive feedback regulation circle in gastric cancer, and this positive feedback regulation is involved in the regulation of gastric cancer cell proliferation and stemness. Interestingly, no significant correlation between expression levels of PTCSC3 and Linc-pint was observed in paracancerous tissue. Therefore, certain pathological factors may be involved to mediate the feedback regulation between PTCSC3 and Linc-pint. It is also worth noting that Linc-pint silencing only partially attenuated the inhibitory effects of PTCSC3 overexpression on cancer cell proliferation and stemness. Therefore, PTCSC3 may interact with multiple factors to participate in the pathogenesis of gastric cancer. Instead of the interaction between two lncRNAs, PTCSC3 is more likely a component of regulation network. More studies are needed to further characterize this network. In this study, we only used CD133+ marker to analyze cancer cell stemness. Our future studies will try to include more stemness markers to further verify our conclusions. Conclusion {#S0005} ========== In conclusion, PTCSC3 and Linc-pint are downregulated in gastric cancer. PTCSC3 and Linc-pint may inhibit gastric cancer by forming a positive feedback regulation circle to inhibit cancer cell proliferation and reduce cell stemness. Data Sharing Statement {#S0006} ====================== The analyzed data sets generated during the study are available from the corresponding author on reasonable request. Ethics Approval and Consent to Participate {#S0007} ========================================== The present study was approved by the Ethics Committee of Fourth Hospital of Hebei Medical University. The research has been carried out in accordance with the World Medical Association Declaration of Helsinki. All patients provided written informed consent prior to their inclusion within the study. Author Contributions {#S0008} ==================== All authors contributed to data analysis, drafting and revising the article, gave final approval of the version to be published, and agree to be accountable for all aspects of the work. Disclosure {#S0009} ========== The authors report no conflicts of interest in this work.

ADA Events {#s1} ========== **59th Annual Advanced Postgraduate Course** **17--19 February 2012** **Hyatt Regency San Francisco** **San Francisco, California** **Contact:** Shirley Ash **E-mail:** <sash@diabetes.org> **2012 Diabetes Professional Educators Conference** **9--10 March 2012** **Country Springs Conference Center** **Waukesha, Wisconsin** **Contact:** Penny Kasprzak **E-mail:** <pkasprzak@diabetes.org> **27th Clinical Conference on Diabetes** **24--27 May 2012** **Hyatt Regency Coconut Point** **Bonita Springs, Florida** **Contact:** Pauline Lowe **E-mail:** <plowe@diabetes.org> **72nd Scientific Sessions** **8--12 June 2012** **Pennsylvania Convention Center** **Philadelphia, Pennsylvania** **Web site:** <http://scientificsessions.diabetes.org> **73rd Scientific Sessions** **21--25 June 2013** **McCormick Place Convention Center** **Chicago, Illinois** **Web site:** <http://scientificsessions.diabetes.org> **Contact for information on ADA events:** American Diabetes Association, 1701 N. Beauregard St., Alexandria, VA 22311. **Tel:** 800-232-3472, select option 1. **Fax:** 703-549-1715 **or** 703-253-4358. **E-mail:** <professionaleducation@diabetes.org>. **Web site:** <http://professional.diabetes.org/ce>. Other Events {#s2} ============ **Circadian Clocks and Metabolic Disease** **2012 UCLA Symposium** **20--22 April 2012** **Lake Arrowhead Conference Center** **Lake Arrowhead, California** **Web site:** <http://www.ccmdsymposium.med.ucla.edu>

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 ============ Bombay phenotype was first discovered by Dr. Bhende in Bombay, India and named accordingly \[[@REF1]\]. The H antigen is located on red blood cells and is the precursor compound of A and B antigens. The A and B allele produce different transferase enzymes which add complex carbohydrates to the H antigen, transforming it into A antigen and B antigen, respectively. In individuals with blood group O there is no functional transferase enzyme to modify the H antigen, therefore it remains unchanged. The h allele is a result of the mutation of the H gene that expresses H antigen in the red blood cells. Individuals with Bombay phenotype inherit the homozygous recessive (hh) genotype instead of the homozygous dominant (HH) or heterozygous (Hh) genotypes of the ABO blood group. As the A and B antigens cannot be formed without the H antigen precursor, their red blood cells also lack these antigens. Consequently, these individuals produce anti-H, anti-A, and anti-B antibodies. Serum from these individuals contain antibodies that react with red blood cells from all O, A, B, and AB blood groups. It is of concern to note that people with Bombay phenotype can only be transfused with red blood cells that lack the H, A, and B antigens, leaving room for either autologous blood or blood from another Bombay blood group \[[@REF2]\]. Here, we report a case where a patient with Bombay blood group presented with upper gastrointestinal bleeding and severe anemia. Once his blood group was determined, donors were reached and he received two successful blood transfusions of Bombay blood group. After stabilizing the patient, he was discharged and follow-up investigations including serum von Willebrand factor antigen, von Willebrand factor functional activity and factor VIII levels were ordered. He was consequently labelled as type 3 von Willebrand disease. This case report emphasizes the significance of both forward (cell) typing and reverse (serum) typing during ABO blood grouping in blood banks. Case presentation ================= A 19-year-old male presented in the emergency department with one episode of melena per day, for one week. It was associated with vomiting, shortness of breath and palpitations. His hemoglobin level on initial complete blood count was 5.80 g/dL, signifying severe anemia according to WHO guidelines \[[@REF3]\]. His lab parameters on admission are presented in Table [1](#TAB1){ref-type="table"}. ###### Laboratory investigations on admission. RBC: Red blood cell; WBC: White blood cell; INR: International normalised ratio; APTT: Activated partial thromboplastin time; SGOT (AST): Serum glutamic oxaloacetic transaminase (Aspartate aminotransferase); SGPT (ALT): Serum glutamic pyruvic transaminase (Alanine aminotransferase). ---------------------------- ------------- ------------------------ Test Result Normal Reference Range Hemoglobin  5.80 g/dL  13.5-18.0 g/dL  RBC total 2.02 m/μL 4.5-6.5 m/μL WBC total 13900/μL 4000-10500/μL Platelet count 474000/μL 150000-400000/μL Prothrombin time 10.40 sec 9.5-11.7 sec INR 1.0 0.8-1.3 APTT 48.60 sec 24.8-36.2 sec Fibrinogen level 171.1 mg/dL 199-463 mg/dL Factor VIII 18.5% 75-216% SGOT (AST) 27 U/L 5-34 U/L SGPT (ALT) 24 U/L 0-55 U/L Alkaline phosphatase 81 U/L 40-150 U/L Total bilirubin 0.22 mg/dL 0.2-12 mg/dL Direct bilirubin 0.102 mg/dL 0.0-0.5 mg/dL Gamma-glutamyl transferase 22 U/L 12-64 U/L C-reactive protein 0.78 mg/L Up to 5.0 mg/L Serum sodium 141 mEq/L 136-145 mEq/L Serum potassium 3.4 mEq/L 3.5-5.1 mEq/L Serum chloride 105 mEq/L 98-107 mEq/L Serum bicarbonate 24 mEq/L 22-29 mEq/L Serum glucose(random) 133 mg/dL \<200 mg/dL Blood urea nitrogen 4.0 mg/dL 8.9-20.6 mg/dL Serum creatinine 0.8 mg/dL 0.72-1.25 mg/dL ---------------------------- ------------- ------------------------ Immediately packed red blood cells (RBCs) were requested from the blood bank. On forward typing his blood group was labeled as O positive and his serum showed strongly positive indirect Coomb's test with a negative direct Coomb's. On extended 11 cell panel antibody testing, his serum demonstrated pan-agglutination which matched with monoclonal panel cells having anti-Kell, anti-Lub, and anti-Kpb antibodies. On cross match with four O negative and four O positive packed RBCs, +4 incompatibility was seen with all. Meanwhile a detailed history of the patient revealed two distinct episodes of epistaxis in childhood and a family history of his paternal grandmother having an increased bleeding tendency. In view of his past history of fresh frozen plasma infusions, it was interpreted that the patient may have multiple alloantibodies in blood leading to gross incompatibility. Considering the urgency of the situation, one unit of the least incompatible (O negative) packed RBCs was issued after washing with normal saline thrice, to the emergency department. Transfusion was started under strict monitoring by the emergency department physicians. After slow transfusion of around 10 ml blood, the patient started shivering and his temperature spiked to 101°F with tachycardia and hypotension. The transfusion was stopped immediately and the patient was given intravenous antihistamine and hydrocortisone. Meanwhile, he was transferred to the intensive care unit (ICU) where he received intranasal desmopressin and intravenous factor VIII. Transfusion reaction workup revealed a grade 4+ pan agglutination in his serum. During repeat blood grouping, forward typing did not demonstrate any reaction to anti-A and anti-B antisera, like a normal O blood group. However, on reverse typing, his serum showed strong agglutination with group O pooled control cells. His post saline wash incompatibility with O negative red cell concentrate showed minor difference from grade +4 agglutination (pre-wash) to grade +3 clumping (post-wash). A fresh RBCs sample from the patient showed negative direct Coomb's test, while fresh serum sample remained positive for indirect Coomb's test. This workup strongly raised the suspicion of Bombay phenotype and his red cells were tested with anti-H lectin, which showed no agglutination. This confirmed his blood group as Bombay phenotype. The reactions observed with Bombay phenotype compared to other blood groups, on forward and reverse typing, are illustrated in Table [2](#TAB2){ref-type="table"}. ###### Forward and reverse grouping with different blood groups. (+) = Agglutination, (-) = No reaction ------------------- ------------------- ------------------- --------------------- ---------- --------- --------- Blood group Anti-A antibodies Anti-B antibodies Anti-A,B antibodies A1 cells B cells O cells A  +  -  +  -  +  -  B  -  +  +  +  -  -  AB  +  +  +  -  -  -  O  -  -  -  +  +  -  Bombay phenotype  -  -  -  +  +  +  ------------------- ------------------- ------------------- --------------------- ---------- --------- --------- Immediately, voluntary donor pools were contacted in blood banks throughout the country. Overnight, a donor with Bombay negative blood group was arranged from Karachi. The packed RBCs were airlifted to Islamabad maintaining the cold chain. After crossmatching with recipient's blood showed no reaction, the donor blood was transfused to the patient. Meanwhile, a distant relative of the patient from a nearby city, with Bombay positive blood group, consented to donate blood at our blood bank. Two days later, another unit of packed RBCs was transfused to the patient. His hemoglobin after two transfusions rose up to 7.40 g/dL. As his melena settled down on supportive therapy, an endoscopy was performed that suggested an underlying hiatal hernia. After surgical consultation, the patient was advised to reduce weight and discharged from the hospital, with a scheduled follow-up visit. In view of the patient's past medical history and family history, during the follow-up visit, a von Willebrand factor antigen, von Willebrand factor functional activity and factor VIII levels were ordered. His von Willebrand factor antigen level was \<2.0%, von Willebrand factor functional activity was \<4.0% and factor VIII level was 18.5%, consistent with type 3 von Willebrand disease. The patient and his family were counselled accordingly and referred to the hematology clinic. Discussion ========== The prevalence of Bombay phenotype is 1:10,000 in India however it is much less prevalent in the Caucasians with an incidence of 1:250,000 \[[@REF4]\]. A study amongst an urban population in Puducherry, India demonstrated that 0.008% population had Bombay blood group and it was strongly associated with consanguineous marriage (66.66%) \[[@REF5]\]. This prompts for pre-marital blood phenotyping for all individuals and discouragement of marriage amongst individuals carrying the h allele. Another descriptive study spanning over a period of six years (2012--2017) comprising 36,964 blood donors at a blood bank in Bangalore, showed a prevalence of 0.005% of Bombay phenotype \[[@REF6]\]. There has not been enough documentation of reported cases of Bombay phenotype in Pakistan. Some sporadic cases have been reported in the past, but there is a need to maintain a separate record of Bombay phenotype individuals in all leading blood banks across the country. Since this blood group is passed to offspring, one way to trace Bombay individuals is to perform thorough screening of family members and relatives of a known case. Once labelled as having Bombay blood group, these individuals should be motivated to become voluntary donors and register themselves with reference regional donation centers. Blood from Bombay blood group should only be reserved for patients with the Bombay blood group phenotype as it is extremely rare. In the event of a surgery, where blood loss is suspected and no Bombay phenotype blood is available, acute normovolemic hemodilution may be employed. This involves removal of blood from a patient after induction of anesthesia while maintaining normovolemia with crystalloids or colloids. The blood collected is stored at room temperature in the operating room and is infused back to the patient within eight hours of collection, allowing platelets and coagulation factors to remain functional \[[@REF7]\]. Other options may include cryopreservation of blood donated by Bombay individuals. As individuals with Bombay phenotype are often misdiagnosed as O blood group on forward typing and reverse typing is not routinely performed in some blood banks, these individuals may be transfused with O blood group in an emergency situation. This may lead to an acute hemolytic transfusion reaction. This can be avoided by providing these patients with wrist bands or a rare blood group card, identifying their blood group. These individuals also require comprehensive counselling, emphasizing the rarity of their blood group and the need to develop a careful behaviour to avoid episodes of bleeding in the future. The importance of counselling is paramount in females where the need for blood transfusion may arise in the setting of child birth. Von Willebrand disease (VWD) is caused by mutations that lead to an impairment in the synthesis or function of von Willebrand factor (VWF). The disease can also be acquired due to various pathophysiologic defects. It is classified as quantitative defect (type 1 and type 3) or qualitative defect (type 2). Type 2 is further subdivided into type 2A, 2B, 2M, and 2N \[[@REF8]\]. A study observed a significant effect of ABO blood group on plasma von Willebrand antigen (VWF:Ag) levels. It demonstrated that VWF:Ag levels in patients with Bombay blood group (median VWF:Ag = 0.69 IU/dL) were significantly lower than in groups AB, A, or B. It further established that VWF:Ag levels in Bombay blood group individuals were even lower than in group O individuals (median VWF:Ag = 0.82 IU/dL), however this difference was statistically not significant \[[@REF9]\]. Various other reports \[[@REF10], [@REF11]\] confirm that plasma VWF:Ag concentration is significantly influenced by the ABO blood group. In this case report too, laboratory investigations on follow-up visit demonstrated a markedly low VWF antigen level in the patient. Therefore, in dealing with a patient of Bombay blood group presenting with a history of bleeding diathesis, the suspicion of von Willebrand disease should be kept high. Conclusions =========== Bombay phenotype is one of the rarest ABO blood groups which often leads to a delayed diagnosis. It is important to perform both forward and reverse blood grouping routinely in all first time O blood group labelled patients. Once identified, these individuals should carry a form of identification of their blood group at all times. Meanwhile, their family members should be screened for Bombay phenotype and affected individuals should be counseled to become voluntary donors. Blood banks also need to create a rare group registry for prompt response in the wake of any emergency situation. The authors have declared that no competing interests exist. Consent was obtained by all participants in this study

Introduction {#sec1-1} ============ Leukoencephalopathy, intracranial calcifications, and cysts (LCC) is a very rare cerebral disorder, first described in 3 children in 1996.\[[@CIT1]\] It has been later reported from around the world in children and adults, with onset up to 59 years.\[[@CIT2]\] The clinical presentation is insidious and variable. Typically, initial symptoms are of raised intracranial pressure, later followed by focal neurologic deficits. All the reported patients have a characteristic triad of calcification in the deep cerebral nuclei and white matter, diffuse leukoencephalopathy, and multiple cystic brain lesions on brain imaging. The histopathologic findings described include a peculiar angiopathy and abundant Rosenthal fibers.\[[@CIT3]\] All these features justify the designation of LCC as a distinct, although extremely rare, nosologic entity. We report an adult case with clinical, radiologic, and pathologic features consistent with LCC from a tertiary care hospital in South India. Case Report {#sec1-2} =========== A 50-year-old man presented in June 2009 with headache of 4 months duration, progressive unsteadiness of gait since 1 month and recurrent vomiting since 1 week. Headache was holocranial, throbbing, almost continuous, and tended to disturb sleep. Cough worsened the headache. He never had diplopia or visual obscurations. He felt unsteady while walking, but gave no history of falls. He did not have tremors or difficulty in using the upper limbs. He had a past history of seizures during childhood, associated with fever, which had not recurred so far. He worked as a vendor for herbal medications, and was continuing to do so until a month prior to the onset of symptoms. Clinical examination showed early papilledema, normal eye movements, bilateral finger--nose and heel--knee incoordination as well as dysdiadochokinesia. He had a spastic-ataxic gait. Complete blood count, sedimentation rate, liver and renal function tests, serum calcium, phosphate, alkaline phosphatase levels, chest radiograph, and abdominal ultrasonogram were within normal limits. Serological tests for HIV 1 and 2 were negative. Magnetic resonance imaging (MRI) of the brain showed multiple bilateral cerebral cystic lesions and a large right cerebellar cyst with mass effect, compressing the 4th ventricle and brainstem. Diffuse T2 hyperintense lesions were seen bilaterally in the subcortical and cerebellar white matter. The cystic lesions enhanced with contrast administration. No hemorrhage was noted \[Figures [1a](#F0001){ref-type="fig"}--[c](#F0001){ref-type="fig"}\]. {#F0001} Lumbar cerebrospinal fluid (CSF) was slightly xanthochromic and showed 26 leukocytes (45% polymorphs, 55% lymphocytes) with protein of 106 mg% and glucose of 149 mg%. CSF cryptococcal antigen was negative. The differential diagnosis considered included cystic metastases, infective cysts, and tumefactive demyelination. We decided to proceed with posterior fossa craniectomy and excision of the right cerebellar cyst. Peroperatively, the cyst contained brownish fluid and showed visible calcification in the wall. A computed tomography of brain was done postoperatively and confirmed the presence of multiple calcifications in the white matter and basal ganglia \[[Figure 1d](#F0001){ref-type="fig"}\]. Histopathology of the excised cyst wall showed intensive gliosis with Rosenthal fibers, prominent angiomatous changes, microcalcifications, and microhemorrhages. Angiomatous changes consisted of numerous small blood filled vessels with many showing hyalinized walls. Gliosis was pilocytic and showed plenty of Rosenthal fibers. Microcalcifications were extensive and there was also concentric fine calcification around blood vessel walls. Foci of microhemorrhages and hemosiderin pigment deposits were also noted in relation to the abnormal blood vessels \[Figures [2a](#F0002){ref-type="fig"}--[d](#F0002){ref-type="fig"}\]. {#F0002} Postoperatively the patient reported improvement in headache, while ataxia persisted. He was discharged on the 8th postoperative day and is being followed-up. Discussion {#sec1-3} ========== We report an adult patient diagnosed with LCC from South India. Our patient has all the characteristic clinical, radiologic, and pathologic features of LCC described in the previous reports. Our patient presented with raised intracranial pressure and cyst-related mass effects, which are the 2 main presenting features of LCC. Large posterior fossa cysts causing pressure effects requiring surgical decompression are frequently noted and may be considered as a characteristic feature.\[[@CIT4]\] Some of the patients tend to be relatively well preserved, with no major cognitive or motor deficits, until they develop symptoms related to the mass effect of large cysts.\[[@CIT5]\] Repeated surgical procedures may be required to relieve symptoms caused by the enlarging cysts. Some patients diagnosed with LCC also have Coat's retinopathy.\[[@CIT3]\] It has been proposed that Coat's retinopathy is a part of the spectrum of this disorder--the combined presentation has been termed as "cerebroretinal microangiopathy with calcification and cysts" or Coat's plus.\[[@CIT6]\] However, our patient did not have any evidence of retinal involvement. Raised CSF protein has been described in a patient diagnosed with LCC and large posterior fossa cysts.\[[@CIT1]\] Our patient also had raised CSF protein, but also had pleocytosis, which has not been previously described, and is of unclear significance--the possibility of an associated sterile meningeal inflammation may be considered. Histopathology of LCC is characterized by angiopathy, calcification, and Rosenthal fibers, and the tissue from our patient showed all these cardinal features. The primary abnormality is probably the obliterative cerebral angiopathy, which involves small vessels. Dystrophic calcifications (via slow necrosis) and formation of cysts and secondary white matter abnormalities may be secondary changes.\[[@CIT1]\] It has also been suggested based on MRI characteristics that the white mater hyperintensities on T2-weighted images represent vasogenic edema affecting the parenchyma.\[[@CIT3]\] In the first reported adult case, genetic analyses did not identify any significant mutations in 2 candidate genes, glial fibrillary acidic protein (GFAP) and cerebral cavernous malformation (CCM), although a mutation of unknown importance was identified in the gene for GFAP.\[[@CIT7]\] The cause of this mysterious disorder remains unknown, but is probably genetic. Further genetic testing may provide more insights. The other case reports are from France,\[[@CIT1]\] Switzerland,\[[@CIT2]\] Brazil,\[[@CIT3]\] Turkey,\[[@CIT4]\] the United States,\[[@CIT5]\] the United Kingdom,\[[@CIT6]\] and Finland.\[[@CIT8]\] We add India to this list, raising the possibility that although rare, LCC has a global distribution. We are indebted to Dr. KG Alexander, Chairman and Chief Physician, Baby Memorial Hospital, for his valuable help and support. **Source of Support:** Nil **Conflict of Interest:** Nil

General topics of cancer vaccination ==================================== The conference started with an introductory lecture by Chris Schmidt, addressing key aspects why cancer vaccines do not work as expected. Initially, cancer vaccines were tested in the systemic disease setting and after obtaining positive results in this patient population, moved to the minimal residual disease (MRD) setting because of the hypothesis that if a vaccine worked in a macroscopic disease setting, it should work better in the MRD setting (less immunosuppression). However, in the MRD setting adjuvant vaccines have often failed, which may be due to too short follow-up periods. On the other hand, it could also be postulated that the systemic disease setting just indicates the availability of large amounts of antigen, where the vaccine can trigger an anti-vaccine T cell response that attack the tumor and in this way activate a second wave of anti-tumor CTL. This relates to the vaccine targets: a meta-analysis of all immunotherapy trials indicated that the rate of objective clinical responses is higher when undefined antigens are used as compared to defined antigens \[[@B1],[@B2]\]. This argues that many targets need to be attacked but that till now: (1) we do not know which are the relevant targets; or (2) other molecules are present in tumor extracts that influence the regulatory environment in a way that defined antigens cannot; or (3) effective immune responses only act at sites of macroscopic tumor. Another reason why cancer treatments in general and vaccines in specific fail to cure patients could be related to the issue of timing of therapy. This was addressed by Brendon Coventry, who presented data indicating that the endogenous anti-tumor immune response follows a cyclical pattern (measured by CRP) which is dependent on antigen persistence and that this cycle could likely potentially correlate with numbers of effector T cells and regulatory T cells (Treg) over time. Preliminary evidence suggests that efficacy of chemo-, radio- or immunotherapy could be boosted by appropriate timing to putative \'therapeutic windows\' in the individual patient\'s CRP cycle \[[@B3]\]. Furthermore, cancer vaccines do not encounter a naïve environment, but instead need to counter tumor-induced tolerogenic mechanisms. Treg are major players in immunosuppression, but in humans there are controversies as to the exact phenotype of these cells. Increased percentages of Treg have been described in various malignancies, but it is also important to take into account that absolute lymphocyte counts and percentage of CD4^+^T cells are also abnormal in cancer patients. Therefore, it is necessary to measure absolute numbers of Treg in blood instead of percentages. To date however, no regimen is available to reproducibly deplete Treg before proceeding to cancer vaccination. This could be due to the fact that we are looking at total Treg, which may not all be functional? This urges us to focus on identifying functional Treg. To this regard, Michael Quinn presented results about TNFRII^+^Treg, which express higher levels of FoxP3 than conventional Treg, express CCR4 (migration to tumor) and CCR7 (migration to lymph nodes -interferes with priming of anti-tumor responses). Furthermore, preliminary results indicated that although total Treg levels increase during chemotherapy, TNFRII^+^Treg are selectively depleted by chemotherapy and the remaining TNFRII^+^Treg express lower levels of FoxP3, that have a reduced suppressive capacity. Feasibility and safety of DC vaccination in cancer patients =========================================================== Most clinical trials conducted to date with DC vaccines are phase I feasibility studies. Despite many alterations in the immune system of cancer patients, all presenters reported that it was feasible to generate the desired numbers of DC to complete the planned vaccination scheme in most patients. However, Allan Dietz pointed out that we need to cautiously report the characteristics of the generated DC, because the phenotype of DC in trials is consistently different than in preclinical studies with normal donors. This deficiency in DC differentiation in tumor bearing patients is independent of tumor type and maturation method and there appear to be tumor-specific conditions for optimal maturation of DC. In most of the presented studies, DC vaccination was safe, well tolerated, with minimal side effects. However, Dagmar Marx and Stefaan Van Gool reported the occurrence of some grade III/IV adverse events in a minority of patients, but this could possibly be related to disease localization in the brain (both studies in brain tumors) \[[@B4],[@B5]\]. Some participants reported the induction of auto-immune effects like vitiligo, but in general most participants agreed that some degree of auto-immunity is probably beneficial for DC vaccine efficacy. Patient selection ================= As outlined above, it was originally hypothesized that DC vaccines should perform better in the MRD setting as compared to patients with widespread disease. However, this does not always appear to be the case, which is possibly due to the altered immune system in cancer patients and the high degree of immunosuppression. Therefore, it would be ideal if we could select patients that are likely to benefit from DC vaccination. To this regard, Angus Dalgleish reported on a study where pre-vaccine sera from responders and non-responders were compared. This led to the identification of molecules that can distinguish responder patients from the non-responder group with 67% sensitivity and 100% specificity. All identified molecules are pro-inflammatory and are increased in the non-responder population; however, no details were given about the exact nature of these molecules. Bart Neyns reported that baseline CRP, LDH, and WHO performance status can also identify melanoma patients likely to benefit from DC vaccinations. Chris Schmidt showed that low S-100B predicts response to treatment in melanoma. Allan Dietz presented data showing that suppressive monocytes are increased in several malignancies and these cells mediate a global immune paralysis. Suppressive monocytes were found to be prognostic for cancer survival independent of therapy \[[@B6],[@B7]\]. Selection of patients on the basis of the CRP inflammatory cycle was also raised by Brendon Coventry as a possibility for patient selection with respect to timing of administration and targeting of therapy, in order to induce the desired immune response. Clinical effects of DC vaccines =============================== Clinical responses are of course largely affected by the setting in which vaccination occurs (measurable disease versus MRD), the type of cancer and the life expectancy of the patients. Furthermore, the evaluation of clinical efficacy can be impeded by the application of concurrent or subsequent therapeutic regimens. Moreover, most trials conducted to date are phase I clinical trials in which clinical efficacy is very difficult to assess. Nevertheless some promising results have been obtained. In general, long-term objective responses (CR or PR) are observed in a minority of patients, while a greater proportion of patients presents with disease stabilization. In particular, Bart Neyns pointed out that stabilization is often followed by disease regression and that clinical responses could be delayed even up to several months after initiation of treatment, indicating an immune-related response pattern (as described for anti-CTLA4 therapy). Therefore, immune-related response criteria (irRC) may be more relevant than RECIST/WHO criteria for the assessment of anti-tumor activity \[[@B8],[@B9]\]. Furthermore, most participants agree that long-term disease stabilization resulting in prolonged survival is also a relevant clinical outcome, which is of benefit for patients. The typical slow response pattern observed also indicates that DC vaccination is no option for patients with rapidly progressing disease and that patient selection is critical. Immune response monitoring after immunotherapy ============================================== The rationale behind DC-based immunotherapy is that injected DC will induce a tumor-specific immune response resulting in tumor shrinkage/clearance. So, ideally we should be able to identify patients that respond to therapy by analyzing the anti-tumor immune response generated by the DC vaccine. However, to date, limited studies show a correlation between immune responders and clinical responders, indicating that either we are not analyzing the right portion of the immune response or that the mode of action (MOA) of DC vaccines is not as expected. Chris Schmidt emphasized that we should monitor responses to the tumor and not only to the vaccine and they also observed that DC from complete responders surprisingly had lower IL-12p70/IL-10 ratios, which is not conform the stereotype of immunogenic DC. Viggo Van Tendeloo and Massimo Di Nicola reported that higher levels of activated NK cells correlated with clinical response, indicating that DC vaccines not only act on the T cell response but also on innate immunity \[[@B10],[@B11]\]. Massimo Di Nicola used killed autologous tumor cells to load DC and observed that killed tumor cell preparations from responding patients showed higher calreticulin and heat shock protein 90 expression compared to non-responders, indicating immunogenic tumor cell death is needed to obtain responses with tumor lysate pulsed DC \[[@B12]\]. The group of Jolanda de Vries developed a skin test where the skin-infiltrating lymphocytes (SKILS) in DTH sites are tested for antigen specificity. The presence of tumor-specific SKILS correlated with survival and can thus be used as a predictor of clinical response \[[@B13],[@B14]\]. Despite the fact that in most studies a correlation between immune response and clinical efficacy could not be established, Antoni Ribas suggested that, in future trials, immune monitoring studies need to be expanded and associated with the observed clinical responses to assess the underlying immune mechanisms and the MOA of DC vaccines. Maximizing DC biology potential =============================== Initial studies on DC vaccination predominantly used immature or cytokine matured DC. However, since then several improvements have been developed to enhance the DC\'s capacity to stimulate T cells and even circumvent immunosuppression. Bart Neyns presented data from so-called TriMix DC, i.e. DC transfected with mRNA encoding constitutively active TLR4 (caTLR4), CD40L and CD70. These TriMix DC are highly mature and secrete high levels of IL-12p70. When co-transfected with TAA encoding mRNA, TriMix DC induce potent TAA-specific CTL in advanced melanoma patients \[[@B15]\]. CD4^+^T cells are essential for the induction of potent CTL response, but data concerning Th cell induction by DC vaccines are scarce. Martin Cannon highlighted that we should try to redirect DC-activated Treg responses to anti-tumor Th17 responses since it was shown in ovarian cancer that these 2 cell types are inversely correlated and that tumor-associated polyfunctional Th17 cells are associated with improved clinical outcome \[[@B16]\]. Therefore, they treated DC with IL-15 and a p38 MAPK inhibitor and these cells are capable of inducing TNFα^+^FoxP3^-^IL-17 secreting CD4^+^T cells that show reduced PD-1 expression and tumor-specific CTL. DC treated with IL-15 and p38 MAPK inhibitor show decreased expression of B7-H1 and reduced IDO activity. Scott Pruitt presented data of local delivery of immune modulators by DC. Therefore, DC were transfected with RNA encoding the GITRL fusion protein and/or anti-CTLA4 mAb. These mediators are then produced locally by DC, thereby preventing side effects that occur with systemic administration as observed with anti-CTLA4 mAb \[[@B17]\]. Previous studies have shown that injected DC poorly traffic to lymph nodes where they should interact with T cells. Jeffrey Weber therefore designed a system to attract T cells to the injection site and induce in this way and \"artificial lymph nodal aggregate\" for T cell priming. To achieve this, DC are adenovirally transduced with CCL21/SLC and pulsed with peptides. The trial is still ongoing, but immunohistochemical analysis of the injection site shows substantial T cell infiltration. Another intriguing approach consists of redirecting the immune response from an anti-tumor response to an anti-viral response. Dagmar Marx and Volker Schirrmacher reported studies combining DC with the oncolytic Newcastle Disease Virus (NDV). NDV mediates lysis of tumor cells, either *in vitro*to pulse DC with viral oncolysates or *in vivo*to provide a supply of tumor antigens in the body of the patients. The tumor cells then present viral antigens to which a more potent immune response can be generated since there is less tolerance and the virus itself also drives DC polarization towards Th1 inducers \[[@B18]-[@B21]\]. Combination strategies ====================== Increasing evidence suggests that DC vaccines on their own are not capable to induce tumor regression in a substantial amount of patients, but should instead be used in combinatorial approaches. Many potential rationales exist for combination of DC with chemotherapy. Chemotherapy could have effects on MDSC and/or Treg, could increase the susceptibility of tumor cell apoptosis or lead to increased tumor cell immunogenicity, as shown by Angus Dalgleish. He also postulates that combination with immune response modifiers like low dose IL-2, Imiquimod or IMiDs would also be promising. Antoni Ribas showed data from a trial combining DC vaccines with the anti-CTLA4 mAb Tremelimumab, in which 4/16 patients experienced long term responses \[[@B22]\]. Jolanda de Vries presented data about the combination of DC vaccines with Daclizumab pre-treatment to deplete Treg, but although TAA-specific T cells could be generated, these T cells had impaired effector functions and the combination did not result in a significant effect on survival \[[@B23]\]. An attractive idea is to combine DC vaccines with adoptive T cell transfer, where DC vaccines could first prime tumor-specific T cells *in vivo*, which could subsequently be expanded *ex vivo*and then be given back to the patients. This type of combination was presented by Isabel Poschke and Gunnar Kvalheim, but studies are still ongoing. This type of combination is also complicated by the high costs. Another rationale is to incorporate DC vaccination in the standard of care (SOC) treatment if available. Surasak Phuphanich and Stefaan Van Gool presented results from this approach, with positive effects on patient survival \[[@B4],[@B5],[@B24]-[@B26]\]. Approved immunotherapeutic vaccines =================================== Provenge^®^or Sipuleucel-T from Dendreon received the first approval for a cell-based immunotherapy and hence is an important step in the development of similar strategies. This therapeutic vaccine was briefly discussed during the meeting. Historically, this vaccine is categorized as a DC vaccine, although it does not consist of \"pure\" DC; hence the term antigen pulsed activated peripheral blood mononuclear cells is probably more correct. The latest phase III trial demonstrated that the effect is merely noted on overall survival (4 month survival benefit), with less evidence of an antitumor effect (1/341 PR, 3% of patients with 50% PSA decrease). Two thirds of patients receiving sipuleucel-T developed antibody responses, and nearly three fourths had T cell proliferative responses. Survival was improved for patients who had an antibody response but not for those with a T cell response. Although these data are encouraging, they also highlight again that still very little is known about the real MOA of such immunotherapeutics. Furthermore, the high cost of such treatments may impact its use and further development \[[@B27],[@B28]\]. It would also be very interesting to investigate whether the effect of Provenge^®^can be enhanced by combination with approaches aiming at modulating the immunosuppressive environment. Discussion ========== The general discussion of the meeting focused on a few key questions to rapidly move DC vaccination forward. The first part of the discussion concentrated on the discrepancy between proof of principle and proof of efficacy. Most participants agree that the desired endpoint is clinical activity (OS, PFS, overall response rate) and that immune monitoring is less important. The main problem with this point of view is that efficacy data are difficult to obtain in phase I clinical trials, so a controlled trial would need to be performed. Other variables that can affect the selection of the desired endpoint is whether the patient population is homogeneous or heterogeneous, whether there is bulk tumor or minimal residual disease\... It is probably best to carefully select patients and assess safety and MOA in a phase I trial and then rapidly move to randomized controlled trials. Alternatively, vaccination could be added to the SOC and comparative effectiveness research could be performed. When focusing on the response rate, responses (CR, PR or even SD) should always be prolonged in time. Besides assessing clinical efficacy, immune monitoring studies are critical to understand the MOA of DC vaccines and also reporting of vaccine characteristics remains crucial in this regard. Another discussion point focused on the question why clinical data are disappointing. Probably this relates to our ignorance about the MOA of DC vaccines. Till now, there are no convincing data about the timing of vaccination, how frequently we need to vaccinate and for which period of time. Nevertheless, the approval by the FDA of Provenge^®^, the first cellular immunotherapeutic, has paved the way for further development of DC vaccines and will certainly boost the field further. Furthermore, we urgently need to get more knowledge about how to efficiently skew immune responses towards the desired phenotype for tumor eradication. Studies to resolve these issues are thus warranted. Furthermore, the concept that DC vaccines should be regarded as bystander therapeutics urges us to go ahead with the rational design of combinatorial approaches with chemotherapeutic regimens, other immunotherapeutic regimens aiming at breaking tolerance, immune response modifiers or targeted therapies (anti-angiogenic molecules, STAT3 inhibitors,\...). Next, the discussion moved to the issue of patient selection for inclusion in DC trials. The rationale argues for inclusion of less advanced patients in which we should then be able to follow a tumor marker to assess efficacy. However, is a patient that can mount an immune response really an end stage patient? Furthermore, it is not ethical to treat patients with DC vaccines if they can still benefit from a SOC treatment. Therefore, it was proposed to try to integrate DC vaccination in the SOC treatment if available or try to combine DC vaccination with available treatment approaches. On the other hand, patients should be carefully selected before inclusion in DC trials to obtain maximal benefit as possible, based on baseline characteristics that are known to correlate with improved outcome and on the aggressiveness of the disease progression. Finally, how can we improve the potential of DC therapy? Ideally, we should move from *ex vivo*generated DC to purified myeloid or plasmacytoid DC or even *in vivo*DC targeting, but before this can be pursued we need to better understand the immunoregulatory network. The use of multiple defined antigens or the whole tumor antigenic spectrum is encouraged to avoid escape and to induce a broad response, even if this implies the risk to induce some auto-immune effects, which seems to be beneficial for DC vaccine efficacy. Furthermore, studies should focus on the induction of broad polyfunctional immune responses encompassing both innate and adaptive immunity. In conclusion, this meeting brought together experts in different aspects of DC vaccine development and provided a platform for exchange of ideas, interesting new findings, encountered barriers and potential innovative new approaches. Hopefully this cross-fertilization between scientists will result in the translation to successful clinical trials that can move forward the DC vaccination approach. List of abbreviations ===================== CR: complete response; CRP: C-Reactive Protein; CTL: cytotoxic T lymphocyte; DC: dendritic cell; DTH: delayed type hypersensitivity; IMiD: immunomodulatory drug; irRC: immune-related response criteria; LDH: lactate dehydrogenase; MDSC: myeloid-derived suppressor cells; MOA: mode of action; MRD: minimal residual disease; NDV: Newcastle disease virus; NK: natural killer cell; OS: overall survival; PFS: progression-free survival; PR: partial response; RECIST: response evaluation criteria in solid tumors; SD: stable disease; SKILs: skin-infiltrating lymphocytes; SOC: standard of care; TAA: tumor-associated antigen; Th cell: T helper cell; TLR: toll-like receptor; Treg: regulatory T cell; WHO: World Health Organization. Competing interests =================== The author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties. No writing assistance was utilized in the production of this manuscript. Appendix: Meeting participants ============================== Zwi Berneman, University Hospital, Antwerp, Belgium Martin Cannon, University of Arkansas for Medical Sciences, Little Rock, USA Raja Choudhury, Karolinska Institute, Stockholm, Sweden Brendon Coventry, University of Adelaide, Adelaide, Australia Angus Dalgleish, St George\'s University of London, London, UK Jolanda de Vries, Nijmegen Centre for Molecular Life Sciences, Nijmegen, The Netherlands Allan Dietz, Mayo Clinic, Rochester, USA Massimo Di Nicola, Fondazione IRCCS Istituto Nazionale Tumori, Milan, Italy Steve Emery, German Cancer Research Center, Tumor Immunology Program, Heidelberg, Germany Rolf Kiessling, Karolinska Institute, Stockholm, Sweden Gunnar Kvalheim, Oslo University Hospital, Oslo, Norway Dirk Lorenzen, Institute for Tumor Immunology, Duderstadt, Germany Dagmar Marx, Institute for Tumor Immunology, Duderstadt, Germany Bart Neyns, Brussels University Hospital, Brussels, Belgium Surasak Phuphanich, Cedars-Sinai Medical Center, Los Angeles, USA Isabel Poschke, Karolinska Institute, Stockholm, Sweden Scott K. Pruitt, Duke University Medical Center, Durham, USA Michael A. Quinn, University of Melbourne, Melbourne, Australia Antoni Ribas, University of California at Los Angeles, Los Angeles, USA Volker Schirrmacher, German Cancer Research Center, Tumor Immunology Program, Heidelberg, Germany Chris Schmidt, Queensland Institute of Medical Research, Brisbane, Australia Inge Marie Svane, Department of Oncology, Copenhagen University Hospital, Denmark Kris Thielemans, Brussels University Hospital, Brussels, Belgium Stefaan Van Gool, Pediatric Neuro-oncology, University Hospital Leuven, Catholic University of Leuven, Belgium Viggo Van Tendeloo, University Hospital, Antwerp, Belgium Jeffrey S. Weber, H. Lee Moffitt Cancer Center, Tampa, USA Acknowledgements ================ The author would like to express her sincere gratitude to the speakers at the Dendritic Cell Therapy for Oncology Roundtable Conference for their open-minded participation to the conference and valuable feedback during the preparation of the report.

Background ========== Anterior cruciate ligament (ACL) injury leads to knee joint instability \[[@B1]\] and failure to return to sports activities at the same level \[[@B2]\]. After ACL injury, rehabilitation can last for six months or longer before an athlete can return to sports activity \[[@B3]\]. Osteoarthritic changes can be found even when ACL reconstruction is performed \[[@B4]\]. Therefore in recent years, prevention of ACL injuries has become a key issue. Most ACL injury prevention training programs are composed of plyometrics, balance training, agility training, and instructions to avoid the characteristic stance that is slight knee flexion and forceful valgus rotation associated with ACL injury \[[@B5]-[@B10]\]. Although the subjects and details of the training programs are different, the results show a decrease in the incidence of ACL injury \[[@B5]-[@B9]\]. Some research studies have investigated the effects of training programs on knee kinematics during landing tasks; however the results were different among the studies \[[@B11]-[@B14]\]. An increase in knee flexion has been reported as a result of conducting plyometric training \[[@B11]\], plyometric or balance training \[[@B12]\], and videotape feedback \[[@B13]\]. Only Myer et al. \[[@B12]\] reported changes in the kinematics of the knee in the frontal plane, i.e., both plyometric and balance training decreased the knee abduction angle during a medial drop landing. In other studies \[[@B11],[@B14]\], no differences in knee valgus were found after plyometric or agility training. Among all these studies, knee kinematics have only been analyzed in the sagittal and frontal plane. Even though tibial rotation is usually observed at the time of ACL injury, the effects of training programs for knee kinematics in the horizontal plane has not yet been analyzed. Other research studies investigated the effects of training programs on electromyography during athletic tasks; however activation of mainly the hip-musculature has been reported \[[@B11],[@B15],[@B16]\]. DeMorat et al. \[[@B17]\] ascertained that aggressive quadriceps loading with the knee in slight flexion produces significant anterior tibial translation and internal tibial rotation and leads to ACL injury. Chappell et al. \[[@B18]\] reported that females tend to have greater quadriceps activation before landing than males. Pre-activation is thought to be important for the dynamic stability of the knee. Thus, it is necessary to evaluate the effects of ACL injury prevention training programs on the electromyography of the quadriceps and hamstrings of female athletes including pre-activation. Thus, the purpose of this study was to determine the effects of a jump and balance training program on knee kinematics including tibial rotation as well as on electromyography of the quadriceps and hamstrings in female athletes. Since the training program aims at reducing ACL injury, subjects were instructed to avoid landing with their body in the characteristic position of ACL injury, which includes slight knee flexion and valgus rotation with either internal or external tibial rotation \[[@B19]-[@B21]\]. Furthermore, for the training program to be successful, hamstrings activation should increase to counteract against aggressive quadriceps loading. Therefore, our hypothesis was that training will increase knee flexion angle, decrease knee valgus, decrease internal tibial rotation, and increase hamstrings activation before ground contact during a single limb landing. Methods ======= Population ---------- Eight female basketball athletes belonging to Waseda University basketball team participated in this study. Exclusion criteria included any lower extremity reconstructive surgery in the past two years or unresolved musculoskeletal disorders that prohibited subjects from participating in sports. The average age of the subjects was 19.4 (0.7) yrs (Mean (SD)), the average height was 1.70 (0.05) m, the average weight was 64.1 (7.8) kg and the average BMI was 22.2 (1.6). Prior to participation, each subject signed a Human Subjects Informed Consent Document approved by Waseda University ethics committee. Experimental Task ----------------- All subjects performed a single limb drop landing from a 30 cm platform as described in a previous study \[[@B22]\]. The subjects were instructed to put their hands on their lower torso, stand on their right foot, and jump 30 cm away from the platform. The subjects were to land on their right foot in a neutral position. Upon landing, each subject was instructed to place their center of mass as far forward as possible in an attempt to limit horizontal motion and land without jumping up. Throughout the experiment, the subjects were barefooted to exclude the influence of differences in their shoes. Subjects were allowed several preparation trials until they could conduct the experimental task precisely. Measurement was continued until three successful trials were accomplished consecutively. We defined a failed trial when the subjects (1) could not maintain a single limb landing position, (2) landed farther than or within 30 cm of the platform, or (3) jumped up from the platform. Subjects took a sufficient rest period between the trials to avoid the effects of fatigue. The experimental task was performed three times: the initial test (Pre-training 1), five weeks later (Pre-training 2), and one week after completing training (Post-training). The training program began two weeks after the second experimental task and lasted for five weeks. Data Collection --------------- All experiments took place at the National Rehabilitation Center for Persons with Disabilities in Saitama, Japan. A seven camera VICON 370 motion analysis system (Oxford Metrics Ink., Oxford, UK) was used to record the 3-D movements of the lower limb. The laboratory was equipped with 2 force plates (9287A; Kistler Japan Co., Ltd., Tokyo, Japan). The motion and force data were recorded at 200 Hz and 1000 Hz, respectively. For each subject, 24 markers of 9 mm diameter were secured to the lower limb using double-sided adhesive tape as described in a previous study \[[@B22]\]. The markers were used to implement the Point Cluster Technique (PCT) \[[@B23]\]. The PCT provides a minimally invasive estimation of the *in vivo*motion of the knee. By using a cluster system of skin markers on a limb segment, the PCT assumes to cancel out the noise resulted from skin deformation. We developed our algorithm of the PCT following the procedure described by Andriacchi et al. \[[@B23]\]. The knee kinematics, including knee flexion/extension, valgus/varus, and internal/external tibial rotation as well as the anterior/posterior tibial translation, were calculated using the joint coordinate system proposed by Grood and Suntay \[[@B24]\]. Simultaneous electromyographic activity of the rectus femoris (RF), biceps femoris (BF) and semimembranosus (SM) were measured. Pre-amplified surface Ag/AgCl electrodes sensors (DelSys, Inc., Boston, USA) were used to detect muscular activity. A single-ended amplifier (gain = 1000) was used, while the common mode rejection rate was -92 dB. Double-sided adhesive strips were used to adhere the electrodes to the subject\'s skin. Additionally surgical tape (NICHIBAN Co., Ltd. Tokyo, Japan) was placed over the electrodes as well as around the thigh and shank to retard movement of the electrodes on the skin that would cause movement artifacts. The surface electromyography (EMG) electrodes were placed at the midpoint between the top of the patella and the anterior superior iliac spine over the muscle belly of the RF as well as at a point located distally at one-third of the distance between the knee-joint space and the ischial tuberosity over the muscle bellies of the BF and the SM. A reference electrode was placed on the head of the fibula. The EMG data was recorded at 1000 Hz. EMG data were recorded 1) while the subjects performed a maximum voluntary knee flexion and extension at 60 degrees of knee flexion against a resistance for three seconds, 2) during a static trial, i.e., while the subjects remained in a standing position for 1 second, and 3) while the subjects performed the single limb drop landing as previously described. Training Program ---------------- Training lasted approximately 20 minutes a day, 3 days a week for 5 weeks. The program was developed based on a thorough review of the literature \[[@B6],[@B25]-[@B27]\] (Table [1](#T1){ref-type="table"}). Because the exisiting programs last for considerably long durations, this program was newly designed in order to decrease the training time to 20 minutes. During training, subjects practiced their fundamental basketball skills and additionally carried out jump and balance training to increase their landing skills. During the first three weeks of training, the technique phase (Phase1) which focused on improving the subject\'s landing technique was implemented. Three basic techniques were stressed: 1) a soft landing with a great bend to the hip, knee and ankle joint; 2) landing on the ball of the foot with the trunk leaning forward; 3) keeping the subject\'s knee neutral without medial motion. The last two weeks of training, the performance phase (Phase2), focused on increasing the intensity of training and on the use of proper techniques throughout several movements. A trainer or therapist attended the training every time and instructed the athletes to learn these landing techniques throughout each training session. ###### Description of the exercises performed during the jump and balance training Exercise Time or Repetitions Exercise Time or Repetitions ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------- ------------------------------------ --------------------- ***Phase1: Technique*** ***Phase2: Performance*** 1\. Squat jumps 20 sec 1\. Squat jumps 20 sec 2\. 180°jumps 20 sec 2\. Scissors jumps 20 sec 3\. Single leg balance 20 sec 3\. Single leg balance and pass 20 sec 4\. Hop jump (both leg) 20 sec 4\. Hop jump (single leg) 20 sec 5\. Broad jump and hold 28 m 5\. Single-leg hop and hold 14 m/leg 6\. Crossover hop, hop, hop, stick 28 m 6\. Crossover hop, hop, hop, stick 28 m **Squat jumps**: Drop into deep knee, hip, and ankle flexion and then take off into a maximal vertical jump. On landing, immediately return to the starting position and repeat the initial jump. **180°jumps**: Initiates a 2-footed jump with a direct vertical motion combined with a 180°rotation in midair, keeping arms away from the body to help maintain balance. When landing, immediately reverses this jump to the opposite direction. **Single-leg balance (and pass)**: This drill is performed on a balance device that provides an unstable surface. Begin by standing on one foot on the device. After the subject has improved, the training drills can incorporate ball catches and passes. **Hop jumps**: Start by standing next to a small square balance board. Hop onto the board and then hop off on the opposite side. Repeat hopping on and off the board. **Broad jump and hold**: Begin by swinging arms forward and jumping horizontally and vertically at approximately a 45°angle to achieve a maximum horizontal distance. The athlete lands with her knees flexed to approximately 90°. **Crossover hop, hop, hop, stick**: Start on a single limb and jump at a diagonal across the body landing on the opposite limb with the foot pointing straight ahead and immediately redirect the jump in the opposite diagonal direction. **Scissors jumps**: Start in a stride position with one foot well in front of other. Jump up, alternating foot positions in midair. **Single-leg hop and hold**: Initiate the jump by swinging the arms forward while simultaneously extending at the hips and knees. The jump should carry the athlete up at an angle of approximately 45°and attain maximal distance for a single-leg landing. The subject is instructed to land on the jumping leg in deep knee flexion. Data Analysis ------------- The coordinate data obtained from the markers were not smoothed because of the expected noise-canceling property of the PCT. In each trial, we calculated the angular displacements of flexion/extension, valgus/varus, and internal/external tibial rotation using the PCT. The reference position for these measurements was obtained during the static trial. Cerulli et al. \[[@B28]\] reported that ACL strain begins to increase prior to landing and reaches a peak that corresponds to the peak ground reaction force. Therefore, we extracted each variable at the time of foot contact, as well as displacement from the event of foot contact to the event of peak vertical ground reaction force (i.e. during the landing). Moreover, we normalized the event of foot contact to the event of peak vertical ground reaction force as 100%. The root mean square (RMS) of the EMG data was calculated for each trial. The EMG activities of the two hamstrings muscles were averaged together to represent the whole activity of the hamstrings (Ham). For the maximum voluntary contraction, the EMG data were recorded when the subject performed a maximum voluntary knee flexion and extension at 60 degrees of knee flexion against a manual resistance for three seconds. If the strength of their knee extension was greater than the manual resistance, some research assistants cooperated to provide additional resistance to their motion. The RMS of the EMG data was calculated and the mean RMS of the middle one second was used to normalize the dynamic contraction recorded during the landing (%MVC). The hamstrings/quadriceps (quadriceps refers to the rectus femoris in this experiment) ratio (HQR) was calculated to define the flexor muscle activation relative to the extensor muscle activation. The HQR was calculated according to procedures outline by Bencke et al. \[[@B15]\]. Average %MVC (RF and Ham) and HQR data output was computed during each of the following time frames: 1) 50 ms before foot contact and 2) 50 ms immediately after foot contact. The time period 50 ms before foot contact indicates pre-activity of these muscles. Considering an electromechanical delay of about 50 ms \[[@B29]\], the pre-activation force corresponds to the time before foot contact. The time period 50 ms after foot contact was chosen since a previous study found that the activity of the knee extensors peaked 46 ms after toe contact while the knee flexors showed minimum EMG activity \[[@B15]\]. It is thought that these activities include spinal reflexive neuromuscular activities \[[@B30]\]. The mean kinematics and electromyographic data of each subject were calculated from 3 successful trials. A repeated ANOVA with post-hoc Bonferroni tests was employed for statistical analysis of the kinematics data. Because the %MVC and HQR data did not distribute normally, a Friedman test with Wilcoxon matched-pairs signed rank test was employed for statistical analysis of these data. Each analysis was used to calculate differences with respect to the three time periods. Differences between preliminary trials 1 and 2 were considered as the changes due to the control session; these changes indicate the effects of learning from the landing task and day-to-day individual variation. Differences between preliminary trial 2 and the post-trial were considered as effects of the training session, which indicates how the training program influenced the subjects\' landing patterns. All statistical comparisons were performed with the level of significance set at p \< 0.05. Intra-class correlation coefficients (ICC (1, 3)), based on multiple same-day trials for kinematics and electromyographic data, and standard error of measurement (SEM) were calculated and are described in Table [2](#T2){ref-type="table"}. These data indicate substantial or almost perfect reliability \[[@B31]\]. ###### ICC (1, 3) and SEM for kinematics and electromyographic data Position at foot contact (deg) ----------- ----------------------------------- ------------------ ------- -------- Flexion Ext. tibial rot. Varus ICC 0.93 0.94 0.95 SEM (deg) 1.80 1.42 0.61 Displacement during landing (deg) Flexion Int. tibial rot. Varus Valgus ICC 0.96 0.87 0.64 0.86 SEM (deg) 1.41 2.18 1.22 0.39 Electromyographic data (%MVC) Rectus femoris Hamstrings ICC 0.79 0.62 SEM (%) 10.84 9.49 ICC: Intraclass correlation coefficients SEM: Standard error of measurement Results ======= Kinematics Data --------------- Mean joint kinematics during the single limb drop landing is illustrated in Figure [1](#F1){ref-type="fig"}. Regarding the position of the knee at foot contact, the knee flexion angle for the Post-training trial was significantly larger than that for the Pre-training 2 trial (p \< 0.01) (Table [3](#T3){ref-type="table"}). Regarding rotational displacements that occurred during landing, the absolute change in knee flexion for the Post-training trial was significantly larger than that for the Pre-training 2 trial (p \< 0.001) (Table [4](#T4){ref-type="table"}). No other significant differences between kinematic data could be detected. The observed statistical power of ANOVA for knee flexion, internal tibial rotation, and knee valgus of the tibia at foot contact was 0.96, 0.13, and 0.12, respectively. While, the observed statistical power of ANOVA for the absolute change in knee flexion, internal tibial rotation, and knee valgus during the landing were 1.00, 0.21, and 0.56, respectively. {#F1} ###### Mean (SE) values of position at foot contact Pre-training 1, Pre-training 2, and Post-training (deg) Flexion External tibial rot. Varus ---------------- -------------- ---------------------- ----------- Pre-training 1 19.5 (2.4) 1.9 (2.2) 1.3 (0.9) Pre-training 2 19.3 (2.5)\* 0.3 (2.5) 0.1 (1.2) Post-training 24.4 (2.1)\* 1.1 (2.0) 0.6 (1.1) \*: p \< 0.01 between Pre-training 2 and Post-training ###### Mean (SE) values of rotational displacement during the landing Pre-training 1, Pre-training 2, and Post-training (deg) Flexion Internal tibial rot. Varus Valgus ---------------- ---------------- ---------------------- ----------- ----------- Pre-training 1 31.6 (2.5) 12.3 (1.4) 1.4 (0.3) 2.8 (0.6) Pre-training 2 34.3 (2.5)\*\* 13.8 (1.8) 1.3 (0.3) 3.7 (0.7) Post-training 40.2 (1.9)\*\* 13.3 (2.0) 1.3 (0.4) 4.1 (0.6) \*\*: p \< 0.001 between Pre-training 2 and Post-training Electromyographic Data ---------------------- For the 50 ms before foot contact, the %MVC of the Ham in the Post-training trial was significantly greater than that in the Pre-training 2 trial (p \< 0.05), while no significant differences could be determined for the RF (Figure [2a](#F2){ref-type="fig"}). On the other hand, for the 50 ms immediately after foot contact, the %MVC of the Ham and the RF were not significantly different between all trials (Figure [2b](#F2){ref-type="fig"}). The HQR were not significantly different between all trials for neither the 50 ms before foot contact nor for the 50 ms immediately after foot contact (Figure [3](#F3){ref-type="fig"}). {#F2} {#F3} Discussion ========== This paper examined the effects of a jump and balance training program designed to prevent ACL injury on both knee kinematics and muscle activity during a single limb drop landing. Our training program consisted of plyometrics, jump and balance training, fundamental skills used in basketball and instructions on proper landing techniques. Each training session lasted for only approximately 20 minutes a day, therefore a trainer, coach or therapist could realistically perform this training program easily during practice or warm-up. This jump and balance training program resulted in increased knee flexion and increased hamstrings activity. While we discuss later about each factors; further, multi-component training, including jump and balance training could result in changes in knee kinematics and femoral muscle activity. The training program was considered to influence knee flexion during landing. However, in tibial rotation and knee varus/valgus, the difference between the training sessions was smaller than the SE and learning effects (difference from the control session). Therefore, the changes in tibial rotation and knee varus/valgus were not considered to be effects of the training program. While the results regarding knee flexion partially proved our hypothesis, our hypotheses that training would decrease knee valgus and internal tibial rotation were not supported. For frontal plane knee motion previous studies \[[@B11],[@B14]\] reported no differences in knee valgus during single limb landing after plyometric or agility training. However, Myer et al. \[[@B12]\] reported that both plyometric and balance training decreased the knee abduction angle during a medial drop landing. There is a possibility that single limb drop landing is not suitable to detect the change in frontal plane knee motion. It is also thought that the number of subjects in this study was low and the statistical power for the valgus. This low statistical power makes it difficult to demonstrate the small changes in knee valgus. For tibial rotation, although we could not compare with previous studies, it is possible that these same reasons impede detecting the change in motion in knee frontal motion (i.e. selection of task and statistical problem). Otherwise, it may be difficult to control tibial rotation, like joint laxity. According to other studies that used video analysis, the position of the knee at the moment of injury is in slight knee flexion and valgus rotation with either internal or external tibial rotation \[[@B19]-[@B21]\]. Teitz \[[@B19]\] reported that the angle of knee flexion at the time of injury was less than 30 degrees. In this study, slight knee flexion, valgus and internal tibial rotation also occurred during the single limb drop landing. It is thought that decreasing tibial rotation and valgus rotation or increasing the angle of knee flexion is effective in preventing this knee position at the moment of ACL injury. Some previous studies \[[@B11],[@B12]\] indicated that training increases the knee flexion angle during both limb landings. The results of this study agree with those previous studies. Onate et al.\[[@B13]\] reported that videotape feedback and instruction increased peak knee flexion during landing. In this study, subjects increased not only peak knee flexion, but also initial knee flexion at foot contact. Because subjects were instructed to go through more knee flexion during training, increased peak knee flexion was not a novel finding. However, continuing the jump and balance training over a period of time also increased initial knee flexion. Thus it can be said that these increases in knee flexion are a direct effect of the jump and balance training. Increased knee flexion angles may lead to changes in the function of the quadriceps and the hamstrings. In slight knee flexion, i.e., less than 30 degrees, contraction of the quadriceps strains the ACL \[[@B32]-[@B34]\]. Hamstrings contraction cannot reduce ACL strain with the knee slightly flexed because these muscles meet the tibia at a small angle \[[@B33],[@B35]\]. On the other hand, at angles of knee flexion greater than 60 degrees, quadriceps contraction does not increase ACL strain \[[@B34]\] and the anterior tibial translation and internal tibial rotation as a result of quadriceps contraction is decreased \[[@B36]\]. Moreover, hamstrings contraction reduces anterior tibial translation and internal tibial rotation at these angles \[[@B37]\]. Therefore, increasing the knee flexion angle during landing may have a beneficial effect to reduce the strain in the ACL. From the results of this study, the jump and balance training program which increased the knee flexion angle during landing also had the effect of decreasing the risk of ACL injury. Hamstrings activity was also increased after training. In the training sessions, subjects were instructed to bend their knees and keep their knees neutral. Subjects were directed to keep a deep knee flexion angle during landing and to stabilize the position of their knee joint on the balance board. It is reported that peak activity of the hamstrings occur between 50-70 degrees of knee flexion \[[@B38]\]. Hamstrings contraction and coactivation of the hamstrings and the quadriceps have an important role to stabilize the knee joint \[[@B37],[@B39]-[@B41]\]. Therefore, the subjects learned the correct posture and landing technique which would increase hamstrings activity. The activity of the hamstrings before foot contact was significantly increased. Recently several researchers have focused on pre-activated muscle patterns in response to anticipated movements and joint loads \[[@B15],[@B42]\]. Pre-activation is important for the dynamic stability of the knee because it provides fast compensation for encountered external loads. Wojtys and Huston \[[@B30]\] described the possibility that pre-activation can provide adequate muscular protection to restrain the knee. Although there are several reflexes that occur after a perturbation \[[@B40]\], the muscle activity is too slow to provide any ligament protection \[[@B43],[@B44]\]. Between preparatory and reactive muscle activation, there is a period of latency that results from electromechanical delay (EMD) \[[@B29]\]. EMD is the delay between neural stimulation of a muscle and the development of muscle tension. Pre-activation is thought to be beneficial for stabilizing the knee after a landing. For this study, although internal tibial rotation and tibial translation were not changed after training, it can be said that the increase in the pre-activity of the hamstrings helped to stabilize the knee joint during landing. There are some limitations in this study. First, we analyzed only the kinematics of the knee joint, even though hip and ankle joint kinematics also play an important role during landing. We examined only the short term effects of the ACL injury prevention training. The incidence of ACL injury during the season was not investigated. We examined the effects of multi-component training. The effects of each training should be examined in the future. We did not consider the effects of menstrual cycle. Our study included control and training sessions, although setting up a control group and an intervention group is ideal to examine the effects of training. However, there were no changes in basketball training during these sessions, which were conducted during the preseason, and training and matches occurred at the same frequency during both sessions. Therefore, the changes in kinematics during the landing could be attributable to the intervened training. Finally, the statistical power for several of the data was relatively low. Using ANOVA at a low statistical power might cause type 2 errors. Although considerable differences in results might be significant, other findings may be overlooked. Future research should be performed to examine the long term effects of the ACL injury prevention program in a large sample size. Conclusions =========== The results of this study indicate that the jump and balance training increased knee flexion and hamstrings activity in female basketball athletes during a single limb drop landing. This program, which includes plyometrics, jump and balance training, fundamental basketball skills, and proper landing instructions, might have partial effects that avoid the characteristic knee position of ACL injury, thereby preventing injury. However, the expected changes in frontal and transverse kinematics of the knee were not observed. Competing interests =================== Grant-in-Aid for Scientific Research (C) (16500394) Authors\' contributions ======================= YN participated in the design of the study and drafted the manuscript. HI developed an algorithm of the PCT. MA and TF conceived of the study, and participated in its design and coordination and helped to draft the manuscript. Each author has read and concurs with the final manuscript\'s contents. Acknowledgements ================ This work was funded by Grant-in-Aid for Scientific Research (C) (16500394) in 2004 and 2005.

Introduction {#s1} ============ The prevalence of diabetes among adults aged ≥65 years is projected to increase dramatically by 2050 ([@B1]). Dementia affects up to 16% of diabetic patients aged ≥65 years and 24% aged ≥75 years ([@B2],[@B3]), and evidence shows that the two conditions share a pathophysiological link ([@B4]--[@B6]). As described in the federally mandated National Plan to Address Alzheimer's Disease ([@B7]), U.S. health care is poorly situated to address the needs of older adults with dementia and common comorbidities such as diabetes. This is evident in the two- to threefold increased odds of severe hypoglycemia and preventable hospitalizations in diabetic patients with comorbid dementia ([@B2],[@B8]). Improving ambulatory care, particularly reducing known risk factors for hypoglycemia, in older patients with coexisting diabetes and dementia is critical, but to date, few such efforts have been made. There is growing consensus that because of their increased risk for hypoglycemia and reduced potential to benefit from tight glycemic control, older diabetic patients with dementia should avoid tight glycemic control (HbA~1c~ \<7% \[53 mmol/mol\]) and instead pursue moderate HbA~1c~ levels of 7% to \<9% (53 to \<75 mmol/mol). Older individuals with dementia often have other risk factors for hypoglycemia, including weight loss, changes in appetite and eating habits, and difficulty following prescribed regimens ([@B9]--[@B11]), and a reduced ability to recognize and respond appropriately to symptoms ([@B9],[@B12],[@B13]). Furthermore, older patients with dementia have an average life expectancy of 2--8 years ([@B14]--[@B16]) and are unlikely to experience benefits of tight glycemic control, which take years to accrue and are less likely in patients with long-standing diabetes ([@B11]). As such, since 2003, guidelines from the U.S. Departments of Veterans Affairs and Defense (VA/DoD) have presented a risk-stratified approach to glycemic control, recommending tight control (HbA~1c~ \<7% \[53 mmol/mol\]) only for patients with a life expectancy of 10--15 years or more and absent/mild microvascular complications ([@B17],[@B18]). The American Diabetes Association and American Geriatrics Society (AGS) subsequently adopted similar recommendations ([@B19],[@B20]), with the AGS in 2013 including the avoidance of tight glycemic control in older adults with comorbidities in its Choosing Wisely campaign ([@B21]). Although others have documented the elevated risk of hypoglycemia in older diabetic patients with dementia ([@B2]), no studies to our knowledge have focused on understanding risk factors for tight glycemic control in patients with comorbid dementia. Choice of antidiabetic medication also may exacerbate risks to type 2 diabetic patients with dementia, especially when HbA~1c~ is tightly controlled ([@B22]). Sulfonylureas and insulin are associated with a high risk of hypoglycemia ([@B19]), and the 2013 AGS guidelines explicitly recommend against the use of glyburide and chlorpropamide in all patients aged ≥65 years ([@B20]). However, sulfonylureas and insulin are recommended as first- and second-line diabetes therapies in the general patient population due to strong evidence of efficacy in lowering HbA~1c~, low cost, market longevity, and low risk of adverse events apart from hypoglycemia ([@B17],[@B19]). As patients develop dementia, the risk of hypoglycemia increases and the risk-to-benefit balance of using these agents changes, especially if HbA~1c~ is tightly controlled. To our knowledge, no prior studies have examined the prevalence of or risk factors for the use of medications with a high risk of hypoglycemia in patients with dementia. In this study, we addressed these gaps in the literature by identifying risk factors for tight glycemic control in older veterans with dementia receiving antidiabetic medication therapy. We also identified the prevalence and characteristics of patients at highest potential risk for hypoglycemia through their use of sulfonylurea and/or insulin after exhibiting tight glycemic control. Such data can help to inform future interventions to improve adherence to the VA diabetes treatment guidelines and AGS Choosing Wisely guidelines recommending against tight glycemic control in this population and enhance safer prescribing for veterans with dementia previously shown to be at especially high risk for severe hypoglycemic events ([@B2]). Research Design and Methods {#s2} =========================== Design and Data Sources {#s3} ----------------------- We conducted a longitudinal, retrospective cohort study using administrative data from the VA health care system and Centers for Medicare & Medicaid Services. Data sources included were VA Medical SAS Datasets, records of dispensed outpatient prescriptions, and laboratory values linked to Medicare Part A, Part B, and enrollment data obtained for a larger study ([@B23],[@B24]). All baseline variables and HbA~1c~ levels were based on fiscal year (FY) 2008 data; the first 120 days of FY2009 served as the follow-up period for determining antidiabetic medication use. The VA Pittsburgh Healthcare System Institutional Review Board approved this study. Sample {#s4} ------ To construct the sample, we first used VA data to identify all veterans who *1*) were age ≥65 years as of the start of FY2008 (1 October 2007) and *2*) were given two or more inpatient or outpatient diagnoses for type 2 diabetes (ICD-9 250.x, 250.x2) in FY2008--2009 (with first diagnosis in FY2008) or received an oral diabetes medication through the VA in FY2008 ([@B25]). We then linked to Medicare claims and enrollment data to further refine the sample ([Fig. 1](#F1){ref-type="fig"}). We applied the Medicare Chronic Conditions Warehouse ICD-9 diagnosis code list for Alzheimer's Disease and Related Disorders (ADRD) to both VA records and Medicare claims to identify patients with dementia ([@B26],[@B27]). The ICD-9 codes in this algorithm include Alzheimer disease, vascular dementia, and a range of other specific related disorders (see [Supplementary Table 1](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-0599/-/DC1) for full list of diagnoses). This definition contains only minor differences from an algorithm shown to have good sensitivity and specificity compared with a gold standard clinical dementia assessment ([@B27]). Patients without an ADRD ICD-9 code who filled a VA prescription for an antidementia medication (galantamine, rivastigmine, donepezil, memantine, or tacrine) in FY2008 were also included. Next, because veterans aged ≥65 years are eligible to enroll in Medicare and use non-VA health care in addition to VA care (yet we did not have access to non-VA prescription drug records), we took two steps to restrict the sample to patients with diabetes managed primarily within the VA: *1*) eliminating all patients whose Medicare records indicated enrollment in a non-VA source of drug coverage during follow-up (i.e., Medicare Part D stand-alone plan, Medicare Advantage plan, employer-sponsored plan eligible for a Centers for Medicare & Medicaid Services retiree drug subsidy) and *2*) excluding all remaining patients with HbA~1c~ levels solely monitored outside the VA, indicated by no HbA~1c~ values in VA records and at least one procedure code for an HbA~1c~ test in Medicare claims. Next, to ensure accurate capture of outpatient medications used in the 120-day follow-up period, we excluded patients who died or resided in a VA or Medicare-covered hospital or nursing home for ≥30 days during the follow-up period and, as a result, may not have needed to refill an outpatient VA prescription during the follow-up period. Finally, we excluded all remaining patients who did not fill at least one VA prescription for antidiabetic medication during follow-up. {#F1} Measures {#s5} -------- ### Outcomes {#s6} We extracted the last available HbA~1c~ value in FY2008 from VA laboratory data to quantify patients' glycemic control as tightly controlled (\<7% \[53 mmol/mol\]), moderately controlled (7% to \<9% \[53 to \<75 mmol/mol\]), poorly controlled (≥9% \[≥75 mmol/mol\]), and not monitored (no HbA~1c~ tests recorded in either VA or Medicare data). We also captured whether the HbA~1c~ value was obtained in an outpatient versus inpatient setting for use in sensitivity analyses. We used outpatient VA drug-dispensing records for the first 120 days of FY2009 to identify patient use of antidiabetic medication associated with a high risk of hypoglycemia, defined as having at least one fill for a sulfonylurea and/or insulin. We used a 120-day window to determine the period prevalence of use of these drugs in close relation to the last HbA~1c~ measurement. A 120-day window was chosen because even individuals obtaining 90-day antidiabetic prescriptions would be expected to fill at least one prescription during this period, allowing for some potential nonadherence. For descriptive purposes, we also captured other specific antidiabetic medication classes (biguanides \[metformin\], thiazolidinediones \[TZDs\], α-glucosidase inhibitors, meglitinides, dipeptidyl peptidase-4 \[DPP-4\] inhibitors, amylin analogs, and GLP-1 receptor agonists) that patients used in this time period and created a summary variable for their overall antidiabetic regimen (noninsulin monotherapy, noninsulin multitherapy, insulin alone, or insulin plus other noninsulin agent). ### Covariate Risk Factors {#s7} We assessed sociodemographic, clinical, and health care utilization factors in relation to tight glycemic control and use of high-risk medications. Patient sex, age (65--74, 75--84, or ≥85 years), and race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, or other) were determined from VA utilization files. Missing VA race/ethnicity values were supplemented with the Research Triangle Institute race code in the Medicare enrollment file ([@B28]). Whether patients had a copay for VA medications in the follow-up period was also extracted. We used the Elixhauser comorbidity measure ([@B29],[@B30]) applied to diagnoses in VA and Medicare files to identify comorbidities present in ≥5% of the present sample as well as recent weight loss (3% of patients) because of an a priori hypothesis that weight loss is an important contributor to tight glycemic control. Detailed information on the specific ICD-9-CM diagnosis codes used to define each comorbid condition can be found at [www.hcup-us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp](http://www.hcup-us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp). We used VA and Medicare records to determine whether patients had at least one inpatient hospital or nursing home stay in FY2008. Finally, we determined whether dementia was documented within the VA, as indicated by either a VA ADRD diagnosis code or a VA prescription fill for an antidementia medication versus documented only in Medicare claims. Analytic Approach {#s8} ----------------- We examined descriptive statistics for all study variables in the full sample and each glycemic control group. To determine the association of sociodemographic, clinical, and health care utilization factors with HbA~1c~ control, we used multinomial logistic regression with robust SEs. Moderate control, which reflects guideline-concordant care, was the reference category. Finally, for patients whose last HbA~1c~ value in FY2008 indicated tight control (*n* = 8,276), we estimated a logistic regression model for use of medication with high hypoglycemic risk. We also conducted a sensitivity analysis using multinomial logistic regression to model use of sulfonylurea but no insulin, use of insulin but no sulfonylurea, and use of both sulfonylurea and insulin as separate outcomes, relative to use of neither agent, and report the results in [Supplementary Table 2](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-0599/-/DC1). Results {#s9} ======= Sample Characteristics {#s10} ---------------------- [Table 1](#T1){ref-type="table"} shows characteristics of 15,880 community-dwelling veterans aged ≥65 years with type 2 diabetes and dementia. Almost all patients were male (99%), and 80% were non-Hispanic white. Comorbidities were common, including hypertension (81%), deficiency anemia (21%), chronic lung disease (19%), and peripheral vascular disease (17%). ###### Characteristics of community-dwelling older veterans with diabetes and comorbid dementia by level of glycemic control Patient characteristics All patients
(*n* = 15,880) Tightly controlled
(*n* = 8,276) Moderately controlled
(*n* = 5,669) Poorly controlled
(*n* = 1,131) Not monitored (*n* = 804) ------------------------------------------------------------------------------ ----------------------------- ---------------------------------- ------------------------------------- --------------------------------- --------------------------- Male sex 15,643 (99) 8,157 (99) 5,581 (98) 1,115 (99) 790 (98) Race/ethnicity  Hispanic 1,242 (8) 566 (7) 475 (8) 137 (12) 64 (8)  White, non-Hispanic 12,629 (80) 6,742 (81) 4,489 (79) 776 (69) 622 (77)  Black, non-Hispanic 1,618 (10) 781 (9) 573 (10) 177 (16) 87 (11)  Other 391 (2) 187 (2) 132 (2) 41 (4) 31 (4) Age  65--74 years 3,857 (24) 1,882 (23) 1,442 (25) 367 (32) 166 (21)  75--84 years 8,745 (55) 4,657 (56) 3,058 (54) 586 (52) 444 (55)  ≥85 years 3,278 (21) 1,737 (21) 1,169 (21) 178 (16) 194 (24) Has medication copay 9,515 (60) 4,990 (60) 3,431 (61) 638 (56) 456 (57) Comorbidities  Congestive heart failure 2,860 (18) 1,416 (17) 1,080 (19) 226 (20) 138 (17)  Heart valve disease 1,168 (7) 648 (8) 394 (7) 69 (6) 57 (7)  Peripheral vascular disease 2,624 (17) 1,412 (17) 943 (17) 166 (15) 103 (13)  Hypertension 12,815 (81) 6,727 (81) 4,646 (82) 945 (84) 497 (62)  Chronic lung disease 2,979 (19) 1,600 (19) 1,032 (18) 212 (19) 135 (17)  Hypothyroidism 1,628 (10) 884 (11) 588 (10) 109 (10) 47 (6)  Renal failure 2,511 (16) 1,266 (15) 941 (17) 211 (19) 93 (12)  Solid tumor without metastasis 2,073 (13) 1,110 (13) 755 (13) 137 (12) 71 (9)  Obesity 1,080 (7) 517 (6) 432 (8) 102 (9) 29 (4)  Weight loss 436 (3) 260 (3) 129 (2) 20 (2) 27 (3)  Fluid and electrolyte disorder 2,016 (13) 1,043 (13) 722 (13) 166 (15) 85 (11)  Deficiency anemia 3,308 (21) 1,821 (22) 1,168 (21) 203 (18) 116 (14)  Psychoses 2,065 (13) 1,105 (13) 709 (13) 166 (15) 85 (11)  Depression 2,463 (16) 1,317 (16) 858 (15) 189 (17) 99 (12) Inpatient stay in FY2008 2,416 (15) 1,246 (15) 837 (15) 243 (21) 90 (11) VA documentation of dementia 11,213 (71) 5,869 (71) 3,958 (70) 818 (72) 578 (72) Last HbA~1c~ value in baseline  year (FY2008)[§](#t1n1){ref-type="table-fn"}  HbA~1c~ (%) 6.8 (6.3--7.6) 6.3 (6.0--6.6) 7.5 (7.2--8.0) 9.8 (9.3--10.7) N/A  HbA~1c~ (mmol/mol) 51 (45--60) 45 (42--49) 58 (55--64) 84 (78--93) N/A Medication use in follow-up period (first 120 days of FY2009) Medication regimen  Noninsulin monotherapy 7,298 (46) 4,942 (60) 1,756 (31) 156 (14) 444 (55)  Noninsulin multitherapy 3,081 (19) 1,438 (17) 1,337 (24) 180 (16) 126 (16)  Insulin alone 3,308 (21) 1,237 (15) 1,502 (27) 421 (37) 148 (18)  Insulin plus other agent 2,193 (14) 659 (8) 1,074 (19) 374 (33) 86 (11) Medication class[\#](#t1n2){ref-type="table-fn"}  Insulin 5,501 (35) 1,896 (23) 2,576 (45) 795 (70) 234 (29)  Sulfonylurea 8,927 (56) 4,690 (57) 3,204 (57) 548 (49) 485 (60)  Metformin 6,487 (41) 3,593 (43) 2,238 (39) 382 (34) 274 (34)  TZDs 826 (5.2) 339 (4) 375 (7) 74 (7) 38 (5)  α-Glucosidase inhibitors 233 (1) 81 (1) 116 (2) 28 (3) 8 (1) Use of medications with high  hypoglycemic risk  No insulin/no sulfonylurea 2,842 (18) 2,063 (25) 598 (11) 42 (4) 139 (17)  No insulin/yes sulfonylurea 7,537 (47) 4,317 (52) 2,495 (44) 294 (26) 431 (54)  Yes insulin/no sulfonylurea 4,111 (26) 1,523 (18) 1,867 (33) 541 (48) 180 (22)  Yes insulin/yes sulfonylurea 1,390 (9) 373 (5) 709 (13) 254 (22) 54 (7) Data are *n* (%) or median (interquartile range). Tightly controlled, HbA~1c~ \<7% (53 mmol/mol); moderately controlled, HbA~1c~ 7 to \<9% (53 to \<75 mmol/mol); poorly controlled, HbA~1c~ ≥9% (≥75 mmol/mol); not monitored, no evidence of having received any FY2008 HbA~1c~ tests in VA or Medicare records. N/A, not applicable. §Presented for the 15,076 patients with HbA~1c~ values in FY2008. \#Data not shown for use of meglitinides, DPP-4 inhibitors, amylin analogs, and GLP-1 agonists; \<1% of the total sample used these agents. The majority (52%; *n* = 8,276) of patients had tight glycemic control, 36% had moderate control, 7% had poor control, and 5% did not have HbA~1c~ monitored in FY2008. Within the tight control group, the mean and median HbA~1c~ value was 6.3% (45 mmol/mol; minimum 3.8% \[18 mmol/mol\], maximum 6.9% \[52 mmol/mol\], interquartile range 6.0--6.6% \[42--49 mmol/mol\]). More than 97% of HbA~1c~ values were obtained in an outpatient setting, and exclusion of inpatient HbA~1c~ values did not substantively affect the distribution of HbA~1c~ values seen in the overall sample or within glycemic control categories. [Table 1](#T1){ref-type="table"} also provides information on the medication regimens and specific medication classes used by the sample. Overall, noninsulin monotherapy was most common (46%) followed by insulin alone (21%), noninsulin multitherapy (19%), and insulin plus another noninsulin agent (14%). Sulfonylureas, metformin, and insulin were the most common classes used, followed by TZDs and α-glucosidase inhibitors. Meglitinides, DPP-4 inhibitors, amylin analogs, and GLP-1 receptor agonists were each used very rarely (\<1% of the sample). Overall, 82% (*n* = 13,038) of the sample used a regimen associated with an increased risk of hypoglycemia, including 47% (*n* = 7,537) using sulfonylurea without insulin, 26% (*n* = 4,111) using insulin without sulfonylurea, and 9% (*n* = 1,390) using both agents ([Table 1](#T1){ref-type="table"}). Among tightly controlled patients, 75% (*n* = 6,213) used such a regimen, including 52% (*n* = 4,317) using sulfonylurea, 18% (*n* = 1,523) using insulin, and 5% (*n* = 373) using both agents. Of the 6,213 tightly controlled patients using either sulfonylureas or insulin, 29% (*n* = 1,811) also took at least one other agent not associated with a high hypoglycemic risk (data not shown). Predictors of Glycemic Control Levels {#s11} ------------------------------------- The multinomial regression model revealed patient age, specific comorbidities, and race/ethnicity to be independently associated with having tight versus moderate glycemic control ([Table 2](#T2){ref-type="table"}). Compared with patients aged 65--74 years, those who were 75--84 or ≥85 years old had higher odds of HbA~1c~ \<7% (53 mmol/mol). Heart valve disease, chronic lung disease, weight loss, and deficiency anemia were associated with increased odds of HbA~1c~ \<7% (53 mmol/mol), whereas congestive heart failure, renal failure, and obesity were associated with lower odds of HbA~1c~ \<7% (53 mmol/mol). Compared with non-Hispanic white patients, Hispanic patients also had lower odds of HbA~1c~ \<7% (53 mmol/mol). Racial/ethnic minority status and having an FY2008 inpatient stay were associated with higher odds of poor glycemic control, whereas older age and deficiency anemia were associated with lower odds. ###### Factors independently associated with level of HbA~1c~ control in community-dwelling older veterans with diabetes and comorbid dementia Tightly controlled
(HbA~1c~ \<7% \[53 mmol/mol\]) Poorly controlled
(HbA~1c~ ≥9% \[75 mmol/mol\]) HbA~1c~ not monitored ---------------------------------------- --------------------------------------------------- ------------------------------------------------- ----------------------- ------ ------------ --------- ------ ------------ --------- Male sex 1.10 0.83--1.46 0.49 1.02 0.59--1.75 0.94 0.81 0.45--1.46 0.48 Race/ethnicity  White non-Hispanic (reference) --- --- --- --- --- --- --- --- ---  Black non-Hispanic 0.91 0.81--1.02 0.11 1.68 1.39--2.04 \<0.001 1.16 0.91--1.49 0.24  Hispanic 0.77 0.67--0.87 \<0.001 1.58 1.28--1.96 \<0.001 0.96 0.72--1.28 0.79  Other 0.94 0.75--1.19 0.62 1.73 1.21--2.49 0.003 1.62 1.07--2.44 0.022 Age  65--74 years (reference) --- --- --- --- --- --- --- --- ---  75--84 years 1.16 1.07--1.26 0.001 0.81 0.70--0.94 0.005 1.29 1.06--1.56 0.011  ≥85 years 1.13 1.02--1.25 0.021 0.66 0.54--0.81 \<0.001 1.43 1.13--1.80 0.002 Has medication copay 0.97 0.90--1.04 0.34 0.99 0.86--1.13 0.83 0.84 0.72--0.99 0.036 Congestive heart failure 0.84 0.76--0.92 \<0.001 1.07 0.90--1.28 0.43 1.09 0.88--1.35 0.42 Heart valve disease 1.16 1.01--1.32 0.033 0.91 0.70--1.20 0.51 1.22 0.91--1.64 0.19 Peripheral vascular disease 1.03 0.94--1.13 0.56 0.86 0.72--1.03 0.11 0.84 0.67--1.05 0.13 Hypertension 0.96 0.88--1.05 0.38 1.08 0.91--1.29 0.37 0.39 0.33--0.45 \<0.001 Chronic lung disease 1.10 1.01--1.21 0.038 1.00 0.84--1.18 0.99 1.08 0.88--1.32 0.48 Hypothyroidism 1.03 0.92--1.15 0.62 0.96 0.77--1.19 0.68 0.60 0.44--0.82 0.001 Renal failure 0.90 0.82--1.00 0.045 1.13 0.94--1.35 0.21 0.78 0.61--0.99 0.045 Solid tumor without metastasis 0.99 0.90--1.09 0.85 0.90 0.74--1.10 0.31 0.67 0.52--0.87 0.003 Obesity 0.83 0.72--0.95 0.007 1.07 0.85--1.35 0.56 0.57 0.38--0.84 0.004 Weight loss 1.36 1.09--1.69 0.006 0.93 0.45--1.19 0.21 1.77 1.14--2.74 0.010 Fluid and electrolyte disorder 0.97 0.87--1.08 0.57 1.11 0.91--1.36 0.29 1.09 0.83--1.41 0.55 Deficiency anemia 1.12 1.02--1.22 0.016 0.76 0.64--0.92 0.003 0.78 0.63--0.98 0.030 Psychoses 1.08 0.97--1.20 0.14 1.04 0.86--1.26 0.68 0.87 0.68--1.11 0.26 Depression 1.05 0.95--1.15 0.35 1.05 0.88--1.25 0.60 0.91 0.72--1.15 0.45 Inpatient stay in FY2008 1.04 0.93--1.15 0.50 1.45 1.21--1.73 \<0.001 0.90 0.70--1.17 0.44 VA documentation of dementia diagnosis 1.05 0.98--1.14 0.18 0.99 0.86--1.16 0.94 0.97 0.82--1.15 0.71 Multinomial logistic regression was used, with moderate control (HbA~1c~ ≥7% \[53 mmol/mol\] and \<9% \[75 mmol/mol\]) as the reference category. OR, odds ratio. Predictors of Use of Medications With High Risk of Hypoglycemia {#s12} --------------------------------------------------------------- Among the 8,276 patients with tight control, several significant risk factors emerged for use of medications with high hypoglycemic risk ([Table 3](#T3){ref-type="table"}). Male sex; black race; age 75--84 or ≥85 vs. 65--74 years; presence of congestive heart failure, peripheral vascular disease, or renal failure; and having had a hospitalization or nursing home stay in FY2008 were all associated with higher odds of use of sulfonylureas and/or insulin versus neither agent. Fluid/electrolyte disorder or depression was also associated with lower odds of use of a high-risk regimen. ###### Independent predictors of use of antidiabetic medications with high risk of hypoglycemia in community-dwelling older veterans with diabetes and comorbid dementia with tight glycemic control OR 95% CI *P* value --------------------------------- ------ ------------ ----------- Male sex 1.77 1.19--2.63 0.004 Race/ethnicity  White non-Hispanic (reference) --- --- ---  Black non-Hispanic 1.27 1.05--1.53 0.015  Hispanic 0.92 0.75--1.13 0.42  Other 0.84 0.60--1.16 0.28 Age  65--74 years (reference) --- --- ---  75--84 years 1.28 1.13--1.45 \<0.001  ≥85 years 1.60 1.37--1.88 \<0.001 Has medication copay 0.93 0.83--1.04 0.20 Congestive heart failure 1.51 1.28--1.78 \<0.001 Valvular disease 0.88 0.72--1.08 0.22 Peripheral vascular disease 1.2 1.05--1.41 0.010 Hypertension 1.10 0.97--1.26 0.14 Chronic lung disease 0.94 0.82--1.08 0.38 Hypothyroidism 0.91 0.77--1.08 0.28 Renal failure 4.60 3.67--5.78 \<0.001 Solid tumor without metastasis 0.97 0.83--1.13 0.67 Obesity 1.01 0.82--1.26 0.91 Weight loss 0.88 0.65--1.20 0.43 Fluid and electrolyte disorder 0.83 0.70--1.00 0.047 Deficiency anemia 0.96 0.84--1.10 0.59 Psychoses 0.94 0.81--1.10 0.47 Depression 0.85 0.74--0.98 0.026 Inpatient stay in FY2008 1.26 1.06--1.49 0.008 VA documentation of dementia 1.09 0.97--1.22 0.16 Tight glycemic control, HbA~1c~ \<7% (53 mmol/mol). OR, odds ratio. Conclusions {#s13} =========== In a national cohort of \>15,000 outpatients with medication-treated type 2 diabetes and dementia, we found that more than one-half had HbA~1c~ \<7% (53 mmol/mol), despite clear guidelines recommending higher glycemic targets. In addition, 75% of tightly controlled patients with dementia used medications that further exacerbated their potential risk for hypoglycemia. Even in the context of heightened hypoglycemia risk resulting from the combination of comorbid dementia and tight glycemic control, the results suggest that the large majority of providers do not substitute sulfonylureas and insulin with lower-risk antidiabetic agents. These findings highlight the need for interventions to encourage appropriate deintensification and alteration of medications when diabetic patients develop dementia, which has not been reflected in previous diabetes quality initiatives. The study has several important strengths. Despite growing numbers of older diabetic patients with dementia and their heightened vulnerability to tight control, few studies have examined glycemic control levels partly because large enough data sources containing HbA~1c~ values for this population are rare outside the VA. A recent study documented high prevalence (45--50%) of tight glycemic control generally among veterans with a range of risk factors for hypoglycemia, including dementia ([@B31]). The present study extends this work by focusing specifically on dementia status and examining risk factors for tight control or sulfonylurea/insulin use and by using Medicare data to help to identify patients with dementia. The incorporation of Medicare claims in identifying dementia patients is critical given past research demonstrating its importance in accurately capturing comorbidities for older veterans ([@B32]). In addition, we believe that the present study is the first to document the very high rate of use of sulfonylureas and insulin in tightly controlled patients with dementia, an important and modifiable contributor to these patients' risk for hypoglycemia. We identified key risk factors for tight glycemic control in patients with dementia. Age ≥75, weight loss, chronic lung disease, and deficiency anemias increase the odds of tight glycemic control, whereas obesity is protective. Providers and patients may not recognize that changes in appetite and weight associated with dementia itself, advancing age, or other comorbidities may make it easier to control blood glucose with less intense antidiabetic regimens or that medication may no longer be required. The results suggest that interventions should encourage the review and deintensification of medication regimens in patients who lose weight or reach advanced ages. We also found that patients with concomitant congestive heart failure or renal failure exhibited lower odds of tight control compared with dementia patients without these comorbidities; these diagnoses are noted in VA/DoD guidelines as indications to avoid tight glycemic control. The explicit mention of the role of dementia in setting glycemic control targets, as in the newly revised 2013 AGS guidelines ([@B20]) and a recently proposed quality indicator for diabetes overtreatment ([@B22]), may be an essential first step in encouraging appropriate diabetes treatment deintensification in this population. Although the present findings suggest some contributing factors to the lack of appropriate deintensification of diabetes treatment in patients with dementia, future research should directly engage providers, patients, and caregivers to uncover their perceptions of how dementia should alter glycemic control targets and barriers to pursuing less intensive targets as well as what role guidelines play in these decisions. The present findings also highlight the need for initiatives to support safer antidiabetic prescribing choices for patients with dementia. The widespread use of sulfonylureas and insulin likely reflects VA/DoD guidelines recommending both options as first-line antidiabetic therapies along with metformin and as add-on therapies to metformin or each other if glycemic goals are not achieved, with no discussion of how dementia should affect medication choice. The greatly increased odds of sulfonylureas and/or insulin use in patients with congestive heart failure and renal failure and those aged ≥75 years also likely reflect VA/DoD (and other) guidelines, which list these conditions and age ≥80 years as contraindications to the use of metformin, leaving sulfonylurea and insulin as the preferred agents over other classes of agents with lower hypoglycemic risk. Although these other antidiabetic medication classes (i.e., TZDs, DPP-4 inhibitors, GLP-1 agonists, α-glucosidase inhibitors) have lower associated risk of hypoglycemia, they have lower efficacy in reducing HbA~1c~ and lack robust evidence regarding their safety profile and associated treatment burden in older patients with dementia. Although these alternative agents were not on the VA national formulary at the time of the study (except for acarbose), they were available to all prescribers when necessary through a nonformulary request process. Because clinical trials comparing various antidiabetic medications in older patients with dementia are unlikely to occur, well-designed observational studies are needed to guide prescribing choices for this rapidly growing population. In the meantime, the present results suggest a need for provider outreach efforts to consider medication options with a lower hypoglycemic risk and encourage the use of the minimally intensive regimen required to achieve moderate glycemic control levels. The results of this study should be interpreted in light of several limitations. First, our focus was on understanding patient characteristics associated with two established risk factors for hypoglycemia (i.e., HbA~1c~ \<7% \[53 mmol/mol\], use of sulfonylureas and/or insulin), and we did not capture data on actual hypoglycemic events. Whether deintensification of therapy in response to tight glycemic control or substitution of insulin and sulfonylurea with other antidiabetic agents results in reduced hypoglycemic events in diabetic patients with dementia is an important direction for future research. In addition, the relationship between glycemic control levels and the progression of dementia is unknown. Whether tight glycemic control hinders or hastens dementia progression should be explored in future studies. Second, although guidelines recommend higher HbA~1c~ targets for older patients with dementia, some patients with mild dementia in otherwise excellent health and good functional status may opt for tighter glycemic targets. Although we did capture extensive information on the presence of comorbid conditions, we were not able to account for patients' functional status or preferences in the analyses. Likewise, insulin therapy may be appropriate for older dementia patients with long-standing diabetes who are unable to attain HbA~1c~ targets with noninsulin agents alone. We did not examine dose of medications; thus, it is possible that some tightly controlled patients were prescribed very low doses. However, by limiting the analysis of hypoglycemic medication use to patients with HbA~1c~ \<7% (53 mmol/mol), such patients face a particularly high risk of hypoglycemia and may benefit from deintensification of insulin dose or a switch to a non--insulin-containing regimen. Although this study is unique in capturing such a large number of patients with both diabetes and dementia as well as important clinical variables like HbA~1c~, we acknowledge that the observational study design, relying solely on administrative data, has inherent limitations. In particular, we were not able to capture the full range of covariates that may be associated with glycemic control and medication choice, such as assistance with at-home diabetes self-management from informal family caregivers or nurse-provided home care. We also did not examine interactions between covariates because of a lack of theory to guide such analyses and relatively small numbers of patients with some comorbid conditions (e.g., weight loss). Future research may consider possible multiplicative effects of specific comorbid conditions or other risk factors identified in this study. Patients may also have had HbA~1c~ tests and medications filled outside the VA that we were unable to capture, although we reduced this possibility by limiting the sample to patients who filled antidiabetic medications in the VA and did not have another major source of drug coverage. In addition, we used a single HbA~1c~ value to classify patients' glycemic control at one point in time and therefore cannot determine the extent to which their glycemic control levels were transient versus stable. We also acknowledge that the data are several years old, and care of dementia patients may have improved since, especially in light of increased attention to the possible risks of tight control and hypoglycemia in older patients with comorbidities. Finally, the results reflect a primarily male veteran population and may not generalize to women or nonveterans. In conclusion, the results show a high rate of intense treatment of diabetes in older patients with comorbid dementia, potentially placing them at elevated risk for hypoglycemia and serious adverse events. Equally disconcerting is the high frequency of use of medications known to cause hypoglycemia. The findings present a compelling need for the development of quality initiatives to encourage review of glycemic targets and antidiabetic medications in this rapidly growing group of patients, especially those aged ≥75 years and those with weight loss. This article contains Supplementary Data online at <http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-0599/-/DC1>. **Funding.** C.T.T., S.Z., M.M., and M.J.F. are supported by VA Health Services Research & Development (CIN 13-405). W.F.G. is supported by a VA Health Services Research & Development Career Development Award (CDA 09-207). The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. **Duality of Interest.** No potential conflicts of interest relevant to this article were reported. **Author Contributions.** C.T.T. contributed to the study concept and design, study supervision, data acquisition, data analysis and interpretation, statistical analysis, drafting of the manuscript, and critical revision of the manuscript for important intellectual content. W.F.G. and X.Z. contributed to the study concept and design, data acquisition, data analysis and interpretation, and critical revision of the manuscript for important intellectual content. C.B.G. contributed to the study concept and design, data analysis and interpretation, and critical revision of the manuscript for important intellectual content. S.Z. contributed to the data acquisition, data analysis and interpretation, and critical revision of the manuscript for important intellectual content. M.M. contributed to the study supervision, data acquisition, data analysis and interpretation, and critical revision of the manuscript for important intellectual content. M.J.F. contributed to the study supervision; administrative, technical, and material support; data analysis and interpretation; and critical revision of the manuscript for important intellectual content. C.T.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. **Prior Presentation.** Parts of this study were presented in poster form at the 2013 Annual Scientific Meeting of the Gerontological Society of America, New Orleans, LA, 20--24 November 2013.

Background {#Sec1} ========== Nature exploits specific proteolytic cleavage as a mechanism to transform inactive precursor proteins into active enzymes. Numerous examples of this exquisite biological strategy can be found, for instance in: the enzymes of the intrinsic and extrinsic coagulation pathways \[[@CR1]\]; the fibrinolytic precursor-enzyme combination of plasminogen and plasmin \[[@CR2]\]; and in the proteins of the complement and caspase cascades \[[@CR3], [@CR4]\]. Enzyme activation by specific proteolytic cleavage allows for tight regulation of the timing, location, and extent of enzymatic activity. From a biotechnological perspective, attempts have been made to emulate this approach, to produce novel proteins with potential therapeutic applications. Dawson et al. mutated the unique activation site of plasminogen, ordinarily cleaved by tissue-type (tPA) or urokinase-type plasminogen activators, to the thrombin recognition site in coagulation factor XI (FXI) \[[@CR5]\]. This change accelerated plasmin formation at the site of pathological blood clots, and the engineered plasminogen was shown to be a superior thrombolytic agent to tPA in dog or rabbit thrombosis models \[[@CR6]\]. Other examples of conceptually related approaches include single chain variable fragment (ScFv) antibody-interleukin 12 (IL-12) chimeric proteins in which the ScFv component blocks biological activity of IL-12 until tumour-enriched proteases release IL-12 \[[@CR7]\], and coagulation factor IX- (FIX-) albumin fusion proteins that circulate together until FIX is activated and released from albumin via a repetition of its natural activation sites \[[@CR8]\]. Hirudin is a small protein found in medicinal leech secretions that potently inhibits the key coagulation enzyme thrombin \[[@CR9]\]. Early in the biotechnological production of recombinant hirudin, it was noted that much of its activity was lost if its N-terminus was blocked, even by small peptides \[[@CR10], [@CR11]\]. This observation was supported and rationalized by the co-crystallization of hirudin with thrombin, which showed the N-terminus of the small protein inserted into the active site canyon of the enzyme \[[@CR12]\]. These findings prompted efforts to engineer a protease-activated "switch" into hirudin, to create a specific way of activating the thrombin inhibitor. Peter et al. fused an anti-fibrin-ScFv to hirudin, separating the fusion protein components with a factor Xa (FXa) cleavage site \[[@CR13]\]. Our laboratory fused human serum albumin (HSA) to hirudin, separating the components with a plasmin cleavage site \[[@CR14]\]. Zhang et al. capped the N-terminus of hirudin with small peptides cleavable by thrombin, FXa, or FXIa \[[@CR15]\]. There is considerable current interest in FXI(a) as a target for antithrombotic therapy \[[@CR16]\]. Existing antithrombotic agents come with bleeding risks. Both mice and human patients deficient in FXI are protected from thrombosis, and do not appear to suffer a strong bleeding tendency \[[@CR17]\]. Pharmacological down-regulation of FXI mRNA by specific antisense oligonucleotides has been shown to be superior to an existing antithrombotic agent in a phase II clinical trial \[[@CR18]\]. For these reasons, in this study we returned to the concept of an activatable hirudin-albumin fusion protein, substituting an FXIa cleavage site N-terminal to the hirudin moiety, and positioning HSA either C-terminal or N-terminal to the FXIa-removable hirudin activity blocking group. We tested the hypothesis that FXIa-activatable hirudin-albumin fusion protein would diminish thrombosis in vivo without promoting bleeding, while maintaining a circulating reserve of latent protein due to the slowly cleared phenotype of HSA. In this study, we found that FXIa-activatable hirudin-albumin had a superior benefit/risk profile to constitutively active hirudin-albumin, but one inferior to our previously reported plasmin-activatable counterpart \[[@CR14], [@CR19]\]. Methods {#Sec2} ======= Construction of expression plasmids and transformation of *Pichia pastoris* {#Sec3} --------------------------------------------------------------------------- Oligonucleotides were obtained through the Institute for Molecular Biology and Biotechnology (MOBIX) at McMaster University (Hamilton, ON). DNA sequencing to confirm the coding sequences of all plasmids was performed by the same facility. Three of four novel expression plasmids were constructed as follows. Plasmid pPICZ9ssHV3HSAH~6~ \[[@CR20], [@CR21]\] was DNA-amplified using sense and antisense paired oligodeoxyribonucleotide primers in reactions catalysed by heat-stable Phusion DNA polymerase (Thermo Fisher Scientific, Waltham, MA, USA). Primer sequences for each construct are given in Table [1](#Tab1){ref-type="table"}. Following PCR, products were restricted with XhoI and XbaI, gel-purified, and ligated to the 5237 bp XhoI-XbaI double digestion fragment of pPICZ9ssHV3HSAH~6~. Following transformation of competent *E. coli* DH5α subclones with appropriate restriction profiles were verified by DNA sequencing. Clones of correct sequence were designated pPICZ9ss(protein), where (protein) was either EPR-HV3HSA, mycEPR-HV3HSA, or mycHV3HSA, as they were designed to direct the synthesis and secretion of these proteins (see Fig. [1](#Fig1){ref-type="fig"}). Note that EPR is tripeptide Glu-Pro-Arg in single letter amino acid code, and that myc is EQKLISEEDL in the same notation. The fourth novel expression plasmid, pPICZ9ssHSAEPR-HV3, was constructed in a similar manner, except that the template was pPICZ9ssHSACHV3H~6~ \[[@CR14]\], and the PCR product produced using the specific primer pair in Table [1](#Tab1){ref-type="table"} was restricted with SnaBI and XbaI and ligated to the major restriction endonuclease digestion product of pPICZ9ssHSACHV3H~6~ \[[@CR14]\] to yield pPICZ9ssHSAEPR-HV3. Each sequence-verified plasmid described above was linearized via SacI digestion and was then used to transform *Pichia pastoris* strain X33 to Zeocin (Invitrogen) resistance as previously described \[[@CR14], [@CR19]--[@CR21]\].Table 1Oligonucleotide sequences. Oligonucleotide primers used in expression plasmid construction. The unique order number, short name, DNA sequence, and purpose is shown. Sense EPR, myc, and mycEPR oligonucleotides were separately paired with antisense myc/EPR oligonucleotide for amplification and restriction with XhoI and XbaI. HSA sense and antisense clEPRHV3 were paired for amplification and restriction with SnaBI and XbaI. Additional details are provided in the Materials and MethodsNumberNameSequencePurpose123,620,623Sense EPR5'-TCTCTCGAG AAAAGAGAGC CTAGAATCAC CTACACAGAC TGC-3'Mutate and amplify codons to alter HV3HSA cDNA to EPR-HV3HSA, provide XhoI site123,620,624Sense myc5'-TCTCTCGAGA AAAGAGAACA AAAACTCATC TCAGAAGAGG ATCTGATCAC CTACACAGAC TGC-3′Mutate and amplify codons to alter HV3HSA cDNA to myc-HV3HSA, provide XhoI site123,620,625Sense mycEPR5'-TCT CTC GAG AAA AGA GAA CAA AAA CTC ATC TCA GAA GAG GAT CTG GAG CCT AGA ATC ACC TAC ACA GAC TGC-3′Mutate and amplify codons to alter HV3HSA cDNA to myc-EPRHV3HSA, provide XhoI siteML-08-6225Antisense myc/EPR5′- CTCTCCAAGC TGCTCGAAAA GCTC-3'Anchor 3′ end of amplification in all 3 cases above, allow inclusion of XbaI site in amplicon890,140,002HSA sense5'-AGGCATCCTGA TTACTCTGTC GTGCTGCTG-3'Anchor 5′ end of amplification for HSAEPR-HV3 construct, allow inclusion of XbaI site in amplicon161,722,273Antisense clEPRHV35' TGTGTACGTAA TTCTAGGCTC GGATCCTAAG GCAGC-3'Mutate and amplify codons specifying EPR cleavage site for HSAEPR-HV3 construct, allow inclusion of SnaBI site in ampliconFig. 1Schematic representation of recombinant proteins employed in this study. All polypeptides contained human serum albumin (HSA) sequences, were expressed in *Pichia pastoris,* and were purified from media conditioned by methanol-induced cultures. Polypeptides are shown in linear format, amino- to carboxyl-terminus, to emphasize their modular design. Protein segments are shown in black boxes (HSA, 585 amino acids), while small proteins (HV3, hirudin variant 3, 66 amino acids, or KPI, the Kunitz Protease Inhibitor domain of protease nexin 2, 57 amino acids) are shown in grey, and peptides (G6, hexaglycine spacer, H6, hexahistidine affinity tag, EPR, FXIa-cleavable tripeptide (i.e. Glu-Pro-Arg, and myc, myc oncoprotein epitope tag, EQKLISEEDL) in white. From top to bottom, the following recombinant, His-tagged proteins are depicted, with their length in amino acids (\[a.a.\], at right of linear depiction): EPR-HV3HSA; HSA-EPRHV3; mycEPR-HV3HSA; mycHV3HSA; KPIHSA; and HSA Expression and purification of albumin fusion proteins {#Sec4} ------------------------------------------------------ Clonal *Pichia pastoris* cell lines transformed with expression plasmids described above, or pPICZ9ssHSACHV3, pPICZ9ssHSAH~6~ \[[@CR14]\], or pPICZ9ssKPIHSA \[[@CR22]\], were used to condition buffered minimal medium supplemented with methanol for 72 h. Following neutralization, clarification and ultrafiltration, albumin fusion proteins were purified from concentrated conditioned media by nickel-chelate affinity chromatography as previously described \[[@CR14], [@CR21]\]. Prothrombin time assays {#Sec5} ----------------------- Prothrombin times (PT) were determined using an STA 4 coagulation analyzer (Diagnostica Stago, Asnières-sur-Seine, France). Citrated normal human pooled plasma (\[NHPP\], Precision Biologic, Dartmouth, NS, Canada) or NHPP immunodepleted of Factors VIII, IX, XI, or XII (Haematologic Technologies, Essex Junction, VT, USA) was combined with purified fusion proteins (45:5 μl by volume) in saline, or saline alone, and incubated at 37 °C for 2 -- 240 min prior to initiation of PT measurement through addition of 50 μl Thromborel S PT reagent containing human placental thromboplastin and calcium chloride (Siemens Healthcare Diagnostics, Oakville, ON, Canada). In some experiments, citrated CD-1 mouse pooled plasma was substituted for NHPP. Two-stage activation and thrombin activity assays {#Sec6} ------------------------------------------------- To determine if albumin fusion proteins possessed latent antithrombin activity, a two-stage assay was employed as previously described \[[@CR14], [@CR19]\]. Briefly, albumin fusion proteins (400 nM) were combined with 300 nM purified protease in PPNE buffer (20 mM sodium phosphate, pH 7.4, 0.1% polyethylene glycol 8000, 100 mM NaCl, 0.1 mM EDTA) for varying times. Purified proteases included FXIa, Factor Xa (FXa), Factor XIIa (FXIIa), plasmin (Enzyme Research Laboratories (South Bend, IN, USA)) and recombinant FVIIa (Niastase, Novo Nordisk Canada, Mississauga, ON, Canada). First-stage reaction products were combined 1:1 with 10 nM thrombin in PPNE for 1 min, and then diluted 10-fold into 100 μM S2238 chromogenic substrate (Chromogenix, Lexington, MA, USA). Colour generation from the amidolysis reaction was followed for 5 min to determine the reaction velocity. In some reactions HEPES-buffered saline (HBS; 25 mM HEPES, pH 7.4, 100 mM NaCl) supplemented with 5 mM calcium chloride was substituted for PPNE. Electrophoresis and immunoblotting {#Sec7} ---------------------------------- SDS polyacrylamide gel electrophoresis was carried out under denaturing and reducing conditions using a Mini-PROTEAN III electrophoresis system following the manufacturer's instructions (BioRad Laboratories, Mississauga, ON, Canada). Replica gels were transferred to nitrocellulose using an iBlot Gel Transfer Device as specified by the manufacturer (Invitrogen, Carlsbad, CA, USA). Immunoblotting was as previously described, using Tris-buffered saline/0.02% Tween 20 (TBST) for washing and 3% BSA in TBST for antibody incubations. Primary antibodies were all monoclonal: mouse anti-HSA (Genway Biotech, San Diego, CA; mouse anti-polyhistidine (Sigma-Aldrich, Oakville, ON, Canada); and mouse anti-myc (Thermo Fisher Scientific, Mississauga, ON, Canada). The secondary antibody was goat anti-mouse IgG, alkaline phosphatase- (AP-) conjugated. Blots were visualized through colour development resulting from the action of AP on substrate nitro blue tetrazolium chloride/5-bromo-4-chloro-3-indolyl-phosphate, toluidine salt (NBT/BCIP). Murine carotid artery thrombosis model {#Sec8} -------------------------------------- The time to occlusion (TTO) of the ferric-chloride treated carotid artery was assessed by Doppler ultrasound as previously described \[[@CR14], [@CR23]--[@CR26]\]. Briefly, CD-1 mice were anesthetized, the carotid artery was exposed surgically, and thrombosis was induced by topical application of a 1 mm^2^ patch of Whatman paper soaked in 10% (*w*/*v*ol) ferric chloride for 3 min. Two min prior to patch application, purified EPR-HV3HSA or mycHV3HSA in saline, or saline vehicle, were injected intravenously at 120 or 40 mg/kg body weight. Arterial blood flow was monitored after patch removal and the TTO defined as the time until flow dropped below 0.1 ml/min. Vessels not occluded at the end of the 60 min observation period were scored with a TTO of 60 min. These and all other experiments with mice were carried out under the terms of an Animal Utilization Protocol (AUP) approved by the Animal Research Ethics Board of the Faculty of Health Sciences of McMaster University. At the close of all experiments, anesthetized mice were euthanized by cervical dislocation as stipulated in the approved AUP. Murine LPS/L-NAME fibrin deposition thrombosis model {#Sec9} ---------------------------------------------------- The deposition of fibrin in the heart and kidney of CD-1 mice treated with lipopolysaccharide (LPS, serotype 0127:B8, Sigma-Aldrich) and vasoconstrictor nitro-L-arginine methyl ester (L-NAME, Alexis, Nottingham, UK) \[[@CR27], [@CR28]\]was assessed using radiological methods as previously described \[[@CR23]\]. Briefly, L-NAME was given intraperitoneally (IP) (50 mg/kg body weight) at 0, 30, 120, and 240 min; the second injection also contained 2 mg/kg body weight LPS. One minute prior to the first IP injection, albumin fusion proteins in saline or saline vehicle were given at 40 mg/kg body weight with 30 million cpm ^125^I-human fibrinogen labeled using the Iodogen method \[[@CR29]\]. Five hours after the first IP injection, mice were euthanized, and hearts and kidneys were excised, washed in ice-cold saline, and organ-associated radioactivity was determined by γ-counting. Murine bleeding models {#Sec10} ---------------------- Blood loss subsequent to injury was determined in two previously described models: liver laceration \[[@CR19]\]; and tail transection model \[[@CR14]\]. In the first model, CD-1 mice were anesthetized and a standard, 5 mm perforating incision was made through an exteriorized liver lobe placed on a tared weigh boat. Shed blood was captured in the weigh boat and combined with shed blood clotted on the lobe surface 15 min after injury. Albumin fusion proteins or saline vehicle were injected intravenously (IV) two minutes prior to injury. In the second model, anesthetized CD-1 mice were also injected IV with albumin fusion proteins or saline vehicle just prior to injury, but the injury in this case was tail transection at a point 1.0 cm from the tip. Shed blood was collected on Whatman paper for 15 min, and subsequently eluted in water. Blood loss was quantified by determining the optical density of the hemolysate at 492 nm wavelength and determined using a standard curve generated from hemolysed whole blood from the same mouse, taken as a terminal sample. Statistical Analysis {#Sec11} -------------------- A *p* value \< 0.05 was considered significant throughout this study. Statistical analysis was performed using computer software (InStat, Version 3.06, GraphPad Software, San Diego, CA, USA). Comparisons between two data sets were made using Student's paired test. For multiple comparisons, one way Analysis of Variation (ANOVA) was employed, with post-tests. Normally distributed data sets with similar standard deviations were assessed with Tukey's post-test, while data sets failing either criterion were assessed using non-parametric (Kruskal-Wallace) ANOVA with Dunn's post-test. Data sets were logarithmically transformed if this transformation allowed them to meet the two criteria above. Results {#Sec12} ======= Recombinant protein design and expression {#Sec13} ----------------------------------------- Novel albumin fusion proteins were designed to test the effects of transient blockade of hirudin activity as shown in Fig. [1](#Fig1){ref-type="fig"}. The N-terminus of HV3 was blocked using a variety of different groups including: the FXIa-cleavable tripeptide EPR (in EPR-HV3HSA); its extended triskaidecapeptide counterpart mycEPR (in mycEPR-HV3HSA); the entire coding sequence of HSA (in HSAEPR-HV3); and noncleavable control decapeptide myc (in mycHV3HSA). Other proteins depicted in Fig. [1](#Fig1){ref-type="fig"} have been previously described, including KPIHSA \[[@CR22]\], a direct FXIa inhibitor-albumin fusion, His-tagged unfused HSA \[[@CR14]\], and constitutively active HV3HSA (also called HV3HSAH~6~) \[[@CR21], [@CR30]\]. All proteins were expressed at similar levels and were purified from conditioned *P. pastoris* media; as shown in Fig. [2a](#Fig2){ref-type="fig"}, they demonstrated relative electrophoretic mobilities consistent with their predicted length in amino acids, given that aberrant migration of HV3-containing polypeptides relative to molecular mass standards has been previously noted and arises from clustered negative charges in the HV3 C-terminal dodecapeptide \[[@CR14], [@CR20], [@CR21]\]. Immunoblotting revealed that all *P. pastoris*-derived fusion proteins reacted with anti-HSA (Fig. [2b](#Fig2){ref-type="fig"}) and anti-hexahistidine monospecific polyclonal antibodies (Fig. [2c](#Fig2){ref-type="fig"}), while only mycEPR-HV3HSA and mycHV3HSA reacted with anti-myc monoclonal antibodies (Fig. [2d](#Fig2){ref-type="fig"}). Immunoblotting results were supported by immunoreactivity with positive controls (e.g. anti-H6 antibodies reacted with recombinant API but not plasma-derived thrombin B chain, Fig. [2c](#Fig2){ref-type="fig"}).Fig. 2Electrophoretic and immunological characterization of fusion proteins.Panel **a** depicts a 10% SDS-polyacrylamide gel electrophoresed under reducing conditions and stained with Coomassie Brilliant Blue (Gel). Lanes contain markers (MW) or pre-stained markers (PS) or 500 ng of purified proteins (identified above the lanes). Recombinant His-tagged Alpha-1 Protease Inhibitor (API) (purified from transformed *E. coli* Top10 as described \[[@CR39]\]), or human plasma-derived α-thrombin B chain (IIa B) served as non-HSA-related or non-HSA, non-His tagged controls. Panels **b**-**d** depict immunoblots of replicate gels identical to that shown in Panel **a**, except that 250 ng of purified proteins were used per lane, and gels were transferred to nitrocellulose and probed with antibodies specific to HSA, the hexahistidine tag, or the myc epitope (Anti-HSA, Anti-H6, or Anti-myc respectively). PS markers were (in kDa): 180; 130; 95; 72; 55; 43; 34; and 26. MW markers were (in kDa): 200; 150; 120; 100; 85; 70; 60; 50 (greater intensity); 40; 30; 25; and 20 Initial screening of FXIa-activatable albumin fusion proteins {#Sec14} ------------------------------------------------------------- We sought an FXIa-activatable, slowly cleared form of hirudin. As in our previous work with plasmin-activatable hirudin and hirudin-albumin fusion proteins, we selected hirudin variant 3 (HV3), one of the most potent known variants of hirudin \[[@CR21], [@CR30]\]. We chose to assess all activatable and non-activatable forms of HV3 as albumin fusion proteins, to ensure slow clearance and an extended time of pharmacodynamic action. To assess if FXIa recognized the EPR-I cleavage site differently if it was positioned N- or C-terminal to HSA, we compared EPR-HV3HSA and HSAEPR-HV3 with respect to FXIa reactivity. We employed a two-stage assay, first challenging the EPR or HSAEPR blocking groups with FXIa, then diluting the first reaction products into a thrombin S2238 chromogenic activity assay. That FXIa in the first stage of the reaction had no effect on S2238 amidolysis was shown by the lack of difference between its presence or absence on the thrombin-catalysed initial rates (see Fig. [3a](#Fig3){ref-type="fig"}) and also by the lack of reaction of S2238 when FXIa was substituted for thrombin (data not shown). As shown in Fig. [3b](#Fig3){ref-type="fig"}, exposure of EPR-HV3HSA, but not HSAEPR-HV3, to FXIa conferred antithrombin activity on the first, but not the second preparation. We next compared two small N-terminal HV3 blocking groups, tripeptide EPR and triskaidecapeptide mycEPR, to determine if one was more readily recognized by FXIa than the other. Fig. [3c](#Fig3){ref-type="fig"} shows that FXIa preincubation did not activate control protein mycHV3HSA, and exposed greater antithrombin activity for EPR-HV3HSA than for mycEPR-HV3HSA (compare middle and lower progress curves in Fig. [3c](#Fig3){ref-type="fig"}). Activation appeared to be specific to FXIa, as it alone, among 5 coagulation or fibrinolytic enzymes that were tested, released antithrombin activity from EPR-HV3HSA (Fig. [3d](#Fig3){ref-type="fig"}; note that the buffer employed in this experiment differed from those shown in the other panels of Figure [3](#Fig3){ref-type="fig"} to accommodate calcium sensitive proteases FVIIa and FXa). In each case, control experiments in which the protease (FVIIa, FXa, FXIIa, Plasmin, or FXIa) was introduced into the S2238 reaction without addition of albumin fusion proteins showed no difference compared to buffer only, activation protease-free results (data not shown). Accordingly, we selected EPR-HV3HSA over mycEPR-HV3HSA and HSAEPR-HV3 for further study.Fig. 3Activation of latent antithrombin activity in fusion proteins. Purified albumin fusion protein preparations were first reacted with specific purified proteases, then diluted into the presence of thrombin and its chromogenic substrate S2238, and the velocity of the thrombin-catalysed amidolysis reaction was measured and expressed as a percentage of the uninhibited control reaction. Panel A: EPR-HV3HSA was reacted with FXIa for zero (open squares) or 60 min (closed triangles) prior to dilution into the S2238 reaction and measurement of the rate of coloured product formation by thrombin over time. Buffer controls introduced phosphate buffer PPNE (see Materials and Methods) and FXIa (open circles) but no fusion protein into the S2238 reaction**.** Data represents the mean ± SEM, *n* = 7. Panel **b**: As in Panel **a**, EPR-HV3HSA or HSAEPR-HV3 proteins were incubated for zero (0′, white bars) or 60 min (60′, black bars) prior to dilution into S2238 reactions and determination of the rate of thrombin-mediated amidolysis. EPR-HV3HSA reactions are the slope of the same progress curves shown in Panel **a** as a percentage of the Buffer control; HSAEPR-HV3 reactions were determined analogously. Panel **c**: The mean (*n* = 6 ± SEM) of thrombin reaction velocities supplemented with activation reactions of the three fusion proteins labelled to the right of the progress curves with FXIa is shown, at six time points, in PPNE buffer as in Panels **a** and **b**. Panel **d**: As in Panel **a** and **b**, except that HBS / 5 mM CaCl~2~ was used as the buffer instead of PPNE and except that EPR-HV3HSA was combined with the proteases identified on the x axis prior to determination of second stage thrombin-mediated reaction velocity (white bars, FVIIa, FXa, FXIIa, or Plasmin; black bars FXIa). Control reaction "FXIa + No EPR-HV3HSA" contained FXIa but no EPR-HV3HSA in the activation reaction. Buffer reaction contained only HBS / 5 mM CaCl~2~ Data represents the mean ± SEM, *n* = 7 Antithrombotic activity of EPR-HV3HSA in the ferric chloride-treated murine carotid artery model {#Sec15} ------------------------------------------------------------------------------------------------ FXIa-activatable EPR-HV3HSA and its non-activatable counterpart, mycHV3HSA, were compared in the ferric chloride-treated thrombosis model in the topically treated murine carotid artery, at two doses. Fig. [4](#Fig4){ref-type="fig"} shows that EPR-HV3HSA administration prior to ferric chloride exposure lengthened the TTO of the artery four-fold versus vehicle, to 44 ± 16 min at 120 mg/kg body weight, and 2.6-fold versus vehicle, to 22 ± 3 min at 40 mg/kg body weight. In contrast, mycHV3HSA administration elicited TTO values that did not differ statistically from saline vehicle at both doses. We selected the lower dose of 40 mg/kg for further experimentation.Fig. 4Time to occlusion of murine ferric chloride-treated carotid arteries. The time to occlusion (TTO) of surgically exposed carotid arteries treated by topical application of 10% (*w*/*v*ol) ferric chloride, for CD-1 mice receiving saline vehicle (circles) or the two albumin fusion proteins identified on the x axis, in saline (EPR-HV3HSA, squares, and mycHV3HSA, triangles). Mice were injected intravenously with protein doses of 120 mg/kg (Panel **a**) or 40 mg/kg (Panel **b**) two minutes prior to application of a ferric chloride-saturated patch (for three minutes). After patch removal blood flow through the artery was monitored for 60 min using ultrasound. Vessels not occluding during the observation period had their TTO recorded as 60 min. Each point in the scatter plots reflects results from a single mouse of a group of 7. Horizontal lines indicate the mean for each group. Lines above groups show statistical significance between groups as determined by ANOVA with post-tests (*p* \< 0.5, \*, *p* \< 0.01, \*\*, *p* \< 0.001, \*\*\*) Antithrombotic activity of EPR-HV3HSA in a mouse model of inflammatory thrombosis {#Sec16} --------------------------------------------------------------------------------- We next examined the ability of EPR-HV3HSA to modulate thrombosis in a murine endotoxemic model, in which co-administration of bacterial LPS and the nitric oxide antagonist L-NAME promotes fibrin deposition in organs \[[@CR23]\]. We monitored ^125^I-labeled fibrin(ogen) in the heart and kidneys of treated mice. Mice treated with saline vehicle just prior to initiation of LPS/L-NAME demonstrated a significant, 4.3-fold elevation in heart fibrin(ogen) relative to mice not receiving LPS/L-NAME, one that was unaltered by administration of mycHV3HSA, but was significantly reduced, by on average 40%, by dosing with EPR-HV3HSA (Fig. [5a](#Fig5){ref-type="fig"} and [c](#Fig5){ref-type="fig"}). The same pattern of results was observed in the kidneys of the same cohorts of mice; the 3-fold elevation in fibrin(ogen) deposition elicited by LPS/L-NAME administration was significantly, albeit partially, reduced, by on average 37%, by treatment with EPR-HV3HSA but not with mycHV3HSA (Fig. [5b](#Fig5){ref-type="fig"} and [d](#Fig5){ref-type="fig"}). In contrast, treatment with 1.0 mg/kg KPIHSA, a fusion protein that inhibits FXIa with inhibitory constants in the low nM range \[[@CR22]\], did not affect the LPS/L-NAME-mediated increases in fibrin(ogen) deposition(data not shown).Fig. 5Deposition of ^**125**^I-fibrin(ogen) in organs of mice treated with LPS and L-NAME. Mice were treated with or without (No LPS/L-NAME) multiple intraperitoneal (IP) injections of LPS and L-NAME. Immediately prior to the IP injection, all mice were injected intravenously with ^**125**^I-fibrinogen in saline vehicle, alone (Vehicle) or combined with 40 mg/kg body weight EPR-HV3HSA or mycHV3HSA. Five hours later, mice were euthanized and the radioactivity present in excised hearts (panels **a** and **c**) or kidneys (panels **b** and **d**) was determined and expressed as the % of the injected dose (panels **a** and **b**) or normalized to the mean of the % of the injected dose in the No LPS/L-NAME cohort (panels **c** and **d**). Bars depict the mean of 5 determinations, each in a separate mouse, ± SD. Lines above bars show statistical significance between groups as determined by ANOVA with post-tests (*p* \< 0.5, \*, *p* \< 0.01, \*\*, *p* \< 0.001, \*\*\*). Scatter plots (**c** and **d**) show the same data as in **a** and **b**, except that each point corresponds to data from a single mouse, normalized to the "No LPS/L-NAME" control Effects of EPR-HV3HSA in murine bleeding models {#Sec17} ----------------------------------------------- Having demonstrated antithrombotic activity of EPR-HV3HSA in two murine models of thrombosis, we next addressed the issue of whether EPR-HV3HSA promoted bleeding. In the liver laceration model, administration of constitutively active HV3HSA just prior to injury resulted in severe bleeding approaching 50% of the murine blood volume (600 ± 150 mg). In contrast, administration of EPR-HV3HSA or mycHV3HSA was associated with statistically indistinguishable post-injury blood losses of 220 ± 60 mg and 140 ± 90 mg, respectively (Fig. [6a](#Fig6){ref-type="fig"}). Similar results were obtained in the tail transection model, which constitutes a less severe hemorrhagic insult; EPR-HV3HSA was associated with significantly lower blood losses than HV3HSA (130 ± 40 μl versus 230 ± 100 μl) which did not differ significantly from those associated with mycHV3HSA administration (54 ± 40 μl) (Fig. [6b](#Fig6){ref-type="fig"}). In the same mice, no significant difference between bleeding times from the transected tail, as opposed to volumes of shed blood from that site, was found between mice treated with HV3HSA or EPR-HV3HSA, while mice treated with mycHV3HSA ceased to bleed significantly earlier than the other two groups (Fig. [6c](#Fig6){ref-type="fig"}). The proportion of mice still bleeding at the end of the observation period was 8/10 and 9/10 for the EPR-HV3HSA and HV3HSA groups, respectively, compared to only 3/10 for the mycHV3HSA-treated group (Fig. [6c](#Fig6){ref-type="fig"}).Fig. 6Blood losses or bleeding time in mice subjected to liver laceration or tail transection. Mice were subjected to liver laceration (Panel **a**) or tail transection protocols (Panels **b** and **c**). Blood losses were measured by weighing shed and/or clotted blood (**a**) or collecting shed blood into water and quantifying hemolysis (**b**). Continuing bleeding from the transected tail was assessed visually every 30 s for 15 min (**c**). Mice were injected intravenously with 40 mg/kg body weight of the purified proteins identified on the x axis of each panel. Scatter plots show each data point from a separate mouse in groups of 5 (panel **a**) or 10 (panels **b** and **c**) mice. Horizontal lines depict the mean for each group. Lines above bars show statistical significance between groups as determined by ANOVA with post-tests (*p* \< 0.5, \*, *p* \< 0.01, \*\*, *p* \< 0.001, \*\*\*) Stability of EPR-HV3HSA in human and murine plasma {#Sec18} -------------------------------------------------- To gain additional insights as to the potential fate of injected EPR-HV3HSA in vivo, we examined its stability in plasma. Using normal human pooled plasma, we compared the PT in the presence of 13.3 μM albumin fusion proteins (the theoretical peak plasma concentration of the fusion proteins following a 1 mg/kg body weight intravenous dose to mice). As shown in Fig. [7a](#Fig7){ref-type="fig"}, plasma samples supplemented with HV3HSA failed to clot during the 400 s observation period in the assay. Mean PTs for plasma samples supplemented with saline, mycHV3HSA, or HSACHV3 were below 25 s, but the corresponding EPR-HV3HSA-containing samples exhibited PT values in excess of 60 s. A concentration curve of HV3HSA PT times after two minutes' preincubation of chelated NHPP with the fusion protein (Fig. [7b](#Fig7){ref-type="fig"}) illustrated that HV3HSA had only a modest effect on the PT until concentrations in excess of 1.5 μM were exceeded; plasma supplemented with 2.5 μM HV3HSA failed to clot. By extrapolation, this finding suggests that 1.5 to 1.75 μM of the total 13.3 μM EPR-HV3HSA became activated within two minutes of exposure to calcium-chelated human plasma, or 11-13% of the administered dose. Fig. [7c](#Fig7){ref-type="fig"} shows PT values for saline and albumin fusion proteins (excepting HV3HSA, to enhance visibility in the \< 80 s PT range). Inclusion of EPR-HV3HSA significantly prolonged the PT by 4-fold, while mycHV3HSA had no significant effect. A modest but significant elevation resulting from HSACHV3 inclusion in the PT was also observed. When we substituted mouse normal pooled plasma for NHPP in PT assays (Fig. [7d](#Fig7){ref-type="fig"}), similar results were observed. Inclusion of 13.3 μM HV3HSA in murine plasma PT assays eliminated clotting, while EPR-HV3HSA significantly increased the PT by 3- to 4- fold, while HSACHV3-containing assays did not differ from saline controls. The 6-fold prolongation in human plasma PT arising from inclusion of 13.3 μM EPR-HV3HSA was retained when pooled plasma depleted of Factor VIII, Factor IX, Factor XI, or Factor XII were substituted for NHPP (5- to 7- fold observed prolongation, data not shown).Fig. 7Stability of albumin fusion proteins in plasma**.** Panel **a**: Prothrombin times (PT) were measured for reactions containing 13.3 μM fusion protein in saline vehicle, or saline alone. Test samples were combined with human normal pooled plasma (NHPP) and preincubated at 37 °C prior to addition of calcium and PT reagent containing tissue factor for times indicated on the x axis. The mean of 12 determinations ± SD is shown for reactions containing test proteins or solutions shown in the legend at right, Panel **b**: As in Panel **a**, but for varying concentrations of purified HV3HSA preincubated with NHPP prior to PT initiation for two minutes. Data points are the mean of seven determinations ± SEM. Note y axis break. Panel **c**: As in Panel **b**, but for reactions supplemented with the proteins or solutions specified below the x axis. \*\*\*, *p* \< 0.001 versus Saline by non-parametric ANOVA with Dunn's post-tests. Panel **d**: As in Panel **c**, but mouse pooled plasma was substituted for NHPP. Note y axis break. \*\*, *p* \< 0.01, and \*\*\*, *p* \< 0.001 versus Saline by non-parametric ANOVA with Dunn's post-tests. Note that results of \> 400 s mean that no clotting was detected at the end of the observation period Discussion {#Sec19} ========== Our initial preferred strategy in this study was to modify the previously described fusion protein HSACHV3 as little as possible, simply substituting a FXIa cleavage site for the plasmin cleavage site GSGIYR- that we had previously positioned between HSA and HV3 \[[@CR14]\]. This geometry would have recreated the hypothetical situation whereby HSACHV3, encountering an activating enzyme in the local environment of a thrombus, released HV3 not only from inhibition by the polypeptide blocking its N-terminus, HSAC, but also physically released HV3 from union with albumin, restoring rapid clearance of the small protein. The only difference would have been substitution of FXIa for plasmin as the putative thrombus-localized activator. However, we found that positioning the EPR cleavage site between HSA and HV3 in fusion protein HSAEPR-HV3 prevented its recognition by FXIa. In contrast, the alternative geometry, represented by fusion protein EPR-HV3HSA, was recognized by FXIa, at least to the extent that detectable antithrombin activity could be liberated in vitro. For in vivo applications, however, we were aware that EPR-HV3HSA activation would liberate HV3HSA, a constitutively active hirudin-albumin fusion protein that would perpetuate or conceivably worsen the narrow therapeutic window of hirudin \[[@CR31]\]. Our ability to activate EPR-HV3HSA was entirely consistent with the findings of Zhang et al., who reported that hirudin variant 2 (HV2) activity was liberated by FXIa cleavage of recombinant protein EH, in which EPR was positioned on the N-terminus of HV2 \[[@CR15]\]. These investigators also noted no activation of EH by FXa or thrombin; we extended this finding to include a lack of reactivity with plasmin, FVIIa, FXa, or FXIIa, supporting the specificity of the EPR hirudin "switch". EPR is nevertheless a very small blocking group, comprising a mere three amino acid residues highly exposed on the N-terminus of both EPR-HV3HSA and EH. To provide additional insulation of the HV3 domain from plasma, we produced fusion protein mycEPR-HV3HSA, expanding the blocking group to 13 amino acids. However, this protein was recognized less effectively by FXIa than EPR-HV3HSA; under conditions in which detectable antithrombin activity was liberated from EPR-HV3HSA in 20 min, 200 min of digestion was required to liberate similar amounts of functional HV3HSA from mycEPR-HV3. We cannot exclude the possibility that another blocking group of the same size, perhaps one with a lesser density of negatively charged residues than the myc epitope (sequence EQKLISEEDL) might be better recognized by FXIa. The FXIa-dependent activation of HV3HSA that we observed in vitro was similar to that we previously observed for HSACHV3 by plasmin \[[@CR14]\] in that high nM concentrations and minutes of incubation were required to generate measurable release of active HV3HSA or HV3. Both for plasmin and for factor XIa, it is difficult to predict what local concentrations might be generated in a thrombus. Free plasmin is generally not detectable in plasma, in part due to the efficiency with which it is incorporated into plasmin-antiplasmin complexes \[[@CR32]\]; and free FXIa has been reported to be present in picomolar concentrations in the plasma of healthy individuals \[[@CR33]\]. Plasma concentration data is not necessarily predictive of local concentrations that could be generated in a thrombus. In the case of HSACHV3, it was possible to show a favourable profile of antithrombotic activity with minimal bleeding risk when in vitro studies were extended into murine in vivo studies \[[@CR19]\]; given this precedent, we decided to examine EPR-HV3HSA activities in vivo, even though the plasmin precursor plasminogen is present in approximately 20-fold excess over factor XI in plasma \[[@CR34]\]. Fusion protein EPR-HV3HSA showed antithrombotic activity in two models of thrombosis in mice, one dependent on ferric chloride, and one independent of this agent. Ferric chloride, although widely used as a prothrombotic treatment, is a potent antioxidant with pleiotropic effects, and it has been suggested that its use should be accompanied in rigorous experimental studies with data from another model \[[@CR35]\]. EPR-HV3HSA, at 120 mg/kg or 40 mg/kg, prolonged the TTO in the ferric chloride-treated carotid artery, while mycHV3HSA had no greater effect than saline vehicle. These results are consistent with the findings of Zhang et al. \[[@CR15]\], who noted a significant increase in the TTO in a rat carotid artery model in which thrombosis was initiated by application of an electrical current, and animals were treated with 4 mg/kg body weight EH, an equimolar dose to the larger EPR-HV3HSA protein at 40 mg/kg. In the LPS/L-NAME model, in which more physiologically relevant thrombotic triggers of inflammation and nitric oxide antagonism are employed, EPR-HV3HSA reduced fibrin deposition in the heart and kidneys. Interestingly, a direct FXIa inhibitor, KPI, also administered as an albumin fusion protein at an identical dose by weight, was without effect. This finding suggests that although FXIa is generated in the LPS/L-NAME model, its inhibition is insufficient to prevent thrombosis, while its use to trigger HV3 liberation and thrombin inhibition was effective. A possible explanation of this effect is that the affinity of KPI for FXIa is much less than that of HV3 for thrombin, with binding constants for the former reported in the nM range \[[@CR36]\] and for the latter interaction in the sub-pM range \[[@CR31]\], a difference of more than three orders of magnitude. While EPR-HV3HSA reduced intra-organ fibrin deposition in the LPS/L-NAME setting, it did not eliminate it; in this regard it was less effective than recombinant activated protein C in our previous use of this model \[[@CR23]\]. EPR-HV3HSA clearly promoted bleeding to a lesser extent than constitutively active HV3HSA, as shown by significantly lower blood losses than those elicited by that protein in both liver laceration and tail transection models. However, mean blood losses were greater in EPR-HV3HSA-treated mice than in mycHV3HSA-treated mice in both models, although the differences did not reach statistical significance. Bleeding times in the tail transection model were significantly lower in mycHV3HSA-treated mice than in HV3HSA- or EPR-HV3HSA-treated mice; therefore this bleeding parameter was not reduced versus constitutively active HV3HSA. In our previous studies with HSACHV3, blood losses were more convincingly reduced to baseline in both bleeding models than in the current study with EPR-HV3HSA \[[@CR19]\]. These observations could reflect the relatively small number of mice that we tested, or they could indicate that FXIa was a less appropriate trigger for HV3 liberation than plasmin, or they could point to instability of the small EPR blocking group in plasma. To distinguish among these possibilities, we examined the stability of EPR-HV3HSA and, to a lesser extent, that of HSACHV3 in plasma. At a fusion protein concentration equivalent to the peak concentration expected in the murine circulation following intravenous injection of 40 mg/kg body weight doses, HV3HSA eliminated the ability of human or murine plasma to clot in the PT assay, while HSACHV3 had no effect in murine plasma and a modest, less than 2-fold prolongation in human plasma. In contrast, EPR-HV3HSA, after only two minutes' exposure to chelated plasma, prolonged the PT by five-fold in human plasma, and four-fold in mouse plasma. Prolonging the incubation in plasma increased the PT, but loss of clotting factor activities over time would also be expected to contribute to such an effect. Comparison to a calibration curve with HV3HSA suggested that approximately 10 -- 15% of EPR-HV3HSA was behaving as if its EPR blocking group had been removed, in both species' plasma. The trivial explanation that proteolysis had taken place during purification of EPR-HV3HSA could be eliminated because EPR-HV3HSA had no antithrombin activity without exposure to purified FXIa or plasma. While we cannot eliminate all proteases specifically involved in coagulation or fibrinolysis in this unexpected activation, its detection in chelated plasma renders their involvement unlikely. Moreover, unexpected activation of EPR-HV3HSA was also noted in plasma depleted of Factor VIII, Factor IX, Factor XI, or Factor XII, and EPR-HV3HSA also showed no sensitivity to purified FVIIa or plasmin. A more likely explanation for these findings is that EPR-HV3HSA, at μM concentrations, lost its EPR blocking group through the action of plasma endopeptidases or exopeptidases unrelated to coagulation or fibrinolysis. Proteomic studies of human plasma have shown that 28% of 722 unique N-terminal blood protein peptides had characteristics consistent with their having arisen through aminopeptidase action, such as laddering \[[@CR37]\]. In future experiments, this problem could conceivably be avoided by rendering EPR more accessible to FXIa in an internal location in a fusion protein like HSAEPR-HV3, either by separating EPR from HSA structures using an optimized polyglycine linker, or by using a different FXIa cleavage site (like TSKTLR in coagulation factor IX \[[@CR38]\]), or both. Conclusions {#Sec20} =========== Fusion protein EPR-HV3HSA's latent antithrombin activity was released by FXIa in vitro to a greater extent than mycEPR-HV3HSA; moving the EPR FXIa cleavage site to the C-terminus of HSA, in HSAEPR-HV3, abrogated FXIa recognition. EPR-HV3HSA was recognized and activated in vivo in mice, as shown by its antithrombotic activity in carotid artery and intra-organ thrombosis models, and the lack of activity of noncleavable mycHV3HSA. EPR-HV3HSA elicited less blood loss than constitutively active HV3HSA in lacerated liver and transected tail mouse models, but failed to attenuate the elevated bleeding time associated with HV3HSA administration. The partial overlap in bleeding promotion with HV3HSA is likely due to the partial instability of EPR-HV3HSA in murine and human plasma. Activation of EPR-HV3HSA in plasma was not due to any of seven tested coagulation or fibrinolytic proteases, but may have arisen due to the action of plasma exopeptidases on the N-terminally exposed EPR tripeptide. Only modest activation of plasmin-activatable HSACHV3 was observed. Although our results supported the use of FXIa as a thrombosis-specific fusion protein activator, N-terminally positioned EPR is a suboptimal biochemical "switch" compared to the internal plasmin-dependent switch in HSACHV3. The authors wish to thank Sharon Gataiance for technical assistance, and Dr. David A. Donkor and Dr. Syed M. Qadri for helpful discussions. Funding {#FPar1} ======= This work was made possible by Grant-In-Aid G-15-0009117 from the Heart and Stroke Foundation of Canada. WPS and VB are members of the Centre for Innovation of Canadian Blood Services, which receives funding from Health Canada. Accordingly, this article must contain the obligatory statement, and "The views expressed herein do not necessarily represent the views of the federal government \[of Canada\]". No funding body had any role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. Availability of data and materials {#FPar2} ================================== Any datasets used or analyzed for the current study which are not included in this published article are available from the corresponding author on reasonable request. Author's contributions {#FPar3} ====================== WPS designed the study, directed experiments, analyzed data, and wrote and revised the manuscript. LJE-S and VB performed all experiments and analyzed data. All authors edited and revised the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar4} ========================================== All experiments with mice in this study were carried out under the terms of an Animal Utilization Protocol approved by the Animal Research Ethics Board of the Faculty of Health Sciences of McMaster University (AUP 16-04-13). Consent for publication {#FPar5} ======================= Not applicable. Competing interests {#FPar6} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#S1} ============ Efficient sensory decision-making in a constantly changing environment requires neuronal circuits to be plastic and rapidly modify their activity. Cortical and subcortical processing are influenced substantially by feedback and neuromodulatory afferents eliciting experience-induced modulation of neuronal excitability lasting from milliseconds to hours ([@B31]; [@B10]; [@B52]; [@B58]; [@B1]). It has been suggested that simultaneous neuromodulation of neural circuits that process sensory, cognitive, and motor information is required to maintain neuronal dynamics for proper decision-making ([@B29]). Therefore, the brain region(s) acting as a modulator should innervate most if not all sensory processing regions. In addition, the brain region should exhibit the ability to modulate neuronal excitability dynamically allowing rapid context-dependent changes in information processing to elicit adequate behavioral outputs. The basal forebrain (BF) emerges as a good candidate to participate as an integrator and neuromodulator source for behavior since it is one of the most important and widely projecting neuromodulatory circuits in the mammalian brain ([@B26]) reaching the entire cortical mantle, hippocampus, and the olfactory system among others ([@B43]; [@B73]; [@B72]). Functionally, it has been linked with attention ([@B34]), arousal ([@B5]), and learning and memory ([@B20]; [@B34]). Specifically, its subnuclei have been proposed to play important roles in components of goal-directed behaviors such as motivational saliency ([@B38]), sensory discrimination ([@B38]; [@B16]; [@B50]; [@B53]; [@B17]), and cortical control ([@B52]). The wide array of neurophysiological and cognitive functions that the BF is involved in correlates with the neuronal complexity found in the region. Among the variety of neuronal types found in the BF ([@B74]), the cholinergic corticopetal projecting neurons have been extensively studied due to the important and dense top--down coordination role acetylcholine plays in cognitive functions such as attention. This idea arose from studies where pharmacological blockade or selective lesions of cholinergic neurons (CNs) in BF produced impairments in attention, memory, and operant conditioning performance ([@B70]; [@B13]; [@B46]; [@B34]; [@B16]; [@B52]; [@B42]; [@B14]; [@B55]). Moreover, in attention-demanding tasks, cholinergic release enhances cue detection and sensory discrimination ([@B59]). Here, we ask whether BF neurons are involved in the decision-making process in a non-cued olfactory-based self-initiated task. We addressed this question by recording the neural activity of the BF while the animals were freely engaged in a go/no--go task with voluntary trial start. Moreover, using optogenetic tagging, we identified CNs among the recorded units offline. Results {#S2} ======= The Firing Rate of Basal Forebrain Neurons Changes Before Initiation of the Trial {#S2.SS1} --------------------------------------------------------------------------------- To study the dynamics of recruitment of BF neurons in animals engaged in a self-initiated decision-making task, we implanted a multielectrode device in the horizontal diagonal band of Broca/magnocellular preoptic (HDB/MCPO) nuclei and proximity in the BF of trained adult mice ([Supplementary Figures S1A,B](#FS1){ref-type="supplementary-material"}). We recorded from HDB/MCPO because these are the only BF nuclei that send projections to the olfactory bulb and olfactory cortex ([@B73]). Animals were trained in a go/no--go olfactory discrimination associative learning task ([@B62]). This task studies the ability of a thirsty rodent to lick to obtain water in response to a rewarded conditioned stimulus (CS^+^) and refrain from licking in response to an unrewarded stimulus (CS^--^) ([Figure 1A](#F1){ref-type="fig"}). The CS^+^ and CS^--^ were odors randomly chosen from an odor set known to elicit neuronal response in the olfactory system ([@B18]) (see section "Materials and Methods"). Each trial is self-initiated 1--1.5 s after the computer detects the mouse entering the odor port. {#F1} We recorded neuronal activity during 200 trial sessions in animals proficient in differentiating between the two odorants (percent of correct responses, ≥80%). We found that single units responded with increases or decreases in firing rate (FR) during trial initiation (tstart). [Figure 1B](#F1){ref-type="fig"} shows examples of scatterplots of spike firing and peristimulus histograms (PSTHs) for two single BF units aligned to trial initiation (tstart) (top, increase in FR; bottom, decrease in FR). We analyzed the time course of FR changes aligned to tstart in 153 units total. A unit was categorized as responsive if the changes in FR assessed for 1 s after tstart were statistically significantly different from the basal FR assessed 1.3--0.3 s before tstart tested with a paired *t* test corrected for multiple comparisons using the false discovery rate (FDR) ([@B12]). We choose this time range to calculate the basal FR, since there appeared to be a change in FR just before the animal entered the port (see below). We found that a substantial fraction of BF neurons (53 out of 153 or 34.6%) exhibited significant increases (43 out of 153 or 28.1%) or decreases (10 out of 153 or 6.5%) in FR when the animal initialized a trial (*p* \< *p*FDR, *p*FDR, the FDR critical significance level per animal, ranged from 0.006 to 0.03; *n* = 153 units, 141 single units and 12 multiunits from 8 mice and 10 sessions). [Figure 1C](#F1){ref-type="fig"} shows on the left side a heatmap illustrating the FR time course for the 53 units that were significantly responsive and on the right side the time when the FR changed by 2 × SD above or below basal FR. Interestingly, for most of the units, the change in FR took place before the animal entered the odor port (mean onset time −260 ms with a 95% bootstrapped confidence interval ranging from −168 to −350 ms) suggesting that BF neurons are involved in behavioral functions associated with trial preparation and anticipation. We found that the number of BF neurons that exhibited a significant change in FR at the start of correct response trials (hits, 36 out of 149 units, −29 or 19.5% increase and 7 or 4.7% decrease their FR; and correct rejections, CR, 38 out of 150 total units recorded on those trials, −30 or 20% increase and 8 or 5.4% decrease their FR) is larger than the number of responsive units in false alarm trials (FA, licking in response to the CS^--^; 4 of 67 total units recorded during FA trials were responsive, 1 unit or 1.5% increase and 3 or 4.4% decrease their FR, [Figures 1D,E](#F1){ref-type="fig"}). A chi-squared test indicated that the difference in the number of responsive units between FA trials and correct trials is significant (*p* \< *p*FDR = 0.05). This shows that engagement of BF neurons during the precue epoch reflects the behavioral outcome of the trial, suggesting that activity of these neurons may play a role in successful discrimination. To further test the relationship between BF neural activity at trial initiation and behavioral outcome, we asked whether the change in BF neuronal activity is affected by engagement in sensory discrimination of rewarded vs. unrewarded odorants. We recorded neuronal activity of animals trained in a go/go task where the mouse is rewarded randomly for 70% of the trials regardless of the identity of the odorant. The key difference with go/no--go is that, in the go/go task, both odorants are rewarded and no sensory discrimination is required to receive the reward. We found that the number of units responsive at port entry was significantly lower in the go/go task compared to the go/no--go task ([Figure 1Eii](#F1){ref-type="fig"} and [Supplementary Figure S2](#FS2){ref-type="supplementary-material"}, chi-squared test *p* \< *p*FDR = 0.05, go/no--go = 53 responsive out of 153 or 34.6%), go/go = 8 responsive out of 44 or 18.2%, all of which increased their FR, suggesting that BF neuronal recruitment before trial initiation may play a role in adequate stimulus discrimination. In the go/go task, the units that responded to trial initialization also exhibited a change in FR before the animal entered the port (mean onset time −175 ms with a 95% bootstrapped confidence interval ranging from 167 ms to −517 ms). We also compared the change in FR after the start of the trial between units that responded to correct and incorrect trials. [Figure 1F](#F1){ref-type="fig"} shows the relationship between change in FR for incorrect trials and the change in FR for correct trials. The data are fit with a line with a slope significantly smaller than one, suggesting that engagement of BF neurons reflects correct behavioral performance ([Figure 1F](#F1){ref-type="fig"}; ANCOVA, *F* = 16.6, Tukey's *post hoc*, *p* = 0.0004, *n* = 19). To further determine whether recruiting BF neurons during trial initiation relates to animal behavior, we calculated *d*′, defined as the difference in the change in FR upon trial initiation normalized by the standard deviation of basal FR (*d*′) per trial. [Figure 1Gi](#F1){ref-type="fig"} shows no difference for *d*′ for hit vs. CR for the go/no--go task ([Figure 1G](#F1){ref-type="fig"}, KS test *p* = 0.11, *n* = 6,289 trials for hit and 6,649 responsive units for CR). In contrast, [Figure 1Gii](#F1){ref-type="fig"} shows a significant difference in *d*′ curves of hit trials in the go/no--go task vs. hit trials in the go/go task (KS test *p* = 3.6 × 10^--8^, *n* = 1,266 trials units for go/go). These data indicate that neurons responded similarly during correct responses that required sensory discrimination but responded differently when no discrimination was required. To study how effective the change in FR during trial initiation is at classifying correct vs. incorrect behavioral outcome, we used the receiver operant characteristic (ROC) analysis ([@B21]) and measured the area under the curve (AUC) for each unit. The higher the AUC (maximum 1), the better the unit differentiates between correct and incorrect responses (AUC of 0.5 indicates no differentiation). We found that units in the go/no--go task were a more effective classifier than in the go/go scenario ([Figure 1Gii](#F1){ref-type="fig"}, *t* test *p* = 0.03, *n* = 153 for go/no--go and 44 for go/go), suggesting that BF neuron activity is related to adequate decision-making in sensory discrimination. Basal Forebrain Neurons Exhibit Changes in Firing Rate When Conditioned Stimulus or Reinforcement Are Delivered {#S2.SS2} --------------------------------------------------------------------------------------------------------------- To determine whether BF neuronal activity is recruited during other epochs of the behavioral trial in our experimental design, we aligned the normalized FR of the recorded units either to delivery of the conditioned stimulus (CS or odor) or the reward. [Figure 2](#F2){ref-type="fig"} shows changes in FR aligned to odor delivery ([Figure 2A](#F2){ref-type="fig"}) or reward delivery ([Figure 2B](#F2){ref-type="fig"}) for mice performing \>80% in the go/no--go task. As described in previous studies ([@B38]; [@B68]; [@B17]), we found that neurons either increase ([Figures 2Ai](#F2){ref-type="fig"}, [iii](#F2){ref-type="fig"}; 12/153 or 7.8%) or decrease ([Figures 2Aii](#F2){ref-type="fig"}, [iii](#F2){ref-type="fig"}, 31/153 or 20.3%), their FR in response to the stimulus. Specifically, 21 out of 150 units (14%) recorded were recruited during CS^+^ delivery and 45 out of 150 (30%) during CS^--^ (*t* test *p* \< *p*FDR, *p*FDR per session ranged from 0.02 to 0.01, 0.003 to 0.04, 0.02 to 0.01 for odor, CS^--^ and CS^+^ and reward, respectively). As shown in the heat map of [Figure 2Aiii](#F2){ref-type="fig"}, some units responded transiently, while others responded with slow sustained changes in FR. {#F2} In addition, units were also recruited during water (reward) delivery (22/148 or 14.9% increase and 8/148 or 5.4% decrease their FR, [Figure 2B](#F2){ref-type="fig"}, *t* test *p* \< *p*FDR, *p*FDR per session ranged from 0.003 to 0.03). Units that responded to reward in this self-initiated task also tended to respond to the conditioned stimulus as described by [@B38] in rodents trained in a cue-oriented task ([Supplementary Figure S2D](#FS2){ref-type="supplementary-material"}). Supporting the idea that the BF could play a role in stimulus discrimination and reward association, the percent of units that responded during odor or reward delivery in the go/go task (when regardless of the odorant 70% of the trials are rewarded) are significantly smaller than the responses in the go/no--go task (28.1% go/no--go vs. 9.1% go/go and 20.3% go/no--go vs. 4.8% go/go during stimulus presentation and water delivery, respectively, [Supplementary Figures S2B,C](#FS2){ref-type="supplementary-material"}; chi-squared test, *p* \< *p*FDR = 0.05, two pairwise comparisons). Basal Forebrain Neurons Become More Responsive to the Stimulus as the Animal Learns to Differentiate Odorants in the Go/No--Go Task {#S2.SS3} ----------------------------------------------------------------------------------------------------------------------------------- Our data suggest that neurons from the BF are required for adequate decision-making and stimulus discrimination in proficient animals, raising the question whether the number of neurons coding for information during the different epochs of the behavioral trial increased as the animal learned to discriminate between rewarded and non-rewarded odors. We compared the change in FR during the different epochs of the trial when the animal was learning to discriminate (= 50% correct trials) and when the animal was proficient in their response to the rewarded odorant (\>80% correct trials). A representative learning curve is shown in [Figure 2C](#F2){ref-type="fig"}. It starts with 50% correct responses, while the mouse gradually becomes proficient until reaching criteria (\>80% correct responses, hits and CRs) within a session. We observe that the number of BF-responsive neurons during trial initialization does not increase as the animals learn to associate the stimulus with the reward ([Figure 2D](#F2){ref-type="fig"}, chi-squared test *p* = 0.37 \> *p*FDR = 0.016, 10 out of 27 units or 37% were responsive during learning and 21 out of 44 or 47.7% were responsive when the animal was proficient). These units are likely engaged during the instrumental shaping of the task that occurs before animals are trained in the go/no--go task and could reflect the motivation and initial attention required to start the trial. After that initial training, animals are trained to lick in response only to the rewarded stimulus that has no hedonic value at the beginning of the session. The number of responsive units when FR is aligned to odorant onset increased dramatically with learning ([Figure 2D](#F2){ref-type="fig"}, 7.4 vs. 38.6%; chi-squared test, *p* = 0.004 \< *p*FDR = 0.016). In contrast, for the reward epoch, we found no statistical difference between the number of responsive neurons before and after the animal became proficient (7.6 vs. 14.6%; chi-squared test, *p* = 0.39 \> *p*FDR = 0.016). Taken together, our results suggest that BF neurons play a role in actively engaging the animal in the task (trial start epoch), in correct odorant discrimination (odor epoch), and responding to the reward (reward epoch) and that learning increases the number of neurons engaged in the odor epoch. A Subset of the Basal Forebrain Cholinergic Neurons That Are Responsive During Trial Initiation or Conditioned Stimulus Epochs Are Cholinergic {#S2.SS4} ---------------------------------------------------------------------------------------------------------------------------------------------- The neuronal makeup of the BF is heterogeneous with glutamatergic, GABAergic, and cholinergic projection neurons, among others ([@B19]; [@B72]). The cholinergics have granted particular attention since they project to the whole cortical mantle and actively participate in cortical plasticity ([@B11]) and sensory processing ([@B39]; [@B53]). This motivated us to determine whether BF CNs were responsive in any epochs of our self-initiated task in proficient animals. To identify CNs, we used optogenetic tagging ([@B30]). We used mice expressing ChR2-EYFP under control of the choline acetyl transferase (ChAT) promoter. Once the behavioral session concluded, we delivered light stimulation (10 trials of 10 50-ms pulses at 5 Hz) through the optic fiber of the optetrode implanted in the BF of mice expressing ChR2-EYFP selectively in CNs (ChR2^+^ neurons, [Figure 3A](#F3){ref-type="fig"}, see section "Materials and Methods"). The pronounced increase in spiking frequency in a subset of units was not observed in control ChAT-Cre animals ([Figure 3B](#F3){ref-type="fig"}) regardless of the frequency of stimulation (1 or 5 Hz), discarding the possibility that light delivery could generate false spikes due to thermal stimulation ([@B28]). {#F3} The BF, however, exhibits intricate local circuitry with abundant cholinergic collaterals terminating in non-cholinergics ([@B19]), raising the possibility that light-responsive neurons *in vivo* might not express ChR2. We confirmed the local connectivity by performing *in vitro* whole-cell patch clamp recordings in acute brain slices from the BF. In the voltage clamp mode, we found that brief light stimulation always elicited inward currents in ChR2^+^ neurons ([Figure 3C](#F3){ref-type="fig"}). In contrast, non-CNs (identified by their lack of ChR2-EYFP fluorescence, ChR2^--^ neurons) located in close proximity to a neuron expressing ChR2-EYFP (ChR2^+^) exhibited an array of responses after being transsynaptically activated by optogenetic activation of CNs. A small number of non-CNs (*n* = 1/10) exhibited an outward current after the cholinergic ChR2^+^ neurons were activated; some (*n* = 3/10) exhibited an inward current or a biphasic response (2/10), and most of them (*n* = 4/10) showed no change ([Figure 3C](#F3){ref-type="fig"}). To obtain information relevant to the correct identification of CNs *in vivo*, we studied the latency for light activation of cholinergic ChR2^+^ and non-cholinergic ChR2^--^ neurons through *in vitro* current clamp ([Figure 3D](#F3){ref-type="fig"}). We found that there was a clear and significant difference in latency of responses between these neurons (18.1 ± 3.5 ms, *n* = 3/9 responded for ChR2^--^ vs. 4.1 ± 0.4 ms, *n* = 7/7 for ChR2^+^, respectively, *t* test, *p* \< 0.001) allowing us to establish 10 ms as a cutoff for maximum latency for neurons that were directly activated by light ([Figure 3E](#F3){ref-type="fig"}), in accordance with [@B30]. Only three out of nine non-cholinergics exhibited action potential generation after activating neighboring ChR2^+^ neurons, while all nine ChR2^+^ neurons responded. *In vivo*, we found that 15 out of 186 units (from go/no--go and go/go tasks) exhibited latency \<10 ms ([Figure 3E](#F3){ref-type="fig"}). In addition, we used two other properties to classify a neuron as cholinergic: (1) it had to exhibit a statistically significant increase in FR after light stimulation in a paired *t* test with correction for multiple comparisons, and (2) it had to display a reliability of response of 100% (they had to spike within 200 ms after light stimulation in all 10 trials; [Figure 3F](#F3){ref-type="fig"} and [Supplementary Figures S3A--C](#FS3){ref-type="supplementary-material"}). CNs, despite their wide and critical role in brain function, are sparsely distributed and account for only 5% of the BF neurons ([@B26]). With our conservative criteria, we classified 6 out of 186 (3.2%) units as cholinergic in accordance with the numbers found in other studies ([@B38]; [@B30]). We found that three of these six optogenetically tagged CNs responded with a significant change in their FR when the mouse decided to enter the port (3/6 units or 50% of responsiveness, paired *t* test *p* \< *p*FDR = 0.025; mean onset latency of FR change, 100 ms ± 100) and when presented with the CS (67% of responsiveness, paired *t* test *p* \< *p*FDR = 0.033, [Figure 4](#F4){ref-type="fig"} and [Supplementary Figure S3D](#FS3){ref-type="supplementary-material"}). We did not find responses to reward in these six neurons ([Figure 4](#F4){ref-type="fig"} and [Supplementary Figure S3E](#FS3){ref-type="supplementary-material"}). Therefore, although CNs are sparse, yielding recording from a small number of units, the changes in FR are clear and consistent from trial to trial, indicating that these neurons are engaged in trial initiation and CS discrimination. {#F4} Discussion {#S3} ========== On the basis of *in vivo* electrophysiological recordings in freely moving and behaving animals, we demonstrated that neurons from the BF are engaged throughout the decision-making process in a goal-directed task. Transient changes in the activity of BF, specifically the HDB/MCPO and proximity, were found during trial initialization anticipating the stimulus, stimulus discrimination, and in reward association in the go/no--go odor discrimination task. Importantly, the number of units displaying changes in FR increased for the stimulus discrimination epochs as the animal learned to discriminate the odorants. Furthermore, the changes in FR were found to be related to correct outcome in the trial, and the number of units that displayed a change in FR decreased in a go/go task where animals receive reward regardless of the odorant, indicating that BF activity plays a role in correct outcome of the trial. Finally, through optogenetic tagging, we found that BF CNs are involved in this processes. Basal Forebrain and Anticipatory Activity {#S3.SS1} ----------------------------------------- The capacity of the brain to correctly respond to environmental cues has been linked in recent years to its ability to predict future outcomes. The anticipatory behavior has been described to improve performance not only by enhancing motor preparedness and reaction time but also by improving perception ([@B48]) and more efficiently processing the upcoming sensory input ([@B2]; [@B32]). Specifically, baseline rates of neurons in HDB/MCPO BF has been shown to be higher during the acquisition phase of an odor--reward association than during spontaneous investigation or the recall phase of an odor reward association ([@B17]). Furthermore, neurons in other nucleus of the BF, the nucleus basalis, responded before stimulus onset and continued for seconds after reward delivery in a whisker-dependent tactile discrimination two-alternative forced choice task ([@B68]). Interestingly, [@B68] observed that the anticipatory modulation in neuronal FR began ∼1 s before the onset of the mechanical deflection of the whiskers, similar to our results, where we observed anticipatory changes in FR of the BF before the animal enters the odor port. They hypothesized that neurons of the nucleus basalis participated in the circuit defining animal's expectations in the task ([@B68]). In addition, neuronal responses with onsets before the first lick were reported in the olfactory tubercle in a self-initiated water-motivated dry lick instrumental task ([@B23]) and an intermodal selective attention task ([@B6]). They found that neurons of the olfactory tubercle fired in anticipation of the expected reward probably to invigorate instrumental training in states of reduced motivation. Finally, in primates, the anticipatory activity of neurons in the caudate nucleus correlates with reward association, expectation, and response latency, probably reflecting the animal's motivational state ([@B35]; [@B71]). In conclusion, there is evidence of anticipatory neuronal activity in different brain regions, suggesting that neuronal activity linked to expectation might play an important role in behavior. Interestingly, we also found that this anticipatory activity was correlated with behavioral performance, supporting an additional role of early neuronal activity in attaining correct stimulus discrimination. This idea follows the line of evidence suggesting that top--down modulation might be as important as the external stimulus information in sensory processing and perception in the visual system ([@B25]) or other sensory-motor modalities, like the tongue--jaw motor cortex, which anticipatory prestimulus activity can be predictive of licking direction in a somatosensory detection task ([@B45]). A recent article found that inhibition of the neuronal activity of the BF using Designer Receptors Exclusively Activated by Designer Drugs (DREADDs) interrupted the ability of rats in increasing their discrimination accuracy in a sustained attention task in response to a high reward probability trials ([@B67]), further suggesting that the BF could play a role in sensory discrimination. In humans, electroencephalogram (EEG)/magnetoencephalogram (MEG) studies have suggested that anticipatory attention could promote desynchronization of oscillatory brain activity ([@B2]; [@B56]), which would enhance perception. Future studies with electrophysiology or imaging of neural activity imaging are necessary to determine the role of neuronal dynamics of the BF in sensory discrimination during reward expectation. Cholinergic Neurons and Anticipatory Activity {#S3.SS2} --------------------------------------------- As mentioned before, the neuronal population of the BF is an intricate heterogeneous network with glutamatergic, GABAergic, and cholinergic projection neurons, among others ([@B19]; [@B72]). Using optogenetic tagging, we identified CNs from our recorded units and found that the activity of CNs were also engaged and modulated during trial initialization, which could participate in the preparation of the decision-making process. The role of slow changes in ACh concentration, in time scales from minutes to hours, is well established based on the finding that CNs are recruited during arousal ([@B5]) and that ACh is slowly released and diffuses through the cortical mantle ([@B47]). However, recent evidence suggests that fast transient changes (milliseconds to seconds) in ACh may regulate neuronal processes affecting decision-making and behavioral performance in instrumentally cued tasks ([@B51]; [@B38]; [@B53]; [@B47]; [@B30]; [@B27]). Here, we found that, in a more naturalistic scenario, where the cholinergic system is not permanently engaged, such as a self-initiated (not-instrumentally cued) behavior in freely moving animals, CNs are also transiently engaged. Our data agree with precue changes in cholinergic release that had been directly measured in the prefrontal cortex using choline-sensitive electrode at the millisecond scale, changes that had been directly correlated with sensory cue detection ([@B51]). Hence, BF with cholinergic and non-cholinergic-projecting neurons might be an important region to participate in anticipatory behavior and improve animal performance. Basal Forebrain and Stimulus Discrimination {#S3.SS3} ------------------------------------------- In addition to the stimulus anticipatory response, we found that BF neurons (including cholinergics) changed their activity when the CS or odor was presented. Afferents from the HDB/MCPO project to the whole olfactory system ([@B72]) and has been proposed that GABAergic and CNs are required for proper stimulus discrimination. GABA released from BF projecting neurons into the olfactory bulb (the first brain region involved in the processing of olfactory information) is required to discriminate between similar olfactory cues, in part by inhibiting local inhibitory neurons in the bulb ([@B50]). On the other hand it has also been proposed through *in vitro* and *in vivo* electrophysiology that ACh is required for olfactory sensory discrimination and odor memory formation ([@B22]; [@B7]; [@B63]). At the circuitry level, it is believed that acetylcholine regulates olfactory information processing by sharpening the olfactory receptive fields of the output neuron of the olfactory bulb ([@B8]; [@B44]) and increasing their firing frequency ([@B57]). At the level of the olfactory cortex, acetylcholine has been implicated in increasing pattern separation ([@B7]) and increasing synchronization in the neuronal output of the bulb, which could lead to a more robust and stable learned olfactory representations in the olfactory cortex ([@B15]). Supporting this idea, we found that when animals were trained in a go/go task, it engaged significantly lower BF neurons during the start of the trial and CS presentation. In other brain regions, cortically implanted choline-sensitive electrode recording in animals performing instrument-initiated detection of a light cue demonstrated that cholinergic neurotransmission is regulated with transient increase within seconds following cue detection superimposed over slower changes in cholinergic activity ([@B51]). These transients are thought to be required for proper cue detection and behavioral output ([@B27]). For instance, optogenetic regulation of BF CNs elicits fast modulation of neuronal activity in visual cortex, enhancing perception in mice responding to grating orientation ([@B53]). Interestingly, in a cue-initiated auditory detection task, optogenetically identified CNs in the BF responded with changes in neuronal activity a few ms after receipt of reward or punishment ([@B30]) and not to any other epoch of the behavioral trial, such as stimulus discrimination. Therefore, depending on the behavioral context, there appears to be differences in the dynamics of cholinergic release. Finally, we found that a substantial number of non-cholinergic BF neurons, but not CNs, responded to water reinforcement. Our finding is consistent with a study that showed that primary reinforcement elicited robust bursting in non-CNs in a go/no--go task initiated by a tone where the animals were freely moving ([@B38]). In summary, we found that in a self-initiated task, BF cholinergic and non-CNs play a role in decision-making and stimulus discrimination. The behavioral response is in part correlated with BF anticipatory precue activity, which opens new targets and time windows to modulate attention. Finally, our data position the BF as a potential information integrator and a common neuronal pathway to elicit a context-adequate behavioral response. Speculation on the Role of Basal Forebrain Modulation on Selective Attention in Olfaction and Vision {#S3.SS4} ---------------------------------------------------------------------------------------------------- What is the role of BF neuron modulation of early sensory processing in the olfactory and visual systems? In the visual system, optogenetic activation of BF CNs increases behavioral performance for mice engaged in the discrimination of vertical vs. horizontal drifting gratings ([@B53]). Interestingly, cholinergic BF stimulation decreases neuronal synchronized of low-frequency oscillations (1--5 Hz) and increases the power of high-frequency gamma oscillations (60--100 Hz). In the olfactory system, chemogenetic inhibition of GABAergic BF modulation of granule cells in the olfactory bulb produced a reversible impairment in the discrimination of structurally similar odors ([@B50]). Optogenetic stimulation of GABAergic BF inputs to olfactory bulb granule cells produces reliable inhibition of these interneurons ([@B50]) that are key in generating gamma frequency oscillations generating synchronized gamma bursts that efficiently stimulate piriform cortex recurrent circuits that transmit the olfactory information in concentration-invariant odor coding ([@B60]; [@B54]; [@B3]). The regulation of gamma oscillations by BF input in these sensory systems raises the question whether BF regulates transmission of information through phase amplitude coupling (PAC) mediating selective attention to specific stimuli ([@B49]). Phase amplitude coupling is defined as gamma bursts of information firing at specific phases of low-frequency theta oscillations (4--12 Hz) ([@B64]; [@B40]; [@B9]). Theta are the most global oscillations in the brain that act as a timekeeper ([@B61]) and are coherent across numerous cortical and subcortical structures arguing for its role in transfer of discrete chunks of information ([@B4]; [@B69]). In the olfactory bulb, contextual odorant identity (is the odorant rewarded?) can be decoded from peak theta-phase referenced power of gamma oscillations in animals proficient in odorant discrimination in the go/no--go task but not in mice that have not learned to discriminate the odorants ([@B41]) arguing for selective attention filtering of information on relevant stimuli through PAC. In the visual system of the macaque monkey, the strength of theta and of theta-rhythmic gamma modulation was markedly reduced by selective attention-altering information transfer through PAC ([@B65]). The engagement of changes in BF activity in different epochs of trials in associative learning tasks shown in this and other studies ([@B38]; [@B53]; [@B30]; [@B17]; [@B23]), and the fact that BF activity modulates oscillatory activity in olfactory bulb ([@B50]) and visual cortex ([@B53]), raises the question whether BF modulates selective attention within sensory systems or intermodally ([@B49]) through modulation of PAC. Whether this is the case requires future studies in the visual and olfactory system of BF regulation of PAC, stimulus decoding by phase-referenced power, and changes in behavioral accuracy by alteration of PAC by modulation of BF activity. Materials and Methods {#S4} ===================== Animals {#S4.SS1} ------- All procedures and experiments were approved by the Institutional Care and Use Committee at the University of Colorado Anschutz Medical Campus in accordance with NIH standards. We used 2- to 6-month-old mice from the Jackson Laboratories bred in-house. Mice were kept with water and food *ad libitum* in a reversed 12 h light cycle, except that, when they were trained for awake behaving recording, they were water restricted (below). To selectively express ChR2 in CNs, we used ChAT-EYFP-ChR2 mice generated by crossing ChAT-Cre mice \[B6;129S6-*Chat*^*TM*\ 2(cre)Lowl^/J, [RRID:IMSR_JAX:006410](https://scicrunch.org/resolver/RRID:IMSR_JAX:006410)\] with Rosa26-floxed-ChR2-EYFP animals \[B6;129S-*Gt(ROSA) 26Sor*^*TM*32(CAG--COP4\*H134R/EYFP)Hze^/J, [RRID:IMSR_JAX:012569](https://scicrunch.org/resolver/RRID:IMSR_JAX:012569)\]. The generated mouse selectively expresses channelrhodopsin 2 (ChR2) under the control of the ChAT promoter. Optetrode Building {#S4.SS2} ------------------ Optetrodes were built as previously shown with custom modifications described in [@B36]. Briefly, four tetrodes consisting of four polymide-coated nichrome wires (diameter, 12.5 μm; Sandvik) were connected to a 16-channel interface board (EIB-16, Neuralynx) and fed through a housing glued to the board. An optic fiber (105 μm diameter, Thor Labs) was also fed through the housing, and the tetrode tips were glued maximizing the distance between them to the end of the bare fiber. Immediately before implantation, the tetrodes were gold plated to an impedance of 200--350 MΩ. Surgery {#S4.SS3} ------- Adult mice were anesthetized with an intraperitoneal injection of ketamine (100 mg/kg) and xylazine (10 mg/kg). Mice were implanted in the BF at coordinates of anterior--posterior (AP) of 0.02 mm and medial--lateral (ML) of −1.625 mm, or AP of 1 mm and ML of −1.500 mm with respect to bregma. On the day of the surgery, the optetrode was implanted 200 μm above the final location, and every day, it was lowered to 50 μm until reaching a final depth of dorsal--ventral (DV) of 5 and 4.9 mm, respectively. A screw was also implanted in the skull in the opposite hemisphere (1 mm right and 2 mm posterior of bregma) to serve as ground connector. Light was delivered through the fiber, and recordings were made in order to verify neuronal light responses. The animals were allowed to recover at least 1 week before experiments were performed. Implant location was corroborated through CT scan imaging. Non-invasive Micro-CT Imaging {#S4.SS4} ----------------------------- All CT imaging protocols were developed at the Animal Imaging Shared Resources (AISR) supported by the University of Colorado Cancer Center (NCI P30CA046934) and the Colorado Clinical and Translational Sciences Institute (NIH/NCATS UL1TR001082). Mice were anesthetized with 2% isoflurane, placed on a warming pad, and inserted into a Siemens Inveon micro-CT scanner (Siemens Preclinical Solutions). A single 3-dimensional (3D) micro-CT image set was acquired for each mouse using Inveon Acquisition Workstation software (IRW v1.5) with the following parameters: 270\_ rotation; 240 rotation steps; charge-coupled device (CCD) readout of 2,304/2,048; 4 binnings for matrix size reduction; exposure time of 30 ms with 80 kV voltage and 450 mA current; with a field of view (FOV) of 30 mm. The 6-min acquisition with middle-to-high magnification resulted in effective isotropic resolution of 54 μm (after the Shepp--Logan reconstruction algorithm). Animals were monitored during recovery from the anesthesia and returned to their cages. The images were read with the RadiAnt DICOM Viewer 1.9.16, and measurements were made from the tip of the electrode to the dorsal, ventral, and medial aspect of the skull, taken in the coronal, sagittal, and horizontal view. With this measurements, the CT scan images were registered into an MRI atlas (AtlasView 1.0, Radiology Department Johns Hopkins University) and finally into the Paxinos Mouse Brain Atlas ([@B24]). Eight out of 14 animals with correct implant locations were considered in the study. Behavior {#S4.SS5} -------- We used instrumental conditioning in freely moving mice in the Slotnick olfactometer ([@B62]; [@B18]). Animals were trained in the go/no--go and go--go behavioral task as explained in detail in [@B18] and [@B37]. Briefly, thirsty animals were trained to discriminate between a rewarded (CS^+^) and unrewarded (CS^--^) odor. Each trial was freely initiated by the mouse entering the odor port and breaking a photodiode beam. Once the trial was started (tstart) 1--1.5 s later, the CS was presented for 2.5 s ([Figure 1A](#F1){ref-type="fig"}). After CS delivery, the animal had to stay in the odor port for at least 500 ms for a trial to be considered completed. If not, it was considered a premature exit, and the trial had to be started again. During CS presentation, the animal learned to lick onto the water port at least once in four 0.5-s segments in response to CS^+^ for a 10-μl water reward. They quickly learned to refrain from licking in response to CS^--^ since no water was rewarded. The animal's performance was evaluated in blocks (maximum of 10 blocks) of 20 trials (10 rewarded and 10 unrewarded, presented at random). Each block's percent correct value represents the percent of trials in which the odors were correctly discriminated and associated with the appropriate behavioral action. Each session included 4--10 blocks of 20 trials. Electrophysiological recordings of the segments where the animal reached criteria (80% of correct responses) were considered in this study. For the go--go task, mice were rewarded at random in 70% of the trials regardless of which of the two odors was presented. The odors used were isoamylacetate, phenylacetate, 2-butatnone, ethyl propionate, ethyl butyrate, and mineral oil, all diluted at 1% in mineral oil. Experiments were performed in the afternoon (1[--]{.smallcaps}5 [PM]{.smallcaps}) under the "light on" cycle. Electrophysiological Recordings and Spike Clustering {#S4.SS6} ---------------------------------------------------- The output of the tetrodes was connected to a 16-channel amplifier (A-M Systems 3500) through a 1× gain headstage (Tucker-Davis Technologies). The signal was amplified 1,000× and was recorded digitally at 24 kHz with a Data Translation DT3010 A/D card in a PC computer controlled with a custom MATLAB (Mathworks) program. Behavioral epochs or events (tstart, CS presentation, water delivery) were also recorded by the A/D board in real time. The spike clustering method was explained in detail in [@B37] Briefly, data were filtered digitally between 300 and 3,000 Hz. With custom-written MATLAB programs, each of the 16 channels was thresholded at three times the standard deviation of the mean. Every spike with amplitude bigger than the threshold was imported into a second program (1 ms record per spike) that performed superparamagnetic clustering and wavelet decomposition of the spikes using 13 different wavelets and three principal components for the analysis and previously described ([@B18]; [@B37]). A single unit was defined as a unit with a refractory period of 1 ms ([@B33]; [@B66]) and a violation \<2% in the inter spike interval (ISI). Data for multi- and single units were used for analysis. For the go/no--go task, we found that out of 156 total units, 141 were single units. In the case of the go--go task, we registered from 1 multiunit and 43 single units. Identified CNs were all single units ([Supplementary Figure S1](#FS1){ref-type="supplementary-material"}). Confocal Imaging {#S4.SS7} ---------------- To visualize ChR2-EYFP expression, mice were intracardially perfused with 4% paraformaldehyde and the brains postfixed overnight in the same fixative at 4°C. Thereafter, the brains were placed in a sucrose solution \[30% in phosphate-buffered saline (PBS)\] until they sank in the solution. Subsequently, they were frozen in dry ice and stored at −80°C. The brains were sliced at 40 μm in a cryostat, mounted on slides, and visualized with a Leica SP5 X confocal microscope. Delivery of Light Stimulus {#S4.SS8} -------------------------- A light pulse protocol was delivered to ChAT-ChR2 mice after a successful behavioral session for optogenetic tagging of CNs. A 473-nm blue laser (Shanghai Laser) was used with a maximal power of 5.3 mW (66.3 mW/mm^2^) measured at the end of the fiber under steady illumination. In the same chamber where the behavior was performed, we delivered 10 pulses of 50-ms duration at a frequency of 5 Hz. The light delivery protocol was repeated 10 times, and only the first pulse on each trial was considered for analysis. Slice Preparation for *in vitro* Whole Cell {#S4.SS9} ------------------------------------------- Choline acetyl transferase-ChR2 mice (2--3 months old) were anesthetized by CO~2~ inhalation and decapitated. Brains were quickly removed and placed in ice-cold oxygenated sucrose slicing solution composed of (in mM): 234 sucrose, 11 glucose, 26 NaHCO~3~, 2.5 KCl, 1.25 NaH~2~PO~4~ 10, MgSO~4~, and 0.5 CaCl~2~ (equilibrated with 95% O~2~ and 5% CO~2~, pH 7.4). Coronal brain slices (300-μm thickness) were prepared using a Leica VT1200S vibratome (Leica Biosystems). Coronal slices were incubated in prewarmed (36°C), oxygenated artificial cerebrospinal fluid (ACSF; in mM): 126 NaCl, 26 NaHCO~3~, 10 glucose, 2.5 KCl, 1.25 NaH~2~PO~4~, 2 MgCl~2~, and 2 CaCl~2~ for at least 30 min before being transferred to the recording chamber, where they will be continuously perfused with ACSF (32°C). Whole Cell Recording {#S4.SS10} -------------------- Positive and negative ChAT-ChR2 neurons in the BF (HDB/MCPO) were visually identified by EYFP expression and differential interference contrast (DIC) on a modified Olympus upright microscope (Scientifca, East Sussex, United Kingdom). Whole cell recording was performed with a Multiclamp 700B amplifier (Molecular Devices Corp.), using recording pipettes with resistance of 3--5 MΩ pulled on a PC10 vertical puller (Narishige International) and filled with intracellular solution containing the following (in mM): 135 potassium gluconate, 20 KCl, 10 HEPES, 0.1 ethylene glycol tetraacetic acid (EGTA), 2 MgATP, and 0.3 NaGTP. Recordings were low-pass filtered at 4 kHz (Bessel filter) and digitized at 10 kHz (Digidata 1440) using pClamp 10.3 software (Molecular Devices Corp.). Series resistances were monitored throughout each voltage-clamp recording with 50 ms and −10 mV steps, and if it changed by \>20%, the data were discarded. Evoked synaptic responses were recorded from ChAT^+^ and ChAT^--^ neurons, and these responses were triggered by light stimulation directly onto the BF area. Light stimulation was evoked by a single mercury-free LED illumination system (CoolED pE-100 series) at 470 nm for 5 ms between 1 and 5% of the maximal intensity of the system. Latencies of evoked responses were analyzed using prewritten code routines in Axograph-X. Data Analysis {#S4.SS11} ------------- To determine the responsiveness of the units to the different events, we aligned all trials to the starting point of the event and calculated the average FR (in Hz). We performed a paired *t* test between the FR 1 s before and after the event and corrected the *p* value for multiple comparisons using the false discovery rate ([@B12]). To display the results, the FR was calculated in 0.1-s bins and normalized per unit to the mean FR, 1.2 s before the beginning of the event for 1 s. To calculate the first bin of responsiveness, we determined the first bin that exhibited a change in normalized FR (either increase or decrease) two times above or below the standard deviation of the mean with a sliding window of six bins used as a baseline. To determine the latency of the response *in vitro* in current clamp mode, the mean latency between the beginning of the light pulse to the peak of the voltage change was measured for 15 trials. For *in vivo* recordings, the mean latency was calculated for 10 trials and defined as the time a spike was detected after the light stimulation and before 200 ms (were another light pulse was given). To identify CNs, the recordings obtained during the behavior and light delivery were processed in batch, and the same units were identified in both recordings. We calculated the latency of the first spike after the first light pulse with a custom program written in MATLAB (Mathworks), with the average of 10 trials defined as the light latency for the unit. To calculate the reliability of the response, a spike had to occur at least once 200 ms before the light was applied, and for 100% reliability, it had to spike during that period of time on each of the 10 trials. To calculate the changes in FR, we calculated the peristimulus time histogram (PSTH) of the 10 trials and performed a pairwise *t* test between baseline (500 ms--1 s) and 30 ms postlight application. The *p* value obtained for all the units was corrected for multiple comparisons ([@B12]). Extracellular recording from the electrodes was used to calculate the local field potential (LFP) in the frequency range from 1 to 100 Hz. Time--frequency power decomposition of the LFP was obtained by means of MATLAB's spectrogram.m function with a 1-s window and 90% overlap. To compare LFP power between genotypes, we utilized Mann--Whitney *U* test with false discovery rate correction for multiple comparisons ([@B12]). Data Availability Statement {#S5} =========================== The datasets generated for this study are available on request to the corresponding author. Ethics Statement {#S6} ================ The animal study was reviewed and approved by the Institutional Animal Care and Use Committee University of Colorado Anschutz Medical Campus. Author Contributions {#S7} ==================== AN-P and DR designed the *in vivo* experiments. AN-P, CC-DR, and MH designed the *in vitro* experiments. AN-P performed the *in vivo* experiments. CC-DR performed the *in vitro* experiments. AN-P and DR performed the data analysis. All authors wrote the manuscript. Conflict of Interest {#conf1} ==================== 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 funded by the NIH grants R01 DC00566 (DR), DC008855 (DR), and NS095311 (MH), CONICYT REDES140231 (DR), and Fondecyt 11150897 (AN-P). We thank Nicole Arevalo and Anan Li for their technical help and Gidon Felsen, Robert Freedman, Abigail Pearson, and E. Mae Guthman for their insightful comments on the manuscript. Supplementary Material {#S10} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fncel.2020.00141/full#supplementary-material> ###### **(A)** Implant location was determined through CT scan imaging (Siemens Inveon animal CT scanner) and posterior electrode registration onto the Paxinos Mouse Brain Atlas **(B)**. The tetrodes can be observed in the BF in the coronal (white arrow) and sagittal CT images (bottom). The resolution of the horizontal CT allows to individually identify single tetrodes (top inset). 10 out of 16 animals were correctly implanted in the BF and included in this study. **(C)** Example of cluster analysis of one tetrode and one session. The spikes features, waveform, cluster size and inter spike interval (ISI) can be observed for a multi-unit (red) and a single unit (blue). **(D)** Bar histogram for all the units recorded in the go/no--go task *in vivo*. Top, single units, bottom, multi units. ###### Click here for additional data file. ###### **(A)** Heatmap of the normalized FR of all the units recorded during a Go/noGo task sorted by the delta FR between 1 s before and 1 s after trial start. During a Go/noGo task 24.2 and 25.3% of units responded to tstart during HIT and CR trials, respectively. Out of the 67 units that were recorded during FA trials, only 5,9% changed their FR in response to the tstart. **(B)** Heat map of the normalized FR of all the neuronas recorded in the Go/Go task. Units that responded to tstart = 18.2%, CS presentation = 9.1% and reward = 4.8% (Chi squared, *p* \< 0.05, corrected for multiple comparisons). **(C)** Comparison of the percentage of neurons responding to tstart, CS and reward presentation between the Go/noGo and Go/Go task. Statistical significance was determined by a Chi squared corrected for multiple comparisons (*p* \< pFDR = 0.0278, the correction was applied at the same time to the graph in [Figure 1E](#F1){ref-type="fig"}). **(D)** Table depicting the responsiveness of all the neurons recorded in the Go/noGo task, sorted by the change in FR exhibited during tstart. Notice that a large percentage of these cells (44.9%) did not change their FR significantly in any of the trial epochs, while 6.1% exhibited responses in all epochs. Out of the 31 units that exhibited a statistical decrease in FR during the stimulus presentation, 14 showed a previous increase during trial initialization, suggesting that previous neuronal activity could affect changes in FR later in the trial. However, 15 additional units showing an increase in FR during the odor epoch, exhibited no change in FR during tstart and two had a decrease in FR in response to trial initialization. In the other hand, out of the 12 units that exhibited an increase in FR during the stimulus presentation, 4 also showed an increase during trial initialization, three a decrease during trial initialization, and 5 had no change during this epoch. ###### Click here for additional data file. ###### **(A)** Cholinergic units exhibited, in addition to a latency of the first spike after light stimulation smaller than 10 ms, a significant increase in FR after light stimulation (*t*-test, *p* \< pFDR = 0.0062, corrected for multiple comparisons) and a reliability of response of 100% **(B)**. **(C)** Cholinergic neurons also exhibited low jitter (mean 4.9 ms). **(D)** Example of a cholinergic neuron responding at trial initialization (tstart). All the trials are aligned to tstart (time = 0 s, dashed black line) and sorted by odor presentation (orange line). **(E)** Top: Example of a cholinergic neuron that did not respond to reward. The PSTH was aligned to reward. Bottom: summary of the normalized FR responses to reward of all the identified cholinergic neurons (*n* = 6). Even though there appears to be a disturbance in the FR a few ms after time = 0, the changes are not statistically significant. ###### Click here for additional data file. [^1]: Edited by: Debra Ann Fadool, Florida State University, United States [^2]: Reviewed by: Daniel W. Wesson, University of Florida, United States; Ricardo C. Araneda, University of Maryland, College Park, United States [^3]: This article was submitted to Cellular Neurophysiology, a section of the journal Frontiers in Cellular Neuroscience

Introduction {#Sec1} ============ Competition is one of the fundamental ecological interactions between species^[@CR1]^. We can observe that coexisting species are competing for the same resources^[@CR2]^. A typical resource competition model which has been recognized widely is the Classical Lotka-Volterra competition^[@CR1],[@CR2]^. The analyses of this equation show that coexistence of two or more species becomes only possible if intraspecific competition is stronger than interspecific competition^[@CR3]^. Otherwise, dynamics leads to the exclusion of one species among *n* species, known as the competitive exclusion principle^[@CR1],[@CR4]^. However, in natural communities many competing species have been coexisting in the same habitat over time, resulting in a high species diversity. Hence, we suspect that there should be some mechanisms for coexistence of competitive species, e.g., spatial structures^[@CR5],[@CR6]^. These models, however, introduce an additional complexity into the mathematical models of classical LV systems. Compared with these complex models, coexistence of multiple species in natural communities seems to be far more ubiquitous. Therefore, a more universal explanation may be worth considering. Crowding effect is considered as one of the ubiquitous mechanisms in any biological populations^[@CR7]--[@CR16]^. A nonlinear density effect at high densities is called crowding effect, while that at low densities, Allee effect^[@CR7]--[@CR17]^. Unlike this nonlinear density effect, in the traditional mathematical models of population dynamics, density effect is usually treated as constant (i.e., linear), and crowding effect is not included. To consider the nonlinear density effects, aggregation models have been developed and studied extensively introducing 'mean crowding'^[@CR8]--[@CR10],[@CR15],[@CR16]^. These models show the coexistence of a few species, but not many species. The 'mean crowding' is a statistical feature of crowding affecting population growth rate (both birth rate and mortality). Here, we develop a simple LV type competition model with crowding effect on mortality only. We assume that the mortality rate of an individual increases with the density of a population. Moreover, using our modified LV competition model with nonlinear crowding effect, we show that multiple species are generally possible to coexist using LV system with crowding effect. This coexistence dynamics should be applicable to insect or animal species. Results {#Sec2} ======= We consider the LV competition system where all interspecific competition rates are unity, i.e., *α*~*ij*~ = *α*~*ji*~ = 1∀*i*. Setting *dx*~*i*~/*dt* = 0, we obtain the zero isoclines for the modified Lotka-Volterra (LV) competition equations with nonlinear crowding effect rate *m*~*i*~. These isoclines are straight lines in the classical LV competition equation when we sketch the graph of the population density 1 with respect to the population density 2 (Fig. [1](#Fig1){ref-type="fig"}, dotted lines). However, we observed that these isoclines turn out to be curved when we include a nonlinear crowding effect *m*~*i*~ (Fig. [1](#Fig1){ref-type="fig"}, solid lines). All four cases of Lotka-Volterra model show convergent-stable coexistence by adding a crowding effect, where an inferior (superior) species always increases (decreases) in densities (Fig. [1](#Fig1){ref-type="fig"}). The equilibrium point is moving from *E*~1~(*d*~*i*~ = 0.0) to *E*~2~(*d*~*i*~ = 0.5) which is also caused by the inclusion of a crowding effect *m*~*i*~, irrespective of other constant parameters (Fig. [1](#Fig1){ref-type="fig"}).Figure 1The zero isoclines and the long-term variation density of the classical LV competition equation with the inclusion of crowding effect for the four classical cases. Crowding effects (Solid line: *d*~*i*~ = 0.5, equilibrium: *E*~2~) enable coexistence in all four cases of classical LV system (Broken line: *d*~*i*~ = 0.0, equilibrium: *E*~1~). (**a**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, (**b**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, (**c**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, and (**d**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$. Colors indicate *x*~1~(red) and *x*~2~ (blue). *K*~*i*~: carrying capacity of species *i*, and *α*~*ij*~: competition coefficient from species *j* to species *i*. Parameters value: *b*~*i*~ = 1.0, *d*~*i*~ = 0.5, *m*~*i*0~ = 0.1, 0 ≤ *δ* ≤ 5. (**a**) *α*~12~ = 0.8, *α*~21~ = 1.2. (**b**) *α*~12~ = 1.2, *α*~21~ = 0.8. (**c**) *α*~12~ = 0.5, *α*~21~ = 0.4. (**d**) *α*~12~ = 1.2, *α*~21~ = 1.3. Initial density *x*~1~ = 0.5, *x*~2~ = 0.2. We used Anaconda package of the software Python 3.6 for our simulation analysis^[@CR22]^. Curved lines in the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{1}-{x}_{2}$$\end{document}$ plane imply that coexistence only take place when the value of *δ* → 1.0 for all *d*~*i*~ \> 0 (Fig. [2](#Fig2){ref-type="fig"}). However, by increasing *δ*, the competitive exclusion reappears in all exclusion cases, where the equilibrium state in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{1}-{x}_{2}$$\end{document}$ plane is strongly curved to return to the originated axis (Fig. [2](#Fig2){ref-type="fig"} except 2c). In contrast, an isocline is straight in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{1}-{x}_{2}$$\end{document}$ plane, where the two species coexist without crowding effects, such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}}$$\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}$${K}_{2} < \frac{{K}_{2}}{{\alpha }_{12}}$$\end{document}$ (Fig. [2c](#Fig2){ref-type="fig"}). Furthermore, Fig. [3a](#Fig3){ref-type="fig"} shows that two competing species can coexist for any small positive value of *δ* for all *d*~*i*~\> 0. Moreover, small positive real number *δ* will make coexistence possible but bigger value indicates the competitive exclusion principle again (Fig. [2](#Fig2){ref-type="fig"} except 2c and Fig. [3](#Fig3){ref-type="fig"}). Figure [4](#Fig4){ref-type="fig"} shows many-species Lotka-Volterra competition dynamics with crowding effects. Temporal dynamics of all species becomes convergent stable by adding crowding effects (Fig. [4a,b](#Fig4){ref-type="fig"}). The effect of *δ* in 5 or 10 species (Fig. [4c,d](#Fig4){ref-type="fig"}) is qualitatively same with the case of two species competition (Fig. [3b](#Fig3){ref-type="fig"}). Thus, many-species LV competition model with crowding effects leads to the stable coexistence of all species (Fig. [4](#Fig4){ref-type="fig"}).Figure 2Modified LV competition system with nonlinear crowding effect using the four possible cases of isoclines for the crowding strength factor *δ*. All non-coexistence patterns (**a,b,d**) move to coexistence as *δ* → 1.0 and return to competitive exclusion as *δ* → 5.0. The right column figures show the equilibrium points in the *x*~1~ − *x*~2~ phase plane. (**a**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, (**b**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, (**c**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$, and (**d**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}},{K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}})$$\end{document}$. Colors indicate *x*~1~(red) and *x*~2~ (blue). Parameter value and software used: see Fig. [1](#Fig1){ref-type="fig"}.Figure 3The effects of *δ* on population density in LV competition model with crowding effect. (**a**) Density ratio (*x*~1~/*x*~2~) of species 1 and 2 plotted on the crowding strength factor *δ* and the basic crowding component constant assuming equal between species 1 and 2 (*d*~1~ = *d*~2~). (**b**,**c**) Density of two species (*x*~1~: red, *x*~2~: blue) plotted against *δ*, where (**b**) 0 ≤ *δ* ≤ 1, (**c**) 0 ≤ *δ* ≤ 5, and *d*~*i*~ = 0.3. Parameters value: *b*~1~ = 1.0, *b*~2~ = 1.8; *α*~*ij*~ = 1.0; *m*~*i*0~ = 0.1. Initial density *x*~*i*~ = 0.5. See Fig. [1](#Fig1){ref-type="fig"} for software.Figure 4Long-term dynamics of many-species LV competition model with crowding effect. (**a**,**b**) Temporal dynamics with the basic crowding component constant (solid line: *d*~*i*~ = 0.3; dotted line (no crowding): *d*~*i*~ = 0.0) and the crowding strength factor *δ* = 0.3, (**c**,**d**) the equilibrium (final) states plotted against the effects of crowding strength factor *δ* (0 ≤ *δ* ≤ 1). (**a**,**c**) 5 species (birth rate: *b*~*i*~ = 1.0 + 0.1(*i* − 1)); (**b**,**d**) 10 species (birth rate: *b*~*i*~ = 1.0 + 0.05(*i* − 1)). Density of *i*-species: *x*~*i*~ (*i* = 1, 2, ..., *n*). Parameter value: *α*~*ij*~ = 1.0; *m*~*i*0~ = 0.1. Initial density *x*~*i*~ = 0.5. See Fig. [1](#Fig1){ref-type="fig"} for software. We also build LV competition models with aggregation effects on mortality that are qualitatively equivalent to the aggregation model of Hartley and Shorrocks^[@CR8]^. We compare them with the current crowding models using the same parameter conditions (Fig. [5](#Fig5){ref-type="fig"}). In the 2-and 5-species dynamics, all species survive and converge to a stable equilibrium in both the crowding model (Fig. [5a,c](#Fig5){ref-type="fig"}) and aggregation model (Fig. [5b,d](#Fig5){ref-type="fig"}). In the 10-species dynamics, all species survive and converge to stable equilibrium in the crowding models (Fig. [5e](#Fig5){ref-type="fig"}), while only five species in the aggregation model.Figure 5Long-term dynamics of 2-, 5- and 10-species LV competition models with two types of nonlinear density effects. (**a**,**c**,**e**) Crowding effect (Eq. [1](#Equ1){ref-type=""}; *d*~*i*~ = 0.3, *δ* = 0.4); (**b**,**d**,**f**) aggregation effect (Eq. [3](#Equ4){ref-type=""}; ε = 0.01). (**a**,**b**) 2 species (*m*~*i*0~ = 0.3); (**c,d**) 5 species (*m*~*i*0~ = 0.3); (**e,f**) 10 species (*m*~*i*0~ = 0.1). Parameter value: *α*~*ij*~ = 1.0; *b*~*i*~ = 1.0 + 0.01(*i* − 1). Initial density: (**a**,**b**) *x*~1~ = 0.15, *x*~2~ = 0.25, (**c**,**d**) *x*~3~ = 0.2, *x*~4~ = 0.12, *x*~5~ = 0.18, (**e**,**f**) *x*~6~ = 0.19, *x*~7~ = 0.22, *x*~8~ = 0.14, *x*~9~ = 0.13, *x*~10~ = 0.26. See Fig. [1](#Fig1){ref-type="fig"} for software. We also consider the effect of nonlinear competition terms^[@CR21]^ with or without crowding effect. We compare them with the current crowding models using the same parameter conditions (Fig. [6](#Fig6){ref-type="fig"}). In their original model, Taylor and Crizer consider a modified Lotka-Volterra model introducing the effect of nonlinear competition on growth rate. Here, we introduce the nonlinear competition on birth rate alone, since crowding effect is introduced only on mortality rate. In this manner, we can compare the effect of these changes separately. In the 2-species dynamics, both species survive and converge to a stable equilibrium in the crowding model with linear or nonlinear competition terms (Fig. [6a,c](#Fig6){ref-type="fig"}). In the current parameter conditions, the LV competition model with nonlinear competition terms lead to the extinction of one species (Fig. [6b](#Fig6){ref-type="fig"}). Note that by changing the conditions, this model lead to the coexistence of the two species^[@CR21]^. In the 5-species dynamics, all five species survive and converge to a stable equilibrium (Fig. [6d,e and f](#Fig6){ref-type="fig"}). Interestingly, the model with nonlinear competition terms converges to the same density for all five species (Fig. [6e](#Fig6){ref-type="fig"}), while the crowding effect lead to different densities among all species (Fig. [6d,f](#Fig6){ref-type="fig"}). In the 10-species dynamics, seven species survive and converge to stable equilibrium in the crowding models (Fig. [6g](#Fig6){ref-type="fig"}). In contrast, all ten species in the LV competition model with nonlinear terms lead to the coexistence of all species with an equal density (Fig. [6h](#Fig6){ref-type="fig"}). By combining the nonlinear competition and the crowding effect, all ten species survive and converge to different densities (Fig. [6i](#Fig6){ref-type="fig"}).Figure 6Long-term dynamics of 2-, 5- and 10-species LV competition models with crowding effect and/or nonlinear competition term. (**a**,**d**,**g**) Crowding effect (Eq. [1](#Equ1){ref-type=""}); (**b**,**e**,**h**) nonlinear competition term; (**c**,**f**,**i**) combination of both (Eq. [3](#Equ4){ref-type=""}). Parameter value: *α*~*ij*~ = 1.0; *b*~*i*~ = 1.0 + 0.1(*i* − 1). (**a,c,d,f,g,i**) *d*~*i*~ = 0.3, *δ* = 0.4. (**a--f**) *m*~*i*0~ = 0.3; (**g--i**) *m*~*i*0~ = 0.1. Initial density: (**a--c**) *x*~1~ = 0.15, *x*~2~ = 0.25, (**d--f**) *x*~3~ = 0.2, *x*~4~ = 0.12, *x*~5~ = 0.18, (**g--i**) *x*~6~ = 0.19, *x*~7~ = 0.22, *x*~8~ = 0.14, *x*~9~ = 0.13, *x*~10~ = 0.26. All simulations run for the same time steps. See Fig. [1](#Fig1){ref-type="fig"} for software. Discussion {#Sec3} ========== Many species compete for precisely the same limited resources to survive^[@CR1],[@CR2]^. Gause's exclusion principle show that multiple competing species cannot coexist in natural communities^[@CR1],[@CR4],[@CR18],[@CR19]^. Only one species, the superior competitor, will survive and other competitors will eventually become extinct. We should note that frequency dependence does not promote the coexistence of multiple species^[@CR20]^. Niche theory suggested that it will only become possible for the competing species to coexist if they have different niche^[@CR2],[@CR3]^. Linear density effects show that coexistence becomes possible under very limited conditions. Hence, we search for mechanisms that will enable coexistence of competitive species^[@CR5],[@CR6]^. As a universal and more biologically founded solution, we consider crowding effect, nonlinear density effects at high densities, in the LV competition systems. Classical LV competition model shows that it is only possible for two species to coexist together if intraspecific competition is stronger than interspecific competition^[@CR3]^. However, our results have shown that inclusion of crowding effect to the classical Lotka-Volterra competition system guarantees the coexistence of two or more species. We have investigated that two species can coexist when we include crowding effect to the classical LV competition system (Fig. [1](#Fig1){ref-type="fig"} (solid lines), Fig. [2](#Fig2){ref-type="fig"} as *δ* → 1.0, Fig. [3a](#Fig3){ref-type="fig"} as *δ* → 1.0, and Fig. [4a,b](#Fig4){ref-type="fig"} (solid lines)) compare to the results of the classical LV competition model which show the competitive exclusion principle (Fig. [1](#Fig1){ref-type="fig"} (dotted lines) and Fig. [4a,b](#Fig4){ref-type="fig"} (dotted lines)). Crowding effect has been recognized in many natural and experimental populations^[@CR7]--[@CR16]^. Note that if the density of population is increased a lot more than the carrying capacity, then crowding effect will kill all competing individuals. We here, introduced the intraspecific crowding effect into the Lotka-Volterra competition model. We have shown that a weak crowding effect make it possible to achieve a stable coexistence of multiple species. Our analysis implies that increased mortality under high density works as elevated intraspecific competition leading to the coexistence. This may be another ubiquitous mechanism for the coexistence of multiple species leading species diversity in nature. We compare the current model with the aggregation model of coexistence^[@CR8]^ (Fig. [5](#Fig5){ref-type="fig"}). Unlike the original aggregation model of Hartley and Shorrocks in which the aggregation reduces growth rates, we only include the aggregation effect on mortality rate. These examples show that the aggregation model becomes difficult to maintain the coexistence of all or many species when the number of species is increased. In contrast, the coexistence of all or many species can be easily achieved in the current crowding model even when the number of species is increased (Fig. [5a,c and e](#Fig5){ref-type="fig"}). The reason why the coexistence becomes difficult when the number of species increased in the aggregation model seems to depend on the combination of the parameters, where the coexistence region is expressed as a polygon in the *n*-species parameter space which can disappear easily when the number of species is increased. The logic is same with linear programming in *n*-dimensional space. In contrast, in the crowding model, the mortality rate of any species increases when its density approaches its carrying capacity. Because of this, increasing mortality rate near carrying capacity will keep all the species at the densities much below their carrying capacity. Thus, the intraspecific competition becomes most severe at or near carrying capacity, resulting in the stable coexistence of all species. We also compare the effects of the nonlinear competition terms with the current crowding effects (Fig. [6](#Fig6){ref-type="fig"}). Unlike the LV model with nonlinear competition effects on growth rates^[@CR21]^, we only include the nonlinear competition effects on birth rate alone since crowding effect is included only on mortality rate, so that these effects can be easily distinguished. In the current parameter conditions, the 2-species model results in the exclusion of one species (Fig. [6b](#Fig6){ref-type="fig"}). However, when the number of species is increased, LV model with nonlinear competition terms will lead to the coexistence of all species with an equal density regardless of the parameter combinations (Fig. [6e,h](#Fig6){ref-type="fig"}). It is not sure whether the stability can be achieved easily when the number of species increased in the nonlinear competition model. However, the equilibrium density is identical for all coexisting species in the model with nonlinear competition terms. This means that the effects of other species-specific parameters are completely cancelled by the introduction of nonlinear competition effect. In contrast, the coexistence of all or many species can be easily achieved in the current crowding model even when the number of species is increased (Fig. [6a,d and g](#Fig6){ref-type="fig"}). By combining the nonlinear competition and the crowding effect, many species survive and converge to different densities (Fig. [6c,f and i](#Fig6){ref-type="fig"}). Thus, both mechanisms can promote the coexistence of many species differently. The most important assumption in our model of crowding effect is the increase in mortality rate at high density. Here, *d*~*i*~ represents the proportion of crowding mortality contribution and *δ* is the power of crowding effect. Therefore, when *d*~*i*~ = 0, this model reduces to the classical LV competition model. This assumption should be one natural way to include the crowding effect. However, there may be many other natural ways to include crowding effect, e.g., the aggregation model of Hartley and Shorrocks^[@CR8]^. We have to wait for empirical studies to verify which way is actually functioned in a natural ecosystem. Note that these functional mechanisms are not exclusive of each other. Thus, the valid mechanisms may be different depending on a natural ecosystem. We should also note that the nonlinearity in the functional responses should be an important factor driving population dynamics and resulting evolution. The nonlinearity in density effect may be biologically inherent and appears as crowding effect at high densities and as Allee effect at low densities. Our studies thus show that the real competitive communities have a much more complicated dynamical system than the classical LV competition system. Methods {#Sec4} ======= Mathematical models {#Sec5} ------------------- We consider the modified Lotka-Volterra (LV) competition equations with crowding effect rate ***m***~***i***~ for species *i*. In our model, we only consider competition between two species. In addition, carrying capacity ***K***~***i***~ is set to be equal to 1, i.e., *K*~*i*~ = 1. The modified LV competition model is shown on the following equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\begin{array}{c}\frac{d{x}_{1}}{dt}={b}_{1}{x}_{1}(1-{x}_{1}-{\alpha }_{12}{x}_{2})-{m}_{1}{x}_{1},\quad {b}_{1}\,{\rm{and}}\,{\alpha }_{12}\,{\rm{are}}\,{\rm{constant}}\\ \frac{d{x}_{2}}{dt}={b}_{2}{x}_{2}(1-{x}_{2}-{\alpha }_{21}{x}_{1})-{m}_{2}{x}_{2},\quad {b}_{2}\,{\rm{and}}\,{\alpha }_{21}\,{\rm{are}}\,{\rm{constant}}\end{array}$$\end{document}$$where ***x***~***i***~ represents the population density of species *i* where *i* = 1, 2. In this model, parameter ***b***~***i***~ represents the birth rate of species *i* while ***α***~***ij***~ represents the effect of species *j* on *i* where *i*,*j* = 1, 2 and *i* ≠ *j*. The crowding effect rate *m*~*i*~ 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}_{i}={m}_{i0}+{d}_{i}{x}_{i}^{\delta },\quad \delta \in (0,\infty )\forall i$$\end{document}$$where parameter ***m***~***i***0~ represents the initial mortality factor of species *i*. Parameter ***d***~***i***~ represents the density-dependent factor of species *i*. In addition, the sum of the initial mortality and density-dependent factor of species *i* must be greater than 0 but less than or equal to 1, i.e.,0 \< *m*~*i*0~ + *d*~*i*~ ≤ 1. Note that, if the initial mortality factor *m*~*i*0~ is zero then nonlinear crowding effect rate *m*~*i*~ will imply that the intraspecific competition is perfectly density-dependent. In addition, if *d*~*i*~ = 0 $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\forall }}i$$\end{document}$ then *m*~*i*~ = *m*~*i*~0 $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\forall }}i$$\end{document}$  which will imply that the modified LV competition model is the same with the classical LV competition model. Following equations 4 and 11 in the paper of Hartley and Shorrocks^[@CR8]^, we arrived with the Lotka-Volterra competition model adding the effect of a few more individuals, shown on the following equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\begin{array}{c}\frac{d{x}_{1}}{dt}={b}_{1}{x}_{1}(1-{x}_{1}-{\alpha }_{12}{x}_{2})-{m}_{10}{{x}_{1}}^{1+\varepsilon }\,,\quad {b}_{1}\,{\rm{and}}\,{\alpha }_{12}\,{\rm{are}}\,{\rm{constant}}\\ \frac{d{x}_{2}}{dt}={b}_{2}{x}_{2}(1-{x}_{2}-{\alpha }_{21}{x}_{1})-{m}_{20}{{x}_{2}}^{1+\varepsilon },\quad {b}_{2}\,{\rm{and}}\,{\alpha }_{21}\,{\rm{are}}\,{\rm{constant}}\end{array}$$\end{document}$$where *ε* is any positive real number and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${m}_{i0}{{x}_{i}}^{1+\varepsilon }$$\end{document}$ is the effect of a few more individuals for all species *i*. Note that, we do not include the crowding effect on birth rates unlike the aggregation model of Hartley and Shorrocks^[@CR8]^. In addition, we also used the modified LV competition model of Taylor and Crizer^[@CR21]^ with the inclusion of nonlinear crowding effect for two species. In their model, they add nonlinear competition terms to prevent the population of species 2 to have a smaller effect on the population of species 1 when the population density of species 1 is very small compare to the population density of species 2 and vice versa. Taylor and Crizer's competition model with nonlinear crowding effect is shown on the following equations:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\begin{array}{c}\frac{d{x}_{1}}{dt}={b}_{1}{x}_{1}(1-{x}_{1}-{\alpha }_{12}{{x}_{2}}^{2})-({m}_{10}+{d}_{1}{{x}_{1}}^{\delta }){x}_{1}\,,\quad {b}_{1}\,{\rm{and}}\,{\alpha }_{12}\,{\rm{are}}\,{\rm{constant}}\\ \frac{d{x}_{2}}{dt}={b}_{2}{x}_{2}(1-{x}_{2}-{\alpha }_{21}{{x}_{1}}^{2})-({m}_{20}+{d}_{2}{{x}_{2}}^{\delta }){x}_{2},\quad {b}_{2}\,{\rm{and}}\,{\alpha }_{21}\,{\rm{are}}\,{\rm{constant}}\end{array}.$$\end{document}$$ Numerical simulations {#Sec6} --------------------- In order to determine the impact of the inclusion of nonlinear crowding effects to the classical Lotka-Volterra equation we simulate the modified LV competition equations using Anaconda package of the software Python 3.6^[@CR22]^. Initially, we determine its effect if there are two competing species in a community and later extend it up to 10 competing species. We also determine the effect when we use small and large values of δ. Moreover, we identify the right combination of *d*~*i*~ and δ that will allow competing species to coexist. Without losing essential qualitative dynamics, we considered the following parameter ranges in our numerical simulations:0 ≤ Initial of *x*~*i*~ ≤ 1 for all *i*;0 \< *α*~*ij*~ ≤ 1 for all *i*, *j*;1 ≤ *b*~*i*~ ≤ 2 for all *im*~*i*0~ = 0.1 or 0.3 for all *i*;*d*~*i*~ = 0,0.1,0.3 or 0.5 for all *i*;0 ≤ δ ≤ 5; and*K*~*i*~ = 1 In addition, we compare the results of the LV competition equation ([1](#Equ1){ref-type=""}) with nonlinear crowding effect ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${m}_{i}={m}_{i0}+{d}_{i}{x}_{i}^{\delta }$$\end{document}$) and without nonlinear crowding effect (*m*~*i*~ = *m*~*i*0~) using the four possible cases of isoclines. The four isocline cases are:$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}},{K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}}$$\end{document}$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{1} < \frac{{K}_{2}}{{\alpha }_{21}},{K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}}$$\end{document}$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}},{K}_{2} < \frac{{K}_{1}}{{\alpha }_{12}}$$\end{document}$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{1} > \frac{{K}_{2}}{{\alpha }_{21}},{K}_{2} > \frac{{K}_{1}}{{\alpha }_{12}}$$\end{document}$ where *K*~*i*~ is the carrying capacity of species *i* and *α*~*ij*~ represents the effect of species *j* on *i* where *i*, *j* = 1, 2 and *i* ≠ *j*. Maica Krizna A. Gavina and Takeru Tahara contributed equally to this work. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work was partly supported by grants-in-aid from the Japan Society for Promotion of Science (nos 22255004, 22370010, 26257405 and 15H04420 to JY; no. 26400388 to SM; nos 14J02983, 16H07075, 17J06741 and 17H04731 to HI; nos 25257406 and 16H04839 to TTo); and the Mitsubishi Scholarship (MISTU1722) to MKAG. M.K.A.G., T.T.a., T.N. and J.Y. conceived the study. M.K.A.G., T.T.a., T.N. and J.Y. built and analyzed the model. M.K.A.G. and T.T.a. built a program and ran the numerical simulations. H.I., S.M., G.I. and T.O. verified the mathematical properties of the models. H.I., T.T.o. and J.Y. developed biological interpretations of the model. M.K.A.G., T.T.a. and J.Y. wrote the manuscript. M.K.A.G. and T.T.a. as the lead authors. All authors reviewed the manuscript and gave final approval for publication. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

{#sp1 .147} {#sp2 .148} {#sp3 .149} {#sp4 .150} {#sp5 .151} {#sp6 .152} {#sp7 .153} {#sp8 .154} {#sp9 .155} {#sp10 .156} {#sp11 .157} {#sp12 .158} {#sp13 .159} {#sp14 .160}

Introduction ============ Parkinson\'s disease (PD), which is both chronic and progressive, is the second most common neurodegenerative disorder, and its prevalence is increasing rapidly.[@B1],[@B2] Health-related quality of life (HrQoL) is considered critical in chronically ill patients, especially in the elderly, and PD research is increasingly focused on factors affecting the HrQoL.[@B3] Although motor features have until recently dominated clinical impact assessments in patients with PD, non-motor symptoms are currently attracting more attention as determinants of HrQoL in PD,[@B4],[@B5] with a growing emphasis on the importance of HrQoL issues when initiating and maintaining treatment.[@B5],[@B6] The 39-item Parkinson\'s disease questionnaire (PDQ-39), which contains 39 items involving 8 discrete dimensions that mainly evaluate HrQoL, is the representative disease-specific tool for assessing HrQoL in PD. The PDQ-39 evaluates mobility, activities of daily living, emotional well-being, stigma, social support, cognition, communication, bodily discomfort and summary index.[@B7],[@B8] The PDQ-39 was originally developed in Britain and has been translated into and validated in many different languages worldwide.[@B9]-[@B13] Because perceptions of HrQoL are subjective and may vary according to cultural and individual backgrounds, translated and validated versions tailored for different cultural contexts are required for research. The aim of the present study was to translate the PDQ-39 into Korean and to validate the reliability and consistency of the new Korean version as well as its feasibility for use in future assessments of HrQoL in the Korean-speaking PD population. Methods ======= Subjects -------- One hundred and two PD patients from 10 movement-disorder centers at university-affiliated hospitals in South Korea who met the clinical diagnostic criteria of the United Kingdom Parkinson\'s Disease Society Brain Bank[@B14] were enrolled in the present study. The obligatory inclusion criterion for all participants was no change in anti-parkinsonian drugs for at least the previous 4 weeks. The exclusion criteria included patients who were younger than 40 years at the onset of PD, had possible secondary causes of PD such as drugs or structural brain lesions that can induce parkinsonism, who were taking medications including antidepressants that could affect cognitive function, or had impaired cognitive function represented by a total Mini-Mental State Examination score of less than 24. All subjects provided written informed consent for unprompted study participation. Ethical approval was obtained from the joint ethics committee of each participating university hospital. Translation ----------- The original PDQ-39 was translated into Korean using forward and backward translation, expert committee review and pretests of the translated Korean version of the PDQ-39 (K-PDQ-39). Two independent bilingual translators translated the original British version of the PDQ-39 into Korean. The translated Korean version was then translated back into British English by another bilingual translator who had no knowledge of the original version of the questionnaire. The back-translated version was then compared to the original version by the same translators. An expert committee (comprising D.Y. Kwon, J.S. Baik and S.B. Koh) reviewed the translated version and modified the draft version until a consensus was reached among them. After pretesting the translated version in four patients, the K-PDQ-39 was finalized and used in subsequent analyses. K-PDQ-39 -------- The K-PDQ-39 comprises 39 items involving 8 discrete dimensions: mobility, activities of daily living, emotional well-being, stigma, social support, cognition, communication and bodily discomfort. The assessment period is the \"previous 1 month.\" Each item is scored on a 5-point Likert scale ranging from 0 (\"never\") to 4 (\"always\"). The total K-PDQ-39 score is expressed as a percentage ranging from 0 to 100. Data collection --------------- All of the authors, who were experts in movement disorders and experienced interviewers from each movement center, conducted the following assessments of subjects in order to assess and qualify non-motor symptoms in PD patients and K-PDQ-39: basic demographics, levodopa equivalent dose (LDED),[@B15] modified Hoehn and Yahr (H&Y) stage,[@B16] Unified Parkinson\'s Disease Rating Scale (UPDRS) parts I, II, and III,[@B17] the Korean version of the Mini-Mental State Examination (K-MMSE)[@B18] for measuring cognitive function, the Korean version of the Montgomery-Asberg Depression Scale (K-MADS), the Epworth Sleepiness Scale (ESS),[@B19] and the non-motor symptoms scale (NMSS).[@B20] Participants returned to outpatient clinics for retests using the K-PDQ-39 in order to assess test-retest reliability over time intervals of from 10 to 14 days, which was a sufficient delay to minimize memory or practice effects. Statistical analysis -------------------- Reliability was assessed in order to measure the internal consistency and stability. Floor and ceiling effects involving less than 20% of the total population per domain were considered acceptable.[@B35] Cronbach\'s α coefficient was calculated to analyze the internal consistency. The criterion value for α was ≥0.70. Test-retest reliability was assessed using the Guttman Split-Half Coefficient analyses with values higher than 0.70 considered indicative of acceptable reliability. The relationships between subscales of the K-PDQ-39 and other variables were analyzed using Spearman\'s rank correlation coefficients, since scores were not evenly distributed. These coefficients were also used to quantify the convergent validity with other clinical scales. SPSS (version 15.0 for Windows, Chicago, IL, USA) was used for all statistical analyses. Results ======= Demographic data ---------------- A total of 102 patients (52 men and 50 women) were recruited for the study. The demographic data are listed in [Table 1](#T1){ref-type="table"}. Among the 102 PD patients, 75 (73.5%) patients had H&Y stages up to 2.0 (32 at stage 1, 12 at stage 1.5, and 31 at stage 2) and 27 (26.5%) had H&Y stages of at least 2.5 (17 at stage 2.5, 9 at stage 3, 1 at stage 4). Finally, 101 PD patients completed the K-PDQ-39 retest. Reliability and validity of the K-PDQ-39 ---------------------------------------- [Table 2](#T2){ref-type="table"} lists the scores in each domain and the total score. The mean total summary index of the K-PDQ-39 was 23.20±18.63% (mean±SD), and ranged from 0% to 74.9%. A higher score in each K-PDQ-39 domain indicates greater discomfort felt by the patient: emotional well-being (27.51±24.11%) and bodily discomfort (26.43±23.38%) were the major complaints, while social support (14.70±21.47%) and communication (13.05±18.49%) were the least problematic. All of the K-PDQ-39 domains showed acceptable ceiling effects. Emotional well-being and cognition domains were within the acceptable range of floor effects (\<20%), but the other six domains ranged from 21.6% to 52.9%. Both the floor and ceiling effects of the SI were 0% ([Table 2](#T2){ref-type="table"}). Each K-PDQ-39 domain showed significant correlations with summary index scores (rS=0.559-0.793, *p*\<0.001). Six of the eight domains met the standards of reliability (Cronbach\'s α coefficient ≥0.70); the exceptions were stigma and social support ([Table 3](#T3){ref-type="table"}). For assessments of test-retest reliability, the Guttman Split-Half Coefficient value of the K-PDQ-39 summary index was 0.919 (*p*\<0.001). and ranged between 0.715 and 0.943 for the eight domains, which satisfied the standards for test-retest reliability (*p*\<0.001) ([Table 3](#T3){ref-type="table"}). The relationships of the K-PDQ-39 and clinical data with rating scales were analyzed. All clinical variables except age, disease duration, LDED, H&Y stage, UPDRS parts I, II, and III, mood status (K-MADS), cognition (K-MMSE) and daytime sleepiness (ESS) showed strong correlations with the K-PDQ-39 summary index (*p*\<0.01). Age was the only factor that was not significantly correlated with the K-PDQ-39 domains; the other factors-including disease duration, LDED, H&Y stage, UPDRS parts I, II, and III, K-MADS, K-MMSE and ESS- were significantly correlated with each K-PDQ-39 domain (*p*\<0.01-0.05) ([Table 4](#T4){ref-type="table"}). The K-PDQ-39 summary index score demonstrated significant relationships with total NMSS scores (rS=0.814, *p*\<0.01; [Fig. 1](#F1){ref-type="fig"} and [Table 5](#T5){ref-type="table"}). [Table 5](#T5){ref-type="table"} indicates that most of the K-PDQ-39 domains were strongly correlated with NMSS items. Discussion ========== There is an increasing need to evaluate HrQoL when managing patients with chronic, degenerative diseases.[@B3],[@B21] We translated the most commonly used and validated instrument for evaluating HrQoL in PD patients, the PDQ-39, into Korean, and found the K-PDQ-39 to be a reliable, valid and useful instrument that was easily applicable to the Korean-speaking PD population. The data of the present study were collected from 10 different movement-disorder clinics and interviews were performed by different investigators, which strengthens the positive results of the K-PDQ-39 reliability and validity in terms of suitability for application in clinical practice. The mean scores in the domains were lower (indicating a lower degree of discomfort) than the results for the original British and other translated, and validated versions, including those in Chinese, Greek, Spanish, American English and Portuguese.[@B7]-[@B13],[@B22] A possible explanation for this result is that the disease severity according to H&Y stage being milder and the stages of PD being earlier in the present study than in the patients included in previous studies, since several studies have demonstrated that mean PDQ-39 scores are positively correlated with disease duration as well as disease severity.[@B8],[@B11],[@B12] Additionally, disease duration and severity primarily contribute to HrQoL through associated motor complications arising from treatment in late-stage PD.[@B23]-[@B25] However, the scores were lowest for social support and communication, as also found in the present study.[@B10]-[@B12] These findings are also reflected in the presence of floor effects in over half of the domains, with no ceiling effects in any of the domains. Social support, communication and stigma domains showed the largest floor effects in this study, which is similar to previous validation studies involving various languages.[@B8],[@B13],[@B34] A cut-off above 0.70 for Cronbach\'s α coefficient is reportedly considered indicative of acceptable internal consistency and reliability.[@B26] The internal consistency of the K-PDQ-39 was satisfactory, since Cronbach\'s α coefficient ranged from 0.58 to 0.80. Two of the eight domains failed to reach the standards, and the average correlations of the items comprising the test were very strong (Cronbach\'s α coefficient=0.913). Test-retest reliability as calculated by the Guttman Split-Half coefficient value was also adequate, with values ranging from 0.715 to 0.943. These results suggest that the K-PDQ-39 is a stable and reliable tool. The relationship between clinical data and K-PDQ-39 was also analyzed. While disease duration, disease severity (H&Y stage and UPDRS), LDED, depressive mood (K-MADS), cognition (K-MMSE), and daytime sleepiness (ESS) were strongly correlated with K-PDQ-39 domains, age was only strongly correlated with the K-PDQ-39 mobility domain. These results, which are consistent with those of previous studies[@B27],[@B28] may be accounted for by several factors. First, patient age has no direct relationship with disease duration or severity, both of which have strong correlations with PD HrQoL.[@B8],[@B11],[@B12] Second, the age range of the participants in the present study was limited to43-84 years (with a mean of 65.3 years), which Soh et al.[@B21] pointed out restricts the ability of HrQoL to discriminate between diverse age groups. Our data also suggest that a longer and more severe case of PD will lower the quality of life. UPDRS (parts I and II) scores were strongly correlated with all items of the K-PDQ-39. The scores for depressive mood and cognition were closely related to nearly all items of the K-PDQ-39, indicating that HrQoL factors are associated with both motor and non-motor symptoms of PD. In particular, the mood status of the patients (as assessed by K-MADS) showed the strongest correlations with all domains of K-PDQ-39 (rS=0.425-0.704, *p*\<0.01). Similar patterns have been found previously, in that the the presence of depressive symptoms was the main factor associated with a poor quality of life.[@B3],[@B29] Concurrently, NMSS items that quantify non-motor symptoms of PD demonstrated strong correlations with K-PDQ-39 scores, which implies that non-motor features of PD are the most important determinants of quality of life in PD.[@B4],[@B5] Although non-motor symptoms are prevalent from the early stages or prestages of motor symptoms of PD, the clinical impact of non-motor aspects on HrQoL has only recently been highlighted.[@B21],[@B30]-[@B33] The first limitation of the present study was the uneven distribution of disease severity, with most participants having mild disease (73.5% of the participants had an H&Y stage of up to 2.0). The late-stage HrQoL of the PD patients therefore might not have been properly evaluated. However, the present study did not aim to assess HrQoL severities in patients, and also showed that the HrQoL had clinical impacts from the relatively early, mild stages of PD. The second limitation of the present study was that other assessments of HrQoL were not compared with the K-PDQ-39, and hence convergent validity might not have been verified. In conclusion, we have demonstrated that the K-PDQ-39 is a useful, valid, and reliable instrument for assessing HrQoL in PD patients. Further detailed investigations of HrQoL in Korean-speaking PD patients should be undertaken using the K-PDQ-39. This work was supported by grants from Sandoz Korea. The authors have no financial conflicts of interest. {#F1} ###### Demographic data of PD patients  H&Y stage: Hoehn and Yahr stage, K-MADS: Korean version Montgomery Asberg Depression Scale, K-MMSE: Korean version Mini-Mental Status Examination, LDED: levodopa equivalent dosage, PD: Parkinson\'s disease, SD: standard deviation, UPDRS: Unified Parkinson\'s Disease Rating Scale. ###### Scores in each item of K-PDQ-39 and correlations of items with total score of K-PDQ-39  K-PDQ: Korean version-Parkinson\'s disease questionnaire. ###### Internal consistency (Cronbach\'s α coefficient) between subitem scores and the summary index score and test-retest reliability (Guttman Split-Half coefficient) of the K-PDQ-39 (*n*=102)  K-PDQ: Korean version-Parkinson\'s disease questionnaire. ###### Spearman\'s correlation of the K-PDQ-39 and clinical features and rating scales  ^\*^Spearman\'s rank correlation, *p*\<0.01, ^†^Spearman\'s rank correlation, *p*\<0.05. ESS: Epworth Sleepiness Scale, H&Y stage: Hoehn and Yahr stage, K-PDQ: Korean version-Parkinson\'s disease questionnaire, LDED: levodopa equivalent dose, MADS: Montgomery-Asberg Depression Scale, MMSE: Mini-Mental Status Examination, NMSS: the non-motor symptoms scale for Parkinson\'s disease, PD: Parkinson\'s disease, UPDRS: the Unified Parkinson\'s Disease Rating Scale. ###### Spearman\'s correlation of the K-NMSS and the subscale items of the K-PDQ-39  ^\*^Spearman\'s rank correlation, *p*\<0.001, ^†^Spearman\'s rank correlation, *p*\<0.05. K-NMSS: Korean version-non-motor symptoms scale for Parkinson\'s disease, K-PDQ-39: Korean version-Parkinson\'s disease questionnaire.

Introduction ============ Peste des petits ruminants (PPR) is a highly contagious viral disease of wild and domestic small ruminants and camels characterized by oculonasal discharge, stomatitis, diarrhea and pneumonia ([@b9]; [@b16]). It is a disease of economic significance because of its transboundary nature, high morbidity and mortality which result in loss of production, abortion, death, limits on export and threat to human food chain ([@b21]). The disease is caused by a virus belonging to the genus *Morbillivirus* of the family *Paramyxoviridae*. The PPR virus (PPRV) is highly contagious and easily transmitted by direct contact through the secretions and/or excretions of infected animals to nearby healthy animals ([@b2]; [@b16]). Contact and movement of animals from affected to unaffected areas play an important role in transmitting PPRV especially where communal grazing system is practiced. The disease was first reported in West Africa in the early 1940s and later recognized as endemic in both West and Central Africa ([@b1]). Currently, PPR is prevalent in Central, Eastern and Western Africa, Asia, and the Near and Middle East ([@b6]; [@b2]; [@b14], [@b15], [@b16]). PPRV has been classified into four lineages that are distinct to different geographical locations based on variable nucleotide sequences of the nucleoprotein (N) gene. Western and Central African PPRV cluster into Lineages I and II, Eastern African and PPRV found in the southern part of Middle East cluster into Lineage III while Asian PPRV mainly cluster into Lineage IV ([@b16]). [@b20] confirmed the absence of PPR antibodies in the first comprehensive serological study done in 1998 to determine the status of PPR among Tanzanian sheep and goats. However, subsequent studies confirmed the presence of PPR antibodies in goats and sheep of Northern Tanzania in 2004 ([@b7]), in 2009 ([@b18]) and between 2008 and 2010 ([@b8]). Antibodies against PPRV have been detected in cattle sampled in 2011 in Northern Tanzania within the Loliondo Controlled Game Area ([@b10]). In 2011, an outbreak of PPR in goats and sheep was reported and confirmed in Southern Tanzania ([@b17]), however, limited data is available about the disease in the rest of Tanzania. The objective of this study was to determine presence of active PPRV infections in Eastern and Northern Tanzania and establish the phylogenetic relationship between Tanzanian and previously characterized PPRV isolates. Materials and Methods ===================== Study area ---------- The present study was conducted in Ngorongoro district in Arusha region located in Northern Tanzania and Mvomero district in Morogoro region located in Eastern Tanzania. The Ngorongoro district was selected owing to occurrence of several PPR outbreaks since 2008 even in the areas where vaccination was practiced. The District is considered to be a risk area because it borders Kenya, from which it is believed that the disease was introduced into Northern Tanzania ([@b8]) and it forms part of the Northern transboundary animal movement route. The Mvomero district was also included in the present study because farmers reported outbreaks of a disease resembling PPR. In Mvomero district, samples were collected from goats in November 2012 and January 2013 in Kauzeni and Dakawa villages, respectively (Fig. [1](#fig01){ref-type="fig"}). In Ngorongoro district, samples were obtained in February and March 2013 from villages bordering the Serengeti National Park including Meshili village within Ngorongoro Conservation Area and Piyaya and Malambo villages within Loliondo Game Controlled Area (Fig. [1](#fig01){ref-type="fig"}). In both Mvomero and Ngorongoro districts, samples were obtained from goats belonging to maasai pastoralists. {#fig01} Sample collection and processing -------------------------------- Nasal and ocular swabs, and whole blood samples were obtained from randomly selected live goats with clinical presentation suggestive of PPR. Whole blood was collected from the jugular vein using EDTA-containing vacutainer tubes (BD Biosciences, Franklin Lakes, NJ, USA). In addition, lungs, intestines, lymph nodes, whole blood and, ocular and nasal swabs were obtained from five goats in Mvomero district and four goats in Ngorongoro district that died naturally. Buffy coat from whole blood was obtained by centrifuging 500 μl of histopaque (Sigma-Aldrich, St. Louis, MO, USA) layered with 1 ml of whole blood at 400 ***g*** for 30 min at 4°C. The upper layer was discarded while the opaque interface containing the buffy coat was carefully collected. Nasal and ocular swabs were collected and placed in universal viral transport medium (BD Biosciences) followed by vortexing to dislodge cells from the swabs and centrifugation at 8000 ***g*** for 5 min at room temperature. Portions of lungs, intestines and lymph nodes from the same animal were pooled before they were homogenized in F-12 basal cell culture medium (Invitrogen, Carlsbad, CA, USA) to obtain a 10% tissue suspension. The tissue supernatant was obtained by centrifugation of tissue suspensions at 8000 ***g*** for 5 min at room temperature. All samples and the buffy coat were stored at −80°C until RNA extraction was performed. PPRV RNA extraction and amplification of PPRV nucleoprotein gene ---------------------------------------------------------------- Viral RNA was recovered from buffy coat, swabs and homogenized tissue samples using a commercial QIAamp viral RNA kit (Qiagen, Hilden, Germany) according to the manufacturer\'s instructions. PPRV nucleoprotein (N) gene was amplified in a 25 μl reaction using AgPath-ID One-Step RT-PCR kit (Applied Biosystems, Courtaboeuf, France) using PPRV-specific primers NP3 (5′-TCTCGGAAATCGCCTCACAGACTG-3′) and NP4 (5′-CCTCCTCCTGGTCCTCCAGAATCT-3′) as previously described by [@b3]. Briefly, a reverse transcription was carried out at 45°C for 30 min before initial denaturation of DNA at 95°C followed by 40 cycles of denaturation at 95°C for 30 s, annealing at 55°C for 30 s and elongation at 72°C for 30 s, and a final extension at 72°C for 10 min. Amplification was carried out in a GeneAmp PCR System 9700 (Applied Biosystems, Foster City, CA, USA) followed by electrophoresis on a 1.5% agarose gel. Agarose was pre-mixed with GelRed nucleic acid stain (Phenix Research Products, Candler, NC, USA) and visualization of PCR products was done using a BioDoc-It imaging system (UVP, Upland, CA, USA). PCR products were treated with exonuclease I and alkaline phosphatase before they were sequenced directly using Big Dye Terminator Cycle Sequencing Kit Version 3.1 (Applied Biosystems). Alternatively, PCR fragments were purified from agarose gels using a NucleoSpin gel and PCR clean-up kit (Macherey-Nagel, Düren, Germany). The gel purified PCR fragments were cloned in a TOPO-T/A plasmid vector (pCR4-TOPO) (Invitrogen). Afterwards, dideoxynucleotide cycle sequencing reaction was performed using standard M13 forward (5′-GTAAAACGACGGCCAG-3′) and M13 reverse (5′-CAGGAAACAGCTATGAC-3′) primers for sequencing of PPRV partial N gene inserts in the TOPO-T/A plasmid vectors using Big Dye Terminator Cycle Sequencing Kit Version 3.1 (Applied Biosystems). Sequencing PCR products were purified by ethanol precipitation and separated on a 3500 Genetic Analyzer (Applied Biosystems). Phylogenetic relationship of PPRV N gene sequences -------------------------------------------------- The forward and the reverse complement nucleotide sequences delimited by forward and reverse primers of several N gene PCR products of PPRV from the same location were aligned to obtain a consensus nucleotide sequence. A set of sequences representing all lineages of PPRV, including newly reported PPRV sequences from African countries, was used for phylogenetic analysis. Sequences were aligned using ClustalW algorithm in BioEdit (Ibis Biosciences, Carlsbad, CA, USA) and clustering pattern was determined by neighbor-joining method using the Kimura-2-parameter option implemented within [mega]{.smallcaps} 5 ([@b19]). Results ======= Clinical signs and post mortem findings observed in goats suspected with PPR ---------------------------------------------------------------------------- The main clinical signs observed in goats suspected with PPR included fever, dullness, diarrhea, lacrimation, matting of eye lids, purulent oculonasal discharges, cutaneous nodules, erosions on the soft palate and gums and laboured breathing. In some animals, raised cutaneous nodules of up to 3 cm in size were observed especially on the neck region (Fig. [2](#fig02){ref-type="fig"}). The rectal temperature of the examined goats ranged between 39.1 to 41°C, with most animals examined having temperatures above 40°C. Post mortem findings included pneumonia, dark red and firm to touch areas in the lungs, haemorrhages and congestion of the intestines, watery intestinal contents and hemorrhages of lymph nodes associated with the respiratory and gastrointestinal systems (Fig. [2](#fig02){ref-type="fig"}). {#fig02} Detection of PPRV by reverse transcription polymerase chain reaction (RT-PCR) ----------------------------------------------------------------------------- Clinical samples obtained from goats suspected of PPR were tested for the presence of PPRV by RT-PCR targeting the N gene using primers NP3 and NP4. Tested samples were scored positive if a PCR product of approximately 300 bp was visualized after agarose gel electrophoresis. Of the tested goats, 31.1% (*n* = 45) and 26.9% (*n* = 26) were positive for PPRV N gene in Mvomero and Ngorongoro districts, respectively. The rate of detection of PPRV was higher in tissues than in buffy coat, and ocular and nasal swabs (Table [1](#tbl1){ref-type="table"}). ###### Distribution of PPRV based on detected of the N gene using reverse transcription polymerase chain reaction (RT-PCR) assay in goat samples collected from Mvomero district located in eastern Tanzania and Ngorongoro district located in northern Tanzania. Depending on availability, more than one sample types from a single animal were tested Location Sample type Positive Negative Total Positive (%) ------------ ------------- ---------- ---------- ------- -------------- Mvomero Swabs 8 23 31 26.0 Whole blood 2 8 10 20.0 Tissues 5 0 5 100.0 Animals 14 31 45 31.1 Ngorongoro Swabs 2 10 12 16.7 Whole blood 3 7 10 30.0 Tissues 2 4 6 33.3 Animals 7 19 26 26.9 Phylogenetic relationship of PPRV --------------------------------- The N gene nucleotide sequences of Tanzanian PPRV obtained from this study were submitted at GenBank and provided with accession numbers KF939643 and KF939644 for Tanzania/Dakawa/2013 and Tanzania/Ngorongoro/2013 strains, respectively. Alignment of the N gene nucleotide sequences of PPRV obtained from Mvomero showed that the sequences were 100% identical. Similarly, the N gene nucleotide sequences of PPRV obtained from positive animals in Ngorongoro were found to be 100% identical. Therefore, single sequence from each area was selected for phylogenetic analysis. However, the 320 nucleotides long N gene sequences of PPRV from Mvomero and Ngorongoro had two nucleotide substitutions. Translation of the sequences showed that one nucleotide substitution was nonsense while the other substitution led to a change of an amino acid at position 60 (G→A). BLASTN analysis of both Dakawa and Ngorongoro N gene nucleotide sequences of PPRV showed 96% identity to PPRV from Sudan and Ethiopia. A phylogenetic tree based on N gene nucleotide sequences was constructed using the PPRV N gene sequences obtained in this study and other viruses representing the four PPRV lineages. The N gene sequences of PPRV from Mvomero and Ngorongoro districts clustered into Lineage III together with other PPRV previously reported from Ethiopia, Sudan, Oman and United Arab Emirates (Fig. [3](#fig03){ref-type="fig"}). {#fig03} Discussion ========== The circulation of PPRV in selected parts of Eastern and Northern Tanzania was investigated in the present study. The results obtained show that PPRV is circulating in goats in Mvomero and Ngorongoro districts as confirmed by molecular detection of PPRV genome, clinical signs and post-mortem findings in goats. Furthermore, molecular characterization of PPRV based on partial amplification of the N gene show that PPRV, belonging to Lineage III, is circulating both in Eastern and Northern Tanzania. Previous studies have confirmed the presence of PPR in Northern Tanzania based on serological detection of antibodies against PPRV in goats, sheep and cattle ([@b18]; [@b7]; [@b8]; [@b10]). However, PPR has not been detected in wild small ruminants in Tanzania neither on molecular nor serological diagnostic bases ([@b10]). The presence of PPRV in Southern Tanzania based on PPRV genome detection using RT-PCR assay has been confirmed ([@b17]). [@b8] sampled goats and sheep between 2008 and 2010 in Northern Tanzania and reported the circulation of PPRV belonging to lineage III. The present study reports the continued circulation of lineage III PPRV in Northern and its incursion in Eastern Tanzania. PPR is believed to have crossed into Tanzania through Northern Tanzania in 2008 ([@b18]; [@b7]). As PPRV obtained from the present study cluster with PPRV from Sudan and Ethiopia, it is reasonable to assume that probably, PPRV crossed into Northern Tanzania from these countries probably through Kenya ([@b8]). It is less likely that PPRV crossed into Tanzania through Uganda, as PPRV lineage I, II and IV but not lineage III have been reported in Uganda ([@b12]). Afterwards, it is believed that PPRV spread to other parts of the country from Northern Tanzania ([@b17]; [@b8]). The results obtained from the present study may indicate that PPRV has spread from Northern to the Eastern Tanzania because of the high genetic identity of the variable N gene. Further in-depth studies are required to elucidate the origin and transmission characteristics of PPRV in Tanzania and neighboring countries. Previous reports showed high seroprevalence of PPR (45.4%) in Northern Tanzania ([@b18]) and PPRV infection of 31.0% in Southern Tanzania ([@b17]). In the present study, PPRV genome was detected in 29.6% and 31.1% of the goats tested in Northern and Eastern Tanzania, respectively. However, this may not be a true reflection of the PPRV prevalence because only goats showing PPR clinical signs were sampled in the present study. PPRV could not be detected in all of the goats tested despite the fact that samples were collected from animals showing PPR clinical presentation. Previous studies have shown that the type of sample used during diagnosis of PPR, stage of infection and the type of gene targeted for RT-PCR may influence the level of positivity ([@b11]). In the present study, a high proportion of positive animals were detected when tissue samples were used in the diagnosis of PPR. However, PPRV was detected from various samples including oculonasal swabs, buffy coat and tissues (lymph nodes, intestines and lungs). The ability to detect PPRV in oculonasal swabs is an important finding because swabs taken from infected animals in the field not only provide a suitable source of viral RNA but also are not subject to the same storage and transport problems associated with tissue samples ([@b5]). Little success on preventing active PPRV infections has been achieved in Ngorongoro district despite multiple vaccination campaigns in the district. Probably PPRV active infections cross from wildlife because domestic animals interact with wildlife in Ngorongoro Conservation Area and the Loliondo Game Controlled Area. However such link has not clearly been established yet ([@b13]), which requires future extensive studies to understand the possible spread between domestic and wild small ruminants. In addition, there is high and frequent movement of goats between Tanzania and Kenya along the Northern border for trade. Trade of live animals at markets has been shown to be an important vehicle for transmission of infectious diseases ([@b4]). The role of animal markets in the transmission of PPR was also reported by [@b17], who found that inadequate infrastructure especially in local animal markets may be facilitating transmission of PPR. Taken together, we conclude that an active PPRV infection is present in Northern and Eastern Tanzania. The characterized strains of PPRV belong to Lineage III and are different from the Nig75/1 (Lineage II) currently being used as vaccine in the region. This study was supported by a grant from the Wellcome Trust (Grant WT087546MA) to the Southern African Centre for Infectious Disease Surveillance (SACIDS), Sokoine University of Agriculture, Morogoro, Tanzania and IPOS project at VIVES University College, Roeselare, Belgium. Research on PPRV in GM, JJW and MM laboratories is supported by a grant from Swedish Research Council (Grant 348-2013-6402). We thank Lawrence Mdimi, Emiliana Z. Biseko, Neema B. Kulaya and Francisca B. Gwandu for their assistance in sampling goats in Dakawa and Yassin Rajabu, Benjamin Kiboge and Idrissa S. Chuma for assisting the sampling of goats in Ngorongoro. TK and ESM were supported by a scholarship from the Wellcome Trust (Grant WT087546MA) to SACIDS, Sokoine University of Agriculture, Morogoro, Tanzania.

*Sr. Editor:* La epidemia por COVID-19 ha supuesto un reto para los sistemas sanitarios de todo el mundo, y ha generado un justificado interés en la comunidad médica por describir y explicar sus manifestaciones mediante un número creciente de artículos científicos, lo que ha supuesto un desafío para revistas, editores y revisores[@bib0050]. En el caso de la patología neurológica y el ictus, la COVID-19 también ha estimulado el espíritu publicador de la comunidad médica, con el objetivo fundamental de dar a conocer la experiencia acumulada y facilitar el trabajo del resto de centros, siempre con la máxima rapidez posible. Así, en una revisión reciente se ha demostrado que aunque la COVID-19 no aumenta el riesgo de ictus, sí incrementa la mortalidad de manera notable[@bib0055]. No obstante, debe ser un deber fundamental de todos los implicados en el proceso de publicación científica la búsqueda del máximo rigor y exactitud, con el fin último de asegurar la mejor asistencia a nuestros pacientes, y a este respecto nos gustaría realizar algunos comentarios sobre dos publicaciones recientes en su revista sobre ictus y COVID. En el reciente y útil artículo de Barrios-López et al.[@bib0060] se describe una serie de 4 casos con ictus agudo tratados en su centro, describiendo de manera exhaustiva los datos clínicos, analíticos y de imagen. No obstante, consideramos necesaria alguna puntualización sobre los estudios de neuroimagen presentados. Así, en la figura 1 se aporta un conjunto de imágenes posprocesadas denominadas «estudio de perfusión cerebral». Sin embargo, no se trata de imágenes fuente de TC, ni de mapas paramétricos al uso, sino de un código de colores fruto de un posproceso automatizado, sin que se aclare el software ni el algoritmo utilizados, datos fundamentales para que el revisor o el lector puedan interpretarlas adecuadamente. Es más, en el pie de foto se hace referencia a un «aumento del tiempo de tránsito medio», cuando en las propias imágenes se hace referencia en la captura de pantalla a Tmax y CBF. Efectivamente, el Tmax ha sido utilizado para cuantificar el volumen de tejido cerebral en riesgo de infarto o penumbra isquémica[@bib0065], [@bib0070], pero no debe ser confundido con el tiempo de tránsito medio o MTT, por sus siglas en inglés, que usan otros software y algoritmos diferentes al ---presumiblemente--- utilizado. El uso inadecuado de la semiología de los valores de perfusión cerebral puede conllevar un error en la interpretación y aplicación de los datos presentados. Así mismo, consideramos cuando menos poco ortodoxa desde el punto de vista de la semiología radiológica la descripción de la figura 2, pues denominar a la figura A como «corte bajo a nivel infratentorial» cuando aparecen gran parte de ambos lóbulos temporales y occipitales sin duda puede llevar a equívocos a los lectores menos experimentados. Paralelamente, Aguirre et al.[@bib0075] exponen de manera excelente un interesante caso de encefalopatía por contraste tras una trombectomía por oclusión de la arteria cerebral media izquierda. En este caso, en la figura 1 se aportan mapas paramétricos de perfusión de tiempo al pico y volumen sanguíneo cerebral, que permiten una correcta comprensión de la fisiopatología al lector. No obstante, la inclusión del código de colores y su correlación con los valores cuantitativos de tiempo y volumen hubiera permitido un acercamiento más preciso al estudio de neuroimagen, pues la gama de colores puede ser modificada durante el posproceso, de manera que sin una escala fija es difícil su valoración y cuantificación. En cuanto a la «imagen H» del TC a las 24 h, los autores hacen referencia a «datos de edematización», que para nosotros no son evidentes. No existe hipodensidad de la sustancia blanca, ni modificación en la diferenciación sustancia blanca-sustancia gris, ni efecto de masa[@bib0080], principales hallazgos semiológicos de edema cerebral en tomografía computarizada. Una mayor precisión en la localización del contraste extravasado (probablemente ganglios basales y córtex hemisférico) también hubiera aportado más exactitud a la publicación, y probablemente más información sobre la fisiopatología del cuadro del paciente. El objetivo de estos comentarios es poner en valor la neuroimagen en las publicaciones científicas, y defender que los neurorradiólogos pueden aportar precisión, experiencia y excelencia en las mismas, dentro del trabajo en equipo y multidisciplinar que debería suponer la publicación científica[@bib0085], [@bib0090].

Background {#Sec1} ========== This study examines whether level of education prior to imprisonment and participation in prison education programs are associated with better health for male and female inmates. We contribute to a large body of work linking educational attainment and adult health \[[@CR1]--[@CR3]\] in which individuals with higher levels of education report fewer chronic diseases, have better mental health, and enjoy longer lives than do adults with lower levels of education \[[@CR4], [@CR5]\]. The resource substitution theory hypothesizes that education will be more strongly associated with health among more disadvantaged populations, such as women relative to men, because education can serve as a "substitute" for the limited resources among these groups \[[@CR6], [@CR7]\]. However, because of the explicit focus on non-institutionalized populations \[[@CR8]\], little is known about the health benefits of education among male and female prisoners. The absence of this research is particularly important because the size of the prison population in the United States has increased more than sevenfold in the past 30 years and the U.S. now has the highest rate of imprisonment compared to all other countries \[[@CR9]\]. Importantly, the female prison population has been increasing at substantially higher rates than the male prison population \[[@CR10]\], incarcerated women have worse health \[[@CR11]\] and complex comorbid conditions \[[@CR12]\] across multiple domains compared to men. In addition, women's prisons have lower programming availability \[[@CR13]\], often struggle to meet the healthcare needs of the prisoners \[[@CR13], [@CR14]\], have higher exposure to sexual violence \[[@CR15]\], and women have specific healthcare needs that are not always provided for in prisons \[[@CR16], [@CR17]\]. As such, examining sex differences in the relationship between education and health among prisoners is important because it provides additional insights into the nature of the association. Methods {#Sec2} ======= Data come from the 2004 Survey of Inmates in State Correctional Facilities (SISCF), which provides a nationally representative sample of U.S. adults incarcerated in state prisons \[[@CR18]\]. The sample design employed a stratified, two-stage selection process. The prison sample was selected from a universe of 1,585 state prisons. Overall, 301 prisons were randomly selected for inclusion in the study. A total of 287 prisons participated (95.3 % prison-level response rate). In the second stage, inmates were randomly selected for participation (*n* = 14,499; 89.1 % respsonse rate). The interviews were conducted using computer assisted personal interviewing and participation was voluntary. Due to missingness on key variables, the final sample size is 13,290 (10,493 men and 2,797 women). Respondents were first asked about *lifetime* diagnoses and then a follow up question ascertained whether or not the respondents still have problems related to the condition (i.e., *current* morbidity). The 11 health conditions include hypertension or high blood pressure, diabetes, heart problems, asthma, kidney problems, stroke, all-cause cancer, arthritis, cirrhosis, sexually transmitted infections (STI), and hepatitis. Lifetime (α = 0.52) and current (α = 0.48) morbidity standardized scale scores are calculated to represent the presence of multiple health conditions. Education prior to entering the current prison term is measured with years of education (0 to 18 years). We further refine this measure by including a series of dummy variables assessing degree attainment by categorizing less than high school, GED, high school diploma, and at least some college. Current education examines whether an inmate participated in a high school education class or obtained a GED during their current prison sentence if they are eligible (i.e. enter prison with less than a GED level of education). All multivariate models control for age, race/ethnicity, employment prior to incarceration, and marital status. We also control for the number of years served to date during the current incarceration episode and whether this is their first incarceration episode. Our analyses are conducted in two parts. First, linear regression models estimate the association between education *prior to entering prison* and *lifetime* morbidity stratifying by gender and controlling for demographic background. Importantly, our models use prison fixed effects to control for contextual and compositional differences across prisons that may confound associations. The second analysis examines educational attainment within prison and includes only those inmates who entered prison with less than a GED level of education. Similar linear models with prison fixed effects estimate the association between attaining a GED *in prison* and *current* morbidity. We then follow with additional descriptive analyses to examine who participates in prison high school classes in more detail. Finally, a supplementary analysis examines the association between education and one of the morbidities: hypertension. All descriptive and multivariate analyses use sampling weights to adjust for the complex sampling design of the study. Results {#Sec3} ======= Table [1](#Tab1){ref-type="table"} presents the descriptive and bivariate (by gender) analysis. Incarcerated women have a significantly higher mean morbidity count than do men. Half of women (52.8 %) report at least one health condition while only one-third of men do (36.9 %). The top 10 % of the unhealthiest women have three current health conditions compared to two for men. The raw frequencies and weighted odds ratios indicate that women have worse health compared to men. Incarcerated women report significantly higher rates for most conditions at the *p* \< 0.001 level excluding hypertension - which is still significant at the *p* \< .01 level - and stroke and cirrhosis - which are not statistically significant.Table 1Health, Education, and Demographic Characteristics of Prisoners Stratified by Gender (Men: *n* = 10,493; Women: *n* = 2,797)MenWomen*p*WeightedConfidence*n* (%)^a^*n* (%)^a^ORInterval (OR)Mean Health Count (se).59 (.01).97 (.02)\< .001Hypertension2197 (21.2)649 (23.6)\< .0091.161.04, 1.29Heart Problems964 (9.3)317 (11.5)\< .0011.321.14, 1.51Diabetes448 (4.3)190 (6.9)\< .0011.751.45, 2.11Asthma1495 (14.4)620 (22.5)\< .0011.781.60, 1.99Kidney Problems592 (5.7)323 (11.7)\< .0012.291.97, 2.65Arthritis1603 (15.5)682 (24.8)\< .0012.061.83, 2.32Stroke475 (4.6)151 (5.5)\< .0531.341.02, 1.76Cancer181 (1.8)278 (10.1)\< .0015.693.71, 8.73Cirrhosis175 (1.7)49 (1.8)\< .7461.480.97, 2.27Hepatitis972 (9.4)432 (15.7)\< .0012.201.85, 2.61STI1226 (11.8)484 (17.6)\< .0013.002.08, 4.31Education Prior Mean Years (se)10.4 (.02)10.7 (.05)\< .001 Less than GED4334 (41.7)1150 (41.4)\< .7660.970.89, 1.06 GED2615 (26.1)555 (20.0)\< .0010.740.66,0 .82 High School Diploma2274 (21.9)606 (21.8)\< .9361.000.90, 1.12 Some College816 (7.8)328 (11.8)\< .0011.601.39, 1.85 College363 (3.5)142 (5.1)\< .0011.531.24, 1.90Mean Age (se)35.3 (.10)35.6 (.17)\< .203Race/Ethnicity White3901 (37.2)1261 (45.1)\< .0011.471.35, 1.61 Black4652 (44.3)998 (35.7)\< .0010.670.61, 0.74 Latino1369 (13.1)352 (12.6)\<.5180.870.77, 0.99 Other571 (5.4)186 (6.7)\< .0141.311.09, 1.57Employed Prior7402 (71.0)1586 (57.2)\< .0010.550.50, 0.60Marital Status Never Married6134 (58.9)1261 (45.2)\< .0010.580.53, 0.63 Married1576 (15.1)505 (18.1)\< .0011.241.11, 1.40 Widowed/Divorced/Separated2762 (26.4)1026 (36.8)\< .0011.641.49, 1.80Mean Time Served (se)4.6 (.06)2.5 (.07)\< .001First Time Incarcerated4796 (46.4)1629 (59.0)\< .0011.671.53, 1.83*OR* Odds Ratio, *CI* 95 % Confidence Interval, *se* Standard Error^a^Raw sample size and frequency unless otherwise stated Although the difference is small in magnitude, incarcerated women have a significantly higher mean years of education compared to men with men averaging 10.4 years of education and women averaging 10.7 years of education (Table [1](#Tab1){ref-type="table"}). Patterns of degree attainment prior to entering prison also differ by gender. While similar proportions of men and women enter prison with less than a GED (41 %), men are significantly more likely to have earned a GED (26.1 %) than women (20.0 %). About 22 % of men and women obtained their high school diploma prior to their current prison stay. Women, though, have significantly higher rates of some college (11.8 %) and at least a four-year college degree (5.1 %) compared to men (7.8 % and 3.5 %, respectively). Results from fixed effects (by prison) regression models examining education *prior to incarceration* and *lifetime* morbidity are presented in Table [2](#Tab2){ref-type="table"}. The findings indicate that increases in years of education are negatively associated with the standardized lifetime morbidity scale for both incarcerated women and men, but that the effect is greater for incarcerated women.[1](#Fn1){ref-type="fn"} Considering that the comorbidity scale is standardized within each sex, it suggests that the effect of one year of additional education among women (*b* = −.017, *p* \< .001) is more than four times the size of the comparable effect among men (*b* = −.004, *p* \< .05). To further evaluate the magnitude of these effects, we standardized the education scores and the same four-fold comparison is evident (*β*~*women*~ = −.04, *se* = .010) and men (*β*~*men*~ = −.01, *se* = .004) but the effect sizes are clearly small. Degree attainment prior to incarceration is only significant for women. Interestingly, women with a GED (*b* = −.088, *p* \< .001) have lower scores on the standardized lifetime morbidity scale compared to women entering prison with a high school diploma.Table 2Results from Regressing Level of Education Prior to Incarceration on Standardized Lifetime Morbidity Scale with Prison-Level Fixed Effect for Men and WomenLifetime Morbidity ScaleMenWomen*bpbpbpbp*Education Prior to Incarceration Years of Education−.004\*−.017\*\*\* Education (High School Diploma Ref.)  Less than High School−.006.036  GED−.010−.088\*\*\*  At Least Some College.002−.036Age.012\*\*\*.012\*\*\*.015\*\*\*.015\*\*\*Race/Ethnicity (White Ref.) Black−.029\*\*\*−.031\*\*\*−.077\*\*\*−.078\*\*\* Latino−.027\*−.027\*−.012−.002 Other.067\*\*\*.066\*\*\*.047.048Employed Prior to Incarceration−.046\*\*\*−.048\*\*\*−.031+−.039\*Marital Status (Never Married Ref.) Married.022\*.022+.038.034 Widowed/Divorced/Separated.020\*.020\*.042+.040+Time Served to Date−.001−.001.004.004First time Incarcerated−.015+−.017\*−.069\*\*\*−.069\*\*\*Model Statistics Number of Inmates9941991426572656 Number of Prisons2252256262 Overall R Squared.14.13.11.11 Rho.042.042.039.039+*p* \< .10; \**p* \< .05; \*\*\**p* \< .001 Table [3](#Tab3){ref-type="table"} presents the results for male and female prisoners who enter prison with less than a high school degree. As with the other analyses, we include prison-level fixed effects and sampling weights. Completing the requirements for a GED is associated with a lower score on the current morbidity scale for men (*b* = −.033, *p* \< .05) but not women. Table [4](#Tab4){ref-type="table"} examines who participates in high school education classes while in prison in more detail. Less than half (43.5 %) of men who entered prison with less than a GED (*n* = 4,334) participated in a high school education class while in prison. There are 60 men in five prisons where no one with less than a GED reported participating in high school classes. It is possible that these prisons do not offer these types of courses so they are dropped from this descriptive analysis. Men who participate in high school education classes have a younger overall age and are overrepresented in the age 18 to 29 category. There are 1,837 men between the ages of 18 and 39 who entered prison with less than GED education and almost half of them (49.4 %) participated in high school education classes compared to only 41.1 % of the 2,083 men between the ages of 30 and 49 and 34.5 % of the 354 men aged 50 years or older.Table 3Results from Regressing Attaining a GED in Prison on Standardized Current Morbidity Scale for Men and Women Entering Prison with Less Than a GED with Prison-Level Fixed EffectsCurrent Morbidity ScaleMenWomen*Bpbp*Attain a GED in Prison−.033\*−.065Age.010\*\*\*.016\*\*\*Race/Ethnicity (White Ref.) Black−.062\*\*\*−.078\* Latino−.071\*\*\*−.045 Other.011−.053Employed Prior to Incarceration−.023+−.027Marital Status (Never Married Ref.) Married.036\*.081+ Widowed/Divorced/Separated.019.015Time Served to Date.001.002First time Incarcerated−.001−.058\*Model Statistics Number of Inmates41281102 Number of Prisons22562 Overall R Squared.13.11 Rho.078.087+*p* \< .10; \**p* \< .05; \*\*\**p* \< .001Table 4Descriptive Analysis Examining Participation in High School Education Classes for Inmates who Enter Prison with Less Than a GED-Level of Education Stratified by GenderMen (*n* = 4,274)Women (*n* = 2,781)No ParticipationParticipate*t/x* ^*2*^*p*No ParticipationParticipate*t/x* ^*2*^*p*Age (se)34.8 (.23)32.4 (.23)7.5\< .00035.0 (.34)33.6 (.42)2.6\< .009Age 18 to 29930 (38.9)907 (48.1)36.0\< .000201 (29.4)181 (38.8)11.2\< .001Age 30 to 491226 (51.3)857 (45.4)14.7\< .000449 (65.6)265 (56.9)9.1\< .003Age 50 or older232 (9.7)122 (6.5)14.6\< .00034 (5.0)20 (4.3).3\< .593Race/Ethnicity White606 (25.4)548 (29.1)7.2\<.007220 (32.2)161 (34.6).7\< .399 Black1244 (52.1)956 (50.7).8\<.362304 (44.4)189 (40.6)1.7\< .191 Latino421 (17.6)274 (14.5)7.5\<.006116 (17.0)80 (17.2).01\< .927 Other117 (4.9)108 (5.7)1.5\<.22944 (6.4)36 (7.7).7\< .398Marital Status Never Married1510 (63.3)1299 (68.9)14.7\< .000366 (53.6)239 (51.2).5\< .489 Married343 (14.4)220 (11.7)6.8\< .009103 (15.1)79 (17.0).8\< .376 Widowed/Divorced/Separated535 (22.3)366 (19.4)5.3\< .021214 (31.3)146 (31.5).00\< .962Employed Prior to Incarceration1614 (67.7)1239 (65.8)1.8\< .176336 (49.1)240 (51.5).6\< .428Years of Education Prior to Incarceration (se)8.9 (.04)9.2 (.04)−4.6\< .0009.1 (.07)9.1 (.07).1\< .907Time Served to Date (se)4.0 (.11)5.3 (.13)−7.6\< .0001.9 (.13)3.0 (.20)−5.0\< .000First time Incarcerated1068 (45.4)899 (48.2)3.4\< .067389 (57.5)277 (60.0).7\< .401Prison Experience Violent Offender1128 (47.2)1072 (56.8)38.9\< .000161 (23.5)169 (36.3)22.0\< .000 Drug Offender501 (21.0)329 (17.4)8.4\< .004196 (28.7)122 (26.2).9\< .357 Job Training Program Participation429 (18.2)38.4 (723)214.2\< .000104 (15.4)150 (32.2)44.9\< .000 Visits from Family/Friends Past Month545 (23.2)566 (30.0)25.6\< .000143 (21.2)151 (32.4)18.0\< .000 Phone Calls from Family/Friends Past Week1910 (81.1)1628 (86.4)21.1\< .000510 (75.7)377 (80.9)4.4\< .037 Written Up for Any Violation1253 (53.4)601 (32.0)195.6\< .000405 (60.2)180 (38.6)51.2\< .000 Compared to men who do not participate in high school education classes, men who participate are overrepresented by whites (29.1 % vs. 25.4 %, *p* \< .01) and those who have never been married (68.9 % vs. 63.3 %, *p* \< .001). Participants have a higher average number of years of school completed prior to entering prison (9.2 vs. 8.9, *p* \< .001) and have been in prison longer (5.3 vs. 4.0, *p* \< .001). Results from a multivariate analysis (not shown) confirm these findings. Men who participate in high school education classes are also more likely to be violent offenders (56.8 % vs. 47.2 %, *p* \< .001) and less likely to be drug offenders (17.4 % vs. 21.0 %, *p* \< .01). They are more likely to also participate in a job training program (38.4 % vs. 18.2 %, *p* \< .001). These men seem to have higher levels of social support in the form of phone calls (86.4 % vs. 81.1 %, *p* \< .001) and visits (30.0 % vs. 23.2 %, *p* \< .001). Finally, they are less likely to be written up for a violation (32.0 % vs. 53.4 %, *p* \< .001). Forty-one percent of all women (*n* = 2,781) entered prison with less than a GED level of education. Among those women, 40.5 % participated in a high school education program with 16.9 % of participants obtaining a GED by the time of the interview. Similar to men, women who participate in high school education classes while in prison tend to be younger (33.6 vs. 35.0, *p* \< .01). Of the 382 women who entered prison with less than a GED education between the ages of 18 and 29, 47.4 % participated in a high school education class. Thirty-seven percent of the 714 women aged 30 to 49 participated and 37.0 % of the women aged 50 or older participated. None of the demographic variables other than age are associated with participating in a high school education class; however, time spent incarcerated is positively associated (3.0 vs. 1.9, *p* \< .001) which was also found in the multivariate analysis (not shown). The findings concerning the additional measures of the prison experience are similar to men. Violent offenders are more likely to participate in high school education classes (36.3 % vs. 23.5 %, *p* \< .001); although being a drug offender is not significant. Female inmates who participate in high school education classes are also more likely to participate in job training programs (32.2 % vs. 15.4 %, *p* \< .001) and to receive visits (32.4 % vs. 21.2 %, *p* \< .001) and phone calls (75.7 % vs. 80.9 %, *p* \< .05) from family and friends. Finally, participants are less likely to be written up for a violation (38.6 % vs. 60.2 %, *p* \< .001). Finally, the analyses thus far have assumed that each health condition is equal in its association with education, but this is likely not the case given different disease progressions and etiologies. Therefore, a supplementary analysis examines the relationship between education and hypertension more closely. Of all of the medical conditions included in this study, hypertension is most likely to be proximately influenced by education. Indeed it appears that education both prior to and during incarceration is negatively associated with hypertension for men. Controlling for age, race, marital status, years incarcerated and first incarceration episode with prison-level fixed effects, each year of education is associated with three percent lower odds (*p* \< .05) of having lifetime hypertension for men. For those men entering prison with at least a GED, their odds of lifetime hypertension are decreased by eight percent (*p* \< .10) controlling for all other factors. We extend our analyses of current hypertension and prison education among only those men who enter prison without a GED or higher degree in order to reduce bias associated with time order. Using a fixed effects logistic regression model, we find that men entering prison with less than a GED level of education who participate in a high school education program while in prison have 19 % lower odds of reporting current problems with hypertension (*p* \< .05)[2](#Fn2){ref-type="fn"} compared to those who did not participate in high school education classes conditioned on availability. For those who earned their GED while in prison, their odds of reporting current problems with hypertension are even lower (*OR* = .71; *p* \< .05). If we reduce this sample to men who are less than 30 years of age (the age group most likely to earn a GED in prison), attaining a GED is associated with 56 % lower odds of reporting problems with current hypertension (*p* \< .05).[3](#Fn3){ref-type="fn"} For women, years of education (*OR* .96, *p* = .247) and having a GED or higher (*OR* = .79, *p* = .163) prior to being incarcerated is not associated with lifetime hypertension. Neither participating in high school education classes (*OR* = 1.07, *p* = .840) nor obtaining a GED while in prison (*OR* = .87, *p* = .813) is associated with hypertension for women. Discussion {#Sec4} ========== Our study demonstrates the importance of education for health among incarcerated adults. Previous research has documented the many benefits of education for inmates which extends to local communities \[[@CR19]\]. While this study focuses on high school education, post-secondary education for inmates is also a growing concern. In 2015 the Obama administration and the U.S. Department of Education announced the Second Change Pell Pilot Program to test new models to allow inmates to receive Pell Grants and pursue the postsecondary education in order to "to create a fairer, more effective criminal justice system, reduce recidivism, and combat the impact of mass incarceration on communities" citing that "for every dollar invested in correctional education programs, four to five dollars are saved on three year re-incarceration costs" \[[@CR20]\]. Our study suggests that improved health may be an additional benefit through potential increases in learned effectiveness, health literacy and ability to engage in health promotion, all of which has the potential to improve community health \[[@CR21]\]. Although not the focus of our study, the findings also point to benefits in the prison experiences for inmates who participate in prison education classes including greater external social support and lower likelihood of receiving a violation. Given these multiple far-reaching benefits of education, prisons should consider expanding basic education for inmates. In this national sample, only about 40 % of inmates who entered prison with less than a GED-level of education participated in high school education classes by the time of the survey. Our study is the first to demonstrate the generalizability of the education-health association beyond the non-institutionalized population. These findings are important because they support the idea that education is a "fundamental cause" of health \[[@CR1]\] even among one of the most select groups in the United States in terms of both health and education. Our results are also consistent with previous research documenting the poorer health status of incarcerated women compared to men \[[@CR11]\] and the gendered nature of the education-health association \[[@CR22]\] where education serves as a protective resource more for women than for men. Further, although not the focus of this paper, it is important to highlight the relationship of education and health compared to the other demographic controls in the study. Specifically, while education operates in a comparable manner to other research of noninstitutionalized adults, the relationship between race and health and marital status and health both operate in directions that are opposite to general findings \[[@CR23], [@CR24]\]. Both black male and female prisoners have a lower cumulative morbidity count compared to white male and female prisoners and prisoners who are currently married have worse health compared to those who are not married. This is important because, again, it speaks to the highly select nature of the incarcerated population but it also indicates further evidence of the robustness of the education-health association. There are two additional points to consider. First, a comprehensive meta-analysis conducted by the RAND Corporation \[[@CR19]\] found that GED programs are the most common education programs in prison, yet all types of education programs available (i.e., GED, adult basic education, postsecondary, and vocational) notably reduce post-release recidivism. The report concluded, however, that data do not exist to evaluate dose--response effects or the specific program characteristics that benefit inmates. The findings from the current study suggest that when moving forward with stronger research designs, additional proximate and distal indicators of program efficacy, such as health-related outcomes, should be considered. These studies could also address causal ordering. In our study we conceptualized health as the outcome. But a more comprehensive study examining the life course of inmates can parse out the dynamic processes of education and health throughout the life span. It should also be noted that the data used in our study are from 2004 when educational programming was more widely available. The 2008 recession affected correctional programming leading to dramatic changes in the number of programs offered, the sizes of the classes, the modes of delivery, and the number of inmate participants \[[@CR19]\]. It is possible that our findings are influenced by period effects, although, research has consistently documented an education-health association in the general population across time and cohorts. Second, this paper discussed selection into prison as a function of gender and the gendered nature of the education-health association. We encourage future research to consider how race-based selection processes \[[@CR25]\] influence education and health in prisons. Regardless of race, high school drop outs are five times more likely to go to prison than high school graduates \[[@CR26]\] and national statistics show that blacks have among the lowest graduation rates for high school students \[[@CR27]\]. The double disadvantage of race and class inequality is striking for incarceration rates. Over 16 % of black men without a high school degree entered prison annually from 1995 to 2001 compared to just 3.4 % of white men without a high school degree, and this disparity grew from previous time periods \[[@CR26]\]. Other research has documented that the cumulative risk of imprisonment by age 34 for black men without a high school education is 68 % \[[@CR28]\]. This translates into 27 % of white prison inmates having not completed high school or their GED compared to 44 % of black prison inmates \[[@CR29]\]. Collectively, this research highlights the differential selection among racial and ethnic minorities into the prison system which may influence the education-health association. Limitations {#Sec5} ----------- This study has several limitations that need to be considered. First, the data are limited to inmates in state correctional facilities. While state prison inmates comprised 1.3 million of the 1.5 million prison inmates at midyear 2011 \[[@CR30]\], it is important to consider that our findings may not be generalizable to all incarcerated persons, especially those in local county jails. Second, this study relies on self-reported health conditions. However, self-report data are an essential and commonly used source of health indicators in research \[[@CR31]\], and the SISCF is the best data set available to answer the research question because it is the only large, nationally representative survey of inmates available in the United States. Third, the data are cross-sectional and do not provide information on onset of health conditions. That is, this study is unable to account for the timing of diagnosis or the severity of symptoms. This study is also limited by our inability to examine the neighborhood contexts that individuals are exposed to prior to prison. Such information may be particularly important since incarcerated persons are drawn from distinct geographic areas \[[@CR32]\] referred to as prison "feeder communities" \[[@CR33]\]. In other words, distinct sociodemographic communities bear the burden of mass incarceration including young poor people of color from disadvantaged neighborhoods \[[@CR32], [@CR34], [@CR35]\]. Therefore, it is not clear whether the study findings reflect education as mitigating the deleterious effects of imprisonment, or a prior association between poor health, low levels of education, and high propensity for incarceration. Conclusion {#Sec6} ========== Results from our study provide additional support for the notion that the association between education and health may be, in part, causally oriented. We do so by focusing on a highly select population and the gendered nature of the education-health association. We encourage future researchers to examine the proximate pathways through which this observed association may operate \[[@CR36]\]. This is especially important considering that incarcerated persons comprise a vulnerable, disadvantaged, and largely unhealthy subset of the U.S. population who may be reflective of the larger marginalized segments of society. Our findings are timely as prisons are having to address correctional healthcare practices. The identified health promotion needs of prisoners include education in health and empowerment, support in adopting health behavior, development of life skills, and education related to specific illnesses, among others \[[@CR37]\]. Our results support the notion that the provision of primary and secondary education to prisoners may be an important element for health promotion and increasing the life chances and longevity of prisoners after they are released back to their communities. The models were estimated using different functional forms of age including age^2^ and age^3^. For women, the coefficient remain unchanged (−.016). For men, the coefficient reduced slightly from−.038 to−.029. Given this, the most parsimonious model is presented. The sample size for this model is 3,498; 531 observations were dropped from the fixed effects model because the prison did not have variation in the outcome variable (current hypertension). The sample size for this model is 390; 786 observations were dropped from the fixed effects model because the prison did not have variation in the outcome variable (current hypertension). The authors are grateful to researchers at the CUPC who made insightful comments on earlier versions of the paper. Publication of this article was funded by the University of Colorado Boulder Libraries Open Access Fund. Funding {#FPar1} ======= Support for this study was provided to the lead author by the NIH Ruth L. Kirschstein National Research Service Award Individual Fellowship (F31 DA037645) funded by the National Institute on Drug Abuse (NIDA), the National Science Foundation (NSF) SBE Doctoral Dissertation Research Improvement Grant (\#1401061), and the NIDA-funded Interdisciplinary Research Training Institute on Drug Abuse at the University of Southern California (R25 DA026401). Additional support was provided to the authors by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) funded University of Colorado Population Center (R24 HD066613). The NIDA, NSF, and NICHD had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. Availability of data and materials {#FPar2} ================================== Data from the Survey of Inmates in State and Federal Correctional Facilities is publically available from the Interuniversity Consortium for Political and Social Research (ICPSR) at the University of Michigan. Authors' contributions {#FPar3} ====================== KM organized and analyzed the data and tabulated the results. KM, RM, and JB all contributed to the conceptualization and writing of the paper. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable Ethics approval and consent to participate {#FPar6} ========================================== Not applicable

{.512} {.513} {.514}

1.. Introduction ================ Advancements in wireless communication modules and Micro-ElectroMechanical systems (MEMs), as well as the huge business potential of widespread low-cost sensing (Internet Of Things---IOT) have motivated the development of small and low power modules equipped with sensors and radios, that are replacing traditional wired sensor systems. These modules can communicate with each other by radio to receive and transmit data and form Wireless Sensor Networks (WSNs). In the last decade, the landscape of WSN applications has been extending rapidly in many fields such as factory and building automation, environmental monitoring, security systems and a wide variety of other commercial and military areas. Platforms for WSNs, including processors, sensors, radios, power supplies, operating systems and protocol stacks, are almost as diverse as their application areas, with only a few standards (such as ZigBee \[[@b1-sensors-14-11070]\], 6LoWPAN \[[@b2-sensors-14-11070]\], *etc.*) that mostly address the lower levels of the radio protocol stack, and not the application API. Furthermore, most of the available sensor nodes on the market (such as Mica \[[@b3-sensors-14-11070]\], Tmote Sky \[[@b4-sensors-14-11070]\], MotionBee \[[@b5-sensors-14-11070]\]) only provide a few on-board blinking Light Emitting Diodes (LEDs) for debugging. Although WSNs have experienced great advancements in the last decade, application development in this domain is still quite challenging and time consuming, particularly for application domain specialists who may not be familiar with low-level processor programming and radio control details. There is a lack of tools that provide modeling, Hardware-In-the-Loop (HIL) simulation, and automatic code generation capabilities for multiple platforms from a single high level abstraction. Just like other embedded systems, WSN applications need to be verified functionally before being implemented on the actual platform in order to find as many bugs as possible at a high level, where fixing them has a lower cost. Moreover, the concept of bridging physical nodes with simulated nodes in an integrated framework is still absent from the WSN domain. Bridging the physical world with the virtual one often broadens the possibilities of accelerating and easing embedded system design. This is even more true for WSNs, where generally the developed applications need to be tested and executed in scenarios involving hundreds to thousands of nodes. Since it is hard to manage physical test beds with huge numbers of nodes, the most common solution is to rely on simulation frameworks that allow the developers to create virtual sensor nodes and then provide various abstraction mechanisms (from languages like C to graphical programming environments) to specify the application which will be executed on the nodes. The foremost drawback of this kind of simulation is the absence of direct interfaces with the physical environment, which allow one to include physical details (like the radio or the sensor interfaces) that are hard to model at a high level. The available functional analysis packages, such as TOSSIM \[[@b6-sensors-14-11070]\] for debugging of TinyOS \[[@b7-sensors-14-11070]\] applications, or OMNeT++ and NS for general-purpose networked applications, fall into two main categories. One is very platform- and OS-specific (such as TOSSIM), while the other includes generic network simulators (such as OMNeT++, NS, *etc.*). Both have significant drawbacks when it comes to complex application development. The first group makes it very expensive to port an application to a different platform (e.g., from TinyOS to a ZigBee \[[@b8-sensors-14-11070]\] compliant platform). The second group still leaves a lot of detailed platform-dependent code to be developed and debugged. Integrated use of a network simulator followed by a platform simulator is the most commonly used path, but it still requires one to port code between a number of environments. Moreover, in case a bug is found at the end, one has to resort to low level debugging, which is extremely time-consuming due to the need to manually maintain the consistency between the various code levels. Our contribution is aimed at easing platform-independent WSN application development without compromising the efficiency of the final implementation. We propose a model-based framework built primarily on top of the MathWorks tools \[[@b9-sensors-14-11070]\]. The engineering approach used within the framework follows a development flow consisting of four different steps. In the first phase, called "design phase," application-level embedded software can be designed and developed using a platform-independent abstraction, namely the Stateflow and Simulink modeling languages, supported by various tools from the MathWorks platform \[[@b9-sensors-14-11070]\]. To remove platform dependency from modeling, we developed a set of generic interfaces (for sending/receiving packets and also for acquiring data from different sensors) and an event-based communication mechanism (based on Stateflow/Simulink signals), which ultimately allows developers to specify applications in Stateflow without knowing in detail the target software and hardware platform. In the second phase, called "simulation," such models can be configured to execute within a fully simulated environment, including very abstract radio communication models and simulated sensors. In the third phase, called "hybrid simulation," exploiting the HIL extensions that we developed, it is possible to replace some simulated nodes with physical ones, in order to easily verify hardware-dependent features. In the fourth phase, the same Stateflow/Simulink models can be used to automatically generate code for different WSN platforms (such as TinyOS, or Ember ZigBee), that run on real-world WSN nodes. The hardware dependent features used in the last two phases are the same: for such reason it is also possible to create WSN deployments consisting of any combination of simulated, partially simulated and real WSN nodes. Thus by using this framework, users can build a hybrid network consisting of virtual and real nodes and then simulate it as a whole. A conceptual view of the simulation scenario is sketched in [Figure 1](#f1-sensors-14-11070){ref-type="fig"}. In this paper, we describe a comprehensive model-based design, simulation and code generation framework for Wireless Sensor Networks by extending work described in \[[@b10-sensors-14-11070],[@b11-sensors-14-11070]\]. In \[[@b10-sensors-14-11070]\], we proposed a single node HIL simulation framework and automatic code generation technique, aimed at providing model-based design capabilities to WSNs. Then in \[[@b11-sensors-14-11070]\] we described a hybrid simulation framework, which mixes virtual nodes with physical nodes. The novel contributions of this paper with respect to \[[@b10-sensors-14-11070],[@b11-sensors-14-11070]\] are in several directions. Firstly, we provide a holistic view of the whole framework by describing both high level application modeling and HIL simulation. In this phase, taking an application of body sensor network as an example, we illustrate in detail how a WSN application can be modeled using our proposed framework. We also provide details of the HIL simulation and automatic code generation process for multiple platforms. Finally, we provide a comprehensive summary of related work. An earlier version of our proposed framework was described in \[[@b12-sensors-14-11070]\]. That approach suffered from lack of modularity and provided limited support for HIL interactions with the simulation model. In this paper we build upon the foundation described in \[[@b12-sensors-14-11070]\], in order to improve modularity and portability, as well as to provide more complete and orthogonal HIL interfaces. The driving goal of this extension is to provide WSN developers with tools to develop platform-independent WSN application through model-based abstractions, test the implementations through simulation, extend the realism of simulations through HIL (thus introducing hardware dependent features) and finally deploy the code on the nodes, through platform-dependent code-generation. The framework is aimed at improving: The time it takes from specification to implementation, by reducing the number of iterations (due to fewer bugs in automatically generated code, and due to a more compact high-level model) and by reducing the manual effort required to obtain the simulation and implementation code.The time it takes to obtain derivative designs, again since the model is more abstract than C code and hence more reusable. Both of these claims are substantiated by a large amount of past work on Model-Based Design (MBD) and Hardware-In-the-Loop (HIL) simulation. *The main contribution of this paper is to show how MBD and HIL can be used for WSN design*, as past work has shown, e.g., for the domain of automotive design. In particular we show that the abstractions (signals, states, transitions, *etc.*) that are useful for MBD of embedded control applications, and that are provided by Simulink and Stateflow, are very close to what is needed for MBD of WSNs. So the only effort required to re-use productively these tools for WSNs is to provide the abstraction layer that we describe in detail in this paper. The remainder of the paper is organized as follows: in Section 2, a review of related work is presented; in Section 3 the overall architecture of the proposed solution is outlined; in Section 4, we illustrate platform independent modeling using a comprehensive example. We outline the HIL methodology in Section 5 and multi-platform code generation in Section 6. In Section 7, we describe how the framework has been tested in a HIL simulation scenario, including code size comparisons between manual and automated implementation. Finally in Section 8, we present conclusions and future research directions. 2.. Related Work ================ In the last decade, much work has been contributed to ease the development of WSN application design using high level abstractions, models and simulation frameworks. According to the established taxonomy in \[[@b13-sensors-14-11070],[@b14-sensors-14-11070]\], they can be classified into two main types (as shown in [Figure 2](#f2-sensors-14-11070){ref-type="fig"}) based on the programming abstractions, namely: node-centric (local behavior) approaches and macro-programming (global behavior) approaches. In node-centric programming approaches, the global application is decomposed into a set of concurrent local node behaviors, and the programmers have to develop explicitly code running on individual nodes (often called "motes"). The node level programming pattern can be further divided into two groups, namely OS-based programming and virtual machine/middleware-based programming. The OS-based programming pattern offers a flexible control of the hardware resources to the WSN application developers by allowing them to directly interact with the device hardware abstractions. TinyOS (with the associated programming language nesC) \[[@b7-sensors-14-11070]\] is one of the earliest examples in this class and has been widely adopted as the software platform running on numerous WSN devices. There are also other commonly used WSN operating systems utilized by the researchers, such as ContikiOS \[[@b15-sensors-14-11070]\], MANTIS \[[@b16-sensors-14-11070]\], SOS \[[@b17-sensors-14-11070]\], *etc.* The appearance of light-weight Java virtual machines, e.g., Darjeeling \[[@b18-sensors-14-11070]\] and Squawk \[[@b19-sensors-14-11070]\], allows designers to develop WSN applications using a high level programming language, with slightly reduced application code performance due to the extra resources consumed by the virtual machine. Other techniques based on virtual machine and middleware are also available to relieve the designers from concerns due to the low level aspects, such as a UML-based approach using a virtual machine. Generally, the virtual machine is an execution environment running on top of the operating system. Macroprogramming approaches introduced a completely different way to develop applications for WSNs. In this kind of approaches, the developers program a flock of motes as a whole, rather than by explicitly writing the software that will run on individual nodes, thus focusing on the overall functionality rather than on implementation and communication details. This can provide flexibility in optimizing the system performance and relieve application developers from dealing with concerns at the mote level. Depending on the scale of the programming entity, the macroprogramming approaches can be classified as group-level (part of WSN) or network-level (entire WSN). Various criteria could be used to define this macroprogramming entity to simplify application development by dealing with data sharing and data aggregation. In macroprogramming, each node belonging to a given group-level programming entity could be considered as a "neighbor" node for other nodes in the same entity. When the group is constructed according to physical closeness (defined by topological distance, *i.e.*, number of communication hops, or geographical distance), it can be called a "neighborhood based group", such as in Abstract Regions \[[@b20-sensors-14-11070],[@b21-sensors-14-11070]\]. On the other hand, if nodes are grouped based on some high level logical properties (e.g., same node type or same physical quantity to be measured), then the group is called a "logical group", as in EnviroTrack \[[@b22-sensors-14-11070]\]. The APIs exposed by a programming group (both neighborhood-based group and logical) can hide the communication inside that group, thus making it easier for the developers to design a collaborative algorithm and analyze its performance. When the group is extended to enclose the entire WSN, then it becomes a network-level programming entity. Solutions such as COUGAR \[[@b23-sensors-14-11070]\], TinyDB \[[@b24-sensors-14-11070]\] and Spine \[[@b25-sensors-14-11070]\] employ this kind of approach. They leverage distributed database techniques and extensible processing and query mechanisms to abstract the underlying network as an entity. Within these solutions, a task is described in high-level languages, injected in the network and transformed into low-level procedures running on each individual node. However, macroprogramming partially reduces the possibility to obtain a fine-grained control over application logic due to the limited expressiveness of high-level task description languages. In contrast to the node-centric and macroprogramming approaches, in this paper we provide rich abstractions that allow one to model application behavior still at the node level, yet in a platform-independent and thus more abstract fashion that classical node-centric approaches. Moreover, users can exploit HIL simulation to connect with real world sensors and networks to perform hybrid network analysis. Finally, the behavior that has been modeled and refined during the simulation phase can be used to generate application code for WSN platforms such TinyOS, Mantis, a commercial ZigBee stack and others. To the best of our knowledge, there is no framework available that can provide these capabilities for WSN application development. Simulation is a de-facto standard first-step in the implementation of a WSN-based solution. However, it is not completely reliable, for many different reasons. On one hand, the degree of realism of the simulation depends on the complexity of the underlying simulation model, which is often based on unrealistic over-simplified assumptions (e.g., regarding the communication models). On the other hand, code executed by the simulator is typically different from the software running on actual WSN nodes, mainly because of differences in the HW platform, which introduces discrepancies regarding simulation timing, concurrency and performance. Finally, even in cases where the simulation code is very similar to the actual code (such as in \[[@b6-sensors-14-11070]\]), the simulation necessarily behaves differently from actual code regarding access to platform-dependent components, such as radio transceivers or sensors. For such reasons, WSN developers are usually forced to engage in time-consuming debug sessions when they move from the simulation to the actual deployment phase. Our hybrid simulation approach, mixing real and simulated objects through HIL interfaces, helps reducing efforts in this intermediate phase. Hybrid Hardware-In-the-Loop simulation is a consolidated approach in several embedded system application domains, such as industrial automation, automotive, and so on \[[@b26-sensors-14-11070]--[@b28-sensors-14-11070]\]. In hybrid simulation, some parts of the system are simulated and other parts are real-world components and sub-systems, integrated through HIL. A co-simulation framework usually is in charge of controlling the interactions between the simulated and the real world. Despite complexity and technical difficulties, hybrid simulation is valuable because it can couple the realism of actual deployment with the flexibility of simulation, in particular keeping the former small scale, while the latter deals with large numbers of nodes. Hybrid simulation has already been used in WSNs, for both testing and debugging purposes. In \[[@b29-sensors-14-11070]\] hybrid simulation is used specifically for testing embedded software for TinyOS-based applications, using a wireless-based control channel. In EmStar/EmTOS \[[@b30-sensors-14-11070]\], hybrid simulation (called "emulation mode") is used as a means to enhance simulation with real radio channels. In \[[@b31-sensors-14-11070]\], hybrid simulation is used to extend TOSSIM capabilities, employing a time-freeze strategy to keep the simulation running at pace with the real world. In \[[@b32-sensors-14-11070]\] the same approach is used, but with the opposite purpose, *i.e.*, to "augment" a real network with a set of simulated nodes. In \[[@b33-sensors-14-11070]\], the authors propose a toolset to support application development for a novel FPGA-based sensor node platform. In this hardware/software co-design approach, a flexible hardware/software boundary can be specified by the user (based on the mechanism of "late binding" of application components) in order to optimize the overall performance. However, the approach is specific to the proposed experimental platform and still requires the application developers to master knowledge about operating systems, communication protocol stacks and integrated circuit design. In \[[@b34-sensors-14-11070]\], a comprehensive extensible meta-data specification and the corresponding meta-model are proposed which support the description of components and configuration of WSNs. Based on this approach, various types of networks can be described at different levels of detail in a service-oriented style, which still requires from the application developers a lot of knowledge of network protocols and platform specifications. In \[[@b35-sensors-14-11070]\], the authors propose a model-based optimization framework dedicated to wireless body sensor networks (WBSN). Based on the research and analysis of the most energy-demanding components in WBSN applications, a multi-objective optimization algorithm is proposed in to find the optimal tradeoffs available in the design space. However, they do not discuss how to manage network simulations and HIL simulation using MBD features. In \[[@b36-sensors-14-11070]\], the authors introduce SenseWeaver, a SysWeaver plug-in that supports model-based design of wireless control applications. They propose a top-down model-based design approach to create wireless sensor-actuator networks. The approach manages complexity and enables the automatic integration of multiple applications. The component libraries allow code to be cleanly integrated and reused across different applications, thus reducing the amount of hand written code and the development time. Even though as described above research has made significant progress towards delivering an integrated tool suite for WSN application modeling, simulation and automatic code generation, many aspects still require more work. Indeed, many of the above mentioned tools and methodologies are not suited for describing complex designs or lack some modeling aspects which are essential to create a fully integrated co-design environment which can actually be used in an industrial setting. As a typical example we can cite approaches based on SystemC modeling \[[@b37-sensors-14-11070]\], which are simple to use and powerful, but require a virtual machine on the node to execute the code, and hence are not optimal for low power and high performance applications. Similarly, while UML-based tools are in widespread use to model and document enterprise software, their adoption in the embedded space is still limited. For this reason, even though we could have started our MBD approach from UML or SysML, in this work we propose using a widely known and widely used industrial language, namely Stateflow/Simulink, as the starting point of our work. Please note that most ideas behind our approach are fully applicable to a UML-based flow. 3.. Hybrid Simulation Framework =============================== The architecture of the framework, implemented using Simulink/Stateflow \[[@b10-sensors-14-11070]\], is depicted in [Figure 3](#f3-sensors-14-11070){ref-type="fig"}. A brief description of the functionality of each component in the framework is given below. The top level contains three "super" blocks which respectively hide and abstract: (1) sensing/actuation; (2) application functionality and (3) communication. 1. **Nodes Block**. This is a container block which abstracts application functionality as a set of cooperating processes (*i.e.*, Stateflow StateCharts or code running on physical nodes). It contains a separate instance for each node included in the scenario. Node objects can be fully simulated, partially simulated or independent. a. Fully simulated nodes are completely handled inside the simulator, as Stateflow Statechart instances. b. Partially simulated nodes are modeled inside the simulation framework but are able to access HIL components (such as sensors, or transceivers) which reside on physical devices. c. Independent nodes are real physical nodes, running code generated from Stateflow, which interact with fully or partially simulated nodes only through the radio channel and a framework-provided interface with the Super-Medium block. 2. **Super-Medium Block**. This is a component devoted to managing all radio communications within the scenario. In particular, it is able to dispatch packets within the simulated world and also between the simulated and the real world, via one or more physical nodes connected to the simulation host, as described below. 3. **Super-Sensor Block.** This is a component devoted to managing all sensing aspects within the scenario. It is able to provide simulated sensor readings and means to access actual sensors through HIL interfaces. In the following subsections, the Nodes, Super-Medium, and Super-Sensor blocks are illustrated in more detail. 3.1.. Nodes Block ----------------- The Nodes Block is an abstraction that contains nodes which will interact with each other in the application scenario. Each node is modeled as a separate instance, structured as shown in [Figure 4](#f4-sensors-14-11070){ref-type="fig"}. The Node Application is a block representing the platform-independent functional model of the WSN application running on each node. The Node Application can interact with the rest of the world through the Packet Reader and Sensing Response Reader, providing a direct connection with the Super-Medium and the Super-Sensor blocks respectively. The Packet Reader behaves as a radio receiver: upon detection of a radio packet, it generates an event called PKT for the Node Application instance, making the payload of the received packet available to it (we will see below how packet transmission is modeled). Similarly, the RxSensRep event is used by the Sensing Response Reader to notify the application of the availability of a sensor reading. A System Clock synchronous signal, finally, provides global timing information to the node components. The "logical" interface to communicate between the node application and other framework components (sensors and medium) is depicted by the block diagram in [Figure 5](#f5-sensors-14-11070){ref-type="fig"}. The node application can communicate with both the "Super-Medium Block" and "Super-Sensor Block" using platform independent APIs to get and put radio packets and sensor data, which can be used transparently in simulation, HIL and code generation modes. The users are able to develop the node application independent of the simulation setup and the node implementation platform, using these APIs and events (e.g., PKT and CLK). These API calls will be translated to the target platform (in the form of OS calls, driver calls, *etc.*) during the code generation phase. Although each node instance is structured according to the same scheme, its actual behavior within the simulation depends on its type. Five different node types are supported in the framework: SIM, HIL_SEN, HIL_RF, HIL_FULL and REAL. The types differ on how they interact with the Super-Medium and the Super-Sensor, as described in the following. Note that the application model is totally unaware and independent of the node type, since the framework handles these aspects transparently. - SIM (SIMulated): this fully simulated node uses both virtual sensors and a simulated radio transceiver. - HIL_SEN (Hardware-In-Loop SENsor): this partially simulated node uses the simulated radio transceiver, but collects sensor data from actual sensing devices through the Super-Sensor. - HIL_RF (Hardware-In-Loop Radio): this partially simulated node collects data from virtual sensors, but uses the actual transceiver through the Super-Medium for sending and receiving packets. - HIL_FULL (Hardware-In-Loop FULL): this partially simulated node executes the application code within the simulation environment, but uses both actual sensors and the actual transceiver. - REAL: this fully independent node, existing only in the physical world, can have active radio communication with other types of nodes, thanks to the Super-Medium, which uses a stub physical node to communicate with it. The application running on it could be coded manually (*i.e.*, using platform-dependent programming languages and operating system, such as TinyOS), or it can execute the automatically generated code from the Stateflow model of the Node Application. During the initial scenario definition phase, users can instantiate any number and combination of the aforementioned node types. For partially simulated nodes, the WSN developer can specify the association between any WSN node instance and HIL interfaces in the Super-Medium and Super-Sensor blocks. This relationship associates partially simulated nodes with actual HIL radio transceivers or sensors. Note that multiple HIL nodes can be associated with the same physical device, but this procedure must be handled carefully, because it can generate inconsistencies (e.g., it is not possible to send two radio packets at the same time using the same radio physical radio transceiver). The Super-Medium node currently resolves the resource contention by dropping the collided communication (which is realistic if the HIL nodes are meant to be in radio contact with each other). The main idea behind our framework is to provide a simple, yet modular and flexible approach to extend high-level, WSN application simulation with HIL, letting the WSN developer specify the hybrid simulation configuration on-the-fly. 3.2.. Super-Medium Block ------------------------ The Super-Medium block manages the exchange of packets within the framework, including any combination of fully simulated, partially simulated, and real nodes. It works by performing "read" and "write" operations over tuples of incoming and outgoing buffers, one tuple per node in the scenario. Packets are exchanged using intermediate buffers which work like placeholders between different components. Depending on simulation parameters such as: (1) the type of the transmitting node; (2) the type of the receiving node; (3) the type of link between the two components; and (4) the simulation radio model configured by the user, a different kind of packet dispatching is performed. For instance, a packet directed to a REAL node might be transferred directly to an actual device through the HIL interface, while a packet exchanged among two SIM nodes is first processed by the simulated radio channel model and then, in case of success (no packet error), simply transferred between the two virtual nodes. A snapshot of the Super-Medium Simulink block is presented in [Figure 6](#f6-sensors-14-11070){ref-type="fig"}. It is composed of four sub-blocks, namely the Simulated Channel and the HIL Channel, plus two blocks used to handle communication from and towards nodes. The role of such blocks is to collect all packets sent by node instances, process them either via simulated radio models (Simulated Channel) or using actual radio channels (HIL channel) and then notify the packet reception event to all nodes that receive packets. ### 3.2.1.. Simulated Channel Currently the Simulated Channel uses a simple implementation based on a link-quality matrix specified in the configuration phase, containing packet loss probability among any couple of nodes. Such implementation can be replaced by other simpler or more complex radio model implementations. The current implementation does not take into account possible interference among transmitted packets, but this can be easily added into the model. ### 3.2.2.. HIL Channel The HIL Channel handles radio communication from and to actual radio nodes. Its role is to intercept and handle packets to be broadcast on the physical radio interface to an HIL_RF, HIL_FULL, or REAL node. The HIL Channel controls, through serial cables, one or more physical nodes used as radio interfaces. A software stub running on the physical devices is responsible for dispatching radio packets from the physical interface towards the framework and vice-versa. When the HIL_Channel component receives a packet through the HIL interface, it is able to decode the packet header and dispatch the packet to the correct destination node (or nodes) according to the association between the node ID and the HIL interface. 3.3.. Super-Sensor Block ------------------------ Since sensor interactions are usually local to a node, its design is much simpler than the Super-Medium, though they share some similarities. Each node instance is provided with an incoming sensor data buffer. When a new sensor reading is available, the Node Application is notified. As described before, the Super-Sensor node is able to receive sensing requests from all node instances. Upon reception of a request, the Super-Sensor generates a sensing value in different ways, according to the node type and the request type. In case of nodes with simulated sensors (SIM, HIL_RF), the sensor reading is generated according to a simulated model (e.g., a sample data file) defined in the initial configuration phase. In case of nodes with HIL sensors (HIL_SEN or HIL_FULL), a request is routed to a specific HIL interface connected to a physical node (which can be the same as the one used by the Super-Medium for the radio, or a different one). A snapshot of the Super-Sensor Simulink block is depicted in [Figure 7](#f7-sensors-14-11070){ref-type="fig"}. The Super-Sensor is composed of three sub-blocks: Simulated Sensor, HIL Sensor and a dispatcher handling data exchange between sensors and the node application. 4.. Platform Independent Algorithm Modeling: An Application Case Study ====================================================================== In this section, we provide a detailed example of how we used Stateflow/Simulink to model a non-trivial WSN algorithm (which could belong to the application or middleware layers) inside the Node Application block (shown in [Figure 4](#f4-sensors-14-11070){ref-type="fig"}). Faithful to the model-based design approach, the application developer uses Stateflow constructs (such as states, transition diagrams, events, function calls *etc.*) which are independent of the specific platform and programming languages (such as TinyOS or implementations of the ZigBee stack) which will be used on the physical nodes. Interactions with the radio channel and the sensors are performed via function calls and events which are part of the framework. As mentioned above, each node can interact with virtual (simulated) or physical (connected by HIL interface) radio and sensors. We provide abstract interfaces for accessing transparently both the virtual and the physical components (radio and sensors), which decouple high-level application modeling from simulation scenario configuration. Hence we can use this platform-independent application model for automatic target code generation as well as for simulation. To illustrate our flow, we use a realistic application that consists of a virtual machine oriented to data processing in Body Sensor Networks which is very similar to Signal Processing in Node Environment (SPINE) \[[@b25-sensors-14-11070]\]. A SPINE node is able to perform dynamically configurable signal processing computations (such as max, min, median, *etc.*) based on collected data sets from sensors. The parameters of these computations (called "features") such as sampling rate, computation window, shift of data set at each new sample, *etc.* can be tuned and the features can be activated or deactivated depending on the application demands. One of the prominent applications of SPINE is to detect body movements, hence a three axis accelerometer is used as an example in our SPINE implementation. Our model of SPINE in Stateflow contains three parallel state machines and 23 Stateflow functions, which implement the following main functionalities: **Scheduler:** It manages the active tasks of the system. A task can be configured to perform the following actions reading a specific sensor at a specified interval (sampling rate), storing the sensor readings into a circular buffer and then calculating features (for example max, median, *etc*.) using the stored data based on a given window/shift value. After computing a feature, it sends the result to the base station via the radio.**Circular buffer:** It is used to store sensor readings in so-called segments. Each segment is used to store sensor readings from a specific sensor and channel (for example the x, y or z axis values of the accelerometer).**Packet processor:** It decodes incoming configuration packets that are sent by the base station to configure the activities of the sensor node.**Features:** Our SPINE model can perform some computations (such as max, min, median, *etc*.) on the data sets stored in circular buffers. The three parallel state machines (taskProcessingEngine, pktProcessingEngine and scheduler) process external events like CLK, PKT, RxSensRep and internal ones like TASK. A snapshot of these state machines is shown in [Figure 8](#f8-sensors-14-11070){ref-type="fig"}. Initially, these three state machines start in parallel and after initialization wait for incoming events to be processed immediately. The base station activates tasks by sending several configuration packets. When pktProcessingEngine receives a PKT event (generated by serial_port_packet_reader), it immediately calls getPktData which copies the packet payload into a local buffer (packetBuffer). Afterwards, it calls parsePktData (shown in [Figure 9](#f9-sensors-14-11070){ref-type="fig"}) which processes different types of configuration packets. In our SPINE implementation, we process four different types of packets that are used to configure the basic SPINE virtual machine: **Packet Type 3:** It contains data that are used to configure the sampling time of individual sensors.**Packet Type 5:** It contains general information to define features such as window (number of samples needed to compute a feature for the first time) and shift (number of samples needed to compute a feature after the first time).**Packet Type 7:** It contains data about the features (max, mean, *etc*.) that need to be activated or deactivated. It also contains information on which sensor and channel (x-axis, y-axis, *etc*.) must be used for these features.**Packet Type 9:** This packet is used to start/stop all the tasks managed by the SPINE engine. For example, The Stateflow function setupSampTimeSensor parses packet type 3. It extracts the sampling scale (millisecond, second or minute) from the payload data. Afterwards, it extracts the sensor code and sampling coefficients and then calculates the sampling time for that sensor and inserts it in the active sensor list. Similarly, setupParamsForSensor extracts the window and shift value for each active sensor, and setupFeature selects the features that need to be calculated. The setupFeature function also assigns which portion of the circular buffer will be used for each specific feature. Finally, the startSpineApp function adds all the features as active tasks in the scheduler. The scheduler, after initialization, waits for the CLK event. At each CLK event, it increments the system timer count and then calls updateTasks (shown in [Figure 10](#f10-sensors-14-11070){ref-type="fig"}). Inside updateTasks, it checks the sampling time for each active task. When a task needs to acquire data from a sensor, it is activated by generating a TASK event for the taskProcessingEngine. The taskProcessingEngine waits for TASK events to process the active task by calling the processingTask function. Inside processingTask, it decides whether the type of the task is an alarm or a feature. If it is a feature, it calls acquireSensorData (shown in [Figure 11](#f11-sensors-14-11070){ref-type="fig"}) to collect data from sensors. The acquireSensorData function uses framework-provided calls to read simulated or physical sensor data. These calls are processed by the Super-Sensor block. Then Sensing Response Reader (shown in [Figure 4](#f4-sensors-14-11070){ref-type="fig"}) generates an RxSensResp event when sensor data are available. Please note that we use an asynchronous (split-phase non-blocking) request/response mechanism for efficiency, but a synchronous (blocking) communication with the sensors can easily be supported by extending our API. The RxSensResp event generated by the *Sensing Response Reader* triggers the sensorReading function (shown in [Figure 12](#f12-sensors-14-11070){ref-type="fig"}) which stores the acquired sensor data into the specified segment of the circular buffer. When adequate data sets are available in the circular buffer, the state machine computes the specified feature (max, mean, *etc*.) of each active task. It then constructs a payload with the feature result and sends it to the Super-Medium by using the sendPacket Stateflow function. After receiving the payload, the Super-Medium broadcasts it according to the network setup. [Figure 13](#f13-sensors-14-11070){ref-type="fig"} depicts more in detail the relationship between the simulation framework, including the simulated node instances, the StateFlow model for the application, and finally the functions that the StateChart calls to perform computations. Each level of the modeling hierarchy can be navigated up and down using the features of the StateFlow model editor. The Stateflow-based or Simulink-based WSN application modeling approach could be used to model applications of any sort. In \[[@b38-sensors-14-11070]\], for example, we modeled a data aggregation algorithm for a cluster-based sensor network using Stateflow. Similarly, WSN applications such as distributed algorithms for dynamic task assignment \[[@b39-sensors-14-11070]\], for resource management using collective intelligence \[[@b40-sensors-14-11070]\] and others could be easily modeled using our proposed approach. In this example, we used Stateflow since it is appropriate for a reactive application like SPINE. Simulink blocks could be used to model more data-intensive applications as well. 5.. Hardware in the Loop Simulation =================================== The HIL interface is used by both the Super-Sensor and Super-Medium blocks to access physical sensors and radios connected to the simulation host e.g., via serial ports. HIL support includes mainly two entities: the stub code executed on the physical node containing the sensors or radio, and the Simulink block (inside the Super-Sensor and Super-Medium blocks) used for accessing this stub via a serial (e.g., RS-232 or usb) cable. The stub contains basic platform-dependent code to join or interact with a ZigBee or TinyOS network, without any application-specific part. When the stub receives a packet from the network, it stores the packet locally and at the next request from the Super-Medium block, it transfers the packet payload over the serial cable. In the same way, when the stub receives a packet payload from the Super-Medium block, it constructs the actual packet and transmits it to the network. The Super-Sensor block interacts in a similar way with the stub for reading sensors. In the following, we will use as an example a three-axis accelerometer included in a node from STMicroelectronics \[[@b5-sensors-14-11070]\]. We will use it for HIL simulation with two supported platforms that are very different from each other: TinyOS and the ZigBee-compliant Ember stack. Note that the same methodology can be used to extend the framework to any number and kind of WSN platforms. As mentioned above, each stub contains device drivers and other basic software to support the underlying platform. In addition, it includes a simple serial protocol implementation that is used to communicate with the Super-Sensor and Super-Medium. This protocol is used for reading data from sensors and also for sending or receiving packets to/from the network respectively. To maintain smooth communication, the serial protocol is implemented using several transactions, all initiated by the master (Super-Sensor and Super-Medium). This is similar to the USB protocol, but can be implemented on top of any kind of connection (e.g., USB, RS-232, *etc*.). The protocol uses the following commands to interact with the blocks: SENDPKT: When the stub receives this command at the starting of a new transaction, it knows that it is going to receive a packet payload from the Super-Medium. The next byte should be the length of the payload followed by consecutive bytes of the payload. After receiving a complete packet, it constructs the physical packet and broadcasts it to the network.GETPKTCNT: After receiving this command from the Super-Medium, the stub sends a byte which contains the number of currently stored received packets.GETPKT: After receiving this command, the stub transfers the received packet to Super-Medium. Afterwards, it removes the packet payload from the queue. When the stub receives a packet from the network, the stub stores it in the local queue and then transfers it to the Super-Medium by using transactions GETPKTCNT and GETPKT.GETACCELXAXIS: After receiving this command from the Super-Sensor, the stub calls either LIS3LgetX.get (on TinyOS) or getAccXAxisValue (on Ember ZigBee) to read the X axis value of the three axis accelerometer. In TinyOS, this is an async command whose result is returned by calling the LIS3LgetX.getDone event handler. Inside this event handler, the stub transfers two consecutive bytes of the result to the Super-Sensor block. In Ember Zigbee, the getAccXAxisValue function is executed in a blocking fashion and directly returns two bytes of the result, which are then sent to the Super-Sensor block. Note how this interaction, modeled as an asynchronous call (AcquireSensorData) followed by an event (RxSensResp)on the Stateflow side, is implemented (1) an asynchronous split-phase call (LIS3LgetX.get/LIS3LgetX.getDone) on a TinyOS node and (2) as a synchronous call which generates the callback event on a ZigBee node. The application programmer can thus ignore the tasking model and other architecture details of the target platform. Our HIL and code generation framework takes care of generating the most efficient implementation.GETACCELYAXIS and GETACCELZAXIS perform the same for the Y and Z axis. 6.. Multi-Platform Code Generation ================================== As mentioned in the previous sections, the WSN algorithm is modeled inside the Node Application block (shown in [Figure 4](#f4-sensors-14-11070){ref-type="fig"}). The whole flow to generate platform specific code for this block is depicted in [Figure 14](#f14-sensors-14-11070){ref-type="fig"}. For each target platform (TinyOS, Mantis, the Ember ZigBee stack implementation) we must create a custom Stateflow target for ANSI C code generation which will automatically generate application code for that platform \[[@b41-sensors-14-11070]\]. The computational bodies of the functions called by the task handler in the Ember ZigBee stack, as well as those of the TinyOS tasks and event handlers, are essentially written in C. So the code generated from Stateflow coder can be directly ported to Ember and TinyOS with almost no modification. Similar to the stubs for the HIL interface, we developed platform specific code for the TinyOS and Ember platforms that will interact with the ANSI C code generated from the Stateflow model of the Node Application block. This code also contains platform specific implementation of all framework-defined calls (sendPacket, getPacketPayload, getAccXAxisValue, *etc*.) and event processing functions for CLK, PKT, RxSensRep, *etc*. In order to pass an event to the Application block, we just call the corresponding event processing function from the framework-provided base code, which in turn calls the generated code of the Node Application block with the corresponding event. In other words, our framework provides a platform for modeling WSN applications in Stateflow that is independent of the underlying OS and programming language. The functions and events provided by that platform, e.g., the sendPacket and getPacketPayload functions and the CLK and PKT event are translated into the underlying OS platform by a very lightweight and efficient application-independent framework layer that we provide, and that is called "base code" in the following. A skeleton of the TinyOS specific base code is shown in Example 1. To send the CLK event to the application, we use a periodic timer (Timer0 in the example). Since this file will be automatically generated by TLC scripts \[[@b42-sensors-14-11070]\], we can decide the duration of the CLK pulse at code generation time. On each occurrence of the periodic timer, the clkEvent (event processing function) will be called, which will pass this event to the Node Application platform-independent code. In TinyOS, the driver implementation for the STMicroelectronics accelerometer is done asynchronously, as mentioned above. ----------------------------------------------------------------------------------- **Example 1: Skeleton of platform base code in TinyOS** ----------------------------------------------------------------------------------- module wsn_applicationM { ...} implementation { .... event void Boot.booted(){ .... call Timer0.startPeriodic(10); //CLK pulse in every 10ms } event void Timer0.fired() { clkEvent(); //Up Call: Sending CLK event to the wsn_application } event void LIS3LgetX.getDone(uint16_t xAxisValue, error_t success){ //Up Call: Returning X-axis value of the ACCELEROMETER to the wsn_application\ sensorReading(accelerometer,xaixs,xAxisValue); return; // Down Call: To read the X-axis value of accelerometer\ void getAccXAxisValue() \@C() \@spontaneous() { call LIS3LgetX.get (); } // Down Call: Construct packet with payload and broadcast it.\ void sendPacket(char\* payload) \@C() \@spontaneous(){ ... } event message_t\* Receive.receive(message_t\* msg, void\* payload, uint8_t len) { // Store packet payload locally pktEvent(); // Up Call: Send PKT event to wsn_application .... } // Down Call: To get the payload of the packet\ void getPktData(char\* payload) \@C() \@spontaneous() {...} } ----------------------------------------------------------------------------------- The Node Application code calls getAccXAxisValue (down-call) to read the x-axis value of the accelerometer. Inside this function we call LIS3LgetX.get, which returns immediately due to the split-phase semantics of TinyOS. Afterwards, when the value is read from the accelerometer, the driver calls LIS3LgetX.getDone with the sensor value. Inside this function, we call sensorReading (up-call with values) which provides the sensor reading to the Node Application. On the other hand, synchronous reading (for the Ember Zigbee stack) is fairly easy because we can call sensorReading inside the getAccXAxisValue call. We followed the same techniques to glue the base code with platform independent code in all the platforms that we support. Platform Specific Code Generation by TLC Scripts ------------------------------------------------ The approach that we explained in this section is automated by appropriate TLC scripts. The application code for the TinyOS and Ember ZigBee platforms looks very different, so the TLC script performs the following tasks to generate platform specific code: Copy into the target files the platform-specific application-independent base code, which includes (1) a type conversion header file that converts standard C types to platform specific types and (2) platform specific implementations of the library functions provided by our framework.Generate platform specific application files by taking different sections (such as includes, defines,functions, *etc.*) from the C code generated from Stateflow, and inserting them into the appropriate location of the platform-specific files.Generate make or configuration files for the platform. 7.. Experimental Results and Code Size Comparison ================================================= In order to illustrate how to design, simulate and test the WSN application based on this framework, we use first a small example, and then provide some more results for the more realistic SPINE application. The small example implements a token ring network in which each wireless sensor node: awaits for the token (transmitted via the radio channel),collects several acceleration sensing values, andtransmits the token to the following node. Tokens are passed according to the node ID. When the node receives the token, it collects three acceleration values on X-axis, four acceleration values on Y-axis and five acceleration values on Z-axis. Then the current active node sends out the token and enters into sleep mode until it is woken up by receiving the token again. After designing the WSN application, the user should set the configuration of the nodes and of the network. A Matlab file is used to perform this customization: the user can easily modify it by hand, but we are also working towards the implementation of a graphical user interface to further simplify this operation. The execution of this script automatically generates the SuperMedium and SuperSensor blocks, and sets up the environment for the simulation. In case of HIL nodes, this operation configures also the hardware interfaces to allow the communication among simulated and physical nodes. The application was tested with a different number of nodes, to give an idea of the scalability of our approach. [Table 1](#t1-sensors-14-11070){ref-type="table"} shows the simulation time for different network sizes: the first column contains the number of nodes in the network. For each size the overall execution time is reported, together with the average execution time for the different node types included. After the model has been developed and simulated, we generated the application code automatically for both the TinyOS and the Ember ZigBee platform. Firstly, by using Stateflow Coder, we generated ANSI C code for the application. Then by executing TLC scripts we added platform dependent code. [Table 2](#t2-sensors-14-11070){ref-type="table"} reports the size of the automatically generated code for both platforms. In this table we also noted the size of the platform base code (without any application code), to give an idea about the relative importance of application-specific and application-independent code in a very small example application. We also generated code for SPINE, which is a more realistic example of WSN application, for both TinyOS and the Ember ZigBee stack, as shown in [Table 3](#t3-sensors-14-11070){ref-type="table"}. To estimate the code size increase due to automation, we also implemented manually the same SPINE functionality both in TinyOS and in Ember ZigBee. In [Table 3](#t3-sensors-14-11070){ref-type="table"}, the increase in code size is evaluated by using [Equation (1)](#FD1){ref-type="disp-formula"}. In the equation, (AG-EA) represents the size of automatically generated application code and (MW-EA) represents the size of manually written application code. Code size increments due to the automation process vary from 4% to 13% for both platforms. Increment = ( ( A G − E A ) − ( M W − E A ) ) ∗ 100 M W − E A A G = Automatically Generated M W = Manually Written E A = Empty Application The advantage of using our framework, as opposed to manually writing the WSN application code, is that after the framework successfully generated correct target code for both platforms (*i.e.*, after we completed the initial debugging for our framework, using the simple token ring application), it took only two weeks for one person to implement the fully complex application (SPINE) described in Section 4. In other words, with our framework we could truly concentrate on spending the time modeling and simulating at the functional level, and then code generation, compilation and execution for two very different platforms was automated and extremely fast. 8.. Conclusions =============== We have described an extensible framework for platform independent application modeling and hybrid simulation for sensor network algorithms based on MathWorks tools. The reason for choosing the MathWorks tools over, for example, TOSSIM, NS, OMNeT++, is that they are widely known and used, both in academia and in industry, and already provide rich libraries for digital signal processing and control algorithm behavior simulation. They also provide extensible mechanisms for efficient code generation and platform-specific retargeting. Possible extensions of this work include providing more library functions to support a broader variety of sensors and platforms, as well as improving the fidelity of the channel model, e.g., by considering more in detail the effects of congestion, noise, obstacles, distance and so on. This paper was produced in a joint collaboration effort by all listed authors. Mohammad Mozumdar and ZhenYu Song are the originators of the ideas demonstrated in this paper and also performed the modeling and analysis work to validate the proposed method. Luciano Lavagno and Alberto L. Sangiovanni-Vincentelli helped by providing valuable insights about the methodology and also provided comments while preparing this paper. The authors declare no conflict of interest. {#f1-sensors-14-11070} {#f2-sensors-14-11070} {#f3-sensors-14-11070} {#f4-sensors-14-11070} {#f5-sensors-14-11070} {#f6-sensors-14-11070} {#f7-sensors-14-11070} {#f8-sensors-14-11070} {#f9-sensors-14-11070} {#f10-sensors-14-11070} {#f11-sensors-14-11070} {#f12-sensors-14-11070} {#f13-sensors-14-11070} {#f14-sensors-14-11070} ###### Simulation time for the token-ring network. **Number of Nodes** **Node Type** **Simulation Time (s)** --------------------- -------------------- ------------------------- 8 Network 551 SIM 0.36 (avg. each) HIL_SENS 256.87 (avg. each) HIL_RF 0.34 (avg. each) HIL_FULL 16.89 (avg. each) 16 Network 2069 SIM 0.50 (avg. each) HIL_SENS 257.04 (avg. each) HIL_RF 0.46 (avg. each) HIL_FULL 257.00 (avg. each) 32 Network 4175 SIM 0.91 (avg. each) HIL_SENS 258.28 (avg. each) HIL_RF 0.91 (avg. each) HIL_FULL 258.08 (avg. each) 64 Network 8396 SIM 1.96 (avg. each) HIL_SENS 259.15 (avg. each) HIL_RF 1.97 (avg. each) HIL_FULL 259.13 (avg. each) ###### Code size and memory usage for the token-ring application. **Software System** **Memory** **Platform Base Code with Libs no Application Code (Bytes)** **Automatic Code Generation (Bytes)** --------------------- ------------ -------------------------------------------------------------- --------------------------------------- TinyOS ROM RAM 16,366 17,958 840 918 Ember ZigBee ROM RAM 87,326 89,662 2738 2790 ###### Code size comparison between TinyOS and Ember ZigBee for SPINE. **Software System** **Memory** **Platform Base Code with Libs no Application Code (Bytes)** **Manual Impl. (Bytes)** **Automatic(Bytes)** **Increment** --------------------- ------------ -------------------------------------------------------------- -------------------------- ---------------------- --------------- TinyOS ROM 9366 19,850 20,814 9.19% RAM 840 1355 1380 4.8% Ember ZigBee ROM 80,101 90,714 92,187 13.97% RAM 2736 3171 3220 11.26%

Today, it is not possible to ignore the role of vision and eye movements involved in postural control and the tonic activity in Human (eg ([@B1]-[@B8])). Vision and gaze signals influence the vestibuloocular, the vestibulospinal and the reticulospinal systems (see (9)). And we cannot ignore the importance of body somatosensory signals on the mobilization of the eye and the visual perception of our surrounding space (eg ([@B10]-[@B15])). In the daily life, balance, posture and movement control is complex. For instance, postural control in quiet upright stance, the Human reference posture and also the basis for body stability during movements and gait (see ([@B16],[@B17])), requires the central integration of visual, vestibular and somaesthetic inputs. The central nervous system performs appropriate coordinate transformations of these inputs and permanently generates adapted muscular response (e.g.([@B4],[@B18])). Various experimental investigations were run to contribute for better understanding about the mechanisms and complexes interaction between vision, oculomotricity, and postural control, notably studies about vertical phoria. When the retinal images are dissociated, vertical heterophoria (VH) is the latent vertical misalignment of the eyes reduced via binocular vision mechanisms, and vertical orthophoria (VO) when there is no misalignment ([@B19],[@B20]). Physiological VH was reported in normal subjects, inferior to one diopter, on average 0.28 diopter, corresponding to 0.16±0.01° ([@B21]). Nevertheless, studies reported that VH experimentally induced in healthy young adult subjects by the insertion of a 1 degree-vertical prism influences postural stability, notably a decrease of postural stability during the quiet upright stance ([@B22]). Other results clearly showed that healthy subjects with VH vs. VO (VH less than 1 diopter) are less stable, but the cancellation of VH with an appropriate prism improves postural stability ([@B23]). In line with this, in nonspecific chronic back pain subjects with an additional comorbidity such as peripheral arthralgia, muscle pain, dizziness, headache, or eyestrain known (e.g. ([@B24],[@B25])), representing up to 85% of back pain ([@B26]), all with VH, the same behavior was recorded ([@B27]). Interestingly, these subjects needed more energy to stabilize their posture compared in healthy subjects, but the energetic cost decreased when VH was canceled ([@B27]). Beside eye refraction problems ([@B19],[@B28]), it was hypothesized that VH, even when small in size, indicates a perturbation of the somaesthetic cues required in the sensorimotor loops involved in postural control and could perhaps indicate the capacity of the central nervous system to integrate these cues optimally, suggesting prevention possibilities (see ([@B23])). It is known that sensorimotor conflict (at least between vision and proprioceptive cues) can induce pain and modify sensory perception in some healthy subjects (29); we hypothesized that non-specific chronic back pain could result from such prolonged conflict in which VH could be a sign, with new theoretical and clinical implications (see ([@B27])). This speculation, beyond to be in line with the experimental model introduced by McCabe et al. ([@B29]), agrees with the suggestions of Harris ([@B30]): i) discordance between awareness of motor intention, muscle and joint proprioception, and vision could lead to pain perception, and ii) when pain exists -- without medical pathologic support as organic lesion, rheumatism, neuropathy nor injuries, sprain or fracture --, instead of treating the painful body part with medication, the therapy management should " be directed at restoring the integrity of cortical information processing". Clinical studies suggested the use of VH detection - via the Maddox Rod Test, one of the more robust for the small vertical deviation ([@B31],[@B32]) - as a landmark in the management of nonspecific chronic pain, non-contact injuries (e.g. tendinitis, muscle pain). For instance, VH can be linked to conflict from the stomatognathic system, the pelvis, piercings (body art, i.e., with jewelry) or refractive error; cancel the conflict most of the time restored VO immediately, diminished pain ([@B33]-[@B35]), improved mobility of spinal and peripheral joints, and normalized behaviour in the balance tests after initial alternation ([@B33],[@B34],[@B36]), but remain to be precisely evaluated -- e.g. with objective measurements of heterophoria (e.g. with the use of an eye tracker) and movement (e.g. 3D motion capture). The exact causal relationship has to be identified. Nevertheless, as previously discussed on theoretical aspects ([@B35]), these various afferences (i.e. visual, vestibular and somaesthetic) project to the cerebellum, the reticular formation, and the vestibular nucleus (see ([@B37])) which are located at the base of the spinal motor neurons and oculomotor efferents (see ([@B38])). Yet, based on these experimental studies and clinical reports, on the impact of afferences required for motor control in various tasks and muscle efficiency performance, the possibility of degradation of postural control in upright stance and possible pain linked to sensorimotor conflict, perspective in optimization of muscular efficiency and at least in nonspecific chronic pain management should be of interest. DISCLOSURE ========== The authors have no financial or proprietary interest in materials presented herein.

Background {#Sec1} ========== Guava (*Psidium guajava* L.) fruit is a berry with edible pericarp tissue as flesh and has excellent antioxidant properties \[[@CR1]\]. Guava is member of family Myrtaceae (possesses \~ 150 species) and has 2n = 22 chromosomes with a genome size of \~ 450 MB \[[@CR2], [@CR3]\]. Guava popularly known as 'Apple of the Tropics' is a native of tropical America from where it was distributed in all tropical and subtropical areas of the world \[[@CR4], [@CR5]\]. India, Mexico, Pakistan, Taiwan, Thailand, Colombia, Indonesia are major producers of guava and a small-scale plantation is done in Malaysia, Australia and South Africa \[[@CR6]\]. Fruiting branches in guava bear three terminal flower buds and the central floral bud develop faster into fruit compared to other two lateral buds. In Northern India subtropics, there are two flowering seasons *viz*. April--May and August -- September with peak anthesis time of flower bud between 5:00--7:30 AM. Guava flowers are hermaphrodite and carry 160--400 bilobed anthers and an ovary which is inferior, syncarpous with axile placentation and subulate terminal style \[[@CR6]\]. Style being longer than filaments, self-pollination is less common and domestic honeybee (*Apis mellifera*) is the chief pollinator \[[@CR7]\]. There are more than 400 guava cultivars grown around the world with variation in fruit pulp and peel color. Fruit pulp color ranges from white to deep pink and fruit skin turns green to yellow or red upon ripening and this character varies among cultivars and depends upon the season \[[@CR7]\]. Guava is India's fourth most important fruit crop after mango, banana, citrus and is popularly known as poor man's apple because of low cultivation cost and high nutritive value. Guava is a climacteric fruit and contains reducing sugars, indigestible lignin fiber and carotenoids that increase as the fruit ripens \[[@CR8]\] with major cell wall hydrolyzing enzymes like polygalacturonases, cellulases and starch hydrolyzing α-, β-amylases \[[@CR9]\]. Guava possesses large quantities of vitamin C \[[@CR6]\], is a rich source of phenolic compounds \[[@CR10]\] and carries secondary metabolites with medicinal properties \[[@CR11], [@CR12]\]. Guava intake induces resistance against infectious agents such as *Staphyloccocus*, scavenge cancer causing free radicals and helps in the structural protein, collagen synthesis which maintains integrity of blood vessels, skin, organs, and bones \[[@CR13]\]. Colored fruits are preferred by the consumer owing to higher nutraceutical properties. Color in fruits and vegetables are controlled by secondary metabolism pathway genes mainly phenylalanine ammonia-lyase (PAL), chalcone synthase (CHS), dihydro- flavonol 4-reductase (DFR), flavanol synthase/flavanone 3-hydroxylase (F3H), UDP-glucose:flavonoid 3-O-glucosyltransferase (UFGT), anthocyanidin synthase (ANS) and transcription factors (TFs) of myeloblastosis (MYB), basic helix-loop-helix (bHLH), tryptophan- aspartic acid (WD) repeats, NAC (NAM, ATAF1/2 and CUC2) and MADS (MCM1, AGAMOUS, DEFICIENS, and SRF) domain \[[@CR14]--[@CR19]\]. For an instance, expression of genes encoding MYB TFs, 4-coumarate-CoA ligase (4CL), Glutathione S transferase (GST), Flavonoid 3′5' hydroxylase (F3'5'H) and WD repeat are expressed at higher levels in the red-fleshed apples compared with green apples in congruence with the higher levels of flavonoid and anthocyanin accumulation in red-fleshed apples \[[@CR20]\]. MADS18 is implicated in regulation of anthocyanin synthesis in red compared to green pear \[[@CR21]\] and a NAC TF named as BLOOD makes a heterodimer with PpNAC1 up-regulating the MYB TFs leading to anthocyanin accumulation in blood-fleshed peach \[[@CR22]\]. Also, 32 red peel-color-related genes have been identified in Longan together with anthocyanin biosynthesis genes \[[@CR23]\]. However, in red-fleshed orange 'Hong Anliu', lycopene accumulation is the primary cause behind flesh color \[[@CR24]\]. Also, a green tomato inbred line BUC30 have mutations in phytoenesynthetase1 (PSY1), STAY-GREEN (SGR), and SlMYB12 genes leading to no carotenoids and no degradation of chlorophylls in green ripe tomatoes compared to KNR3 red-fruits \[[@CR25]\]. No such studies have so far been conducted in guava. Also, there exists enormous gene sequence variation among species that generating consensus sequence-based markers and validation is labor intensive and non-targeted. Developing new colored genotypes with desirable agronomic traits by hybridization without marker assisted selection for color related genes is a time-consuming process. So, generating expressed genic sequence information at genome wide level is important to expedite gene cloning and tapping in color trait controlling loci from agronomically less preferred colored guava cultivars (owing to low yields and/or lesser shelf life). Tissue specific comparative gene expression within a genotype and comparison to contrasting genotypes by RNA-Seq is an alternate targeted approach in the absence of gold standard genome assembly. To generate a global gene expression landscape in guava we generated RNA-Seq libraries from leaf, flower buds and fruit tissue of green skinned/white pulped table purpose guava cv. Allahabad Safeda (AS). In another cv. Apple Color (AC) fruit peel color changes from green to apple color (reddish) at fruit picking stage and peel becomes leathery within 3--5 days in winter season. Pink pulp cv. Punjab Pink (PP) is commercially grown for red nectar and the color develops during maturation process (immature fruits have white pulp) probably owing to the chromoplast development as found in other similar genotypes \[[@CR26]\]. Comparative RNA-seq of leaf, flower and fruit at various developmental stages of AS, red vs green peel of AC and pink pulp of PP vs white pulp of AS in current study enhances our understanding of color development in guava and identifying important color controlling candidate genes. Most importantly this study provides the first de novo transcriptome of guava setting a stage for guava genomics at genome wide scale. Results {#Sec2} ======= We have developed the first de novo reference transcriptome assembly of guava, performed gene annotations, compared different fruit development stages to understand molecular pathway (s) in fruit ripening and compared 3 different genotypes with variable coloration in pulp and fruit skin/peel to understand the fruit color development pathway in guava (Fig. [1](#Fig1){ref-type="fig"} & Fig. [2](#Fig2){ref-type="fig"}). Allahabad Safeda (AS) is the widely grown table purpose guava cultivar of India and has green foliage. Figure [1](#Fig1){ref-type="fig"}a shows that the floral buds at all the growth stages of AS are green in color, and exhibits white colored petals as flower opens. Immature and mature fruits of AS both have white pulp and green skin. During ripening fruit skin turns yellow within 3 days after harvesting and stays yellow thereafter. Punjab Pink (PP) has darker green foliage (Fig. [1](#Fig1){ref-type="fig"}b) and floral buds compared to AS. Although pulp color in PP is white in immature fruit but turns pink in mature fruit (Fig. [1](#Fig1){ref-type="fig"}b). Apple Color (AC) has green foliage, green floral buds, white flowers and white pulp of immature and mature fruits but the skin of fruit changes its color from green to crimson red (apple color) at maturation within 3--5 days in winter season (Fig. [1](#Fig1){ref-type="fig"}c). We have compared the RNA-Seq (methods) of Allahabad Safeda leaf and shoot tip (LSt), mixed flower buds (MFb) and mixed fruits (MFr) to understand the landscape of molecular changes in fruit development of guava. We have also compared the immature (ImF), mature (0DF), ripe (3DF), and over-ripe (7DF) fruit growth stages to understand maturation and ripening of guava fruit. To identify inducible genes resulting into apple color development in colored genotypes, we compared red vs green skin of AC and mature fruit of AS to PP. Fig. 1Leaves, flower buds and fruits color comparison in Allahabad Safeda, Punjab Pink and Apple Color - CISH G5 genotypes of guavaFig. 2Experimental set up of de novo transcriptome assembly of *Psidium guajava* L. *cv.* Allahabad Safeda and functional annotation of transcripts. LSt -- Leaf and Shoot tip tissue (Immature leaf, Mature leaf and shoot apex), MFb -- Mixed Flower bud tissue (six developmental stages), MFr -- Mixed Fruit tissue (Immature, just harvested, 3 days ripe and 7 days ripe fruit -- with seed and peel), ImF -- Immature Fruit (80 day before harvesting, without seed), 0DF -- Zero Day Fruit (Mature just harvested, without seed), 3DF -- Three Days after harvesting Fruit (ripe fruit, without seed), 7DF -- Seven Day after harvesting Fruit (over-ripe fruit, without seed) RNA-Seq data generation, de novo transcriptome assembly and annotation {#Sec3} ---------------------------------------------------------------------- The pair end libraries from different tissue types of AS, AC and PP were sequenced and 137.3, 20.24 and 20 million raw reads of 100 bp each were generated, respectively (Additional file [1](#MOESM1){ref-type="media"}: Table S1). The low quality sequences were filtered at quality score ≥ 30 and 120 million high quality reads of AS belonging to 13 libraries were used for generating de novo reference transcriptome using Trinity assembler \[[@CR27], [@CR28]\]. A total of 279,792 transcripts belonging to 84,206 components/genes with N~50~ of 3603 bp were obtained (Table [1](#Tab1){ref-type="table"}). Benchmarking Universal Single-Copy Orthologs (BUSCO) analysis \[[@CR29]\] with eudicots identified 93.7% (1987/2121) complete BUSCO genes, 4.5% (95) fragmented orthologs and 1.8% (39) orthologs as missing (Additional file [6](#MOESM6){ref-type="media"}: Figure S1). Blast search against the nr protein database identified homologs for 219,924 transcripts. Protein family search identified 140,061 protein family domains. Gene ontology assessment with Blast2GO assigned gene ontology terms to 116,629 transcripts (Table [1](#Tab1){ref-type="table"}; Fig. [3](#Fig3){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S1), where biological process consists of 87,954 transcripts, cellular components of 82,820 and molecular function of 96,308 transcripts (Fig. [3](#Fig3){ref-type="fig"}, Additional file [7](#MOESM7){ref-type="media"}: Figure S2). Table 1Allahabad Safeda transcriptome assembly statistics**Transcriptome Assembly** Contigs/Transcripts279,792 Components/Genes84,206 % GC content43.08 Contig N503603 Assembly length (MB)647.4**Functional annotation** Transcripts with homologs219,924 Match with predicted protein9958 Match with hypothetical protein7790**Protein Family annotation** Transcripts with Pfam domains140,061**Gene Ontology Annotation** Transcripts with assigned GO terms116,629  Biological Processes87,954  Cellular Component82,820  Molecular Function96,308Fig. 3Distribution of assembled transcripts in the gene ontology classes of biological processes, molecular functions and cellular components. Bars are scaled to log values Differential expression analysis of leaf, flower and fruit of Allahabad Safeda {#Sec4} ------------------------------------------------------------------------------ In AS 2777 transcripts representing 2139 genes were found differentially expressed in mixed fruit (MFr) vs mixed flower buds (MFb), mixed fruit vs leaf & shoot tip (LSt) and mixed flower bud vs leaf & shoot tip (Data S[2](#MOESM13){ref-type="media"}). Clustering analysis shows a high correlation among the replicated samples \> 0.96 for LSt, \> 0.93 for MFb and \> 0.97 for MFr (Fig. [4](#Fig4){ref-type="fig"}a; Additional file [2](#MOESM2){ref-type="media"}: Table S2). Fig. 4Differentially expressed transcripts in Allahabad Safeda tissues **a** Heatmap and hierarchal clustering **b** Venn diagram in tissue types viz. Leaf and shoot tip (LSt), Mixed flower buds (MFb) and Mixed fruit tissue (MFr). R1, R2 and R3 are the three RNA-Seq biological replicates We identified 2125 differentially expressed transcripts (DETs) in MFr compared to LSt, with 971 being up and 1154 down regulated. In MFr and MFb comparison, 1445 DETs were found, of which 719 were up-regulated and 726 down regulated. However, 660 DETs were identified between MFb and LSt, with 447 up and 213 down-regulated (Additional file [13](#MOESM13){ref-type="media"}: Data S2). Only 33 transcripts among the DETs were common among the three tissues types (Fig. [4](#Fig4){ref-type="fig"}b). In order to identify genes involved in fruit development, top 20 up-regulated transcripts were selected from MFr comparison to LSt and/ or MFb and 7 genes were found common with \> 10 Log2FC. Interestingly, putting together transcripts of these two comparisons all were found co-upregulated and none of the genes was found down regulated (Additional file [3](#MOESM3){ref-type="media"}: Table S3). The most up-regulated gene (comp27411_c1) represented by six transcripts Hydroxycinnamoyl CoA shikimate (quinate hydroxycinnamoyltransferase, HCT) belonging to BAHD family of acyl-CoA-dependent acyl-transferases controls lignin \[[@CR30], [@CR31]\] and cutin biosynthesis \[[@CR32]\]. Cinnamyl alcohol dehydrogenase (CAD) important for lignin biosynthesis \[[@CR33], [@CR34]\], expansins involved in cell wall loosening \[[@CR35]\], ABC transporter encoding ATP dependent channels \[[@CR36]\], Palmitoyl transferase involved in fatty acid oxidation \[[@CR37]\], 1-aminocyclopropane-1-carboxylate oxidase (ACO) an ethylene biosynthesis gene \[[@CR38]\], Subtilisin-like protease with a role in plant-pathogen interactions \[[@CR39]\], 9-cis-epoxycarotenoid dioxygenase (NCED) a major Abscisic Acid biosynthesis gene \[[@CR40], [@CR41]\] and Rbcx, a Rubisco assembly chaperon \[[@CR42]\] are the top protein families represented by up-regulated transcripts in guava fruit (Additional file [3](#MOESM3){ref-type="media"}: Table S3). Metabolic pathway analysis of fruit tissue in comparison to leaf and flower {#Sec5} --------------------------------------------------------------------------- The metabolic and regulatory pathway analysis of fruit, the major sink in comparison to the strongest source, the leaf was performed with MAPMAN software \[[@CR43]\] (<http://mapman.gabipd.org>) with all DETs at FDR \< 0.001. Differential 2125 transcripts were found significantly regulated in fruit compared to leaf (Fig. [5](#Fig5){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S2). General metabolism analysis showed that transcripts involved in light reactions, C3 cycle, photosynthesis, tetrapyrole pathway (controlling chlorophyll biosynthesis), starch synthesis, amino acid biosynthesis except phenylalanine (input for secondary metabolism), lipid degradation, raffinose biosynthesis, cell wall associated leucine rich repeat and arabinogalactan-proteins are down-regulated in fruit. Importantly sucrose biosynthesis, gluconeogenesis, conversion of starch to reducing sugars like glucose and fructose, wax biosynthesis, phenylalanine generation, glycolipid synthesis (for generating mono and di galactosyl diacylglycerol for food reserve storage in seeds), cellulose synthesis, trehalose biosynthesis, mitochondrial electron transport chain, cell wall degradation pectate lyases (PeLs) and polygalacturonases (PGs) are up-regulated in fruit. Discreet furcation of these pathways in fruit tissue are in general concordance with its biological role of alluring birds for seed dispersal \[[@CR16]\]. However, pectin esterases involved in plant cell wall modification and subsequent breakdown and long chain fatty acid biosynthesis genes catalyzing the cutin synthesis exhibited a mixed response (Fig. [5](#Fig5){ref-type="fig"}a). Fig. 5MAPMAN pathway distribution of differentially expressed transcripts of fruit (MFr) vs leaf (LSt) **a** metabolism overview **b** regulation overview **c** cellular response overview **d** proteasome and autophagy. Up- and Down- regulated DETs are represented with blue and red squares, respectively with log2 transformed values (scale for **b** and **c** is same) Regulation overview analysis with MAPMAN shows that most of the transcripts mapping to ABA, Ethylene, Cytokinin, Gibberellins (GA) and Salicylic acid (SA) signal transduction pathways were up-regulated whereas Jasmonate (JA), Auxin, and Brassinosteroid (BR) were down-regulated with few transcripts showing up-regulation (Fig. [5](#Fig5){ref-type="fig"}b). Ethylene biosynthesis and signal transduction genes, 1-aminocyclopropane-1-carboxylate synthase (ACS), ACC oxidase 3, Ethylene receptor 2 (ETR2), ethylene response factor ERF-1, basic helix-loop-helix (bHLH) TF and pyridoxine biosynthesis gene PDX1.2 were found up-regulated. ABA biosynthesis and signaling factors including NCED and ABA binding factor 4 (ABF4), B3 domain containing high-level expression of sugar-inducible gene 2 (HSI2), highly ABA-induced 1 (HAI1), hypostatin resistance 1 (HYR1), a UDP glycosyltransferase (UGT) and GRAM domain family protein were highly up-regulated, only ABA-responsive TB2/DP1 (HVA22 family protein) showed down-regulation. Auxin leucine-rich repeats (LRR), F-box TIR receptor, TCP family, ARF and AUX/IAA TFs were found down regulated indicating the auxin signaling down-regulation in guava fruit development. Interestingly, Brassinosteroid insensitive (BRI) encoding receptor kinase is up-regulated indicating overall up-regulation of BR signal transduction and responses. We identified 40 TF families with multiple transcripts belonging to MYB, MADS, HB, WRKY, ARF, bHLH, AP/EREBP, bZIP, NAC, AUX/IAA, B3, Jumonji and, Polycomb. These families showed both up and down regulation, indicating their importance in modulation of fruit development. Sucrose cytosolic invertase 2 (CINV2), responsible for conversion of sucrose to monosaccharides like fructose and glucose showed up-regulation and is in line with increase in sucrose catabolism in developing fruits. Cellular response analysis depicts down regulation of transcripts belonging to biotic stress and is in line with fruits being more prone to pathogen and insect damage in comparison to leaves (Fig. [5](#Fig5){ref-type="fig"}c). Phytoene synthase (PSY) and lycopene beta cyclase (lcy-b) responsible for accumulation of α and β- carotene shows over-expression indicating up-regulation of carotenoid biosynthesis pathway. Ubiquitin and autophagy dependent degradation pathways (Fig. [5](#Fig5){ref-type="fig"}d) showed up-regulation of 44 transcripts, emphasizing increased protein turnover process. Near similar results were obtained in a comparison of fruit vs flower transcripts (Additional file [13](#MOESM13){ref-type="media"}: Data S2). Up-regulation of secondary metabolites during fruit ripening {#Sec6} ------------------------------------------------------------ We compared RNA-Seq at different fruit maturity and ripening stages in AS. Comparison of mature fruit 0DF to immature fruit ImF identified 220 differentially regulated transcripts, with 75 showing up-regulation and 145 showing down regulation (Additional file [13](#MOESM13){ref-type="media"}: Data S3). However, at ripening 3DF vs 0DF, 366 transcripts were differentially regulated with 232 up-regulated and 144 down-regulated (Additional file [13](#MOESM13){ref-type="media"}: Data S4). Interestingly, during over-ripening 7DF vs 3DF only 11 transcripts showed differential regulation with only one down regulated (Additional file [13](#MOESM13){ref-type="media"}: Data S5). The major up-regulated genes in mature vs immature fruit (Additional file [8](#MOESM8){ref-type="media"}: Figure S3; Additional file [13](#MOESM13){ref-type="media"}: Data S3) include Alpha-Expansin, cellulose synthase, phospho-enol-pyruvate carboxylase kinase, β-amylase, PSY, CAD and COMT family of lignin biosynthesis genes and other o-methyl transferases. However, flavonoid pathway genes other than lignin biosynthesis, pectin methylesterases, light reactions, calvin cycle and photorespiration were down-regulated. In ripe vs mature fruit (Additional file [9](#MOESM9){ref-type="media"}: Figure S4 A; Additional file [13](#MOESM13){ref-type="media"}: Data S4) there is upregulation of transcripts for cellulose synthase, expansins, increased fatty acid synthesis and elongation, PSY, phenylalanine biosynthesis genes arogenate dehydratase, flavonoid biosynthesis related transcripts like UGT - Hypostatin Resistance 1 (HYR1), Flavonoid 3′,5′-hydroxylase 2 (F3′5′H), Phenylalanine ammonia-lyase 3 (PAL 3) and 4-coumarate\--CoA ligase 2 (4CL2). All the transcripts belonging to ABA, BR, Ethylene, Cytokinin, and SA were up-regulated, while AUX and TCP transcripts related to auxin and GA Insensitive (GAI) were down-regulated (Data S[4](#MOESM13){ref-type="media"}, Figure S[4](#MOESM9){ref-type="media"} B). Transcripts corresponding to TF families of WRKY, AP2, bHLH, PHOR1 (ubiquitin ligase activity), MYB and C2C2.CO like were found up-regulated. Also, Ubiquitin and autophagy dependent protein turnover pathway were up-regulated (Data S[4](#MOESM13){ref-type="media"}, Figure S[4](#MOESM9){ref-type="media"} C) as well. Apple color in fruit skin is derived from up-regulation of secondary metabolism {#Sec7} ------------------------------------------------------------------------------- Apple color skin of AC guava develops within a short time period of \~ 3--5 days in winter season during fruit maturation. Comparison of FPKM value of transcripts belonging to red vs green skin, identified only 52 DETs indicating very specific pathways involved in fruit color development (Fig. [6](#Fig6){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S6). Interestingly, all the transcripts of phenylpropanoid and lignin pathway showed up-regulation indicating the color development in the skin of guava is result of over expression of phenypropanoid and lignin biosynthesis pathway (Additional file [10](#MOESM10){ref-type="media"}: Figure S5; Additional file [13](#MOESM13){ref-type="media"}: Data S6). Comparison of PP mature fruit with AS mature fruit identified 19 DETs with 9 up-regulated and 10 down-regulated transcripts (Additional file [13](#MOESM13){ref-type="media"}: Data S7). However, only omega-hydroxypalmitate O-feruloyl transferase-like indicated a footprint of secondary metabolism pathway. Fig. 6Heatmap of log transformed FPKM values of differentially expressed transcripts in red peel compared to green peel in Apple Color cultivar. Blue arrow represents up-regulated and red represents downregulated transcripts in red peel Fruit color development in colored guava genotypes, is concomitant with ripening process {#Sec8} ---------------------------------------------------------------------------------------- We hypothesized that pink pulp / apple color skin development may share the fruit ripening related genes as the color development in colored cultivars is concomitant with ripening process. The cluster analysis of DETs in 5 comparisons viz. AS ImF vs 0DF, 3DF vs 0DF, 7DF vs 3DF, PP ImF vs 0DF, and AC_RP vs AC_GP was carried out. AS mature fruit, PP mature fruit, AC green peel and AC red peel clustered in single group of mature fruit stages. AS 3DF and AS 7DF made another group with mostly similar expressions except 11 DETs. Third cluster consisted of AS ImF and PP ImF, again a result of similar fruit stage (Additional file [11](#MOESM11){ref-type="media"}: Figure S6; Additional file [13](#MOESM13){ref-type="media"}: Data S8). qRT-PCR for two well-known candidates for coloration in fruits and vegetables PAL and PSY 2 was carried out to see the reproducibility of differential FPKM expression values. Expression of PAL in AS fruit was maximum at yellow peel color ripe stage (\~ 9.5X in 3DF_AS compared to ImF_AS; Fig. [7](#Fig7){ref-type="fig"}a). However, expression in mature Punjab Pink fruit (0DF\_ PP with pink pulp) was \~ 1.8 fold higher as compared to mature Allahabad Safeda fruit (0DF_AS). Interestingly, expression in red peel of Apple Color (AC_RP) was found the highest (\~ 1.5 X compared to 3DF_AS). These results indicated that the expression of PAL increases with ripening, but is also genotype dependent and might have contribution towards red coloration in peel of Apple Color genotype. However, Phytoene Synthase 2 (Fig. [7](#Fig7){ref-type="fig"}b) shows a general trend of increase in expression with maturity and ripening in all the three genotypes. These results also indicated that observations recorded in our comparative transcriptomic data are in line with qRT-PCR analysis. Fig. 7qRT-PCR assay of universal fruit color determining factors in cereals, fruits and vegetables **a** Phenylalanine Ammonia-Lyase, **b** Phytoene Synthase 2 and candidate genes for color development in apple color skin of guava cv. Apple Color and pink pulp in Punjab Pink **c** & **d** (R,S)-reticuline 7-O-methyltransferase-like, **e** Glycerol-3-phosphate acyltransferase 5, **f** Peamaclien **g** CTP synthase-like, **h** Monodehydroascorbate chloroplastic, **i** Probable 2- oxoglutarate-dependent dioxygenase AOP1, **j** Methionine synthase, **k** Secoisolariciresinol dehydrogenase, **l** BEL1-like homeodomain 1, **m** Aminocyclopropane-1-carboxylate oxidase 1-like, **n** Uncharacterized protein LOC104449412 (transcript id are given in brackets). Histone 3 is used as an internal reference. The fold change data is normalized to Allahabad Safeda immature fruit set to unity. Tissue types compared are: ImF_AS -- Immature fruit of Allahabad Safeda, 0DF_AS -- Mature fruit of Allahabad Safeda at harvesting stage, 3DF_AS -- Fruit 3 days after harvesting, 7DF_AS -- Fruit 7 days after harvesting, ImF_PP -- Immature fruit of Punjab Pink, 0DF_PP -- Mature fruit of Punjab Pink at harvesting stage, AC_GP -- Green peel of Apple Color fruit 5 days before harvesting stage of fruit, AC_RP -- Red peel of Apple Color fruit at harvesting stage. Error bars represents ±Standard error with *n* = 3 Candidate genes for color development in guava {#Sec9} ---------------------------------------------- To identify the genes involved in color development, GO enrichment of DETs was performed (Additional file [12](#MOESM12){ref-type="media"}: Figure S7). For the potential candidates FPKM values were compared in red peel of Apple Color AC_RP vs green peel AC_GP and mature fruit of Punjab Pink 0DF PP vs mature fruit of Allahabad Safeda 0DF AS for all the tissue types. The FPKM values showed higher expression of genes at color turning stages (Additional file [4](#MOESM4){ref-type="media"}: Table S4). Interestingly, the FPKM comparative analysis indicated reticulin o-methyltransferase (responsible for converting alkaloid reticulin to laudanine) as top candidate gene. The moderate expression of transcripts was present in green and maturing fruits of AS. However, huge increase in expression of reticulin o-methyltransferase in red peel of AC and mature PP fruit suggested increased levels of such alkaloids in colored fruit tissues. Maximum FPKM expression value of glycerol-3-phosphate acyltransferase 5 (GPAT 5), peamaclein, CTP synthase-like, chloroplastic monodehydroascorbate (MDA), probable 2-oxoglutarate-dependent dioxygenase AOP1 (2OG-AOP1) and methionine synthase (MS) corresponding transcripts in red peel of guava suggested interplay of several candidate genes for red coloration in peel of guava. Higher expression of transcripts for Secoisolariciresinol dehydrogenase (SDH), BEL1-like homeodomain 1(BLH1) in PP indicates additional candidates for red color in pulp of guava. Three transcripts (R, S)-reticuline 7-O-methyltransferase-like (comp25759_c1_Seq1, comp25759_c1_Seq5 & comp25759_c1_Seq11) showed high homology, so common primers were designed for qRT-PCR. Results show that expression of (R, S)-reticuline 7-O-methyltransferase Like (RML) transcripts increases to 100X at 0DF_AS compared to ImF_AS set to unity (Fig. [7](#Fig7){ref-type="fig"}c). However, expression did not change in 3DF_AS compared to 0DF_AS and reduced to 20X in 7DF_AS. Interestingly, expression in ImF_PP is \~3X compared to ImF_AS and increases to \~ 300 X in 0DF_PP. Expression in AC_GP is \~21X but show an increase to \~300X in AC_RP, almost to similar levels as in 0DF_PP suggesting RML, a candidate for coloration in both guava peel and flesh. Surprisingly, expression of RML\_ comp25759_c1_seq4 (Fig. [7](#Fig7){ref-type="fig"}d) stayed at low level in AS and PP at all the stages, however, its expression is \~30X in AC_GP and reaches to 80X in AC_RP. In AC this result indicated genotype specificity of RML members. High expression of RML family members emphasized their role in color development in guava. High expression of G3PAT, SDH, AOP1, MS and SDH in AC_RP indicated the additional candidates for color development in guava (Fig. [7](#Fig7){ref-type="fig"}). In general, qRT-PCR for all these candidates supported our transcriptome based FPKM results. Discussion {#Sec10} ========== Fruit ripening and maturation involves ABA, ethylene and secondary metabolite up-regulation {#Sec11} ------------------------------------------------------------------------------------------- Climacteric fruits show increased rate of respiration and ethylene biosynthesis which in turn triggers the activity of enzymes like PGs, PeLs and, [p]{.ul}ectate [me]{.ul}thylesterases (Pme). Process of ripening is accelerated by the conversion of complex polysaccharides into simple sugars leading to increased sugar to acid ratio concomitant with textural and color changes. Conversion of l-aminocyclopropane-l-carboxylic (ACC) acid from S-adenosylmethionine (SAM) is catalyzed by ACC synthase and is the rate limiting step in ethylene biosynthesis \[[@CR44], [@CR45]\]. Expression value comparison in AS, PP and AC showed differential increase in transcripts corresponding to ACO 1, ACO 2, ACS and ETR2 during fruit maturation (Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9). ABA biosynthesis and signaling is also found up-regulated during fruit maturation and ripening \[[@CR46]\]. We observe increased expression of ABA biosynthesis gene NCED, receptor Pyrabactin resistance 1 like 9 (PYL9) and TF ABA insensitive 5 (ABI5) during fruit ripening in all 3 cultivars (Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9). Ethylene and ABA are stress hormones and up-regulates defense system of plants against pathogens by stimulating phenylpropanoid pathway, pathogenesis-related proteins and inducing systemic resistance \[[@CR47]\]. Expression of pathogenesis related PR-4 is high in fruit tissues with the highest expression in red peel and PR-10 in green peel (Additional file [13](#MOESM13){ref-type="media"}: Data S9). Fig. 8DETs mapped to fruit maturation and ripening in Allahabad Safeda, Punjab Pink and Apple Color. Ethylene biosynthesis and signaling (ACO, 1-aminocyclopropane-1-carboxylate oxidase; ERF2, ethylene-responsive transcription factor 2; ACS, 1-aminocyclopropane-1-carboxylate synthase), Abscisic acid biosynthesis and signaling (NCED3, 9-cis-epoxycarotenoid dioxygenase 5; PYL9, abscisic acid receptor PYL9; ABI5, Abscisic acid-insensitive 5 7; CYP707A3, abscisic acid 8-hydroxylase 3), secondary metabolism (CAD1, probable cinnamyl alcohol dehydrogenase 1; CHS, chalcone synthase; Cinnamate beta-D-glucosyltransferase-like; F3'5'H, flavonoid 3,5 -hydroxylase; F3H, flavanone 3-hydroxylase; FLS1, flavonol synthase flavanone 3-hydroxylase; IFR, isoflavone reductase; Isoflavone 2 -hydroxylase-like; PAL, phenylalanine ammonia-lyase; PSY1, phytoene synthase chloroplastic-like; UGE, UDP-glucose 4-epimerase; UGT/UFGT, anthocyanidin 3-O-glucosyltransferase), carbohydrate metabolism (β-amylase) and fruit softening (PG, polygalacturonase; Pectinesterase Inhibitor,; PL, pectate lyase 4). The color scale on the right represents the log-transformed FPKM values Guava is rich in secondary metabolites. Expression of genes controlling secondary metabolites like isoflavone reductase (IFR) is maximum in immature fruits, flavanone 3-hydroxylase (F3H) is high in leaf, flower and immature fruits and flavonoid 3, 5 -hydroxylase (F3'5'H) 2-like in leaf and flower tissue (Additional file [13](#MOESM13){ref-type="media"}: Data S9). Interestingly, expression of PAL, isoflavone 2 -hydroxylase-like and flavonol synthase are high during fruit maturation with maximum expression in ripe guava (Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9). Hydrolysis of starch by β-amylase generates maltose leading to sweet flavor in ripened fruits \[[@CR8]\]. Expression of β-amylase increased at 0DF and 3DF while reduced at over-ripe stage 7DF. Also, the expression is much higher in AC peel and pink PP fruit (Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9). High expression of β-amylase underscores the sweet-smelling nature of guava at fruit ripening in general and specifically higher in AC and PP cultivars compared to AS. Synthesis of lignin monomers involve the phenylpropanoid pathway initiated by PAL and followed by Cinnamate 4-hydroxylase (C4H), 4CL, HCT, Caffeoyl CoA 3-O-methyltransferase (CCoAOMT), Cinnamoyl CoA reductase (CCR), Caffeic acid 3-O-methyltransferase (COMT) and CAD genes. Several CAD family members generally show up-regulation in the fruit flesh and respond to ABA, ethylene and various biotic and abiotic stresses as observed in melon \[[@CR33]\]. However, expression of COMT gene is directly associated with increase in lignin content and is maximum in immature AS fruit. HCT and CAD genes control the crucial step in suberin \[[@CR30], [@CR31]\] and cutin biosynthesis \[[@CR32]\]. HCT and CAD genes are the most up-regulated transcripts (Additional file [3](#MOESM3){ref-type="media"}: Table S3) in MFr vs LSt. Their expression increases during maturation and was the highest at ripe fruit stage followed by reduction at over-ripe stage indicating lignin, suberin and cutin synthesis during maturation and ripening. PGs expression increases during ripening. We also identified the highest expression of transcript corresponding to PGs in overripe fruit of AS, and high expression in red peel of AC compared to green peel and mature compared to immature fruit in PP (Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9). Results of comparative transcriptome analysis in this manuscript are in concordance with protein and metabolite analysis reported in different fruit species supporting that this first transcriptome of guava will play a promising role in setting the stage for functional genomics in guava. Our findings of gene up-regulation during ripening of guava from immature fruits are in concordance with existing literature and are summarized in Fig. [9](#Fig9){ref-type="fig"}. Maturation of guava involves the wave of ABA biosynthesis and signaling followed by ethylene biosynthesis and accompanies the secondary metabolites accumulation, upregulation of carbohydrate metabolism and cell wall degradation enzymes. Fig. 9Concordant model for ripening of guava in Allahabad Safeda and color development in colored cultivars Punjab Pink and Apple Color based on RNA-Seq (FPKM) and/or qRT-PCR results. Abscisic Acid (ABA) biosynthesis and signaling (NCED3, 9-cis-epoxycarotenoid dioxygenase 3; PYL9, Pyrabactin like 9; ABI5, Abscisic acid-insensitive 5)**,** Ethylene biosynthesis (ACS, 1-aminocyclopropane-1-carboxylate synthase; ACO, 1-aminocyclopropane-1-carboxylate oxidase), secondary metabolism (SM) (CAD1, cinnamyl alcohol dehydrogenase 1; PAL, phenylalanine ammonia-lyase; PSY1, phytoene synthase 1; FLS1, flavonol synthase 1; GT2, cinnamate beta-D-glucosyltransferase, CYP81E1: Isoflavone 2 -hydroxylase), carbohydrate metabolism (CM) (β-amylase) and cell wall degradation (CWD) (PG, polygalacturonase) pathways interact for fruit ripening in guava. ABA, Ethylene, SM, CM and CWD further interact with (R,S)-reticuline 7-O-methyltransferase (RML), glycerol-3-phosphate acyltransferase 5 (GPAT5), Peamaclein, CTP synthase, Monodehydroascorbate chloroplastic (MDAC), 2- oxoglutarate-dependent dioxygenase (2OG) AOP1, Secoisolariciresinol dehydrogenase (SDH) and BEL1-like homeodomain 1 (BEH1) for color developement in colored genotypes (R,S)-reticuline 7-O-methyltransferases are the potential candidates for color development in guava {#Sec12} --------------------------------------------------------------------------------------------------- Transcript FPKM value and qRT-PCR (Fig. [7](#Fig7){ref-type="fig"}; Additional file [4](#MOESM4){ref-type="media"}: Table S4) of RML, GPAT 5, Peamaclein, CTP synthase-like, chloroplastic MDA, 2OG-AOP1, MS and SDH genes show that expression level of these are maximum in AC fruit skin begging the question if increased expression of these genes is responsible for colored guava skin. RMLs are involved in tetra hydrobenzyl isoquinoline alkaloid production in poppy at the step of reticuline to laudenine conversion \[[@CR48]\]. Up-regulation of 4 isomeric transcripts of RML in peel of guava suggests that red guava fruits may serve as a good source for extracting alkaloids like reticulin and laudenin rather than a much-regulated source like poppy. Although the reticulin and laudenin are extracted from poppy for medicinal purposes as an anti- diarrheal, anti-dysenteric, anticough, the use of guava fruit for similar utility have been described before \[[@CR10], [@CR13]\]. BLAST search of four RML transcripts identified its homologs in eucalyptus, grape and cork oak tree with 74--86% identity. Expression of all the four RML transcripts is the second highest in PP 0DF- mature fruit (pink pulp) after AC_RP - red peel (Additional file [4](#MOESM4){ref-type="media"}: Table S4). GPAT 5 is required for synthesis of suberin in Arabidopsis root and seed coat \[[@CR49]\]. Increased expression of HCT and GPAT 5 indicates escalated suberisation occurring in fruit coat in a short duration accompanying the ripening process. Peamaclein originally identified in peach fruit, is a recently identified gibberellin induced fruit allergen conserved among various fruits \[[@CR50]\]. Higher expression of peamaclein transcripts in red peel (Additional file [4](#MOESM4){ref-type="media"}: Table S4) might be an indicator of high fruit allergen prevalence in red peel. CTP synthase-like is an enzyme associated with cytosine synthesis. Increased CTP cellular level in yeast has been associated with increased phospholipid synthesis via the Kennedy pathway \[[@CR51]\]. This suggests increased levels of phospholipids in apple color fruit skin. MDA can be reduced to ascorbate through reactions by MDA reductase (MDAR) in chloroplast and help relieving the oxidative stress by reactive oxygen species (ROS) evolving from the photosynthesis reactions \[[@CR52]\]. Overexpression of MDAR in tomato has been associated with stress enhanced tolerance to temperature and methyl viologen-mediated oxidative stresses \[[@CR52]\]. High expression of chloroplastic MDA transcript indicates the release of ROS in the fruit peel during color change. 2OG-AOP1 is involved in glucosinolate synthesis in Arabidopsis. Interestingly, 2OG is also the second largest enzyme family in plants involved in hormone and flavonoid biosynthesis \[[@CR53]\]. Increased expression of 2OG-AOP1 is convincingly in the line of increased expression of flavonoid biosynthesis related genes during color turning in AC. Methionine not only acts as a building block for protein synthesis but also serves as immediate precursor of S-adenosylmethionine, a major methyl-group donor in transmethylation reactions and as an intermediate for biosynthesis of polyamines and ethylene \[[@CR54]\]. Up-regulation of methionine synthase, an enzyme catalyzing the last step of methionine biosynthesis supports the increased demand of ethylene biosynthesis during fruit ripening in general and highest expression in red skin (Fig. [7](#Fig7){ref-type="fig"}j; Additional file [4](#MOESM4){ref-type="media"}: Table S4). SDH converts (2)-secoisolariciresinol into (2)-matairesinol in *Forsythia intermedia*, a precursor for the biosynthesis of antiviral and anticancer agent, podophyllotoxin \[[@CR55], [@CR56]\]. Although, SDH is identified in a comparison of AS 0DF to PP 0DF but the expression was the highest in AC red skin with \> 3-fold expression increase compared to green skin (Fig. [7](#Fig7){ref-type="fig"}k; Additional file [4](#MOESM4){ref-type="media"}: Table S4). Presence of RML and SDH high expression in red guavas indicated colored guava being a rich source of rare secondary metabolites. BLH1 have been implicated in leaf and ovule development in Arabidopsis \[[@CR57]\] and overexpression of apple BEL1-Like MDH1 in Arabidopsis \[[@CR58]\] leads to reduced fertility and other pleiotropic effects, however no implication in color development are reported. RNAi of *Sl*BEL1- like 11 in tomato increased the chlorophyll content \[[@CR59]\] and delayed fruit ripening. Increased expression of BEL1 in PP pulp might have role in chromoplast development. Comp27248_c1_seq24 is another highly expressed transcript in PP 0DF that does not have any other functional evidence but shows 90% identity with *Eucalyptus grandis* uncharacterized LOC104449412. Up-regulation of DETs in red skin of guava *cv.* Apple color implicates cross talk among myriad dynamic processes {#Sec13} ----------------------------------------------------------------------------------------------------------------- Up-regulation of Mannan endo-1,4-beta-mannosidase (β-mannanase) indicates utilization of mannans in producing mannose oligosaccharides from lignocellulosic polysaccharides \[[@CR60]\]. Kinesis light chains (KLCs) are involved in microtubule mediated organelle transport in eukaryotes \[[@CR61]\]. Up-regulation of KLC corresponding transcripts indicate high metabolic activity during skin color alteration in guava. EG45-like domain proteins belong to plant natriuretic peptides (PNPs), a class of systemically mobile molecules distantly related to expansins with implication in regulating water and ions homeostasis and participation in plant immune response \[[@CR62]\], so are in line with increased demand in leathery red skin. (−)-germacrene D synthase is involved in biosynthesis of a sesquiterpene germacrene D generally obtained from cubebe oil \[[@CR63]\]. Golgi α-mannosidase II is a key glycosyl hydrolase in the N-linked glycosylation pathway \[[@CR64]\], indicating N-glycosylation role in reddening of skin. E3-ubiquitin ligases catalyze final step of ubiquitin conjugation demanding tight regulation to ensure accurate substrate ubiquitylation and finally degradation via 26 S proteasomal pathway marking an important checkpoint for protein turnover \[[@CR65]\]. Up-regulation of 2 transcripts belonging to E3 ubiquitin- ligase SHPRH in apple color skin development indicated high turnover of specific proteins. Omega-6 fatty acid destaurases or Fatty Acid Destuarse 2 (FAD2) desaturates oleic acid to linoleic acid (18:2) for enhancing stress tolerance in endopasmic reticulum in response to abiotic stresses \[[@CR66]\]. Omega-6 fatty acid endoplasmic reticulum isozyme 2-like up-regulation in red and leathery skin might be an indicator of better protection to biotic and abiotic stresses. Arabinogalactan peptides (AGPs) in wheat are encoded by grain softness protein genes \[[@CR67]\]. Wheat-AGPs derived down-regulation of GSPs in RNAi lines increases the grain hardness and decreases viscosity of aqueous extracts. Such extension like activities in the reddening skin might be attributes of AGPs as Arabinogalactan peptide 13-like is up-regulated. Knock-down of FAB1A/B, 1-phosphatidylinositol-3-phosphate 5-kinase in Arabidopsis causes defect in the membrane recycling by auxin transporters \[[@CR68]\]. FAB1 also plays an important role in endosomes maturation for mediating cortical microtubule association of endosomes \[[@CR69]\]. Increased expression of 1-phosphatidylinositol-3-phosphate 5-kinase/FAB1D during color change is suggestive of increased auxin transport. Hemicellulose 4-O-methyl glucuronoxylan is a major component present in the secondary cell walls of eudicots. Arabidopsis GXMT catalyzes 4-O-methylation of the glucuronic acid substituents in hemicellulose \[[@CR34]\]. Up-regulation of GXMT 1-like indicates hemicellulose synthesis at color change stage. Phenylpropanoid pathway in colored guava {#Sec14} ---------------------------------------- Anthocyanin biosynthesis genes like F3H \[[@CR70]\], Isoflavone-2′-hydroxylase \[[@CR71]\], Anthocyanidin 3-O-glucosyltransferase 2-like / UGT \[[@CR72]\], 4-Coumarate:CoA ligase and PAL are up-regulated in red skin compared to green skin (Fig. [6](#Fig6){ref-type="fig"}; Fig. [8](#Fig8){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S6). However, expression of multiple transcripts corresponding to these genes are also up-regulated during fruit ripening in AS and PP, pointing the shared phenomenon between fruit ripening and color accumulation. However, doubled expression of PAL in AC red skin compared to AS 3DF (Fig. [7](#Fig7){ref-type="fig"}a; Additional file [13](#MOESM13){ref-type="media"}: Data S9) compels the idea that increased anthocyanin accumulation in red skin might be the result of PAL accumulation. PAL is the check point between primary and secondary metabolism and activity increases in response to biotic and abiotic stresses. PAL in combination with other phenylpropanoid enzymes like 4-Coumarate-CoA ligase, Chalcone synthase, Cinnamic acid 4-hydroxylase, F3H, Flavonol synthase, Stilbene synthase, Isofavone synthase, Resveratrol synthase etc. has been used in different hosts, like *Escherichia coli*, *Saccharomyces cerevisiae*, *Pseudomonas putida* and *Streptomyces spp.* to synthesize a wide range of phenylpropanoid-derived compounds like flavanones, naringenins, kaempferol, quercetin, stilbenes and many more \[[@CR73]\]. Preliminary work of oral or subcutaneous administration of PAL to Phenyl Ketone Uric patients leading to substantial reduction of plasma L-Phe levels has been reported \[[@CR73]\]. Increased expression of PAL in red peel indicates a potential new source of this enzyme in guava. Expression increase in PSY and ACO (Fig. [7](#Fig7){ref-type="fig"}; Additional file [13](#MOESM13){ref-type="media"}: Data S9;) during ripening in AS, PP and AC supports high correlation in color development and maturation in guava but might need interaction with specific factors like RML / GPAT 5 / peamaclein / CTP synthase / MDA chloroplastic / 2OG-AOP1 / SDH / BLH1 (Fig. [9](#Fig9){ref-type="fig"}). Conclusions {#Sec15} =========== Tissue specific and genotypic comparative transcriptome analysis of guava reported here corresponds to the metabolic changes expected and observed in climacteric fruit crops. The transcriptome assembly generated in this study will set the stage for functional genomics in guava, the second most important fruit crop of Northern India. Comparative transcriptome sequence analysis of non-colored vs colored guava cultivars will lead to identification of simple sequence repeat, Insertion-deletion and single nucleotide polymorphism-based markers and be utilized in linkage mapping and other genetic studies. Notably, identification of candidate gene-based markers for red color will aid in generating polymorphic markers and prove a boon for marker assisted breeding for color trait in guava. Methods {#Sec16} ======= Original source of the plant materials {#Sec17} -------------------------------------- Punjab Pink is a hybrid between Portugal x L-49 = F1 x Apple Color and was developed by Punjab Agricultural University, Ludhiana and released in year 2009. CISH-G5/Lalima is an apple color selection of Central Institute of Sub-tropical Horticulture, Lucknow, Uttar Pradesh, India. Allahabad Safeda is an open pollinated seedling selection from Uttar Pradesh, India. The three cultivars were clonally propagated and raised in the mother block of Regional Fruit Research Station - PAU, Bahadurgarh, Patiala, Punjab, India. The 10 years old clonal propagated trees growing in fruit research orchards of PAU, Ludhiana, India have been used in current study. Source of fruits or plant material and tissue sample collection {#Sec18} --------------------------------------------------------------- In the present study we used *Psidium guajava* L. local cultivars Allahabad Safeda (AS), Apple Color (AC: CISH-G5) and Punjab Pink (PP) 10 years old trees growing in Orchards of PAU, Ludhiana, India. Allahabad Safeda is the most prominent cultivar grown throughout India for table purpose. AS fruits are green skinned at immature and harvesting stage (mature fruit, 0DF), starts turning yellow and skin turns completely yellow within 3 days after picking (ripe fruit, 3D) and becomes over-ripe thereafter. Foliage and flower buds of AS, PP and AC are all green however pulp of PP turns to deep pink on maturity (0DF), whereas immature fruit pulp is white. Fruits of AC turn their skin color from green (−5DF) to reddish (0DF) in winter season (Fig. [1](#Fig1){ref-type="fig"}). In AS, we collected actively growing immature young leaves, fully expanded mature leaves, actively growing shoot tips, flower buds at six different developmental stages from immature to mature (1 day before anthesis) and 0D opened flower, 80DPA (immature fruit) with seeds and without seed (ImF), fully mature ready to harvest fruit tissue with seeds and without seeds (0DF), fruit tissue 3 days after harvesting without seeds (3DF) and fruit tissue 7 days after harvesting without seeds (7DF) in three replicates. All the tissues were flash frozen in liquid N~2~ and stored in − 80 °C before proceeding to RNA extraction. PP immature fruit (80 DPA) without seed (ImF_PP), mature fruit without seed (0DF_PP) and AC green peel (AC_GP, −5D fruit) and red peel (AC_RP, 0D fruit) were also harvested, similarly. RNA extraction {#Sec19} -------------- Total RNA was extracted from all the tissues using Spectrum™ Plant Total RNA Kit (Sigma-Aldrich) followed by on-column DNase I digestion (Sigma-Aldrich) for removing DNA contamination. RNA integrity was analysed on 1.2% agarose denaturing gel as described before \[[@CR74]\]. High quality RNA from different tissues was pooled in equimolar ratios to reduce the number of libraries to be sequenced. Briefly, total RNA extracted independently from immature young leaves, fully expanded mature leaves and actively growing shoot tips was pooled in equal amount to make one RNA sample Leaf shoot tip (LSt), 6 flower bud stages and opened flower sample- Mixed flower bud (MFb), 80 DPA immature fruit (ImF), 0D fruit (0DF), 3D ripe fruit (3DF) and 7D overripe fruit (7DF) sample- Mixed fruit (MFr). All the fruit stage samples used to pool MFr were a slice of guava fruit containing peel, pulp and seed altogether. However, independent fruit samples ImF, 0DF, 3DF and, 7DF were also sampled with skin but without seed. So, in total 7 samples representing 7 tissue types of AS were LSt, MFb, MFr, ImF, 0DF, 3DF and 7DF. For PP and AC we extracted 2 RNA samples each ImF_PP & 0DF_PP and AC_GP & AC_RP, respectively. Library preparation, RNA sequencing and reference assembly generation {#Sec20} --------------------------------------------------------------------- Bioanalyzer (Agilent) was used to asses RNA quality before preparing RNA-Seq libraries. Libraries were constructed with ribo-zero treated RNA from 3 biological replicates of AS LSt, MFb, MFr and a single sample for AS ImF, 0DF, 3DF, 7DF, ImF_PP, 0DF_PP, AC_GP & AC_RP. RNA had RIN value ≥8.3 and TruSeq Stranded RNA Library Prep kit (Illumina) was used with an insert size of \~ 300 bp. The paired end (PE) libraries were sequenced on Illumina HiSeq2500. 100 bp high quality PE reads were generated. All the libraries were run in a single lane to avoid any discrepancies while calculating differential expression (FPKM). Read quality was assessed using FASTQC toolkit \[[@CR75], [@CR76]\]. Adapter and low quality sequences were trimmed at minimum PHRED quality score 30 using Trimmomatic read filtering tool \[[@CR77]\]. de novo RNA-seq assembly was generated by pooling AS libraries of LSt, MFb, MFr, ImF, 0DF, 3DF, 7DF with Trinity transcriptome assembler \[[@CR27]\] as no reference assembly for guava is available. Benchmarking Universal Single-Copy Orthologs (BUSCO) analysis was performed using BUSCO pipeline with eudicot model to estimate the completeness of transcriptome assembly. Functional annotation of de novo assembled Allahabad Safeda Transcriptome {#Sec21} ------------------------------------------------------------------------- The assembled transcripts were annotated on the basis of corresponding homologs identified from BLASTX \[[@CR78]\] program with search against NCBI protein "nr" database at e-value of 1e^− 3^. Gene ontology (GO) terms associated with transcripts were determined using BLAST2GO program \[[@CR79]\]. GO enrichment was done among the DETs in specific comparisons using GOseq \[[@CR80]\]. KEGG annotations were also assigned using Blast2GO KEGG mapping. Transdecoder tool in Trinity package was used to identify longest open reading frame (ORF) and protein families were assigned by searching against the Pfam database using pfamscan \[[@CR81]\]. Datafile with transcript ID description, GO annotation and enzyme name are provided in Additional file [13](#MOESM13){ref-type="media"}: Data S1. Measuring gene expression and identification of differentially expressed genes/transcripts {#Sec22} ------------------------------------------------------------------------------------------ The trimmed reads were mapped to reference transcriptome and abundance of transcripts was measured in FPKM values using RSEM estimation tool, for each sample \[[@CR82]\]. Trimmed mean of M-values (TMM) normalization for libraries was performed with Trinity. The expected count values were used for determining the differential expression with edgeR \[[@CR83]\] on biological replicated data. Default dispersion value of 0.1 was used to calculate the differential expression for samples with no replicates. Transcripts exhibiting ≥4-fold change (Log2FC) in expression and \< 0.001 false discovery rate (FDR) were considered significant DETs among different tissues, developmental stages and /or genotypes. Pathway analysis {#Sec23} ---------------- DETs in tissues/genotypes were put together in a fasta file and pathway annotation were determined using online Mercator analysis tools (<http://mapman.gabipd.org>) \[[@CR84]\]. DETs were binned into functional categories by Mercator. MAPMAN software \[[@CR43]\] was used for graphical representation of metabolic and signaling pathways. The results of metablolic pathway analysis described in manuscript can be reproduced by installing MAPMAN and uploading the data set in supplementary files on a computer following authors guidelines described in original manuscript \[[@CR43]\]. qRT-PCR analysis {#Sec24} ---------------- Two micro grams of RNA was reverse transcribed by reverse transcriptase (Maxima First Strand cDNA Synthesis Kit for RT-qPCR containing oligo (dT)18 and random hexamer primers; Thermo-Scientific, Surrey, UK), in a 25 μl reaction further diluted to 100 μl with nuclease free water. Diluted cDNA template (2 μL) was used for a 15 μl PCR reaction. Gene-specific primers (Additional file [5](#MOESM5){ref-type="media"}: Table S5) for candidate genes were used to amplify cDNAs. Histone 3 (comp27670_c0_seq1) was used as internal control. qRT-PCR was performed using PowerUp™ SYBR™ Green Master Mix on a LightCycler® 480 Instrument. Fold change was calculated as described before \[[@CR85]\]. Oligonucleotide primers (Additional file [5](#MOESM5){ref-type="media"}: Table S5) were designed using Primer3 design (<http://frodo.wi.mit.edu/>). Supplementary information ========================= {#Sec25} ###### **Additional file 1: Table S1.** Description of RNA-Seq paired-end data through Illumina high-throughput sequencing. ###### **Additional file 2: Table S2.** Correlation matrix values among Leaf Shoot tip (LSt), Mixed Flower bud (MFb) and Mixed Fruit (MFr) tissue samples and the replicates. ###### **Additional file 3: Table S3.** Expression of top 20 co-up regulated transcripts in Allahabad Safeda fruit tissue compared to leaf and flower with FDR \< 0.001. ###### **Additional file 4:Table S4.** FPKM value of top differentially regulated transcripts of red peel vs green peel of Apple Color (AC) and Punjab Pink (PP) mature red fruit vs Allahabad Safeda (AS) mature fruit in development stages of AS, PP and AC. ###### **Additional file 5: Table S5.** Primer Sequences for candidate genes/internal control for qRT-PCR analysis. ###### **Additional file 6: Figure S1.** Summary of conserved orthologous genes (BUSCO) in the assembled guava transcriptome. ###### **Additional file 7: Figure S2.** Blast2GO distribution of assembled transcripts into A) Biological processes B) Cellular component C) Molecular function. ###### **Additional file 8: Figure S3.** Metabolic overview with MAPMAN analysis of differentially expressed transcripts of mature fruit (0DF) vs Immature fruit (ImF) of cv. Allahabad Safeda. Up- and Down- regulated DETs are represented with blue and red squares, respectively with log2 transformed values. ###### **Additional file 9: Figure S4.** MAPMAN analysis of differentially expressed transcripts of ripe fruit (3DF) vs mature fruit (0DF) of cv. Allahabad Safeda A) Metabolic Overview B) part of regulation overview C) proteasome and autophagy. Up- and Down- regulated DETs are represented with blue and red squares, respectively with log2 transformed values. ###### **Additional file 10: Figure S5.** MAPMAN analysis of differentially expressed transcripts of apple color skin vs green skin of Apple Color -- CISH-G5, shows up-regulation of the secondary metabolism pathway. Up- and Down- regulated DETs are represented with blue and red squares, respectively with log2 transformed values. ###### **Additional file 11 : Figure S6.** Cluster analysis of differentially expressed transcripts among fruit stages of Allahabad Safeda (AS), Apple Color (AC) and Punjab Pink (PP) immature fruit (ImF), mature fruit (MF), 3 days after harvesting (3DF), 7 days after harvesting (7DF), green peel and red peel. ###### **Additional file 12: Figure S7.** Gene Ontology enrichments between A) red and green peel of Apple Color B) mature fruit of Punjab Pink and Allahabad Safeda. ###### **Additional file 13.**Guava_Reference_Transcriptome_Data files. **Data S1.** Functional Annotation of Allahabad Safeda Transcriptome Assembly. **Data S2.** Differential expression among different tissue types, mixed flower buds vs leaf & shoot tip (MFb vs LSt), mixed fruit stages vs leaf & shoot tip (MFr vs LSt) and mixed fruit stages vs mixed flower buds (MFr vs MFb) of Allahabad Safeda. **Data S3.** Differential expression between mature but unripened fruit (0DF) vs immature fruit (ImF) of Allahabad Safeda. **Data S4.** Differential expression between 3 days ripe fruit (3DF) vs mature unripened fruit (0DF) of Allahabad Safeda. **Data S5.** Differential expression between 7 days over-ripe fruit (7DF) vs 3 days ripe fruit (3DF) of Allahabad Safeda. **Data S6**. Differential expression between red peel vs green peel of Apple Color. **Data S7.** Differential expression between Punjab Pink mature fruit (PP_0DF) vs Allahabad Safeda mature fruit (AS_0DF). **Data S8.** FPKM values of DETs among fruit and peel stages of 3 genotypes. **Data S9.** Expression of genes involved in important pathways for fruit maturation, ripening and pathogenesis in different tissues of 3 guava genotypes. AC : Apple Color/CISH-G5/Lalima AC_GP : Apple Color green peel AC_RP : Apple Color red peel AS : Allahabad Safeda BUSCO : Benchmarking Universal Single-Copy Orthologs DETs : Differentially expressed transcripts FPKM : Fragments Per Kilobase Million ImF : Immature fruit ImF_PP : Immature fruit of Punjab Pink LSt : Leaf and shoot tip MFb : Mixed flower buds tissue MFr : Mixed stage fruit tissue PP : Punjab Pink 0DF_PP : Mature fruit of Punjab Pink 0DF : Mature ready to harvest fruit 3DF : 3 days after harvesting fruit 7DF : 7 days after harvesting fruit **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/s12864-020-06883-6. Authors are thankful to the subDIC center (BTISNET, DBT) PAU for providing computational facility. Authors thank Dr. M.R. Dinesh, Director-Indian Institute of Horticultural Research, Bengaluru, and Dr. Binay Panda, Ganit Labs, Bengaluru for useful discussions and Mr. Bhupinder Singh, Mr. Manish Jindal and Dr. Parva Sharma for their technical support. AM, KS and MISG conceived the idea; AM, KS, ISY, NKA and MISG planned experiments; MISG, NKA and RSB maintained and provided the guava genotypes; AM, NKA and RSB collected the plant tissue; MM and AM optimized the RNA extraction method from various tissue types & prepared RNA for sequencing; ISY and AM analyzed the RNA-Seq data and did bioinformatics analysis; AM, PC and MM planned and conducted qRT-PCR for validating the assembly and color responsive candidate genes; WE and PK supported Bioinformatics data analysis; AM, ISY and PC wrote the paper. All the coauthors read the manuscript and approved it. Experiments were financially supported by initial grant to Amandeep Mittal from self-financing scheme of Punjab Agricultural University, Ludhiana and Department of Biotechnology Grant no BT/PR24373/AGIII/103/1012/2018. Trinity generated transcripts of length \> 200 bp submitted to NCBI Transcriptome Shotgun Assembly (TSA) has accession no. GGPP00000000. RNA-seq data of *cv.* Allahabad Safeda is submitted under Bioproject PRJNA472130 with 13 biosamples (SAMN09227265--77). Raw reads are submitted under Short Reads Archive (SRA) database for LSt (SRR7186630, SRR7186631, SRR7186633), MFb (SRR7186632, SRR7186634, SRR7186635), MFr (SRR7186629, SRR7186636, SRR7186637), ImF (SRR7186628), 0DF (SRR7186640), 3DF (SRR7186639) and 7DF (SRR7186638). Reads for Punjab Pink (PP ImF - SRR7471728 & PP 0DF - SRR7471727) and Apple Color -- CISH G5 (AC_GP - SRR7471740 & AC_RP - SRR7471739) can also be found at NCBI - SRA. Not applicable. Not applicable. Authors declare no financial and non-financial competing interest.

{.273} {.274} {.275} {.276} {.277} {.278} {.279} {.280} {.281} {.282} {.283} {.284} {.285} {.286} {.287} {.288}

Handled by Dr S. Bustin 1. Introduction {#sec0005} =============== As technology has advanced, transcriptomics at the single cell level has become not only possible but preferable due to greater recognition of sample heterogeneity. Single cell experiments are becoming increasingly common in the form of RNA sequencing, qPCR, and digital PCR (dPCR). It is broadly presumed that the measurements are becoming more accurate with these new methods but one must be preemptively cautious and take note of the variability and uncertainty in transcriptomics data. Transcriptomics measurements almost invariably include a reverse transcription (RT) step, where RNA transcripts are used as templates to generate cDNA transcripts for quantification. This significantly complicates data interpretation as techniques are not directly measuring RNA transcript number, and results are therefore dependent on the efficiency of the RNA to cDNA conversion. Alternative RT-free methodologies exist and involve direct sequencing of RNA or hybridisation of probes to individual RNA molecules. However, these methods also have limitations as they are currently expensive and still struggling with accuracy and throughput, and poor hybridisation efficiency \[[@bib0005], [@bib0010], [@bib0015], [@bib0020]\]. Problems with the reverse transcription step are many \[[@bib0025], [@bib0030], [@bib0035], [@bib0040]\]. A multitude of research articles have been published that address the effects of modifying individual components or steps of the RT reaction, providing a resource for RT efficiency optimization in experimental design. These modifiable parameters include but are not limited to priming strategy \[[@bib0040],[@bib0045]\], choice of RT enzyme \[[@bib0030],[@bib0035],[@bib0045], [@bib0050], [@bib0055], [@bib0060]\], choice of PCR priming site \[[@bib0030]\], target RNA concentration \[[@bib0030],[@bib0040],[@bib0045]\], background RNA concentration \[[@bib0040], [@bib0045], [@bib0050]\], and RNA quality \[[@bib0030]\]. Results reported from such studies are often inconsistent; one of the few undisputed findings to come from collating this research is that the effects of changing these parameters appear to be gene-dependent \[[@bib0025],[@bib0035],[@bib0040],[@bib0055],[@bib0065]\]. Strategies to improve reverse transcription have been addressed in some detail using population-based RT-qPCR experiments, and many recommendations have been made based on these results. Here, we explore RT methods a step further by examining this problem in the context of single cell analyses using absolute quantification by digital PCR (dPCR). The underlying and consistent experimental and analytical focus is to investigate the efficiency and variability of RT-dPCR in order to determine the consequences of the reverse transcription step in this experimental system. 1.1. The problem of efficiency and variability in reverse transcription {#sec0010} ----------------------------------------------------------------------- A large proportion of transcriptomics is concerned with relative differences between samples. In such scenarios, simplifying analysis by assuming global 100% efficiency may be justified. The relatively recent release of dPCR with claims of accurate direct quantification point towards the ability to use this system in instances where absolute numbers are important. For example, accurate interpretation of data to attain absolute numbers is both relevant and critical in validating a model especially if low numbers of factors are present and ratios of different factors are important. In this situation, it is imperative to understand the efficiency and variability of the system to properly interpret the data. Several published articles have addressed this question by attempting to put a value on RNA-to-cDNA conversion efficiency, yet results vary widely with different experimental conditions. Some cited efficiency ranges are 49--114% \[[@bib0070]\], 50--77% \[[@bib0075]\], 0--102% \[[@bib0050]\], and 39--65% \[[@bib0045]\]. This wide variety effectively illustrates the problem and is likely a combined outcome of the many parameters that are different within and between tests, including the specific transcripts measured. In addition to variable efficiencies across different transcripts, one must consider the reproducibility of reverse transcription for a single, particular transcript. In a study of RT efficiency variability, Linden et al \[[@bib0035]\] showed that some genes had much more variability in efficiency than others, and did not correlate with the general transcription efficiency in each reaction. One of these genes was ACTB, a commonly used reference gene. Similar results have been reported elsewhere \[[@bib0040]\]. The issue of reproducibility is of particular concern in single cell studies where there is little scope for replication to help average away technical differences. Reproducibility is also of great relevance to other areas dealing in absolute quantification of RNA, such as the increasing interest in using RT-dPCR for clinical applications (for example the detection of RNA biomarkers) \[[@bib0070],[@bib0080]\]. This highlights the importance of characterizing assay variability in order to avoid drawing unreliable conclusions from results \[[@bib0030]\]. 1.2. Sensitivity to variations in reaction conditions {#sec0015} ----------------------------------------------------- In designing transcriptomics experiments, optimal conditions allow the best efficiency (closest to 100%) with the lowest variability. Considering the range of outcomes reported across different studies, it is clear that the performance of the RT step is greatly influenced by the context of the experiment, dismissing the possibility of a one-size-fits-all approach to RT optimization. Therefore, it is important to note that previous data published in the literature is not always transferrable, and often certain optimization choices are not compatible with the proposed experiment. Examples of such constraints include avoiding gene-specific primers when one needs to store cDNA for later, yet-to-be-decided analysis, or the necessity for random primers when studying non-adenylated transcripts. In the case of single-cell digital PCR, there are many caveats and constraints. For example, with the direct-lysis single-cell RNA preparation method used for the studies presented here, there is no way to modify or even evaluate some characteristics of the sample, such as target RNA concentration or RNA quality. Another area of significant limitation with single cell analysis is replication. Tichopad et al \[[@bib0085]\] note that use of sampling replicates can estimate the boundaries of technical noise, however only a single sample is available with single cell studies. Using dPCR, Sanders et al \[[@bib0075]\] showed that efficiency problems extend to this method and again, that this effect is gene-dependent. Consequently, while studies have shown the accuracy of digital PCR for DNA quantification \[[@bib0090],[@bib0095]\], the technique does not negate the issues identified with reverse transcription. The authors suggest that a calibrant sample with defined value could help account for effects of enzyme efficiency, inhibitors and molecular dropout, while noting that due to the differences between targets, only gene-specific calibrators would be appropriate. Most efforts in reverse transcription optimization studies up until now have focused on RT-qPCR. Some previously identified improvements could conceivably affect the dPCR output. Previous qPCR findings relevant to single cell studies include that the use of background RNA was shown to exert some of its effect via the qPCR step \[[@bib0050]\] and that dilution of the RT reaction greatly influenced subsequent qPCR, especially in the presence of low background RNA \[[@bib0050]\]. The transferability of these effects from qPCR to dPCR is unknown and exploring such space is constrained by the nature of working with single cells, given limited sample and limits of detection at the single-cell scale. It is likely that some variability and reduced efficiency will always remain, regardless of how well a reverse transcription reaction has been optimized. Clearly, rather than relying on previous publications for 'best-practice' protocols to allow one to ignore the problem, we should use prior published data as a baseline from which to work to better understand and accept the limitations of our own system and build this into our individual data interpretation. This study was conducted as an example of how reverse transcription efficiency can be taken into consideration when performing a transcriptomics experiment. First, a non-exhaustive optimization experiment was run using information from the literature to determine optimal RT conditions in our specific system. Subsequently, a range of efficiency and variability values corresponding to a number of genes of interest was determined, which can then be integrated into downstream analyses of data. Finally, this data is used to guide a brief discussion of the implications of the efficiency data in the context of higher-throughput transcriptomics methods. 2. Results and discussion {#sec0020} ========================= 2.1. Optimization of RT conditions {#sec0025} ---------------------------------- In this study we explored four genes linked to myeloid haematopoiesis and the common reference gene ACTB. Based on previous literature \[[@bib0045],[@bib0050]\], first, a limited optimization experiment was performed, testing a small number of conditions considered most likely to improve the reverse transcription efficiency from our standard protocol. The design included a fully factorial test of three different RT enzymes (SuperScript III VILO Kit, Life Technologies; Superscript II, Life Technologies; Protoscript, New England Biolabs), three concentrations of random hexamer primers (6 uM/as directed; 25 uM; 100 uM), and two concentrations of background yeast RNA (10 pg; 250 ng). Each condition was tested with duplicates of two different transcripts at a known concentration, and measured the efficiency by digital PCR. SSIII VILO LR, LH represents our standard protocol. Results, displayed in [Fig. 1](#fig0005){ref-type="fig"}, indicated that in our system, SuperScript III VILO was the best performing enzyme, and the addition of extra random hexamers to 25 uM improved upon the efficiency compared with our standard protocol. The addition of high concentration of yeast RNA did not improve efficiency in this case.Fig. 1Comparison of efficiency values from RT optimization experiment. SSIII VILO: SuperScriptIII VILO Kit; SSII: SuperScriptII; LR: low yeast RNA (10 pg); HR: high yeast RNA (250 ng); LH: low hexamer (6 uM/as directed); MH: mid hexamer (25 uM); HH: high hexamer (100 uM). Results are presented as mean of duplicate data points with standard deviation. RT efficiency \> 100% is possible given random hexamer primers were used.Fig. 1 These results support previous observations that emphasise the gene-dependent nature of reverse transcription efficiency, as there was a doubling in efficiency with one transcript while only minimal improvement with the other using the increased hexamer concentration. This suggests that the determined optimal condition from the tested conditions above may differ for any other transcripts of interest and there will not be a condition optimal across all scenarios. This position underscores the specified goal of identifying best possible conditions given practical and system-based constraints, and in parallel to identify associated efficiency and variability values and incorporate them into the data analysis. 2.2. Identifying variability and efficiency values {#sec0030} -------------------------------------------------- Once the 'optimal' RT conditions were defined, variability and efficiency tests with IVRS (in vitro RNA synthesis)-produced transcripts were run for our five genes of interest. ACTB was included as a comparison given its wide use as a reference gene despite having been shown to exhibit high RT variability \[[@bib0035]\]. First, 10 RT replicates per transcript were measured using a single dilution at three concentrations (10 fg, 1 fg and 100 ag/reaction) to determine inherent variability of the RT step for each transcript. The coefficient of variation (CV) was less than 12% for all transcripts at the 10 fg and 1 fg level (see [Fig. 2](#fig0010){ref-type="fig"}A and [Table 1](#tbl0005){ref-type="table"}). Despite the fact that these values incorporate both RT and PCR variability, they compare favourably with PCR component-only CV values reported in the literature \[[@bib0070],[@bib0080]\] and are within the guidelines for dPCR equipment specifications for repeated readings (Bio-Rad specifies QX200 precision as ±10%). However, at the 100 ag level the CV ranged from 12 to 35% depending on the transcript, indicating an increased variability at this concentration. It is uncertain whether this increase is primarily driven by an inherent increase in RT variability at the lower concentration or stochasticity associated with small molecule numbers, for example when pipetting from the master mix, and is likely to be some combination of the two factors.Fig. 2Variability and efficiency tests for transcripts of interest using EvaGreen dPCR. A) Variability. Ten replicates were performed per concentration for each transcript (nine for ACTB 100 ag), and coefficient of variation calculated for each. B) Efficiency. Five dilution replicates were performed per concentration for each transcript (four for ACTB 10 fg, ACTB 1 fg, ACTB 100 ag, CEBPA 1 fg, PU.1 1 fg), and mean efficiency values calculated for each. More information can be found in [Table 1](#tbl0005){ref-type="table"} and values used for calculations are included in the Supplementary Information.Fig. 2Table 1Variability and efficiency values for transcripts of interest using EvaGreen dPCR. Data used for graphs in [Fig. 2](#fig0010){ref-type="fig"} are shaded.Table 1 Based on these results, RT variability was considered to be minimal for the majority of the reactions, and RT replicates were not performed for the following steps. It is important to note for data analysis that measurements at the 100 ag level (54--130 transcripts/reaction for our particular genes) carry significant variability and modest differences between cells should be approached with caution. Subsequently, to determine efficiency values while keeping in mind the possibility of pipetting inconsistencies during dilution, five dilution replicates of each transcript were measured over a range of three concentrations, corresponding to a theoretical medium, high and very high transcript expression level in a single cell \[[@bib0100]\]. The medium level of expression was considered to be towards the lower limit of what was detectable using EvaGreen dPCR with a reasonable signal-to-noise ratio (unpublished results). Replicate average efficiency values were similar across concentrations for each transcript ([Fig. 2](#fig0010){ref-type="fig"}B), although the error is much more pronounced at low concentration (100 ag/transcript). Similar to results reported above, CV of all transcripts at 10 fg and 1 fg level is less than 15%, indicating minimal impact from dilution uncertainty when compared with RT replicates and giving confidence in calculated efficiency values. However, at the 100 ag level the coefficient of variation for some transcripts was above 60%, showing additional stochasticity introduced with dilution in addition to RT, and casting doubt on the utility of using standards for measuring transcript levels at this low concentration. Significantly, there is a wide range of efficiency values across the different transcripts, ranging from a combined average of 120% for EPOR to 55% for GCSFR. This illustrates how important knowledge of this value is for interpreting data in a quantitative setting. It is clear that data obtained without adjustment of this discrepancy is inaccurate. This is especially problematic in a quantitative experiment where the aim is to enumerate absolute numbers of transcripts rather than relative values. 2.3. Incorporation into downstream analyses {#sec0035} ------------------------------------------- The specific calculated efficiency values for synthetic transcripts can be incorporated as a normalisation factor into dPCR experiments performed in an identical manner. To arrive at a corrected absolute transcript number, measured transcript numbers from the dPCR experiment can be divided by the efficiency value for each transcript. For single cell experiments, the measured transcript numbers are likely to be at the low end of the concentrations tested where the efficiency values are highly variable. Therefore, we would recommend using the mean efficiency value for all concentrations tested. Of course, this also indicates large variability in the single cell reactions, and all data should be approached with this presumption. With an estimate of the variability present in the reaction at a defined expression level, the information can be incorporated into significance calculations and interpretation of results by both using the variance as bounds on model input variables and by capturing the effect this may have on the model output. In measuring higher transcript numbers (i.e. a very highly expressed transcript at single cell level or an experiment involving more than one cell per reaction) it might be preferable to use the efficiency value determined for the concentration closest to that being measured, assuming low variability for that particular concentration. 2.4. Comparison across assays {#sec0040} ----------------------------- Digital PCR experiments may be run using two different detection chemistries, EvaGreen and probes. It is easy to switch between the two methods as they are performed in a highly similar manner with the same hardware and procedure. These parallel methods provide a means to explore the impact of detection chemistry on efficiency and variability values. To test the applicability of efficiency values as a useful tool across PCR detection methods, probe-based dPCR assays were run for three targets (GATA1, PU.1 and CEBPA) using the same dilution series as the EvaGreen experiment, with the addition of a 10 ag concentration to leverage the improved signal-to-noise ratio of the probe assay. Substantial differences were observed between the efficiency values obtained from the two experiments, as seen in [Fig. 3](#fig0015){ref-type="fig"}; data for the probe experiment is shown in Supplementary Figure S1 and Supplementary Table S1.Fig. 3Comparison of efficiency values obtained with EvaGreen and probe assays. Results are presented as mean of replicate data points with standard deviation. There is a discrepancy in GATA1/PU.1 ratio in the higher concentration probe results not evident in the EvaGreen results.Fig. 3 While the values for CEBPA are consistent, GATA1 and PU.1 show a divergence in efficiency values between the two assays. While the efficiencies of reverse transcription of the two transcripts are reasonably similar using the EvaGreen assay, they are markedly different using the probe-based assay, such that the ratio of GATA1 to PU.1 is highly affected depending on the approach used. Based on these values, adoption of the EvaGreen efficiency value for probe assay data would overestimate GATA1 concentration by 60% relative to PU.1 concentration (ratio of GATA1/PU.1 normalisation factors = 0.86 for EvaGreen assay, 0.53 for probe assay). This would substantially bias the outcome of any model where quantitative data is expected. This calculation illustrates the necessity of determining empirically the performance of the assay in each specific situation and negates the possibility of broadly applying efficiency values between platforms. This is not unexpected given the evidence outlined above regarding the sensitivity of RT reactions to modifications in protocol. 2.5. Implications for high-throughput technologies {#sec0045} -------------------------------------------------- As demonstrated here, measurement of reverse transcription efficiency should be incorporated into any experiment involving quantitative RNA assessment. While adopting this recommendation is relatively easy for smaller scale experiments, it becomes unworkable for more high-throughput approaches such as single cell sequencing. As single-cell sequencing experiments are becoming more common, there is increasing discussion about the appropriate method of analysing such data. Using a well-studied, homogeneous cell line, Marinov et al. \[[@bib0100]\] demonstrated a 6-fold range in RNA content between single cells, leading to the assertion that in such a single-cell setting the often-used expression unit 'fragments per kilobase per million mapped reads' (FKPM) is misleading and absolute transcripts should be counted. Consequently, this and other studies \[[@bib0105],[@bib0110]\] have attempted to calculate absolute transcript numbers from sequencing data using spike-in standards. While spike-in RNA standards (such as ERCC spike-ins \[[@bib0115]\], Sequins \[[@bib0120]\], and SIRVs \[[@bib0125]\]) allow calculation of a global efficiency value based on a pre-determined subset of synthetic genes for assay performance and quality control purposes, the data presented here indicate this can not be extrapolated to estimate efficiency values for specific genes of interest in the data. Indeed, it has previously been reported using bulk RNA sequencing that reproducible transcript-dependent discrepancies show absolute measurements using RNA-seq are inaccurate \[[@bib0130]\]. Therefore, while the addition of spike-ins to calculate global efficiency may allow an improvement on current methods for analysing single cell gene expression, it still only gives a relative value for comparing between samples that cannot be compared across transcripts, and is not able to give absolute quantitative information about RNA molecules in a cell. Instead, synthetic standards should be incorporated to gain efficiency values for subsequent validation tests for significant findings. 3. Conclusions {#sec0050} ============== The data presented in this publication and other works cited here reiterate gene-to-gene variability in reverse transcription efficiency and highlight the necessity of considering RT-efficiency when working with quantitative data. The wide range of values calculated for the six synthetic RNAs tested may have a significant impact on the quantitative analysis of the downstream transcriptomics data. Despite the between-gene variability, efficiency values are sufficiently reproducible at all but the lowest concentration within the confines of a particular protocol such that this variability can be mitigated through incorporation of synthetic standards as controls. Standards specific to each experiment are necessary as the efficiency values have been shown to be highly sensitive to alterations of even a single component of the workflow. Certain assays are limited in throughput (for example dPCR which allows for low-level multiplexing) and it is reasonable to include controls for all the genes investigated in an assay. We suggest including a synthetic transcript for each transcript measured across at least 3 biologically relevant concentrations in technical replication to gain a metric for normalisation of data, and find the lower limit for reproducibility and detection of signal above noise. While this is not possible for higher-throughput technologies such as single-cell sequencing, standards should be included in targeted follow-up validation assays if quantitative claims are made. Measurement of lowly-expressed transcripts using dPCR at the single cell level is likely to be highly variable, and analysis of technical variability should be factored in to final conclusions. 4. Materials and methods {#sec0055} ======================== 4.1. Synthetic RNA standards {#sec0060} ---------------------------- ### 4.1.1. Construction of synthetic standards {#sec0065} Synthetic RNA standards were made for the following transcripts: GATA1, PU.1, CEBPA, GCSFR, EPOR, ACTB. Lengths of RNA transcripts ranged from 1428 to 3414 nucleotides. Full-length mRNA has been shown to be the most accurate method of setting up RNA standards \[[@bib0045]\]. Therefore, transcripts were amplified using Phusion High-Fidelity DNA Polymerase (New England Biolabs) from cDNA isolated from haematopoietic cells with primers designed at the ends of the mRNA transcript with leeway to ensure acceptable primer design. Primers used for amplification are outlined in Supplementary Info. cDNA was cloned into pGem-T Easy plasmid (Promega) and IVRS performed using linearized plasmid and HiScribe T7 High-Yield RNA Synthesis Kit (New England Biolabs). In vitro-synthesised RNA was isolated using Nucleospin RNA II kit (Macherey-Nagel), treated with TURBO DNase (Ambion), and cleaned up using RNA Clean & Concentrator Kit (Zymo). Sample quantity and quality were assessed using Qubit RNA Assay (ThermoFisher Scientific) and RNA Pico Bioanalyzer chip (Agilent) respectively, and stored in Lo-Bind tubes (Eppendorf) at -80**°**C. Stock concentrations were between 100 ng/uL and 700 ng/uL. All experiments were performed within six months of RNA generation. ### 4.1.2. Dilution of synthetic standards {#sec0070} Evagreen and probe dPCR efficiency test: A single dilution series was performed for each transcript for the 10 fg and 1 fg variability test. Samples were diluted to ˜10 ng/uL in water and measured again with Qubit RNA Assay (ThermoFisher Scientific). From this working stock, aliquots of 1 ng/uL were prepared for all transcripts. A series of 1 in 10 dilutions in a final volume of 20 u L was performed until desired concentrations were reached. All dilutions were performed in Lo-Bind tubes (Eppendorf). 5 dilution replicates: A separate dilution series comprising five replicates per transcript, beginning with five aliquots at 1 ng/uL, was performed for the EvaGreen and probe efficiency tests as outlined above. Dilution 1 for each of these transcripts was also used for the 100 ag variability test. Dilutions were prepared on the same day as each of the studies were conducted. 4.2. cDNA synthesis {#sec0075} ------------------- ### 4.2.1. cDNA synthesis for RT optimisation {#sec0080} Reverse transcription reactions were set up and run in 96-well Twin-tec semi-skirted LoBind plates (Eppendorf). Reactions were designed to measure 100 ag of PU.1 and GATA1 RNA in 5 uL reactions containing 5 mg/mL UltraPure BSA (Ambion), 5U SUPERaseIn RNase Inhibitor (ThermoFisher Scientific), 1X RT Reaction Mix/Buffer, 0.5X RT enzyme, variable concentrations of yeast RNA (10 pg and 250 ng; Ambion), and variable concentrations of random hexamers (6 uM/as directed, 25 uM and 100 uM; Integrated DNA Technologies). Primer concentration in the 1× SSIII VILO Reaction Buffer was not specified and for the purposes of this experiment was assumed to be 6 uM. SuperscriptII and Protoscript reactions also required the addition of 500uM dNTP Mix (New England Biolabs) and 10 mM DTT (ThermoFisher Scientific). A denaturation step at 65 °C for 5 min was performed before the addition of RT enzyme. The temperature profile of the SSIII VILO reaction was 25**°**C for 10 min, 50**°**C for 30 min, 55**°**C for 25 min, 60**°**C for 5 min, and 70**°**C 15 min. Temperature profile of SSII and Protoscript reactions were 25**°**C for 10 min, 42 **°**C for 30 min, 48**°**C for 25 min, 50**°**C for 5 min and 70**°**C for 15 min. Upon completion of each reverse transcription (RT) reaction, 96-well plates were spun to recover individual reaction volumes and cDNA was stored at −80 °C for up to two weeks. After thawing on ice, all RT reactions were transferred to a single new plate for dPCR. ### 4.2.2. cDNA Synthesis for RT efficiency/variability test {#sec0085} Reverse transcription reactions were set up and run in 96-well Twin-tec semi-skirted LoBind plates (Eppendorf). Reactions were designed to measure 10 technical replicates of three concentrations (100 ag, 1 fg, 10 fg) of six target transcripts (PU.1, GATA1, CEBPA, GCSFR, EPOR, ACTB) in 5 uL reactions containing 5 mg/mL UltraPure BSA (Ambion), 5U SUPERaseIn RNase Inhibitor (ThermoFisher Scientific), 1X VILO Reaction Mix, 0.5× SSIII enzyme, 10 pg yeast RNA (Ambion), and 19 uM additional random hexamers (Integrated DNA Technologies). A denaturation step at 65 °C for 5 min was performed before the addition of RT enzyme. The temperature profile was 25**°**C for 10 min, 50**°**C for 30 min, 55**°**C for 25 min, 60**°**C for 5 min, and 70**°**C 15 min. Upon completion of each cDNA synthesis reaction, 96-well plates were spun to collect volume and cDNA was stored at 4 °C for up to four days. 4.3. Digital PCR {#sec0090} ---------------- ### 4.3.1. EvaGreen dPCR {#sec0095} Samples were prepared for dPCR in 22 uL reactions containing cDNA, 1× ddPCR^™^ EvaGreen Supermix (Bio-Rad), and primers. For targets PU.1, GATA1, CEBPA, and ACTB primers were designed and validated in-house and used at 200 nM for each oligo (Integrated DNA Technologies). For targets GCSFR and EPOR, PrimePCR^™^ EvaGreen Assay (Bio-Rad) were used at 1× primer mix. Primer sequences (or context sequences for commercial assays) are provided in the Supplementary Information. No template RT controls contained only yeast RNA and were included for each primer set. No-RT controls were performed previously for each transcript and confirmed absence of DNA. Droplets were created using an Automated Droplet Generator (BioRad) followed by the recommended PCR thermocycling protocol using a C1000 Thermal Cycler (Bio-Rad): 95 °C for 10 min, followed by 40 cycles of 95 °C for 30 s and 58 °C for 60 s, and a final signal stabilization cycle of 4 °C for 5 min and 90 °C for 5 min. [A]{.smallcaps} QX200^™^ Droplet Reader (BioRad) was used for signal detection. ### 4.3.2. Probe dPCR {#sec0100} Samples were prepared for dPCR in 22 uL reactions containing cDNA, 1X ddPCR^™^ Supermix for Probes (No dUTP; Bio-Rad) and 1X PrimePCR^™^ ddPCR^™^ Expression Probe Assay primers/probe mix (Bio-Rad): FAM-PU.1, HEX-GATA1 and HEX-CEBPA. PU.1 and GATA1 were run as duplex samples, CEBPA as singleplex. Context sequences are provided in the Supplementary Information. No template RT controls contained only yeast RNA and were included for each primer set. Droplets were created using an Automated Droplet Generator (BioRad) followed by the recommended PCR thermocycling protocol using a C1000 Thermal Cycler (Bio-Rad): 95 °C for 10 min, followed by 40 cycles of 94 °C for 30 s and 55 °C for 60 s, and a final incubation at 98 °C for 10 min. [A]{.smallcaps} QX200^™^ Droplet Reader (BioRad) was used for signal detection. ### 4.3.3. dPCR data analysis {#sec0105} QuantaSoft^™^ Analysis Pro analysis software (Bio-Rad) was used to determine absolute transcript numbers. A threshold for defining positive droplets was set manually by comparison with control samples. The number of positive droplets was used by the software to perform a Poisson correction to give an absolute number of transcripts per microliter. These results were multiplied by the total sample reaction volume of 22 uL for a final absolute quantification of a given target. 4.4. Calculations {#sec0110} ----------------- ### 4.4.1. Calculation of efficiency values {#sec0115} The exact sequence of each transcript were determined by taking the plasmid sequence starting at the final G nucleotide of the T7 promoter and continuing through to the final base before cleavage at the linearization site. Molecular weights of each synthetic RNA standard was calculated according to transcript sequence using the following formula:where An, Un, Cn, and Gn are the numbers of A, U, C, and G bases, respectively, and the additional 159 corresponds to the weight of the 5′ triphosphate group. This value was used to determine the number of transcripts present in each tested concentration. The numbers calculated for each tested concentration can be found in the Supplementary Information. Numbers of transcripts detected by dPCR for each sample were divided by the theoretical number in the reaction to arrive at an efficiency value. ### 4.4.2. Calculation of variability values {#sec0120} The coefficient of variation for each transcript was calculated by dividing the standard deviation of the number of transcripts per reaction for 10 replicates by the mean, expressed as a percentage. Funding source {#sec0125} ============== This work was supported by Stem Cells Australia. Appendix A. Supplementary data {#sec0135} ============================== The following are Supplementary data to this article: Supplementary data associated with this article can be found, in the online version, at <https://doi.org/10.1016/j.bdq.2018.12.002>. [^1]: These authors contributed equally

Introduction {#Sec1} ============ Nowadays, there are all kinds of complex systems with specific functions in the real world such as online social systems, medical systems and computer systems. These systems can be abstracted into networks with complex internal structures, called complex networks. The research of complex networks has received more and more attention due to the development of the Internet. Community structure \[[@CR6], [@CR23]\] is a common feature of complex networks, which means that a network consists of several communities, the connections between communities are sparse and the connections within a community are dense \[[@CR10]\]. Mining the community structure in the network is of great significance to understand the network structure, analyze the network characteristics and predict the network behavior. Thus, community detection has become one of the most important issues in the study of complex networks. In recent years, a great deal of research is devoted to community detection in networks. Most community detection methods are used to identify non-overlapping communities (i.e., a node belongs to only one community). The main approaches include graph partitioning and clustering \[[@CR9], [@CR10], [@CR13]\], modularity maximization \[[@CR1], [@CR20]\], information theory \[[@CR12], [@CR25]\] and non-negative matrix factorization \[[@CR16], [@CR27]\]. The Kernighan-Lin algorithm \[[@CR13]\] is a heuristic graph partitioning method that detects communities by optimizing the edges within and between communities. GN algorithm \[[@CR10]\] is a representative hierarchical clustering method, which can find communities by removing the links between communities. Blondel et al. proposed the Louvain algorithm \[[@CR1]\], which is a well-known optimization method based on modularity. It is used to handle large-scale networks due to low time complexity. Liu et al. \[[@CR16]\] put forward a community detection method by using non-negative matrix factorization. Zhao et al. \[[@CR33]\] introduced the idea of granular computing into the community detection of network and proposed a community detection method based on clustering granulation. The existing non-overlapping community detection algorithms have made great achievements, but these algorithms only use the traditional two-way decisions \[[@CR29], [@CR30]\] method (the acceptance or rejection decision) to deal with the overlapping nodes between communities. Compared with the two-way decisions method, the three-way decisions theory (TWD) \[[@CR28]\] adds a non-commitment decision. The main idea of TWD is to divide an entity set into three disjoint regions, which are denoted as positive region (POS), negative region (NEG) and boundary region (BND) respectively. The POS adopts the acceptance decision, the NEG adopts the rejection decision, and the BND adopts the non-commitment decision (i.e., entities that cannot make a decision based on the current information are placed in the BND). For entities in the BND, we can further mine more information to realize their final partition. The introduction of non-commitment decision can effectively solve the decision-making errors caused by insufficient information, which is more flexible and closer to the actual situation. How to deal with the boundary region has become a key issue for three-way decisions community detection. At present, the commonly used methods to process the boundary region include modularity increment \[[@CR20]\] and similarity calculation \[[@CR2], [@CR8]\]. But these methods only take advantage of the local features of the network, without considering the information of the divided communities and the similarity of the global structure. Therefore, how to tackle the boundary region effectively is a challenge. In this paper, we propose a three-way decisions community detection model based on weighted graph representation (WGR-TWD). The graph representation can well transform the global structure of the network into vector representation and make the two nodes in the boundary region that appear in the same community more similar by using the weight. Firstly, the multi-layered community structure is constructed by hierarchical clustering. The target layer is selected according to the extended modularity value of each layer. Secondly, all nodes are converted into vectors by weighted graph representation. Finally, nodes in the boundary region are divided into positive or negative region based on cosine similarity. Thus, non-overlapping community detection is realized. The key contributions of this paper can be summarized as follows: We use weighted graph representation to obtain the global structure information of the network to guide the processing of the boundary region, which gets a better three-way decisions community detection method.Based on the knowledge of the communities in the target layer, we make the two nodes connected by a direct edge in the boundary region more similar by using frequency of appearing in the same community as the weight. Then the walk sequences are constructed according to the weight of the edge. Finally, the Skip-Gram model is used to obtain the vector representation of nodes. Therefore, the weighted graph representation method is realized.We evaluate the effectiveness the proposed model WGR-TWD on real-world networks compared with the baseline methods. The experimental results show the superior performance of our model. The rest of this paper is organized as follows. We introduce related work in Sect. [2](#Sec2){ref-type="sec"}. We give the detailed description of our algorithm in Sect. [3](#Sec5){ref-type="sec"}. Experiments on real-world networks are reported in Sect. [4](#Sec8){ref-type="sec"}. Finally, we conclude the paper in Sect. [5](#Sec13){ref-type="sec"}. Related Work {#Sec2} ============ Community Detection of Hierarchical Clustering {#Sec3} ---------------------------------------------- Hierarchical clustering method has been widely used in community detection due to the hierarchical nature of the network structure. This approach can be divided into two forms: divisive method and agglomerative method. The divisive method removes the link with the lowest similarity index repeatedly, while the agglomerative method merges the pair of clusters with the highest similarity index repeatedly. These two methods eventually form a dendrogram, and communities are detected by cutting the tree. The research of community detection based on hierarchical clustering has received widespread attention from scholars. Girvan and Newman proposed the GN algorithm \[[@CR10]\], which is a typical divisive method. Clauset et al. \[[@CR5]\] proposed a community detection algorithm based on data analysis, which is a representative agglomerative method. Fortunato et al. \[[@CR9]\] presented an algorithm to find community structures based on node information centrality. Chen et al. proposed the LCV algorithm \[[@CR4]\] which detects communities by finding local central nodes. Zhang et al. \[[@CR32]\] introduced a hierarchical community detection algorithm based on partial matrix convergence using random walks. Combining hierarchical clustering with granular computing, we introduce an agglomerative method based on variable granularity to build a dendrogram. Given an undirected and unweighted graph $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G= \left( V,E \right) $$\end{document}$, where *V* is the set of nodes, *E* denotes the set of edges. The set of neighbor nodes to a node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\end{document}$ is denoted as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N\left( v_i \right) = \left\{ v_j\in V|\left( v_i,v_j \right) \in E \right\} $$\end{document}$, the set of initial granules 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}$$H^{1}=\left\{ C_{1}^{1},C_{2}^{1},...,C_{p}^{1} \right\} $$\end{document}$. The formation process of the initial granules is as follows. First, we calculate the local importance of each node in the network. The local importance of a node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\end{document}$ is defined as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} I\left( v_{i} \right) = \frac{\left| Z \right| }{\left| N\left( v_{i} \right) \right| }, \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}$$Z=\left\{ v_{j} \in N\left( v_{i} \right) \mid d\left( v_{j} \right) \le d\left( v_{i} \right) \right\} $$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d\left( v_{i} \right) $$\end{document}$ is the degree of node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\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}$$\left| \cdot \right| $$\end{document}$ denotes the number of elements in a set. Second, all important nodes are found according to the local importance of nodes. The node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\end{document}$ is an important node if $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left( v_{i} \right) > 0$$\end{document}$. Finally, for any important node, an initial granule is composed of all neighbor nodes of the important node and the important node itself. After all the initial granules are obtained, the hierarchical clustering method based on variable granularity is described. The clustering coefficient between the two granules is defined as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} f\left( C^m_i,C^m_j \right) = \frac{\left| C^m_i\cap C^m_j \right| }{min\left\{ \left| C^m_i \right| ,\left| C^m_j \right| \right\} },C^m_i,C^m_j\in H^m \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}$$\left| C^m_i\cap C^m_j \right| $$\end{document}$ denotes the number of common nodes in granules $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^m_i$$\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}$$C^m_j$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$min\left\{ \left| C^m_i \right| ,\left| C^m_j \right| \right\} $$\end{document}$ is the smaller number of nodes in granules $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^m_i$$\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}$$C^m_j$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^m$$\end{document}$ is the granules set of the *m*th ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m= 1,2,...$$\end{document}$) layer. The collection of clustering thresholds is denoted as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda =\left\{ \lambda _{m},m=1,2,... \right\} $$\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}$$\lambda _{m}$$\end{document}$ is the clustering threshold of the *m*th layer. In order to automatically obtain the clustering threshold of each layer, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda _{m}$$\end{document}$ is defined as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{m} = med\left\{ f\left( C_{i}^{m},C_{j}^{m} \right) |\forall C_{i}^{m},C_{j}^{m}\in H^{m} ,C_{i}^{m}\cap C_{j}^{m}\ne \varnothing \wedge i\ne j \right\} \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}$$med\left\{ \right\} $$\end{document}$ is a median function. The clustering process is as follows. Firstly, for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\forall C^m_i,C^m_j\in H^m$$\end{document}$, the clustering coefficient between them is calculated. Then the clustering threshold $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda _{m}$$\end{document}$ of the current layer is calculated. And the maximum clustering coefficient is found, which is denoted as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f\left( C^m_\alpha ,C^m_\beta \right) $$\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}$$f\left( C^m_\alpha ,C^m_\beta \right) \geqslant \lambda _{m}$$\end{document}$, the two granules $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C^m_\alpha $$\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}$$C^m_\beta $$\end{document}$ are merged to form a new granule and the new granule is added to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{m+1}$$\end{document}$. Otherwise, all the granules in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^m$$\end{document}$ are added to $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{m+1}$$\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}$$H^m$$\end{document}$ is set to empty. For each layer, repeat above clustering process until all nodes in the network are in a granule. Therefore, a dendrogram is built. DeepWalk {#Sec4} -------- Traditional network representation usually uses high-dimensional sparse vectors, which takes more running time and computational space in statistical learning. Network representation learning (NRL) is proposed to address the problem. NRL aims to learn the low-dimensional potential representations of nodes in networks. The learned representations can be used as features of the graph for various graph-based tasks, such as classification, clustering, link prediction, community detection, and visualization. DeepWalk \[[@CR24]\] is the first influential NRL model in recent years, which adopts the approach of natural language processing by using the Skip-Gram model \[[@CR18], [@CR19]\] to learn the representation of nodes in the network. The goal of Skip-Gram is to maximize the probability of co-occurrence among the words that appear within a window. DeepWalk first generates a large number of random walk sequences by sampling from the network. These walk sequences can be analogized to the sentences of the article, and the nodes are analogized to the words in the sentence. Then Skip-Gram can be applied to these walk sequences to acquire network embedding. DeepWalk can express the connection of the network well, and has high efficiency when the network is large. The Proposed Algorithm {#Sec5} ====================== Weighted Graph Representation {#Sec6} ----------------------------- To effectively deal with overlapping communities in the target layer, a weighted graph representation approach is proposed. At first, a weighted graph is constructed according to the community structure of the target layer. The weights of edges in an unweighted graph are defined as follows$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W_{ij}=1.0+\sigma _{ij}/N_{c} \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}$$\sigma _{ij}$$\end{document}$ is the number of communities in which nodes $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\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}$$v_j$$\end{document}$ appear in a community at the same time, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{c}$$\end{document}$ is the total number of communities in the target layer. After that, an improved DeepWalk (IDW) model is used to acquire the vector representation of all nodes in the graph. Unlike DeepWalk, the IDW model constructs the walk sequences according to the weight of the edge. The greater the weight, the higher the walk probability. Assume that the current walk node is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\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}$$v_{j}\in N\left( v_{i} \right) $$\end{document}$, then the walk probability from node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\end{document}$ to node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_j$$\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} P\left( v_{i}\rightarrow v_{j} \right) = \frac{W_{ij}}{\sum \limits _{v_{k}\in N\left( v_{i} \right) }W_{ik}}. \end{aligned}$$\end{document}$$After obtaining all the walk sequences, the Skip-Gram model is used to learn the vector representation of nodes from the walk sequences. The objective function of IDM is as follows$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \min \limits _R\sum \limits _{-\omega \leqslant j\leqslant \omega ,j\ne 0}-log P\left( v_{i+j}|R\left( v_{i} \right) \right) \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}$$R\left( v_{i} \right) $$\end{document}$ is the vector representation of node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_i$$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega $$\end{document}$ is the window size which is maximum distance between the current and predicted node within a walk sequence. Thus, the vector representation of all nodes in the network is obtained. The WGR-TWD Algorithm {#Sec7} --------------------- Fig. 1.The framework of the proposed method. We will present the proposed WGR-TWD algorithm in this section. Figure [1](#Fig1){ref-type="fig"} shows the overall framework of the proposed algorithm. Our algorithm consists of two parts: the construction of multi-layered community structure and boundary region processing. The first part, we employ the hierarchical clustering method based on variable granularity to construct a multi-layered community structure according to Sect. [2.1](#Sec3){ref-type="sec"}. Some overlapping communities exist in the multi-layered community structure because of clustering mechanism, so we use the extended modularity (EQ) \[[@CR26]\] to measure the partition quality of each layer. It is defined as follows$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} EQ= \frac{1}{2m}\sum _{i}\sum _{u\in C_{i},v\in C_{i}}\frac{1}{O_{u}O_{v}}\left( A_{uv}-\frac{d_{u}d_{v}}{2m} \right) \end{aligned}$$\end{document}$$where *m* is the number of edges in the network, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{i}$$\end{document}$ represents a community, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O_{u}$$\end{document}$ is the number of communities that node *u* belongs to, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{uv}$$\end{document}$ is the element of adjacent matrix, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d_{u}$$\end{document}$ is the degree of node *u*. A larger EQ value means better performance for overlapping community division. Thus, we select the layer corresponding to the largest EQ value as the target layer. The second part introduces the method of dealing with overlapping communities in the target layer. Since there are overlapping communities in the target layer, we need to further divide the target layer to achieve non-overlapping community detection. Therefore, the three-way decisions theory (TWD) is introduced to handle overlapping communities. Based on the idea of TWD, we define the overlapping part in the communities as boundary region (BND), and the non-overlapping part as positive region (POS) or negative region (NEG). And our goal is to process nodes in the BND. First of all, we adopt the weighted graph representation method to learn the vector representation of all nodes in the network. After that, the nodes in the BND are divided into the POS or NEG by using cosine similarity. Suppose the vector of node *u* is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u= \left( x_{1},x_{2},...,x_{n} \right) $$\end{document}$, node *v* is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v= \left( y_{1},y_{2},...,y_{n} \right) $$\end{document}$, then the cosine similarity is defined as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S\left( u,v \right) = \frac{\sum \limits _{i=1}^{n}x_{i}y_{i}}{\sqrt{\sum \limits _{i=1}^{n}\left( x_{i} \right) ^{2}}\cdot \sqrt{\sum \limits _{i=1}^{n}\left( y_{i} \right) ^{2}}}. \end{aligned}$$\end{document}$$For arbitrary node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{i}$$\end{document}$ in the BND, find out all communities containing node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{i}$$\end{document}$ in the target layer, calculate the average value of cosine similarity between node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{i}$$\end{document}$ and non-overlapping nodes in each community as the similarity between node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{i}$$\end{document}$ and this community, then join node $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{i}$$\end{document}$ into the community corresponding to the maximum similarity and update the community structure of the target layer. Repeat the above operation until all nodes in the BND are processed. The WGR-TWD algorithm is described in Algorithm 1. Experiments {#Sec8} =========== Datasets {#Sec9} -------- We test the performance of our method on eight real-world datasets in which each dataset is described as follows, and the main information of those datasets are shown in Table [1](#Tab1){ref-type="table"}. Zachary's karate club \[[@CR31]\]. This is a social network of friendships between 34 members of a karate club at a US university in the 1970s. Dolphin social network \[[@CR17]\]. It is an undirected social network of frequent associations between 62 dolphins in a community living off Doubtful Sound, New Zealand. Books about US politics \[[@CR22]\]. A network of books about US politics published around the time of the 2004 presidential election and sold by the online bookseller Amazon.com. Edges between books represent frequent co-purchasing of books by the same buyers. American college football \[[@CR10]\]. A network of American football games between Division IA colleges in 2000. Email communication network \[[@CR21]\]. It is a complex network which indicates the email communications of a university. The network was composed by Alexandre Arenas. Facebook \[[@CR15]\]. The network was collected from survey participants using Facebook app. Geom \[[@CR11]\]. The authors collaboration network in computational geometry. Collaboration \[[@CR14]\]. The network is from the e-print arXiv and covers scientific collaborations between authors papers submitted to High Energy Physics Theory category.Table 1.Information of datasets.NetworkNodesEdgesReal clustersKarate34782Dolphin621592Polbooks1054413Football11561312Email11335451UnknownFacebook403988234UnknownGeom734311898UnknownCollaboration987725998Unknown Baseline Methods {#Sec10} ---------------- In this paper, two representative algorithms are chosen to compare with the proposed WGR-TWD, as shown below:Modularity increment (MI) \[[@CR3]\]. A hierarchical clustering method based on variable granularity, and the overlapping nodes between communities are divided according to modularity optimization.DeepWalk \[[@CR24]\]. It is a network representation learning method. This approach is used to handle the overlapping communities in the target layer. Evaluation Metrics {#Sec11} ------------------ We employ two widely used criteria to evaluate the performance of community detection algorithms. The first index is modularity (*Q*) \[[@CR5]\], which is often used when the real community structure is not known. *Q* is defined as follows$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Q= \frac{1}{2m}\sum _{i,j}\left[ A_{ij}-\frac{d_{i}d_{j}}{2m} \right] \delta \left( c_{i},c_{j} \right) \end{aligned}$$\end{document}$$where *m* is the number of edges in the network, *A* is the adjacent matrix, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d_{i}$$\end{document}$ is the degree of node *i*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_{i}$$\end{document}$ represents the community to which node *i* belongs, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( c_{i},c_{j} \right) = 1$$\end{document}$ when $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_{i}= c_{j}$$\end{document}$, else $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \left( c_{i},c_{j} \right) = 0$$\end{document}$. The higher the modularity value, the better the result of community detection. Another index is normalized mutual information (NMI) \[[@CR7]\], which is defined as follows$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} NMI= \frac{-2\sum _{i=1}^{C_{A}}\sum _{j=1}^{C_{B}}C_{ij}log\frac{C_{ij}n}{C_{i.}C_{.j}}}{\sum _{i=1}^{C_{A}}C_{i.}log\frac{C_{i.}}{n}+\sum _{j=1}^{C_{B}}C_{.j}log\frac{C_{.j}}{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}$$C_{A}$$\end{document}$ ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{B}$$\end{document}$) denotes the number of communities in partition *A* (*B*), $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{ij}$$\end{document}$ is the number of nodes shared by community *i* in partition *A* and by community *j* in partition *B*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{i.}$$\end{document}$ ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{.j}$$\end{document}$) represents the sum of elements of matrix *C* in row *i* (column *j*), and *n* is the number of nodes in the network. A higher value of NMI indicates the detected community structure is closer to the real community structure. Experimental Results {#Sec12} -------------------- In the networks with known real partition (the first four small networks), we use two indicators (*Q* and NMI) to evaluate our algorithm. Table [2](#Tab2){ref-type="table"} presents the community detection results of the proposed algorithm and the baseline algorithms on networks with known real partition. We can see that our method obtains best results on the Karate and Football datasets. On the Dolphin dataset, the *Q* value of our method is better and the NMI value is second to the DeepWalk method. On the Polbooks dataset, our method performs not well because the connections between nodes are sparse which is difficult to mine the structural information of the network.Table 2.Experimental results on networks with known real partition (under EQ criterion).NetworkIndexMIDeepWalkWGR-TWDKarate*Q*0.3600.360**0.371**NMI0.8370.837**1.000**Dolphin*Q***0.385**0.379**0.385**NMI0.814**0.889**0.830Polbooks*Q*0.439**0.442**0.441NMI**0.469**0.4480.453Football*Q*0.4800.545**0.558**NMI0.6260.726**0.732** To further verify the effectiveness of the proposed algorithm, the MI method is used to deal with each layer in the multi-layered community structure. And we select the layer corresponding to the maximum *Q* value as the target layer. The experimental results are shown in Table [3](#Tab3){ref-type="table"}. Compared with Table [2](#Tab2){ref-type="table"}, Table [3](#Tab3){ref-type="table"} can obtain higher *Q* value. Combined with Tables [2](#Tab2){ref-type="table"} and [3](#Tab3){ref-type="table"}, our method can get better community detection results compared with the baseline methods.Table 3.Experimental results on networks with known real partition (under *Q* criterion).NetworkIndexMIDeepWalkWGR-TWDKarate*Q*0.3910.391**0.401**NMI0.6060.620**0.700**Dolphin*Q*0.5100.520**0.523**NMI**0.619**0.6130.618Polbooks*Q***0.514**0.4920.509NMI**0.501**0.4630.489Football*Q*0.582**0.5960.596**NMI0.9040.899**0.911** We also conducted experiments on four large networks. On these networks, the real partition is unknown. Therefore, we only use modularity to evaluate the performance of different methods. Table [4](#Tab4){ref-type="table"} shows the community detection results of the proposed method and baseline methods. On the first three networks, we can see that our method obtains better results compared with the two baseline methods. On the Collaboration dataset, the MI method achieves the best performance which is a little bit higher than our method. The main reason is that the Collaboration network is very sparse which leads to poor vector representation of the nodes. In conclusion, the proposed method effectively addresses the problem of non-overlapping community detection in networks.Table 4.Modularity values on networks with unknown real partition.NetworkMIDeepWalkWGR-TWDEmail0.5370.538**0.542**Facebook0.7710.770**0.772**Geom0.7020.710**0.711**Collaboration**0.723**0.7180.720 Conclusion {#Sec13} ========== In this paper, we propose a method for three-way decisions community detection based on weighted graph representation. The target layer in multi-layered community structure is selected according to the extended modularity value of each layer. For the overlapping communities in the target layer, the weighted graph representation can well transform the global structure into vector representation and make the two nodes in the boundary region more similar by using frequency of appearing in the same community as the weight. Finally, the nodes in the boundary region are divided according to cosine similarity. Experiments on real-world networks demonstrate that the proposed method is effective for community detection in networks. This work is supported by the National Natural Science Foundation of China (Grants Numbers 61876001) and the Major Program of the National Social Science Foundation of China (Grant No. 18ZDA032).

There is little risk of cell biologists\' getting bored in the 21^st^century, but it is worth considering a few of the questions they might hope to have solved by 2100, if not before. Fundamental features of cells ============================= Despite all the efforts of the 20^th^century there are many areas of basic cell biology that remain mysterious. Some of these questions, such as how secreted proteins get across the Golgi, or the source of the membranes of autophagosomes, have been debated for so long that a further century might not be enough to reach agreement. Other phenomena such as lipid rafts, exosomes, or unconventional secretion seem to have divided the protagonists into camps of believers and skeptics who consider each other\'s views too eccentric to even engage. Other questions, however, have started to receive attention so recently that there may not yet be sufficient entrenched views to impede progress. These include the question of how organelles and cytoskeletal structures, and indeed the cell itself, maintain constant size and shape. In addition, the extent of non-vesicular transport of lipids between organelles has only recently been appreciated, and exciting recent work has revealed the importance of membrane contact sites \[[@B1]\], and also suggested that the transport of cholesterol to the plasma membrane is mediated by oxysterol binding proteins via a counter-current of phosphoinositides \[[@B2]\], although perhaps inevitably an opposing view has formed that oxysterol binding proteins do not actually perform cholesterol transport but instead activate an as yet unidentified transport machinery \[[@B3]\]. Variations on a theme of Henrietta Lacks ======================================== Much of the work of the molecular era has concentrated on readily tractable cells, notably immortalized mammalian tissue culture cells, or powerful genetic systems such as yeast. Although many important general principles have emerged from these studies, they represent only a tiny fraction of the rich diversity of cell types that populate the protozoal world, or make complex multicellular creatures from oak trees to chimpanzees. One key challenge will be to understand how the basic machinery of organelles and cytoskeletal systems that all cells share is then regulated and enhanced to achieve this astonishing diversity. What cells should we care about? ================================ It is unlikely that the planet\'s tax payers will be willing to pay for enough cell biologists to investigate every last intriguing invertebrate or bizarre bikont, and thus future work is likely to focus on particular key cells types, especially those found in tax payers themselves. Much of our body is made up of sheets of polarized epithelial and endothelial cells, whose shapes form tissues, and whose polarity allows the regulation of our internal fluids. How these cells are polarized is beginning to be understood, but there is still much to be learned about how proteins are directed to the different sides of these cells, and how their cytoskeletons are regulated to direct the changes in cell shape that form and maintain tissues. An equally challenging and crucial question concerns the formation of the neurons and glia that are converting this text into your consciousness. We have a good, if not complete, understanding of how synapses work, but understand little of how membrane traffic and the cytoskeleton work together to establish and maintain the extraordinary cellular architecture of the brain. Life without growth =================== Another feature of the cells that most cell biologists study is that you find you have more of them when you return to the lab in the morning. This is of course very useful, but it has meant that relatively little work has been done on the cell biology of post-mitotic or quiescent cells. Such cells form the majority of our tissues, and in addition to their cell-type-specific features, their lack of growth makes it likely that their membrane traffic and cytoskeletal systems will share features that are distinct from those in cells that must be continuously expanding up to division. Temperature, the forgotten variable =================================== Cycling through this morning\'s heavy frost deepened my gratitude at being a homeotherm, and the cells growing in my lab can also look forward to another day of consistent incubation. However, not all organisms are so lucky and indeed for much of evolutionary history organism body temperature must have varied depending on the time of day and season. Indeed, even many extant vertebrates, including some mammals, do not maintain a constant body temperature. Thus, there must be mechanisms to ensure that biological processes are robust to temperature changes, which is likely to be a particular issue for membranes where fluidity varies with temperature. Moreover, it may place a constraint on what sorts of cellular mechanisms could have emerged in evolution before we reached our homeothermic state. For instance, phase separations are coming back into fashion as a mechanism for organising membranes and the cytoplasm, but phase behaviour can be highly sensitive to temperature changes, and so to be biologically useful it must be of a type that is robust to the sorts of temperature fluctuations that our remote ancestors would have experienced on a daily basis.

Background {#Sec1} ========== Cholera is a bacterial disease caused by infection of small intestine by *Vibrio cholerae*, characterized by a variety of diarrhea, abdominal cramp and dehydration. Most common route of infection is through contaminated water and foods, and due to the environmental nature of transmission, the control is complicated in tropical environment where clean water is not easily accessible. From 2016, a large epidemic of cholera has been seen in Yemen, and stool samples have tested positive for *V. cholerae*, serotype Ogawa. The incidence once declined in early 2017, but a bigger epidemic was triggered from early-mid April 2017 with unprecedented size of cases \[[@CR1]\]. Essential resources including vaccines, fluids and antibiotics have been allocated. Mathematical modeling studies have contributed to better understanding of the cholera transmission dynamics \[[@CR2]--[@CR7]\]. Many models during the 2010 Haitian cholera outbreak explicitly accounted for the presence of asymptomatic individuals and also the transmission through water environment, because these features, especially the former have been the issue of unrecognized part of the outbreak \[[@CR8]\]. Models have been frequently utilized for theoretically optimizing resource allocation of antibiotics and oral vaccines \[[@CR2]--[@CR8]\] in collaboration with public health experts \[[@CR9]\]. Such modeling effort has been extended to real-time analysis and explicit assessment of predictive ability during the outbreak in Haiti \[[@CR10], [@CR11]\]. However, in the context of Yemen epidemic 2017, a modeling effort for both now-casting and forecasting, elevating situation awareness of both experts and the public, has yet to be made using a parsimonious modeling approach, possibly with real time updates. The present study aims to forecast the cholera epidemic in Yemen, 2017, explicitly addressing the reporting delay and ascertainment bias. Methods {#Sec2} ======= Epidemiological data {#Sec3} -------------------- The present study analyzed the 2017 epidemic that we assumed to have started from 16 April 2017 (the first date of week 16) which coincided with a new increase in the reported incidence in 2017. Situation report of the cholera outbreak, as represented by Weekly Update and Epidemiology Bulletin, has been managed by the World Health Organization (WHO) Regional Office for the Eastern Mediterranean Regional Office, and the present study used weekly counts of suspected cases and deaths as reported in the Epidemiology Bulletin \[[@CR12]\]. Suspected case of cholera adhered to the WHO's classical case definition. That is, a case of cholera was suspected when a patient aged 5 years or more develops acute watery diarrhea, with or without vomiting \[[@CR13]\]. The so-called case fatality risk (CFR), which was calculated in each week as the ratio of weekly count of deaths to that of cases, was also retrieved. Modeling methods {#Sec4} ---------------- Let *c* ~t,∆t~ be weekly "reported" incidence in week *t* which took place in ∆*t* days since the first date of corresponding week: we account for ∆*t* because ∆*t* greatly varied by Epidemiology Bulletin. Let *j* ~t~ be the actual (unbiased) incidence in week *t*. That is, here we explicitly distinguish reported incidence *c* from actual incidence *j*. Setting the latest value of CFR in week 26 as the baseline (i.e. 1.0), the relative CFR of week *t* is denoted by *α* ~t~. *F* ~∆t~ is the cumulative distribution function of the time from illness onset to reporting. We assume that the actual (unbiased) CFR is a constant without any interventions or treatment and that all deaths are reported. Moreover, we assume that *α* ~t~ in and after week 26 remains to be the value of 1. The expected value of the weekly incidence is described as$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ E\left[{c}_{t,\varDelta t}\right]=\frac{j_t}{\alpha_t}{F}_{7\left(s-t\right)+\varDelta t}, $$\end{document}$$where *s* represents the latest week of observation (for *t* ≤ *s*). *F* ~∆t~ was assumed to follow an exponential distribution with mean *δ* days, i.e., *F* ~∆*t*~ = 1-exp(−∆*t* /*δ*), and thus, the product of *j* and *F* captures the reporting delay structure in empirical data. *F* ~∆t~ is important in interpreting the updated epidemic curve every week. As we assume that the WHO's weekly CFR estimate mirrors the frequency of case ascertainment in each week, *α* ~t~ adjusts time-dependent variations in case ascertainment. The unbiased incidence, *j* ~t~ in week *t* is obtained from the cumulative incidence, *J*(*t*):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {j}_t=J(t)-J\left(t-1\right), $$\end{document}$$where *J*(*t*) was assumed to be described by two different parsimonious models. One is the logistic growth curve, i.e.,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ J(t)=\frac{K}{1+ \exp \left(-r\left(t-{t}_i\right)\right)}, $$\end{document}$$where *K* is referred to as the carrying capacity, representing the cumulative incidence at time infinity, i.e., our interest in the present study, *r* is the growth rate and *t* ~i~ is the point of inflection, corresponding to the peak timing of an epidemic. The other model is a slightly more general model, referred to as the generalized logistic model or the so-called "Richards model", i.e.,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ J(t)=\frac{K}{{\left[1+g \exp \left(-r\left(t-{t}_i\right)\right)\right]}^{\frac{1}{g}}}, $$\end{document}$$where *g* is the parameter that partially determines the point of inflection on vertical axis. The model (4) is known as flexible with only 4 parameters and has been widely applied to capture the temporal distribution of epidemics of variety of diseases \[[@CR14]--[@CR17]\]. Assuming that the observed data in week *t*, *c* ~t,∆t~, follows a Poisson distribution, the likelihood function to estimate parameters of abovementioned models (i.e., 1 parameter for *F* and 3 or 4 parameters for *J*) is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ L\left(\theta; D\right)=\frac{E{\left[{c}_{t,\varDelta t}\right]}^{c_{t,\varDelta t}} \exp \left(-E\left[{c}_{t,\varDelta t}\right]\right)}{c_{t,\varDelta t}!}, $$\end{document}$$where *θ* represents the population parameter and *D* the observed data. The maximum likelihood estimates were obtained by minimizing the negative logarithm of (5). Profile likelihood based 95% confidence intervals (CI) were used. To compare model (3) and (4), Akaike Information Criterion (AIC) was employed. Given empirical data for each observed week, we implemented real time forecasting and sequentially updated it every week. The latest forecast that we present is based on the dataset from week 16 to 26, 2017. The forecast of unbiased cumulative number of cases was obtained from parameterized models (3) and (4). Additionally adjusting the reporting delay, we also produced forecasts of weekly incidence that is expected to be observed in week 27, 28, 29 and 30 in WHO's report. Ethical considerations {#Sec5} ---------------------- The present study analyzed data that is publicly available. As such, the datasets used in our study were de-identified and fully anonymized in advance, and the analysis of publicly available data without identity information does not require ethical approval. Results {#Sec6} ======= Figure [1](#Fig1){ref-type="fig"} shows the epidemic curve. The incidence continuously increased from week 16, the first date of which is 16 April 2017 which we use as the initial date of the 2017 epidemic in the following analysis. Counting from 27 April 2017 (from which WHO counted \[[@CR12]\]), a total of 356,591 suspected cases and 1802 deaths were reported as of 17 July 2017. The ratio of weekly deaths to cases (i.e., the so-called CFR in the WHO's report) in the week of 25 June 2017 was 0.27, while that in the week of 16 April 2017 was 1.98. Setting the latest value of CFR (=0.27) as the baseline, those in the week of 16th, 23rd and 30th of April 2017, *α* ~t~, were 7.39, 5.62 and 7.85, respectively, which we assume that they reflect small case ascertainment in April. Incidence of the latest week 26 in Fig. [1a](#Fig1){ref-type="fig"} may look indicating that the epidemic turned into decreasing trend, but Fig. [1b](#Fig1){ref-type="fig"} implicates that always the incidence in the latest week is underestimated in its initial report due to reporting delay. In the following week, the current weekly incidence likely increases.Fig. 1Weekly incidence of the cholera epidemic, Yemen, 2017. **a** Weekly number of suspected cholera cases (*left vertical axis*) along with the case fatality risk (CFR) calculated as the ratio of weekly count of deaths to cases. Week 16 on *horizontal axis* corresponds to 16 April 2017. **b** Reported and updated epidemic curves from 22 May to 3 July 2017. Each epidemic curve represents the weekly number of suspected cases reported by the date of report (i.e., 22, 27, 30 May, 4, 11, 19 and 26 June and 3 July, respectively). Differences in the number of cases reported in the same week between two or more curves reflect the delay in reporting. Week 16 on horizontal axis corresponds to 16 April 2017 Figure [2](#Fig2){ref-type="fig"} shows the real-time fitting results of our model. Both logistic and generalized logistic curves qualitatively captured the observed patterns of reported incidence, and that the overall patterns of agreement have not varied with time. AIC values of using the latest data were 1,986,063 and 2,045,441 for logistic and generalized logistic models, respectively, favoring the simpler model (3). The estimated mean reporting delay were 4.4 days (95% CI: 3.2, 5.8) and 4.0 (95% CI: 2.9, 5.3) days for logistic and generalized logistic models, respectively, and the growth rate was 0.06 (95% CI: 0.05, 0.07) per day and 0.18 (95% CI: 0.09, 0.65) per day, respectively, for these models. Parameter *g* of Richards model was estimated at 6.9 (95% CI: 2.4, 33.2).Fig. 2Real time forecasts of suspected patients of cholera epidemic, Yemen, 2017. The weekly reported incidence that were released on 11, 19, 26 June and 3 July 2017 were compared with forecasts of logistic model and Richards model (i.e., generalized logistic model) on figures **a**), **b**), **c**) and **d**), respectively. Week 16 in 2017 corresponds to the week starting from 16 April and it is the beginning of the second wave of cholera epidemic in Yemen, 2017 Figure [3](#Fig3){ref-type="fig"} shows the forecasted *J*(*t*), representing the actual cumulative incidence. Carrying capacity, or the cumulative incidence at the end of the epidemic, was estimated at 790,778 (95% CI: 700,495, 914,442) cases and 767,029 (95% CI: 690,877, 871,671) cases, respectively, by using logistic and generalized logistic models. Counting days from 1 January 2017 as day 0, the inflection point was estimated at day 168.0 (95% CI: 165.6, 171.1) and 173.3 (95% CI: 169.1, 180.7), respectively, for these models, indicating that the latest week (i.e. week 26) is about the time just after experiencing the inflection point of the epidemic curve.Fig. 3Predicted actual cumulative incidence of cholera epidemic, Yemen, 2017. Days (*horizontal axis*) are counted from 1 January 2017. For both forecasts (one using logistic model and the other using Richards (or generalized logistic) model), the 95% confidence intervals, *dashed lines*, were obtained using the 95% confidence intervals of carrying capacity (note that this is not the result from Bootstrapping of all parameters for simplicity) Figure [4](#Fig4){ref-type="fig"} shows the forecasting result of observed incidence data using model (1), as derived from logistic (Fig. [4a](#Fig4){ref-type="fig"}) and generalized logistic (Fig. [4b](#Fig4){ref-type="fig"}) models. The weekly incidence that is expected to be reported in week 27 and afterwards are predicted to decrease over time. The logistic curve yielded slower decline with evident reporting delay compared with Richards model.Fig. 4Predicted weekly reported incidence to be reported in week 27, 28, 29 and 30, 2017. **a** Logistic model and **b** Richards (generalized logistic) model. Parameter estimates of both models were obtained from the datasets from week 16 to 26 Discussion {#Sec7} ========== The present research study responded to the cholera epidemic in Yemen, 2017 in real time. Using the weekly incidence of suspected cases, updated as a revised epidemic curve every week, the reporting delay was explicitly incorporated into the model. Moreover, using the CFR (as calculated by the WHO which was implemented in a biased manner \[[@CR18]\]), ascertainment bias was adjusted, enabling us to parameterize the family of logistic curves for modeling the unbiased incidence. As a result, it was estimated that we have just passed through the epidemic peak by week 26, and the unbiased cumulative incidence was forecasted to range from 690 to 910 thousand cases by the end of the epidemic. Doing so, while high incidence nearby epidemic peak may lead Yemen's people to feel that the epidemic is somewhat uncontrollable due to massive number of reported cases, we have been able to objectively demonstrate that the epidemic peak is likely over and elevate people's situation awareness through our "now-casting" approach. To our knowledge, this modeling study is the first to elevate the situation awareness of the cholera epidemic in Yemen, 2017 via statistical forecasting of the epidemic. Employing parsimonious models with small number of parameters, the observed patterns of incidence and reporting were appropriately captured and we were successful in obtaining real time forecasts. While the epidemic peak was not directly identifiable as of week 26, 2017 only from reported data, the model has radically indicated that the inflection point has plausibly been passed. Moreover, jointly using the relative CFR with the abovementioned model with reporting delay, we assumed that the time period with high CFR involved relatively low ascertainment, enabling us to strongly feel that the qualitative patterns of unbiased epidemic curve can be captured. Epidemiological modelers will experience similar epidemics in the future. What to be learnt from our modeling exercise are two folds. First, if the weekly incidence is reported in the up-to-the-minute manner and updated in later weeks, not a single data point but the entire epidemic curve must be precisely updated-and-reported. In the Yemen case, we initially intended to implement real time modeling using daily incidence data, but this intent was in vain mainly due to the absence of regular updates in the epidemic curves showing daily incidence. Second, not only the epidemic curve, but death data should also be consistently updated over time, so that it helps researchers to adjust ascertainment bias. Carrying capacity, or the cumulative incidence at the end of the epidemic, is worthy of further discussions. We estimated that *K* ranges from 690 to 910 thousand cases, but that value is dependent on our choice of the weekly CFR in week 26 (i.e. CFR = 0.27). If the actual CFR is greater than 0.27, our estimated *K* may be an overestimate. On the other hand, considering that the empirically observed CFR does not capture mild and asymptomatic infected individuals that are never reflected in the denominator of the CFR estimation, the estimated *K* may be a considerable underestimate. Several limitations must be noted. First, we were not able to assure the quality of suspected cases at an individual level. Sometimes, weekly incidence in a specific week has decreased (rather than increase due to reporting delay) as the epidemic curve was updated, and in such an instance, we had to adopt the smallest latest value regardless of the week of update, because we had to avoid any decrease in the observed incidence as a function of week through minimum arbitrariness for adjustment. Second, the validity of forecast is limited due to the use of parsimonious phenomenological model. Prediction approaches using more mechanistic models would be beneficial. Third, our analysis rests on the dataset of entire Yemen, and we have not been able to dig into more detailed heterogeneous data. Spatio-temporal heterogeneity is known to play a key role in cholera transmission \[[@CR19], [@CR20]\] and that could allow examining the impact of environmental predictors on the transmission dynamics \[[@CR21]\]. Fourth, our model cannot incorporate the population impact of interventions on the transmission dynamics, due to phenomenological nature of the model that we used. While several modeling research subjects remain, the present study acted as the first to contribute to improving situation awareness of cholera epidemic in Yemen, 2017. Although the epidemic is predicted to start to decline, it is vital that ongoing resource allocations and countermeasures are tightly conducted to minimize potential number of victims. Conclusions {#Sec8} =========== The present study responded to the cholera epidemic in Yemen, 2017 in real time. Employing parsimonious models with small number of parameters, the observed patterns of incidence and reporting were appropriately captured. It was estimated that we have just passed through the epidemic peak by the end of week 26, and the weekly incidence was predicted to start to decrease in the following weeks. The unbiased cumulative incidence was forecasted to range from 690 to 910 thousand cases by the end of the epidemic. AIC : Akaike Information Criterion CFR : Case fatality risk CI : Confidence interval WHO : World Health Organization Not applicable. Funding {#FPar1} ======= HN received funding from the Japan Agency for Medical Research and Development, the Japanese Society for the Promotion of Science (KAKENHI Grant Numbers 16KT0130, 16 K15356 and 17H04701), the Japan Science and Technology Agency CREST program (JPMJCR1413) and the RISTEX program for Science of Science, Technology and Innovation Policy. BY received financial support from China Scholarship Council. The funders had no role in the study design, data collection and analysis, decision to publish, or manuscript preparation. Availability of data and materials {#FPar2} ================================== Collected datasets are publicly available from reference \[[@CR12]\]. HN conceived the study. HN, ST, BY, TY and YA conceptualized the study design, collected the data, formulated mathematical model and performed statistical analyses. HN drafted the early version of the manuscript. ST, BY, TY and YA drafted figures. All authors gave comments on the revised manuscript and approved the final version of the manuscript. Authors' information {#FPar3} ==================== The authors are experts with interest in Infectious Disease Epidemiology and also in Theoretical Epidemiology, and the team of lead author is led by professor from Hokkaido University Graduate School of Medicine. Ethics approval and consent to participate {#FPar4} ========================================== Not applicable. Consent for publication {#FPar5} ======================= Not applicable. Competing interests {#FPar6} =================== The authors declare that co-author H. Nishiura is the Editor-in-Chief of Theoretical Biology and Medical Modelling. This does not alter the authors' adherence to all the Theoretical Biology and Medical Modelling policies on sharing data and materials. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

See editorial on page 293. SummaryWe have identified 5 molecularly and clinically relevant subtypes of the CpG island methylator phenotype (CIMP) in colorectal cancer. We show that CIMP-high cancers segregate into distinct subgroups, which display different frequencies of *BRAF* and *KRAS* mutation. These CIMP subtypes are associated with important clinical and molecular features, are correlated with mutations in different epigenetic regulator genes, and show a marked relationship with patient age. Colorectal cancer is a heterogeneous disease characterized by distinct genetic and epigenetic changes that drive proliferative activity and inhibit apoptosis. The conventional pathway to colorectal cancer is distinguished by *APC* mutation and chromosomal instability, and accounts for approximately 75% of sporadic cancers.[@bib1], [@bib2] The remaining colorectal cancers arise from serrated polyps and have activating mutations in the *BRAF* proto-oncogene, frequent microsatellite instability (MSI), and the CpG island methylator phenotype (CIMP).[@bib2], [@bib3] The development of CIMP is critical in the progression of serrated neoplasia.[@bib3] It is well established that CIMP can result in the silencing of key genes important for tumor progression, including the tumor-suppressor gene *CDKN2A* and the DNA mismatch repair gene *MLH1*.[@bib4], [@bib5] Gene silencing mediated by *MLH1* promoter hypermethylation impairs mismatch repair function, which leads to MSI.[@bib5] CIMP can be detected using a standardized marker panel to stratify tumors as CIMP-high, CIMP-low, or CIMP-negative.[@bib3] Activation of the mitogen-activated protein kinase signaling pathway as a result of the *BRAF* mutation is associated highly with CIMP-high. CIMP-high cancers frequently arise proximal to the splenic flexure and are more common in elderly female patients,[@bib2], [@bib3] whereas CIMP-low cancers have been associated with *KRAS* mutation.[@bib6], [@bib7] More recently, consensus molecular subtyping (CMS) was proposed for classifying colorectal cancers based on transcriptional signatures. Guinney et al[@bib8] identified 4 major molecular subtypes (CMS1--CMS4). CMS1, or MSI immune subtype, is characterized by MSI, *BRAF* mutation, and enhanced immunogenicity. CMS2 can be distinguished by chromosomal instability and WNT pathway perturbations. CMS3, or metabolic subtype, is characterized by *KRAS* mutation, CIMP-low status, and infrequent copy number alterations. CMS4, or mesenchymal subtype, shows high copy number aberrations, activation of the transforming growth factor-β signaling cascade, stromal infiltration, and the worst overall survival. The relationship between CIMP and CMS subtypes is currently unclear. Methylation is not a phenomenon distinct to neoplasia. Changes in the epigenome also occur with age and in response to environmental factors.[@bib9], [@bib10] We previously showed that the promoter region of certain genes becomes increasingly methylated in normal colonic mucosa with age.[@bib9] CIMP-high cancers are identified primarily in older patients,[@bib2] hence, age-related hypermethylation might prime the intestinal epigenome for serrated neoplasia-type colorectal cancers. Methylation also is critical in the progression of serrated pathway precursors to invasive cancer, primarily through methylation of MLH1 at the transition to dysplasia.[@bib11], [@bib12] Thus, the natural history of the cancer within the colorectum may dictate the methylation profile of the cancer once malignancy develops. DNA methylation alone can be insufficient to induce transcriptional repression.[@bib13] Gene repression also is associated with repressive histone marks such as the H3K27me3 mark,[@bib14] which is catalyzed by the polycomb-repressor-complex 2. Modification of histone tails is catalyzed by a series of enzymes including epigenetic readers, which scan for histone modifications; writers, which effect the addition of a modification; and erasers, which are responsible for the removal of histone marks. Mutations in genes encoding epigenetic enzymes have been shown to occur frequently in cancer.[@bib15] Although DNA methylation is associated classically with gene silencing, the relationship between DNA methylation and histone modifications has not been fully elucidated, and the role of somatic mutations in enzymes that catalyze these epigenetic processes has not been examined comprehensively. In this study, we define the extent and spectrum of DNA methylation changes occurring in colorectal cancers and relate this to key clinical and molecular events characteristic of defined pathways of tumor progression. We investigate the role of DNA methylation in the modulation of gene transcription, and assess mutation of genes encoding epigenetic regulatory proteins. Results {#sec1} ======= Clinical and Molecular Features of the Consecutive Cohort in Comparison With the Cancer Genome Atlas Cohort {#sec1.1} ----------------------------------------------------------------------------------------------------------- Genome-wide DNA methylation levels were assessed in 216 unselected colorectal cancers ([Table 1](#tbl1){ref-type="table"}). The mean age of patients at surgery was 67.9 years. Twenty-nine of 216 (13.4%) cancers had a *BRAF* V600E mutation, and 75 of 216 (34.7%) cancers were mutated at *KRAS* codons 12 or 13. Mutation of *BRAF* and *KRAS* were mutually exclusive. Patients with *BRAF* mutated cancers were significantly older than patients with *BRAF* wild-type cancers (mean age, 74.9 vs 66.9 y; *P* = .01). *TP53* was mutated in 78 of 185 (42.2%) cancers. MSI was associated significantly with *BRAF* mutation (18 of 29 *BRAF* mutant vs 9 of 187 *BRAF* wild-type cancers; *P* \< .0001). By using the Weisenberger et al[@bib3] panel to determine CIMP status, 24 of 216 (11.1%) were CIMP-high, 44 of 216 (20.4%) were CIMP-low, and 148 of 216 (68.5%) were CIMP-negative. CIMP-high was associated significantly with *BRAF* mutation compared with *BRAF* wild-type cancers (19 of 29 vs 5 of 186; *P* \< .0001). CIMP-low was associated significantly with *KRAS* mutation compared with *KRAS* wild-type cancers (26 of 75 \[34.6%\] vs 18 of 141 \[12.8%\]; *P* \< .001).Table 1Clinicopathologic Details of the 216 Colorectal Adenocarcinomas as Stratified for Methylation-Based CIMP Clustering, Measured on Illumina HM450 Arrays, Using the 5000 Most Variable CpG Sites That Were Not Hypermethylated in Normal Mucosal TissuenCIMP-H1CIMP-H2CIMP-L1CIMP-L2CIMP-Neg*P* valueTotal, n2162322526653Mean age, *y*67.975.273.470.166.861.9\<.0001Sex Male100 (46.4%)5 (21.7%)9 (40.9%)24 (46.2%)35 (53.0%)27 (50.9%).11 Female116 (53.7%)18 (78.3%)13 (59.1%)28 (53.8%)31 (47.0%)26 (49.1%)Site Proximal75/213 (35.2%)19 (82.6%)13 (59.1%)20 (39.2%)15 (23.4%)8 (15.1%)\<.0001 Distal96/213 (45.1%)4 (17.4%)6 (27.3%)21 (41.2%)32 (50.0%)33 (62.3%) Rectal42/213 (19.7%)03 (13.6%)10 (19.6%)17 (26.6%)12 (22.6%)CIMP status CIMP-high24 (11.1%)16 (69.6%)3 (13.6%)3 (5.8%)2 (3.0%)0\<.0001 CIMP-low44 (20.4%)6 (26.1%)13 (59.1%)16 (30.8%)8 (12.1%)1 (1.9%) CIMP-neg148 (68.5%)1 (4.3%)6 (27.3%)33 (63.5%)56 (84.8%)52 (98.1%)Mutation *KRAS* mutant75 (34.7%)4 (17.4%)12 (54.5%)34 (65.4%)19 (28.8%)7 (13.2%)\<.0001 *BRAF* mutant29 (13.4%)17 (73.9%)2 (9.1%)6 (11.5%)4 (6.0%)0 (0%)\<.0001 *TP53* mutant77/185 (41.6%)12/21 (57.1%)6/21 (28.6%)18/45 (40.0%)22/54 (40.7%)19/44 (43.2%).45Microsatellite instability MSI26 (12.0%)11 (47.8%)1 (4.8%)8 (15.4%)6 (9.1%)0\<.0001 MSS190 (88.0%)12 (52.2%)21 (95.2%)44 (84.6%)60 (90.9%)0CMS CMS135 (16.2%)16 (69.6%)4 (18.2%)5 (9.6%)9 (13.6%)1 (1.9%)\<.0001 CMS268 (31.5%)04 (18.2%)10 (19.2%)30 (45.5%)24 (45.3%) CMS353 (24.5%)3 (13.0%)12 (54.5%)21 (40.4%)10 (15.2%)7 (13.2%) CMS460 (27.8%)4 (17.4%)2 (9.1%)16 (30.8%)17 (25.8%)21 (39.6%)Stage I30/1110/155/11 (45.5%)8/30 (26.7%)13/35 (37.1%)4/20 (20.0%).15 II33/1117/15 (46.7%)1/11 (9.1%)10/30 (33.3%)10/35 (28.6%)5/20 (25.0%) III34/1116/15 (40.0%)4/11 (36.4%)7/30 (23.3%)11/35 (31.4%)6/20 (30.0%) IV14/1112/15 (13.3%)1/11 (9.1%)5/30 (16.7%)1/35 (2.9%)5/20 (25.0%)LINE170.368.7568.9672.0570.4569.67.38[^1][^2] We collected a subset of 32 matched noncancerous mucosal samples from patients in the consecutive cohort. The mean age of patients within the cohort of matched normal samples was 68.9, and was not significantly different than the mean age of patients in the wider cohort (*P* = .71). Methylation-Based Clustering Shows 5 Subtypes of Colorectal Cancer With Distinct Clinical and Molecular Features {#sec1.2} ---------------------------------------------------------------------------------------------------------------- We examined the extent and spectrum of DNA methylation changes in these 216 colorectal cancers using Illumina HumanMethylation450 BeadChip arrays (Illumina Inc, San Diego, CA). Five clusters were identified by recursively partitioned mixed model (RPMM) clustering ([Figure 1](#fig1){ref-type="fig"}). These included 2 clusters with high levels of methylation that we have designated CIMP-H1 and CIMP-H2; 2 clusters with intermediate levels of methylation, CIMP-L1 and CIMP-L2; and a single cluster with low levels of methylation, CIMP-neg. There was a significant stepwise increase in age between clusters concordant with increasing genomic methylation (CIMP-neg, 61.9 y; CIMP-L2, 66.8 y; CIMP-L1, 70.1 y; CIMP-H2, 73.4 y; and CIMP-H1, 75.2 y; *P* \< .0001) ([Table 1](#tbl1){ref-type="table"}).Figure 1Methylation heatmap of unselected 216 colorectal cancers using the 5000 most variable β values in CpG sites that were not hypermethylated in normal mucosal tissue. Clustering was performed using the RPMM R package. Clustering showed 5 distinct clusters, termed CIMP-H1, CIMP-H2, CIMP-L1, CIMP-L2, and CIMP-Neg. This was faithfully recapitulated in TCGA. The CIMP-H1 subgroup comprised 23 of all 216 (10.6%) cancers and was enriched for female patients (18 of 23, 78.3%; *P* \< .0001) and for tumors located proximal to the splenic flexure (19 of 23, 82.6%; *P* \< .0001). We observed no differences in cancer stage at diagnosis and methylation cluster. The CIMP-H1 cluster was strikingly enriched for cancers with features characteristic of serrated neoplasia, including *BRAF* mutation (17 of 23, 73.9%; *P* \< .0001), CIMP-H status was determined using the Weisenberger et al[@bib3] marker panel (16 of 23, 69.6%; *P* \< .0001), MSI (11 of 23, 47.8%; *P* \< .0001), and consensus molecular subtype CMS1 (16 of 23, 69.6%; *P* \< .0001) ([Table 1](#tbl1){ref-type="table"}, [Figure 1](#fig1){ref-type="fig"}). *TP53* was mutated in 12 of 21 (57.1%) CIMP-H1 cluster cancers. CIMP-H2 cluster cancers also frequently arose in the proximal colon (consecutive cohort, 13 of 22; 59.1%). CIMP-H2 cancers were KRAS mutant more often than CIMP-H1 cancers (54.5% vs 17.4%), and were less often TP53 mutant when compared with the rest of the cohort (28.6%). The incidence of MSI within these cancers was low (4.8%). The frequency of the metabolic CMS3 subtype was higher than in the other CIMP subtypes (54.5%). CIMP-H2 cancers were significantly less likely to be identified as CIMP-high using the Weisenberger et al[@bib3] MethyLight panel when compared with CIMP-H1 cancers (13.6% vs 69.6%; *P* \< .001). CIMP-L1 cancers were significantly enriched for KRAS mutation (65.4%; *P* \< .0001), and were identified equally in the distal and proximal colon. These cancers were rarely MSI (15.4%), and were often the CMS3 (40.4%) or CMS4 (30.8%) subtype. CIMP-L2 cancers mutate KRAS with relative infrequency when compared with CIMP-H2 and CIMP-L1 cancers (28.8%), and are significantly enriched for distal colonic and rectal locations (50% and 26.6%, for distal and rectal locations, respectively; *P* \< .0001). The proportion of CMS2 cancers was significantly higher in CIMP-L2 cancers when compared with CIMP-H1, CIMP-H2, and CIMP-L1 cancers (*P* \< .001). The frequency of distal colonic location was the highest among CIMP-neg cancers (62.3%) and were identified in patients with the youngest mean age (61.9 y). We did not identify a BRAF mutation in any CIMP-neg cancers. CMS2 and CMS4 were the most frequent CMS subtypes in CIMP-neg cancers (45.3% and 39.6%, respectively). The proportion of CMS4 was highest in CIMP-neg cancers when compared with other subtypes (*P* \< .001). We sequenced hotspots on exons 11 and 15 of *BRAF*, codon 61 in *KRAS*, and exon 18 in *EGFR* in CIMP-H1/H2 cancers that were wild-type at BRAF V600E and KRAS codons 12 and 13, however, we did not identify any mutations in these regions. Validation of the Association Between CIMP Subtype and Clinical and Molecular Features in The Cancer Genome Atlas {#sec1.3} ----------------------------------------------------------------------------------------------------------------- DNA methylation was previously measured using the HumanMethylation 450 array in 392 colorectal cancers from The Cancer Genome Atlas (TCGA) project.[@bib16] We observed several differences in the TCGA cohort when compared with the consecutive Royal Brisbane and Women\'s Hospital (RBWH) cohort. The mean age of patients at the time of diagnosis was significantly lower in the TCGA cohort when compared with the consecutive cohort (64.5 vs 67.9; *P* \< .01). Male sex was slightly over-represented (199 of 373; 53.4%). The distribution of cancers throughout the colon was significantly different in the TCGA cohort. Cancers in the TCGA were significantly enriched for proximal location in comparison with the RBWH cohort (47.0% vs 35.2%; *P* \< .01), and less likely to be located in the distal colon (40.3% vs 45.1%; *P* \< .01) or rectum (12.7% vs 19.7%; *P* \< .01). There were many similarities between the TCGA and RBWH cohorts. The frequency of *BRAF* mutations was 9.4%, and was not significantly different from the proportion observed in the RBWH cohort. Likewise, there was no significant difference in the frequency of *KRAS* mutations between the cohorts (40.1% vs 34.7%, for TCGA and RBWH cohorts, respectively). The proportion of microsatellite unstable cancers was not significantly different between the 2 cohorts (15.9% vs 12%; *P* = .1). Despite underlying differences in the clinical and molecular features of the cohorts, unsupervised clustering using the same methods as was used in the RBWH cohorts also resulted in the 5 distinct CIMP clusters identified in the TCGA series ([Table 2](#tbl2){ref-type="table"}, [Figure 1](#fig1){ref-type="fig"}). There was a similar, striking association between CIMP subtype and biological age (*P* \< .0001). In keeping with the RBWH cohort, increasing CIMP in the TCGA cohort was associated with proximal colonic location (*P* \< .0001), and was correlated inversely with distal and rectal locations (*P* \< .0001 and *P* \< .05, for distal and rectal locations, respectively). The distribution of *KRAS* mutations in CIMP subtypes followed a similar bell-shaped distribution, and were most common in CIMP-L1 cancers (48 of 81; 59.3%), and least common in CIMP-H1 (5 of 22; 26.3%) and CIMP-negative cancers (21 of 102; 20.6%). Notably, *KRAS* mutation was more common in CIMP-H2 cancers when compared with CIMP-H1 cancers in the TCGA cohort (43.6% vs 26.3%).Table 2Clinicopathologic and Molecular Details of 374 Colorectal Adenocarcinomas From TCGA Stratified for CIMP SubtypenCIMP-H1CIMP-H2CIMP-L1CIMP-L2CIMP-neg*P* valueTotal, n37419 (5.1%)39(10.4%)81 (21.7%)133 (35.6%)102 (27.3%)Mean age, *y*64.572.267.866.564.557.1\<.0001Sex Male1997 (36.8%)21 (53.8%)47 (58.0%)74 (55.6%)50 (49.5%)NS Female17412 (63.2%)18 (46.2%)34 (42.0%)59 (44.4%)51 (50.5%)Site Proximal16717 (100%)28 (84.8%)53 (67.9%)53 (40.8%)16 (16.5%)\<.0001 Distal143f04 (12.1%)18 (23.1%)57 (43.8%)64 (65.9%) Rectal4501 (3.0%)7 (9.0%)20 (15.4%)17 (17.5%)Mutation BRAF3510 (52.6%)19 (48.7%)5 (6.2%)1 (0.8%)0\<.0001 KRAS1505 (26.3%)17 (43.6%)48 (59.3%)59 (44.4%)21 (20.6%)\<.0001 TP5323410 (52.6%)19 (48.7%)44 (54.3%)85 (63.9%)76 (74.5%).01Microsatellite instability MSI5110 (52.6%)17 (50%)11 (16.7%)7 (6.2%)6 (6.7%)\<.0001 MSS2699 (47.4%)17 (50%)55 (83.3%)105 (93.8%)83 (93.3%)CMS CMS14210 (58.8%)20 (69%)9 (14.3%)3 (2.8%)0 (0%)\<.0001 CMS21212 (11.8%)1 (3.4%)25 (39.7%)48 (45.3%)45 (51.1%) CMS3454 (23.5%)4 (13.8%)16 (25.4%)14 (13.2%)7 (8%) CMS4951 (5.9%)4 (13.8%)13 (20.6%)41 (38.7%)36 (40.9%)Stage I543 (15%)9 (23.7%)16 (20.8%)11 (8.7%)15 (16%)\<.01 II1339 (45%)18 (47.4%)32 (41.6%)50 (39.4%)24 (25.5%) III1195 (25%)11 (28.9%)20 (26%)46 (36.2%)37 (39.4%) IV503 (15%)0 (0%)9 (11.7%)20 (15.7%)18 (19.1%)[^3][^4] In both cohorts, CMS2 cancers were most frequent in CIMP-L2 (TCGA, 45.3%; RBWH, 45.5%) and CIMP-negative (TCGA, 51.1%; RBWH, 45.3%). Likewise, CIMP-neg cancers were strongly enriched for the CMS4 subtype in both cohorts (TCGA, 40.9%; RBWH, 39.6%) In contrast to the RBWH cohort, CIMP-H1 cancers were less frequent overall (TCGA, 5.1%; RBWH, 10.6%) and *BRAF* mutation was associated with CIMP-H1 and CIMP-H2 (CIMP-H1: TCGA, 52.6%; RBWH, 73.9%; CIMP-H2: TCGA, 48.7%; RBWH, 9.1%). Perhaps as a consequence of the increased frequency of *BRAF* mutations in TCGA CIMP-H2 cancers, MSI was significantly more enriched in CIMP-H2 cancers in the TCGA cohort (50%). Although we did not identify any association between stage and CIMP subtype in the RBWH cohort, late-stage disease was associated significantly with decreasing CIMP in the TCGA cohort (stage IV: CIMP-H1, 15%; CIMP-H2, 0%; CIMP-L1, 11.7%; CIMP-L2, 15.7%; and CIMP-neg, 19.1%; *P* \< .01). The Colorectal Cancer Methylome Is Altered in Comparison With Normal Mucosa {#sec1.4} --------------------------------------------------------------------------- We identified differentially methylated probes in each cluster compared with 32 normal mucosal samples that matched a subset of cancers in the unselected series ([Table 3](#tbl3){ref-type="table"}). In all 4 CIMP clusters (CIMP-H1, -H2, -L1, and -L2), the number of differentially hypermethylated CpG sites greatly exceeded those that were hypomethylated ([Table 3](#tbl3){ref-type="table"}). By contrast, in the single CIMP-negative cluster, hypomethylation was more common than hypermethylation. Probe hypermethylation was most frequent in the CIMP-H1 cluster, including 21,168 hypermethylated probes occurring within 5165 unique CpG islands. Of these, 4333 also were hypermethylated in CIMP-H2, whereas 832 were uniquely hypermethylated in CIMP-H1. An additional 523 CpG islands were uniquely hypermethylated in the CIMP-H2 cluster relative to CIMP-H1. The highest number of hypomethylation events was seen in the CIMP-H2 cluster compared with all other clusters (*P* \< .0001), with the majority occurring in open sea regions of the genome.Table 3Distribution of Differentially Hypermethylated Probes in Reference to CpG Islands Vs Normal Mucosal TissueCpG locationCIMP-H1CIMP-H2CIMP-L1CIMP-L2CIMP-neg+-+-+-+-+-Island21,01120419,65142611,2971185685127754162South Shore319658630031359125342651328478242North Shore4745890464118852095617911420184346South Shelf2297431811620835744933119238North Shelf2807382591660925915834235246Sea20568396172115,575647681229741891043428Total31,51711,55729,45322,52515,46791387513569311744662[^5][^6] Next, we examined the impact of our chosen β value change threshold on the number of differential methylation events we were able to detect. Shifting the β value change threshold to 0.3 substantially reduced the number of differentially methylated probes identified (to 47.1%, 47.8%, 24.9%, 13.4%, and 5.8% of the probes identified at 0.2 for CIMP-H1 to CIMP-neg, respectively). When we increased the threshold to 0.4 we saw a similar, and more drastic, reduction in our ability to identify differentially methylated probes (DMPs) (18.9%, 19.5%, 4.1%, 1.2%, 0.3% of probes identified at 0.2 for CIMP-H1 to CIMP-Neg, respectively). There was a significant relationship between CIMP subtype and the magnitude of the DMPs identified (*P* \< .0001). We compared the probes that were differentially hypermethylated (vs normal mucosa) in the RBWH cohort with those differentially hypermethylated in the TCGA cohort. There was a remarkable degree of overlap in differentially methylated loci. In CIMP-H1, 80.2% of differentially hypermethylated loci were detected in both the RBWH and TCGA cohorts. Of the remaining 7481 probes, 6009 were detected solely in the TCGA and 1472 in the RBWH cohorts. We hypothesized that the β cut-off value (\>0.2 mean β value difference vs normal) may have resulted in the filtering of many of the probes that were detected in 1 cohort only. Indeed, of the 7481 DMPs detected in 1 cohort only, the methylation level of 98.5% was statistically significantly different from normal colonic mucosa in the other cohort, but were filtered as a result of the difference in the β cut-off value. This was consistent across all CIMP subtypes. The events that were recognized in 2 independent cohorts are likely to be bona fide differential methylation events. These data indicated that the selection of an appropriate difference in the β cut-off value is critical and that applying stringent cut-off values may significantly increase the type II error rate when reporting differentially methylated events. CIMP Subtypes Are Associated With Different Stromal Immune Cell Composition {#sec1.5} --------------------------------------------------------------------------- We hypothesized that CIMP subtypes may differ in their stromal cell type composition. We used CIBERSORT to deconvolute the relative composition of immune cells in the tumor microenvironment.[@bib17] CIMP-H1 cancers were enriched for M1 macrophages in comparison with all other CIMP subtypes, with the exception of CIMP-L2 cancers (*P* \< .01 vs CIMP-H2, *P* = .02 vs CIMP-L1, and *P* = .01 vs CIMP-neg). CIMP-H2 cancers were enriched for resting CD4 T memory cells (*P* \< .01), and were depleted for M1 macrophages (*P* = .01). Mast cells were associated inversely with DNA methylation subtype, with mast cells contributing least to the immune microenvironment in CIMP-H1 cancers and increasing in a stepwise manner from CIMP-H1 to CIMP-neg (*P* = .01). Conversely, natural killer cells were associated with CIMP-H cancers (analysis of variance, *P* \< .05), but did not differ between CIMP-H1 and CIMP-H2. CIMP-H1 and CIMP-H2 Cancers Can Be Delineated by Expression Profiles {#sec1.6} -------------------------------------------------------------------- To examine the extent to which CIMP-H1 and CIMP-H2 are transcriptionally distinct, we analyzed differential expression for each cluster with respect to normal mucosa using Illumina HT-12 expression arrays. We then performed single-sample gene set enrichment analysis[@bib18] to evaluate enrichments in the Hallmark gene set[@bib19] in individual samples (false-discovery rate \[FDR\] corrected, *P* \< .05). We identified 10 gene sets significantly enriched in CIMP-H1 cancers, 7 of which were related to the immune response ([Figure 2](#fig2){ref-type="fig"}). The bile acid metabolism gene set was significantly enriched in CIMP-H2 cancers. In TCGA we did not identify any significant differences in immune response or bile acid metabolism. This may be owing to the increased frequency of *BRAF* mutant MSI cancers in CIMP-H2 cancers in TCGA.Figure 2Differentially regulated hallmark gene sets between CIMP-H1 and CIMP-H2 cancers as assessed by single-sample gene set enrichment analysis. IL, interleukin; ssGSEA, single sample gene set enrichment analysis. Relationship Between Promoter Hypermethylation and Gene Transcriptional Activity {#sec1.7} -------------------------------------------------------------------------------- To determine the frequency of which DNA hypermethylation in promoter regions controls transcription of downstream genes, we examined the transcript levels for genes where the promoter was hypermethylated relative normal mucosa. Although promoter methylation was most common in CIMP-H1 and CIMP-H2 clusters ([Figure 3](#fig3){ref-type="fig"}*A*), these subgroups had the lowest proportion of genes in which hypermethylation correlated with reduced transcript expression (13.9% and 15.6%, respectively). This inverse relationship continued for CIMP-L1 (18.9%), CIMP-L2 (19.9%), and with the CIMP-negative cancers, with reduced transcription in 22.7% of hypermethylated promoters (*P* \< .0001) ([Figure 3](#fig3){ref-type="fig"}*B*). We observed a similar relationship between gene transcription and promoter methylation in cancers in TCGA. In TCGA, the proportion of methylated genes that resulted in gene transcription repression did not differ between CIMP subtypes.Figure 3(*A*) Number of differentially methylated promoters in each CIMP cluster vs the cohort of normal mucosal samples. (*B*) The proportion of methylation events within each cluster that resulted in gene repression at the transcript level. We considered that loci that were methylated and repressed in multiple CIMP clusters may be genes that are important for cancer development. Strikingly, of the 1273 genes that were methylated and repressed in at least 1 CIMP cluster, 82.3% were methylated and repressed in 2 or more CIMP clusters, 16.9% silenced in 3 or more CIMP subtypes, and 8.0% in all 4 CIMP subtypes (excluding CIMP-negative). We identified 21 tumor-suppressor genes, as per the Network of Cancer Genes (NCG)6.0 database, that were recurrently methylated and silenced in 3 or more CIMP subtypes ([Table 4](#tbl4){ref-type="table"}).Table 4Tumor-Suppressor Genes That Were Recurrently Methylated and Repressed in More Than 3 CIMP SubtypesGene nameDescription*PCDH9*Protocadherin 9 (source: HGNC symbol; Acc: HGNC: 8661)*CDO1*Cysteine dioxygenase type 1 (source: HGNC symbol; Acc: HGNC: 1795)*MAL*Mal, T-cell differentiation protein (source: HGNC symbol; Acc: HGNC: 6817)*EPB41L3*Erythrocyte membrane protein band 4.1-like 3 (source: HGNC symbol; Acc: HGNC: 3380)*AKAP12*A-kinase anchoring protein 12 (source: HGNC symbol; Acc: HGNC: 370)*NDRG4*NDRG family member 4 (source: HGNC symbol; Acc: HGNC: 14466)*LIFR*LIF-receptor α (source: HGNC symbol; Acc: HGNC: 6597)*SCUBE2*Signal peptide, CUB domain, and EGF-like domain containing 2 (source: HGNC symbol; Acc: HGNC: 30425)*TMEFF2*Transmembrane protein with EGF-like and 2 follistatin-like domains 2 (source: HGNC symbol; Acc: HGNC: 11867)*DUSP26*Dual-specificity phosphatase 26 (source: HGNC symbol; Acc: HGNC: 28161)*C2orf40*Chromosome 2 open reading frame 40 (source: HGNC symbol; Acc: HGNC: 24642)*SFRP1*Secreted frizzled-related protein 1 (source: HGNC symbol; Acc: HGNC: 10776)*UCHL1*Ubiquitin C-terminal hydrolase L1 (source: HGNC symbol; Acc: HGNC: 12513)*IKZF1*IKAROS family zinc finger 1 (source: HGNC symbol; Acc: HGNC: 13176)*CADM2*Cell adhesion molecule 2 (source: HGNC symbol; Acc: HGNC: 29849)*CXCL12*C-X-C motif chemokine ligand 12 (source: HGNC symbol; Acc: HGNC: 10672)*IRF4*Interferon regulatory factor 4 (source: HGNC symbol; Acc: HGNC: 6119)*ZBTB16*Zinc finger and BTB domain containing 16 (source: HGNC symbol; Acc: HGNC: 12930)*CHFR*Checkpoint with forkhead and ring finger domains (source: HGNC symbol; Acc: HGNC: 20455)*SLIT2*Slit guidance ligand 2 (source: HGNC symbol; Acc: HGNC: 11086)*ZFP82*ZFP82 zinc finger protein (source: HGNC symbol; Acc: HGNC: 28682)[^7] Polycomb-Repressive Complex 2 Occupancy at Hypermethylated CpGs Is Correlated Inversely With Global Hypermethylation {#sec1.8} -------------------------------------------------------------------------------------------------------------------- Supressor Of Zeste 12 (SUZ12) occupancy is a surrogate for polycomb-repressor complex 2 occupancy and in embryonic stem cells this has been shown to associate with transcriptional repression of hypermethylated loci.[@bib6], [@bib20] Consistent with this, we observed an increase in the number of methylated CpG sites that overlap with SUZ12-occupied regions with increasing CIMP cluster (*P* \< .0001) ([Figure 4](#fig4){ref-type="fig"}*A*). Conversely, and in keeping with our findings with promoter methylation, an inverse association between the proportion of hypermethylated loci genes that overlapped with SUZ12-occupied sites with increasing CIMP cluster was observed (*P* \< .0001) ([Figure 4](#fig4){ref-type="fig"}*B*). This further supports our finding that although DNA hypermethylation occurs more frequently with increasing CIMP cluster, these methylation events are more likely to result in gene silencing in CIMP-negative cancers.Figure 4(*A*) Proportion of SUZ12-occupied regions in hESC1 cells that contained hypermethylated probes in respective CIMP clusters. (*B*) Proportion of differential hypermethylation events that overlapped with Polycomb Repressive Complex-2 (PRC2)-occupied regions. CIMP-H1 and CIMP-H2 Promoter Methylation Is Defined by the Enrichment of Distinct Transcription Factor Binding Sites {#sec1.9} -------------------------------------------------------------------------------------------------------------------- Transcription factor binding sites often contain CpG sequences and therefore are a target of DNA methylation, which may explain some of the effects of methylation on transcription. To explore whether DNA methylation is targeted to specific transcription factor binding sites we performed an enrichment analysis using the CentriMo[@bib21] tool to examine the 2-kb region immediately upstream of hypermethylated genes. There were 128 significantly enriched binding sites that overlapped in CIMP-H1 and CIMP-H2 cancers. An additional 323 sites were uniquely enriched in CIMP-H1 cancers and an additional 330 sites in CIMP-H2 cancers. SMAD4 and FOXP3 (adjusted *P* values = 1.2 × 10^-24^ and 4.1 × 10^-23^, respectively) were the most significantly enriched motifs in CIMP-H1 cancers. SPDEF, FLI1, and NKX6 (adjusted *P* values = 7.2 × 10^-30^, 1.1 × 10^-16^, and 3.5 × 10^-16^, respectively) were most significantly enriched in CIMP-H2 cancers. [Table 5](#tbl5){ref-type="table"} presents the top 10 enriched consensus binding sites that were exclusive to CIMP-H1 and CIMP-H2.Table 5Motifs That Were Most Significantly and Exclusively Enriched at Methylated Promoters in CIMP-H1 and CIMP-H2CIMP-H1CIMP-H2Motif nameMotifRaw *P* valueAdjusted *P* valueMotif nameMotifRaw *P* valueAdjusted *P* valueSmad4TGTCTRGM1.2E-211.2E-24SPDEF_DBD_2GTGGTCCCGGATTAT7.2E-337.2E-30FOXP3_DBDRTAAACA4.1E-204.1E-23UP00142_1VNTAATTAATTAABGSG2.4E-202.4E-17FOXP3RTAAACA4.1E-204.1E-23FLI1_full_2ACCGGAAATCCGGT1.1E-191.1E-16POU2F2_DBD_2HWTRMATATKCAWA4.5E-194.5E-22UP00200_1GVWAATTAATTAMYBBG3.5E-193.5E-16Zscan4_primaryDHNATGTGCACAYAHWN1.2E-181.3E-21NHLH1_DBDCGCAGCTGCS2.1E-182.1E-15HOXC10GTCRTAAAAH1.3E-181.3E-21ERG_full_2ACCGGAWATCCGGT4.8E-184.8E-15Bbx_secondaryHVWNNGTTAACASHNRV3.1E-163.1E-19MA0680.1TAATCGATTA8.7E-188.6E-15Foxc1_DBD_1GTAAAYAAACA1.3E-151.3E-18PAX7_DBDTAATYRATTA1.4E-161.4E-13 Gene Bodies of Wnt Pathway Antagonists Are Resistant to Methylation {#sec1.10} ------------------------------------------------------------------- We further explored gene bodies that were unmethylated but had more than 10 CpG island probes, and performed pathway analysis to identify pathways that were devoid of gene body methylation. There were 6 pathways that were significantly enriched among these genes, including the WNT signaling pathway ([Figure 5](#fig5){ref-type="fig"}). The WNT signaling pathway was most heavily enriched. *PCDHA6*, *PCDHGA2*, *PCDHA7*, and *PCDHA2* contained 36, 15, 10, and 20 gene body CpG island probes, respectively, which were all unmethylated. These protocadherins have been implicated in the regulation of the WNT signal and may act as a tumor-suppressor gene. Likewise, *AXIN1*, a gene critical to the β-catenin destruction complex, contained 11 unmethylated intragenic CpG Island (CGI) probes. *TCF3*, a WNT pathway repressor, contained 19 unmethylated intragenic CGI probes. We considered whether gene body methylation within WNT antagonists could alter gene transcription, however, we did not observe any differences in expression profiles of these genes vs normal mucosa tissue, and they were not expressed in normal mucosa tissue. In the remaining WNT genes we did not identify any consistent expression changes.Figure 5Pathways significantly enriched for genes that contained CpG islands that were devoid of methylation in both CIMP-H clusters. VEGF, vascular endothelial growth factor. Oncogenes Are Significantly More Likely Than Tumor-Suppressor Genes to Undergo Gene Body Methylation in CIMP-H1 and CIMP-H2 Cancers {#sec1.11} ----------------------------------------------------------------------------------------------------------------------------------- Gene body methylation is correlated positively with gene expression.[@bib22] We examined hypermethylation in gene body CpG islands, defined as a minimum of 2 probes in the CpG island as hypermethylated relative to normal (*P* \< .01) and there was a mean absolute difference in β values vs normal of greater than 0.2 to evaluate whether gene body methylation was a phenomena enriched in oncogenes of CIMP-H--type cancers, or was driven more nonspecifically by CIMP itself. In total, 239 genes were annotated as known oncogenes, and 239 as known tumor-suppressor genes in the NCG6.0 cancer gene database.[@bib23] Of these, 121 tumor suppressors and 116 oncogenes had a CpG island within the gene body that was probed on the array. In CIMP-H1 cancers, 21.5% (20.2% in TCGA) of oncogenes had significant gene body methylation in reference to normal, by comparison, significantly fewer tumor-suppressor genes underwent gene body methylation (12.4% in the RBWH cohort, *P* \< .05; 8.1% in TCGA; *P* \< .001). Likewise, gene body methylation was significantly more likely to occur in oncogenes than tumor-suppressor genes in CIMP-H2 cancers (23.3% vs 11.6%; *P* = .01). The gene expression of 5 oncogenes in CIMP-H1 and CIMP-H2 differed significantly from normal mucosa (FEV, BCL2, and KIT were down-regulated and PAX3 and SND1 were up-regulated in CIMP-H1; LMO2 and CTNND2 were down-regulated and SND1, CNTTA2, and TLX1 were up-regulated in CIMP-H2). [Table 6](#tbl6){ref-type="table"} presents the oncogenes that had significantly higher gene body methylation in CIMP-H1 and CIMP-H2 cancers compared with normal colonic mucosa.Table 6Oncogenes With Significantly Higher Methylation Within the Body of the GeneCIMP-H1CIMP-H2GeneExpressionDescriptionGeneExpressionDescription*FEV*Down-regulatedFEV, ETS transcription factor*LMO2*Down-regulatedLIM domain only 2*BCL2*Down-regulatedBCL2, apoptosis regulator*CTNND2*Down-regulatedCatenin Δ 2*KIT*Down-regulatedKIT proto-oncogene receptor tyrosine kinase*SND1*Up-regulatedStaphylococcal nuclease and tudor domain containing 1*PAX3*Up-regulatedPaired box 3*CTNNA2*Up-regulatedCatenin α2*SND1*Up-regulatedStaphylococcal nuclease and tudor domain containing 1*TLX1*Up-regulatedT-cell leukemia homeobox 1*LMO2*No differenceLIM domain only 2*PREX2*No differencePI-3,4,5-trisphosphate-dependent Rac exchange factor 2*RSPO3*No differenceR-spondin 3*RSPO3*No differenceR-spondin 3*CTNND2*No differenceCatenin delta 2*RET*No differenceRet proto-oncogene*TLX3*No differenceT-cell leukemia homeobox 3*LMO1*No differenceLIM domain only 1*SIX1*No differenceSIX homeobox 1*FLT3*No differenceFms-related tyrosine kinase 3*HOXC13*No differenceHomeobox C13*CACNA1D*No differenceCalcium voltage-gated channel subunit α1 D*LMO1*No differenceLIM domain only 1*WWTR1*No differenceWW domain containing transcription regulator 1*ZNF521*No differenceZinc finger protein 521*CHST11*No differenceCarbohydrate sulfotransferase 11*SALL4*No differenceSpalt like transcription factor 4*PAX3*No differencePaired box 3*ZEB1*No differenceZinc finger E-box binding homeobox 1*FLT4*No differenceFms-related tyrosine kinase 4*PREX2*No differencePI-3,4,5-trisphosphate dependent Rac exchange factor 2*CXCR4*No differenceC-X-C motif chemokine receptor 4*OLIG2*No differenceOligodendrocyte transcription factor 2*TLX3*No differenceT-cell leukemia homeobox 3*SMO*No differenceSmoothened, frizzled class receptor*TAL1*No differenceTAL bHLH transcription factor 1, erythroid differentiation factor*FLT3*No differenceFms related tyrosine kinase 3*SIX1*No differenceSIX homeobox 1*GATA2*No differenceGATA binding protein 2*HOXC11*No differenceHomeobox C11*TLX1*No differenceT-cell leukemia homeobox 1*OLIG2*No differenceOligodendrocyte transcription factor 2*TAL1*No differenceTAL bHLH transcription factor 1, erythroid differentiation factor*MYOD1*No differenceMyogenic differentiation 1*CACNA1D*No differenceCalcium voltage-gated channel subunit a1 D*ZEB1*No differenceZinc finger E-box binding homeobox 1*MYOD1*No differenceMyogenic differentiation 1*HOXC13*No differenceHomeobox C13*CTNNA2*No differenceCatenin a2*ZNF521*No differenceZinc finger protein 521*CHST11*No differenceCarbohydrate sulfotransferase 11*SMO*No differenceSmoothened, frizzled class receptor*NR4A3*No differenceNuclear receptor subfamily 4 group A member 3*GATA2*No differenceGATA binding protein 2*NR4A3*No differenceNuclear receptor subfamily 4 group A member 3[^8] Loci Marked by the PRC2 Complex in Human Embryonic Stem Cells Are Prone to Gene Body Methylation During Cancer Development {#sec1.11a} -------------------------------------------------------------------------------------------------------------------------- Polycomb Repressive Complex-2 (PRC2) marking in human embryonic stem cells has been shown previously to overlap significantly with promoter hypermethylation in colorectal cancers.[@bib6] We hypothesized that a similar phenomenon would occur with regard to gene body hypermethylation. In CIMP-H1 and CIMP-H2 cancers, 30.59% and 31.04%, respectively, of loci marked with H3K27me3 in human embryonic stem cells developed significant gene body hypermethylation ([Table 7](#tbl7){ref-type="table"}) (*P* = 1.34 × 10^-280^ for CIMP-H1 and *P* = 2.5 × 10^-300^ for CIMP-H2 overlap). We observed a lesser, but still highly significant, overlap between H3K27me3 marked loci and gene body methylation in CIMP-L1 (13.1%; *P* = 6.11 × 10^-122^) and CIMP-L2 (8.5%; *P* = 1.6 × 10^-78^) cancers, but did not observe any correlation in CIMP-neg cancers, which likely is owing to the scarcity of which gene body methylation occurs in these cancers. We observed similar overlaps for embryonic ectoderm development (EED) targets, SUZ12 targets, and PRC2 targets.Table 7Overlap Between Genes Marked by the PRC2 Complex and H3K27Me3 in hEScells and Genes That Undergo Significant Gene Body Methylation in Colorectal Cancer DevelopmentGene set nameCIMP-H1CIMP-H2CIMP-L1CIMP-L2Overlap fractionFDR *P* valueOverlap fractionFDR *P* valueOverlap fractionFDR *P* valueOverlap fractionFDR *P* valueBENPORATH_ES_WITH_H3K27ME330.59%1.34E-28031.04%2.50E-30013.06%6.11E-1228.50%1.60E-78BENPORATH_EED_TARGETS30.70%3.91E-26731.07%1.12E-28412.81%8.75E-1128.66%8.47E-77BENPORATH_SUZ12_TARGETS30.92%5.05E-26430.73%9.67E-27312.91%1.29E-1108.48%2.02E-72BENPORATH_PRC2_TARGETS37.27%1.04E-21838.04%8.59E-23516.41%4.56E-9811.04%2.05E-66[^9] Epigenetic Regulator Gene Mutations Are Common in TCGA Cancers {#sec1.12} -------------------------------------------------------------- Mutations in epigenetic modifier genes have been shown previously to modulate transcriptional profiles in cancer.[@bib15] We assessed the mutational frequency of 719 epigenetic regulator genes in cancers from the TCGA colon adenocarcinoma and rectal adenocarcinoma projects using the CIMP subtypes identified earlier. For these analyses we included only mutations that were truncating in nature (nonsense or indels), were predicted to alter splicing, or were predicted to have a deleterious effect by PolyPhen.[@bib24] Overall, 92.8% of cancers had a deleterious mutation in an epigenetic regulator gene (347 of 374). There were 94.7% and 100% of CIMP-H1 and CIMP-H2 cancers that had at least 1 mutation in an epigenetic regulator. The proportion of CIMP-L1, CIMP-L2, and CIMP-negative cancers with deleterious mutations in these genes was slightly lower (93.8%, 89.5%, and 93.1%, respectively), however, these proportions were not significantly different from CIMP-H1 or CIMP-H2. Of the 719 genes we investigated, 95.7% were mutated in at least 1 cancer (688 of 719). [Figure 6](#fig6){ref-type="fig"} shows the most commonly mutated epigenetic regulators in each cluster. Mutations were least common in cancers classified as CIMP-neg, with increasing global methylation being associated with a concordant increase in epigenetic mutational load. However, when we examined epigenetic mutation frequency in relation to microsatellite instability, there was no significant relationship between CIMP cluster and epigenetic mutation frequency, indicating that the differences observed between CIMP clusters may be driven by the increasing frequency of microsatellite instability in CIMP clusters with higher genomic methylation.Figure 6High-impact mutations in epigenetic regulator genes are frequent in cancers with higher genomic methylation. Del, deletion; Ins, insertion. CIMP-H1 and H2 Subtypes Have Similar Mutational Patterns in Epigenetic Regulator Genes {#sec1.13} -------------------------------------------------------------------------------------- We examined the top 25 mutated epigenetic regulator genes in CIMP-H1 and CIMP-H2 to identify mutational targets that are common to CIMP-H and those that are exclusive to either the CIMP-H1 or CIMP-H2 subtypes. This was not influenced by MSI, which was equally represented in these cancer subtypes (53% CIMP-H1, 50% CIMP-H2). A total of 31.6% of these genes were identifiable in the top 25 epigenetic mutational targets in both CIMP-H1 and CIMP-H2. Such genes included 4 histone lysine methyltransferases (*SETD1B*, *KMT2A*, *KMT2B*, and *KMT2D*), the SWItch/Sucrose Non-Fermentable (SWI/SNF) complex member *ARID1A*, and the chromohelicase domain gene *CHD7*. Thirteen genes were identified in the top 25 mutated epigenetic regulators in CIMP-H1, but not CIMP-H2, these included the DNA demethylases *TET1* (mutated in 15.8% of CIMP-H1 cancers vs 10.3% of CIMP-H2 cancers) and *TET3* (mutated in 26.3% of CIMP-H1 cancers vs 10.3% of CIMP-H2 cancers). Mutations in histone lysine demethylase *KDM2B* were enriched in CIMP-H1 cancers (mutated in 36.8% of CIMP-H1 cancers vs 7.7% of CIMP-H2 cancers; *P* = .01). In contrast, 13 genes were found in the top 25 mutated epigenetic regulators of CIMP-H2 but not CIMP-H1. The *NCOR1* transcription factor was mutated in 20.5% of CIMP-H2 cancers compared with 5.3% of CIMP-H1 cancers, and the cohesin complex subunit *NIPBL* in 15.4% of CIMP-H2 cancers, despite not being identified as mutated in any CIMP-H1 cancer. Epigenetic Regulator Gene Mutation Exclusivity Supports the Dichotomization of CIMP-L Clusters {#sec1.14} ---------------------------------------------------------------------------------------------- We used a similar approach (top 25 epigenetic gene mutations) to investigate whether CIMP-L1 and CIMP-L2 subtype cancers also target similar epigenetic regulators for somatic mutation. Here, 11 epigenetic regulator genes were commonly mutated in both CIMP-L1 and CIMP-L2. The histone lysine methyltransferases *KMT2B* and *KMT2C* were among the top 25 mutated epigenetic regulators in both CIMP-L1 and CIMP-L2, however, the frequency of mutation in both *KMT2B* and *KMT2C* was lower in CIMP-L2 cancers (*KMT2B* CIMP-L1, 11.8%; CIMP-L2, 5.7%; *KMT2C* CIMP-L1, 10.5%; and CIMP-L2, 6.5%), but this was not statistically significant. There was a nonsignificant trend (*P* = .06) for increased *ASH1L* mutation in CIMP-L1 cancers (13.2%) vs CIMP-L2 cancers (4.9%). Fourteen genes were in the top 25 mutated epigenetic regulators of CIMP-L1 or CIMP-L2 alone. *SETD1B*, a histone lysine methyltransferase identified as a commonly mutated gene in CIMP-H cancers was mutated in 6 CIMP-L1 cancers, but was only mutated in a single CIMP-L2 cancer (*P* \< .01). Likewise, we identified recurrent *ARID1A* mutations in CIMP-L1 (9.2%), however, we identified significantly fewer in CIMP-L2 cancers (1.6%; *P* \< .01). The SWI/SNF Complex Is a Commonly Aberrantly Mutated Chromatin Remodeling Complex in CIMP-H1, CIMP-H2, and CIMP-L1 Cancers {#sec1.15} -------------------------------------------------------------------------------------------------------------------------- Next, we examined the SWI/SNF complex (*MARCA2*, *ARID1A*, *ARID1B*, *ARID2*, *PBRM1*, *SMARCB1*, and *SMARCA4*) for high-impact somatic mutations. Mutations in any of the SWI/SNF subunits occurred in 19.06% of cancers. An *ARID1A* mutation was the most frequent genetic alteration of the complex (6.7%). We observed a number of recurrently mutated positions in *ARID1A*, including 6 frameshift deletions at codon 2141, 4 deletions at codon 1850, and 3 deletions at codon 1072 ([Figure 7](#fig7){ref-type="fig"}). *ARID2* was mutated in 6% of cancers, but unlike *ARID1A* we did not identify any recurrently mutated positions. The distribution of the mutations between CIMP subtypes was significantly skewed toward subtypes with higher overall methylation (*P* \< .0001). SWI/SNF mutations were observed in 50% of CIMP-H1 cancers, and 38.5% of CIMP-H2 cancers. A total of 26.3% of CIMP-L1 samples mutated a SWI/SNF member, and in contrast to CIMP-H1 and CIMP-H2, the most frequently mutated member of the complex was *SMARCA4* (11%). The R885C mutation was observed in 3 cancers in CIMP-L1. Mutations in SWI/SNF subunits were similarly infrequent and significantly less prevalent in CIMP-L1 and CIMP-neg (10.6% and 11.6%, respectively; *P* \< .0001).Figure 7High impact mutations in ARID1A are common in colorectal adenocarcinomas. Del, deletion; Ins, insertion. Synthetic lethality in the SWI/SNF complex was established previously.[@bib25] CIMP-H1, CIMP-H2, and CIMP-L1 cancers may be more vulnerable to treatments targeting the other element of the SWI/SNF complex. To test whether 1 SWI/SNF mutation confers dependency on other SWI/SNF subunits in vitro, we correlated exome capture data from 15 cell lines[@bib26] with cell line--dependency data from Meyers et al.[@bib27] Five cell lines had an *ARID1A* truncating mutation and these were significantly more dependent on *ARID1B* expression for survival (0.31 vs 0.06; *P* \< .05). The Frequency of Genetic Perturbation of Chromodomain Helicase DNA Binding Genes Is Associated With DNA Methylation {#sec1.16} ------------------------------------------------------------------------------------------------------------------- CHD genes are members of another chromatin remodeling family. High-impact CHD family gene mutations were present in 22.4% of colorectal cancers in the TCGA. CHD mutations were markedly more common in CIMP-H1 and CIMP-H2 cancers. Family members were mutated in 50% and 51.3% of CIMP-H1 and CIMP-H2 cancers, respectively. *CHD7* was the most frequently altered gene in CIMP-H1 (33% of cancers), and *CHD8* in CIMP-L2 (22%). CHD mutations were less common, but still frequent, in CIMP-L1 cancers (19.7%). In these cancers, *CHD4* was the most commonly mutated gene (8%). The frequency of CHD mutations continued to decline as concordant with DNA methylation. The frequency of CHD mutations in CIMP-L2 was 11.7%, and was lower than the frequency observed in CIMP-neg cancers (15%). We examined the CHD genes for recurrently mutated positions. At the *CHD7* locus, which was mutated in 5.5% of cancers, we observed 5 frameshift deletions (D2988fs del 3) at the 3' end of the gene. This mutation has been observed in a number of colorectal cancer cell lines. For *CHD3*, *CHD4*, and *CHD9* we observed 3 recurrently mutated positions at R540fs del 16, R975H, and F760fs del 16. Discussion {#sec2} ========== Remodeling of the epigenome is fundamental to colorectal cancer progression. One of the most common epigenetic phenomena altered throughout carcinogenesis is the DNA methylation landscape. Here, we aimed to better understand the extent and heterogeneity of aberrant DNA methylation in colorectal cancers, and characterize the interplay between DNA methylation, somatic variation in epigenetic regulator genes, and gene transcription. Through the genome-scale interrogation of the largest unselected and consecutive series of colorectal cancers to date, we identified 5 clinically and molecularly distinct DNA methylation subtypes. The 5 subtypes identified in this study are highly correlated with key clinical and molecular features, including patient age, tumor location, microsatellite instability, and oncogenic mitogen-activated protein kinase mutations. We show that cancers with high DNA methylation show an increased preponderance for mutating genes involved in epigenetic regulation, and namely those that are implicated in the chromatin remodeling process. Hinoue et al[@bib6] previously reported the presence of 4 colorectal cancer methylation subgroups by assessing 125 colorectal cancers using Illumina 27K DNA methylation arrays. In the present study, we have considerably increased the power to assess subgroups based on differential methylation by studying 216 unselected cancers using the Illumina 450K DNA methylation platform. The Illumina 450K DNA methylation platform is capable of assessing more than 10 times more CpG sites and thus can identify methylation subtypes more robustly. A major difference of our study was the identification of 2 discrete CIMP-high subtypes: CIMP-H1 and CIMP-H2. The dichotomization of these CIMP-H cancers identified a homogeneous subgroup of CIMP-H1 cancers with an average age of 75 years, striking over-representation of female sex, and *BRAF* mutant cancers arising in the proximal colon. The newly identified CIMP-H2 subtype encompasses more *KRAS* mutant cancers than CIMP-H1, and the majority of cancers in this subtype would be CIMP-low using the 5-marker CIMP panel proposed by Weisenberger et al.[@bib3] Our genome-scale analyses of both our cohort and the TCGA indicate this is not the case. Together, our CIMP-H1/H2 clusters represent 21% of our unselected cohort, and 16.3% of the TCGA cohort. Collectively, the current findings indicate that CIMP is more prevalent than previously thought, and classification of cancers using existing panels may not identify all CIMP-high colorectal cancers. We observed a consistent increase in patient age with CIMP cluster, from 62 years in CIMP-neg cancers to 75 years in CIMP-H1 cancers. This is in contrast to the Hinoue et al[@bib6] study. The variance in our assay was mostly contained in uniquely mapping probes that were not present in the Illumina HumanMethylation27 BeadChip array used by Hinoue et al.[@bib6] Numerous studies have shown age-related methylation in different tissues[@bib9], [@bib28], [@bib29] and we previously identified hypermethylated loci in the colons of patients even with no history of colonic disease.[@bib9] In the present study, we detected a significant correlation between methylation and patient age. After removal of all probes that were significantly hypermethylated in normal mucosal tissue, we still observed distinct, age-linked clustering. This association was faithfully reproduced in cancers from TCGA. The subtype with the highest degree of methylation (CIMP-H1) was strongly associated with mutations in the *BRAF* oncogene. *BRAF* mutations are a hallmark of the serrated neoplasia pathway, and indicate that these cancers probably arose in serrated precursor lesions. We previously showed that the colonoscopic incidence of sessile serrated adenomas does not differ between patients aged in their 30s and patients who are much older, whereas *BRAF* mutant cancers were restricted to older individuals,[@bib30] suggesting these *BRAF* mutant polyps may have limited malignant potential in young patients. We also reported a striking association between patient age and CIMP in sessile serrated adenomas.[@bib31] Here, we report that the vast majority of *BRAF* mutant cancers in both the RBWH and TCGA cohorts are CIMP-H and arise in older individuals. Collectively, these findings suggest that sessile serrated adenomas may be relatively benign in young patients. In older patients with more advanced DNA methylation changes in the colon, the risk of progression to cancer will be significantly greater. Recently, we recapitulated this process in a murine model for serrated neoplasia and showed that early onset *Braf* mutation leads to the temporal accumulation of DNA methylation and ultimately to malignancy.[@bib32] Additional studies are necessary to fully determine the natural history of *BRAF* mutant cancers, and elucidate the determinants of malignant potential to inform the development of patient-centric surveillance for young and older patients who present with sessile serrated adenomas. Differential CpG island and shore hypermethylation were the most frequently observed methylation events in the study. Probes on the north and south CpG shelves, as well as those in the open seas, frequently were hypomethylated across most cancers. The implications of hypomethylated CpG dinucleotides outside of CpG islands are unclear. We did not observe any relationship between hypomethylation and gene transcription, however, it is possible that hypomethylation of specific regions of the genome may affect chromatin accessibility elsewhere and hence may modulate transcription in a trans-acting manner. Open sea hypomethylation was also the most frequent methylation event in CIMP-neg cancers. These are predominately conventional pathway cancers with a high degree of chromosomal instability. One hypothesis that may explain this association is that hypomethylation outside of CpG islands may predispose to copy number changes in these cancers.[@bib33], [@bib34] Functional studies are necessary to explore the implications of shelf and open sea hypomethylation and whether this is relevant to the cancer development process for these cancers. There were marked differences in transcriptional deregulation of key cancer-related pathways between methylation clusters. CIMP-H1 cancers activated several immune pathways, including those involved in the interferon response, inflammatory response, and complement signaling, consistent with the over-representation of CMS1 cancers in this group. This likely is owing to the higher mutational burden in these cancers, largely driven by the increased incidence of epigenetically induced microsatellite instability. MSI cancers have been associated with greater immune infiltrate and hence some of this signaling may originate in the stromal immune cells, rather than from within the tumor itself.[@bib35] In the RBWH cohort, CIMP-H2 cancers were uniquely enriched for altered bile acid metabolism, consistent with the previously described relationship between silencing of the farnesoid X bile acid receptor in *KRAS* mutant cancers.[@bib36] Bile acids are more concentrated in the proximal colon and metabolism is influenced by the gut microbiome.[@bib37] The increased bile acid metabolism signaling in this group of cancers may identify a subset of cancers that have arisen owing to aberrant bile acid accumulation. We did not observe such an effect in the TCGA cohort. This may be owing to the increased frequency of *BRAF* mutant MSI cancers in CIMP-H2 in TCGA. A better understanding of the role of bile acid signaling in *KRAS* mutant cancers of the proximal colon may have therapeutic implications for this cancer subgroup. Paradoxically, despite observing less differential methylation, we observed an increase in gene silencing that correlated with promoter hypermethylation in the least methylated cancer clusters. This may indicate that promoter hypermethylation in CIMP-L1/2 and CIMP-neg cancers is more specifically selected based on a functional advantage in these cancers. Alternatively, the increased frequency of mutations in epigenetic regulators of CIMP-H1/2 cancers may result in a reduced capacity to induce gene repression at certain loci. This may be owing to the loss of a repressive histone-modifying enzyme, or mutation of locus-specific repressive transcription factors. Methylation alone may be insufficient to induce gene repression in certain instances. Instead, relevant chromatin remodeling and histone modifications, such as the addition of the repressive PRC2 mark, may be required in tandem with methylation changes to reduce gene expression. Indeed, we showed that PRC2 occupancy was most frequently related to transcriptionally repressed and methylated genes in the CIMP-neg subgroup. We also observed instances of promoter methylation that correlated with increased gene transcription. It is possible that some transcription factors preferentially bind methylated DNA,[@bib38] and that binding sites for these transcription factors become available after promoter methylation. These data may indicate that the genomic context of methylation is important for determining whether gene expression changes will occur. In TCGA, however, we were unable to discern any significant differences in the proportion of methylated and repressed genes vs all methylated genes between CIMP subtypes. This may be owing to technological differences between the array-based methods used to evaluate gene transcription in the current study and the RNA sequencing-based methods used in TCGA. Direct comparisons between the expression values derived from each of these studies is difficult and should be approached with caution. A major novel finding of the current study was the discovery that gene body methylation may be a major driver of serrated tumorigenesis, and that this may be mediated by H3K27me3 histone marks. Gene body hypermethylation recently was correlated with increased oncogene expression.[@bib22] Here, we identified many well-characterized oncogenes, such as *BCL2* and *ZEB1*, with methylation of their gene bodies in CIMP-H1/2 cancers, and noted a significant preference for the methylation of gene bodies of oncogenes compared with tumor-suppressor genes. We also identified Wnt pathway antagonists that are resistant to gene body methylation. In the present study, we did not identify distinct transcriptional differences in these Wnt pathway antagonists. It is possible that gene body methylation affects other aspects of the transcriptional process that were not assessed in this study, such as splicing and isoform switching. Alternatively, this gene body methylation may be a stochastic result of the overall increase in aberrant DNA methylation in these cancers. The epigenome is regulated by proteins that interact with histones or DNA. We assessed the coding sequence of 719 epigenetic regulator genes in the TCGA data set. The chromodomain-helicase-DNA (CHD) binding protein family was a frequent mutational target in CIMP-H1 cancers. Recently, Fang et al[@bib39] showed that CHD8 operates in a transcriptional repression complex to direct methylation in the setting of *BRAF* mutation. In the current study we showed *BRAF* and *CHD8* mutations were associated with CIMP-H1. Thus, these data suggest that *CHD8* mutation may enhance repression complex activity in the setting of *BRAF* mutation, resulting in hypermethylation. Moreover, CHD8 has been associated with the CCCTC-binding factor (CTCF) protein, which is essential for promoter-enhancer looping and regional insulation. *CHD8* mutations may influence CIMP by decreasing the ability of CTCF to insulate regions of the genome, and could encourage methylation spreading throughout the genome.[@bib40] Similarly, we report frequent mutations in different members of the CHD family. *CHD7* was the most mutated CHD gene, and some positions in the *CHD7* locus were recurrently mutated. Tahara et al[@bib41] identified mutations in *CHD7* and *CHD8* in 42% of CIMP1 colorectal cancers. The functional consequences of *CHD7* mutations are unclear. In pancreatic duct adenocarcinoma, *CHD7* expression has been shown to correlate with gemcitabine sensitivity.[@bib42] The most commonly mutated CHD gene in CIMP-L1 cancers was *CHD4*. Recently, Xia et al[@bib43] in 2017 proposed an oncogenic role for CHD4 through facilitating the hypermethylation of tumor-suppressor genes. In contrast, Li et al[@bib44] in 2018 showed that *CHD4* mutations that promote protein degradation enhance stemness and contribute to the progression of endometrial cancers via the transforming growth factor-β signaling cascade. Indeed, we identified 3 mutations at the R975H hotspot of *CHD4* that were studied by Li et al[@bib44] and a number of other mutations that were predicted to be damaging. It is not possible to conclude from our data whether these mutations promote the hypermethylation proposed by Xia et al,[@bib43] and therefore support the oncogenic role of the protein or whether the enhanced protein degradation and increased stemness proposed by Li et al[@bib44] is the predominant purpose of these mutations. Chromatin remodeling is an essential process whereby condensed euchromatin is modified in a context-specific manner to give rise to regions of heterochromatin that can be actively transcribed. Chromatin remodeling is driven by a series of complexes that are able to enzymatically catalyze reactions that modify histone tails and, in turn, modulate the accessibility of the chromatin. In mammalian cells, 5 key chromatin-modifying complexes predominate, the CHD binding complex, the INO80 complex, the SWI/SNF complex, Imitation SWItch (ISWI) complex, and the NuRD complex.[@bib45] Here, we have examined the frequency of mutations in the SWI/SNF complex, which has been shown previously to be perturbed in various cancers. Interestingly, half of CIMP-H1 and more than 25% of CIMP-H2 and CIMP-L1 cancers harbored somatic mutations in SWI/SNF members that were predicted to be deleterious. We hypothesized that mutation of 1 member of the subunit would increase the reliance of the cancer on other otherwise redundant subunits. To test this hypothesis we used public colorectal cancer cell line dependency data in conjunction with mutational data, and identified a strong dependency conferred upon ARID1B after genetic perturbation of ARID1A. These data support the investigation of SWI/SNF inhibitors to exploit synthetic lethality presented by SWI/SNF mutations in CIMP-L1 cancers. Although we have shown associations between genomic methylation and SWI/SNF mutations, and between mutations of SWI/SNF members and synthetic lethality, functional causation is difficult to infer from our study. Collectively, these data indicate a need for further functional experiments to elucidate the role of these mutations in the carcinogenic process of CIMP-H1, CIMP-H2, and CIMP-L1 cancers, and to determine whether the potential synthetic lethalities they create can be exploited. We leveraged the publicly available DNA methylation data from the TCGA project to validate findings in our consecutive cohort. Key findings, including relationships between CIMP subtype and age, proximal location, *BRAF* mutation, and *KRAS* mutation also were identified in an analysis of the TCGA data. In our unselected and consecutively collected series we observed a strong relationship between the *BRAF* mutation and CIMP-H1 and the KRAS mutation and CIMP-H2. Although *BRAF* was still enriched in the TCGA CIMP-H1 cancers, and *KRAS* among the CIMP-H2 cancers, we observed a higher proportion of *BRAF* mutant CIMP-H2 cancers in the TCGA cohort. The increased proportion of *BRAF* mutant/CIMP-H2 cancers skewed these cancers toward a preference for microsatellite instability, and the CMS1 subtype. It is notable that more than 40% of CIMP-H2 cancers in the validation cohort are *KRAS* mutant, and, of these, the majority are microsatellite stable and follow similar CMS patterns to that observed in our consecutive series. The discrepancies observed between the 2 cohorts may be owing to structural differences in each cohort. The mean age of patients in our study was 3.4 years older than those in the TCGA cohort. Cancers were identified most often in the distal colon of the patient, as is typical for colorectal cancers,[@bib46] however, in contrast, the TCGA consisted of a marked over-representation of proximal cancers (47.7%). It is important to recognize the limitations of our study. First, our samples were collected in a consecutive manner in which there was sufficient sample available for DNA and RNA analyses. This excluded very small cancers and those in patients in whom surgery was not possible. This presents a slight bias, however, this is standard practice and unavoidable in studies of this nature. As technologies improve and analyses are possible on smaller amounts of tissue it will be important to replicate the key findings of this study. Moreover, because we collected fresh tissue we were not able to make any assessments of tumor purity. One alternative would have been to perform analyses on formalin-fixed, paraffin-embedded samples, in which we could perform accurate histologic assessments of the purity of the samples. Although the Illumina HM450 platform and newer platforms such as the EPIC arrays are amenable to formalin-fixed, paraffin-embedded--derived DNA, co-extraction of high-quality RNA from formalin-fixed, paraffin-embedded remains challenging. We note that the findings of this study are largely correlative and as such we cannot draw causation from our data. In depth, mechanistic follow-up evaluation is necessary to fully examine many of the key associations we have identified in the present study. Another limitation of our study was the use of normal mucosal samples from patients with cancer. Field DNA methylation defects have been reported in colorectal cancer.[@bib47] Thus, we cannot exclude the possibility that field DNA defects impacted our analysis. In the current study, we performed all analyses on bulk tissue samples. As such, we have collected the DNA methylome and transcript profile of an aggregate of cells that includes epithelial cells, immune cells, and stromal cells. The interplay between these cell types is crucial and it is important to note that some of the expression and methylation differences observed here may be driven by any of the cells in the bulk cell sample. Conclusions {#sec3} =========== The past decade has heralded an era in which the importance of the cancer epigenome increasingly is recognized, in which treatments targeting different epigenetic modifications are entering the clinic and improving patient outcomes. Although early strategies targeting epigenetic modifications in colorectal cancers largely have proved ineffective, it has become apparent that a comprehensive understanding of the epigenetic drivers of cancer will be crucial in the rational design of clinical trials and the development of precision medicine strategies. Here, we have identified 5 clinically and molecularly distinct subgroups based on a comprehensive assessment of a large, unselected series of colorectal cancer methylomes. We have validated these subtypes in an additional cohort of 374 cancers from TCGA. In contrast to earlier studies, we identified 2 clinically and molecularly distinct CIMP-H clusters. We observed a striking association between genomic methylation and age, which further supports the investigation of the epigenetic clock in serrated neoplasia risk. We identified an association between gene body methylation CIMP-H cancers, which may be mediated by H3K27me3 histone marks. Our interrogation of the coding regions of epigenetic regulatory genes shows that they frequently are mutated in colorectal cancers and this may be partially influenced by the degree of genomic methylation. Our analyses have identified potentially druggable vulnerabilities in cancers of different methylation subtypes. Inhibitors targeting synthetic lethalities, such as SWI/SNF component inhibitors for those with *ARID* mutations, should be evaluated because these agents may be clinically beneficial to certain patient subsets. Methods {#sec4} ======= Patient Samples {#sec4.1} --------------- Colorectal cancer (N = 216) and matched normal (N = 32) samples were obtained from patients undergoing surgery at the Royal Brisbane and Women's Hospital in Brisbane, Australia, in a consecutive manner between 2009 and 2012. Tissue was snap-frozen in liquid nitrogen to preserve sample integrity. Written informed consent was obtained from each patient. The study protocol was approved by the Royal Brisbane and Women's Hospital and QIMR Berghofer Medical Research Institute Research Ethics Committees. TCGA colon adenocarcinoma exome and methylation data (N = 278) were used for independent validation.[@bib16] DNA and Messenger RNA Extractions {#sec4.2} --------------------------------- DNA and messenger RNA (mRNA) were extracted simultaneously from approximately 30 mg of homogenized tissue using the AllPrep DNA/RNA Kit (QIAGEN, Hilden, Germany) in accordance with the manufacturer's protocols. Double-stranded DNA concentration was assessed using the PicoGreen quantitation assay (Thermofisher Scientific, Waltham, MA). mRNA quality was measured using the Bioanalyzer 2100 platform (Agilent, Santa Clara, CA). Microarray analysis was performed on samples with a RNA integrity number greater than 7. Molecular Characterization of Cancer Samples {#sec4.3} -------------------------------------------- Cancer sample DNA was analyzed for the *BRAF* V600E mutation using allelic discrimination as previously reported.[@bib48] In addition, we assayed mutations in *KRAS* codons 12 and 13, and *TP53* exons 4 to 8 using previously reported methods.[@bib49], [@bib50] We assessed CIMP status by methylation-specific polymerase chain reaction using the 5-marker panel (CACNA1G, IGF2, NEUROG1, RUNX1, and SOCS1) proposed by Weisenberger et al.[@bib3] Samples were considered CIMP-high if 3 or more markers were methylated, CIMP-low if 1 or 2 markers were methylated, and CIMP-negative if no markers were methylated. MSI was assessed using the criteria of Nagasaka et al[@bib51] in which instability in 1 or more mononucleotide markers, and 1 or more additional nonmononucleotide markers, using the marker set reported by Boland et al,[@bib52] was indicative of MSI, the remainder being microsatellite stable. *LINE1* methylation was assessed using pyrosequencing as per Irahara et al.[@bib53] CIMP-high cancers that were both *KRAS* and *BRAF* wild-type at hotspot codons were Sanger sequenced for *BRAF* exons 11 and 15 (exon 11, forward: 5'-TTCCTGTATCCCTCTCAGGCA-3', reverse: 5'-AAAGGGGAATTCCTCCAGGTT-3'; exon 15, forward 5'-GGAAAGCATCTCACCTCATCCT-3', reverse 5'-TAGAAAGTCATTGAAGGTCTCAACT-3'), *KRAS* codon 61 (forward: 5'-TCCAGACTGTGTTTCTCCCTTC-3', reverse: 5'-TGAGATGGTGTCACTTTAACAGT-3'), and *EGFR* exon 18 (forward: 5'-ATGTCTGGCACTGCTTTCCA-3', reverse: 5'-ATTGACCTTGCCATGGGGTG-3'). DNA Methylation Microarray {#sec4.4} -------------------------- Genome-scale DNA methylation was measured using the HumanMethylation450 BeadChip array (Illumina). The BeadChip array interrogates cytosine methylation at more than 480,000 CpG sites. A total of 500 ng DNA was bisulfite-converted using the EZ-96 DNA Methylation Kit (Zymo Research, Irvine, CA) per the manufacturer's protocol. Whole-genome amplification and enzymatic fragmentation was performed on post-treatment DNA, which subsequently was hybridized to the array at 48°C for 16 hours. Arrays were scanned using the iScan System (Illumina). Gene Expression Microarray {#sec4.5} -------------------------- Gene expression levels for more than 47,000 transcripts were measured for all samples using the HumanHT-12 v3 Expression BeadChip array (Illumina). Total mRNA (500 ng) was reverse-transcribed, amplified, and biotinylated using the TotalPrep-96 RNA Amplification Kit (Illumina). The labeled complementary RNA (750 ng) was hybridized to the array followed by washing, blocking, and staining with streptavidin-Cy3. Arrays were scanned on the iScan System and the data were extracted using GenomeStudio Software (Illumina). Data Analysis {#sec4.6} ------------- Methylation microarray data were checked for quality against parameters provided by Illumina using the GenomeStudio Software package. IDAT files were read into the R environment using Limma.[@bib54] We used subset-within-array normalization to correct for biases resulting from type 1 and type 2 probes on the array. We used the BEclear R package to assess for probe-level batch effects and excluded probes that were significantly batch-affected (n = 1072) from downstream analysis. We filtered probes that had a detection of *P* \> .05 in more than 50% of samples, as well as probes that were on the X or Y chromosome, where the CpG site was within 10 bp of a single-nucleotide polymorphism, or where a probe mapped to the genome ambiguously. At the conclusion of filtering, 377,612 probes remained and were used for subsequent analyses. The RPMM clustering method[@bib55] was used for unsupervised clustering. To capture cancer-specific methylation we followed the methods used based on TCGA.[@bib56] DNA methylation drift with age has been characterized in a number of different normal and cancerous tissues.[@bib10] To limit confounding from methylation that occurs through age, probes with a mean β value greater than 0.3 in normal samples were excluded from clustering analysis. A total of 144,542 probes were unmethylated (mean β value, \<0.3) in normal mucosa, of these the 5000 probes with the greatest variance in tumor samples were selected for clustering. The RPMM clustering method is particularly suited to analysis of methylation data generated from the HumanMethylation450 array because output β values are between 0 and 1, and can be modeled using a β-like distribution.[@bib55] We accessed level 1 DNA methylation data from the TCGA project and performed an identical analysis as mentioned earlier for validation. For motif analysis, the CentriMo tool was used.[@bib21] CentriMo identifies over-represented motifs within sequences, correlating these with known DNA protein-binding motifs.[@bib21] β values were transformed to M values using the following formula: M = log~2~ (β/\[1 - β\]). For differential methylation analysis vs the subset of normal mucosal samples, a probe was considered to be differentially methylated in a comparison if the Benjamini--Hochberg[@bib57] adjusted *P* value for the comparison was less than 0.05 and had an average absolute Δβ ≥ 0.2 vs normal mucosal samples. For examination of methylation in oncogenes and tumor-suppressor genes we consulted the NCG6.0 cancer gene database.[@bib23] For these analyses we included only cancer genes that were annotated in NCG6.0 without ambiguity (were not annotated as both tumor-suppressor genes and oncogenes) and those that we probed on the array. Expression data were preprocessed and normalized using quantile normalization with the Limma R package. For between-group comparisons the empiric Bayes function was used, and adjusted for multiple testing using the Benjamini--Hochberg[@bib57] method to control for FDR and avoid type 1 errors. We examined gene expression in the TCGA by accessing level 3 expression data in Fragments Per Kilobase of transcript per Million reads (FPKM) format from Genome Data Commons.[@bib58] We used Limma to perform a voom transformation to correct for heteroscedasticity and examine differential expression against normal colonic mucosal samples using the same methods as used in the consecutive series. We considered 0.05 to be the FDR threshold for significance. For integrated expression and methylation data analysis, genes were considered to be methylated if 1 probe within 2 kb upstream of the gene transcription start site was methylated differentially by FDR and had an average Δβ ≥ 0.2 at that site. If a gene met this criterion, and had a significant FDR corrected *P* value for the cancer vs normal expression value, it was predicted to be influenced by methylation. Single-sample gene-set enrichment analysis was used for between-groups comparisons of transcriptomes.[@bib18] We used the CIBERSORT algorithm to compute the relative proportion of stromal cells within each subtype.[@bib17] The CMS classifier package was used to classify cancers into CMS as previously reported.[@bib8] To examine the mutational frequency of epigenetic regulators, level 3 somatic variant data were downloaded from the Genome Data Commons portal. Silent variants were discarded and variants in epigenetic regulator genes present in the EpiFactors database extracted for further analyses. We assessed the potential pathogenicity of missense mutations using the PolyPhen2 tool.[@bib59] PolyPhen2 predicts functional effects of missense mutations by examining how evolutionarily conserved the affected residue is, and computes the likelihood that the event will induce a structural change. Only variants that were predicted to be probably or possibly damaging were retained. Variants predicted to be benign were not included as part of these analyses PRC2 and Methylation Overlap Analysis {#sec4.7} ------------------------------------- Polycomb occupancy was inferred from SUZ12 Chromatin Immunoprecipitation (ChIP) sequencing data from hESC1 cells analyzed as part of the Encyclopedia of DNA Elements (ENCODE) ENCODE consortium.[@bib60] SUZ12 was chosen as a surrogate for PRC2 occupancy because previous studies have indicated that it is an essential subunit of the PRC2 complex.[@bib20], [@bib61] The overlap function within BedTools[@bib62] was used to overlap differentially methylated probes within each cluster vs normal with regions where SUZ12 was bound in hESC1 cells, producing a list of regions where methylation and PRC2 occupancy co-occurred. Synthetic Lethality Analysis {#sec4.8} ---------------------------- Cell line dependency data from Meyers et al[@bib27] was correlated with colorectal cancer cell line mutation data.[@bib26] Synthetic lethal relationships were inferred if a high-impact mutation (truncating mutations or those in splice sites) occurred in 1 subunit of a molecular complex, and the cell line had relatively higher dependence values on other subunits when compared with cell lines that lacked a mutation. Cell lines were grouped as having a mutation in a specific gene and those not having a mutation, and a Student *t* test was performed on dependence values in every other subunit within the complex. Statistical Analysis {#sec4.9} -------------------- For statistical analyses a combination of different types of software were used, including R and GraphPad Prism 7 (GraphPad Software, San Diego, CA). The Fisher exact test was used for hypothesis testing on 2 × 2 contingencies. The Pearson chi-squared test was used to compare contingencies larger than 2 × 2. The Student *t* test or the Wilcoxon rank-sum test was used to compare continuous variables where appropriate. One-way analysis of variance was used for continuous variable comparisons with more than 2 groups. Supplementary Material {#appsec1} ====================== Supplementary Material The authors are thankful for the insightful comments offered by the reviewers and for their contribution in greatly improving the manuscript. The microarray data have been deposited in the ArrayExpress database at EMBL-EBI ([www.ebi.ac.uk/arrayexpress](http://www.ebi.ac.uk/arrayexpress){#intref0010}) under accession number E-MTAB-8148 (expression) and E-MTAB-7036 (methylation). **Author contributions** Lochlan Fennell performed bioinformatic and statistical analyses on the data, was involved in the conceptualization aspects of the study, and prepared the manuscript; Troy Dumenil performed molecular and bioinformatic analyses and revised the manuscript for content; Gunter Hartel was involved in the statistical and bioinformatic analysis of the data; Katia Nones was involved in the bioinformatic analysis of methylation data and revised the manuscript for content; Catherine Bond performed molecular analyses and revised the manuscript for content; Diane McKeone was involved in molecular analyses; Lisa Bowdler processed the microarrays; Grant Montgomery processed the microarrays; Leesa Wockner was involved in bioinformatic analyses; Kerenaftali Klein was involved in bioinformatic analyses; Ann-Marie Patch was involved in bioinformatic analyses of The Cancer Genome Atlas exome data; Stephen Kazakoff was involved in bioinformatic analyses of The Cancer Genome Atlas exome data; John Pearson was involved in bioinformatic analyses of The Cancer Genome Atlas exome data; Nicola Waddell was involved in bioinformatic analyses of The Cancer Genome Atlas exome data; Pratyaksha Wirapati performed consensus molecular subtype analysis; Paul Lochhead performed Long Interspersed Nuclear Element-1 methylation assays and analysis; Yu Imamura performed Long Interspersed Nuclear Element-1 methylation assays and analysis; Shuji Ogino performed Long Interspersed Nuclear Element-1 methylation assays and analysis and provided supervision for this aspect of the study; Renfu Shao supervised the study and was involved in the conceptualization aspects of the study; Sabine Tejpar performed Consensus Molecular Subtype analysis and provided supervision for this aspect of the study; Barbara Leggett supervised the study, was involved in conceptualization aspects of the study, revised the manuscript for content, and secured funding for the study; Cheng Liu performed molecular analyses and revised the manuscript for content; Jennifer Borowsky performed molecular analyses and revised the manuscript for content; Isabell Hoffmann performed bioinformatic analysis and revised the manuscript for content; and Vicki Whitehall conceptualized the study, performed statistical analyses, revised the manuscript for content, secured funding for the study, and provided overarching supervision of the study. **Conflicts of interest** The authors disclose no conflicts. **Funding** This work was supported through funding from the National Health and Medical Research Council (1050455 and 1063105), the US National Institutes of Health (R01 CA151933 and R35 CA197735), and Pathology Queensland. Also supported by a Senior Research Fellowship from the Gastroenterological Society of Australia (V.W.); and by a Research Training Program Living Scholarship from the Australia Government and a Top-Up award from QIMR Berghofer and Australian Rotary Health (L.F.). [^1]: NOTE. *P* values reported were obtained using analysis of variance for continuous variables and chi-squared analysis for categoric variables. [^2]: MSS, microsatellite stable. [^3]: NOTE. *P* values reported were obtained using analysis of variance for continuous variables and chi-squared for categoric variables and represent the *P* value for an association between all subtypes and the feature in question. [^4]: MSS, microsatellite stable. [^5]: NOTE. Cancers were stratified for CIMP clustering. Differential methylation was deemed as an absolute β value change of more than 0.2 and an FDR corrected *P* value less than .01 compared with 32 normal colorectal mucosal samples. [^6]: +, differential hypermethylation; -, differential hypomethylation. [^7]: Acc, accession number; BTB, Broad-Complex, Tramtrack and Bric a brac; EGF, epidermal growth factor; HGNC, Human Genome Organisation Gene Nomenclature Committee; LIF, leukocyte inhibitory factor; NDRG, N-Myc downregulated gene. [^8]: BCL, B-cell lymphoma; bHLH, basic helix-loop-helix; ETS, E26 transformation specific; FEV, fifth ewing variant; PI, phosphatidylinositol; SIX, Sineoculis homeobox homolog; TAL, T-cell acute lymphocyctic. [^9]: NOTE. The overlap fraction represents the gene bodies that are methylated (k) divided by the number of genes marked by each respective mark in hES cells (K) (k/K). The FDR corrected *P* value was obtained through modeling a hypergeometric distribution (k-1, K, N-K, n; where k is the number of genes methylated in each cluster; K is the number of genes in the gene set; N is the number of genes in the human genome; and n is the number of genes in the query set) using the compute overlaps tool on the Gene Set Enrichment Analysis (GSEA) web portal using the Benporath gene sets, which were obtained though ChIP-on a Chip analysis of human embryonic stem cells.

GENOME ANNOUNCEMENT {#s0} =================== *Escherichia coli* is a common commensal of the human and animal gut. *E. coli* strain WG5 is used as a host to detect somatic coliphages, proposed to be an indicator of human and animal fecal contamination of water, sediments, and sludge, as described in ISO method 10705-1 (2000) ([@B1]). This is an easily applicable and affordable method for water quality and contamination management in water treatment facilities ([@B2]). Other studies employ a similar method using strain WG5 as a host to detect temperate infectious phages from food ([@B3], [@B4]). The applicability of the method is based on the high sensitivity of the *E. coli* WG5 host to infection by somatic coliphages ([@B5]). *E. coli* WG5 ([@B6]) is a nalidixic acid-resistant mutant of *E. coli* C, also known as strain CN, and is publicly available in the ATCC (ATCC number 700078). *E. coli* strain WG5 possesses an attenuated host restriction-modification system and contains only the core part of the lipopolysaccharide (LPS), increasing susceptibility to phage infections ([@B5], [@B7]). Bacterial DNA was isolated using a DNeasy blood and tissue kit (Qiagen), according to the manufacturer's protocols. DNA libraries were prepared using a TruSeq Nano kit (Illumina) and sequenced on a MiSeq platform (2 × 300 bp). In parallel, genomic DNA was used to prepare barcoded DNA with Native barcoding kit 1D (product number EXP-NBD103; Oxford Nanopore Technologies), together with Ligation sequencing kit 1D (SQK-LSK108; Oxford Nanopore Technologies). DNA was sequenced using R9.4 chemistry (FLO-MIN106; Oxford Nanopore Technologies), and the raw signal was base called using Albacore 1.2.6. The sequences were assembled *de novo* into a single contig using the Unicycler hybrid assembler ([@B8]), with default settings. The WG5 strain has a circular complete genome of 4,592,887 bp. Genome annotation was acquired from NCBI Prokaryotic Genome Annotation Pipeline ([@B9]), which revealed 4,657 genes, 4,536 coding sequences, 22 rRNAs (5S, 16S, and 23S), 87 tRNAs, and 22 noncoding RNAs. Multilocus sequence types were identified using MSTL ([@B10]). No horizontally identified antibiotic resistance genes were detected by using ResFinder ([@B11]). The PHASTER prophage finder ([@B12]) identified three prophage regions, with two incomplete prophage regions and one questionable prophage region. Accession number(s). {#s1} -------------------- The genome sequences have been deposited in GenBank under the accession number [CP024090](http://www.ncbi.nlm.nih.gov/nuccore/CP024090). The version described in this paper is the first version. **Citation** Imamovic L, Misiakou M-A, van der Helm E, Panagiotou G, Muniesa M, Sommer MOA. 2018. Complete genome sequence of *Escherichia coli* strain WG5. Genome Announc 6:e01403-17. <https://doi.org/10.1128/genomeA.01403-17>. This research was funded by the EU H2020 ERC-20104-STG LimitMDR (grant 638902), the Danish Council for Independent Research Sapere Aude Program DFF-4004-00213, and the European Union Seventh Framework Programme (FP7-KBBE-2013-7-singlestage, no. 613745, Promys). M.O.A.S. acknowledges additional funding from the Novo Nordisk Foundation Center for Biosustainability and the Lundbeck Foundation. M.-A.M. acknowledges Spanish Ministerio de Innovación y Ciencia (AGL2016-75536-P), the Agencia Estatal de Investigación (AEI), and the European Regional Fund (ERF).

Introduction {#s1} ============ Research comparing abstract words (e.g., effect) and concrete words (e.g., table) has been abundant. The concreteness effect is often revealed, whereby people perform better with concrete words than with abstract words on a variety of tasks such as memory recall, lexical decision, sentence processing, and translation [@pone.0114421-Kounios1]--[@pone.0114421-West1]. In recent years, another semantic feature, emotionality, has received much attention. Studies have shown that emotionally arousing words demonstrate processing advantage over neutral words [@pone.0114421-Citron1]. Research on concept representation offers perspective on the distinctions between abstract, concrete and emotion words. A concrete concept, such as table or television, usually has a core referent, whereas an abstract concept, such as effect or consultation, is usually represented as a situation or a scenario with a number of key elements [@pone.0114421-Barsalou1]. Among these elements, subjective experience, i.e., what one thinks and how one feels in a certain situation, is essential for the representation of abstract concepts [@pone.0114421-WiemerHastings1]. For example, the representation of consultation should involve an agent with a motive of seeking information, and possibly also a feeling of eagerness. Therefore, one\'s own experience, direct or indirect, with such situation as well as one\'s understanding about this experience must be an integral part of forming and developing the proper concept. Just as Barsalou pointed out, introspection plays an important role in the representation of abstract concepts [@pone.0114421-Barsalou2]. In this regard, emotion concepts (e.g., joy, guilt, sadness, and excitement) may be considered an extreme case, the formation and development of which rely primarily on introspective effort. People reliably differ in their levels of self-insight, i.e., awareness and understanding of one\'s subjective experiences including cognitive or affective states and processes [@pone.0114421-Grant1]. Consider the aforementioned representational differences between concrete concepts and abstract concepts, including emotion concepts as the special case in the abstract domain, individual variations in self-insight might be reflected as differences in semantic knowledge about abstract words in general and emotion words in particular. Specifically, a high level of self-insight should be reflected as rich semantic knowledge about words that denote abstract concepts involving cognitive or affective experiences and particularly about words that denote complex emotions, which may eventually translate into speedy and effortless processing of these words. Intuitively, introspective clarity or insight may not be as relevant to the processing of concrete words. However, concept formation or word learning in general requires reflecting upon one\'s own perception, memory, and thought processes. For example, car is an obvious concrete word, but driving experience must be part of our knowledge about cars. In addition, the process of learning to read, to write, and to understand the meaning of this word is also an experience subjective to introspection. Consequently, efficient processing of concrete words might also relate to one\'s level of insight. It thus can be speculated that self-insight is relevant to word processing in general, but the extent to which it is relevant may vary across different word categories, depending on how much introspective knowledge is involved in acquiring the concept. The present study utilized the ERP technique to examine the effect of self-insight on semantic processing for three categories of words: abstract neutral words (e.g., logic and effect), concrete neutral words (e.g., teapot and table), and emotion words (e.g., despair and guilt). It was expected to reveal individual differences in word processing, which were associated with different levels of self-insight. After an extensive review of the literature about emotional words, Citron indicated that research on individual differences, particularly among healthy population, has been lacking [@pone.0114421-Citron1]. In fact, individual differences in semantic processing in general are very much in need of investigation. This is especially true when it comes to words denoting abstract entities or events. This study was therefore expected to contribute to the literature along these lines. Furthermore, through examination of the relations between self-insight and ERP responses to different word categories, findings of this study may reveal potential electrophysiological evidence for representational differences across word categories. Structural and functional differences between the male brain and the female brain, including the sexually dimorphic nature of the amygdala, have been well documented [@pone.0114421-Cahill1]. Gender thus appears particularly relevant to the aim of the present study to examine individual differences across and within different word categories, including emotion words. In particular, converging evidence has clearly shown that women experience more self-conscious emotions such as guilt relative to men [@pone.0114421-ElseQuest1], which indicates that women and men may differ in their levels of introspective knowledge regarding at least some aspects of their affective experiences. Gender is one of the few individual differences variables that have been investigated for semantic processing. However, research on gender differences along the time course of word processing is limited. Wirth et al. analyzed participants\' N400 responses in a word priming task [@pone.0114421-Wirth1]. Female participants were found to always engage in more elaborative processing of semantic information than did male participants. Daltrozzo et al. obtained the same results in the N400 time window, but reported a lack of significant gender difference in P300 [@pone.0114421-Daltrozzo1]. The reason for the lack of evidence for gender variation at the early stage could be due to possible insensitivity of the early ERP component. Alternatively, it could be due to the specific word samples chosen by the studies. As Daltrozzo et al. pointed out, men and women responded to different categories of words differently [@pone.0114421-Daltrozzo1]. To achieve a certain level of balance across different semantic categories, their study incorporated various types of words including a small portion of abstract words and erotic words. Similarly, the study by Wirth et al. included 50% abstract words and 50% concrete words [@pone.0114421-Wirth1]. However, neither study conducted gender comparison within each of the different word categories. A study by Sass et al. examined gender differences in early ERP components in response to pleasant words, threat words, and neutral words [@pone.0114421-Sass1]. They found that men showed preferential processing of threat words at about 100 ms post word onset, while women showed preferential processing of threat words not until 300 ms post word onset. Extending from these earlier studies, the present study incorporated gender as another individual differences variable along with level of self-insight to investigate their effects on early processing of emotion words, abstract neutral words, and concrete neutral words. With these different word categories, the study was expected to extend the knowledge about the timing of gender variation in semantic processing across word categories and potential gender contrast in the relationship between self-insight and semantic knowledge. In sum, this study aimed to examine the effects of gender and self-insight on the early stage of word processing with abstract neutral words, concrete neutral words, and emotion words. The study utilized a paradigm to obtain an early ERP component, the Recognition Potential (RP), which is considered an index for semantic activation, peaking between 200 and 250 ms [@pone.0114421-MartnLoeches1], [@pone.0114421-Rudell1]. With a mixed word sample, gender difference in processing level has been shown emerging in the N400 time window [@pone.0114421-Wirth1], [@pone.0114421-Daltrozzo1]. With emotionally arousing words, research found gender difference before 300 ms post stimuli [@pone.0114421-Sass1]. It was therefore predicted that the RP would reveal gender difference in early processing of emotion words. However, Rudell and Hua showed that the RP was sensitive to individual differences in reading ability [@pone.0114421-Rudell2]. Therefore, should gender difference exist in early word processing in general, not just in the processing of emotion words, the RP would also reveal gender variation in response to abstract and concrete neutral words. In such case, women were expected to show a greater processing level relative to men [@pone.0114421-Wirth1], [@pone.0114421-Daltrozzo1]. The investigation of the relationship between level of self-insight and word processing was exploratory in nature. Based on the above discussion that representations of abstract words and emotion words contain essentially introspective knowledge relative to that of concrete words, a tentative prediction was that level of self-insight would show stronger association with the processing of abstract words and emotion words than with the processing of concrete words. Gender variation might exist with regard to the relation between self-insight and word processing, particularly for emotion words given that women and men showed differential responses to this type of words at early processing stage and different levels of introspective knowledge about this type of concepts [@pone.0114421-ElseQuest1], [@pone.0114421-Sass1]. As for RP differences across word categories, based on past research [@pone.0114421-MartnLoeches2], [@pone.0114421-MartnLoeches3], it was predicted that concrete neutral words would elicit greater RP than abstract neutral words. In addition, in one study with similar experimental paradigm, Hinojosa et al. found in the 225--300 ms time window greater negativity for some words with emotional significance relative to neutral words [@pone.0114421-Hinojosa1]. It was therefore predicted that emotion words might lead to greater RP than abstract and concrete neutral words. Finally, one thing needed to be pointed out is that many studies on the effect of emotionality in word processing intermixed emotion words (e.g., despair, guilt, joy, sadness) that denote emotion states or processes with emotional words (e.g., success, hostility, flower, and enemy) that denote non-emotion concepts but evoke emotional responses [@pone.0114421-Citron1]. Because emotional words can vary in abstractness, these studies often employed the factorial design to examine both abstractness and emotionality, and often reveal complex, yet inconsistent, interactions between the two factors [@pone.0114421-Kaltwasser1]--[@pone.0114421-Palazova1]. Acknowledging complexity of the interplay between abstractness and emotionality in word processing, for the specific purpose of this research, the experiment described below utilized only emotion words in order to detect potential individual differences in the representation and processing of emotion concepts relative to abstract and concrete neutral concepts. As indicated by prior research on emotion words, they are abstract in nature [@pone.0114421-Altarriba1], [@pone.0114421-Altarriba2]. Therefore, the present study employed a one-way design to compare the three types of words. Method {#s2} ====== Ethics Statement {#s2a} ---------------- This study was approved by the Institutional Review Board of the Imaging Center for Brain Research of Beijing Normal University. All participants gave written informed consent prior to the experiment and were paid for their participation. Participants {#s2b} ------------ Fifty-five college students were recruited for the experiment. The final analysis excluded five participants with signals from electrodes of interest interfered by unknown noise, another participant with a response accuracy rate lower than 75%, and yet another participant with no accurate responses due to misunderstanding of task instruction. This resulted in 48 participants (25 females) with an average age of 22.06 (*SD* = 1.82) ranging from 19 to 26. They were all native speakers of Chinese, right-handed, with normal or corrected to normal vision. As one of the recruitment criteria, right-handedness was determined based on participant report, and further verified by the experimenter through observation during the course of the experiment. Experimental Stimuli {#s2c} -------------------- Following the paradigm described by earlier studies of RP [@pone.0114421-MartnLoeches2], the experimental stimuli included real words, non-words, and background stimuli. There were three types of real Chinese two-character words (all nouns): 48 abstract words, 48 concrete words, and 48 emotion words (28 negative words and 20 positive words). These real words were sampled from a database that contained 715 Chinese two-character words with normative information on frequency, part of speech, number of strokes, and emotionality ratings. The emotionality was assessed using a 5.0 scale (0 =  "neutral" and 4 =  "high emotional value"). For the purpose of this study, abstractness ratings were collected using a 6.0 scale (1 =  "highly concrete" and 6 =  "highly abstract") from a group of 21 Chinese native speakers who did not participate in the main experiment. Considering that this rating process can be sometimes difficult, an even number (6.0) was chosen for this bipolar scale (highly concrete versus highly abstract) in order to prevent potentially a large number of convenient responses by selecting the mid-point of the scale (e.g., 4 on a 7.0 scale). The properties of the sampled abstract, concrete and emotion words are displayed in [Table 1](#pone-0114421-t001){ref-type="table"}. The three types of words were matched on frequency (*F*\<1) and number of strokes (*F* = 2.30, *p*\>.10). Abstract words and concrete words were also matched on emotionality ratings (*p*\>.45), both rated lower than emotion words (both *p*s\<.001). Abstract words and emotion words were matched on abstractness ratings (*p*\>.27), both rated higher than concrete words (both *p*s\<.001). In addition, to constitute a semantic judgment task, 48 animal words were collected with their numbers of strokes matched with the three types of real words. 10.1371/journal.pone.0114421.t001 ###### Properties of the three types of real words. {#pone-0114421-t001-1} Property Word Type Mean SD --------------------- ---------------- ------- ------ frequency (log) abstract words 3.03 1.50 concrete words 2.84 1.24 emotion words 3.02 1.20 number of strokes abstract words 16.75 3.39 concrete words 17.98 4.23 emotion words 18.42 4.18 emotionality rating abstract words 0.16 0.11 concrete words 0.10 0.11 emotion words 2.40 0.36 abstractness rating abstract words 4.07 0.56 concrete words 1.46 0.31 emotion words 4.20 0.23 There were 48 non-words. A non-word was created by cutting two Chinese characters each into half and then re-arranging the four pieces. As a result, these non-words appeared to resemble real two-character words, but were devoid of meaning. There were 240 background stimuli. A background stimulus was created by cutting a pair of two Chinese characters into 32 pieces and re-arranging them randomly to form an image without word-like characteristics or meanings. Exemplars for each type of stimuli are listed in the [S1 Appendix](#pone.0114421.s001){ref-type="supplementary-material"}. Measure for Level of Self-insight {#s2d} --------------------------------- The Self-Reflection and Insight Scale (SRIS) [@pone.0114421-Grant1] served to assess individual differences in self-insight. It contains 20 statements. Respondents indicate how much each statement describes their own characteristics by rating on a 6.0 scale. The SRIS consists of two subscales, self-reflection (SRIS-SR) and insight (SRIS-IN). SRIS-SR was designed to measure "the inspection and evaluation of one\'s thoughts, feelings, and behaviors," while the SRIS-IN was intended to measure "the clarity of understanding of one\'s thoughts, feelings, and behaviors" [@pone.0114421-Grant1]. Grant et al. indicated that scores from SRIS-SR were not always positively associated with, and thus not reflective of, the level of insight [@pone.0114421-Grant1]. Therefore, for the purpose of this study, only the summation score collected from SRIS-IN was used to assess individual differences in self-insight. In this study, participants\' insight scores ranged from 15 to 45, with a mean of 32.17 (*SD* = 6.07). There was not a significant difference between men (*M* = 33.17, *SD* = 6.56) and women (*M* = 31.24, *SD* = 5.56), *p*\>.27. Procedure {#s2e} --------- All participants signed a written informed consent form prior to the experiment. The experiment was carried out in a soft-lighted and soundproof recording room. All stimuli were presented white-on-black, 3.5 cm high and 6.6 cm wide, in the middle of the computer screen. Participants sat about 100 cm from the screen. The Rapid Stream Stimulation paradigm [@pone.0114421-Hinojosa2], [@pone.0114421-Rudell3] was employed with a stimulus onset asynchrony (SOA) of 250 ms. The screen refresh rate was set at 60 Hz, and each stimulus was presented for 250 ms. The experiment included three sessions. Each session contained 16 abstract words, 16 concrete words, 16 emotion words, 16 animal words, 16 non-words, and 80 background images. Each session began with a string of asterisks (\*\*\*\*\*\*), where participants were instructed that they could blink as much as they needed, and that they could begin the presentation of the stimuli by pressing a key when they were ready. Then, there were eight sequences in a session. Each sequence contained ten test stimuli, including two abstract words, two concrete words, two emotion words, two animal words, and two non-words presented in random order. Each word or non-word was preceded by four to six background stimuli. The presentation order of the sequences within each session was randomized, so was the presentation order of the sessions. Participants were instructed to press a key as quickly and accurately as possible whenever they detected an animal word. All stimuli were presented only once. During the experiment, there were two breaks, with duration determined by the participants. To ensure that the participants understood the task, they practiced with a different set of stimuli before the experiment. After the experiment, they filled out the Self-Reflection and Insight Scale [@pone.0114421-Grant1]. The whole experiment lasted about 1.5 hours. EEG Recording and Data Analysis {#s2f} ------------------------------- Participants\' electroencephalograms (EEG) were recorded from a 32-channel Quik-cap (NeuroScan, Inc.) with the right mastoid as reference. Vertical eye movements were recorded by two electrodes symmetrically placed above and below the left eye. Horizontal eye movements were monitored by two electrodes placed on the left and right outer canthi. Impedances of all electrodes were kept below 5 kΩ. The sample rate was 500 Hz with a band-pass of 0.05--100 Hz. The data were re-referenced offline using the global average reference method, which was proved to be the best way to obtain the RP [@pone.0114421-MartnLoeches2]. The continuous data were segmented from 100 ms pre-stimulus to 600 ms post-stimulus for abstract words, concrete words, emotion words, and non-words. Data were filtered offline with a low pass of 30 Hz (24 dB). The mean voltage of the 100 ms pre-stimulus interval acted as the baseline for ERP measurement. Wrong trials and trials contaminated by eye blinks, eye movements, or muscle potentials exceeding ±100 µV at any scalp electrode were excluded from analysis. Thus, about 6.66% of trials were excluded. The segmented data were then averaged for each word type within each participant. Data extraction and analyses focused on average recordings from P7, P8, O1, and O2 electrodes, the areas where RP was found most evident [@pone.0114421-MartnLoeches2], [@pone.0114421-Hinojosa2], [@pone.0114421-Iglesias1]--[@pone.0114421-Zhang1]. The peak latency and peak amplitude value of RP were measured in the 150--400 ms time window that covered the presentation time of one stimulus and included the most negative peak of RP [@pone.0114421-Rudell2], [@pone.0114421-MartnLoeches2]. More specifically, for each participant, a peak value and the corresponding peak latency were extracted from the most negative peak within the time window under each word condition at each of the four electrode sites. To examine gender difference in RP peak latency across different word conditions, a mixed 4×2 ANOVA with word condition (abstract, concrete, emotion, and non-word) and gender as factors was conducted on RP latency at each of the four electrode sites of interest: P7, P8, O1, and O2. The same analysis was applied to examine gender difference in RP peak amplitude. For the above analyses, the Geisser-Greenhouse correction for non-sphericity was applied when appropriate, with unadjusted degrees of freedom presented. To examine the relationship between insight level and the RP, correlational analysis was first performed. Regression analysis then further examined the role of self-insight in semantic activation of real words by partialing out individual variation in response to non-words. Results {#s3} ======= Gender Differences {#s3a} ------------------ ### Latency {#s3a1} [Fig. 1](#pone-0114421-g001){ref-type="fig"} presents the RP at the four chosen electrode sites. The mixed 4×2 ANOVA with word condition and gender as factors revealed a significant word condition effect only at P8, *F* (3, 138)  = 7.42, *p*\<.001. The same effect seemed to approach significance at P7 and O2, both *p*s = .07. Bonferroni *t*-tests showed that, at P8, abstract words (*M* = 225 ms) and emotion words (*M* = 220 ms), but not concrete words, elicited shorter RP latency than did non-words (*M* = 234 ms), both *p*s\<.04. In addition, emotion words (*M* = 220 ms) had shorter RP latency than concrete words (*M* = 227 ms), *p* = .02. No gender effect or gender by word interaction approached significance at any sites. {#pone-0114421-g001} ### Peak amplitude {#s3a2} To examine potential gender difference in level of processing, the 4×2 mixed ANOVA with word condition and gender as factors was applied to RP peak amplitude at P7, P8, O1, and O2, respectively. [Fig. 2](#pone-0114421-g002){ref-type="fig"} contrasts the RP between two gender groups. At O2, female participants (*M* = −7.63 µV) showed significantly greater RP than male participants (*M* = −6.04 µV), *F* (1, 46)  = 4.72, *p* = .035. The absence of gender by word interaction indicated that this difference was consistent for all real word conditions and the non-word condition, which may suggest that, compared to men, women engage in deeper processing in response to both written words and word-like visual stimuli. {#pone-0114421-g002} The ANOVA also revealed a significant main effect of word condition at P7, *F* (3, 138)  = 12.06, *p*\<.0000005. Bonferroni *t*-tests showed that all three real word conditions (*M* = −7.22 µV for abstract words, *M* = −7.39 µV for concrete words, and *M* = −7.55 µV for emotion words) led to significantly greater RP than did the non-word condition (*M* = −6.55 µV), all *p*s\<.005, which indicated that semantic activation was evident at P7. No other effects appeared significant at this or other sites. Level of Self-insight {#s3b} --------------------- To reveal potential gender contrast in the relationship between self-insight and word processing, correlational analysis was first performed for each gender group. For the male group, insight scores appeared negatively related to RP latency at all sites, but most evidently at P7 for abstract words, *r* (23)  = −.478, *p* = .021, and concrete words, *r* (23)  = −.441, *p* = .035. Higher insight scores were associated with shorter latency for both types of words ([Fig. 3](#pone-0114421-g003){ref-type="fig"}). For the female group, insight scores showed a significant positive correlation with RP peak amplitude at O2 for abstract words, *r* (25)  = .461, *p* = .020, and emotion words, *r* (25)  = .482, *p* = .015. Lower insight scores were associated with greater RP for both types of words ([Fig. 4](#pone-0114421-g004){ref-type="fig"}). {#pone-0114421-g003} {#pone-0114421-g004} To test whether insight level accounts for variation in semantic activation instead of simply variation in RP responses to word and non-word stimuli in general, regression analysis was applied to further examine the effect of insight scores on RP responses to the real words identified in the above correlational analysis. Specifically, for the male group, RP latency at P7 in response to abstract and concrete words was regressed upon insight score after partialing out variation in RP latency in response to non-words at this electrode site. After controlling for RP latency of non-words, insight scores no longer accounted for a significant amount of variation in RP latency for abstract words (Δ*R^2^* = .07, *F* (1, 20)  = 3.14, *p* = .09) or concrete words (Δ*R^2^* = .05, *F* (1, 20)  = 2.21, *p* = .15). For the female group, RP peak amplitude at O2 in response to abstract and emotion words was regressed upon insight score after partialing out variation in RP peak amplitude in response to non-words at O2. After controlling for variation in non-words, insight scores still accounted for a significant amount of variation in RP peak amplitude for abstract words (Δ*R^2^* = .07, *F* (1, 22)  = 8.34, *p* = .009), and for emotion words (Δ*R^2^* = .10, *F* (1, 22)  = 7.17, *p* = .014). Discussion {#s4} ========== Through examining the RP, an early ERP component, during a semantic judgment task, this study investigated the effects of two individual differences variables, gender and level of self-insight, on early semantic processing. The experiment utilized emotion words, abstract neutral words, and concrete neutral words. Results indicated that women showed a greater level of processing, as assessed by RP amplitude, in response to real word and word-like stimuli than did men. Level of self-insight demonstrated differential relations with RP responses of men versus women. For men, the effect of self-insight manifested as difference in speed of processing, whereas for women, the effect manifested as difference in level of processing. Most notably, for women, level of self-insight accounted for a significant amount of variation in semantic activation for abstract neutral words and emotion words. Individual Differences {#s4a} ---------------------- Early research has shown that women tend to engage in more elaborative semantic processing [@pone.0114421-Wirth1], [@pone.0114421-Daltrozzo1], [@pone.0114421-MeyersLevy1]. However, with a mixed word sample, gender difference did not become evident until the N400 time window [@pone.0114421-Daltrozzo1]. This study found that, as early as 200--250 ms post stimuli, female participants showed greater electrophysiological activities than male participants in response to real word and word-like visual stimuli. Further, gender difference did not just manifest at the early stage of processing for semantically salient emotion words, as shown by Sass et al. [@pone.0114421-Sass1]. It was evident across three word categories. The RP is considered an index representing the process of analyzing visual form to gain semantic access [@pone.0114421-MartnLoeches1]. Therefore, gender difference revealed here in both real words and non-words suggests that women start to engage in deeper processing than men not only before semantic integration as indexed by N400 in previous studies, but also before actual semantic access is attained. In the literature, the time course and mechanism of verbal processing by which the male brain and the female brain differ have not been fully investigated [@pone.0114421-Wirth1]. The findings of this study add to the literature with regard to the timing of gender differences in semantic processing. It awaits future research to further investigate whether such early gender variations manifest specifically in response to word-like visual stimuli, and what characteristics of visual stimuli trigger such differential reactions. Gender differences were further evidenced in the relationship between level of self-insight and semantic processing. This finding was astonishing in that the two gender groups were qualitatively different in terms of how level of self-insight influenced the RP waveform. To better depict this contrast between the female and the male participants, within each gender group, a median split based on their insight scores divided participants further into two groups, the high insight group and the low insight group. [Fig. 5](#pone-0114421-g005){ref-type="fig"} displays RP responses by the high and low insight groups for male versus female participants. {#pone-0114421-g005} For the female group, at the electrode site O2, higher level of self-insight was found significantly associated with smaller RP amplitude in response to abstract words (e.g., logic, feature, and tradition) and emotion words (e.g., despair, sadness, and gratitude). This was true even after controlling for individual variation in response to non-word visual stimuli. That is, level of self-insight can explain a significant amount of variation in the level of semantic activation for abstract and emotion words. Individuals who reported lower levels of self-insight showed enhanced activation to these two types of words, whereas those who reported higher levels of self-insight showed relatively reduced activation to these words. Statistically, this relationship did not hold for the processing of concrete words (e.g., teapot, window, and vehicle). The differential relationship of insight to the processing of different categories of words may be a reflection that the level of introspective knowledge involved in word processing varies as a function of these word categories. As discussed earlier, research has indicated that subjective experiences are an integral part of the representation of abstract concepts [@pone.0114421-WiemerHastings1]. Intuitively, this seems particularly the case for the representation of emotion concepts, whereas the representation of concrete concepts may rely to a much lesser degree on introspective knowledge about one\'s subjective experiences. Recent research [@pone.0114421-Kousta1], [@pone.0114421-Vigliocco1] on abstract concepts seems to offer even further explanation for the parallel between abstract and emotion words revealed by the present study. These studies have shown that one\'s emotional experiences play a crucial role in the representation of abstract concepts. That is, involvement of emotional experiences, rather than other types of subjective experiences, in the representations of abstract and emotion words could be the primary reason why processing both types of words were found closely related to self-insight. The finding that insight appeared more relevant to the activation of abstract words and emotion words relative to concrete words therefore may be considered further evidence for the representational differences across word categories. However, it is not clear why this was only reflected in the RP of female participants. It might be possible that the greater RP of this gender group rendered the relationship between insight and processing level more detectable. This certainly requires further research to be addressed. The revealed association of insight with the processing of abstract and emotion words among women was correlational in nature, and it needs future research to explore potential causality. On one hand, it is possible that an inherent self-perception of uncertainty about one\'s own subjective experiences results in more extensive activation of introspective knowledge during processing of these words, and hence an augmented level of electrophysiological reaction. On the other hand, it is also possible that a predisposed tendency to engage in enhanced electrophysiological reactions to words that evoke introspective knowledge leads to a self-perception of uncertainty about relevant experiences. More likely the two factors are intertwined during the course of mental development. Research on the relationship between these factors may have implications for intervention with cognitive and affective disorders. For the male group, the relationship between insight level and RP responses to word processing appeared less conclusive. In contrast to the female group, a higher level of self-insight was found associated with shorter RP latency, particularly evident in response to abstract and concrete words at the electrode site P7. However, insight scores did not account for latency variation in response to these real words above and beyond individual variation in response to non-word visual stimuli. This may indicate that level of self-insight is not a reliable predictor for latency of semantic activation. Alternatively, this may be partially due to the fact that insight scores among male participants appeared negatively skewed. While the majority of the male participants\' scores were well above the mid-point of the insight scale, three participants reported substantially lower insight scores ([Fig. 3](#pone-0114421-g003){ref-type="fig"}), which may or may not be an unbiased reflection of their actual insight levels. Analysis excluding these three participants revealed even stronger correlation between insight and RP latencies for these words. While subjectivity is inherent to self-report measures, a larger sample may help elucidate the relationship of insight and processing speed among male participants in future research. For male participants, the finding that higher insight tended to be associated with shorter latency in response to word stimuli (Pearson\'s *r*s ranging from.33 to.48 for emotion, concrete, and abstract words) may prompt one to ponder whether language proficiency, particularly vocabulary knowledge, is the underlying factor for the relationship between insight and the speed of semantic judgment. Rudell and Hua indeed reported an association between RP and language proficiency as assessed by GRE verbal test [@pone.0114421-Rudell2]. Their participants in the fourth quartile of GRE verbal scores generated the longest RP latency and the smallest RP amplitude. This association appeared to hold for both men and women. However, in the present study, low insight scores were associated with longer RP latency only for the male group. For the female group, low insight scores were associated with greater RP amplitude. This contrast suggests that verbal proficiency and insight are two factors that, though possibly partially overlapping, differ in the mechanisms by which they influence the RP waveform. Future research is needed to investigate the mechanism through which insight, gender, and verbal proficiency interact in the process of language processing. Different Word Categories {#s4b} ------------------------- This study found that the three real word categories elicited comparable levels of RP. The lack of difference between abstract words and concrete words was in contrast to Martin-Loeches et al. [@pone.0114421-MartnLoeches2], which utilized the same experimental paradigm and revealed that concrete words elicited greater RP than abstract words. The following might have contributed to this discrepancy. First, the present study included emotion words as a separate category of comparison, and excluded emotion words from the abstract word category, which resulted in a word sample with different semantic features from that of the earlier study. Further, to distance abstract and concrete words from emotion words, the emotionality ratings of these two categories of words were designed to fall on the lower end of the scale. Under this constraint, matching emotionality, frequency, and stroke number between abstract and concrete conditions further restricted word sampling. As a result, the contrast in abstractness ratings between the two conditions was relatively limited compared with the rating contrast in the earlier study. A numerical comparison of rating difference between the two studies showed that this was indeed the case. Second, the fact that the two studies were conducted in two different languages, Spanish and Mandarin, might also have been a factor. Specifically, non-words in the present study contained recognizable radicals and thus, to a large degree, preserved word-like visual features. A more careful control for visual form similarity between abstract words and concrete words relative to non-words may help to detect semantic processing differences between these two real word conditions. Further research is necessary to investigate RP amplitude in response to the processing of different semantic features as well as the application of RP paradigm to research in different languages. Finally, emotion words did not elicit greater RP than did abstract and concrete words. In the literature about word emotionality, more studies have been done through examining another early ERP component, the early posterior negativity (EPN). Some studies reported that the EPN was amplified by emotionally significant words around 250 ms post word onset [@pone.0114421-Hinojosa1], [@pone.0114421-Palazova1], [@pone.0114421-Kissler1]. However, other studies using emotional words did not find the EPN [@pone.0114421-Kaltwasser1], [@pone.0114421-Kanske1]. The EPN has been thought to signify allocation of attentional resources [@pone.0114421-Potts1], [@pone.0114421-Schacht1], and functionally linked to the RP [@pone.0114421-Kissler2]. Therefore, the lack of difference between emotion words and neutral words in the present study could be due to the semantic judgment task, which might to some extent have diverted attention away from word emotionality. Further, the fact that the present study utilized only emotion words (not emotionally significant words) might also have been a factor. Unfortunately, there does not seem to be research looking into processing differences between words that denote specific emotion states or processes and words that have general emotional connotations. A follow up study seems appropriate to address this issue. Limitations {#s4c} ----------- As emphasized earlier, the RP is an early ERP component indexing early stage of word processing. The findings of this study likely reflect mainly the automatic and implicit stage of semantic processing. To gain a more comprehensive understanding about processing differences across various word categories and individual differences in word processing, later ERP components certainly need to be investigated. The Rapid Stream Stimulation (RSS) paradigm in this study is generally utilized to enhance the RP [@pone.0114421-Rudell3]. A different design is needed to obtain more standard later ERP components, which will be adopted in future research to examine the more controlled and explicit stage of semantic processing. This study employed a semantic judgment task with animal words serving as the target words to be responded to by participants. There might be a possibility that this design could have distorted the differences between abstract, concrete, and emotion words. More specifically, animal words denote a type of concrete concept that might share greater conceptual similarities with concepts denoted by concrete words than those denoted by abstract words or emotion words. Consequently, word processing during the experiment might have been biased toward concrete words, but away from abstract and emotion words. It is not clear how much difference this could make at the early stage of word processing. However, for future research, a design to induce more balanced processing across word categories would be helpful. Lastly, the primary goal of this study was to explore individual differences in word processing. Although emotion words were included as one of the word categories for investigation, valence was not taken into consideration. Roughly half positive words (e.g., joy and excitement) and half negative words (e.g., guilt and despair) constituted the category of emotion words. In the literature of word emotionality, valence has been shown a complex factor as evidenced by inconsistent reports with regard to its modulations of emotionally sensitive ERP components such as the EPN and the late positive component [@pone.0114421-Hinojosa1]--[@pone.0114421-Palazova1]. Therefore, it should be acknowledged that separate investigations of positive emotion words and negative emotion words would help to enhance our understanding about processing differences across word categories, and possibly to reveal further individual variations. Conclusions {#s5} =========== This study found gender difference in the level of early stage semantic processing. In addition, level of self-insight exhibited a differential relationship with word processing for men versus women. Generally speaking, for men, high insight appeared to be associated with a greater processing speed. For women, low insight appeared to be associated with a greater processing level. These relations were found to vary as a function of word categories with different semantic features, e.g., concreteness and emotionality, which may reflect representational differences across these word categories. Although many questions remain to be answered about the mechanism by which gender, self-insight, and semantic knowledge interact in word processing, the individual variations revealed in this study highlight the need to take into consideration subject variables when utilizing electrophysiological measures in related research. Acknowledgments {#s6} =============== The authors would like to thank Wuwei Fan, Jie Lin, and Haoyun Zhang for their assistance with data collection. We also sincerely thank Kathy Brode and Christopher Twilley for proofreading this manuscript. Supporting Information {#s7} ====================== ###### **Exemplar words for each type of experimental stimuli.** The abstract word shown here means logic in English, the concrete word teapot, the emotion word despair, and the animal word frog. (TIF) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: XX TG. Performed the experiments: CK. Analyzed the data: XX CK TG. Contributed reagents/materials/analysis tools: XX CK TG. Wrote the paper: XX CK TG.

Introduction {#sec1_1} ============ The resistance arteries (\<300 μm in internal diameter) are essential for mediating the autoregulation of flow and the stabilization of capillary pressure. An important component of this protective mechanism, called the myogenic response, allows these vessels to change diameter in response to alterations in intraluminar pressure. Besides resulting in enhanced myogenicity of the resistance arteries \[[@B1]\], sustained high blood pressure is associated with changes in cardiovascular structure. Resistance vessels develop a narrowed lumen and a thickened vascular wall, but without an apparent increase in cross-sectional area. This process, known as inward eutrophic remodeling, occurs as a response to prolonged (myogenic) constriction in order to protect the downstream vasculature, and it is thought to be an energetically favored mechanism to preserve a reduced lumen diameter for long periods. Such a structural adaptation involves a migratory process, whereby vascular smooth-muscle cells (VSMCs) reposition in the vascular wall \[[@B2]\], and it can be observed in most forms of hypertension including onset and milder hypertensive states. When hypertrophy is observed, the rise in blood pressure is often fulminant and severe. Autoregulatory mechanisms are then inadequate to normalize wall stress. Recently, it has been reported that the development of small-artery hypertrophy is an adverse prognostic sign \[[@B3]\]. The integrin/extracellular-matrix (ECM) axis transfers tensile forces exerted by blood pressure across the cell membrane. Integrins are cell surface ECM receptors that can heterodimerize in a noncovalent fashion to form 24 different αβ receptors which recognize a wide variety of cellular and ECM components present in the arterial extracellular space including collagen, fibronectin and laminin \[[@B4]\]. In resistance arteries, specific integrin subtypes are initially utilized for the mechanotransduction of pressure \[[@B5]\], while others mediate the migration of VSMCs towards a narrowed lumen \[[@B6],[@B7]\]. Engagement of integrins upon migration activates intracellular signaling complexes found at focal adhesion (FA) sites \[[@B8]\]. To date, there has been no conclusive evidence that FA sites, which contain active tyrosine kinases (e.g. focal adhesion kinase, FAK) and other cytoskeletal associated phosphoproteins observed in migrating cells in vitro \[[@B9]\], exist in vivo in arterial VSMCs. However, it is clear that the migratory process of arterial VSMCs in vivo is more subtle and is limited to the elongation of tapered cells and an increase in cell overlap \[[@B2]\] compared with the more pronounced movement of cells in vitro. Eutrophic remodeling is a relatively rapid process that is followed by the fixation of cells within the vascular ECM scaffold and it requires active transglutaminases \[[@B10]\]. Recently, we described a key role for αVβ3 integrin in eutrophic remodeling and other studies have described its function in myogenic constriction \[[@B6]\]. There is increasing evidence that FAK is an important mediator for cell attachment and mechanosensing \[[@B11]\]. The Src family kinases are also regarded as key components of mechanotransduction to special domains via the cytoskeleton when force is applied to fibronectin-binding integrins such as αVβ3 \[[@B12]\]. Moreover, FAK/Src association with specific integrin subunits seems to regulate particular cellular functions: for example, α5β1 appears to regulate the L-type calcium current, which mediates myogenic constriction via Src and FAK signaling \[[@B13]\], whereas the association of FAK/Src with β3 integrin subunits seems to be important for cell migration \[[@B14]\]. In addition, evidence is accumulating for the initiation of FAK phosphorylation in the growth response of arteries exposed to elevated intraluminal pressure \[[@B15]\]. We wished to test the hypothesis that FAK/Src signaling is crucial in mechanotransduction and the eutrophic inward remodeling of resistance arteries as a response to elevated pressure. This study examined the activation levels of FAK/Src in arteries exposed to elevated pressure in vitro and ex vivo, and we report differential FAK Y397 phosphorylation levels present at the integrin β subunits of resistance arteries from hypertensive TGR(mRen2)27 rats as pressure rises and eutrophic inward remodeling occurs. Ex vivo pressurization of arterial segments which exhibit myogenic tone confirmed that FAK Y397 phosphorylation is increased as a response to elevated pressure. Methods {#sec1_2} ======= Animals {#sec2_1} ------- The TGR(mRen2)27 rat carries a DBA/2J Ren2 transgene and develops hypertension with vascular structural changes from eutrophic inward remodeling \[[@B6],[@B16]\]. These animals, together with Sprague-Dawley (SD) normotensive control animals, were studied by immunohistochemistry and Western blot at 4 and 5 weeks of age. Ex vivo studies were performed on arteries harvested from SD rats at 5 weeks of age. On the day of study, rats were sacrificed by stunning and cervical dislocation. All animal procedures were performed in accordance with the UK Home Office Regulated Procedures on Living Animals (Scientific Procedures) Act 1986. Blood Pressure Measurement {#sec2_2} -------------------------- Blood pressure measurements of TGR(mRen2)27 and SD control animals were performed using tail-cuff plethysmography under light fluothane anesthesia. Vessel Morphology and Integrin Antagonism {#sec2_3} ----------------------------------------- We performed studies on the small arteries harvested from animals sacrificed at 5 weeks of age, because hypertension-induced eutrophic remodeling takes place from the age of 4 weeks and is completed by week 5, as described previously \[[@B6]\]. Segments of 2nd-order arteries (approx. 200 µm) were isolated from the proximal region of the mesenteric bed, mounted on a wire myograph and held in physiological salt solution (PSS; 119 mM NaCl, 4.7 mM KCl, 25 mM NaHCO~3~, 1.17 mM KH~2~PO~4~, 1.17 mM MgSO~4~, 0.026 mM EDTA, 1.6 mM CaCl~2~ and 5.5 mM glucose) at 37°C and gassed with 5% CO~2~ and 95% O~2~ to maintain a pH of 7.4. Subsequent morphological parameters of live vessels such as the media-to-lumen ratio, growth and remodeling indices were then measured and calculated, as previously described \[[@B17]\]. In order to study the effects of integrin αV antagonism, TGR(mRen2)27 rats were injected intraperitoneally at the age of 4 weeks with cRGDfV or cRADfV peptide (10 mg/kg; Calbiochem) twice daily for 5 days before sacrifice \[[@B6]\]. These act as an αV integrin-specific inhibitory peptide, thereby inhibiting integrin functions or inactive control peptide, respectively \[[@B18],[@B19],[@B20]\]. Arteries were then harvested for morphological measurements using wire myography. Phosphorylated FAK Y397 Western Blot Analysis {#sec2_4} --------------------------------------------- Western blot analysis was performed according to the methodology described by Laemmli \[[@B21]\]. Vessels were dissected and protein extracted on ice in radioimmunoprecipitation assay buffer containing phosphatase and protease inhibitors to prevent changes in the phosphorylation status of the proteins under investigation. Anti-phosphorylated (p)FAK Y397 antibody (1:1,000, 0.5 μg/ml; AB4803, Abcam, Nottingham, UK), with a total (pan)-FAK antibody as a control for loading (1:1,000, 1.0 µg/ml; AB1311, Abcam), was used to detect expression in 25 µg of total protein. pFAK Y397 expression levels were corrected for the amount of total (pan)-FAK. Densitometric analysis was performed on a BioRad-GS690 scanner. Integrin αV Immunoprecipitation {#sec2_5} ------------------------------- An amount of 5 μg of each antibody was used to precipitate integrin subunits from 250 μg of total resistance artery radioimmunoprecipitation assay extract. The antibodies used for immunoprecipitation were: β1 (AB1952), β3 (AB1932) and α5 (aba5b) integrin antibodies (Chemi-Con, Moses Lake, Wash., USA). Agarose IgG/protein A (Sigma-Aldrich, UK) was used to bind and identify coprecipitated complexes. Subsequent Western blot analysis was performed using FAK Y397 (1:1,000) antibody and Src Y418 (1:500, 0.5 μg/ml; Abcam). Protein A-HRP was used for chemiluminescent detection of coprecipitated proteins \[[@B22]\]. Ex vivo Pressure Arteriography {#sec2_6} ------------------------------ Arterial segments from the proximal region of the cremasteric artery from SD animals were dissected and mounted between two glass microcannulae in physiological salt solution in a pressure myograph (Living Systems Instrumentation, Burlington, Vt., USA). Experiments were carried out at 32°C. PP2 and PP3, i.e. a potent inhibitor of c-Src and an inactive analog (negative control) of PP2, respectively \[[@B23],[@B24]\], were used at 1 μM (Calbiochem, Cambridge, UK). Only arteries that exhibited pressure-mediated constriction were included in the study. After pressurization at either 60 or 120 mm Hg for 1 h, arteries were quickly fixed in ice-cold 4% paraformaldehyde solution, left overnight, washed in 70% ethanol and prepared for histology/immunofluorescence. Immunohistochemical preparations of arteries ex vivo and in vitro were the same. Immunofluorescence Localization {#sec2_7} ------------------------------- Vessels for immunofluorescence were dissected on ice and snap-fixed in ice-cold 4% paraformaldehyde. Paraffin-embedded arteries were sliced at 4 μm, dewaxed and rehydrated. The pFAK Y397 integrin (1:100; AB4803, Abcam) and a nonimmune control IgG antibody were used separately, followed by donkey anti-rabbit antibody with conjugated TRITC (1:200, 1 μg/ml; Jackson ImmunResearch Laboratories, West Grove, Pa., USA). Diamidino-2-phenylindole (DAPI, 1 pg/ml) (Jackson ImmunoResearch Laboratories) was used as a blue fluorescent nuclear stain. A Leica DM6000 epifluorescence microscope was used at ×1,000 magnification in conjunction with Leica FW4000 software for image capture and analysis. Superimposing fluorescence and phase-contrast microscopy images allowed the visualization of the VSMC membrane and the internal elastic lamina which indicates the smooth-muscle/endothelium boundary. Statistical Analysis {#sec2_8} -------------------- Statistical analysis was performed using MS Excel. Values are expressed as mean ± standard deviation. Differences between data were tested using the two-tailed unpaired homoscedastic Student t test. p \< 0.05 was considered statistically significant. Results {#sec1_3} ======= Blood Pressure {#sec2_9} -------------- Between 4 and 5 weeks of age, mean systolic blood pressure in TGR(mRen2)27 rats was significantly increased from 121 ± 7.9 to 148 ± 3.0 mm Hg (n ≥ 6, p \< 0.02). SD control rats showed no age-dependent increase in blood pressure which ranged from 80.9 ± 8.8 to 93.4 ± 8.1 mm Hg. Pressures in TGR(mRen2)27 rats were significantly higher than those in SD rats at both time points (p \< 0.0001). The administration of cRGDfV and cRADfV peptides had no effect on the blood pressure rise seen in TGR(mRen2)27 rats (data not shown). Vascular Morphology {#sec2_10} ------------------- The cross-sectional area of TGR(mRen2)27 rat arteries at 5 weeks was unchanged compared with at 4 weeks of age, with a remodeling index of 97% indicating the development of eutrophic inward remodeling due to the media-to-lumen ratio being significantly increased (p \< 0.01). The structural parameters of SD rat arteries did not change over 7 days. cRGDfV treatment reduced the remodeling index to 9% and there was evidence of hypertrophic growth, with a calculated growth index of 17%. Treatment with cRADfV had no effect on remodeling, as predicted (data not shown). Western Blot {#sec2_11} ------------ Western blot analysis of pFAK Y397 showed significant rises in autophosphorylation levels in resistance arteries from TGR(mRen2)27 animals treated with cRADfV control peptide (approx. 1.4-fold, p \< 0.05), compared with vessels from SD normotensive control animals (fig. [1a, b](#F1){ref-type="fig"}). Surprisingly, an increase in pFAK of approximately 2-fold was observed in TGR(mRen2)27 rat arteries treated with the active cRGDfV peptide (p \< 0.01) compared with vessels from normotensive animals treated with the same peptide (n ≥ 4). Relative pFAK Y397 levels, established by densitometry, were corrected by comparison to total (pan)-FAK (fig. [1a, b](#F1){ref-type="fig"}). Immunofluorescence {#sec2_12} ------------------ Phospho-FAK Y397 immunofluorescent staining of the arteries from normotensive SD animals was faint but present throughout the medial smooth-muscle cells (n ≥3; fig. [2b, d](#F2){ref-type="fig"}). The level of pFAK Y397 was observed to be increased in the VSMCs of TGR(mRen2)27 rat (n ≥ 3) resistance arteries (fig. [2a](#F2){ref-type="fig"}). Superimposing phase-contrast images onto fluorescent signals allowed the visualization of the VSMC medial layer, the internal elastic lamina and intense fluorescent pFAK Y397 regions of staining present in TGR(mRen2)27 rat VSMCs (fig. [2c](#F2){ref-type="fig"}). This is in contrast with fluorescence of total (nonphosphorylated) FAK (fig. [3a--d](#F3){ref-type="fig"}), which is expressed diffusely throughout the VSMCs of the medial layer (fig. [3c](#F3){ref-type="fig"}). FAK was localized to FA sites as identified by coimmunohistochemistry for FAK Y397 and vinculin, a membrane protein found associated with the FA sites of the elastic lamina (fig. [4](#F4){ref-type="fig"}). Control incubations with nonimmune IgG did not result in any detectable signal (fig. [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}). The treatment of animals with the cRGDfV peptide (10 mg/kg) showed localization of pFAK within the medial layer of VSMCs and more intense staining in the treated TGR(mRen)27 rat arteries than in those from SD rats (n ≥ 3; fig. [5a](#F5){ref-type="fig"}) and more diffuse than in untreated arteries (fig. [2a](#F2){ref-type="fig"}), thereby supporting quantified data from our observations from Western blot analysis (fig. [1](#F1){ref-type="fig"}). Figure [5](#F5){ref-type="fig"} a, c clearly shows intense Y397 staining is throughout the medial layer of arteries. In contrast, cRGDfV treatment of normotensive SD animals did not have any effect on pFAK Y397 levels (fig. [5b, d](#F5){ref-type="fig"}). Inspection of arterial segments undergoing FAK Y397 phosphorylation staining, which were pressurized at 120 mm Hg for 1 h (and then developed myogenic-tone ex vivo) exhibited increases in FAK Y397 phosphorylation compared with vessels pressurized at 60 mm Hg (n ≥ 5). No passive morphological alterations of vessels to pressure were observed at this stage (data not shown). Phosphorylation of FAK Y397 at 120 mm Hg was abrogated when arteries were coincubated with 1 μM of PP2. In contrast, coincubation of PP3 did not have any effect (fig. [6](#F6){ref-type="fig"}). Integrin Coimmunoprecipitation {#sec2_13} ------------------------------ pFAK Y397 was coprecipitated with integrin β3 and, to a lesser extent, with integrin α5β1 (fig. [7a](#F7){ref-type="fig"}) from the protein extracts of SD rat arteries. However, no FAK was coprecipitated with the general subpopulation of β1 integrins (fig. [7a](#F7){ref-type="fig"}). In contrast to the protein extracts of normotensive SD resistance arteries, pFAK Y397 coprecipitation with all the integrins, i.e. β1, β3 and α5β1, was observed in the arteries of TGR(mRen2)27 animals (fig. [7b](#F7){ref-type="fig"}). Coimmunoprecipitation of the protein extracts from resistance arteries of 5-week-old cRGDfV-treated TGR(mRen2)27 animals (which inhibited eutrophic inward remodeling but showed hypertrophy) did not show pFAK cobinding with β3 or α5β1 integrins, but interacted solely with other subpopulations of β1 integrins (fig. [7c](#F7){ref-type="fig"}). In conjunction with FAK, c-Src was prominently coprecipitated with integrin α5β1 and, to a lesser extent, with integrin β3 (fig. [8a](#F8){ref-type="fig"}). In contrast, c-Src is not coprecipitated with the α5β1 integrins of resistance arteries of hypertensive TGR(mRen2)27 animals, but with integrins β1 and β3 (fig. [8b](#F8){ref-type="fig"}). c-Src, however is not recruited to the β3 cytoplasmic domain when αVβ3 binding to the ECM with the cRGDfV peptide is inhibited (fig. [8c](#F8){ref-type="fig"}). Discussion {#sec1_4} ========== Previously, we showed that the hypertension-mediated inward eutrophic remodeling of TGR(mRen2)27 rat resistance arteries commenced at 4 weeks of age and was complete at 5 weeks \[[@B6]\]. This process was dependent on the integrin αVβ3 (the only VSMC β3 integrin) because administration of cRGDfV peptide, which preferentially interferes with the binding of this integrin-heterodimer to the ECM \[[@B18],[@B19],[@B20],[@B25]\], prevented the remodeling process but enhanced hypertrophy. We also investigated the localization of FAK and its activation status in the smooth-muscle cells of TGR(mRen2)27 rat resistance arteries and determined its association with β1 or β3 integrin subunits. We established that low basal levels of FAK Y397 phosphorylation in normotensive SD rat mesenteric arteries are predominantly associated with the β3 integrin subunit and that active c-Src associates with integrin α5β1. FAK Y397 phosphorylation was significantly increased in hypertensive TGR(mRen2)27 resistance arteries and recruited to FA sites after blood pressure rose and once eutrophic inward remodeling was complete. In addition, both pFAK Y397 and active c-Src associated with β1 and β3 integrins in the resistance arteries of TGR(mRen2)27 animals. However, the inhibition of remodeling by the administration of cRGDfV peptides resulted in the hypertrophy of VSMCs and a further increase of FAK Y397 phosphorylation. This process was associated with integrins containing the β1 subunit only. We can conclude from these data that FAK phosphorylated at tyrosine position 397 (the autophosphorylation site) is present at both the integrin β1 and β3 cytoplasmic sites in the resistance arteries of hypertensive TGR(mRen2)27 animals, in contrast to SD controls, which exhibit relatively low pFAK Y397 levels only associated with the integrin β3 subunit. We cannot rule out the possible involvement of FAK Y397 phosphorylation in subsequent length/autoregulation processes during the hypertension remodeling processes downstream of the integrin β3 subunit. However, it is now thought that the initial role, in the light of the FAK Y397 phosphorylation observed, is in mechanotransduction, as cRGDfV peptides not only inhibited eutrophic remodeling but also encouraged further FAK Y397 phosphorylation during the onset of hypertension in this model. Therefore, the increase of FAK Y397 phosphorylation observed is more likely to occur as a consequence of increases in wall tension, rather than acting as a direct intermediate of eutrophic inward remodeling. Besides aiding adhesion and migration \[[@B26]\], β1 and β3 integrins are both responsible for the intracellular transfer of these forces in the VSMCs of resistance arteries \[[@B27],[@B28]\]. This conclusion is strengthened further by how the arterial segments of the cremaster muscle, at a high pressure (120 mm Hg), exhibit a prompt induction of FAK Y397 phosphorylation (\<1 h) when compared with vessels pressurized at much lower levels and completely independent of neurohumoral influences \[[@B29]\]. There is the possibility that the increase of FAK Y397 phosphorylation observed in the resistance arteries of the young TGR(mRen2)27 hypertensive animals was induced by enhanced activity of the renin-angiotensin system (RAS; reviewed in \[[@B30]\]). However, this seems unlikely because at 5 weeks of age the resistance arteries of these animals have eutrophically remodeled without evidence of hypertrophy \[[@B6]\], similar to the morphological adaptations observed in a low-renin model of hypertension \[[@B31]\] and the BMH-2 mouse model which develops hypertension independently of the RAS \[[@B32]\]. The hallmark of a locally activated vascular RAS in older animals is vascular hypertrophy \[[@B30],[@B33]\]. Again, ex vivo pressurization has determined that FAK Y397 phosphorylation is likely to be associated with an increase in pressure, in an Src-dependent manner, without the possible interference of neurohumoral influences. However, we cannot completely rule out the contribution of RAS activity to FAK Y397 phosphorylation in the TGR(mRen2)27 resistance arteries \[[@B34],[@B35]\]. Various studies have reported the association of FAK with different β integrin subunits; β1, β3 and β5 cytoplasmic domains all contain sufficient information to trigger FAK phosphorylation and these specific regions have been identified \[[@B36]\]. Different cellular systems in vitro utilize different integrins for FAK activation, e.g. β3-but not β1-integrin engagement in cultured cardiomyocytes and VSMCs is accompanied by FAK activation in FAs \[[@B14],[@B37]\]. In contrast, fibroblasts cultured on collagen employ β1 integrins only for FAK activation \[[@B38]\]. In response to the hepatocyte growth factor, VSMCs are known to utilize both β1 and β3 integrins for FAK activation and subsequent migration \[[@B39]\]. In this study, using the TGR(mRen2)27 rat model \[[@B16]\], it became apparent that both β1 and β3 integrins are involved in VSMC FAK recruitment/signaling when pressure becomes elevated, in contrast to normotensive controls. However, the FAK Y397 phosphorylation observed which is associated with β-integrin subunits in response to elevated pressure is not exclusively increased when eutrophic remodeling processes take place, but is already apparent at the initial stages when pressure becomes elevated. Clinical Perspectives {#sec1_5} ===================== It has been established that both β1 and β3 integrin subtypes are required not only in the adhesion of VSMCs to the ECM of the vascular wall but also for pressure-mediated signaling, resulting in vascular myogenic constriction and remodeling in hypertension \[[@B6],[@B27]\] (summarized in fig. [9](#F9){ref-type="fig"}). In this study, we identified FAK Y397 phosphorylation as an early event as a result of elevated pressure in vivo and ex vivo. The identification of FAK Y397 phosphorylation in an Src-dependent manner is an important first step in resolving the complex signaling cascades that underlie integrin-mediated vascular adaptations of resistance arteries in hypertension. The relevance of unraveling molecular components in specific signaling events, in the context of pressure mechanotransduction in hypertension, will enlighten the design of pharmacological agents to prevent hypertension-mediated target-organ damage. Recent data have suggested that the morphological changes in the arterial wall of small blood vessels are associated with cardiovascular prognosis. The normal response to untreated hypertension is eutrophic inward remodeling, i.e. a rearrangement of the vascular wall architecture without any evidence of growth \[[@B17]\]. When hypertrophy supervenes, this heralds the breakdown of normal homeostatic mechanisms such as autoregulation and, as such, there is clear evidence that this is the structural alteration associated with an adverse prognosis \[[@B3]\]. This study was funded by the Wellcome Trust. We thank Mrs. Maureen Speed for her expert secretarial assistance. {#F1} {#F2} {#F3} {#F4} {#F5} {#F6} {#F7} {#F8} {#F9}

Introduction ============ Human performance is affected by both the genetic makeup of the individual and the environmental factors. Current research and interest in sports genetics focus on genetic variants that may make a significant contribution to the individual's performance. It is known that personal traits such as endurance, strength, power, muscular coordination, and psychological willingness and motivation, all have a genetic background.[@b1-geg-7-2015-001] The most frequently investigated genetic polymorphisms in terms of athletic performance or predisposition to athletic capacity are angiotensin-1 converting enzyme (ACE) gene and α-actinin-3 (ACTN3) gene.[@b2-geg-7-2015-001] Variants in these genes have been reported to be associated with elite athletic performance and with quantitative physical performance traits in the general population.[@b3-geg-7-2015-001]--[@b5-geg-7-2015-001] *ACTN3* was the first structural gene specific to skeletal muscle that has been associated with athletic performance.[@b6-geg-7-2015-001] The actinins, major and important Z line functional and structural proteins in sarcomers, are members of the actin-binding protein family.[@b7-geg-7-2015-001] *ACTN3* codes for ACTN3 in humans, and its expression is restricted to fast, type 2 fibers.[@b8-geg-7-2015-001] North et al[@b9-geg-7-2015-001] reported that a deficiency of α-actinin protein because of a premature stop codon in *ACTN3* alters the designation for the amino acid arginine to a stop codon at position 577 (R577X; dbSNP rs1815739) in exon 16 and may contribute to the possibility of differential fitness. The *ACTN3* 577R allele and 577RR genotype are associated with top-level, power-orientated athletic performance in a wide array of ethnic groups.[@b10-geg-7-2015-001] The *ACTN3* R577X polymorphism has been reported to be associated with elite athletic status,[@b6-geg-7-2015-001] endurance athletes,[@b11-geg-7-2015-001] and many other groups.[@b12-geg-7-2015-001]--[@b16-geg-7-2015-001] *ACE* is located on chromosome 17q23 and comprises 26 exons and 25 introns. It contains a polymorphism due to an insertion (I) or a deletion (D) of a 287 bp Alu sequence in intron 16, resulting in the three genotypes of insertion/insertion (II), insertion/deletion (ID), and deletion/deletion (DD).[@b17-geg-7-2015-001] The I/D polymorphism is associated with circulating and tissue ACE levels. Individuals homozygous for the D allele had higher tissue and plasma ACE concentrations than heterozygotes and II homozygotes.[@b18-geg-7-2015-001] Many of the case--control studies reported that success in speed-strength disciplines, such as short-distance running, long jump, high jump, and short-distance swimmers, is associated with *ACE* DD genotype.[@b19-geg-7-2015-001],[@b20-geg-7-2015-001] On the other hand, individuals with the II genotype have a lower ACE serum concentration and have more success in endurance-related disciplines such as medium- and long-distance running, race walking, and rowing.[@b21-geg-7-2015-001],[@b22-geg-7-2015-001] Two different studies involving Spanish and Lithuanian football players showed that players had a significantly higher percentage of the *ACE* ID genotype when compared to the nonathletic population.[@b23-geg-7-2015-001],[@b24-geg-7-2015-001] Previous studies, including related genetic polymorphisms and football players, are very limited, especially in Turkish subjects. The aim of this study was to determine the genotype and allele distribution of *ACTN3* and *ACE* in Turkish male football players and assess the impact on predisposition to football. Materials and Methods ===================== Study subjects -------------- All players enrolled for the study had Turkish ancestry and were members of a Turkish football team. The study was conducted in accordance with the principles of the Declaration of Helsinki II. Üsküdar University Ethical Committee approved the study protocol, and written informed consent was obtained from each player once prospective participants understood and accepted the aim and protocol of the study. Genotyping ---------- DNA isolation was carried out by using High Pure Polymerase Chain Reaction Template Preparation Kit (Roche Diagnostics) using peripheral blood. The region of interest, *ACTN3*, was amplified using the following primers: forward 5′-CTG TTG CCT GTG GTA AGT GGG-3′ and reverse 5′-TGG TCA CAG TAT GCA GGA GGG-3′, as described previously.[@b5-geg-7-2015-001] Polymerase chain reaction (PCR) was performed by initial denaturation at 95°C for five minutes, followed by 35 cycles of denaturation at 95°C for 30 seconds, annealing and extension at 72°C for one minute, and a final extension for seven minutes at 72°C. Genotyping of the *ACTN3* R577X was maintained by restriction length polymorphism method. The 290-bp amplicons were digested by DdeI (New England Biolabs) as recommended by the manufacturer. Digested fragments were separated on 10.0% polyacrylamide gel electrophoresis and visualized under UV light by ethidium bromide staining. The wild-type allele, 577R, showed fragments of 205 and 85 bp, whereas the variant allele, 577X, showed fragments of 108, 97, and 85 bp ([Fig. 1](#f1-geg-7-2015-001){ref-type="fig"}). For the *ACE* genotype, conventional PCR amplifications were carried out. Primers 5′-CTGGAGACCAC TCCCATCCTTTCT-3′ and 5′-GATGTGGCCATCACAT TCGTCAGAT-3′ were used for the amplification. The final volume of the PCR mixture was 50 μL and contained 50--100 ng genomic DNA, 1 mM of each primer, 50 mM KCl, 1 mM deoxynucleotide triphosphate (dNTP), 1.5 mM MgCl~2~, 10 mM Tris--HCl, pH 8.0, and 1 U Taq DNA polymerase. An initial denaturation at 94°C for five minutes was followed by annealing at 58°C for one minute and extension at 72°C for two minutes. Amplification was finalized with 30 cycles: denaturation at 94°C for one minute, annealing at 58°C for one minute, and extension at 72°C for two minutes, followed by a final elongation at 94°C for one minute, annealing at 58°C for one minute, and extension at 72°C for seven minutes. Amplicons were separated by electrophoresis on a 2% agarose gel and visualized under UV light after ethidium bromide staining. Electrophoresis gave rise to three possible patterns: a 490-bp band (II genotype), a 190-bp band (DD genotype), or both 490- and 190-bp bands (I/D genotype) ([Fig. 2](#f2-geg-7-2015-001){ref-type="fig"}). Amplicons had 190 bp in the presence of the D allele and 490-bp fragment in the presence of the I allele. Results ======= The percentage of the *ACE* genotype in the examined players was 16, 44, and 40 for II, ID, and DD genotypes, respectively. For the *ACTN3* genotype, the respective frequencies were 20, 36, and 44 for XX, RX, and RR. An allelic count gave rise to 19 (38%) I and 31 (62%) D alleles for *ACE* and 31 (62%) R and 19 (38%) X alleles for *ACTN3*. [Table 1](#t1-geg-7-2015-001){ref-type="table"} lists the genotype and allelic frequencies of *ACE* and *ACTN3*. According to the genotypes, nine different combinations were found, five of the players had DD + RR, the same number of the players had ID + RR, four players had ID + RX, three players had DD + RX, two players had DD + XX, the same number of the players had ID + XX, two players had II + RX, one player had II + XX, and one player had II + RR (data not shown in table). Discussion ========== Recent studies have indicated that several genes are involved in determining the performance of the players, both physiologically and psychologically.[@b25-geg-7-2015-001] Genetic models could be developed and used to find the optimal genetic endowment of a player to help scientists establish which genetic polymorphisms are advantageous for proper performance in different sports types. Therefore, the creation of genomic databases will be very useful for sport scientists. In this study, we solely examined 25 senior football players in terms of *ACE* and *ACTN3* polymorphisms and aimed to associate these polymorphisms with predisposition to football. In our study cohort, the RR genotype for *ACTN3* genotype was higher than XX and RX. The R allele, considered to be the wild-type allele of the gene and associated with the rapid contraction of the sarcomers, was also high when compared to X allele. Egorova et al[@b26-geg-7-2015-001] reported similar results in Russian football players; they examined 240 football players, and 46.25% had the RR genotype. Santiago et al[@b16-geg-7-2015-001] analyzed 60 Brazilian football players, and in their cohort, 48.3% of the players had RR genotype. Pimenta et al[@b27-geg-7-2015-001] aimed to compare acute inflammatory responses, muscle damage, and hormonal variations with eccentric training in soccer athletes and examined 37 professional soccer players of which 40.5% were RR. To the best of our knowledge, this is the first report in Turkish male professional soccer players, and our results were in agreement with previous studies, indicating R allele and RR genotype superiority for soccer players. In our cohort, 44% of the players had ID genotype, 40% of the players had DD genotypes for *ACE*, and 84% of the players had at least one D allele. The allelic count revealed D allele as 62% in players. D allele is responsible for high ACE concentration and is associated with success in speed-strength disciplines.[@b17-geg-7-2015-001] Gineviciene et al[@b24-geg-7-2015-001] examined 199 Lithuanian football players and reported similar results to ours. They showed that ID was the highest genotype (46.7%), and together with DD genotype, the percentage of the players having at least one D allele was 76.3%. Juffer et al[@b23-geg-7-2015-001] analyzed 54 male professional soccer players, and ID was the most detected genotype. Unlike our results, Egorova et al[@b26-geg-7-2015-001] examined 213 Russian football players and found the frequency of ID and DD genotypes as 28.6% and 50.7%, respectively, giving both genotypes an overall frequency of 79.3%, which is similar to our results. The cumulative effect of genotypes in human metabolism had always received great attention. In an attempt to find the optimal genotypes, several researchers have examined candidate genes separately, in different populations, to create a pool of genomic data. Genotype combinations are very useful for identifying a certain metabolism, or a part of a metabolism, if they are part of the same reaction. For example, *ACTN3* and *ACE* have crucial roles for their metabolisms, but they do not have roles in the same biologic pathway. But it is important to understand performance-enhancing polymorphisms to create an optimal genomic score for being an elite athlete. In our cohort, five of the players had the DD + RR genotype combinations and the same number had ID + RR genotypes. The D and R alleles were found together in these genotypes, the most important similarity between these genotypes. On the other hand, one player had the II + XX genotype, thought to be associated with endurance capacity. The number of the subjects examined in the current study is the main limitation. We hope this preliminary study will guide new studies, with extended numbers of players in different kinds of sports to evaluate the effect of the combination of these polymorphisms more accurately. Conclusion ========== In this study, we examined the distribution of *ACE* and *ACTN3* polymorphisms in male Turkish professional soccer players for the first time. *ACTN3* RR genotype and R allele and *ACE* ID genotype and D allele dominated our cohort. These polymorphisms are well known and have been examined in different populations. Therefore, we did not compare the genomic results with sedentary people; rather, we examined the combined effect of these polymorphisms. According to our previous and current results, we hypothesize that elite soccer players tend to have a power/strength-oriented genotype. These polymorphisms, alone or in combination with the additional polymorphisms, should be taken into account when deciding a genomic score profile for success in sports. **ACADEMIC EDITOR:** Christian Bronner, Editor in Chief **PEER REVIEW:** Four peer reviewers contributed to the peer review report. Reviewers' reports totaled 526 words, excluding any confidential comments to the academic editor. **FUNDING:** Authors disclose no funding sources. **COMPETING INTERESTS:** Authors disclose no potential conflicts of interest. Paper subject to independent expert blind peer review. All editorial decisions made by independent academic editor. Upon submission manuscript was subject to anti-plagiarism scanning. Prior to publication all authors have given signed confirmation of agreement to article publication and compliance with all applicable ethical and legal requirements, including the accuracy of author and contributor information, disclosure of competing interests and funding sources, compliance with ethical requirements relating to human and animal study participants, and compliance with any copyright requirements of third parties. This journal is a member of the Committee on Publication Ethics (COPE). Provenance: the authors were invited to submit this paper. **Author Contributions** Conceived and designed the experiments: KU. Analyzed the data: CS, TB. Wrote the first draft of the manuscript: KU. Contributed to the writing of the manuscript: CS. Agree with manuscript results and conclusions: KU, CS, TB. Jointly developed the structure and arguments for the paper: KU, CS. Made critical revisions and approved final version: All authors. All authors reviewed and approved of the final manuscript. {#f1-geg-7-2015-001} {#f2-geg-7-2015-001} ###### Genotypic and allelic distribution of the *ACE* and *ACTN3* in the examined players. GENOTYPE ALLELE FREQUENCY ------------------ -------- ---------- ------------------ -------- ------- ------- *ACE* (n = 25) Number 4 11 10 19 31 Percentage (%) 16 44 40 38 62 **XX** **RX** **RR** **X** **R** *ACTN3* (n = 25) Number 5 9 11 19 31 Percentage (%) 20 36 44 38 62

ANNOUNCEMENT {#s1} ============ Cryptosporidium parvum is one of the two major species of the protozoan parasite Cryptosporidium that infect humans and cause gastrointestinal disease ([@B1]). C. parvum is considered zoonotic and has a wide host range, in which cattle and small ruminants are the predominant reservoir hosts ([@B2]). However, some subtypes of C. parvum, such as the IIc gp60 subgroup, are considered to be human adapted ([@B2]), because they are mainly reported in humans, with only a few reports in European hedgehogs ([@B3]). Whole-genome sequencing (WGS) provides a means for comparing isolates to identify markers important in distinguishing routes of transmission and potential virulence traits for better epidemiological analysis and risk assessment. The objective of this work was to sequence a zoonotic isolate of C. parvum. We sequenced a human isolate of C. parvum, UKP1, isolated at the Public Health England and Public Health Wales Cryptosporidium Reference Unit and identified as gp60 type IIaA17G1R1 (GenBank accession no. [JX971701](https://www.ncbi.nlm.nih.gov/nuccore/JX971701)). This subtype has a global distribution and has been observed in humans, cattle, pigs, and sheep ([@B4], [@B5]). Zoonotic linkages with cases in England and Wales have been demonstrated with IIaA17G1R1 (4). Oocysts were semipurified from stool using a salt flotation and hypochlorite treatment. DNA was extracted with the QIAamp DNA extraction kit (Qiagen, Hilden, Germany) and whole-genome amplified with the REPLI-g kit (Qiagen) before 454 GS FLX Titanium and Illumina HiSeq 2500 sequencing. A total of 472.6 Mbp, representing 1.6 million reads, were obtained from the 454 GS FLX Titanium sequencing, and 12.8 Gbp, representing 26 million reads, were produced on an Illumina MiSeq instrument. The quality of the reads was examined using FastQC ([@B6]). Illumina and 454 reads were mapped against a reference C. parvum isolate (GenBank accession no. [NZ_AAEE00000000](https://www.ncbi.nlm.nih.gov/nuccore/NZ_AAEE00000000)) using Bowtie 2 v. 2.3.3.1 ([@B7]), and the mapped reads were then *de novo* assembled using SPAdes v. 3.11.0 ([@B8]). The initial assembly was used in a second round of assembly using the \--trusted-contigs flag with SPAdes. Iterative polishing of the assembly was done by mapping the reads back to the assembly with Bowtie 2 and correcting them with Pilon 1.22 ([@B9]). The 8 C. parvum chromosomes assembled into 14 contigs with a total genome size of 8,881,956 bp, a G+C content of 30.20%, an *N*~50~ value of 1,092,230 bp, and a largest contig length of 1,333,759 bp. Data availability. {#s1.1} ------------------ This whole-genome shotgun project has been deposited at DDBJ/EMBL/GenBank under the accession no. [PYCJ00000000](https://www.ncbi.nlm.nih.gov/nuccore/PYCJ00000000), and raw sequence reads are available under the BioProject no. [PRJNA439211](https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA439211). We are grateful to Brian Boyle (Genomics Sequencing Platform, IBIS, Université de Laval, Quebec City, Quebec, Canada) for whole-genome amplification and generating the 454 and Illumina sequence data. Funding for this work was provided by the Public Health Agency of Canada. [^1]: **Citation** Nash JHE, Robertson J, Elwin K, Chalmers RA, Kropinski AM, Guy RA. 2018. Draft genome assembly of a potentially zoonotic *Cryptosporidium parvum* isolate, UKP1. Microbiol Resour Announc 7:e01291-18. <https://doi.org/10.1128/MRA.01291-18>.

Introduction {#sec1} ============ Gamma-delta T cells represent a small subset of normal human T cells that possess a distinct T-cell receptor (TCR) on their surface. Their exact function has not yet been determined, but these cells are known to play a role in both the adaptive and innate immune system.[@bib1] The World Health Organization currently recognizes 2 subtypes of γδ T-cell lymphoma, primary cutaneous γδ T-cell lymphoma and hepatosplenic γδ T-cell lymphoma,[@bib2] both of which are rare. Primary cutaneous γδ T-cell lymphoma has an aggressive clinical behavior and poor prognosis, with a meager response to multiagent chemotherapy. However, with the advent of detecting TCR-γδ in paraffin-embedded sections, a few reports have shown examples of indolent γδ T-cell proliferations within the skin, namely in pityriasis lichenoides (PL) and lymphomatoid papulosis (LyP).[@bib3], [@bib4], [@bib5] In contrast to primary cutaneous γδ T-cell lymphoma, a retrospective study of TCR-γδ expression in cases of PL and LyP found that such cases have a benign clinical course, with most patients achieving eventual resolution of the lesions and none having lymphoma.[@bib4] Here we describe such an example, with a rare case of γδ LyP type D histopathologically mimicking primary cutaneous γδ T-cell lymphoma but having an indolent clinical course. This case further highlights the expanding classification scheme of LyP and aims to bring awareness to this newly described entity to allow distinction from primary cutaneous γδ T-cell lymphoma. Case report {#sec2} =========== An otherwise healthy 46-year-old white woman presented with a 3-month history of a rash, primarily on the extremities. The lesions began as erythematous papules, which eventually ulcerated and then self-resolved. The patient denied pruritus or tenderness associated with the lesions and had no systemic symptoms. She was initially treated with a 3-week course of prednisone with mild improvement. Outside biopsy found a markedly atypical lymphoid infiltrate composed of T cells that were weakly CD8^+^, positive for granzyme and TIA-1, and TCR-βF^-^. The cells also expressed TCR-γ and TCR-δ, leading to a diagnosis of primary cutaneous γδ T-cell lymphoma. The patient was subsequently referred to our institution for further workup. At the time of presentation to our clinic, most of her lesions had resolved with only a few erythematous papules remaining on the forearms and lower legs ([Fig 1](#fig1){ref-type="fig"}).Fig 1Clinical presentation of a patient with γδ lymphomatoid papulosis type D. Small erythematous papules involving the extremities. The lesions are asymptomatic, last approximately 1 month and then self-resolve. Repeat biopsy of the left forearm found an atypical epidermotropic lymphoid infiltrate expressing CD2, CD3, CD7, CD8, TIA-1, CD30, diminished CD5, TCR-γ, TCR-δ, and lack of CD4 and βF1 ([Fig 2](#fig2){ref-type="fig"}, *A-C*). Complete blood count, metabolic panel, and lactate dehydrogenase were significant only for mildly elevated liver function. Flow cytometry showed 17% of the total T-cell population expressing TCR-γδ, with an absolute count of 221 cell/μL. The immunophenotype of the γδ TCR expressing cells was not aberrant: CD2^+^, CD3^+^, CD5^+^ (diminished), CD7^+^, CD8^+^ (diminished minor subset), CD16^+^ (diminished), CD45^+^, CD56^+^ (diminished), CD57^+^ (subset), and TCR-γδ. They were CD4^-^, CD19^-^, CD26^-^, CD38^-^, and negative for TCR-αβ. After review with the hematology/oncology department, this minor increase in γδ T cells was deemed physiologic. Peripheral blood smear showed only rare atypical large granular lymphocytes. Polymerase chain reaction for T-cell receptor gene rearrangement performed on the peripheral blood was negative for any clonal *TRG* gene rearrangement or T-cell clones. Put together, these findings ruled out peripheral involvement of a lymphoproliferative process. The patient declined systemic treatment. She continues to get intermittent, small, erythematous papules that last approximately 1 month and then self-resolve without treatment. She follows regularly with the hematology/oncology and dermatology departments and will have repeat testing for complete blood count with differential, metabolic panel, lactate dehydrogenase and leukemia/lymphoma immunophenotyping annually. At 10-month follow-up, the patient had no evidence of systemic disease.Fig 2Histopathologic findings in γδ lymphomatoid papulosis, type D. Biopsy reveals epidermotropic enlarged and atypical lymphoid cells (**A**, Hematoxylin-eosin stain; original magnification: ×200). These cells express TCR gamma (**B**, ×200) and TCR delta (**C**, ×200). Discussion {#sec3} ========== Primary cutaneous γδ T-cell lymphoma accounts for less than 1% of all cutaneous T-cell lymphomas. Patients typically experience extranodal and nodal dissemination, with 5-year overall survival rates ranging from 11% to 15%. This lymphoma typically presents with plaques, tumors, ulcers, or panniculitis. Histologically, the lesions are typically composed of atypical medium-to-large sized lymphocytes expressing CD3 and CD2, with variable expression of C7 and CD56, and strong positivity for cytotoxic proteins such as granzyme B, perforin, and TIA-1. CD8 positivity is seen rarely, but most cases are negative for both CD4 and CD8. The T cells express the γδ receptor by definition and are negative for TCR-βF1.[@bib6], [@bib7], [@bib8] Our patient showed these histopathologic features; however, the atypical lymphocytes showed epidermotropism and were CD30^+^ and CD8^+^. Epidermotropism and CD30 positivity are common in LyP type D but would typically not occur in γδ T-cell lymphoma. The complete histologic picture as well as the clinical course of relapsing and remitting erythematous papules and lack of systemic symptoms led to the ultimate diagnosis of LyP, specifically LyP type D. LyP is a chronic, recurrent, CD30^+^ lymphoproliferative disorder with various histopathologic subtypes (A-F), described. The diagnosis of LyP can be histologically challenging because of its variability and overlapping features of other cutaneous lymphomas. Clinical correlation is crucial for diagnosis and appropriate management. Type D is a rare subtype of LyP, typically composed of CD8^+^ T cells, which can histologically mimic aggressive epidermotropic CD8^+^ cytotoxic T-cell lymphoma.[@bib9], [@bib10] Although LyP type D is rare, the expression of TCR-γδ is even more uncommon. This type of LyP was first reported by Morimura et al[@bib3] in 2011, and subsequently TCR-γδ expression has been found in a few additional cases of PL and LyP.[@bib3], [@bib4], [@bib5] To date, all reported cases have shown an indolent course (follow-up ranging from 3-64 months). With advancing immunohistochemistry and gene rearrangement techniques, more cases of γδ LyP type D will likely emerge. Clinical awareness of this indolent entity is prudent given the poor prognosis of primary cutaneous γδ T-cell lymphoma. Unlike patients with primary cutaneous γδ T-cell lymphoma, patients with LyP do well and do not require aggressive intervention. However, as with other cases of LyP, patients require long-term follow-up and should be routinely screened for other systemic lymphomas. Although histologically similar, the clinical behavior of γδ LyP type D differs greatly from that of primary cutaneous γδ T-cell lymphoma. This newly described entity should be kept in mind, as clinicopathologic correlation is necessary for correct diagnosis. The authors kindly thank Dr Geraldine S. Pinkus for the TCR-γ and -δ immunohistochemical studies used in this report. Funding sources: None. Conflicts of interest: None disclosed.

Friedrich Nietzsche once wrote, "That which does not kill us makes us stronger."[@bb0005] The purpose of this article is to apply this philosophy to infectious and communicable diseases. In the recent scramble for Ebola virus readiness, health care providers adopted drastic measures to protect patients and staff. Using hindsight, what lessons can be extrapolated for other infectious and communicable diseases within the context of emergency nursing? Moreover, what are ways in which emergency departments might prepare for future pandemics? Infectious and communicable diseases are not new. The current generation of emergency nurses has adapted to human immunodeficiency virus, severe acute respiratory syndrome, avian flu, annual influenzas, Ebola virus, and other microorganisms. However, one newer and confounding variable is the speed at which remote germs are urbanized. Specifically, more humans spread germs at faster rates, as demonstrated in the [Figure](#f0005){ref-type="fig"} .[@bb0010] We have entered an era in which risks are higher and everyone has more to lose. This opportunity for reflection is one we cannot afford to ignore.FigureGrowing populations and shorter travel time has urbanized remote infectious diseases at faster rates. Modified From Rodrigue J-P. The geography of transport systems. Retrieved from <https://people.hofstra.edu/geotrans/eng/ch1en/conc1en/circumnavigation.html>, and from Tverberg G. World energy consumption since 1820 in charts. Retrieved from [http://gailtheactuary.files.wordpress.com/2012/03/world-population-1820-to-2010.png](https://people.hofstra.edu/geotrans/eng/ch1en/conc1en/circumnavigation.html){#ir0895} . Accessed January 12, 2014. Early Ebola virus response plans were developed under chaotic circumstances. These circumstances took a toll on scarce hospital and human resources. Pandemics are forecasted to appear at faster rates, and thus time for preparation between pandemics will grow shorter as a result. Emergency departments need a proactive framework or formula from which to stage and mobilize future responses. The following sections describe "5 S's" (screening, segregating, suiting, sharing appropriate information, and sanitizing) to consider during pandemics to prioritize and solidify safety. Screening {#s0005} ========= Patients suspected of having a disease may not exhibit a hallmark symptom or may be in denial. In 13% to 25% of Ebola cases, patients have been afebrile. A lesson learned was that emergency departments should screen for more disease symptoms than the one "hallmark Ebola symptom" of fever.[@bb0015] Rigor must be a priority when constructing screening questions. Without validity and reliability of screening instruments, one breach in screening can expose more ED patients and staff than one breach in personal protective equipment (PPE) or PPE donning and doffing procedures. Segregating {#s0010} =========== Privacy screens and physical barriers have limitations. Negative flow rooms and private toilets[@bb0020] are recommended but may not be available. However, a designated anteroom is recommended to ensure proper PPE staging and buddy monitoring.[@bb0025], [@bb0030] Having a designated anteroom provides for cold, warm, and hot areas and pass-through of supplies. Extrapolations from the construction industry are helpful. Consider interdisciplinary walkthroughs, such as those done prior to remodeling. Emergency departments should have access to portable high-efficiency particulate air (HEPA) filter machines[@bb0035] to counteract droplets when aerosolization is a concern. HEPA filtering should be considered for prescreening areas, waiting rooms, ambulance entries, or other common areas to protect staff and other patients. Suiting {#s0015} ======= Staff input and brainstorming into the PPE process is important. Not every employee is suited or built for frontline care. Hospitals were advised to solicit volunteers for Ebola response teams and containment unit patient care. Ebola caregivers must be highly engaged and devoted to strict isolation techniques.[@bb0040] Visual displays of the equipment and the suiting process should be posted. Detailed checklists should be provided for buddies or monitors so they can verbalize each step and initial the checklist once it is completed. Lists of all staff or visitors and the time of entry and exit should be kept. Staff should have the option of wearing the N-95 mask inside the powered air purifying respirator. Although outer hose surfaces are regularly decontaminated, it is important that inside hose surfaces be cleaned between uses and that training regarding cleaning be provided. For example, the literature shows that shared opium and narcotic pipes are a vehicle for transmission of tuberculosis, syphilis, and other communicable pathogens.[@bb0045] Sharing Appropriate Information {#s0020} =============================== The communications process should be streamlined to remove chaos. It is advisable to create "speed dial" access to your institution's infection prevention or infectious disease department and local or state departments of health so they can be contacted quickly if potential cases are identified. Distribution lists should be created so a group page, text, or e-mail message can be sent via emergency response software. This step empowers staff to focus on patients rather than spending time guessing who should be called. Sanitizing {#s0025} ========== When in doubt, risk versus benefit principles should be used.[@bb0050] When built-in ED bathrooms are not present, commode chairs are the next best option. However, whenever commode chairs are used, disposable commode pails are the recommended standard of care for bedside toileting,[@bb0055] with use of strict wipe-down techniques. Manipulation of infected waste increases microbe shedding and aerosolization. In addition to red bag use, triple bagging of infected waste is preferred compared with double bagging. Conclusion {#s0030} ========== The speed at which remote germs are urbanized has complicated ED infection control processes. The need for the use of standard precautions remains a fundamental practice in emergency nursing and a high priority to reduce the transmission of infectious and communicable diseases.[@bb0060] ED staff should prepare for infectious and communicable disease pandemics by using a safe, structured, and prioritized approach such as the 5 S's: screening, suiting, segregating, sharing appropriate information, and sanitizing. Molly B. Delaney, *Member, Twin Cities Chapter,* is ED Nurse Manager, East Bank, University of Minnesota Medical Center, Minneapolis, MN. Laura Reed is Chief Nurse Executive, University of Minnesota Medical Center, Minneapolis, MN. **Section Editors:** Andrew D. Harding, MS, RN, CEN, NEA-BC, FACHE, FAHA, FAEN, and Kathryn C. Whalen, DNP, MSN, RN, FAHA **Submissions** to this column are encouraged and may be sent to **Andrew D. Harding, MS, RN, CEN, NEA-BC, FACHE, FAHA, FAEN**<ADHardingRN@gmail.com> or **Kathryn C. Whalen, DNP, MSN, RN, FAHA**<katewhalenrn@aol.com>

Child undernutrition in all its forms is widespread in much of the developing world. Among children less than 5 years of age, 165 million are stunted, 51 million are wasted, 90 million are subclinically deficient in vitamin A and 18 % suffer from Fe-deficiency anaemia (IDA)^(^ [@ref1] ^)^. Many of these forms of undernutrition have long-term consequences. For example, IDA among children of pre-school age is linked to poorer cognitive, motor and social-emotional function in adolescence^(^ [@ref2] ^,^ [@ref3] ^)^. Animal models point to Fe deficiency causing direct biological deficits in neurometabolism, myelination and neurotransmitter function^(^ [@ref2] ^)^. IDA in pre-school children can be addressed in a number of ways. Nutrition-specific interventions and programmes addressing IDA focus on improving diet, supplementation and fortification^(^ [@ref1] ^,^ [@ref4] ^)^. Within this approach, provision of free supplements such as multiple-micronutrient powders (MMP)^(^ [@ref5] ^)^ -- single-dose sachets containing Fe and other micronutrients in a powdered form, which can be sprinkled on to foods prepared in the household -- has proved successful in reducing IDA among children of pre-school age in several developing countries including Bangladesh^(^ [@ref6] ^)^, Ghana^(^ [@ref7] ^)^ and Haiti^(^ [@ref8] ^)^. However, nutrition-specific interventions such as free direct provision of supplements^(^ [@ref1] ^)^ have limited sustainability and scalability; market-based strategies that rely on voluntary purchase and use by households are needed^(^ [@ref9] ^)^. These market-based strategies rely on the availability of MMP or other supplements in pharmacies or shops; household incomes that are high enough for these supplements to be affordable; and knowledge among caregivers that these supplements will benefit their children^(^ [@ref9] ^,^ [@ref10] ^)^. Conditional on MMP availability, nutrition-sensitive interventions may be one means of ensuring households have sufficient income to purchase supplements^(^ [@ref1] ^,^ [@ref10] ^)^. Nutrition-sensitive interventions draw on complementary sectors (such as agriculture, social protection and health) to affect the underlying determinants of nutritional status, including factors underlying household decisions related to nutrition^(^ [@ref10] ^)^. Social protection interventions -- specifically, programmes that provide cash or in-kind transfers to poor households -- are seen as particularly promising vehicles for addressing undernutrition in all its forms, as these reach large numbers of poor households which may be particularly constrained in nutrition-related decisions^(^ [@ref10] ^)^. However, assessments of the impact of these interventions on IDA are limited in number^(^ [@ref10] ^)^. These are dominated by one type of social protection programme, conditional cash transfers, and one region of the world, Latin America^(^ [@ref10] ^,^ [@ref11] ^)^. These programmes, where cash transfers were accompanied by the provision for pre-school child consumption of either micronutrient powders (Mexico) or Fe tablets (Honduras and Nicaragua), showed mixed results^(^ [@ref11] ^)^. Anaemia in pre-school children was reduced in Mexico but there was no effect in either Honduras or Nicaragua^(^ [@ref11] ^)^. Further, to the best of our knowledge, it has not been determined if raising incomes is sufficient to increase use of MMP or other micronutrient supplements, or if this needs to be complemented by activities that improve caregivers' knowledge of micronutrient deficiencies and how these can be addressed. The present paper provides new evidence on the impact of a nutrition-sensitive social protection intervention on the use of supplements that address IDA among children. We assessed whether transfers received as part of a social protection intervention are sufficient to induce use of supplements, whether the addition of a high-quality nutrition behaviour change communication (BCC) intervention has additional effects on use and whether the form of the transfer, cash or food, affects the use of these supplements. We assessed this using two cluster-randomized controlled trials in Bangladesh, varying the type of transfer provided -- cash or food -- with or without BCC. Methods {#sec1} ======= Programme description {#sec1-1} --------------------- Between March 2012 and May 2014, the Transfer Modality Research Initiative (TMRI) implemented two cluster-randomized controlled trials (RCT): one in rural areas of the north-west region of Bangladesh ('North RCT') and one in the south ('South RCT')^(^ [@ref12] ^)^. In the North, study villages were randomly assigned to a control group or one of four treatment arms in which beneficiaries received one of the following monthly: (i) a cash transfer of 1500 Taka ('Cash'), approximately \$US 19 and equivalent to approximately 25 % of average monthly household consumption expenditures of poor rural households in Bangladesh^(^ [@ref12] ^)^; (ii) an equal-value food ration consisting of rice, lentils and micronutrient-fortified cooking oil ('Food'); (iii) a half cash transfer and half food ration ('Cash & Food'); or (iv) the cash transfer along with nutrition BCC ('Cash + BCC'; see [Figs 1](#fig1){ref-type="fig"} and [2](#fig2){ref-type="fig"} for further details). The control group did not receive food, cash or nutrition BCC. In the South, study villages were also randomly assigned to a control group or one of four treatment arms; the first three treatment groups were the same as in the North. The final treatment group in the South was different: beneficiaries received the food ration along with nutrition BCC ('Food + BCC'). All transfer payments and BCC activities were implemented for 24 months. Quantities were chosen so that the value of the food ration was equal to the value of the cash provided in the cash treatment arm, as of March 2012. Payments were made to mothers who met study inclusion criteria at baseline; specifically, they lived in poor households (see 'Sample design' section for details) and had at least one child aged 0--24 months.Fig. 1Flow diagram of participant participation in the North randomized controlled trial conducted in north-west Bangladesh, March 2012--May 2014 (BCC, (high-quality nutrition) behaviour change communication) Fig. 2Flow diagram of participant participation in the South randomized controlled trial conducted in south Bangladesh, March 2012--May 2014 (BCC, (high-quality nutrition) behaviour change communication) The nutrition BCC consisted of a suite of intensive activities. The core activity consisted of a weekly, one-hour group session of the ten beneficiaries in each village with a trained community nutrition worker (CNW). These sessions covered a defined series of topics: nutrition, diet diversity and health; handwashing, hygiene and health; diet diversity and micronutrients; breast-feeding; complementary foods for children aged 6--24 months; feeding and treatment of children with diarrhoea; maternal nutrition; encouraging homestead food production; and women's status and relationships with influential family members such as husbands and mothers-in-law (e.g. negotiating intrahousehold relationships related to feeding pre-school children). Methods used to deliver this information included presentations, question and answer, interactive call and answer, songs and chants, practical demonstrations and role playing. Micronutrient deficiencies and how these could be addressed were discussed approximately once per month. CNW made home visits to beneficiaries, to follow up on topics discussed during the group sessions and to discuss specific concerns that mothers might have. While attendance at the BCC sessions was a condition for receipt of transfers in the 'Cash + BCC' and 'Food + BCC' arms, this was a soft condition. When a mother missed a session, the CNW followed up with a home visit to ascertain why the session had been missed and discuss the session's material; no beneficiary was dropped from the study for failing to attend sessions. MMP and Fe supplements were not directly provided as part of the BCC activities. However, the BCC emphasized the importance of micronutrients including Fe in the diets of children aged 6--59 months and encouraged use of MMP and other Fe supplements. MMP, specifically a five-ingredient micronutrient powder (branded as *Pushtikona* and containing 12·5 mg ferrous fumerate, 0·16 mg folic acid, 0·3 mg vitamin A (acetate), 30 mg vitamin C and 5 mg zinc gluconate), was available through the 'Bangladesh Sprinkles Program', a partnership between non-governmental organizations BRAC and the Global Alliance for Improved Nutrition (GAIN)^(^ [@ref13] ^)^. At baseline, a single-dose sachet of *Pushtikona* cost 1·55 Taka^(^ [@ref14] ^)^. Over a 6-month period, sixty doses are recommended^(^ [@ref13] ^)^, implying that the per month cost of *Pushtikona* is 15·5 Taka or 1·03 % of the monthly cash transfer. Other Fe supplements such as tablets or syrups were also available through health clinics or pharmacies in many study areas. Sample design {#sec1-2} ------------- We used a cluster-randomized controlled trial design; see [Figs 1](#fig1){ref-type="fig"} and [2](#fig2){ref-type="fig"}. In each region, five sub-districts (*upazilas*) were selected from a list of *upazilas* where, in 2010, the proportion of households living below the lower poverty line in Bangladesh (approximately 1200 Taka per person per month^(^ [@ref15] ^)^) was 25 % or more. In each *upazila*, we obtained a list of all villages (the smallest administrative unit in Bangladesh, comprising approximately 250 households). Using a random number generator, each village was assigned a random number. Villages were sorted in ascending numerical order; the first fifty villages in this sorted list were assigned to the 'Cash' group, the next fifty villages were assigned to the 'Food' group, the next fifty villages were assigned to the 'Cash & Food' group, the next fifty villages were assigned to the 'Cash + BCC' group in the North and to the 'Food + BCC' group in the South, and the next fifty villages were assigned to the 'Control' group. In these selected villages, a census was carried out. From these data, a list of households was constructed that were considered poor (i.e. their predicted level of consumption -- based on a score calculated using information on the age and education of the household head, housing characteristics, ownership of consumer durables, land ownership and household livelihoods -- lies below the lower poverty line for rural Bangladesh); would have a child aged 0--24 months by the time the intervention began; and were not receiving benefits from any other social safety net interventions. These were the eligible households for participation in the study. Ten households meeting these three conditions were randomly selected from each village using simple random sampling, giving a total sample size of 2500 households targeted in each of the two studies. Study design and participants {#sec1-3} ----------------------------- The baseline survey was carried out in March--April 2012, prior to the first payment. The principal survey instrument was a multi-topic household questionnaire that included questions on maternal knowledge of infant and young child feeding (IYCF) and access to and use of micronutrients. Mothers of children under the age of 24 months at baseline (or if absent, the primary female caregivers, referred to hereafter as 'mothers') and their spouses provided answers to different sections based on who was most informed, with men interviewed by male enumerators and women interviewed by female enumerators. Mothers responded to all questions on micronutrient supplements. An endline survey was conducted in April 2014 during the final month of payments. Community questionnaires were administered to capture information on local infrastructure and access to services. CNW who implemented the nutrition training were also surveyed at endline. Measures {#sec1-4} -------- We considered five outcome measures. These are: (i) whether the mother could identify at least one adverse consequence of insufficient Fe intake by infants and young children; (ii) whether the mother was familiar with MMP; (iii) for each child aged 6--59 months, whether the mother had ever mixed MMP into the food the child consumed; (iv) for each child aged 6--59 months, whether the mother had at any time during the last 7d mixed MMP into the food the child consumed; and (v) for each child aged 6--59 months, whether the mother had at any time during the last 7d mixed MMP into the food the child consumed or provided an Fe supplement in some other form (tablets, syrup). Each outcome was set equal to 1 when the answer was yes and equal to 0 when the answer was no. Outcomes (i) and (ii) were collected at baseline and endline. For children aged 6--59 months, outcomes (iii), (iv) and (v) were collected at endline. We used additional variables for attrition analysis and robustness checks, as well as for covariates in impact analysis. These included baseline maternal characteristics (age and grade of formal schooling), child characteristics (age in months and sex) and household characteristics (log value (in Taka) of production assets and consumer durables, household size, whether the household is female headed) from the household surveys; as well as locality characteristics (number of months during which local roads were impassable, the presence of individuals from non-TMRI organizations who also provided information on aspects of IYCF nutrition (called 'IYCF promoters')) from the community surveys. Sample size calculations {#sec1-5} ------------------------ We calculated the *ex post* statistical power for the outcomes specific to the paper. Under the conservative assumption that baseline covariates included in analysis would provide no explanatory power for these outcomes, setting a significance level of 0·05 and statistical power of 0·80, and using outcome-specific means, standard deviations and intracluster correlations from the baseline data, we estimated that this sample provides sufficient statistical power to detect the following: a 10 percentage point increase in mothers who can identify one reason why Fe deficiency in children is a concern; an 8 percentage point increase in the likelihood that a mother has heard of MMP; a 10 percentage point increase in the likelihood that a child ever consumed MMP; a 5 percentage point increase in the likelihood that in the last 7d a child consumed MMP; and a 5 percentage point increase in the likelihood that in the last 7d a child consumed MMP, Fe tablets or Fe syrup. Actual power would differ if parameters changed between baseline and endline or if inclusions of baseline covariates led to power gains. Statistical analysis {#sec1-6} -------------------- Statistical analyses were conducted in the statistical software package STATA version 15.0. Household, maternal and child characteristics were compared across treatment and control arms separately for the North RCT and the South RCT. Using ANOVA, variables were considered balanced if *P*\>0·05 for *F* tests comparing means for continuous variables and percentages for dichotomous variables across all treatment and control groups. We assessed whether attrition was non-random by estimating a probit model where the dependent variable equalled 1 if the household attrited between baseline and endline, 0 otherwise. We included as covariates household treatment status, the baseline household and maternal characteristics used as controls for our impact estimates, baseline maternal knowledge of the consequences of Fe deficiency and baseline maternal awareness of MMP. Parameter estimates were converted to marginal effects^(^ [@ref16] ^)^. A covariate had a statistically significant impact on the probability of a household attriting if *P*\<0·05. Standard errors accounted for clustering at the village level^(^ [@ref16] ^)^. We conducted ANCOVA estimates for the impact of different treatment arms at endline on mothers' knowledge of the consequences of insufficient Fe intake and on their familiarity with MMP controlling for baseline levels of maternal knowledge regarding the consequences of Fe deficiency, as well as other baseline maternal characteristics (age and grade of formal schooling) and household characteristics (log value (in Taka) of production assets and consumer durables; household size; whether the household is female headed). Single-difference impact estimates were used to measure the impact of different treatment arms on the child-level outcomes. These controlled for the same baseline maternal and household characteristics listed above as well as child age (in months) and sex. Because the outcomes are dichotomous variables, we used a probit estimator. Estimated coefficients were transformed into marginal effects; for our treatment variables which are dichotomous, these were obtained by calculating the predicted change in our dichotomous outcomes when we changed the value of the dummy variable for treatment from 0 to 1^(^ [@ref16] ^)^. We report a pseudo *R* ^2^ statistic to assess goodness-of-fit^(^ [@ref16] ^)^. Standard errors account for clustering at the village level^(^ [@ref16] ^)^. Impacts were considered statistically significant if *P*\<0·05. We used Wald *χ* ^2^ statistics to test for differences in impacts across treatment arms^(^ [@ref16] ^)^. Robustness checks included assessing whether the impact estimates were sensitive to the inclusion or exclusion of control variables, disaggregating the sample by maternal and child characteristics, disaggregating by locality characteristics such as market accessibility and the presence of non-TMRI IYCF promoters, and whether the results were sensitive to the use of alternative estimation methods (logits, linear probability models and linear probability models with union fixed effects (unions are the geographic unit above a village)). Results {#sec2} ======= Programme implementation {#sec2-1} ------------------------ Quantitative data collected throughout the intervention indicated that the TMRI transfers and BCC were implemented as designed. Women in the BCC treatment arms attended nearly all weekly sessions. Average attendance in the North among women in BCC treatment arms was forty-seven sessions per year and in the South forty-eight sessions per year -- with each session lasting approximately an hour. When a session was missed, 83 % of respondents reported that the CNW followed up with a home visit. Further, CNW delivering the BCC messages were knowledgeable. The endline survey of CNW included a fourteen-question quiz on key nutrition messages they were supposed to provide to beneficiaries regarding exclusive breast-feeding; the introduction of complementary foods; the importance of diet diversification; micronutrients; and water, sanitation and health. The mean score out of 14 was 13·2 in the North and 13·5 in the South. Trial profile and attrition {#sec2-2} --------------------------- In the North RCT, 2500 households were invited to participate in the study. There were two refusals, yielding a baseline sample of 2498 households ([Fig. 1](#fig1){ref-type="fig"}). At endline, across all arms we obtained complete data for 2408 households; these household contained 2698 children aged 6--59 months. The household attrition rate was 3·6 % with most of this arising from out-migration. Attrition probit models found that households in the 'Cash & Food' treatment group were 2·4 percentage points less likely to attrit relative to control households (*P*=0·02) and female-headed households were 3·1 percentage points more likely to attrit (*P*=0·03). No other variables were associated with attrition in the North. In the South RCT, 2500 households were invited to participate in the study. There were six refusals, yielding a baseline sample of 2494 households ([Fig. 2](#fig2){ref-type="fig"}). At endline, across all arms we obtained complete data for 2432 households; these household contained 2775 children aged 6--59 months. The household attrition rate was 2·5 % with most of this arising from out-migration. In the South, analysis of attrition found no association between treatment variables, household or maternal characteristics and attrition. One community characteristic, months in which roads were passable, was associated with lower attrition. Each month that a village road was passable reduced the likelihood of attrition by 0·1 % (*P*=0·04). In the North, across all treatment arms the sample is balanced across baseline household characteristics, baseline maternal characteristics (including knowledge of Fe deficiency in children and awareness of MMP), and endline child age and sex ([Table 1](#tab1){ref-type="table"}). In the South, the sample is balanced across baseline household characteristics (except for land ownership), endline child characteristics, baseline maternal age and knowledge of Fe deficiency in children, but not for baseline maternal schooling or awareness of MMP ([Table 1](#tab1){ref-type="table"}).Table 1Baseline characteristics of households, mothers and children by study and treatment arm, Bangladesh, March 2012--May 2014'Cash''Food''Cash & Food''Cash + BCC''Food + BCC''Control'Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}*P*North RCTLog value of per capita household assets6·841·046·911·036·750·986·881·02----6·881·060·11Land owned (decimals[\*](#tab1fn1){ref-type="fn"})13·837·614·933·412·437·213·432·7----15·032·60·74Household size (no. of persons)4·91·54·91·54·91·454·91·5----4·91·40·98Female head (%)5·923·79·729·67·626·68·027·2----6·725·10·17Mother's age (years)30·49·930·910·330·28·930·59·5----30·49·90·85Mother's schooling (grade)2·33·02·22·92·12·82·12·9----2·43·00·49Mother knows about Fe deficiency (%)79·740·381·638·878·541·181·339·0----77·441·90·34Mother heard of MMP (%)24·142·827·948·928·545·224·443·0----25·143·40·30Child's age (months)36·79·936·010·335·711·235·910·7----35·510·50·34Child is female (%)47·950·047·350·048·550·049·750·0----46·549·90·87No. of mothers sampled483485490475--475--No. of children sampled534531559541--533--South RCTLog value of per capita household assets7·131·067·201·087·301·10----7·211·117·231·110·23Land owned (decimals[\*](#tab1fn1){ref-type="fn"})17·832·718·133·426·761·4----22·848·922·448·4\<0·01Household size (no. of persons)5·41·55·51·65·41·7----5·31·65·61·70·12Female head (%)8·928·512·433·010·831·1----12·332·912·132·60·28Mother's age (years)31·39·832·010·632·411·6----31·110·331·510·50·26Mother's schooling (grade)2·73·02·62·92·93·0----3·13·13·23·20·03Mother knows about Fe deficiency (%)60·948·860·349·061·048·8----59·249·262·448·50·87Mother heard of MMP (%)17·137·713·431·116·637·3----13·033·619·239·40·02Child's age (months)35·910·434·410·935·411·2----34·711·735·710·50·11Child is female (%)53·449·945·649·848·350·0----46·249·948·450·00·08No. of mothers sampled490489482--487484--No. of children sampled539566551--563556--[^1][^2] Impacts on maternal knowledge of iron deficiency and multiple-micronutrient powders {#sec2-3} ----------------------------------------------------------------------------------- In the North, mothers' knowledge of the adverse consequences of Fe deficiency increased by 11·9 percentage points in the 'Cash + BCC' treatment arm (*P*≤0·01) relative to the control group ([Table 2](#tab2){ref-type="table"}). It increased by 6·2 percentage points in the 'Cash & Food 'treatment arm (*P*=0·04). There was no statistically significant change in any other treatment arm. Awareness of MMP was 29·0 percentage points higher in the 'Cash + BCC' treatment arm (*P*≤0·01) relative to the control group. It was 11·4 percentage points higher in the 'Cash' treatment arm (*P*=0·02) relative to the control group. Awareness of MMP was 17·5 percentage points higher in the 'Cash + BCC' treatment arm compared with the 'Cash' treatment arm, with this difference being statistically significant (*P*≤0·01).Table 2Impact estimates of treatment arms on maternal knowledge by study and treatment arm, Bangladesh, March 2012--May 2014Mother can identify one reason why Fe deficiency in children is a concernMother has heard of MMPMarginal effect[se]{.smallcaps}*P*Marginal effect[se]{.smallcaps}*P*North RCT'Cash'0·0390·0290·120·1140·0460·02'Food'0·0130·0290·670·0350·0450·45'Cash & Food'0·0620·0280·040·0560·0450·22'Cash + BCC'0·1190·024\<0·010·2900·036\<0·01Pseudo *R* ^2^0·0190·051South RCT'Cash'0·0030·0250·890·0070·0410·87'Food'0·0340·0230·16−0·0130·0390·72'Cash & Food'−0·0020·0280·94−0·0080·0410·85'Food + BCC'0·0920·024\<0·010·2220·044\<0·01Pseudo *R* ^2^0·0010·028[^3] In the South, knowledge of the adverse consequences of Fe deficiency increased by 9·2 percentage points for mothers in the 'Food + BCC' treatment arm (*P*≤0·01) relative to the control group ([Table 2](#tab2){ref-type="table"}). There was no statistically significant change in any other treatment arm. Awareness of MMP was 22·2 percentage points higher in the 'Food + BCC' treatment arm (*P*≤0·01) relative to the control group. For the 'Food' treatment arm, there was no statistically significant impact of knowledge of MMP (*P*=0·72). Awareness of MMP was 22·1 percentage points higher in the 'Food + BCC' treatment arm compared with the 'Food' treatment arm with this difference being statistically significant (*P*≤0·01). In each of the North and the South, results for both knowledge of adverse consequences of Fe deficiency and awareness of MMP are robust to the inclusion or exclusion of maternal and household characteristics. Impact estimates on each treatment arm remained virtually unchanged when we estimated using a linear probability model or a union-level linear probability fixed-effects regression. In each region, we disaggregated the sample by maternal age, maternal education, household assets, the number of months during which local roads were impassable and the presence of a non-TMRI IYCF promoter. We found no statistically significant differences in these impacts across the different sub-samples. Impacts on child consumption of multiple-micronutrient powders and other iron supplements {#sec2-4} ----------------------------------------------------------------------------------------- MMP or other Fe supplements were available in 88 % of villages in the North and 78 % of villages in the South. In the North, children aged 6--59 months in the 'Cash' treatment arm were, at endline, 12·6 percentage points more likely to have ever consumed MMP (*P*=0·01) relative to the control group ([Table 3](#tab3){ref-type="table"}). Children in the 'Cash + BCC' treatment arm were 32·0 percentage points more likely to have ever consumed MMP (*P*≤0·01) relative to the control group. Children in the 'Cash' and the 'Cash + BCC' treatment arms were more likely to have consumed MMP in the last 7d relative to the control group by 5·2 and 16·9 percentage points, respectively (both *P*≤0·01). All treatment arms had, relative to the control group, a statistically significant increase in the likelihood that children consumed MMP, Fe tablets or Fe syrup in last week with the magnitude of the impact estimates ranging from 7·0 percentage points ('Food') to 22·3 percentage points ('Cash + BCC'). For each of these three outcomes, we tested the null hypothesis that the impacts of the 'Cash' and the 'Cash + BCC' treatment arms were equal; in each case we did not accept this null hypothesis (*P*≤0·01).Table 3Impact estimates of treatment arms on children's consumption of multiple-micronutrient powders (MMP) and other iron supplements by study and treatment arm, Bangladesh, March 2012--May 2014Child has ever consumed MMPChild has consumed MMP in the last 7dChild has consumed MMP, Fe tablets or Fe syrup in the last 7dMarginal effect[se]{.smallcaps}*P*Marginal effect[se]{.smallcaps}*P*Marginal effect[se]{.smallcaps}*P*North RCT'Cash'0·1260·0480·010·0520·0240·010·0900·033\<0·01'Food'0·0330·0490·500·0260·0190·120·0700·027\<0·01'Cash & Food'0·0360·0470·440·0370·0240·060·0800·031\<0·01'Cash + BCC'0·3200·044\<0·010·1690·033\<0·010·2230·039\<0·01Pseudo *R* ^2^0·0550·1070·073South RCT'Cash'0·0140·0320·650·0000·0090·980·0280·0170·07'Food'0·0390·0340·240·0000·0090·980·0050·0150·75'Cash & Food'0·0210·0330·510·0010·0080·920·0270·0180·10'Food + BCC'0·1190·037\<0·010·0390·015\<0·010·0710·022\<0·01Pseudo *R* ^2^0·0370·0700·033[^4] In the South, children aged 6--59 months in the 'Food + BCC' treatment arm were, at endline, 11·9 percentage points more likely to have ever consumed MMP (*P*≤0·01) relative to the control group and were 3·9 percentage points more likely to have consumed MMP in the last 7d (*P*≤0·01; [Table 3](#tab3){ref-type="table"}). For these outcomes, no other treatment arm in the South had statistically significant impacts. Relative to the control group, 'Food + BCC' (7·1 percentage points) had a statistically significant effect on the likelihood that children consumed MMP, Fe tablets or Fe syrup in the last week (*P*≤0·01). We rejected the null hypotheses that the impacts of 'Food' and 'Food + BCC' on children ever consuming MMP, on consumption of MMP in the last 7d and on consumption of MMP, Fe tablets or Fe syrup were equal (*P*≤0·01) in all cases. In both the North and the South, we disaggregated the child samples by child sex and age, by maternal education, by maternal age and by household wealth. In all cases, we did not reject the null hypothesis that the impact of 'Cash + BCC' (in the North) or 'Food + BCC' (in the South) did not differ across these disaggregations. We also disaggregated by the number of months during which local roads were impassable and the presence of a non-TMRI IYCF promoter, and found no statistically significant differences in these impacts across these different sub-samples in either the North or the South. Discussion {#sec3} ========== We found that a transfer (food or cash) accompanied by high-quality nutrition BCC improved mothers' average knowledge of Fe deficiency and awareness of MMP, as well as significantly increased the likelihood of their children aged 6--59 months consuming MMP or some other Fe supplement (tablets, syrup) in the preceding week. The BCC drove these effects: in all cases, improvements were statistically significant relative to not only the control group, but also relative to the group that received the corresponding transfer only ('Cash' in the North; 'Food' in the South). In the North, receiving a cash transfer alone also significantly increased mothers' awareness of MMP, as well as children's likelihood of consuming MMP ever or in the preceding week. It is possible that the receipt of cash resulted in mothers frequenting markets or health centres where these supplements were sold, and this exposure resulted in improved awareness and use of MMP. However, no similar effect was observed in the South and, as noted above, in the North the impact of receiving cash was smaller than that of receiving both cash and nutrition BCC. Our study has strengths. Implementation of the interventions was of high quality and as designed. Our analyses were based on longitudinal data with rich information on over 5000 children, their mothers, their households and their localities. Attrition was low. Our findings were robust to changes in estimation approach and disaggregation. Our study also has weaknesses. Given the design of the RCT, we cannot directly compare impacts across the North and the South. Neither RCT included a 'BCC only' arm. For two outcomes (mother can identify one reason why Fe deficiency in children is a concern; mother has heard of MMP), we have baseline and endline data, but for the other three (child has ever consumed MMP; child has consumed MMP in the last 7d; child has consumed MMP, Fe tablets or Fe syrup in the last 7d), we only have endline data. We did not measure anaemia status. Attrition in the North was significantly associated with being in the 'Food & Cash' arm; however, the magnitude is small (2·4 percentage points less likely to attrit relative to control households) and given the pattern of impacts, does not appear to drive the results. Some household and maternal characteristics showed statistically significant differences at baseline in the South; however, magnitudes of difference were small, and these characteristics along with others were included as control variables in impact estimation. Existing evidence suggests that free provision of MMP is effective in reducing IDA among children of pre-school age but leaves knowledge gaps on alternative delivery platforms with greater sustainability and scalability^(^ [@ref9] ^,^ [@ref17] ^,^ [@ref18] ^)^. The one study of a market-based approach that we are aware of, in which MMP were sold by front-line health workers in Bangladesh, found that household poverty and variable household awareness of MMP were important constraints to effectiveness^(^ [@ref19] ^)^. Moreover, increasing awareness through counselling on IYCF was also insufficient to meaningfully increase uptake^(^ [@ref19] ^)^. Our study complements this work, by examining in two RCT how increasing household income, with or without also increasing household awareness of MMP and other Fe supplements, affects households' purchase and use of the supplements, in a setting where they are widely available in the market. We find that simply increasing income of mothers in localities where MMP were widely available through either cash or in-kind transfers had no or at best small effects on increasing the use of MMP or related supplements. However, combining transfers with intensive high-quality BCC that increased mothers' knowledge of Fe deficiency and awareness of supplements led to significant increases in the likelihood that their children aged 6--59 months had ever consumed MMP (32 and 11·9 percentage points, *P*≤0·01), consumed MMP in the preceding week (16·9 and 3·9 percentage points, *P*≤0·01) and consumed either MMP or an Fe supplement in the preceding week (22·3 and 7·1 percentage points, *P*≤0·01). These results suggest that relaxing only the income constraint of poor households will not in general be sufficient to increase uptake of MMP and related supplements. However, relaxing both the income and awareness constraints together may be effective in a setting where the supplements are widely available. Taken together with the previous study in which relaxing only the awareness constraint appeared to be inadequate, our findings suggest that effective, scalable platforms may need to twin interventions that increase income with interventions that increase awareness of MMP and other supplements. The results of our study build the evidence base for social protection programming as a promising platform to increase the number of children, particularly children in poor households, who could be reached with MMP. These findings provide proof of concept that nutrition-sensitive interventions through social protection -- and particularly transfers with BCC added -- can be a promising way to advance progress on micronutrient deficiencies such as IDA, allowing a scale beyond what may be feasible or sustainable for nutrition-specific interventions such as free distribution of MMP. *Acknowledgements:* The authors thank Data Analysis and Technical Assistance (DATA) Ltd for its careful data collection work. *Financial support:* This study received financial support from the German Ministry for Economic Cooperation and Development (BMZ); the UK's Department for International Development (DFID) through its funding of the Transform Nutrition (TN) consortium; the Swiss Agency for Development and Cooperation (SDC); the United Nations Development Programme (UNDP); and the US Agency for International Development (USAID). None of these funding agencies had any role in the design and conduct of the study; collection, analysis and interpretation of the data; or the preparation, review and approval of the manuscript. *Conflict of interest:* None. *Authorship:* J.H., A.A. and S.R. jointly formulated the research question and designed the study. A.A. oversaw data collection. J.H. analysed the data. J.H., A.A. and S.R. collaborated on drafting and revising the paper. *Ethics of human subject participation:* This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the Institutional Review Board of the International Food Policy Research Institute, Washington, DC. The study was also reviewed in Bangladesh by the Ministry of Food and Disaster Management who issued Letters of Authorization to conduct the surveys. As both household heads and mothers of the index children were interviewed, verbal consent to participate in the study was received from both. Verbal consent was witnessed and formally recorded. The study was registered with ClinicalTrials.gov (study ID: NCT02237144). [^1]: BCC, (high-quality nutrition) behaviour change communication; RCT, randomized controlled trial; MMP, multiple-micronutrient powders. [^2]: A decimal is 0·1 ha. [^3]: MMP, multiple-micronutrient powders; RCT, randomized controlled trial; BCC, (high-quality nutrition) behaviour change communication. Marginal effects of probit models reported. ANCOVA estimates at endline control for baseline levels of maternal knowledge regarding the consequences of Fe deficiency, as well as other baseline maternal characteristics (age and grade of formal schooling) and household characteristics (log value (in Taka) of production assets and consumer durables; household size; whether the household is female headed). Standard errors account for clustering at the village level. [^4]: RCT, randomized controlled trial; BCC, (high-quality nutrition) behaviour change communication. Marginal effects of probit models reported. Single-difference estimates at endline control for baseline levels of maternal knowledge regarding the consequences of Fe deficiency, other baseline maternal characteristics (age and grade of formal schooling), household characteristics (log value (in Taka) of production assets and consumer durables; household size; whether the household is female headed) and child characteristics (age in months and sex). Standard errors account for clustering at the village level.

Introduction {#Sec1} ============ Electric gating---as an instrument to modulate properties of materials via control of carrier density---became famous in the middle of the 20th century, when research of W. Shockley, W.H. Brattain, and J. Bardeen was lauded by the Nobel prize in Physics in 1956. Nowadays, it lies in the technological foundation of our civilization---field effect transistor, where electric gating controls semiconductor channel and switches device between on- and off-states. However, application of electric gating techniques for a long time was restricted to semiconducting materials, where carrier density is low enough to be tuned by the gate voltage. Nowadays, electric gating is also used to control devices in the emerging field of two-dimensional materials like graphene and transition metal dichalcogenide monolayers. While carrier density for two-dimensional materials can be high, carriers are located at the interface---region where electric gating has the most influence. In contrast to the cases described above, metals are bulk materials, and have high carrier density at the same time. Thus, significant modulation of properties of metals by electric gating was mostly considered out of reach. While electric gating controls carrier density and directly influences conductivity of the material, it can be also applied to control any property of the material that depends on the position of the Fermi level. A few prominent examples include electric gating controlled insulator--superconductor^[@CR1]^ and paramagnet--ferromagnet transitions^[@CR2]^. Lack of such degree of control over physical phenomena in metals, presented a big disadvantage to the functionality of metallic devices. However, success in the tuning of the Curie temperature of ultrathin Co gave hope of finally achieving such a degree of control even in metallic materials^[@CR3]^. In 2006, Saitoh et al. converted the pure spin current into the electric charge current using the inverse spin Hall effect (ISHE) in metallic Pt layer^[@CR4]^, while Valenzuela et al. detected the same effect in the aluminum channel of lateral spin valve^[@CR5]^. The ISHE originates from the spin--orbit interaction inside the material. Due to the spin--orbit interaction, the scattering direction of the carriers depends on their spin direction. Thus, the ISHE couples spin current with the electric charge current and allows their interconversion: the longitudinal pure spin current generates transversal charge current (the ISHE) and vice versa (the spin Hall effect). After the initial prediction^[@CR6]^ and experimental detection^[@CR7]^ it took almost 20 years until the importance of the effect was recognized by the scientific community. Pt became the dominant material choice to introduce the spin--charge conversion in studied systems due to its large spin Hall angle (which characterizes the efficiency of the conversion between spin and charge currents) and easy fabrication^[@CR8]--[@CR11]^. In recent years, novel spin--charge conversion effects like spin--charge conversion due to the electric field and spin--orbit coupling at the interface between two materials^[@CR12]^, or spin--charge conversion via the spin--momentum locking in topological insulators^[@CR13]--[@CR15]^ challenged the dominance of the ISHE generated by Pt and other heavy metals^[@CR16]^. However, though some of the aforementioned systems possess higher spin--charge conversion efficiency than ISHE in Pt, at the moment they are very sensitive to interface quality or not robust at room temperature. Through the years, there were many attempts to achieve control over the spin--charge conversion process. Such control was achieved using electric bias tuning of Schottky barrier in semiconductors^[@CR17]^ and electric gating in the various two-dimensional systems^[@CR18]--[@CR21]^. However, as discussed above, in contrast to semiconductors and two-dimensional systems, electric gating control over spin--charge conversion via ISHE in metals remained a formidable challenge. Recent studies showed that it is possible to change the ISHE of Pt through composition control of the sample: by either adjusting the number of the scattering centers^[@CR22],[@CR23]^, or substituting part of Pt atoms with another element^[@CR24],[@CR25]^. However, tuning of the ISHE in the heavy metals within a single device remained elusive so far. In this paper, we report the largest to date modulation of the metal resistivity in ultrathin Pt film through the careful control of Pt thickness and an ionic gate technique. We show that such control over the carrier density allowed us to tune reversibly and reproducibly the amplitude of the ISHE in Pt over two orders of magnitude---a result that can be used in spin-torque and other spintronics devices that use spin--charge conversion. Results {#Sec2} ======= Resistivity and carrier modulation measurements {#Sec3} ----------------------------------------------- Figure [1](#Fig1){ref-type="fig"} shows the dependence of the Pt resistivity *ρ*~Pt~ in logarithmic scale on the inverse thickness 1/*d*~Pt~ and thickness *d*~Pt~ (a and b, respectively), where each filled circle represents an individual sample. The solid blue line shows the resistivity calculated using Eq. (23) from the literature^[@CR26]^ (which takes into account scattering by both grain boundaries and film surfaces) with values *p* = 0.8---fraction of carriers specularly scattered at the surface of Pt layer, bulk resistivity *ρ*~∞~ = 40 µΩ cm, mean free path *λ*~mfp~ = 10 nm and the grain boundary penetration parameter *ζ* = 0.25^[@CR26]^. Experimental data follows the theoretical calculation, which shows that Pt samples were successfully fabricated down to the smallest thickness. The sharp increase in the Pt resistivity with the decreasing thickness is due to the increased contribution from the surface and grain boundaries scattering in the thin layers. The atomic force microscopy measurements (Supplementary Note [11](#MOESM1){ref-type="media"}) also confirmed the continuous nature of the Pt films.Fig. 1Thickness dependence of the Pt resistivity. Dependence of the Pt resistivity *ρ*~Pt~ on **a** the inverse thickness 1/*d*~Pt~, **b** the thickness *d*~Pt~. Note the logarithmic scale of *y*-axis. Filled circles show the experimental data, line shows the resistivity calculated using Eq. (23) from the literature^[@CR26]^ and values *ρ*~∞~ = 40 µΩ cm, *λ* = 10 nm, *p* = 0.8, *ζ* = 0.25^[@CR26]^. Resistivity of the Pt films increased with the decreasing thickness due to increased contribution from the surface and grain boundaries scattering. See Supplementary Note [3](#MOESM1){ref-type="media"} for further details Figure [2](#Fig2){ref-type="fig"} shows the schematic image of the samples used in the study. Thin Pt films (in the thickness range from 1.5 to 20 nm) were grown on top of the insulating Gadolinium Gallium Garnet (GGG)/Yttrium Iron Garnet (YIG) substrates. We modulated charge carrier density in our Pt devices with electric top gate using the ionic liquid technique, which is also commonly referred to as the electric double-layer transistor method. During the measurement, a sample with the gate was mounted in a vertical position into the cavity of the electron spin resonance system. Figure [3](#Fig3){ref-type="fig"} shows the ratio of the resistivity of the samples measured at gate voltage *V*~G~ = −2 V and *V*~G~ = +2 V: *k* = *R*(*V*~G~ = −2 V)/*R*(*V*~G~ = +2 V). A simple calculation using the free electron Drude theory of metals that assumes one free electron per atom gives an estimation of the atom and carrier density in Pt *n* = 6.6·10^22^ cm^−3^. However, band calculations predict smaller number of 0.4 6s-band electrons per Pt atom^[@CR27]^. Experimentally even lower values of 0.24 conduction electrons per Pt atom were measured in thin Pt films, with bulk carrier density calculated to be *n* = 1.6·10^22^ cm^−3[@CR28]^. While Pt is a two-band conductor with dominating carriers from the closed s-like Γ-electron and open d-like X-hole Fermi surfaces^[@CR27],[@CR29],[@CR30]^, the difference in effective mass and specular reflection of light electron carriers and heavy hole carriers can lead to electron band dominated conduction in thin films^[@CR28],[@CR31]^. Using formula $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = \sqrt {\frac{3}{{8{\mathrm{\pi }}}}} \left( {\frac{{\sigma _\infty }}{{\lambda _{{\mathrm{mfp}}}}}\frac{h}{{{\mathrm{e}}^2}}} \right)^{3/2}$$\end{document}$, where e---elementary charge, *h*---Planck constant, and values of *σ*~∞~ and *λ*~mfp~ obtained from the thickness dependence of resistivity, value of the bulk carrier density in our Pt is estimated to be *n* = 6·10^21^ cm^−3^, which is the same with result calculated for thin Pt films from the data reported in the literature^[@CR26]^. Carrier density---modulated by the gate voltage in 2 nm-thick sample---is estimated to be *n* = 4.2·10^21^ cm^−3^ at *V*~G~ = −2 V, and *n* = 7.9·10^21^ cm^−3^ at *V*~G~ = 2 V. Thus, induced by the ionic gel carrier density is Δ*n* = ± 2·10^21^ cm^−3^, which gives induced sheet carrier density of Δ*n*~sh~ = Δ*n*·*d*~Pt~ = 4·10^14^ cm^−2^. This value is within the range of commonly reported carrier density modulation using the ionic liquid: the order of 10^14^ carriers/cm^2^ is routinely achieved, with the highest reported values larger than 10^15^ carriers/cm^2[@CR32]^. Using relation between *n* and *σ* from the above, we can theoretically estimate $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k = \frac{{R|_{V_{\mathrm{G}} = - 2{\mathrm{V}}}}}{{R|_{V_{\mathrm{G}} = + 2{\mathrm{V}}}}} = \left( {\frac{{d_{{\mathrm{Pt}}} + {\mathrm{\Delta }}n_{{\mathrm{sh}}}/n}}{{d_{{\mathrm{Pt}}} - {\mathrm{\Delta }}n_{{\mathrm{sh}}}/n}}} \right)^{2/3}$$\end{document}$, where *n* = 6·10^21^ cm^−3^. Figure [3](#Fig3){ref-type="fig"} shows experimentally measured *k* for samples with various Pt thickness (purple filled circles), and theoretically estimated *k* assuming the same induced sheet carrier density in all devices Δ*n*~sh~ = 4.8·10^14^ cm^−2^ (blue line). In agreement with theoretical calculation, all devices showed increased resistivity modulation factor *k* with the decreased thickness (see [Supplementary Notes](#MOESM1){ref-type="media"} [4](#MOESM1){ref-type="media"}, [5](#MOESM1){ref-type="media"}, [9](#MOESM1){ref-type="media"}, [14](#MOESM1){ref-type="media"} for further discussion on the resistivity modulation using ionic gel, and [Supplementary Note 15](#MOESM1){ref-type="media"} for detailed discussion on the effect of the charge screening on gate modulation). Additionally, in thin films resistivity contribution due to the grain hopping can possibly be modulated by the gate voltage application. However, measurements of the temperature dependence of the resistivity indicate that its contribution at 250 K (temperature of the spin pumping and ISHE measurements) is on the order of only a few percent even in 2 nm-thick films (see [Supplementary Notes](#MOESM1){ref-type="media"} [16](#MOESM1){ref-type="media"} for detailed discussion). We achieved the resistivity modulation *k* in our samples up to 280%: this is more than one order of magnitude larger than previously reported in other studies^[@CR31],[@CR33]--[@CR35]^.Fig. 2Layout of the experiment. **a**--**c** Schematic representation of the carrier density modulation in Pt channel by using top ionic gel gate. **a** At the negative gate voltage *V*~G~ \< 0, ions inside the ionic gel form negatively charged layer at the interface with Pt, thus decreasing the number of electron carriers available in the channel; **b** at *V*~G~ = 0, ionic gel is unordered and carrier density in the channel is not modulated; **c** at *V*~G~ \> 0, positively charged layer is formed at the interface with Pt leading to increased number of electron carriers in the Pt channel and decreased resistance. **d** Under the ferromagnetic resonance condition of YIG layer, out-of-plane spin current **j**~s~ is injected into Pt channel via spin pumping. **e** Inverse spin Hall effect, ISHE, inside the Pt channel converts out-of-plane spin current **j**~s~ into the in-plane charge current **j**~c~. **f** Schematic top view of the sample. Electromotive force generated by the ISHE is detected from the Ti/Au electrodes at the ends of the sample. See [Supplementary Notes](#MOESM1){ref-type="media"} [1](#MOESM1){ref-type="media"}, [17](#MOESM1){ref-type="media"} for further details on sample structure and fabricationFig. 3Thickness dependence of the Pt resistivity modulation. Dependence of the resistivity modulation factor *k* = *R*(*V*~G~ = −2 V)/*R*(*V*~G~ = +2 V) on the thickness of Pt layer *d*~Pt~. Blue solid line shows theoretical calculation assuming the same number of carriers induced in all devices, as described in the main text; purple filled circles are experimental data, each circle corresponds to a separate sample. The blue filled circle is the averaged resistivity modulation *k* obtained for 2.5 nm sample from 15 gate voltage sweeps (see Supplementary Note [5](#MOESM1){ref-type="media"} for details). In agreement with the theoretical calculation, devices showed increased resistivity modulation factor *k* with the decreased thickness. Note the logarithmic scale of *x*-axis Spin--charge conversion measurements {#Sec4} ------------------------------------ For the ISHE measurements, after setting the gate voltage, we cooled the sample from room temperature to 250 K. Ionic leak current, comparable to the spin--charge conversion current at room temperature in Pt (on the order of nA), is completely suppressed at 250 K, when ion molecules in the ionic gel become immobile. For the details about ionic gel preparation and measurement procedure, see Methods section. Magnetic field of the microwaves applied to the cavity with sample drives magnetization of the ferrimagnet YIG layer into precession at the certain value of the external magnetic field, known as the ferromagnetic resonance field **H**~FMR~^[@CR36]^. Precession of the magnetization induces transfer of the angular momentum from YIG into the adjacent Pt layer without any charge transfer, i.e., generates pure out-of-plane spin current **j**~s~, where spin direction is determined by the direction of the applied static magnetic field. This pure spin current generation method is commonly referred to as the spin pumping^[@CR37],[@CR38]^. The pure out-of-plane spin current is converted into the in-plane charge current via the ISHE, which is measured at Ti/Au electrodes located at the opposite sides of the samples. Figure [4d](#Fig4){ref-type="fig"} shows the electromotive force generated in the 2 nm-thick Pt sample under the gate voltage of 0 V during the microwave absorption in YIG layer. The generated pure spin and charge currents are proportional to the absorbed microwave power^[@CR39]^, thus generated electromotive force follows the Lorentzian shape of the microwave absorption spectrum. Direction of the injected spins **σ** is reversed together with the direction of the external magnetic field (characterized by angle *θ*~**H**~), which results in the sign change of the spin--charge conversion current: **j**~c~∝**j**~s~×**σ**. In agreement with the ISHE theory, sign of the generated electromotive force was reversed between *θ*~**H**~ = 0° (blue filled circles) and *θ*~**H**~ = 180° (purple filled circles). Subtracting electromotive force data for the opposite directions of the external magnetic field removes spurious contributions independent of magnetic field (for example, the Seebeck effect). The amplitude of the ISHE voltage was extracted from the fitting of the magnetic field-averaged electromotive force $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(V_{\theta_{\bf {H}}}=0^{\circ}-V_{\theta_{\bf {H}}}=180^{\circ})/2$$\end{document}$ using symmetrical and asymmetrical Lorentzian components (see [Supplementary Notes](#MOESM1){ref-type="media"} [6](#MOESM1){ref-type="media"}--[8](#MOESM1){ref-type="media"}, [13](#MOESM1){ref-type="media"} for more details, examples of the averaging and fitting procedure). Figure [4a, b](#Fig4){ref-type="fig"} shows a change in the amplitude of the current and voltage generated by the ISHE in the 2 nm-thick Pt sample. In contrast to the microwave absorption spectrum (Fig. [4i--n](#Fig4){ref-type="fig"}), electromotive force generated via the ISHE (Fig. [4c--h](#Fig4){ref-type="fig"}) was strongly modulated with the application of the gate voltage. The amplitude of the spin--charge conversion current was tuned from the *I*~ISHE~ = *V*~ISHE~/*R* = 3.0 nA at *V*~G~ = −2 V to *I*~ISHE~ = 0.1 nA at *V*~G~ = 2 V. In our best sample, we achieve modulation of the *V*~ISHE~ and *I*~ISHE~ from 100% at *V*~G~ = −2 V down to 0.8% and 1.7%, respectively, at *V*~G~ = 2 V. Figure [5c](#Fig5){ref-type="fig"} shows reproducibility of the modulation of the spin--charge conversion current in two different sweeps of the gate voltage for the same sample and in the different sample with the same Pt thickness *d*~Pt~ = 2 nm. These results show the successful achievement of control over the ISHE in a metallic material.Fig. 4Gate modulation data for the 2 nm-thick Pt sample. **a** Dependence of the ISHE current amplitude, *I*~ISHE~, on the applied gate voltage *V*~G~. **b** Dependence of the ISHE voltage amplitude, *V*~ISHE~, on the applied gate voltage *V*~G~. **c**--**h** Electromotive force measured for the direction of the external magnetic field *θ*~**H**~ = 0° (blue filled circles) and *θ*~**H**~ = 180° (purple filled circles), dashed black lines show corresponding experimental data points in **b** (*V*~ISHE~(*V*~G~)); **i**--**n** microwave absorbance spectrum at the ferromagnetic resonance; and **o**--**t** drain-source current *I*~DS~ dependence on the drain-source current voltage *V*~DS~; at a set gate voltage *V*~G~, where **B**′ = *µ*~0~(**H**−**H**~Center~). **c**, **i**, **o** *V*~G~ = −2.0 V; **d**, **j**, **p** *V*~G~ = 0 V; **e**, **k**, **q** *V*~G~ = 1.0 V; **f**, **l**, **r** *V*~G~ = 1.2 V; **g**, **m**, **s** *V*~G~ = 1.4 V; **h**, **n**, **t** *V*~G~ = 2.0 V. In contrast to the microwave absorption spectrum (**i**--**n**), electromotive force generated via the ISHE (**c**--**h**) was strongly modulated with the application of the gate voltageFig. 5Gate control of the spin--charge conversion in thin Pt films. **a** Resistance modulation under application of the gate voltage *V*~G~ for the *d*~Pt~ = 2 nm sample. **b** Electromotive force detected from *d*~Pt~ = 2 nm sample for *V*~G~ = −2.0 V and *V*~G~ = 2.0 V averaged over opposite directions of the external magnetic field (*θ*~**H**~ = 0° and *θ*~**H**~ = 180°) to remove spurious contributions; \|**B**′\| = *µ*~0~(**H**−**H**~Center~). **c** Comparison of the normalized spin--charge conversion current *I*~ISHE~/*I*~ISHE~^max^ between different devices, where *I*~ISHE~ = *V*~ISHE~/*R*---amplitude of the generated via the ISHE spin--charge conversion current. Blue filled circles---Device A with *d*~Pt~ = 2 nm, purple filled circles---Device A with *d*~Pt~ = 2 nm remeasured, green filled circles---Device B with *d*~Pt~ = 2 nm (*V*~ISHE~(*V*~G~) and *R*(*V*~G~) can be found in Supplementary Note [10](#MOESM1){ref-type="media"}), red filled circles---device with *d*~Pt~ = 2.5 nm, gray filled circles---device with *d*~Pt~ = 10 nm. Comparing to the 10 and 2.5 nm devices, 2.0 nm devices showed larger modulation of the ISHE current, as expected from the carrier density modulation mechanism Discussion {#Sec5} ========== In the following paragraph, we show that the ISHE observed in our samples is intrinsic in nature and originates from the inter-d-band excitations. The ISHE spin--charge conversion current is given by the equation^[@CR40]^, *I*~ISHE~ = *wθ*~SH~*λ*~s~tanh(*d*⁄(2*λ*~s~))(2e⁄ℏ)*j*~s~^0^, where *w*---channel width (same for all samples), *θ*~SH~---spin Hall angle, *λ*~s~---spin diffusion length, *d*---channel thickness, ℏ---reduced Planck constant, *j*~s~^0^---injected spin current density at the YIG/Pt interface. Also, due to the Elliott--Yafet spin relaxation mechanism^[@CR41],[@CR42]^ in Pt *σ*~SH~ = *σθ*~SH~∝*λ*~s~*θ*~SH~, where *σ*~SH~ and *σ* are spin Hall and electrical conductivities, respectively^[@CR22],[@CR23],[@CR43]^. Disregarding small changes in the factor tanh(*d*/2*λ*~s~) between different samples (it is close to 1 for all channel thicknesses because of the scaling of the spin diffusion length with resistivity of Pt samples^[@CR22],[@CR23],[@CR43]^), one arrives at: *I*~ISHE~ ≅ *Aσ*~SH~ *j*~s~^0^, where *A* = *w*(*λ*~s~/*σ*)tanh(*d*⁄(2*λ*~s~))(2e⁄ ℏ) can be considered as a constant across the samples. Since YIG surface treatment and sputtering conditions for Pt were identical across all samples, we can assume similar injected spin current density *j*~s~^0^ at the Pt/YIG interface. Thus, generated spin--charge conversion current *I*~ISHE~ from sample to sample was solely controlled by the spin Hall conductivity (SHC) *σ*~SH~ of the sample: *I*~ISHE~ ≅ *A*′*σ*~SH~, where *A*′ = *Aj*~s~^0^ is a constant. The ISHE in Pt was theoretically predicted^[@CR8],[@CR44],[@CR45]^ and experimentally confirmed^[@CR10],[@CR11]^ to be dominated by the intrinsic mechanism, unless the superclean regime is entered (*ρ*~Pt~ \< 15 μΩ cm), where extrinsic ISHE cannot be neglected anymore^[@CR23]^. Large spin--orbit splitting lifts the double degeneracy of the d-bands near the L and X points at the Fermi level in Pt. Intrinsic ISHE originates from the interband scattering between these orbitals^[@CR45]^, and---according to the band calculations---SHC should exhibit a sharp decrease with increasing resistivity^[@CR8],[@CR44]^. In contrast, in the previous experimental studies, SHC was measured to be independent of Pt resistivity^[@CR22],[@CR23],[@CR46]^. While the spin Hall angle of Pt varies greatly among different studies, it was shown to originate from the linear scaling of the spin Hall angle with the resistivity of the sample^[@CR22],[@CR23],[@CR43]^, leaving SHC *σ*~SH~ = *σθ*~SH~ unaffected by changes in the resistivity. Thus, to the best of our knowledge, the theoretically predicted dependence of the SHC on the resistivity of material has never been observed experimentally neither in Pt, nor in other materials. Our high-resistivity samples address this discrepancy between the theory and experiment, and access the resistivity-dependent regime of SHC. All experimental studies so far were carried out on the low-resistivity Pt samples, where interband excitations that govern SHC are controlled by the ℏ/Δ, where Δ is the spin--orbit-induced splitting of d-bands. In contrast to the low-resistivity regime (*ρ*~Pt~ \< 40 μΩ cm), in the high-resistivity regime (*γ*≫Δ) interband excitations are governed by the quasiparticle lifetime ℏ/*γ*, which is roughly inversely proportional to resistivity. Figure [6](#Fig6){ref-type="fig"} shows the dependence of the *σ*~SH~(*ρ*) ≅ *I*~ISHE~(*ρ*)/*A*′ (calculated using *I*~ISHE~ measured at *V*~G~ = 0 V and *A*′ = 0.05 nA Ω cm) on the resistivity of the samples, which was controlled by the thickness of the Pt layer (see Fig. [1](#Fig1){ref-type="fig"}). SHC showed a strong decrease with the resistivity in our samples that followed the *ρ*^−2^ dependence (Fig. [6](#Fig6){ref-type="fig"} dashed blue line) theoretically predicted for the *σ*~SH~^[@CR8],[@CR44]^. Our results provide experimental evidence for the inter-d-band excitations origin of SHC in Pt.Fig. 6Dependence of the spin Hall conductivity on the resistivity of Pt. Purple filled circles show the spin Hall conductivity (SHC) calculated from the experimentally measured spin--charge conversion current *I*~ISHE~, assuming *A*′ = 0.05 nA Ω cm; blue filled circles show SHC *σ*~SH~ from the band calculations of Kontani et al.^[@CR44]^. The experimentally measured decrease of SHC follows the theoretically predicted *ρ*^−2^ dependence of the SHC in the high-resistivity regime (dashed blue line)^[@CR8],[@CR44]^. The SHC values are given in ℏ/e units Keeping in mind the inter-d-band transitions nature of the SHC, we discuss a possible mechanism of the observed strong suppression of the ISHE at *V*~G~ \> 0. As discussed above, in thin films contribution of the s-like electrons to conduction is dominant, while the contribution from the d-like carriers is small due to the specular reflection and large effective mass. However, the main scattering mechanism for the s-like conduction electrons is phonon-induced scattering into d-like empty states at the Fermi level, which depends strongly on the density of the d-states. At the same time, the density of states of the d-bands affects the SHC governed by the inter-d-band transitions. Interestingly, band calculations showed that density of the d-states in Pt sharply decreases above the Fermi level^[@CR47]^. Thus, the upshift of the Fermi level at positive *V*~G~ should lead to decrease in the resistivity due to the increased number of electrons at Γ point and the decreased scattering rate through the d-states, and decrease in the SHC due to the decreased inter-d-band scattering because of the moving away from the points where the split d-bands are close to each other and the decreased number of d-states. This is consistent with our experimentally observed results where the decrease in the sample resistance is followed by the decrease in the spin--charge conversion current generated through the ISHE (Fig. [5](#Fig5){ref-type="fig"}). Such reduction of the SHC with the tuning of the Fermi level was predicted for the bulk Pt, though large suppression of the SHC was estimated to occur on the shift of Fermi level on the order of 1 eV^[@CR45]^, which is larger than expected in our case. We hope that our results will motivate theoretical studies on the Fermi level dependence of the SHC in ultrathin Pt films, where increased scattering and lower carrier density, together with the few-atomic layer thickness of the film, can lead to a difference in the SHC in comparison with bulk Pt, which can help to explain the sharper dependence of the SHC on the position of the Fermi level. Finally, we note that the anomalous Hall effect (AHE) induced by the proximity to ferrimagnetic insulators^[@CR48]^ or gate voltage application was reported in thin Pt films^[@CR49]^. Such induced magnetic proximity effect can lead to a reduction of the SHC in Pt^[@CR50]^. While the mechanism of the induced magnetic moments that causes the AHE in Pt is still not completely understood, the AHE in Pt was only present at low temperatures below 200 K, and a large part of the gate-induced resistance and the AHE modulation was irreversible^[@CR48],[@CR49]^. In contrast, we show reversible control over the resistance and the ISHE at *T* = 250 K (see Supplementary Note [5](#MOESM1){ref-type="media"} and Fig. [5c](#Fig5){ref-type="fig"}). Hence, magnetically induced moments in Pt are expected to emerge at temperature lower than used in our experiments. To confirm this, we carried out Hall measurements in our samples. We observed a clear non-linear component that can be attributed to the AHE only at 10 K (Supplementary Note [12](#MOESM1){ref-type="media"}). Moreover, the negative magnetoresistance in Pt was also attributed to the emergence of the magnetic moments^[@CR49]^. We observe switching from the positive to the negative magnetoresistance in *d*~Pt~ = 2 nm sample only at 10 K. Thus, magnetic effects appear in our system at much lower temperature than the 250 K, at which spin pumping and spin--charge conversion experiments were performed. In other study, change in sign of the spin--charge conversion in Pt-based spin-torque structures was reported with the thickness of the Pt layer^[@CR51],[@CR52]^. However, it originated from the spin--charge conversion at the interface between Pt and oxidized CoFeB layer, which is absent in our case. Our results provide insight into the fascinating physics of ultrathin Pt films and spin--charge conversion. Through the ionic gel gate, we demonstrated reversible control over the resistivity of Pt film that is one order of magnitude larger than was achieved in previous studies. Such control over the carrier density allowed us to tune the ISHE in Pt by two orders of magnitude---a result that can be used in the gate-tunable spin--charge converters, spin-torque, and other types of spintronics devices. For example, it opens an exciting possibility of the gate-controlled spin--orbit torque magnetoresistive random-access memory (SOT-MRAM), where the spin current generated by the spin--charge conversion in the heavy metal exerts a torque on the free magnetic layer^[@CR53]^. Methods {#Sec6} ======= Sample fabrication procedure {#Sec7} ---------------------------- Below is the description of the preparation and measurement procedure for each YIG/Pt sample used in the study. The GGG/YIG (1.3 µm-thick, 3 mm-long, and 1 mm-wide) (Granopt, Japan) substrate was polished with agglomerate-free alumina polishing suspension (50 nm particle size), and then annealed at 1000 °C in the air atmosphere for 90 min. The Pt layer was sputtered on top of YIG in Ar plasma at a rate 0.6 Å/s. Afterwards, the Ti(5 nm)/Au(100 nm) electric pads were formed on the sides of the sample by the electron beam evaporation. Ionic gel was prepared using mixture with weight ratio 9.3:0.7:20 of the PS-PMMA-PS polymer (Polymer Source, USA), DEME-TFSI ionic liquid (Kanto Chemical, Japan) and Ethyl Propionate (CH~3~CH~2~COOC~2~H~5~, Nacalai Tesque, Japan). Insulating double-side adhesive tape was placed on the sides of the Pt channel (inside the area covered by Ti/Au electric pads) to provide additional mechanical support for the gate electrode film, on top of which it was placed. Gate electrode film was mounted after the application of the ionic gel and was located directly above the Pt channel. See [Supplementary Notes](#MOESM1){ref-type="media"} [1](#MOESM1){ref-type="media"}, [17](#MOESM1){ref-type="media"} for further details on sample structure and fabrication. Measurement procedure {#Sec8} --------------------- For the measurement, sample was mounted in the TE~011~ cavity of the electron spin resonance system (JEOL JES-FA200). The applied microwave power was set to 1 mW, and the microwave frequency to *f* = 9.12 GHz. Gate voltage was set at room temperature; after the development of the electric double layer in the ionic gel, sample was cooled to 250 K and *I*--*V* characteristics, FMR and ISHE measurements were carried out. Constant nitrogen gas flow was supplied to the cavity with sample, which was only stopped during the refilling of the liquid nitrogen vessel. Schematic layout of the measurement procedure can be also found in Supplementary Note [2](#MOESM1){ref-type="media"}. Data availability {#Sec9} ----------------- Data measured or analyzed during this study are available from the corresponding authors on reasonable request. Electronic supplementary material ================================= {#Sec12} Supplementary Information **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Sergey Dushenko, Masaya Hokazono. Electronic supplementary material ================================= **Supplementary Information** accompanies this paper at 10.1038/s41467-018-05611-9. This work was supported in part by MEXT (Innovative Area "Nano Spin Conversion Science" KAKENHI No. 26103003), Scientific Research (S) "Semiconductor Spincurrentronics" (No. 16H06330), and Grant-in-Aid for Young Scientists(A) No. 16H06089. S.D. acknowledges support by JSPS Postdoctoral Fellowship and JSPS KAKENHI Grant No. 16F16064. Authors are grateful to T. Takenobu and J. Pu for the advice on the ionic gel preparation and application. S.D. and M.S. designed and supervised the experiment; M.H. prepared the samples and carried out the measurements; S.D. guided the measurements, processed and analyzed the data, and wrote the manuscript; all of the authors contributed to the discussion of the results. Competing interests {#FPar2} =================== The authors declare no competing interests.

Introduction {#s1} ============ Head and neck cancers accounts for five percent of all tumors, and half of them occur specifically in the oral cavity (Kademani, [@B22]). Oral squamous cell carcinoma (OSCC) is a subset of head and neck squamous cell carcinoma, constituting over 90% of all oral cancers (Tandon et al., [@B59]). Despite advances in surgical techniques, adjuvant radiotherapy, and chemotherapy, the incidence of OSCC appears to be increasing worldwide, and the 5-year overall survival rate remains low, at approximately 50--60%. Smoking, drinking, and chewing betel are the main risk factors for oral cancer (Lin et al., [@B32]). Other possible risk factors may include viral infection, fungal infection, and chronic periodontitis, whereas some cases cannot be clearly explained by any known risk factors (Sanjaya et al., [@B51]; Hubbers and Akgul, [@B21]; Rischin et al., [@B50]; Shaikh et al., [@B55]). In the 1990s, researchers demonstrated the pathogenic role of *Helicobacter pylori* in gastric cancer, linking carcinogenicity with bacteria for the first time (Marwick, [@B35]). Subsequently, many studies evaluated the relationships between bacteria and cancer in other organs. For example, an increased risk of gallbladder carcinoma is associated with *Salmonella typhi* infection (Scanu et al., [@B52]). Oral carcinogenesis is also associated with bacteria (Khajuria and Metgud, [@B23]). Previous studies based on bacteria culture and biochemical identification demonstrated that Gram-negative anaerobes (*Fusobacterium* spp., *Prevotella* spp., etc.) were present more frequently on the tumor surface of OSCC (Nagy et al., [@B39]; Bolz et al., [@B6]), but only semi quantitative or qualitative estimation of oral microflora were obtained. Then PCR technology and DNA-DNA hybridization were used to describe oral microflora, but each experiment could only find very limited bacterial changes (Tateda et al., [@B60]; Morita et al., [@B37]; Mager et al., [@B34]). With the emergence of next-generation sequencing (NGS), 16S rDNA sequencing promoted the study of associations between microbial flora and OSCC. Pushalkar et al. studied the saliva microbiome of OSCC patients and proposed their potential application as a diagnostic tool to predict oral cancer (Pushalkar et al., [@B46]). Zhao et al. observed that a group of periodontitis-correlated taxa was significantly enriched in OSCC samples (Zhao et al., [@B70]). Other studies have reported that *Fusobacterium nucleatum, Pseudomonas aeruginosa* (Al-Hebshi et al., [@B3]), and *Fusobacterium periodonticum* (Yang et al., [@B67]) are associated with OSCC development. The control samples used for OSCC microbiota study usually include tumor adjacent or contralateral tissue, and healthy subjects (Al-Hebshi et al., [@B2]). The key advantage of self-normal control over healthy subjects is that the person-related factors that are easy to affect oral flora, like genotype and diet, can be avoided. During the process of oral carcinogenesis, the local microenvironment is altered (Koontongkaew, [@B24]). So we want to determine the microbiota composition of OSCC surface and compare it with the contralateral normal tissues. We collected samples from tumor sites and contralateral normal tissues in the buccal mucosal of 50 patients with OSCC, mainly from East China. We used strict screening criteria, made clear pathological diagnosis before sampling, and conducted careful specialized examination on the lateral anatomical position to ensure no visible lesions. Our study aimed to determine the characteristics of oral microflora on OSCC tumor sites, which has implications for further mechanistic exploration, and can be used as a biomarker to predict OSCC with high diagnostic accuracy. Materials and Methods {#s2} ===================== Sample Collection ----------------- Fifty patients with SCC of the oral buccal mucosa (median age: 61 years; 63% men and 37% women) were recruited from the Department of Oral and Maxillofacial-Head and Neck Oncology of the Ninth People\'s Hospital (Shanghai, China), from January 2018 to July 2018. The diagnosis criteria of OSCC were confirmed by clinical presentation and pathologic diagnosis and all patients were diagnosed with OSCC for the first time and they did not have any history of cancer. For this study, we collected bilateral buccal mucosal tissues of the same patient with OSCC, thus, 100 oral tissue samples (50-paired samples) were obtained from non-tumor (50) and tumor sites (50). Patients were not on antibiotics for 1 week before sampling and had no history of other oral mucosal diseases or severe systemic disorders. Patients were prevented from drinking and eating for at least 2 h before sampling. According to a well-defined clinical protocol, the surface of the tumor site and the opposite healthy side of the oral mucosa were separately scraped 10 times. We used disposable sterile nylon flocking swab (cy-98000, Hua Chen Yang Incorporate, Shenzhen, China) for sampling and stored it in the prepared oral swab preservation solution (mainly Tris, EDTA and antiseptic) to prevent DNA degradation. All samples were kept on ice and transported to the laboratory within 2 h after collection; they were stored at −80°C at the laboratory before subsequent use. The study was approved by the Medical Ethical Committee of the Shanghai Institute of Planned Parenthood Research. Written informed consent was obtained from all participants involved in this study. DNA Extraction, Polymerase Chain Reaction (PCR) Amplification, and 16S rRNA Gene Sequencing ------------------------------------------------------------------------------------------- Genomic DNA was extracted using a TIANamp Swab DNA kit (Tiangen Biotech, China). The V3-4 hypervariable region of the 16S rRNA genes was amplified using the primers 338F (5′-CCTACGGGNGGCWGCAG-3′) and 806R (5′-GACTACHVGGGTATCTAATCC-3′) with a TransStart Fastpfu DNA Polymerase (TransGen, Beijing, China). Cycling conditions were as follows: 5 min at 95°C; 20 cycles of 45 s at 95°C, 30 s at 55°C, and 30 s at 72°C; and a final extension step for 10 min at 72°C. Each sample was PCR amplified in triplicate. All amplicons were purified with a QIAquick PCR Purification Kit (Qiagen, Valencia, CA, USA) and quantified on a Qubit instrument (Life Technologies). Samples were then pooled with equal concentrations, and 2 × 300 bp paired-end sequencing was performed for pooled amplicons on an Illumina MiSeq instrument. Bioinformatics and Statistical Analysis --------------------------------------- Paired-end 16S rRNA gene sequences were assembled using Mothur (version 1.41.1) (Schloss et al., [@B53]). The following criteria were used for sequences assembly and filter: homopolymers \<8, average Qscore \>25, window size = 50, ambiguous bases (N) = 0, and sequence length \>350 bp. The sequences alignment was performed using the SILVA reference databases (V132) (Quast et al., [@B48]); VSEARCH algorithm was used to identify the chimeric sequences; contaminant sequences were filtered based on the RDP trainset database, which was provided by Mothur; next, matric distances were generated, and the DNA sequences were clustered into operational taxonomic units (OTUs) at 97% similarity using the "cluster" command of Mothur. RDP classifier (80% threshold) (Wang et al., [@B64]) assigned the taxonomy to each OTU based on the Ribosomal Database Project (Cole et al., [@B9]). Assessments of community richness, evenness, and diversity (Shannon, Simpson, Shannoneven, Simpsonenven, ACE, and Chao indices and Good\'s coverage) were also performed using Mothur. Differences in features (taxonomy and OTUs) between the control and tumor tissues were determined using STAMP (Parks et al., [@B43]). Differences in bacterial diversity were assessed using analysis of molecular variance (AMOVA). For microbiome function prediction, classification information of each OTU was generated using the classify.seqs and classify.otu command, based on OTU representative sequences, related abundance information of OTU, and GreenGenes database. Then the biom file was generated using "make.biom" of Mothur. The biom file was analyzed using the online program Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt, <http://huttenhower.sph.harvard.edu/galaxy/>) (Langille et al., [@B30]), and KEGG pathway hierarchical categories level 3 was chosen for the predicted function analysis. Representative sequences of OTUs were used as query sequence to define species through BlastN against HOMD RefSeq V15.1, Silva SSU database (V132), and the online NCBI database with more than 99% identity and the highest total score (Quast et al., [@B48]; Escapa et al., [@B13]). Results {#s3} ======= Bacterial Populations and Core Microbiome in Oral Samples --------------------------------------------------------- The clinical characteristics of the study subjects are listed in [Table 1](#T1){ref-type="table"}. Two oral microbiota samples (one from the OSCC lesion and one from the healthy/control site) from each patient were collected for analysis. After quality filtering, 3,267,929 16S rRNA genes (from 50 patients) were identified for subsequent analysis. The average sequencing depth was 32,679 (18,353\~39,526) reads per sample. A minimum size of 18,353 was selected as a baseline for normalization to avoid statistical bias. In total, 2,983 OTUs (97% similarity) were observed from all samples. The sequencing depth (Good\'s coverage \> 98%) was sufficient to undertake microbiota analysis with OSCC and control groups. ###### Clinical characteristics of subjects. **Total** **Male** **Female** ----------------------------------------------------- ----------- ---------- ------------ **AGE**    Mean 60.7 60.3 61.4 **Sex** 50 32 (64%) 18 (36%) **SITES**    Left 28 18 (36%) 10 (20%)    Right 22 14 (28%) 8 (16%) **CLINICAL STAGE[^\*^](#TN1){ref-type="table-fn"}**    I 23 13 (26%) 10 (20%)    II 16 10 (20%) 6 (12%)    III 8 7 (14%) 1 (2%)    IV 3 2 (4%) 1 (2%) **DRINKING**    Previous 20 16 (32%) 4 (8%)    Current 17 14 (28%) 3 (6%)    Non-drinker 13 2 (4%) 11 (22%) **SMOKING**    Previous 17 15 (30%) 2 (4%)    Current 9 8 (16%) 1 (2%)    Non-smoker 24 9 (18%) 15 (30%) **BETEL NUT CHEWING**    Previous 4 3 (6%) 1 (2%)    Current 2 2 (4%) 0    Non-chewer 44 27 (54%) 17 (34%) *The criteria of clinical stage are based on the AJCC Cancer Staging Manual*. Our analysis showed that 99.0% of the oral microbiota was aligned into 13 phyla. Additionally, 95.6% of the oral microbiota was clustered into 82 families, and 91.0% was aligned into 162 genera. At the phylum level, the common bacteria *Firmicutes, Proteobacteria, Bacteroidetes, Fusobacteria*, and *Actinobacteria* were dominant in both the OSCC and control groups. At the family level, 17 families were identified as the major taxa and core microbiota co-existing in the OSCC and control groups, accounting for over 91.2% of the microbiome in both groups ([Table 2](#T2){ref-type="table"}). Among the 17 families, *Streptococcaceae, Prevotellaceae, Neisseriaceae, Pasteurellaceae, Fusobacteriaceae* and *Veillonellaceae* were dominant (\> 69.7% of the entire microbiome). Among the 162 genera, 41 were the dominant genera (with each genus comprising \>0.1% of the total microbiome), including *Streptococcus, Neisseria, Haemophilus*, and *Prevotella* ([Table 3](#T3){ref-type="table"}). Among the 41 dominant genera, 11 ubiquitous (core) genera were consistently found in all samples and comprised more than 62.4% of the total microbiome. ###### Dominant families and significant differences between the OSCC and Control groups computed by STAMP. **Phylum** **Family** **OSCC** **Control** **Enriched in** ---------------- ------------------------------ ---------- ------------- ----------------- Firmicutes Streptococcaceae 12.13% 25.54% Control Bacteroidetes Prevotellaceae 17.85% 11.99% OSCC Proteobacteria Pasteurellaceae 12.57% 12.55% Proteobacteria Neisseriaceae 9.92% 10.36% Fusobacteria Fusobacteriaceae 11.03% 3.29% OSCC Firmicutes Veillonellaceae 5.25% 6.87% Fusobacteria Leptotrichiaceae 4.21% 3.31% Bacteroidetes Porphyromonadaceae 3.47% 2.47% Bacteroidetes Flavobacteriaceae 3.65% 1.76% OSCC Actinobacteria Micrococcaceae 1.32% 3.73% Control Firmicutes Bacillales_Incertae Sedis XI 1.91% 2.53% Proteobacteria Burkholderiaceae 0.88% 2.83% Firmicutes Lachnospiraceae 2.12% 1.24% OSCC Firmicutes Peptostreptococcaceae 1.92% 0.41% OSCC Proteobacteria Campylobacteraceae 1.66% 0.61% OSCC Actinobacteria Actinomycetaceae 0.64% 1.61% Control Firmicutes Carnobacteriaceae 0.67% 1.54% Control ###### Dominant genera and significant differences between the OSCC and Control groups computed by STAMP. **Phylum** **Genus** **OSCC** **Control** **Feature** ----------------------------- ---------------------------------------------- ---------- ------------- ------------------- Firmicutes Streptococcus 12.10% 25.50% Ubiquitous (core) Proteobacteria Haemophilus 8.65% 10.82% Bacteroidetes Prevotella 11.02% 7.92% Ubiquitous (core) Proteobacteria Neisseria 7.92% 8.96% Ubiquitous (core) Fusobacteria Fusobacterium 10.98% 3.27% Ubiquitous (core) Firmicutes Veillonella 3.05% 5.33% Ubiquitous (core) Fusobacteria Leptotrichia 4.04% 3.25% Bacteroidetes Alloprevotella 4.79% 2.30% Ubiquitous (core) Bacteroidetes Porphyromonas 3.13% 1.95% Ubiquitous (core) Actinobacteria Rothia 1.32% 3.72% Ubiquitous (core) Bacteroidetes Capnocytophaga 3.43% 1.58% Ubiquitous (core) Firmicutes Gemella 1.91% 2.53% Ubiquitous (core) Proteobacteria Lautropia 0.88% 2.83% Proteobacteria Aggregatibacter 2.59% 0.92% Proteobacteria Campylobacter 1.66% 0.61% Firmicutes Granulicatella 0.67% 1.53% Ubiquitous (core) Actinobacteria Actinomyces 0.59% 1.43% Firmicutes Selenomonas 1.31% 0.70% Candidatus Saccharibacteria Saccharibacteria \_genera \_incertae \_sedis 0.67% 1.30% Spirochaetes Treponema 1.29% 0.38% Firmicutes Peptostreptococcus 1.18% 0.22% Firmicutes Lachnoanaerobaculum 0.61% 0.40% Firmicutes Peptococcus 0.79% 0.16% Firmicutes Catonella 0.81% 0.12% Actinobacteria Corynebacterium 0.20% 0.66% Firmicutes Dialister 0.49% 0.17% Bacteroidetes Tannerella 0.20% 0.44% Firmicutes Parvimonas 0.46% 0.12% Firmicutes Filifactor 0.36% 0.14% Firmicutes Solobacterium 0.31% 0.12% Proteobacteria Morococcus 0.16% 0.25% Firmicutes Peptostreptococcaceae \_incertae \_sedis 0.36% 0.04% SR1 SR1_genera \_incertae \_sedis 0.21% 0.18% Firmicutes Oribacterium 0.12% 0.23% Firmicutes Stomatobaculum 0.10% 0.21% Proteobacteria Pseudomonas 0.25% 0.04% Actinobacteria Atopobium 0.18% 0.10% Firmicutes Megasphaera 0.08% 0.20% Firmicutes Abiotrophia 0.15% 0.08% Firmicutes Lactobacillus 0.07% 0.13% Proteobacteria Cardiobacterium 0.06% 0.14% Changes in Bacterial Composition Between the OSCC and Control Groups -------------------------------------------------------------------- We compared the oral microbiota profiles of the OSCC and control groups. AMOVA showed significant differences in microbiota between the two groups (*P*~AMOVA~ \< 0.001). Species evenness and diversity were significantly higher in the OSCC group than in the control group ([Figure 1](#F1){ref-type="fig"}). Principal component analysis (PCA) was conducted to visualize the different diversities of microbiota in the two groups ([Figure 2](#F2){ref-type="fig"}). {#F1} {#F2} Among the 17 major families, 10 families showed significant differences (*P* \< 0.05) between the OSCC and control groups. The families *Streptococcaceae, Micrococcaceae Actinomycetaceae* and *Carnobacteriaceae* were decreased in the OSCC group, whereas another six families were increased in the OSCC group ([Table 2](#T2){ref-type="table"}). Both *Neisseriaceae* and *Pasteurellaceae* were present at over 10% in the OSCC and control groups, without significant differences between groups. Thus, these two families were stable and common microbiota in the oral cavity. Among the 41 dominant genera, 21 genera showed significant differences (*P* \< 0.05) between the OSCC and control groups. Eight genera, including *Streptococcus, Veillonella* and *Rothia*, showed significant decreases in the OSCC group (*P* \< 0.05), whereas 13, including *Fusobacterium, Alloprevotella* and *Porphyromonas*, showed significant increases ([Table 3](#T3){ref-type="table"}, [Figure 3](#F3){ref-type="fig"}). The genus *Neisseria, Prevotella* and *Gemella* were stable and common microbes (occupying about 20% in both the OSCC and control groups). Taken together, the results of these analyses indicated that the composition of the core bacteria present in the oral cavity was significantly altered in OSCC. {#F3} The top 50 species (OTUs) were chosen to identify differentially enriched species within groups using STAMP. In total, 14 species showed differences between two groups, such as *Streptococcus oralis*. Four species were decreased, whereas 10 species were increased in the OSCC group ([Figure 4](#F4){ref-type="fig"}). {#F4} Predicted Functional Changes in the Microbiomes of the OSCC and Control Groups ------------------------------------------------------------------------------ We used PICRUSt to predict and compare potential changes in microbial functions between the two groups. In the metabolism category, 45 Metabolism pathways and 14 pathways related to Genetic Information Processing were identified as having significant differences (*P* \< 0.05) between the OSCC and control groups ([Figure 5](#F5){ref-type="fig"}). Analysis revealed the relative abundance of genes associated with proinflammatory bacterial component, such as lipopolysaccharide biosynthesis; and genes involved in metabolism of cofactors and vitamins, such as Porphyrin and chlorophyll metabolism, were significantly increased in cancer sites. Genes participating in carbohydrate metabolism and PTS transport were significantly decreased. Other pathways, in particular the genes related to cell motility, such as bacterial chemotaxis and flagellar assembly, were remarkably enriched in the OSCC group. {#F5} Discussion {#s4} ========== As a part of the digestive tract, oral cavity includes diverse microorganisms (Segata et al., [@B54]), and oral microbiota is a complex microbial community (Lamont et al., [@B28]). The oral microbiota plays an important role in human health, and dysbiosis of oral microbiota can lead to a variety of systemic diseases (Olsen and Yamazaki, [@B41]). Since changes in gut microbial composition may contribute to cancer initiation and progression (Vivarelli et al., [@B63]), oral microbial dysbiosis may also be involved in the occurrence and development of oral cancer. In this study, we aimed to determine the relationships between oral buccal mucosal microbial profile and OSCC. Through the analysis, we found significant changes in microbiota between tumor sites and contralateral normal tissues in the buccal mucosa. In terms of the composition of the oral microbiota, *Firmicutes, Proteobacteria, Bacteroidetes, Fusobacteria*, and *Actinobacteria* were the five dominant phyla in the mouth. Six families (*Prevotellaceae, Fusobacteriaceae, Flavobacteriaceae, Lachnospiraceae, Peptostreptococcaceae*, and *Campylobacteraceae*) and 13 genera, including *Fusobacterium, Alloprevotella* and *Porphyromonas*, were enriched in cancer tissues, whereas *Streptococcus, Veillonella*, and *Rothia*, were significantly decreased in cancer tissues. Ten species showed significantly increased abundances in cancer lesions. These species included *Fusobacterium nucleatum, Prevotella intermedia, Aggregatibacter segnis, Peptostreptococcus stomatis*, and *Catonella morbi*, which reside in the oral mucosa as commensals but may be opportunistic pathogens with potential correlations with OSCC ([Figure 6](#F6){ref-type="fig"}). {#F6} *Fusobacterium nucleatum*, a known pathogenic oral species showing a 5.88% increase in cancer lesions, has been reported to enhance oral cancer progression via direct interactions with oral epithelial cells (Binder Gallimidi et al., [@B5]). Al-hebshi reported the associations of *F. nucleatum* and *Pseudomonas aeruginosa* with OSCC (Al-Hebshi et al., [@B3]). However, in our analysis, *P. aeruginosa* was not enriched in cancer lesions. A metatranscriptome analysis revealed that *Fusobacteria* was the main phylum causing increased expression of virulence factors in the oral microbiome of OSCC patients, and *F. nucleatum* was the most active bacterium expressing putative virulence factors in the tumor sites (Yost et al., [@B68]). *F. nucleatum* infection is prevalent in human colorectal carcinoma (Castellarin et al., [@B7]), and various mechanisms are involved in the process. For example, *F. nucleatum* can suppress host immunity leading to the carcinogenesis of colorectal cancer. *F. nucleatum* inhibits T cells and NK cells function by directly interacts with human CEACAM1 (Gur et al., [@B16]; Wu et al., [@B65]). *Prevotella intermedia* and *Porphyromonas gingivalis* are considered to be the pathogen of periodontitis (Mysak et al., [@B38]; Zhang et al., [@B69]; Hsiao et al., [@B19]). Studies have reported that repeated periodontitis were associated with increased risk of OSCC (Li et al., [@B31]; Shin et al., [@B56]). Both of them can secrete peptides (Lisi et al., [@B33]; Zhang et al., [@B69]; Eftekhari et al., [@B12]). It is reported that proteases can act as signaling molecules through activation of proteinase-activated receptors (PARs) (Van Spaendonk et al., [@B62]), which involves cell proliferation and apoptosis, autoimmunity (Lisi et al., [@B33]), cytokine production, microenvironment inflammation, pain and epithelial barrier function (Amadesi and Bunnett, [@B4]). There was an increase of peptidases in tumor sites by functional analysis. Proteases produced by bacteria can degrade host tissue like extracellular matrix (ECM), destruct host physical barriers, and modulate host immune response, finally contributing to the onset and progression of tumors (Alfano et al., [@B1]). Several bacterial inflammatory processes, such as lipopolysaccharide synthesis, flagella assembly, and bacterial chemotaxis, also have roles in mediating inflammation in cancer (Al-Hebshi et al., [@B3]; Perera et al., [@B44]). Chronic inflammation of the oral cavity, usually caused by microorganisms, has been observed at various stages of OSCC (Pushalkar et al., [@B46]; Chen et al., [@B8]). Poor oral hygiene can also lead to periodontitis, and many studies have shown that poor periodontal health status, such as gingivitis and periodontitis, can be a direct or indirect risk factor for oral cancer (Tezal et al., [@B61]; Perera et al., [@B45]). Thus, disruption of the balance between microbes and human hosts can increase the risk of many diseases, including cancer, regardless of external factors (such as the use of alcohol and cigarettes) or pathogenic microbial infections. Lipopolysaccharide (LPS), composed of lipids and polysaccharides, is a component of the cell wall of Gram-negative bacteria. Functional analysis in our study revealed a significant increase in LPS biosynthesis in OSCC sites. LPS has been reported to enhance OSCC progression and migration (Kurago et al., [@B27]; He et al., [@B17]). In innate and adaptive immunity, LPS is recognized by LPS binding protein (LBP) and Toll-like receptor 4 (TLR4) to stimulate cytokine transcription (Park and Lee, [@B42]) as part of the recognition of pathogen-associated molecular patterns (PAMPs) (Kumagai and Akira, [@B26]), thus causing the LPS-induced inflammation. In T cells, TLR4-ligand LPS stimulated the TLR directing the cells toward type 1 polarization and expressed suppressor of cytokine signaling (SOCS) 1, and thus suppressed IL-10 expression (Ghosh et al., [@B14]). IL-10 was considered as a switch from tumor-promoting inflammation to antitumor immunity, and deficient IL-10 signaling developed tumors spontaneously and at high rates (Oft, [@B40]; Talero et al., [@B58]). LPS could activate TLR4 signaling in tumor cells and help tumor cells escape attack from cytotoxic lymphocyte (CTL) and natural killer (NK) cells (Huang et al., [@B20]). The functional prediction of oral bacterial communities also revealed enrichment of genes involved in bacterial chemotaxis and flagellar assembly. This result is consistent with the report of Al-Hebshi et al. who performed functional analysis of the microbiome associated with OSCC based on the V1-V3 region of 16s rDNA, and proposed that bacterial flagella is a potent inflammatory structure like LPS, and bacterial chemotaxis play an important role in cancer-related inflammation (Al-Hebshi et al., [@B3]). The decrease in the phosphotransferase system (PTS), glycolysis and galactose metabolism might reflect the community response to reduced sugar source on the tumor surface, since increased glucose uptake is essential for OSCC cells to survive (Eckert et al., [@B11]). We also emphasized the relationship between bacteria, chronic inflammation, and tumors. Inflammation is a defensive response, which restores tissue injury and eliminates pathogenic agents. Transient inflammation is thought to be part of the body\'s immune defenses against pathogens, but persistent inflammation can lead to cancer (Mirjalili and Kheirollahi, [@B36]; Crusz and Balkwill, [@B10]). Once the balance in bacterial communities is broken, the dominance of pathogens or a significant increase in biomass will result in an inflammatory defense response in the human body. For example, *Porphyromonas*, especially *Porphyromonas gingivalis*, are obligatorily anaerobic, and their fermentation end products are associated with chronic inflammation (Gibson and Genco, [@B15]). However, if inflammation is unregulated or continuous, it can become chronic, which can induce malignant cell transformation in the surrounding tissue. This process involves a variety of inflammatory factors and signaling pathways, as a result, chronic inflammation works in carcinogenesis, tumor growth, epithelial mesenchymal transition (EMT), angiogenesis, and metastasis (Landskron et al., [@B29]). The tumor microenvironment is hypoxic; most of the bacteria that changed significantly in oral cancer were anaerobes. Pathogenic bacteria can promote the occurrence and development of malignant tumors. The tumor microenvironment can selectively promote the growth of specific bacteria. We should not neglect taxa with low abundance but showing significant increases in OSCC because these taxa might be taken as "keystone" microbes and may have stronger virulence and thus play a greater role in the development of cancer. The genus *Peptostreptococcaceae incertae sedis*, occupying only 0.04% in normal oral buccal mucosa, increased to 0.36% in OSCC sites. *Catonella morbi*, a Gram-negative anaerobic bacillus, is involved in primary endodontic infections (Siqueira and Rocas, [@B57]). Increased abundance of *Catonella* spp. and *Catonella morbi* in patients with chronic obstructive pulmonary disease (COPD) and periodontitis compared with that in patients without COPD (Wu et al., [@B66]) suggests that this periodontitis-associated bacteria may be related to oral cancers. *Gemella morbillorum*, a facultative anaerobic Gram-positive coccus of the phylum *Firmicutes*, is highly associated with OSCC tumor sites (Pushalkar et al., [@B47]) and has been cultured from both deep-tissue specimens and corresponding superficial tissues of OSCC samples (Hooper et al., [@B18]). *Campylobacter rectus* also plays pathogenic role in human periodontitis (Rams et al., [@B49]), and chronic inflammation may be a possible trigger for OSCC (Crusz and Balkwill, [@B10]). In a previous case report, a patient with OSCC was reported to also be suffering from advanced chronic periodontitis infected with *Campylobacter rectus, Porphyromonas gingivalis, Peptostreptococcus micros*, and *Fusobacterium nucleatum* (Kruger et al., [@B25]). In summary, we found that OSCC tissues exhibited a unique microbiota compared with contralateral normal tissues. From our findings, we propose that recolonization of bacteria may disrupt the equilibrium between the resident oral microbiota and the host. This may be a key link through which commensal oral bacteria promote oral cancer. Accordingly, these findings may provide insights into the development of vaccines and/or antimicrobial therapies to prevent OSCC. Alternatively, the significant association of bacteria with OSCC may have clinical utility in screening for cancer. Further studies are needed to explore these possibilities. Data Availability Statement {#s5} =========================== The sequence data have been submitted to the NCBI Sequence Read Archive (Accession Number: [PRJNA533177](PRJNA533177)). Ethics Statement {#s6} ================ The studies involving human participants were reviewed and approved by the Medical Ethical Committee of Shanghai Institute of Planned Parenthood Research. The patients/participants provided their written informed consent to participate in this study. Author Contributions {#s7} ==================== LZ, YL, and HZ contributed to the conception, design, data acquisition, analysis, and interpretation, drafted and critically revised the manuscript. CZ contributed to the conception, design, data analysis, and interpretation, and critically revised the manuscript. All authors gave final approval and agree to be accountable for all aspects of the work. Conflict of Interest -------------------- 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.** The work was supported by the National Natural Science Foundation of China (No. 81771127). [^1]: Edited by: Georgios N. Belibasakis, Karolinska Institutet (KI), Sweden [^2]: Reviewed by: Nezar Al-hebshi, Temple University, United States; J. Christopher Fenno, University of Michigan, United States [^3]: This article was submitted to Microbiome in Health and Disease, a section of the journal Frontiers in Cellular and Infection Microbiology [^4]: †These authors have contributed equally to this work

Ivens ABF, Gadau A, Kiers ET, Kronauer DJC. Can social partnerships influence the microbiome? Insights from ant farmers and their trophobiont mutualists. Mol Ecol. 2018;27:1898--1914. <https://doi.org/10.1111/mec.14506> 29411455 1. INTRODUCTION {#mec14506-sec-0001} =============== Across the tree of life, animals form partnerships with microbes, allowing them to colonize new habitats (Dubilier, Bergin, & Lott, [2008](#mec14506-bib-0027){ref-type="ref"}; Mueller, Mikheyev, Hong, et al., [2011](#mec14506-bib-0072){ref-type="ref"}), utilize unique metabolic pathways (Pauli et al., [2014](#mec14506-bib-0079){ref-type="ref"}; Pinto‐Tomas et al., [2009](#mec14506-bib-0081){ref-type="ref"}; Raychoudhury et al., [2013](#mec14506-bib-0088){ref-type="ref"}), increase protection against natural enemies (Kaltenpoth et al., [2014](#mec14506-bib-0051){ref-type="ref"}; Rangan et al., [2016](#mec14506-bib-0087){ref-type="ref"}) and even boost their reproductive output under certain ecological conditions (Montllor, Maxmen, & Purcell, [2002](#mec14506-bib-0067){ref-type="ref"}; Oliver, Degnan, Burke, & Moran, [2010](#mec14506-bib-0076){ref-type="ref"}). Hosts show a huge range of dependencies on these microbial partners, with some becoming so tightly associated that the formerly independent partners evolve into a single integrated organism (Fisher, Henry, Cornwallis, Kiers, & West, [2017](#mec14506-bib-0031){ref-type="ref"}; Gruber‐Vodicka et al., [2011](#mec14506-bib-0036){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; Van Leuven, Meister, Simon, & McCutcheon, [2014](#mec14506-bib-0111){ref-type="ref"}; West, Fisher, Gardner, & Kiers, [2015](#mec14506-bib-0113){ref-type="ref"}). In these cases, physical, genomic and metabolic integration can drive partner interests to be closely aligned, leading to mutual dependence and loss of autonomy (Gruber‐Vodicka et al., [2011](#mec14506-bib-0036){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; Kiers & West, [2015](#mec14506-bib-0053){ref-type="ref"}; Moran, McCutcheon, & Nakabachi, [2008](#mec14506-bib-0069){ref-type="ref"}). Often, these partnerships form the basis of evolutionary innovation, with the microbes' services allowing their hosts to evolve traits and behaviours to tap into novel resources (Joy, [2013](#mec14506-bib-0050){ref-type="ref"}; Moran, [2007](#mec14506-bib-0068){ref-type="ref"}). This includes farming behaviour, in which hosts promote and control growth, reproduction and often dispersal of the symbiotic microbes or other organisms on which they rely for food (Brock, Douglas, Queller, & Strassmann, [2011](#mec14506-bib-0008){ref-type="ref"}; Chomicki & Renner, [2016](#mec14506-bib-0017){ref-type="ref"}; Hata & Kato, [2006](#mec14506-bib-0038){ref-type="ref"}; Ivens, [2015](#mec14506-bib-0045){ref-type="ref"}; Mueller, Gerardo, Aanen, Six, & Schultz, [2005](#mec14506-bib-0071){ref-type="ref"}; Pauli et al., [2014](#mec14506-bib-0079){ref-type="ref"}). To successfully access new habitats and resources, however, partnerships need to be reliable (Chomicki, Janda, & Renner, [2017](#mec14506-bib-0016){ref-type="ref"}; Meseguer et al., [2017](#mec14506-bib-0066){ref-type="ref"}; Mueller, Mikheyev, Solomon, & Cooper, [2011](#mec14506-bib-0073){ref-type="ref"}; Simonsen, Dinnage, Barrett, Prober, & Thrall, [2017](#mec14506-bib-0103){ref-type="ref"}; Sudakaran, Salem, Kost, & Kaltenpoth, [2012](#mec14506-bib-0110){ref-type="ref"}). Especially in cases where microbes do not become physically integrated with their hosts, such as farming mutualisms, environmental context plays a major role in the availability and suitability of particular microbial consortiums (Kaltenpoth et al., [2014](#mec14506-bib-0051){ref-type="ref"}; McFall‐Ngai, [2008](#mec14506-bib-0065){ref-type="ref"}; Poulsen, Fernandez‐Marin, Currie, & Boomsma, [2009](#mec14506-bib-0083){ref-type="ref"}). While we know that context matters, it is unclear what factors are most important in driving the reliability of particular host--microbe associations. For example, phylogenetic relatedness and associated traits, such as transmission mode and compatibilities, are likely important, such that closely related species have more similar microbiota than distantly related species, (Anderson et al., [2012](#mec14506-bib-0001){ref-type="ref"}; Brucker & Bordenstein, [2013](#mec14506-bib-0010){ref-type="ref"}; Currie et al., [2003](#mec14506-bib-0020){ref-type="ref"}; Groussin et al., [2017](#mec14506-bib-0035){ref-type="ref"}; Henry, Maiden, Ferrari, & Godfray, [2015](#mec14506-bib-0039){ref-type="ref"}; Sanders et al., [2014](#mec14506-bib-0099){ref-type="ref"}). However, social partnerships of the host with other animals, such as intimate mutualistic farming relationships, are also key. It is well known that physical interactions with other organisms can influence an organism\'s symbiotic microbiome, sourcing and reinforcing specific microbial associations (Gonella et al., [2015](#mec14506-bib-0033){ref-type="ref"}; Lax et al., [2014](#mec14506-bib-0057){ref-type="ref"}; Macke, Tasiemski, Massol, Callens, & Decaestecker, [2017](#mec14506-bib-0061){ref-type="ref"}; Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}; Sintupachee, Milne, Poonchaisri, Baimai, & Kittayapong, [2006](#mec14506-bib-0104){ref-type="ref"}; Song et al., [2013](#mec14506-bib-0106){ref-type="ref"}; Stahlhut et al., [2010](#mec14506-bib-0108){ref-type="ref"}). Often the relative importance of these factors is difficult to untangle because it is challenging to find examples of distantly related species that share nearly identical social partnerships and physical environments. The recent characterization of a set of overlapping farming mutualisms allows us to look more closely at the role of phylogeny versus social partnerships in determining host--microbe symbiotic associations. These mutualisms involve two types of honeydew‐producing insects that are farmed by several ant species (Figure [1](#mec14506-fig-0001){ref-type="fig"} and Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Subterranean *Lasius* and *Brachymyrmex* ants farm multiple species of aphids and mealybugs (their "trophobionts") often in the same underground root chambers for "milk" (i.e. honeydew) and, occasionally, "meat" for protein (Ellison, Gotelli, Farnsworth, & Alpert, [2012](#mec14506-bib-0029){ref-type="ref"}; Ivens, [2015](#mec14506-bib-0045){ref-type="ref"}; Ivens, Kronauer, Pen, Weissing, & Boomsma, [2012a](#mec14506-bib-0046){ref-type="ref"}; Pontin, [1978](#mec14506-bib-0082){ref-type="ref"}). It is likely that the ants depend on the aphid and mealybug honeydew as their sole sugar supply, as well as amino acids via honeydew and predation (Ivens, [2015](#mec14506-bib-0045){ref-type="ref"}; Ivens et al., [2012a](#mec14506-bib-0046){ref-type="ref"}; Pontin, [1978](#mec14506-bib-0082){ref-type="ref"}; Way, [1963](#mec14506-bib-0112){ref-type="ref"}). In return for these nutritional benefits, the ants actively protect the trophobionts against predators and provide hygienic services that are key to trophobiont survival (Bach, [1991](#mec14506-bib-0003){ref-type="ref"}; El‐Ziady & Kennedy, [1956](#mec14506-bib-0030){ref-type="ref"}; Ivens, [2015](#mec14506-bib-0045){ref-type="ref"}; Paul, [1977](#mec14506-bib-0078){ref-type="ref"}; Way, [1963](#mec14506-bib-0112){ref-type="ref"}; Zwölfer, [1958](#mec14506-bib-0115){ref-type="ref"}). ![*Lasius* ants tending (a) *Rhizoecus* mealybugs, (b) *Prociphilus* aphids and (c) a mixed live stock of mealybugs and aphids on the underside of rocks covering their nests in Millbrook, New York (photos: A.B.F. Ivens) \[Colour figure can be viewed at <http://www.wileyonlinelibrary.com>\]](MEC-27-1898-g001){#mec14506-fig-0001} These farmed mealybugs (Figure [1](#mec14506-fig-0001){ref-type="fig"}a) and aphids (Figure [1](#mec14506-fig-0001){ref-type="fig"}b) reside in nests of the same host ants and are therefore engaged in similar social partnerships, meaning they are farmed under nearly identical abiotic and biotic conditions (Figures [1](#mec14506-fig-0001){ref-type="fig"}c and Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). In addition, both groups of organisms rely heavily on bacterial endosymbionts that facilitate their ability to feed off similar sugar‐rich, but otherwise nutrient‐poor, plant phloem sap (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Husnik & McCutcheon, [2016](#mec14506-bib-0042){ref-type="ref"}; Oliver et al., [2010](#mec14506-bib-0076){ref-type="ref"}). In aphids, the primary, obligate, endosymbiont is *Buchnera aphidicola* (Enterobacteriaceae), which is known to be vertically transmitted and to cospeciate with its insect hosts (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Jousselin, Desdevises, & D\'acier, [2009](#mec14506-bib-0049){ref-type="ref"}; Nováková et al., [2013](#mec14506-bib-0075){ref-type="ref"}). *Buchnera* is often complemented by secondary, facultative, endosymbionts such as *Serratia symbiotica* and *Hamiltonella defensa* (both Enterobacteriaceae; Henry et al., [2013](#mec14506-bib-0040){ref-type="ref"}, [2015](#mec14506-bib-0039){ref-type="ref"}; Russell & Moran, [2006](#mec14506-bib-0093){ref-type="ref"}). Similarly, mealybugs harbour the primary, obligate and vertically transmitted endosymbiont *Candidatus* Tremblaya princeps (Betaproteobacteria, hereafter referred to as *Tremblaya*). *Tremblaya*, in turn, often carries another intracellular Gammaproteobacterium such as *Moranella* (von Dohlen, Kohler, Alsop, & McManus, [2001](#mec14506-bib-0023){ref-type="ref"}; Husnik & McCutcheon, [2016](#mec14506-bib-0042){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; McCutcheon & von Dohlen, [2011](#mec14506-bib-0064){ref-type="ref"}). These Sternorrhyncha endosymbionts typically occur intracellularly. Extracellular (gut) bacteria have so far only been described for a couple of species and are thought to occur as opportunists or pathogens rather than specialised beneficial symbionts (Clark, Daniell, Wishart, Hubbard, & Karley, [2012](#mec14506-bib-0018){ref-type="ref"}; Grenier, Nardon, & Rahbé, [1994](#mec14506-bib-0034){ref-type="ref"}; Harada, Oyaizu, & Ishikawa, [1996](#mec14506-bib-0037){ref-type="ref"}; Sreerag, Jayaprakas, Ragesh, & Kumar, [2014](#mec14506-bib-0107){ref-type="ref"}). Despite similarities in habitat, overall lifestyle, and their general dependence on endosymbionts, mealybugs and aphids, as well as their respective endosymbionts, are only distantly related in phylogenetic terms. This provides the unique opportunity to explore the influence of phylogeny versus social partnerships. Given their reliance on vertically transmitted endosymbionts, it is expected that the aphid and mealybug microbiota are strongly correlated with the trophobionts' respective phylogenies. However, their overlapping social partnerships could also contribute to shaping the trophobionts' microbiota, for example by increasing opportunities for horizontal transfer by feeding on the same host plants, as has previously been described for transfer of bacterial symbionts among other arthropod taxa (Gonella et al., [2015](#mec14506-bib-0033){ref-type="ref"}; Sintupachee et al., [2006](#mec14506-bib-0104){ref-type="ref"}; Stahlhut et al., [2010](#mec14506-bib-0108){ref-type="ref"}). Because multiple species of both groups reside in the ants' nests, we can begin to tease apart these different factors. Like the farmed trophobionts, the different species of ants face similar selection pressures from farming the same trophobionts in similar, and rather extreme habitats. For example, they likely harbour gut bacteria that allow them to live off the sugary honeydew of their trophobionts (Russell, Sanders, & Moreau, [2017](#mec14506-bib-0095){ref-type="ref"}). We might, therefore, expect the different ant species to show overlap in their core symbiotic microbiota. Alternatively, the symbiotic microbiota of the ants may be correlated with phylogeny, resulting in a unique microbiome in each ant species. Testing this hypothesis is challenging because the exact phylogenetic relationships of the *Lasius* ants have yet to be worked out in more detail. However, if we find different ant species sharing microbiota, then this would suggest a potentially larger role for the ants' social environment. Here, we test whether the symbiotic microbiomes of interacting aphids, mealybugs and ants are exclusively a function of phylogeny, or whether the social farming partnerships also have some predictive power. To this end, we used a DNA barcoding approach on ants, aphids and mealybugs collected from nests of five North American ant species in Millbrook, New York. We asked (i) do the different trophobionts farmed by ants share a similar microbiome and (ii) do ants that farm the same trophobionts share a core symbiotic microbiome with each other, and even with their trophobionts? 2. MATERIALS AND METHODS {#mec14506-sec-0002} ======================== 2.1. Species diversity and sample collection {#mec14506-sec-0003} -------------------------------------------- The mutualistic network is composed of two groups of insects: the ant farmers in the genera *Lasius* (Formicinae) and *Brachymyrmex* (Formicinae) (five species, *L. claviger, L. umbratus*, and *L. nearcticus* \[common\], as well as *L. flavus* and *B. depilis* \[rare\]) (Ellison et al., [2012](#mec14506-bib-0029){ref-type="ref"}). The group of farmed trophobionts consists of two types of Sternorrhyncha: *Rhizoecus* (Pseudococcidae) mealybugs (unclassified species 1--5) and at least nine *Prociphilus* (Eriosomatinae) (*P. probosceus, P. fraxinifolii, P. longianus, P. erigeronensis* and cryptic species therein) and Pemphigini aphids (very rare, with only a single observation between 2013 and 2017; Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}, S. A. Schneider, personal communication). Generally, most aphid species are found in nests of all ant species at similar frequencies, given their differences in abundance. Mealybugs *R. *spp. 3 and 4 are also both found in nests of the three most common ants, but the rarer mealybug species *R. *spp. 1, 2 and 5 are restricted to the nests of the rarer ant species *L. flavus* and *B. depilis* (Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Ants, aphids and mealybugs were collected from 143 unique ant nests between 2013 and 2016, mostly during the months of April---June (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}). The vast majority of sampling took place in Millbrook, New York, USA (41.767897, −73.750848) with the exception of two sets of samples from Annandale‐on‐Hudson, New York. All nests were marked, and their GPS coordinates logged for future resampling. Immediately after collection, we stored most insect samples in absolute EtOH at −30°C awaiting further analysis. Of each trophobiont chamber (defined as a cluster of aphids or mealybugs on a rock surface \[Figure [1](#mec14506-fig-0001){ref-type="fig"}\] or tree root), one individual was stored in 70% EtOH, then heated for 2 minutes at 60°C and stored at room temperature in preparation for slide mounting for taxonomic purposes. Slide mounted specimens are stored at De Vrije Universiteit Amsterdam and are available upon request. From each sample containing ants, we used one individual worker for species identification, combining *COI* barcoding with morphological identification of the subsequently pinned specimen (Ellison et al., [2012](#mec14506-bib-0029){ref-type="ref"}). Specimens stored at The Rockefeller University, New York, are available upon request. Samples were then subjected to different barcoding approaches as follows: Illumina MiSeq microbiome profiling, Sanger sequencing of insect hosts and specific bacterial endosymbionts, or a combination of both (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}). Samples that repeatedly failed to amplify or that yielded mixed traces in Sanger sequencing, a sign of contamination, were excluded from further analysis. In total, this study includes data on 602 samples: 21 controls (1 positive, 20 negative), two honeydew samples, 129 ants, 340 aphids and 111 mealybugs (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}). In total, 446 samples were included for microbiome profiling in one or more of four MiSeq runs (MS1, MS2MS3 \[consisting of two MiSeq lanes\], MS4, MS5). These samples included 20 controls, 129 ants (*Brachymyrmex depilis n* = 2, *Lasius flavus n *=* *3, *L. claviger n *=* *55, *L. nearcticus n *=* *14 and *L. umbratus n *=* *55), 76 mealybugs (*Rhizoecus* sp.1 *n *=* *8, *Rhizoecus* sp.2 *n *=* *5, *Rhizoecus* sp.3 *n *=* *42, *Rhizoecus* sp.4 *n *=* *21; *Rhizoecus* sp.5 was not included due to its rarity) and 219 aphids (*Prociphilus fraxinifolii n *=* *1, *P. probosceus n *=* *8, *P. erigeronensis n *=* *107 and *P. longianus n *=* *103; *P. caryae* and the unclassified Pemphigini aphids were not included due to rarity). All mealybug and aphid samples and one worker per ant nest included in the MiSeq runs were subsequently subjected to targeted Sanger sequencing of their mitochondrial DNA (mtDNA) and one or more specific endosymbionts, together with 156 (*n = *1 negative control, *n = *35 mealybugs and *n *=* *121 aphids) additional samples. Only those (*n *=* *350) that successfully yielded high‐quality sequences from both insect mtDNA and endosymbiont DNA were included in Figures [2](#mec14506-fig-0002){ref-type="fig"} and [3](#mec14506-fig-0003){ref-type="fig"}, and Figures S4--S15 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}. {ref-type="supplementary-material"}). Each species harbours a specific strain of *Candidatus* Tremblaya (b) and *Sodalis* (c) endosymbionts, and *Rhizoecus* sp1. and sp2. also harbour specific *Serratia symbiotica* strains (d). (e) The rooted NJ‐tree based on 601 bp *COI* mitochondrial gene fragments (*n *=* *218) shows ten clusters (colour‐coded) among the root aphids, which all harbour specific *Buchnera aphidicola* strains (d). In addition, most individuals of *Prociphilus longianus* A also harbour a *Serratia symbiotica* strain (d). The mitochondrial phylogeny (e) remains insufficiently resolved for the *P. erigeronensis* clade, and branch location within this clade has been rearranged to reflect haplotype‐specificity of *Buchnera* strains within this clade. The original aphid *COI* NJ‐tree with sample labels is given in Figure S10 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}. Trees in (a) and (e) are based on Tamura--Nei distances (scale bars), only relative distance (not to‐scale) of endosymbiont strains is given for clarity (b, c, d and f). For to‐scale NJ‐trees based on bacterial *16S rRNA*/*rpS15* and *rpS15‐16S rRNA* intergenic spacer fragments with sample labels and, for the mealybug endosymbionts, outgroups, see Figures S6, S8, S12, S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"} \[Colour figure can be viewed at <http://www.wileyonlinelibrary.com>\]](MEC-27-1898-g002){#mec14506-fig-0002} {ref-type="fig"}b), trophobiont‐derived bacteria (in blue, including *Buchnera aphidicola, Serratia symbiotica* and unclassified Enterobacteriaceae) and potential core bacteria *Wolbachia* (purple) or *Oxalobacter* (pink) over all individuals per ant species. Only bacteria belonging to the core microbiome were included in this analysis. For complete reads of individual ant workers and separate OTUs, see Figures S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}. (b) NJ‐tree of Tamura--Nei distances (scale bar) between 293 bp and 1,426 bp bacterial *16S rRNA* fragments showing close phylogenetic relationship between the five clusters of Acetobacteraceae OTUs (here named strains aab_L1‐L5, colour coded) found during three MiSeq runs (MS1, MS2MS3, MS4) in ants in this study, and those previously described from other ant guts (aab1 and aab2 from *Camponotus chromaiodes*, A_53_40_5\_16S from *Linepithema humile,* Hu et al., [2017](#mec14506-bib-0041){ref-type="ref"}; \#128 from *Formica occulta,* Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}). (c) Rooted NJ‐tree (outgroup removed for clarity) based on Tamura--Nei distances (scale bar) of 685 bp *COI* mitochondrial gene fragments (*n *=* *60) shows five species of ants (four *Lasius* and one *Brachymyrmex*)*,* each represented by colour‐coded clades (original *COI* NJ‐tree with sample labels and outgroup in Figure S14 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Four of these species harbour Acetobacteraceae strains aab_L1‐aab_L5 \[Colour figure can be viewed at <http://www.wileyonlinelibrary.com>\]](MEC-27-1898-g003){#mec14506-fig-0003} 2.2. DNA extraction {#mec14506-sec-0004} ------------------- Insect and bacterial DNA for barcoding and microbiome profiling were extracted from whole aphid and mealybug specimens and from ant gasters only. All extractions were performed under the following sterile conditions, with the exception of 30 mealybugs and 114 aphid samples (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}). In the sterile protocol, all extractions were conducted under a flow hood using sterile consumables, to prevent contamination with environmental bacteria. Prior to extraction, all specimens were surface sterilized by immersing them individually for 30 s in 5% bleach, followed by 30 s in autoclaved 1 × PBS solution. Individual aphids, mealybugs and ant gasters (separated using autoclaved disposable razors) were then placed in sterile 1.5‐ml tubes with 180 μl enzymatic lysis buffer (20 mM Tris‐CL, PH 8.0, 2 mM Sodium EDTA, 1.2% Triton X100 and 20 mg/ml lysozyme) and a sterile stainless steel bead (5 mm), followed by homogenization in a QIAGEN TissueLyzer II for 3 min at 30 Hz. Next, we extracted insect and bacterial DNA simultaneously using the QIAGEN DNeasy Blood & Tissue kit, using the manufacturer\'s modified extraction protocol that includes pretreatment for gram‐negative bacteria to prevent extraction bias against these bacteria. One negative control was included in each extraction batch. All samples included in MiSeq run MS1 were extracted using the manufacturer\'s unmodified protocol. In MiSeq run MS4, two honeydew samples were included, as a pilot for screening for the bacterial presence in the honeydew that is transferred from aphids to ants. The samples were collected from *P. longianus* aphids (nest M172) and obtained by lightly touching the aphid\'s abdomen with a minute pin, shortly after the aphids had been collected from the field. This "milking" yielded \~0.5 μl honeydew per aphid, taken up in 0.5 μl Drummond Microcaps^**®**^ microcapillary tubes (Sigma‐Aldrich Co. LLC) and stored at −30°C. For DNA extraction, the samples were, after defrosting, each added to 100** **μl lysis buffer and further processed using the protocol described above. For 30 mealybugs and 114 aphids, mostly stemming from sampling events which yielded only a limited number of individuals and therefore stored in 70% EtOH, we used a modified extraction protocol. This "regular" protocol (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}) allowed for preservation of the sample for future slide mounting for morphological identification if necessary. In this protocol, samples were extracted under standard (nonsterile) conditions, without surface sterilization and without homogenization, to leave the specimen intact for slide mounting. Under these conditions, contamination with environmental bacteria cannot be excluded and, therefore, these samples were not included in MiSeq microbiome profiling and only used for targeted sequencing of specific endosymbionts. 2.3. DNA sequencing {#mec14506-sec-0005} ------------------- Trophobiont and ant mtDNA were amplified using standard primers targeting the *COI* region (mealybugs: primers "Jerry" CI‐J‐2183 & "Ben" C1‐N‐2568, 405 bp; aphids and ants: primers LCO1490 & HCO2198, 605 bp and 685 bp, respectively; Table S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}; Brady, Gadau, & Ward, [2000](#mec14506-bib-0007){ref-type="ref"}; Folmer, Black, Hoeh, Lutz, & Vrijenhoek, [1994](#mec14506-bib-0032){ref-type="ref"}; Simon et al., [1994](#mec14506-bib-0101){ref-type="ref"}). PCR products were subsequently purified and sequenced in both directions using Sanger sequencing, outsourced to Macrogen Inc. (New York, USA). We characterized the insects' bacterial communities based on the *V3V4* (initial MiSeq run MS1) or *V4* region (subsequent MiSeq runs MS2MS3, MS4 and MS5) of the bacterial *16S rRNA* gene, using a protocol modified from Caporaso et al. ([2011](#mec14506-bib-0014){ref-type="ref"}, [2012](#mec14506-bib-0013){ref-type="ref"}). Briefly, we amplified the target region using 2.5 μl aliquots of the extractions as DNA template with 0.5 μl of 10 μM primers (*V3V4*: 16S_V3V4F & 16S_V3V4R, *V4*: 16S_V4F & 16S_V4R, each extended with Illumina overhang adapter sequences for later multiplexing using an Illumina Nextera Index kit; Illumina, [2013](#mec14506-bib-0044){ref-type="ref"}; Kozich, Westcott, Baxter, Highlander, & Schloss, [2013](#mec14506-bib-0055){ref-type="ref"}), 12.5 μl of 2X KAPA HiFi Hotstart Readymix to a total PCR volume of 25 μl, ran at 55**°**C annealing temperature for 26 cycles (34 cycles in MS4). This amplification PCR was followed by an indexing PCR to allow for multiplexing of all samples in a single MiSeq run. These 50 μl PCR cocktails consisted of 5 μl PCR product taken from the amplification PCR as template and 5 μl of forward and reverse Nextera Index barcodes and were run at 55°C annealing temperature for 8 cycles. Samples were then purified for final library construction. Because of their higher yield in bacterial DNA, samples in runs containing mostly mealybug and aphid PCR products (MS2MS3, MS5) were purified and normalized with two rounds of the SequalPrep™ Normalization Plate Kit (ThermoFisher Scientific), followed by a final concentration step using 0.6× Agencourt AMPure XP beads (Beckman Coulter) on pools of 24 samples. For samples in MiSeq runs MS1 and MS4, we pooled index PCR products of eight samples of similar PCR product concentration, estimated based on visual inspection of an electrophoresis gel and then used 0.6× Agencourt AMPure XP beads (Beckman Coulter) only for purification. After DNA quantitation of all purified product pools using Qubit™ (ThermoFisher Scientific), they were further pooled and normalized to equimolar concentrations for sequencing on an Illumina MiSeq sequencer using 250 bp (300 bp in MS1), pair‐end reads at the Rockefeller University Genomics Resource Center. To enable construction of higher resolution phylogenies of endosymbionts found with the MiSeq microbiome profiling, we designed novel primers targeting 700+ bp fragments of bacterial *16S rRNA* of *Buchnera aphidicola* and *Serratia symbiotica* in *Prociphilus* root aphids and *Sodalis* in *Rhizoecus* mealybugs. For *B. aphidicola,* we first used universal eubacterial *16S rRNA* primers (10F & 1507R, Munson et al., [1991](#mec14506-bib-0074){ref-type="ref"}) to obtain longer sequences based on which we could design primers specific to these strains. For *S. symbiotica* and *Sodalis,* we aligned the OTU sequences obtained during our MiSeq profiling to longer sequences of the five closest related sequences published in [genbank]{.smallcaps}. Sequences were then aligned, and primers designed using the respective functions in [geneious]{.smallcaps} ^®^ 9.1 (Biomatters Ltd.) and targeting regions for primer design that were highly similar across clades and would target regions of maximal possible length. This resulted in primers Buch_proF & Buch_proR for *B. aphidicola*, SerPro2F & SerPro2R for *S. symbiotica* and 16SSodF & 16SSodR for *Sodalis* (see Table S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"} for primer details). For *Tremblaya,* we amplified fragments of *rpS15* and the adjacent *rpS15‐16S rRNA* intergenic spacer using previously published primers C‐16S‐F & C‐16S‐R (Baumann, Thao, Hess, Johnson, & Baumann, [2002](#mec14506-bib-0005){ref-type="ref"}; Malausa et al., [2011](#mec14506-bib-0062){ref-type="ref"}). We then amplified endosymbiont DNA under PCR conditions optimized for each specific primer pair (see Table S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"} for PCR conditions). PCR products were then purified and sequenced using Sanger sequencing in both directions by Macrogen Inc. (New York, USA). 2.4. Sequence processing and curation {#mec14506-sec-0006} ------------------------------------- All MiSeq results were analysed using the most recent release of the software package [mothur]{.smallcaps} and reference database [ribosomal database project]{.smallcaps} (RDP) at the time of analysis (MS2M3: [mothur v1.35.1]{.smallcaps}, RDP14; MS4: [mothur v1.36.0]{.smallcaps}, RDP14; MS5: v1.39.1, RDP14 and MS1: [mothur v1.39.5]{.smallcaps}, RDP16) (Cole et al., [2014](#mec14506-bib-0019){ref-type="ref"}; Schloss et al., [2009](#mec14506-bib-0100){ref-type="ref"}). We used a MOTHUR pipeline modified from (Kozich et al., [2013](#mec14506-bib-0055){ref-type="ref"}; Lukasik et al., [2017](#mec14506-bib-0060){ref-type="ref"}). The full, annotated script of the most recent analysis (MS1) can be found in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}. Briefly, pair‐end reads were first joined into contigs. Then, all sequences were curated from sequencing errors by removing all sequences that were 50 bp shorter or longer than the expected product size, showed homopolymers longer than 8, did not well align to the targeted *16S rRNA* reference region or were estimated to be chimeras by [uchime]{.smallcaps} (Edgar, Haas, Clemente, Quince, & Knight, [2011](#mec14506-bib-0028){ref-type="ref"}). In addition, all singletons (or, in MS1, sequences with copy number \<3) were removed, assuming these were sequencing artifacts. This curated set of sequences was then clustered into operational taxonomic units (OTUs) at the 97% level using the "average neighbour" algorithm as implemented in [mothur]{.smallcaps} and identified taxonomically using the [rdp]{.smallcaps} reference. After final removal of all sequences derived from chloroplasts, mitochondria, Archaea or Eukaryota, this analysis resulted in one table per MiSeq run, with read counts per OTU for each multiplexed sample. The purpose of our MiSeq screens was detection of endosymbiont presence in each of our screened host species. After the initial sequence curation, we therefore curated the data further to minimize false positive OTU calls for our samples, which could be caused by one of three technical issues: (i) the presence of contaminants in extraction and amplification reagents (Russell et al., [2017](#mec14506-bib-0095){ref-type="ref"}; Salter et al., [2014](#mec14506-bib-0096){ref-type="ref"}), (ii) sequencing errors resulting in novel OTUs and (iii) "leakage" between multiplexed samples due to sequencing errors in the Illumina overhang adapter sequences (see Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"} for definitions and curation details for each issue). Lastly, samples were omitted as "failed" when their total read number was lower than 10% of the average read number for their type of sample (Figure S2 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). All runs included blank negative controls, technical replicates and biological replicates (samples collected in the same sampling event). MS1 also included a *Cephalotes* ant worker as positive control (not shown). As additional quality check, we confirmed that in all runs, negative controls were mostly blank except for contaminant and "leakage" reads, and technical and biological replicates yielded very similar results. Lastly, we verified successful normalization by checking read number distribution per sample and removed the sample with least reads of each replicate pair from the final results (Figures S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). All forward and reverse sequence pairs generated by the sequencing of insect mtDNA and targeted sequencing of specific endosymbionts were joined into contigs and then manually curated for sequencing errors, trimmed and aligned in [geneious]{.smallcaps} ^®^10.2.2 (Biomatters Ltd.). 2.5. Data analysis {#mec14506-sec-0007} ------------------ Sequence consensus alignments generated in [geneious]{.smallcaps} ^®^ were then used to construct neighbour‐joining distance trees, using the built‐in tree builder function of [geneious]{.smallcaps} ^®^ with the Tamura--Nei distance model and 1,000 bootstrap replicates. In addition, [raxml]{.smallcaps} trees were constructed using nucleotide model GTR gamma, the rapid‐hill climbing algorithm and 100 bootstrap replicates using the [raxml]{.smallcaps} 8.2.11 plug‐in in [geneious]{.smallcaps} ^®^ (Stamatakis, [2014](#mec14506-bib-0109){ref-type="ref"}). Estimating the phylogenies using [raxml]{.smallcaps} instead did not qualitatively alter the conclusions (see Figures S5, S7, S9, S11, S13, S15 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Trees were edited for readability in [figtree v1.4.2]{.smallcaps} (<http://tree.bio.ed.ac.uk/software/figtree/>). Consensus sequences for each alignment as well as the most abundant genotype observed for each reported OTU in our MiSeq analyses were matched against sequences previously deposited in [ncbi genbank]{.smallcaps} using their [blast]{.smallcaps} ^®^ search. The sequences with the maximum "total score" were reported as "closest match," provided the information stored under their accession number was still available. If not, we reported the next sequence listed. For each matching sequence, we reported its source, as well as the % of sequence identity to the queried sequence and its Expect (E)‐value. The E‐value gives the likelihood of the match having occurred by chance given the database size. The closer to 0, the more significant the match is. We estimated host--microbe specificity using the $H_{2}^{\prime}$ network specificity metric, adopted from ecological network theory (Blüthgen, Menzel, & Blüthgen, [2006](#mec14506-bib-0006){ref-type="ref"}; Ivens, von Beeren, Blüthgen, & Kronauer, [2016](#mec14506-bib-0048){ref-type="ref"}). This metric estimates the specificity of a bipartite network of two interacting species groups based on the number of times each species‐to‐species interaction is observed, taking into account the total number of possible interactions. $H_{2}^{\prime}$ values can range from 0 (generalist network) to 1 (specialist network). We estimated $H_{2}^{\prime}$ values for each host--bacteria bipartite network by compiling a host‐by‐bacterium matrix of the observation numbers of each possible combination. Network metrics were then calculated using the software [r]{.smallcaps} version 3.4.1 (R Development Core Team, [2011](#mec14506-bib-0085){ref-type="ref"}) and [r]{.smallcaps} package [bipartite]{.smallcaps} 2.08 (Dormann, Gruber, & Fründ, [2008](#mec14506-bib-0024){ref-type="ref"}). We tested for statistical significance by comparing the observed $H_{2}^{\prime}$ values to those of 10,000 randomized networks of equal size (using <http://rxc.sys-bio.net>; Blüthgen et al., [2006](#mec14506-bib-0006){ref-type="ref"}; Patefield, [1981](#mec14506-bib-0077){ref-type="ref"}). 3. RESULTS {#mec14506-sec-0008} ========== 3.1. Specificity of mealybug symbiotic microbiomes {#mec14506-sec-0009} -------------------------------------------------- First, we asked whether the symbiotic microbiomes of the five *Rhizoecus* species were unique and best explained by phylogeny, or whether their microbiomes were correlated with overlapping social partnerships with farming ants. Using Illumina MiSeq sequencing of *16S rRNA*, we first screened the four most common mealybug species for the presence of internal bacteria. This broad‐scale screen showed that these *Rhizoecus* mealybugs harbour a simple microbiome, solely consisting of endosymbionts. We found three groups of bacterial operational taxonomic units (OTUs) at the 97% level: those belonging to *Tremblaya*, those of *Sodalis* (Gammaproteobacteria) and a single OTU of *Serratia symbiotica* (Figure S2, Tables S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The closest matching bacterial sequences to longer fragments of these three OTUs (obtained with targeted sequencing, see below) currently included in the NCBI nucleotide database all stem from associates of other insects. The closest *Tremblaya* match is from the mealybug *Planococcus ficus* (at 90% identity, E‐value = 0.0), *Sodalis* matches *Sodalis glossinidius* found in the Tsetse fly (95% identity, E‐value = 0.0) and *Serratia symbiotica* matches a known secondary endosymbiont of aphids (99% identity, E‐value = 0.0) (Figure S2, Tables S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}; Chen, Wang, Chen, & Qiao, [2015](#mec14506-bib-0015){ref-type="ref"}; López‐Madrigal, Latorre, Moya, & Gil, [2015](#mec14506-bib-0059){ref-type="ref"}; Matthew, Darby, Young, Hume, & Welburn, [2005](#mec14506-bib-0063){ref-type="ref"}). The MiSeq results already showed variation at the \~300 bp resolution, with several different OTUs belonging to the same bacterial taxon. To examine our question further at the genotype level, we used custom‐designed primers targeting \~700--1,000 bp *16S rRNA* fragments of *Sodalis* and *Serratia*, and *rpS15* and the *rpS15‐16S rRNA* intergenic spacer in *Tremblaya*. The targeted sequencing showed that each of the five mealybug species harbours their own genotype (hereafter referred to as strain) of *Tremblaya* (*n *=* *61 mealybugs) and *Sodalis* (*n* = 59 mealybugs; Figure [2](#mec14506-fig-0002){ref-type="fig"}b,c and Figures S6--S9 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). In addition to these well‐known mealybug endosymbionts, mealybugs belonging to *Rhizoecus* sp. 1 and 2 also invariably harbour species‐specific strains of *Serratia symbiotica* (Figure [2](#mec14506-fig-0002){ref-type="fig"}c, Figure S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). These data demonstrate that the endosymbiotic microbiomes of all five mealybug species are highly species‐specific (mealybug‐endosymbiont bipartite network specificity *H'* ~*2*~ = 1, *p* \< .001). Thus, we found no evidence of a shared symbiotic microbiome among species, nor was there any clustering according to which ant species farmed the mealybugs. This suggests that indeed phylogeny rather than social partnership is correlated with microbiome diversity in mealybugs. 3.2. Specificity of aphid symbiotic microbiomes {#mec14506-sec-0010} ----------------------------------------------- Second, we asked whether the symbiotic microbiome of the farmed aphids followed a similar pattern of species‐specificity or whether social partnerships with the same farming ants played a role. Our aphid MiSeq screening covered the eight most common *Prociphilus* root aphids found in the ants' nests. The results showed that these aphids harbour a maximum of two endosymbionts: *Buchnera aphidicola*, the primary endosymbiont of all aphids and also, like the mealybugs, *Serratia symbiotica,* the secondary endosymbiont known from several other aphids (Henry et al., [2015](#mec14506-bib-0039){ref-type="ref"}). The closest matching sequences of these OTUs stem from other aphids: *Prociphilus ligustrifoliae* (*B. aphidicola,* 98% identity, E‐value = 0.0) and *Stomaphis longirostris* (*S. symbiotica,* 99% identity, E‐value = 0.0; Figure S2, Tables S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The single, rare, sample of *Prociphilus fraxinifolii* harboured an unclassified Enterobacteriaceae, a potential endosymbiont as its closest match is a bacterial endosymbiont of the scale insect *Coelostomidia pilosa* (98% identity, E‐value 1 × 10^−137^; Figure S2 and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Lastly, we observed a handful of isolated cases of likely pathogenic infections with Microbacteriaceae and other actinomycetes (Figure S2 and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Mapping the aphid microbiome to its phylogeny at strain level resolution was again achieved by targeted sequencing of *B. aphidicola* and *S. symbiotica* using custom‐designed primers. The results show that each strain of *B. aphidicola* (*n *=* *218) was limited to a single *COI* clade or species of aphid (Figures [2](#mec14506-fig-0002){ref-type="fig"}f and Figure S10--S13 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Currently, the mitochondrial phylogeny of *P. erigeronensis* (Figure [2](#mec14506-fig-0002){ref-type="fig"}e) remains insufficiently resolved to assess clade‐specificity of *Buchnera* strains. However, each aphid in this clade has one of three types of *Buchnera,* and importantly, each of the aphid mitochondrial haplotypes is always associated with the same *Buchnera* haplotype, suggesting that these might, in fact, represent separate aphid lineages as well. More data are needed to verify this. All included aphids were also screened for the presence of *S. symbiotica*, but only aphids in the clade *P. longianus* A tested positive for this endosymbiont, confirming the initial results of our MiSeq screen (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}, Figures S2 and S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). *Serratia symbiotica* was found in the majority of the aphids in this clade, but not in all of them (22 of 28 screened *P. longianus* A aphids) (Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}, Figures [2](#mec14506-fig-0002){ref-type="fig"}d and Figures S2, S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Together, this suggests that also the aphid endosymbiotic microbiome is characterized by a strong phylogenetic signal, with one‐to‐one clade‐specificity of endosymbiont genotypes (aphid endosymbiont bipartite network specificity *H'* ~*2*~ = 1, *p* \< .001) and no evidence for an effect of social partnerships. 3.3. Specificity of ant symbiotic microbiomes {#mec14506-sec-0011} --------------------------------------------- We next looked within ants and asked whether the five trophobiont‐farming subterranean ant species *Brachymyrmex depilis, Lasius flavus, L. claviger, L. nearcticus* and *L. umbratus* each harbour their own species‐specific microbiome or whether farming of the same aphids and mealybugs results in a shared microbiome across ant species, or with their trophobionts. To address this question, we amplified bacteria from individual ant worker gasters of all species for subsequent *16S rRNA* MiSeq sequencing. Amplification rates varied widely across samples, both between species (Figure S2, S3 and S17 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}) and within‐ and between ant colonies (Figure S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}), suggesting that none of the species consistently contained symbiotic bacteria at significant levels. These generally low bacterial densities were confirmed visually by fluorescent microscopy on SYBR‐green‐stained ant gut extracts in a limited number of samples, following Sanders et al. ([2017](#mec14506-bib-0098){ref-type="ref"}). Owing to these varying amplification rates, we were only able to sequence the microbiome of 129 of the 160 ant workers initially sampled, with the species *Lasius claviger* and *L. umbratus* best represented (Figure [3](#mec14506-fig-0003){ref-type="fig"}a). The symbiotic microbiome sequencing revealed the presence of a relatively simple set of OTUs at the 97% level with significant presence in one or more individuals (Figures S2, S3 and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). After contaminant removal, the remaining OTUs grouped into four categories: noncore OTUs, which were observed only occasionally, potential core OTUs (overall rare OTUs showing significant presence in certain species or ant colonies), core OTUs (making up the majority of reads across species and colonies) and lastly, trophobiont‐derived bacteria. The first and second categories of noncore and potential core OTUs occasionally infect individual workers or ant colonies. The noncore OTUs mostly included *Streptococcus*,*Diplorickettsia, Entomoplasmatales, Spiroplasma* and *Lactobacillus* (Figures S2, S3 and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). All microbiome reads belonging to this category were excluded from further analysis. We also found significant presence of *Wolbachia* (exclusive to both included *B. depilis* workers) and *Oxalobacter* (four *L. claviger* workers from two nests), which were therefore deemed potential core OTUs (Figures [3](#mec14506-fig-0003){ref-type="fig"}a and Figures S2, S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The third category, the core OTUs, contains members of the sugar‐processing Acetobacteraceae (aab). These bacteria were consistently found in *Lasius* ant gasters, in all three MiSeq runs that included ant samples (MS1, MS2MS3, MS4) (Figure [3](#mec14506-fig-0003){ref-type="fig"}). Phylogenetic comparison of all Acetobacteraceae OTUs found in the various runs shows that these OTUs cluster together into five groups, hereafter called strains aab_L1‐L5 (Figure [3](#mec14506-fig-0003){ref-type="fig"}b). The closest [ncbi genbank]{.smallcaps} matches to these OTUs are to Acetobacteraceae found in other ants with sugar‐rich diets. Inclusion of these sequences in our phylogeny reveals close relationship between Acetobacteraceae found across these different ants, with aab1 and aab2 from *Camponotus chromaiodes* matching closer than 97% to our observed strains aab_L1 and aab_L2, respectively, and thus being the same strains, under the 97% identity definition, as those found in *Lasius* ants (Figure [3](#mec14506-fig-0003){ref-type="fig"}b and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}; Brown & Wernegreen, [2016](#mec14506-bib-0009){ref-type="ref"}; Hu et al., [2017](#mec14506-bib-0041){ref-type="ref"}; Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}). Interestingly, the three most common ant species *L. claviger, L. nearcticus* and *L. umbratus* share the same three Acetobacteraceae strains (aab_L1, aab_L2, aab_L3), while the three analysed individuals of *L. flavus* exclusively harboured strains aab_L4 and aab_L5 (Figure [3](#mec14506-fig-0003){ref-type="fig"}a,c). The fourth and final category of bacteria observed in ant gasters is that of potentially trophobiont‐derived bacteria (Figure [3](#mec14506-fig-0003){ref-type="fig"}a, in blue). This category included the occasional observation of an unclassified Enterobacteriaceae, which matches closest to a secondary endosymbiont of giant scale insect *Coelostomidia pilosa* and was also found in the single sample of *P. fraxinifolii* we screened (Table S2 and Figure S2 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}; Dhami, Buckley, Beggs, & Taylor, [2013](#mec14506-bib-0022){ref-type="ref"}). Whether this is indeed a trophobiont--ant‐transferred endosymbiont merits further study. The most notable two bacteria in this category are *Buchnera aphidicola* and *Serratia symbiotica*, which we found at significant levels in at least three ant workers. These were included in MiSeq run MS4, which did not include any aphid or mealybug samples, so the possibility of these reads being the result of "leakage" because of multiplexing can be ruled out. In addition, the presence of *S. symbiotica* was confirmed in these ant samples using our specific *Serratia* primers and the sequence matched those observed in *P. longianus* A samples most closely (Figure S16 and Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Whether these are transient bacteria that the ants obtained from their trophobionts directly, either by preying on them or from transfer via honeydew, or whether these observations indicate colonization of ants by these bacteria is unknown. The two honeydew samples, collected from *P. longianus* B, included in our microbiome analysis did not show significant presence of any of the OTUs observed in the ants or their trophobionts (Figure S2 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). With *B. aphidicola* being an obligate endosymbiont of aphids, its occasional occurrence in ant workers is therefore best explained by recent consumption of aphids. Thus, we observe indications of both microbiome species‐specificity (the $H_{2}^{\prime}$ of the ant‐microbiome network being 0.421 (*p* \< .0001), which is mostly driven by the apparent exclusive association of the two rare ant species *B. depilis* and *L. flavus* with *Wolbachia* and two unique strains of Acetobacteraceae, respectively) and influence of social partnerships (e.g. three Acetobacteraceae strains shared among the other three *Lasius* species) on the microbiome composition of these five ant species (Figure [3](#mec14506-fig-0003){ref-type="fig"}a,c). Interestingly, the three *Lasius* species that share Acetobacteraceae strains have also been found to farm the same two mealybug species (*Rhizoecus* spp. 3 and 4), indicating more similar social partnerships among these species than between them and *B. depilis* and *L. flavus*, which have never been found to farm *Rhizoecus* spp. 3 and 4 (Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The small sample sizes of the latter two ant species, however, do not allow us to draw any firm conclusions regarding this potentially interesting correlation. 4. DISCUSSION {#mec14506-sec-0012} ============= We took advantage of a recently characterized subterranean symbiosis in which five species of ants farm multiple species of trophobionts (mealybugs and aphids) inside their nests, often simultaneously (Figure [1](#mec14506-fig-0001){ref-type="fig"} and Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). We first asked whether trophobionts that are farmed by similar hosts have similar microbiomes, or whether their microbiomes are correlated with their respective phylogenies. This latter hypothesis is expected given the ancient relationships between the trophobionts and their endosymbionts (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}). Our results confirmed that in the trophobionts, phylogeny is the key correlate. Both mealybugs and aphids harbour phylosymbiotic microbiomes that are highly species‐specific, composed of multiple endosymbionts that closely diverged with their insect hosts (Figure [2](#mec14506-fig-0002){ref-type="fig"}). None of the observed endosymbiont strains are shared among different groups or species of trophobionts. Despite the similarity in habitat and available food sources due to being farmed by overlapping ant species in similar, isolated underground environments, we found that the trophobionts' symbiotic microbiome is exclusively correlated with their phylogeny. Next, we asked whether this is also the case for the farming ants. Do the symbiotic microbiomes of different ant species that farm the same trophobionts share core components with each other and their trophobionts, or is microbiome composition species‐specific and correlated with ant phylogeny? We found that the three most common *Lasius* ant species, which all farm the same trophobionts, share a trio of sugar‐processing bacteria (Figure [3](#mec14506-fig-0003){ref-type="fig"}), suggesting a potential role of social partnerships. In addition, a limited data set on two additional ant species, *Lasius flavus* and *Brachymyrmex depilis*, which farm a slightly different set of trophobionts (Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}), indicates that these may harbour bacteria exclusively associated with them, including sugar‐processing bacteria. Despite the potential microbial transfer between the trophobionts and the ants via the trophobionts' honeydew, we also did not find any evidence for consistently shared microbiomes between ants and their trophobionts. The high level of species‐specificity in the trophobionts' symbiotic microbiome is most likely explained by two characteristics: (i) the highly specialized metabolic function of these microbes within their unique hosts and (ii) the predominant, if not exclusive, vertical mode of transmission of the endosymbionts. The majority of the bacterial endosymbionts observed in the trophobionts are closely related to those previously found in related hosts. In these well‐studied examples, the endosymbionts provide vital metabolic functions specific to their hosts (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Husnik & McCutcheon, [2016](#mec14506-bib-0042){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; Oliver et al., [2010](#mec14506-bib-0076){ref-type="ref"}). For example, *Tremblaya* and *Sodalis* endosymbionts were found in all *Rhizoecus* mealybug species. *Tremblaya* is known to perform indispensable metabolic roles, such as production of essential amino acids (Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}). In other systems, both *Tremblaya* and its mealybug host have undergone genome reduction over evolutionary time, making them codependent due to complementary genome function (Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; McCutcheon & von Dohlen, [2011](#mec14506-bib-0064){ref-type="ref"}). This mealybug--*Tremblaya* partnership, in fact, can constitute a hierarchical symbiosis, in which the *Tremblaya* endosymbiont contains its own endosymbiotic Gammaproteobacterium. The best‐studied example is *Morenella,* which fulfils several essential genome functions of its own (von Dohlen et al., [2001](#mec14506-bib-0023){ref-type="ref"}; Husnik et al., [2013](#mec14506-bib-0043){ref-type="ref"}; McCutcheon & von Dohlen, [2011](#mec14506-bib-0064){ref-type="ref"}). Over evolutionary time, these Gammaproteobacteria have repeatedly been replaced, also with *Sodalis*‐like bacteria (Husnik & McCutcheon, [2016](#mec14506-bib-0042){ref-type="ref"}). The trio of perfectly matching mealybug, *Tremblaya*, and *Sodalis* phylogenies reported here is therefore in line with a nested symbiosis of *Rhizoecus* mealybugs and *Tremblaya* endosymbionts, which in turn contain *Sodalis* endosymbionts. While we would need to confirm this physical nestedness using for example fluorescent in situ hybridization (FISH) (von Dohlen et al., [2001](#mec14506-bib-0023){ref-type="ref"}), our results are in accordance with a long history of co‐evolution, with little potential for colonization and sharing of endosymbionts. Similarly, all root aphids in the system contain species‐specific genotypes of *Buchnera aphidicola*, a well‐studied, obligate primary endosymbiont of aphids, well‐known for its vital metabolic functions, most importantly in the production of essential amino acids (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}). The long‐term symbiosis between *Buchnera* and its aphid hosts is further characterized by the evolution of a specific organ inhabited by *Buchnera*, the bacteriocyte (Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Jousselin et al., [2009](#mec14506-bib-0049){ref-type="ref"}). Like in the mealybugs, this integrated, physically "closed" symbiosis leaves little room for frequent exchange or colonization by shared symbionts. The high congruence between host phylogenies, based on mitochondrial *COI* sequences, and the symbiont *16S rRNA* phylogenies, is likely the result of transmission mode. Vertical transmission of the endosymbionts from mother to daughter trophobionts drives the evolution of species‐specific high‐dependency symbiosis (Fisher et al., [2017](#mec14506-bib-0031){ref-type="ref"}). Vertical transmission has long been established as the sole transmission mode in both the mealybug--*Tremblaya* complex and for *Buchnera* in aphids (von Dohlen et al., [2001](#mec14506-bib-0023){ref-type="ref"}; Douglas, [1998](#mec14506-bib-0025){ref-type="ref"}; Jousselin et al., [2009](#mec14506-bib-0049){ref-type="ref"}). In addition to strong endosymbiont‐specificity at the mitochondrial level, species‐specificity even at the host nuclear genetic level is likely aided by predominant, if not exclusive, asexual reproduction in all focal species of mealybugs and aphids. This is because the absence of a sexual cycle will reduce nuclear genetic recombination levels and, therefore, preclude novel endosymbiont‐nuclear gene combinations. During our 5 years of field work (2013--2017), trophobiont males were never observed (A. B. F. Ivens, personal observation). Mealybugs often reproduce exclusively clonally, with males being only rarely observed or absent (Ross & Shuker, [2009](#mec14506-bib-0089){ref-type="ref"}). Likewise, aphids are cyclic parthenogens, which commonly become fully parthenogenetic when they forego their annual sexual cycle (Ivens, Kronauer, Pen, Weissing, & Boomsma, [2012b](#mec14506-bib-0047){ref-type="ref"}; Simon, Rispe, & Sunnucks, [2002](#mec14506-bib-0102){ref-type="ref"}). The subterranean lifestyle provides a relatively stable environment, and being associated with ants may have further facilitated the loss of the sexual cycle in aphids (Ivens, [2015](#mec14506-bib-0045){ref-type="ref"}; Ivens et al., [2012b](#mec14506-bib-0047){ref-type="ref"}; Law & Lewis, [1983](#mec14506-bib-0056){ref-type="ref"}; Wulff, [1985](#mec14506-bib-0114){ref-type="ref"}). Surprisingly, we observed an additional species‐specific endosymbiont, *Serratia*, to be present in some clades of both aphids and mealybugs. While *Serratia symbiotica* is a known secondary, facultative, endosymbiont of other aphids, most notably of Lachnids (Burke, Normark, Favret, & Moran, [2009](#mec14506-bib-0012){ref-type="ref"}; Henry et al., [2015](#mec14506-bib-0039){ref-type="ref"}; Russell, Latorre, Sabater‐Muñoz, Moya, & Moran, [2003](#mec14506-bib-0092){ref-type="ref"}), our study is the first to observe *S. symbiotica* in mealybugs. The three observed *Serratia* strains cluster together, away from *Serratia* known from five other aphid hosts (Figure S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Importantly, our finding of all individuals of *Rhizoecus* spp. 1 and 2 invariably harbouring species‐specific *Serratia* genotypes indicates a potential long‐term association of these mealybugs with this third endosymbiont, meriting further exploration. In some aphids, this bacterium has become an obligate endosymbiont, providing metabolic functions complementary to the resident *Buchnera* strain and potentially being on its way to replacing *Buchnera* (Burke & Moran, [2011](#mec14506-bib-0011){ref-type="ref"}; Meseguer et al., [2017](#mec14506-bib-0066){ref-type="ref"}; Pérez‐Brocal et al., [2006](#mec14506-bib-0080){ref-type="ref"}). *Serratia* has been shown to confer several ecological advantages such as heat‐stress tolerance and nutritional benefits (Koga, Tsuchida, & Fukatsu, [2003](#mec14506-bib-0054){ref-type="ref"}; Montllor et al., [2002](#mec14506-bib-0067){ref-type="ref"}; Russell & Moran, [2006](#mec14506-bib-0093){ref-type="ref"}). It is overrepresented in monophagous aphid species and in those feeding on specific *Acer* trees (Henry et al., [2015](#mec14506-bib-0039){ref-type="ref"}). In our study, we found *Serratia* to be exclusively associated with *Prociphilus longianus* clade A aphids, although not all individual aphids belonging to this clade contained *Serratia*. Field observations and preliminary root barcoding results suggest that *Prociphilus longianus* aphids primarily feed on *Quercus*, possibly also *Acer* (A. B. F. Ivens, personal observation). It is conceivable that in this species, *Serratia* confers an ecological benefit in this feeding niche, but more work is needed to clarify the exact root feeding niches of these organisms. One potential explanation for the presence of *Serratia* in both aphids and mealybugs is that it can be horizontally transmitted, even via host plants (Burke et al., [2009](#mec14506-bib-0012){ref-type="ref"}; Henry et al., [2013](#mec14506-bib-0040){ref-type="ref"}, [2015](#mec14506-bib-0039){ref-type="ref"}; Oliver et al., [2010](#mec14506-bib-0076){ref-type="ref"}). With aphids and mealybugs frequently sharing host ant nests, and even nest chambers and, consequently, feeding niches, *Serratia* could then even be transmitted across taxon boundaries (Figure [1](#mec14506-fig-0001){ref-type="fig"}c). However, the strong mitochondrial clade‐specificity of *Serratia* in both aphids and mealybugs (Figure [2](#mec14506-fig-0002){ref-type="fig"} and Figure S16 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}) suggests that if there ever was horizontal transfer, it must have been historical, and since that time, there has been a transition to vertical transmission and strict host‐specificity. Such historic transfer has been shown before in whiteflies harbouring aphid endosymbiont‐like, but diverged, symbionts (Darby, Birkle, Turner, & Douglas, [2001](#mec14506-bib-0021){ref-type="ref"}). In a recently published survey of endosymbionts in above‐ground aphids, there was a trend towards a higher *Serratia* prevalence in ant‐farmed aphid species coinciding with a marked absence of two other facultative aphid endosymbionts, *Hamiltonella defensa* and *Regiella insecticola* (Henry et al., [2015](#mec14506-bib-0039){ref-type="ref"}). This pattern can potentially be explained by *Serratia* conferring nutritional benefits complementary to the protective benefits provided by farming ants, while the other endosymbionts confer benefits redundant with ant protection, such as protection against parasitoids (Henry et al., [2015](#mec14506-bib-0039){ref-type="ref"}). While their natural enemies remain unknown at this point, the matching results of our survey indicate that the same mechanisms may be at play in root aphids, but this would need to be further verified by broader screening of both ant‐tended and nonant‐tended root aphids. In contrast to these highly specific and somewhat diverse microbial associations in the trophobionts, we find little diversity and specificity in the microbiomes of the ant farmers. Instead, the microbiomes of the focal ant species *Brachymyrmex depilis*,*Lasius flavus*,*L. claviger, L. nearcticus* and *L. umbratus* are simple, with limited diversity. Many of our screened ant workers lacked detectable levels of bacteria, and overall, we only observed seven core OTUs (Figure [3](#mec14506-fig-0003){ref-type="fig"}). The observed low bacterial density and diversity in *Lasius* ants were expected given past work on other ant microbiomes, which display similar patterns (Hu et al., [2017](#mec14506-bib-0041){ref-type="ref"}; Moreau & Rubin, [2017](#mec14506-bib-0070){ref-type="ref"}; Ramalho, Bueno, & Moreau, [2017](#mec14506-bib-0086){ref-type="ref"}; Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}, [2017](#mec14506-bib-0095){ref-type="ref"}). While there are exceptions representing a number of highly specialized ant--microbe associations in a handful of specific ant clades (Anderson et al., [2012](#mec14506-bib-0001){ref-type="ref"}; Lukasik et al., [2017](#mec14506-bib-0060){ref-type="ref"}; Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}, [2017](#mec14506-bib-0095){ref-type="ref"}; Sanders et al., [2014](#mec14506-bib-0099){ref-type="ref"}), overall this low density and diversity is in line with previous work (Sanders et al., [2017](#mec14506-bib-0098){ref-type="ref"}). We identified the most prevalent bacteria in the surveyed *Lasius* species as belonging to the Acetobacteraceae, a family of acetic acid‐producing bacteria that thrive in sugar‐rich environments (Figure [3](#mec14506-fig-0003){ref-type="fig"}). These bacteria are generally found in hosts with sugar‐rich diets, such as the honeydew on which these *Lasius* ants predominantly feed (Ano, Toyama, Adachi, & Matsushita, [2008](#mec14506-bib-0002){ref-type="ref"}). Indeed, the observed OTUs are closely related to those previously published from the guts of other ants feeding on carbohydrate‐rich diets, including honeydew‐feeding *Camponotus* carpenter ants, *Formica* wood ants and *Linepithema* Argentine ants (Brown & Wernegreen, [2016](#mec14506-bib-0009){ref-type="ref"}; Hu et al., [2017](#mec14506-bib-0041){ref-type="ref"}; Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}) (Figure [3](#mec14506-fig-0003){ref-type="fig"}, Tables S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). In addition, these bacteria are related to *Asaia* bacteria, recently found in *Pseudomyrmex* and *Tetraponera* ants, which both feed on another sugar‐rich diet of extra floral nectar (Kautz, Rubin, & Moreau, [2013](#mec14506-bib-0052){ref-type="ref"}; Samaddar et al., [2011](#mec14506-bib-0097){ref-type="ref"}). Their prevalence among these specific ants, in combination with shown experimental increase in bacterial density with sugar‐rich diets, suggests that Acetobacteraceae aid in the ants' digestion of sugary honeydew, although more functional work is needed (Hu et al., [2017](#mec14506-bib-0041){ref-type="ref"}). In addition to sugar‐processing abilities, *Asaia* found in *Tetraponera* ants has been suggested to play a role in nitrogen fixation (Samaddar et al., [2011](#mec14506-bib-0097){ref-type="ref"}). This additional functionality could also be present in their relatives colonizing the *Lasius* ants in this study. The ant--Acetobacteraceae symbiotic relationship has previously been suggested to be quite old and specialized (Brown & Wernegreen, [2016](#mec14506-bib-0009){ref-type="ref"}). *Camponotus* strains aab1 and aab2 were shown to be members of a monophyletic Acetobacteraceae clade that is highly specific to ants, in particular the subfamily Formicinae, to which also *Lasius* ants belong. Our observation of these same strains in *Lasius* ants suggests a more wide‐spread distribution than hitherto thought, spanning across several genera within the Formicinae. We did not observe any Acetobacteraceae in the two screened workers of *Brachymyrmex depilis*. Instead, both workers, collected from the same nest, showed exclusive colonization by *Wolbachia*, constituting the only two observations of this bacterium in our study. *Wolbachia* is a common symbiont of ants (Brown & Wernegreen, [2016](#mec14506-bib-0009){ref-type="ref"}; Kautz et al., [2013](#mec14506-bib-0052){ref-type="ref"}; Russell, [2012](#mec14506-bib-0090){ref-type="ref"}), and its potential roles include manipulation of the ants' reproduction as well as beneficial roles such as nutritional aid and protection (Russell et al., [2017](#mec14506-bib-0095){ref-type="ref"}). With *Wolbachia*\'s prior record as an ant symbiont with a large potential to impact the ecology and evolution of its hosts, a possible *Brachymyrmex--Wolbachia* relationship merits further study. In addition, we cannot exclude the possibility that the presence of *Wolbachia* in relatively high abundance (making up around 50% of the sequencing reads, Figure S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}) in these two ant workers precluded detection of any Acetobacteraceae present at lower abundance. Future experiments using Acetobacteraceae strain‐specific primers to screen larger sample numbers of each of the five focal ant species will help resolve this issue. A second observed potential core bacterium was *Oxalobacter*, with significant presence in workers of two *L. claviger* colonies (Figure S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The function of *Oxalobacter* in ants remains unknown, but it has been attributed a beneficial role in humans (Barnett, Nazzal, Goldfarb, & Blaser, [2016](#mec14506-bib-0004){ref-type="ref"}). Contrasting previous findings of strong phylogenetic signal of ant--microbe associations at higher taxonomic levels (Anderson et al., [2012](#mec14506-bib-0001){ref-type="ref"}; Russell et al., [2009](#mec14506-bib-0094){ref-type="ref"}, [2017](#mec14506-bib-0095){ref-type="ref"}), our survey shows only weak phylogenetic signal of microbial occurrence in these trophobiont‐farming ants. Although we find differences at the genus level, with the limited sample of two *Brachymyrmex* workers exclusively containing *Wolbachia*, all ants surveyed within the genus *Lasius* harbour Acetobacteraceae, with three of these strains being shared by several ant species. The only species‐specific microbiome observed within *Lasius* is that of the three screened *L. flavus* ant workers containing two additional Acetobacteraceae strains, while the other species share their three strains. Because of the rarity of this species, our sample size was limited. This finding needs further confirmation using the approach of targeted Acetobacteraceae sequencing described above, but it points to a potentially interesting association. Rather than a phylogenetic signal, this distribution pattern may be best explained by the ants' social partnerships. The three most common *Lasius* species not only share three strains of Acetobacteraceae but also exclusively farm *Rhizoecus* spp. 3 and 4 mealybugs, while *L. flavus* and *Brachymyrex* have only been found associated with *R. *spp. 1, 2 and 5 (Figure S1 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). These differences in food source resulting from different honeydew produced by different mealybugs could then possibly be driving maintenance of different symbiotic microbiomes. In summary, our results confirm a predominant correlation with phylogeny over one with social partnerships in the trophobiont microbiomes and, in contrast, point to a potentially important effect of social partnerships in the formation of the *Lasius* ant microbiomes. These findings are both supported by recent work on the microbial communities of a tropical ant--plant--hemipteran symbiosis in which two different ant species farm the same two scale insect trophobiont species and house them in domatia of the same species of ant--plant (Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}). The effect of social partnerships on trophobiont microbiomes was limited, with the scale insects both harbouring their own microbiome. In contrast to our findings, however, the microbiomes of both ant species were very species‐specific (Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}). This is likely because one of the two focal species belongs to the genus *Cephalotes*, which is known for the strong phylogenetic signal of its microbiome (Sanders et al., [2014](#mec14506-bib-0099){ref-type="ref"}). In contrast, the microbiome of the other ant species, *Azteca*, seems to be more transient and displays lower bacterial abundances. This is more in line with the microbiomes observed in the *Lasius* and *Brachymyrmex* ants in our study (Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}; Sanders et al., [2014](#mec14506-bib-0099){ref-type="ref"}). These differences in phylogeny‐correlated microbiomes versus social partnership‐correlated microbiomes have been described as "closed" versus "open" symbioses. In "closed" symbioses, the start of an association coincides with birth (e.g. vertical transmission of a symbiont). In "open" symbiosis, on the other hand, microbial colonization and birth are decoupled (e.g. horizontal transmission) (Douglas, [2015](#mec14506-bib-0026){ref-type="ref"}). In open cases, the hosts may be colonized by novel microbial partners over their life time, providing scope for a large influence of social partnerships, but perhaps less strict dependency (Fisher et al., [2017](#mec14506-bib-0031){ref-type="ref"}). Yet, even in a system that is mostly marked by closed symbioses, Pringle and Moreau ([2017](#mec14506-bib-0084){ref-type="ref"}) also observed "microbial leakage," where bacterial OTUs overlap between samples originating from the different organisms. This "leakage" can either be explained by a technical issue, namely mistakenly assigning sequencing reads in multiplexed MiSeq runs which include samples stemming from different organisms, or alternatively by ecological microbial transfer in species interactions among different taxa (Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}). Our employed stringent sequence curation protocol (Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"}) was designed to prevent the former technical issue in all multiplexed runs. We could therefore only observe true ant--trophobiont overlap in microbiomes in the single "ant‐only" run MS4, and likely because of this reason, we observed it to a very limited extent only. These observations may indeed be explained by ecological microbial transfer among the focal organisms. For example, transfer can happen when ants prey on their trophobionts. This mode of transfer likely resulted in the remnants of trophobiont‐derived *Buchnera* and *Serratia* we observed in several ant workers (Figure [3](#mec14506-fig-0003){ref-type="fig"}, Figures S2 and S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). Likewise, previous observations of *Serratia* in honeydew‐feeding *Formica* ants were best explained by the ants' predation on their aphid livestock (Sirviö & Pamilo, [2010](#mec14506-bib-0105){ref-type="ref"}). Alternatively, transfer could occur when ants consume trophobiont‐produced honeydew that contains bacteria (Leroy et al., [2011](#mec14506-bib-0058){ref-type="ref"}). We did, however, not detect any bacteria in the two screened honeydew samples (Figure S2 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"}). The predominant absence of trophobiont‐derived symbionts in the screened ants matches previous findings in which a suite of trophobiont (lycaenids, Sternorrhyncha) symbionts was found to be absent from ants (Russell et al., [2012](#mec14506-bib-0091){ref-type="ref"}). Regardless of the mode of transfer, merely observing bacterial presence by *16S rRNA* sequencing does not distinguish passive transfer of bacteria through consumption from active colonization of the host ant\'s gut by bacteria. Future studies employing dietary manipulation of ants in combination with FISH microscopy could aid distinguishing between these two possibilities. Interestingly, Pringle and Moreau ([2017](#mec14506-bib-0084){ref-type="ref"}) observed most OTU overlap between one species of scale insects and the plant domatia it was housed in, suggesting that the physical environment can play a role in microbiome formation. In future, we will include samples from trophobiont chamber walls in *Lasius* nests to further test this idea. This study suggests that the role of social partnerships in shaping a host\'s symbiotic microbiome is variable and likely dependent on whether the partnership is a "closed" symbiosis, with strict vertical transmission (Douglas, [2015](#mec14506-bib-0026){ref-type="ref"}; Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}). This is in line with previous studies finding a large effect of social partnerships on microbiome formation in the case of "open" symbioses where microbes are acquired from the environment (Lax et al., [2014](#mec14506-bib-0057){ref-type="ref"}; Pringle & Moreau, [2017](#mec14506-bib-0084){ref-type="ref"}; Song et al., [2013](#mec14506-bib-0106){ref-type="ref"}). The species‐specific patterns we observed may also be the result of historic events derived from shared social partnerships. Because in ant--trophobiont relationships, both animal hosts are nutritionally so intricately connected, it is potentially the microbiome that sets the boundaries of viable mutualisms, with only those ant--trophobiont partnerships persisting that harbour complementary microbiomes. Future studies, including those employing experimentally manipulated ant and trophobiont microbiomes, will be able to shed further light on the emerging question of whether it is, in fact, the microbes that indirectly govern maintenance of the higher level animal‐animal social partnerships in which they play such essential nutritional roles. DATA ACCESSIBILITY {#mec14506-sec-0014} ================== All DNA sequences of OTUs and strains as reported in Table S3 in Appendix [S1](#mec14506-sup-0001){ref-type="supplementary-material"} are accessible via [genbank]{.smallcaps}, and accession numbers are provided in the respective table.Unique microbial DNA sequences from the NJ‐trees in Figure [2](#mec14506-fig-0002){ref-type="fig"} are accessible via [genbank]{.smallcaps}, and accession numbers are provided in Table S1 in Appendix [S2](#mec14506-sup-0002){ref-type="supplementary-material"} with representative samples.Complete MiSeq count tables, taxonomies, fasta sequence files and [mothur]{.smallcaps} scripts are available from the Dryad Digital Repository: <https://doi.org/10.5061/dryad.t2q12>.All curated sequences, sequence alignments and original NJ and RaxML trees from Geneious are available from the Dryad Digital Repository: <https://doi.org/10.5061/dryad.t2q12>.Network tables and R scripts for the network specificity analyses of the aphid--*Buchnera* and ant--Acetobacteraceae networks are available from the Dryad Digital Repository: <https://doi.org/10.5061/dryad.t2q12>.All raw MiSeq sequencing reads are accessible via the Sequence Read Archive at <https://www.ncbi.nlm.nih.gov/sra/SRP128691>. AUTHOR CONTRIBUTIONS {#mec14506-sec-0015} ==================== A.B.F.I. and D.J.C.K. designed the study, with input from E.T.K.; A.B.F.I. and A.G. collected field samples and performed laboratory analyses; A.B.F.I. performed data analysis; A.B.F.I. wrote the manuscript with support from E.T.K. and D.J.C.K.; D.J.C.K. supervised the project. Supporting information ====================== ######   ###### Click here for additional data file. ######   ###### Click here for additional data file. The authors thank Jacob Russell, Guillaume Chomicki and two anonymous reviewers for their constructive comments on the manuscript; Christoph von Beeren, Leonora Olivos‐Cisneros, Sean McKenzie, Peter Oxley, Piotr Lukasik, Marten van de Sanden and Stephanie Weldon for providing advice and support on laboratory procedures and data analysis; Tim Gale, Mark Wolek and Kate Leitch for providing help in the field, and the Rockefeller University Center for Field Research in Ethology and Ecology and the Cary Institute of Ecosystem Studies for granting access to their grounds for field sampling. We also thank the Rockefeller Genomics Resource Center and Connie Zhao for providing advice and logistic support. The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007--2013) under Grant Agreement 624145, a Niels Stensen Fellowship, and a Rubicon Fellowship 825.13.025 from the Netherlands Organisation for Scientific Research (all three to A.B.F.I). The project was further sponsored by the Iris and Junming Le Foundation and the Rockefeller University Center for Clinical and Translational Science grant \#UL1 TR000043 from the National Center for Advancing Translational Sciences, National Institutes of Health Clinical and Translational Science Award Program and by the Sackler Center for Biomedicine and Nutrition Research, through the generosity of the Sackler Foundation (both to A.B.F.I), Netherlands Organisation for Scientific Research Vidi Grant 864.10.005 (to E.T.K.), European Research Council (ERC) Grant Agreement 335542 (to E.T.K.) and a Rockefeller University Summer Undergraduate Research Fellowship (to A.G.).

1. Introduction {#sec1} =============== Cerebral ischemia/reperfusion (I/R) injury is a complex pathological process in the nervous system, resulting in high disability and mortality worldwide, with significant clinical and socioeconomic impacts \[[@B1]\]. The complex pathobiological mechanisms of this medical problem include inflammation, apoptosis, oxidative damage, and ionic imbalances \[[@B2]\]. Although reperfusion after ischemia is essential for cell survival, it may have numerous negative consequences such as microvascular damage, cell dysfunction, and cell death. It can attract leukocytes and cause the release of several proinflammatory mediators and the induction of microglia and macrophages \[[@B3]\]. Excessive neuroinflammation can increase brain damage and bring about many secondary complications that influence stroke outcome; therefore, anti-inflammation is considered a target for ischemic stroke. A neurovascular unit (NVU), consisting mainly of microvessels, astrocytes, neurons, extracellular matrix, and other types of gliocytes, is defined as a complete functional and structural unit of the brain \[[@B4]\]. Not only do neurons suffer from strokes, but the microvasculature and gliocytes are also involved \[[@B5]\]. Consequently, protecting different cell types simultaneously in the NVU is necessary for I/R injury therapy \[[@B6]\]. Troxerutin, a naturally occurring flavonoid, is known mainly because of its anti-inflammatory, antioxidative, antithrombotic, antineoplastic, and antiapoptotic activities \[[@B7], [@B8]\]. Cerebroprotein hydrolysate (with abundant bioactive peptides) was found to facilitate the distribution of troxerutin and had a positive synergistic effect with troxerutin against acute ischemic stroke \[[@B9]\]. The present study aimed to investigate the protective effects of troxerutin and cerebroprotein hydrolysate injections (TCHis) on oxygen-glucose deprivation and reoxygenation- (OGD/R-) inducing NVU dysfunction and the possible mechanism. 2. Methods {#sec2} ========== 2.1. Animals {#sec2.1} ------------ Adult Wistar rats (3 months old) were purchased from Peking Vital River Laboratory Animal Ltd. Three female rats were mated with one male rat in each cage, and the pregnant females were kept individually. The rat pups were used for further experiments. All experiments were performed in accordance with China\'s Guidelines for Care and Use of Laboratory Animals. 2.2. Primary Cell Cultures {#sec2.2} -------------------------- Primary cells were extracted from the rat pups and routinely cultured in conditioned incubators (37°C/5% CO~2~). The isolation procedure was performed according to the methods used by Xue et al. (2013) \[[@B10]\] and Wang et al. (2015) \[[@B11]\]. Primary cortical neurons (N) were prepared from Wistar newborn rats (less than 24 h). In brief, the cerebral cortex was digested with 0.125% trypsin for 10 min at 37°C and the cell suspension was passed through a 75 *μ*m pore filter. Cells were harvested and seeded on poly-D-lysine (Sigma Aldrich, MO, USA) precoated plates in Neurobasal Medium (Invitrogen, CA, US) containing 2% B27 supplement (Invitrogen), 1% penicillin--streptomycin, and 2 mM L-glutamine. Experiments were performed for 8 days*in vitro*. Primary astrocytes (A) were extracted from 1- to 2-day-old rat pups, as described previously with a few modifications (Saini MG et al., 2011). In brief, the cerebral cortex was digested with 0.125% trypsin for 10 min at 37°C, and the cell suspension was then passed through a 75 *μ*m pore filter. Cells were seeded in the DMEM/F12 medium containing 10% fetal bovine serum (FBS) (NQBB, Australia) and 1% penicillin--streptomycin. After 7--10 days, the cultures were shaken at 37°C at a speed of 260 rpm for 16 h to remove contaminating microglia and oligodendrocytes. The third passage of astrocytes was used for the following study. Primary brain microvascular endothelial cell (rBMEC) cultures were established from 7-day-old rat pup brain tissues (B), which were extracted and homogenized with type II collagenase/DNaseI (Sigma) for 1 h. Microvessels were separated after density centrifugation (spun at 1000 g/min for 20 min) in 20% bovine serum albumin at 4°C. The microvessels were then digested using a collagenase/dispase solution (Roche Applied Science, Mannheim, Germany) containing DNaseI for 1 h and suspended in a DMEM high-glucose medium containing 20% FBS, 10 ng/mL basic fibroblast growth factor, 30 U/mL heparin, 2 mM glutamine, and 1% penicillin--streptomycin. Cells were then seeded into gelatin (1%) coated flasks, and the third passage of rBMECs was used in this study. 2.3. Establishment of NVU*In Vitro* {#sec2.3} ----------------------------------- The NVU model was established according to the previous report using purified normal morphological cells ([Figure 1](#fig1){ref-type="fig"}) \[[@B12]\]. Briefly, after the neurons had grown in a six-well culture plate with a density of 0.5 × 10^5^ cells/cm^2^ for 2 days, the purified astrocytes with 1.5 × 10^5^ cells/cm^2^ were seeded on the outer side of the insert membrane, which faced the bottom of the well. After 4 h for astrocyte adhering, the insert was placed into the well with neurons. Two days later, rBMECs (1.0 × 10^5^ cells/cm^2^) were seeded in the inner side of the insert membrane. After being cocultured for 3--5 days, the NVU model was prepared for the following experiments (Figures [1](#fig1){ref-type="fig"} and [2](#fig2){ref-type="fig"}). 2.4. Four-Hour Leakage Detection {#sec2.4} -------------------------------- Blood--brain barrier (BBB) permeability was evaluated by performing a 4-hour leakage experiment. After 3 days of coculture, the upper inserts were filled with the medium, while the level of the medium in the plates was maintained 0.5 cm lower than the level of the medium in the upper inserts. Inserts with no cells were used as a control. After 4 h, changes in the level of the medium in the top inserts were observed ([Figure 2](#fig2){ref-type="fig"}). 2.5. Establishment of OGD/R Damaging NVU and TCHi Treatment {#sec2.5} ----------------------------------------------------------- The prepared NVU cells were cultured in a conditional medium \[glucose-free, 98.5 mM NaCl, 10.0 mM KCl, 1.2 mM MgSO~4~, 0.9 mM Na~2~HPO~4~, 6.0 mM NaHCO~3~, 1.8 mM CaCl~2~, 40 mM sodium lactate, and 20 mM HEPES (Sigma) at a pH of 6.8\] and placed in an anaerobic incubator (BINDER CB150, Germany) with conditions of 5% CO~2~, 0.2% O~2~, and 37°C for 2 h (named as Model group). Then, cultures were switched to completely normal conditions with TCHi at concentrations of 10 *μ*M, 100 *μ*M, or 1000 *μ*M for 2 h \[TCHi, Jilin Sihuan Pharmaceutical Co. Ltd., was dissolved in sterile phosphate-buffered saline (PBS)\]. Cells cultured in media with PBS under normoxic conditions were used as a control (named as CK group). 2.6. Transmission Electron Microscopy {#sec2.6} ------------------------------------- The method of using the transmission electron microscope (TEM) was similar to what Liu et al. (2011) \[[@B13]\] and Garbuzova-Davis et al. (2007) \[[@B14]\] mentioned in their study. The cocultures were postfixed in 1% osmium tetroxide (Electron Microscopy Sciences, Inc., PA, USA) in 0.1 M PB for 1 h at room temperature. Following osmication, they were dehydrated in a graded series of acetone dilutions (30%, 50%, 70%, and 95% acetone in water), allowing 10 min for each change. Three 10 min changes in 100% acetone were made, and the cocultures were transferred to a 50 : 50 mix of acetone: LX112 epoxy resin embedding mix (Ladd Research Industries, Burlington, VT). Subsequently, the cocultures were infiltrated with this mix for 1 h under vacuum and were transferred to a 100% LX112 embedding mix and infiltrated for 1 h on a rotator. Two more 1 h infiltration steps were performed with fresh changes of the embedding mixture. Cocultures were further infiltrated in a fresh embedding medium at 4°C overnight. The following day, the tissues were infiltrated in two additional changes of embedding medium at room temperature, 4 h per change, and then embedded in a fresh change of resin in tissue capsules. The blocks were polymerized at 70°C in an oven overnight and then trimmed and sectioned with a diamond knife on a Reichert Ultracut E ultramicrotome (Leica Microsystems, Inc., IL, USA). Thick sections cut at 0.35 m were placed on glass slides and stained with 1% toluidine blue. Thin sections were cut at 80--90 nm, placed on copper grids, and stained with uranyl acetate and lead citrate. The sections were examined and photographed with a Philips CM10 TEM (FEI, Inc., OR, USA) at 60 kV. The arrangement of parenchymal cells and cerebral microvessel endothelial cells on the Transwell filter was detected in the cocultures. 2.7. Western Blotting {#sec2.7} --------------------- Cells in the NVU were individually scraped down and lysed on ice for 10 min. After centrifugation (13,000*g*, 4°C), the resulting supernatant was saved as the cytoplasmic extract sample and the nuclear pellet was prepared for a nuclear extract sample. The samples were separated by 10% sodium dodecyl sulfate--polyacrylamide gel electrophoresis (SDS-PAGE) (P0012A, Beyotime, China) and then transferred to polyvinylidene difluoride membranes. The membranes were blocked for 1 h and incubated overnight at 4°C with the following antibodies: rabbit polyclonal antibody against GAP-43 (1 : 1000, 8945S, CST, China), rabbit polyclonal against AQP-4 (1 : 1000, ab31721, Abcam, China), rabbit polyclonal antibody against Claudin-5 (1 : 1000, ABT45, Millipore, China), and mouse polyclonal antibody against Tubulin (1 : 200, sc-5286, Santa Cruz, China). Membranes were incubated with a secondary goat anti-rabbit/mouse antibody (1 : 3,000, Service, China) for 1 h at 37°C. Immunoreactive bands were observed using the ECL detection system (Bio-Rad, Beijing, China). 2.8. Immunocytochemical Analysis {#sec2.8} -------------------------------- For immunocytochemical analysis, cells were blocked with 10% normal goat serum in PBS containing 0.1% Triton X-100 (Sigma) for 30 min before incubation with primary antibodies for 18 h at 4°C: Rabbit polyclonal antibody against Anti-GAP-43 (1 : 100, 8945S, CST, China), rabbit polyclonal against AQP-4 (1 : 100, ab31721, Abcam, China), rabbit polyclonal antibody against claudin-5 (1 : 100, ABT45, Millipore, China), goat monoclonal antibody against caspase-3 (1 : 100, EB07286, Everest Biotech, UK), mouse monoclonal antibody against P53 (1 : 100, ab26, Abcam, China), and rabbit monoclonal antibody against Bax (1 : 250, ab32503, Abcam, China) at 4°C overnight. The sections were subsequently incubated with Sheep Anti-Mouse IgG H&L secondary antibody (Texas Red) (1 : 500, ab6806, Abcam, China), Donkey Anti-Goat IgG H&L (Alexa Fluor 647) (1 : 200, ab150135, Abcam, China), and Alexa Fluor 488 Monkey Anti-Rabbit IgG (H + L) antibody (1 : 1000, A21206, Life Technologies, Beijing, China) at room temperature for 2 h. Subsequently, the sections were incubated with the DAPI dyeing kit (1 : 500, C0060, Solarbio, Beijing, China) for 15 min. The sections were covered with coverslips for microscopic observation after antibody incubation. Images were acquired using a Nikon Eclipse (TE2000-E) inverted C1 confocal microscope (Nikon Instruments, Minato, Tokyo, Japan) equipped with an oil immersion 60x objective with 1.4 numerical aperture (Nikon) or a Zeiss video microscope (Zeiss AG, Oberkochen, Germany) equipped with a plan Neofluar 10x/0.3 numerical aperture. Antibodies of Bax, P53, caspase-3, and DAPI were used for the triple labeling of apoptosis. 2.9. Enzyme-Linked Immunosorbent Assay {#sec2.9} -------------------------------------- The contents of the tumor necrosis factor-*α* (TNF-*α*), interleukin-1*β* (IL-1*β*), interleukin-6 (IL-6), vascular cell adhesion molecule 1 (VCAM-1), and intercellular adhesion molecule 1 (ICAM-1) of the supernatant were detected by a sandwich enzyme-linked immunosorbent assay (ELISA) according to the manufacturer\'s protocol. Absorbances were measured at 495 nm using a microplate ELISA reader (Bio-Rad Model 680 microplate reader, Multiskan FC, Thermo Fisher Scientific Oy, Vantaa, Finland). Each final value was quantified against a standard curve calibrated with known amounts of protein. 2.10. Statistical Analysis {#sec2.10} -------------------------- All the results were repeated at least three times. All pictures were analyzed by Adobe Photoshop software. Data were statistically analyzed by analysis of variance (using IBM SPSS 17.0). Results were expressed as mean ± SD. *P* \< 0.05 was regarded to be statistically significant. 3. Result {#sec3} ========= Morphology and specific identification for three types of cells in the coculture system: as shown in [Figure 2](#fig2){ref-type="fig"}, pictures of the same field were obtained in visible light under the fluorescent inverted microscope. Neuronal cells were cultured at the bottom of the Transwell filter ([Figure 2(a)](#fig2){ref-type="fig"}), rBMECs ([Figure 2(b)](#fig2){ref-type="fig"}) were seeded on the upper side of the Transwell filter of the inserts, and astrocytes were seeded on the opposite side of the Transwell filter of the inserts ([Figure 2(c)](#fig2){ref-type="fig"}). The schematic drawing of the triple cell coculture system is shown in [Figure 1](#fig1){ref-type="fig"}. The transendothelial electrical resistance (TEER) of different models indicated that the coculture system had an acceptable BBB function ([Figure 2(d)](#fig2){ref-type="fig"}). TEM findings demonstrated that BBB appeared normal in rBMEC; meanwhile, tight junctions and desmosomes were close and adjacent ([Figure 3](#fig3){ref-type="fig"}). 3.1. Effects of TCHi on Cell Survival in NVU Cells after OGD/R {#sec3.1} -------------------------------------------------------------- As shown in [Figure 4](#fig4){ref-type="fig"}, neurons, astrocytes, and rBMECs had typical damage manifestations after OGD/R. However, with the treatment of TCHi at doses of 10 *μ*M, 100 *μ*M, and 1000 *μ*M, these manifestations were weakened. Western blotting was analyzed to confirm the effect of TCHi, and the findings were similar to those of immunocytochemical analysis (Figures [4](#fig4){ref-type="fig"} and [5](#fig5){ref-type="fig"}). 3.2. Effects of TCHi on Inflammatory Cytokines in the NVU Model after OGD/R {#sec3.2} --------------------------------------------------------------------------- OGD/R-induced inflammation was inhibited by TCHi in the NVU model. As shown in [Table 2](#tab1){ref-type="table"}, IL-1*β*, IL-6, and TNF-*α* levels decreased significantly in TCHi 10 *μ*M (*P*\< 0.01) and TCHi 100 *μ*M (*P* \< 0.01). Apart from IL-1*β* and IL-6 (*P* \< 0.01), TCHi 1000 *μ*M showed a more obvious effect on TNF-*α* levels (*P* \< 0.001). As well as regulating inflammatory cytokines, cell adhesion molecules, such as vascular cell adhesion molecule 1 (VCAM-1) and intercellular adhesion molecule 1 (ICAM-1), which have the potential to recruit peripheral leukocytes and other cytokines, were upregulated by OGD/R. ELISA results ([Table 1](#tab2){ref-type="table"}) demonstrated that TCHi 10 *μ*M (*P* \< 0.05), TCHi 100 *μ*M (*P* \< 0.05), and TCHi 1000 *μ*M (*P* \< 0.05) ameliorated the increase in the VACM-1 level significantly. The inhibitory effect on ICAM-1 seemed even more apparent in TCHi 1000 *μ*M (*P* \< 0.01). 3.3. Effects of TCHi on Antiapoptosis in the NVU Model after OGD/R {#sec3.3} ------------------------------------------------------------------ Whether TCHi had effects on apoptosis was further checked. Proapoptotic factors, Bax, p53, and caspase-3, were detected using immunocytochemical analysis. [Figure 6](#fig6){ref-type="fig"} showed that TCHi at concentrations of 10 *μ*M, 100 *μ*M, and 1000 *μ*M suppressed all these factors. 3.4. Effects of TCHi on the Ultrastructure of NVU Cells after OGD/R {#sec3.4} ------------------------------------------------------------------- The ultrastructures of neurons and rBMECs were observed by TEM to confirm the effectiveness of TCHi. [Figure 7](#fig7){ref-type="fig"} indicated that both neurons and rBMECs showed apoptotic signs after OGD/R. Cells exhibited shrinkage of shape, irregular nuclei, diffused distribution of heterochromatin, autophagosome appearance, and so on. At concentrations of 1000 *μ*M, TCHi reversed all these signs. 4. Discussion {#sec4} ============= Our study demonstrated that an injection of TCH could ameliorate OGD/R, inducing NVU dysfunction. The mechanism of TCH treatment was to suppress inflammation and apoptosis. All these findings in the NVU model implied the underlying therapeutic effect of TCHi against cerebral ischemic strokes, which might explain positive clinical findings for TCHi \[[@B9]\]. Neuroinflammation is a complex inflammatory process in the central nervous system, which is thought to play an important defensive role against various pathogens \[[@B15]\]. However, an aberrant inflammatory response is known to be a type of reperfusion injury after a stroke \[[@B16]\]. Several cytokines and chemokines are released after an ischemic brain injury. The most extensively studied of the proinflammatory cytokines include IL-1*β*, IL-6, and TNF-*α* \[[@B16]\]. This study measured the concentrations of IL-1*β*, IL-6, and TNF-*α* in the OGD/R model and found that TCHi has an inhibitory effect on all these cytokines. An inflammatory process consists of both the activation of resident cells of the central nervous system and the infiltration of peripheral leukocytes into the ischemic brain tissue \[[@B17]\]. Cell adhesion molecules are involved in the trafficking and recruitment of leukocytes to activate ischemic endothelia in strokes, which could worsen ischemic brain injuries \[[@B18]\]. ICAM-1 and VCAM-1 are reported to be upregulated by the proinflammatory cytokines TNF-*α* and IL-1*β* \[[@B19], [@B20]\]. The results of the present study revealed that the anti-inflammatory properties of TCHi in an OGD/R-damaged NVU model were due to the reducing levels of ICAM-1 and VCAM-1 as well as cytokines. Apart from inflammation, apoptosis is another focus in the pathogenesis of brain injuries \[[@B17]\]. The regulation of the apoptosis process is essential to maintain the balance between cell survival and cell death, which is also important in ischemic injuries \[[@B21]\]. Proapoptotic proteins, such as Bax, p53, and caspase-3, were reported to take part in I/R injuries earlier \[[@B21]\]. The current study demonstrated that TCHi could be an effective therapy against apoptosis. It has an impact on the regulation of the apoptosis process via decreasing Bax, p53, and caspase-3 levels in an OGD/R-damaged NVU model. Not only is BBB the key component of NVU, but also it is the most important structure to maintain cerebral homeostasis and correct neuronal function \[[@B22]\]. BBB integrity in the present study was reflected in TEER, tight junctions, and other ultrastructures of NVUs. TCHi was found to alleviate the BBB breakdown in the OGD/R model. Previous researchers have confirmed the anti-inflammatory and antiapoptotic roles of troxerutin in chronic diseases and diabetic models \[[@B23], [@B24]\]. The findings of this study revealed that TCHi, a new compound of troxerutin, had a good therapeutic effect in acute attacks such as cerebral I/R injuries. This study still has limitations. Inflammatory cytokines and cell adhesion molecules were found to be upregulated in NVUs for the*in vitro* OGD/R model; however, the infiltration of peripheral leukocytes and other factors in the central nervous system could not be demonstrated. As a result, further*in vivo* studies should be performed to show whether TCHi had the same effects as in the*in vitro* model. 5. Conclusion {#sec5} ============= TCHi protected the main types of cells of NVUs*in vivo* and*in vitro* depending on anti-inflammation, antiapoptosis, and BBB. The data implies that TCHi is a candidate medicine to treat cerebral ischemic stroke. The authors thank Professor Xiaoyu Xu and M.D. Qiang Xue for their technical support for the*in vitro* NVU model. Disclosure ========== Hóngyi Zhào and Yu Liu are co-first authors in this article. Conflicts of Interest ===================== The authors declare that they have no conflicts of interest. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} ###### Expression of cytokines in the cell culture medium of NVU (mean ± SD, *N* = 6). Groups IL-1*β* (pg/mL) IL-6 (pg/mL) TNF-*α* (pg/mL) ---------------- ---------------------- ---------------------- ------------------------ CK 280.09 ± 13.36 30.07 ± 5.81 29.56 ± 3.57 Model 568.13±33.36^\#\#\#^  433.01±20.56^\#\#^ 156.49 ± 17.98^\#\#\#^ TCHi 10 *μ*M 485.57 ± 27.18^*∗∗*^ 387.44 ± 30.66^*∗*^ 88.43 ± 7.00^*∗∗*^ TCHi 100 *μ*M 446.92 ± 18.57^*∗∗*^ 305.90 ± 22.69^*∗∗*^  85.51±2.99^*∗∗*^ TCHi 1000 *μ*M 314.29 ± 20.45^*∗∗*^ 289.75 ± 17.49^*∗∗*^ 36.06 ± 3.48^*∗∗∗*^ ^*∗*^ *P* \< 0.05, ^*∗∗*^*P* \< 0.01, and ^*∗∗∗*^*P* \< 0.001 relative to Model; ^\#\#^*P* \< 0.01 and ^\#\#\#^*P* \< 0.01 relative to CK. ###### Expression of cell adhesion molecules in the cell culture medium of NVU (mean ± SD, *N* = 6). Groups VCAM-1 (pg/mL) ICAM-1 (pg/mL) ---------------- ---------------------- ----------------------- CK 29.03 ± 2.31 21.69 ± 4.40 Model  142.46±6.33^\#\#\#^ 90.31 ± 15.63^\#\#\#^ TCHi 10 *μ*M 119.88 ± 9.17^*∗*^ 73.44 ± 14.61^*∗*^ TCHi 100 *μ*M 104.94 ± 4.13^*∗*^ 45.19 ± 22.69^*∗∗*^ TCHi 1000 *μ*M 104.19 ± 4.81^*∗*^ 38.97 ± 7.46^*∗∗*^ ^*∗*^ *P* \< 0.05 and ^*∗∗*^*P* \< 0.01 relative to Model; ^\#\#\#^*P* \< 0.001 relative to CK. [^1]: Academic Editor: Kuo-Tong Liou

Introduction {#Sec1} ============ Plants are rich and vital source of a large variety of pharmaceutically and industrially important natural metabolites^[@CR1]^. *Stevia rebaudiana* Bert. (Stevia, family-Asteraceae), is a shrubby, perennial plant species^[@CR2]^, popular worldwide for its ability to accumulate considerably high level of several commercially important steviol glycosides (SGs; up to \~20% of total dry weight)^[@CR3],[@CR4]^. These SGs have been used as an alternative natural sweetener and are effective for controlling important modern lifestyle diseases (diabetes, obesity, cardiac blockage and hypertension)^[@CR5],[@CR6]^. Based on carbohydrate moiety and its position, SGs have been classified as Steviosides, Rebaudiosides (A--E), Dulcosides, Steviobiosides, and Rubusosides^[@CR3],[@CR7]^. Being \~300 times sweeter than sucrose with low glycemic index, Stevioside and Rebaudioside-A are among the commercially most popular SGs^[@CR8]^. Therefore, despite being native to South America (Paraguay, Argentina and Brazil), Stevia cultivation has been expanded globally to China, Japan, Australia, Canada, USA and India^[@CR2],[@CR9]^. In India, it is mainly cultivated in Rajasthan, Kerala, Maharashtra, Orissa and Himachal Pradesh, and has been expanded to the other parts of the country^[@CR10]^. SGs synthesis utilize the combined metabolic flux of cytosolic mevalonic acid (MVA) and plastidal methyl erythritol 4-phosphate (MEP) pathways^[@CR11]^ (Figure [S1](#MOESM1){ref-type="media"}). Geranylgeranyl pyrophosphate (C-20), the common precursor for the synthesis of all diterpenoids, is produced by geranylgeranyl pyrophosphate synthase (GGPPS) after condensing four isoprene units. The introduction of ent-cyclization by ent-copalyl pyrophosphate synthase (CPPS) specify the metabolic flux towards ent-diterpenoids such as steviol glycosides. Involvement of ent-kaurane synthase (KS) and ent-kaurane oxidase (KO) leads to the synthesis of ent-kaurenoic acid^[@CR3],[@CR11]^. This ent-kaurenoic acid is the last shared intermediate of SGs and gibberellic acids (GAs) biosynthesis. The action of two different ER-membrane located cytochrome P450 monooxygenases (CYPs): ent-kaurenoic acid hydroxylase (KA13H) and ent-kaurenoic acid oxidase (KAO), results in the formation of steviol and GA12, respectively^[@CR12]^. Further, cytosolic glycosylation of steviol by four cytosolic UDP-glucosyltransferases (UGTs) gives rise to different types of steviol glycosides, while, GA12 acts as a precursor for the synthesis of all kinds of GAs^[@CR13]^. The shared synthesis route with GAs and involvement of multiple cellular compartments makes SGs biosynthesis a more complex process. Several physiological and phytochemical studies indicate the higher accumulation of SGs in vegetative phase till appearance of floral bud followed by a declining in flowering phase^[@CR14],[@CR15]^. Although, change in SGs content has been highlighted in previous studies, nonetheless, global molecular mechanism to identify key candidates that influence SGs accumulation during different phases of plant development have not been elucidated, so far. Thus, it becomes imperative to understand developmental phase dependent expression pattern of the genes involed in SGs biosynthesis, and to identify other putative key candidates. *De novo* transcriptome sequencing using various NGS platforms have emerged as a robust, efficient and cost-effective approach to understand genome-wide expression patterns in non-reference plants^[@CR16],[@CR17]^. In the current study, global transcriptome sequencing approach was adopted to understand the effect of developmental phase transitions on the expression of the genes required for SGs biosynthesis. Further, efforts were also made to identify and classify putative candidates such as CYPs and UGTs that assist the modification and diversification of secondary metabolism. Further, Protein-protein interactome (PPI) network analysis of CYPs and UGTs with genes involved in SGs biosynthesis was performed to identify the presence of putative hub proteins that may directly or indirectly regulate the SGs accumulation. The current study will provide a comprehensive genomic resource for manipulating SGs accumulation through genetic engineering, and implementation of molecular breeding approaches for dissection of major agronomic traits and varietal improvement programs in Stevia. Results {#Sec2} ======= Illumina sequencing and *de novo* assembly {#Sec3} ------------------------------------------ To study gene expression pattern in the leaf tissues during developmental phase transitions in Stevia (Figure [1](#Fig1){ref-type="fig"}), three cDNA libraries (LV: leaf tissue in vegetative phase, LB: leaf tissue in bud phase, and LF: leaf tissue in flowering phase) were sequenced using Illumina GAIIx platform. After quality assessment and data filtering (removal of low quality, contaminated reads and adaptor sequences), 17,055,744, 14,299,157 and 17,610,069 filtered reads were obtained for LV, LB and LF, respectively. Further, to improve the *de novo* assembly and downstream annotations, in-house (unpublished) high quality filtered reads of young floral bud (B; 18,300,946) and fully bloomed flower tissues (F; 15,027,649) were also included (Table [1](#Tab1){ref-type="table"}). Out of 101.6 million raw reads, a total 82,293,555 filtered reads were *de novo* assembled into 41,262 transcripts with average length, N~50~ and CG content of 922 bases, 1,244 bases and 39.3%, respectively (Figure [S2](#MOESM1){ref-type="media"}, Table [S1](#MOESM1){ref-type="media"}). To validate the quality of *de novo* assembly, mapping of high quality filtered reads to the assembled transcripts resulted a high alignment rate of 86.82% (71,340,904 mapped reads, Table [1](#Tab1){ref-type="table"}). Secondly, alignments of publicly available EST sequences of Stevia (5,548) obtained mapping rate of 95.71%. The raw reads of Illumina sequencing for all the samples have been deposited in National Centre for Biotechnology Information (NCBI) Sequence Read Archive (SRA) with accession number SRP094030 under BioProject PRJNA355055.Figure 1Schematic representation of methodology adopted for comparative leaf transcriptome analysis and function annotation during developmental phase transition. Leaf tissues from each respective node of three biological replicates were pooled together, hence, total six leaf tissues were used for RNA isolation for each developmental phase. Equimolar concentration from six RNA samples was used for library preparation. Abbreviations are as follows: LV (leaf tissues in vegetative phase), LB (leaf tissues in bud phase), LF (leaf tissues in flowering phase), SGs (steviol glycosides), GAs (gibberellic acids), CYPs (cytochrome P450 monooxygenases), and UGTs (UDP-glucosyltransferases). Table 1Characteristics distribution of different types of reads (raw, filtered and mapped) obtained in Illumina sequencing.SamplesTotal raw readsTotal filtered reads% of filtered readsTotal mapped reads% of mapped readsMultiple mapped reads% of multiple mapped readsLV211541081705574480.631474065486.43695210.4LB179088541429915779.841274672789.14622090.4LF212340361761006982.931506902185.57816200.5B227111701830094680.581592633985.21775480.5F185926801502764980.831318995787.77638300.4Total**1016008488229356580.967134090486.823547280.4** Functional annotation and classification of assembled transcripts {#Sec4} ----------------------------------------------------------------- To obtain the comprehensive functional insights of assembled transcripts, five main databases (NCBI's non-redundant, Swiss-Prot, TAIR 10, KEGG and PTF) were used to search homologs of Stevia transcripts using the BLASTx. Out of the total 41,262 transcripts, 29,436 (71.34%), 28,467 (70.0%), 23,154 (56.11%), 8,888 (21.54%) and 5,683 (13.77%) were annotated against NCBI's nr, TAIR10, Swiss-Prot, PTF and KEGG database, respectively, with 2337 transcripts common in all annotations (Figure [2A](#Fig2){ref-type="fig"}). This also revealed highest homology with some of the well-explored plant systems like *Vitis vinifera* (15.92%), *Coffea canephora* (10.91%), *Solanum tuberosum* (6.28%), *Theobroma cacao* (5.50%), *Erythranthe guttata* (4.75%), *Jatropha curcas* (3.95%) and *Populus trichocarpa* (3.92%). However, due to limited genomic information, only 103 transcripts were annotated with *Stevia rebaudiana* entries in NCBI's nr database **(**Figure [2B](#Fig2){ref-type="fig"} **)**.Figure 2Summary of functional annotation of Stevia transcripts. (**A**) Venn diagram representing the abundance of annotations with five different protein databases, (**B**) Histogram representing the species wise homology distribution of Stevia sequences in NCBI's nr protein database annotations. (**C**) Histogram representing three broad categories, Cellular components, Molecular function; and Biological process. X-axis elucidating different GO categories, Y-axis (left) indicating the percentage of annotation to each GO category, Y-axis (right) depicted the scale for GO terms in a single GO category. Gene Ontology (GO) has been widely used to assign functional terms to uncharacterized sequences obtained by transcriptome sequencing^[@CR18]^. A total of 16,853 transcripts were successfully assigned to 65,751 GO terms (47 functional groups), in which 26,656 (40%) classified into 23 categories of biological processes, 24,569 (37%) into 10 functional categories of cellular component, and 14,526 (27%) into 14 functional categories of molecular function (Figure [2C](#Fig2){ref-type="fig"}). More interpretations revealed that the 'cellular process' (GO:0009987) in biological processes, 'cell part' (GO:0044464) in cellular component and 'binding' (GO:0005488) in molecular function were found to be the most abundant functional groups with 8,219, 6,098 and 5,791 GO counts, respectively. For cellular component, a high proportion of annotations were given to 'cell' (GO:0005623) and 'organelle' (GO:0043226); while 'metabolic process' (GO:0008152) and 'response to stimulus' (GO:0050896) were more abundant in biological process. Interestingly, categories which specify the events related to plant development and phase transition like 'developmental process' (GO:0032502), 'reproduction' (GO:0000003), and 'reproductive process' (GO:0022414), were also found significantly represented (Figure [2C](#Fig2){ref-type="fig"}). Annotation with KEGG database can facilitate the biological function of the genes/pathway distributions^[@CR19]^. A total of 5,683 transcripts revealed the significant match with default statistical parameters, and were assigned to 332 different KEGG pathways. Of these, metabolic pathways (819 hits), biosynthesis of secondary metabolites (352), photosynthesis (44), starch and sucrose metabolism pathway (38), plant hormone signal transduction (37) and carbon fixation (24), were the most enriched pathways (Table [S2](#MOESM2){ref-type="media"}). The KEGG pathway analysis also identified 23 pathways representing gene network for secondary metabolites biosynthesis (Table [2](#Tab2){ref-type="table"}). Of these, terpenoid backbone synthesis (30 hits), diterpenoid biosynthesis (8 hits) and monoterpenoid biosynthesis (3 hits) were reportedly involved in SGs biosynthesis. In order to identify the enzymes actively involved in various biological pathways in Stevia, the assembled transcripts were assigned their respective EC number by mapping against the KEGG database. Among them, members of Transferases class with 1563 hits were the most abundant followed by Oxidoreductases (903), Hydroxylases (770), Lyases (326), Ligases (262) and Isomerases (192) (Figure [S3](#MOESM1){ref-type="media"}).Table 2Details of pathways involved in plant secondary metabolites synthesis revealed in KEGG database annotation.S. No.Pathway IdPathway descriptionNumber of hitsNumber of transcripts1.ko00900Terpenoid backbone synthesis30482.ko00130Ubiquinone and other terpenoid-quinone biosynthesis20303.ko00940Phenylpropanoid biosynthesis18654.ko00100Steroid biosynthesis18235.ko00906Carotenoid biosynthesis17236.ko00640Propanoate metabolism14207.ko00941Flavonoid biosynthesis9138.ko00960Tropane, piperidine and pyridine alkaloid biosynthesis9159.ko00950Isoquinoline alkaloid biosynthesis81310.ko00904Diterpenoid biosynthesis8911.ko00905Brassinosteroid biosynthesis5612.ko00945Stilbenoid, diarylheptanoid and gingerol biosynthesis52013.ko00909Sesquiterpenoid and triterpenoid biosynthesis5514.ko00945Stilbenoid, diarylheptanoid and gingerol biosynthesis52015.ko00908Zeatin biosynthesis4416.ko00902Monoterpenoid biosynthesis3517.ko00232Caffeine metabolism2318.ko00944Flavone and flavonol biosynthesis2219.ko00966Glucosinolate biosynthesis2320.ko00903Limonine and pinene degradation2321.ko00523Polyketide sugar unit biosynthesis2222.ko00942Anthocyanin biosynthesis1123.ko00281Geraniol degradation12 Identification of the transcription factors encoding transcripts {#Sec5} ---------------------------------------------------------------- Transcription factors are the key gene regulators that control the expression of their targeted genes in eukaryotes through binding at respective promoter region^[@CR20]^. For understanding the role of different transcription factors in plant metabolism, Stevia transcripts were annotated with Plant Transcription Factor (PTF) database with e-value of 10^−5^. A total of 8,888 transcripts were found to harbor the transcription factor domains which were further classified into 58 transcription factor families. Among these, MYB family (1,053) was the most represented, followed by bHLH (681), MYB-related (577) and NAC (515) families (Figure [3](#Fig3){ref-type="fig"}).Figure 3Pie-chart is representing the details and abundance of transcripts encoding different plant transcription factor family. Identification and classification of CYPs and UGTs {#Sec6} -------------------------------------------------- Recent studies illustrated the role of CYPs and UGTs in diversification of plant secondary metabolisms consequent to specific requirements of plant^[@CR21],[@CR22]^. A total 307 transcripts exhibited homology against 124 different CYP proteins in Swiss-Prot annotation, including KO (CYP 701A3), KAO (CYP 88A3) and KA13H, well-known CYPs reportedly involved in SGs/GAs synthesis. Considering, KO was annotated with two homologs \[(*Arabidopsis thaliana* ent-kaurene oxidase (Q93ZB2) and *Oryza sativa* ent-kaurene oxidase 2 (Q5Z5R4)\], while KAO and KA13H were annotated against O23051 and Q0NZP1, respectively (Table [S3](#MOESM3){ref-type="media"}). Furthermore, a total 118 transcripts were showing annotation for 45 UGTs including all three known UGTs of SGs biosynthesis with Q6VAB0, Q6VAA6 and Q6VAB4 for UDP 85C2, 74G1 and 76G1, respectively (Table [S4](#MOESM4){ref-type="media"}). Global gene expression dynamics of leaf tissues during developmental phase transitions {#Sec7} -------------------------------------------------------------------------------------- Understanding the differential gene expression in tissue-specific manner is a common practice to identify and analyze the important genes/gene networks. For this, the filtered reads from three different RNA-Seq libraries were separately mapped to the assembled transcripts and were further normalized as RPKM (number of reads per kilobase per million mapped reads). Based on RPKM values, global gene expression of LV, LB and LF leaf tissues were classified into four different categories: the transcripts with RPKM value \<2 (low expression), the transcripts with RPKM value 2.1--20 (moderate expression), the transcripts with RPKM value 20.1--100 (high expression), and the transcripts with RPKM value \>100 (very high expression) (Figure [S4A--C](#MOESM1){ref-type="media"}). Transcripts with a high level of expression were maximum in LB (5,978) followed by LF (5,863) and LV (5,622), while transcripts with very high expression were maximum for LF (1,395) followed by LB (1,313) and LV (1,244). The pair-wise differential gene expression analysis with edgeR statistics revealed 4,274, 5,380 and 3,498 transcripts were found differentially expressed in LV *vs* LB, LV *vs* LF, and LB *vs* LF combinations, respectively (Figure [S4D--F](#MOESM1){ref-type="media"}). Interestingly, comparison of LV *vs* LF resulted in a discrete alteration in the DGEs as 3,456 transcripts were up-regulated in LV while such transcripts were only 1,824 in LF. Similarly, in case of LB *vs* LF, a total 2,637 transcripts were up-regulated in LB, while only 861 transcripts showed higher expression in LF tissue (Table [S5](#MOESM5){ref-type="media"}). Identification of genes involved in steviol glycosides biosynthesis pathway {#Sec8} --------------------------------------------------------------------------- SGs are diterpenoid derivatives and share the biosynthesis route with GAs (Figure [4A](#Fig4){ref-type="fig"}). The precursor isoprene unit (5-C) is contributed by the bi-directional cross-talk between two well-characterized pathways: plastidal MEP and cytosolic MVA pathway^[@CR11]^. Annotations of assembled transcripts facilitated the identification of all the genes for these two basic pathways (Table [S6](#MOESM6){ref-type="media"}). The isoprene unit (IPP/DMAPP) is polymerized into the diterpene precursor (GGPP) with the help of a plastidal enzyme GGPPS. Further, ent-cyclization, a process unique to SGs/GAs synthesis is performed by CPPS enzyme. The catalytic action of KS followed by hydroxylation with ER-membrane located CYP protein- KO, results in the synthesis of ent-kaurenoic acid^[@CR3]^. These two crucial enzymes of ent-diterpenoid biosynthesis were also successfully identified in the present study. Ent-kaurenoic acid is the last common intermediate in SGs and GAs synthesis and the introduction of hydroxyl (--OH) group at a different position by the action of two different CYPs segregate this into two different precursor molecules. KA13H, a member of CYPs protein family located at the ER membrane introduces --OH group at 13C position to form precursor skeleton for SGs synthesis known as "steviol". While, the addition of an-OH group at 7^th^ position resulted into the synthesis of GA12 that act as a precursor for all gibberellins (Figure [4A](#Fig4){ref-type="fig"}). Further, Steviol undergoes the process of glycosylation performed by four UGTs to produce an array of SGs. However, GA12 is processed by different types of oxidases to produce different bioactive GAs (GA1, GA3 and GA4). Except for one unknown UGT, all the other genes participate in SGs and GAs biosynthesis were identified in our data (Table [S6](#MOESM6){ref-type="media"}).Figure 4Diagrammatic representation and differential expression pattern of gene(s) involved in SGs/GAs biosynthesis. (**A**) MVA (Cytosolic) and MEP (Plastid)pathways is representing all the genes (Table [S7](#MOESM7){ref-type="media"}), Dotted arrows depicting bioconversions reported *in vitro *only^[@CR13]^, (**B**) Heatmap representing differential gene expression patterns of these genes in different leaf tissues during developmental phase transitions. Abbreviations are as follows: DXS (1-deoxy-D-xylulose-5-phosphate synthase), DXR (1-deoxy-D-xylulose 5-phosphate reductoisomerase), CMS (2-C-methyl-D-erythritol 4-phosphate cytidylyltransferase), CMK (4-diphosphocytidyl-2-C-methyl-D-erythritol kinase), MCS (2-C-methyl-D-erythritol 2,4-cyclodiphosphate synthase), HDS (4-hydroxy-3-methylbut-2-enyl diphosphate synthase), HDR (4-hydroxy-3-methylbut-2-enyl diphosphate reductase), AACT (acetyl Co-A acetyltransferase), HMGS (HMG-CoA synthase), HMGR (HMG-CoA reductase), MK (mevalonate kinase), MPK (phosphomevalonate kinase), MDD (diphosphomevalonate decarboxylase), IDI (isopentenyl-diphosphate delta-isomerase), GGPPS (geranylgeranyl pyrophosphate synthase), CPPS (ent-copalylpyrophosphate synthase), KS (ent-copalyl diphosphate synthase), KO (ent-kaurene oxidase), KA13H (ent-kaurenoic acid 13-hydroxylase), UGT 85C2 (UDP-glycosyltransferase 85C2), UGT 74G1 (UDP-glycosyltransferase 74G1), UGT 76G1 (UDP-glycosyltransferase 76G1), UGT? (unknown UGT), KAO (ent-kaurenoic acid oxidase), GA 20-O (gibberellin 20 oxidase), GA 3-O (gibberellin 3 oxidase), and GA 2-O (gibberellin 2 oxidase). The comparative expression analysis revealed the abundance of transcripts encoding the enzymes, involved in SGs and GAs biosynthesis and were found to be differentially expressed. The rate limiting enzymes, such as HMGR (MVA pathway) and DXS (MEP pathway), were more expressed in LV compared to LB and LF tissues, while other genes showed relatively similar expression pattern during developmental phase transitions. The genes for ent-kaurenoic acid synthesis (KS and KO) showed similar expression, while expression of CPPS was higher in LV. The expression of SGs synthesis specific genes (KA13H, UGT 85C2, UGT 74G1 and UGT 76G1) was comparatively higher in LV as compared to LB and LF tissues. However, genes related to GAs biosynthesis (KAO, GA20O and GA3O) were expressed more in LF as compare to LV and LB tissues (Figure [4B](#Fig4){ref-type="fig"}). To validate the transcriptome data, 14 genes specifically required for SGs and GAs biosynthesis were selected for qRT-PCR analysis. The relative expression of these selected genes was calculated using GAPDH as an endogenous reference gene. qRT-PCR analysis also depicted the similar gene expression patterns as found in RNA-Seq data. The GGPPS was equally expressed while HMGR and CPPS were more in LV as compared to LB and LF tissues. KA13H and three UGTs were more expressed in LV followed by a decreasing trend along with the plant's maturation (in LB and LF leaf tissues), whereas genes for GAs biosynthesis were more expressed in LB and LF tissues as compared to LV (Figure [5](#Fig5){ref-type="fig"}).Figure 5Histograms representing the comparative expression ratios obtained from RNA-seq data and qRT-PCR of key genes involved in SGs/GAs biosynthesis in LV, LB and LF tissues. The X-axis represents different combination of three leaf tissues for comparative expression analysis, Y-axis represents the fold change in RNA-seq and qRT-PCR analysis. Protein-protein interactome network analysis {#Sec9} -------------------------------------------- PPI network have become an effective approach for understanding the complex processes and solving many biological problems such as signaling, pathway identification and prediction of protein functions, and relationships between various kinds of proteins with different functions^[@CR23]^. Considering the role of CYPs and UGTs in plant metabolic diversification^[@CR24]^, 507 transcripts (including CYPs, UGTs and proteins involved in SGs/GAs biosynthesis) represented by 488 TAIR IDs were utilized for PPI network analysis to understand the influence of these diversifying proteins on SGs biosynthesis (Figure [6A](#Fig6){ref-type="fig"}). Network analysis revealed that 183 Stevia orthologs were interacting with 637 nodes having 2153 edges (with average numbers of neighbors: 6.760; clustering coefficient: 0.484) (Table [S7](#MOESM7){ref-type="media"}). Interestingly, we found that AACT and HMGS (the initial enzymes of MVA pathway) were interacting with 64 and 25 neighbors, respectively. DXS (the first enzyme of MEP pathway) was found to interact with 8 other proteins in interactome network. GGPPS, an initial enzyme of SGs/GAs specifying diterpenoid biosynthesis was connected to 9 other proteins and GA20O, GA3O and GA2O were interacting with 1, 5 and 3 neighbors, respectively. The CYP 701A3 (KO) was showing interaction with 4 other proteins. KA13H, specify ent-kaurenoic acid to SGs biosynthesis was interacting with 6 other proteins. CYP 88A3 (KAO) shifts flux towards GAs synthesis, was connected to 8 neighbors including three other CYPs expressing in Stevia (CYP 71A21, CYP 71A22 and CYP 71A25). Interestingly, only two out of three known UGTs involve in SGs biosynthesis were identified in our network analysis. UGT 85C2 was interacting with three (AT1G78270, AT5G12890, AT4G34138) while, UGT 76G1was interacting with two neighbors (AT3G16520 AT3G46670). Absence of UGT 74G1 in the major interactome network may be due to the uniqueness of this UGT to Stevia.Figure 6Protein-protein interactome (PPI) network analysis (**A**) CYPs and UGTs interactions with genes involved in SGs/GAs biosynthesis genes, (**B**) Heatmap representation of CYPs and UGTs with ≥5 interactions and differentially expressed in LV, LV and LF. Further, the CYPs and UGTs having ≥5 neighbours (putative hub proteins) in PPI network analysis and having higher expression (RPKM \>20) were analyzed to find their influences on SGs/GAs biosynthesis during developmental phase transitions (Table [S8](#MOESM8){ref-type="media"}). We found several CYPs (81D3, 72A15, 98A3, 704 A2, 98A3, 77A1 and 82G1) and UGTs (74E2, 92A1 and 74D1) showing significant interactions and higher expression in LV as compare to LB and LF. Similarly, many CYPs (72A3, 71A22, 71A24, 72A8, 706A7, 71B35, 89A5 and 72A14) and UGTs (85A2, 73C4 and 83A1) were highly expressed in LB and LF as compared to LV tissues (Table [S8](#MOESM8){ref-type="media"}). UGT 92A1 was interacting with UGT 85C2 of SGs synthesis and was also expressed more in LV tissue. Moreover, CYP 71A22 was interacting with KAO (the CYP protein that shifts metabolic flux towards GAs synthesis) and showed higher expression in LB and LF (Figure [6B](#Fig6){ref-type="fig"}). Discussion {#Sec10} ========== The genetic resources have been extensively explored for plant systems that are the source of many bioactive metabolites used in pharmaceutical, nutraceutical and flavor industries^[@CR25]^. Considering multiple advantages of NGS sequencing to unravel the molecular/regulatory networks involved in developmental phase transitions^[@CR16],[@CR26]--[@CR28]^, *de novo* transcriptome sequencing approach was adopted to illustrate the mechanisms involve in altering SGs content in leaf tissues during vegetative, budding and flowering phases. A total of 5.92 Gb transcriptome data and *de novo* assembly statistics (41,262 transcripts with an average length of 922 bases, N50; 1,244 bases) used to determine the efficiency of transcriptome sequencing, and was found to be comparable with earlier studies in Stevia^[@CR29],[@CR30]^. The transcripts length (269--12,230 bases) with the abundance of \>1,200 bases long transcripts (8,895 transcripts) (Figure [S2](#MOESM1){ref-type="media"}), signify the presence of complete transcripts in our data. Furthermore, higher percentage of EST mapping (95.71%) and mapped reads (86.82%) to assembled transcripts significantly validate the high quality of *de novo* assembly^[@CR31]^. Functional annotations of 71.34% transcripts with NCBI's nr protein database suggest that a larger part of the data was annotated in this study. Nonetheless, 11,826 transcripts could not find homology in NCBI's nr database, owing to the lack of Stevia specific protein information. Gene Ontology (GO) analysis showed that the abundance of transcripts involved in cellular process, binding and catalytic process categories, including other processes participated in plant developmental events. In KEGG annotation, a total of 23 pathways were found to be actively involved in secondary metabolites synthesis including pathways of SGs biosynthesis (Table [2](#Tab2){ref-type="table"}), complemented the fact that about 15--25% of a plant genome is engaged for encoding proteins/enzymes involved in natural and secondary metabolite biosynthesis pathways^[@CR32]^. Transcription factors are the key gene regulators to control gene expression during developmental phase transitions^[@CR20],[@CR33]^. Predominant expression of members of MYB, bHLH, MYB-related and NAC families in Stevia expressome suggests their vital role in regulating secondary metabolism, cellular morphogenesis and plant growth regulators responsive signaling pathways^[@CR34],[@CR35]^. Except for few small RNAs based studies^[@CR36],[@CR37]^, no transcription factor family expansion has reported in Stevia. However, the members of bHLH (basic helix--loop--helix) family were reported to participate in phytochrome signaling during vegetative to reproductive phase transition in *Arabidopsis* ^[@CR38]^. Similarly, MYC2 regulators (bHLH family) are known to be involved in many plant defense mechanisms in *Nicotiana attenuate* ^[@CR39]^, while many WRKY proteins were involved in secondary metabolism^[@CR40]^, biotic and abiotic stress tolerance^[@CR41]^, trichome development and senescence^[@CR42]^. Likewise, the regulatory role of NAC proteins was also documented in various developmental processes, defense, and abiotic stress responses^[@CR43]^. Both SGs and GAs, follow the common biosynthesis route by consuming 5C unit (IPP/DMAPP) contributed by MVA (cytosolic) and MEP (plastidal) pathway through a bi-directional cross talk^[@CR11]^ (Figure [S1](#MOESM1){ref-type="media"}). With the exception of HMGR and DXS, expression of all the genes involved in MVA and MEP pathways remained unaffected during phase transitions (Figure [4A,B](#Fig4){ref-type="fig"}). However, HMGR and DXS, the initial gene(s) of terpenoid biosynthesis recorded higher expression in LV as compared to LB and LF. Recently, a report in *Medicago truncatula* pointed that stress induced TRITERPENE SAPONIN BIOSYNTHESIS ACTIVATING REGULATOR1 (TSAR1) and TSAR2 (bHLH type) as positive regulators of HMGR governed by physiological and environmental conditions^[@CR44]^. Differential expression pattern of DXS under strict regulation during different developmental phases was also illustrated in the previous studies^[@CR45]^. Regulation of these rate limiting steps of terpenoid biosynthesis possibly essential for maintaining the equilibrium between primary and secondary metabolism during developmental phase transitions. The relatively higher expression of KA13H, UGT 85C2, UGT 74G1 and UGT 76G1, during LV supports the concept of optimum SGs accumulation during vegetative phase. While higher expression of GAs-specific genes (KAO, GA20O and GA3O) in LB and LF suggest the shift of ent-kaurenoic acid flux towards GAs synthesis. This could consequently be the potential cause for the reduction in SGs content during the onset of flowering event. Similar observations were also recorded in previous studies in Stevia^[@CR14],[@CR15],[@CR41],[@CR46]^. Furthermore, the current findings also provide insights into the regulatory mechanism for precise utilization of ent-kaurenoic acid between SGs and GAs biosynthesis. Generally, gibberellins are involved in many processes throughout the plant life-span but their main function has been exclusively studied during shoot apical meristem (SAM) to floral meristem (FM) transition, and altered expression of GAs biosynthesis related genes from vegetative to reproductive phase transitions supports this concept. Secondary metabolites are involved in plant acclimatization during different stresses and developmental events, therefore, synthesis and accumulation of the bioactive molecules are straightly influenced by both physiological and environmental conditions^[@CR47]^. The involvement of CYPs and UGTs in the bioconversion and diversification of secondary metabolism has been one of the considerable interests in the recent past^[@CR22],[@CR24],[@CR25]^. CYPs (EC 1.14.−.−) are monooxygenases that introduce hydroxyl (--OH) group in a regiospecific manner to provide a modification site for further diversification events^[@CR24]^. It is evidenced from the current data that a large proportion of Stevia genome was found to be engaged in the synthesis of such proteins. UGTs (EC 2.4.1.−) are glucosyltransferases and perform the integration of activated nucleotide sugar moieties in an acceptor molecule at specific positions to define their bioactivity, solubility and inter and/or intra-cellular transports^[@CR22]^. Interestingly, 118 transcripts were annotated to encode 45 UGTs involved in several glycosylation processes including SGs biosynthesis. PPI network analysis became a useful way to identify putative hub proteins, and revealing their relatedness and interactive actions during signaling and regulatory mechanisms^[@CR48],[@CR49]^. Using *Arabidopsis* PPI network, we analyzed the interactive influence of CYPs and UGTs on the SGs/GAs biosynthesis which brought about the higher number of interactions for initial enzymes (AACT, HMGS and DXS) of terpenoid biosynthesis, providing insights about the controlled energy flow between primary and secondary metabolism in Stevia. Interaction of KAO (CYP 88A3), which converts ent-kaurenoic acid into GA12, with three other CYPs and their co-expression during reproductive phase (LB and LF) is an indication of their putative role in GAs biosynthesis. Similarly, the interaction of UGT 85C2 with UGT 92A1 and their co-expressive attributes during vegetative phase (LV) suggest their influence on SGs synthesis. Furthermore, CYPs and UGTs with more expression in LV and significant numbers of neighbours (putative hub proteins) in PPI network were found to be involved in various processes, which directly or indirectly helpful for higher SGs biosynthesis in leaf tissues during vegetative phase. CYP 98A3 was reported to be involved in para- and meta-hydroxylation of cinnamates, which are the part of the chemical defense system to protect vegetative tissues from herbivory attacks^[@CR21]^. Likewise, CYP 77A1 was found to be involved in anthocyanin synthesis that may require to cope up with different oxidative stresses^[@CR50]^. These processes are also necessary for proper plant development and therefore, can influence overall SGs content. Contrarily, higher expression of CYP 82C4 and 707A2 during reproductive phase possibly involved in circadian rhythm^[@CR51]^ and ABA catabolism to reduce its antagonistic action against GA-signaling^[@CR52]^, respectively. Hence, expression of these CYPs may have a positive influence on vegetative-reproductive phase transition. Constitutive expression of CYP 707A3 irrespective of phase transition (LV, LB, LF), further signifies its role in plant cell metabolism^[@CR53]^. Furthermore, elevated expression of several UGTs during vegetative phase was known to be involved in cell-cycle regulation and phytohormones signaling, wherein, UGT 92A1 and UGT 74D1 reported to be involved in auxins and cytokinins glycosylation controlling their concentration, and root gravitropism^[@CR54],[@CR55]^. Likewise, UGT 74E2, an H~2~O~2~ induced protein that acts on indole-3-butyric acid (IBA) to maintain auxin homeostasis and signaling, and integrate reactive oxygen species (ROS)^[@CR56]^. While, UGT 85A2, a membrane-associated protein which express predominately in actively dividing cells^[@CR57]^ signifies its role in cell cycle regulation. Conclusion {#Sec11} ========== In this study, comparative leaf transcriptome demonstrates the advantages of high throughput genomics to accelerate the genome-wide ascertainment of the key gene(s) and regulators for the dissection of complex developmental phase transitions involved in SG biosynthesis in Stevia. Coordinated utilization of ent-kaurenoic acid between SGs and GAs synthesis evident by differential expression and quantitative validations of important genes of MVA and MEP pathway also indicates the presence of a mechanism for homeostatic balance between primary and secondary metabolism. Conclusively, developmental phase dependant expression of many genes (HMGR, DXS, KA13H), transcription factors (MYB, bHLH, WRKY, NAC) and different sets of diversifying enzymes (CYPs and UGTs), can be considered as the putative candidates for manipulating SGs content in Stevia. Further, identified CYPs (124) and UGTs (45) can be the potential targets for plant engineering practices, understanding the evolutionary pattern of secondary metabolism and other important pathways in Stevia. These results represent the first step towards dissection of the complex molecular mechanisms involved in SGs biosynthesis in leaf tissue during developmental phase transition in Stevia. This study provides abundant genomic resources and potential candidates for futuristic studies to upscale SG biosynthesis, and implementation of molecular breeding strategies for genetic improvement of this plant species. Materials and Methods {#Sec12} ===================== Plant materials and RNA isolation {#Sec13} --------------------------------- Stevia genotypes (CSIR-IHBT-ST-04) were cultivated under long day (16-hr light/8-hr dark) and 60% humidity conditions at 25 °C in growth chamber (Weiss Technk UK Ltd). Considering the growth and developmental period of Stevia (from May-January)^[@CR58]^, leaves from 1^st^ to 6^th^ successive nodes of three phenotypically healthy plants were harvested at the end of July (LV), in the mid of September (LB), and at the end of October (LF) (Figure [1](#Fig1){ref-type="fig"}). Leaves of each node (1^st^ to 6^th^) from three genotypes were pooled and snap-frozen in liquid nitrogen to store at −80 °C. Total RNA was isolated from each pooled leaf sample of three phases using iRIS method^[@CR59]^. The isolated RNA samples were resolved on 1% denaturing agarose gel to assess their integrity followed by quantification with NanoDrop™ 2000 (Thermo Scientific, Lithuania). cDNA library preparation and Illumina sequencing {#Sec14} ------------------------------------------------ Equimolar concentration from six RNA samples were pooled for respective phases (LV, LB and LF) and was used for cDNA library preparation. In total, three cDNA libraries were constructed using TruSeq RNA library kit (Illumina, USA) as per the manufacturer's instructions. Briefly, magnetic beads with Oligo (dT) were used for isolating mRNA, then the purified mRNA was fragmented into shorter fragments and reverse transcribed with Superscript II Reverse transcriptase (Invitrogen, USA) by priming with random hexamers to synthesize the first strand of cDNA. The second strand was synthesized using DNA polymerase I and the overleft single strands were removed by RNase H treatment. The cDNA was cleaned up using Agencourt® AMPure® XP beads (Backman Coulter, USA). Adapters were ligated to the cDNA molecules after end repair and single nucleotide (A) addition followed by washing to remove excess adaptors. The quality of all the libraries was ascertained using Agilent 2100 Bioanalyzer (Agilent Technologies USA) and quantified using Qubit^®^ 2.0 fluorometer (Invitrogen, USA). An equimolar concentration of the three libraries was used for transcriptome sequencing. Finally, the libraries were sequenced on Illumina GAIIx platform following manufacturer's recommendations to generate 72 bp paired-end reads. Similar sampling and sequencing approach was adopted for generating in-house transcriptome data for young unopened floral bud and full bloomed flower tissues collected during bud phase and flowering phase, respectively. *De novo* sequence assembly, validation and functional annotation {#Sec15} ----------------------------------------------------------------- After Illumina sequencing, raw reads captured in image form were converted to the readable FASTQ format by base calling method using CASAVA package (ver. 1.8.2). High quality reads were obtained after adaptor removal and quality filtering with default parameters (minimum probability for a read to contain zero errors = 75%, minimum average Phred score for a sequence read = 20, and minimum Phred score for each base of a read = 10) using NGS QC Toolkit^[@CR60]^. For improving the quality of *de novo* assembly, filtered reads from in-house transcriptome data (unpublished) for young unopened floral bud and full bloomed flower tissues were also used along with the reads obtained from three libraries (LV, LB and LF). CLC Genomics Workbench (ver. 6.5, CLC Bio, Denmark, <http://www.clcbio.com>) was used to assemble high quality reads with default parameters (trimming quality score = 0.05, similarity fraction = 0.8, mismatch cost = 2, insertion/deletion cost = 3) and a minimum transcript length of 300 bp^[@CR28]^. Further, to validate the quality of *de novo* assembly, we used two deferent approaches^[@CR61]^. Firstly, high quality reads were mapped on the assembled transcripts using Bowtie2 tool (ver. 2.2.4)^[@CR62]^ and secondly, by aligning available EST sequences (<https://www.ncbi.nlm.nih.gov/nucest/?term=stevia%20rebaudiana>) over assembled transcripts using the BLASTn algorithm. For functional annotation, *de novo* assembled transcripts were subjected to the BLASTx algorithm (e-value cut off of ≤1e^−5^)^[@CR63]^ against different databases such as NCBI's nr, Swiss-Prot (<http://www.expasy.ch/sprot>), TAIR 10, and PTF database ver. 3.0 (<http://planttfdb.cbi.pku.edu.cn/>) to retrieve the top hits showing highest sequence similarity. The transcripts having homologs in TAIR10 database were assigned specific GO terms to classify them into three broad categories (biological processes, molecular functions and cellular components) using AgriGO toolkit^[@CR64]^. To identify and characterize the active metabolic pathways in Stevia, the KEGG database (<http://www.genome.jp/kegg>) was used. Identified enzymes were assigned their respective enzyme commission (EC) numbers and further classified into six major classes namely, Oxidoreductases, Transferases, Hydrolases, Lyases, Isomerases and Ligases. Furthermore, CYPs and UGTs, the putative key candidates for diversification of plant metabolism during developmental phase transition^[@CR21]^ were identified from Swiss-Prot (<http://www.expasy.ch/sprot>) annotation followed by their classification in respective families. Statistical analysis and identification of differentially expressed genes {#Sec16} ------------------------------------------------------------------------- To compute the transcript abundance, the filtered reads of three libraries (LV, LB and LF) were aligned individually to *de novo* assembled transcripts using Bowtie2 tool (ver. 2.2.4)^[@CR60]^. The expression level of each transcript was measured in terms of RPKM^[@CR65]^. edgeR, a Bioconductor package based on negative binomial distribution^[@CR66]^, was used to identify differentially expressed genes in the pair-wise comparative analysis of different leaf tissues with false discovery rate (FDR) \< 0.05 and log2 fold change ≥2. Quantitative polymerase chain reaction (qRT-PCR) validation {#Sec17} ----------------------------------------------------------- To support the efficacy of gene expression in RNA-Seq analysis, key genes of SGs and GAs biosynthesis were selected for qRT-PCR validation. Gene specific primers were designed using BatchPrimer3 (<http://probes.pw.usda.gov/batchprimer3/>) and their related information is listed in Table [S9](#MOESM8){ref-type="media"}. Total RNA was isolated from leaf tissues of respective phases (LV, LB and LF) followed by removal of genomic DNA contamination using DNase I (Thermo Scientific, Lithuania) treatment. 2 µg of purified RNA was used for reverse transcription to prepare cDNA using RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Scientific, Lithuania). qRT-PCR was performed with a StepOnePlus™ Real-Time PCR System (Applied Biosystems, USA) in a 20 µl reaction volume containing 200 ng cDNA, Power SYBR® Green PCR Master Mix (Applied Biosystems, USA) and gene-specific primers. GAPDH was used as an internal control to maintain the equality of template in all reactions. Expression analysis of all the genes was performed in triplicate and relative gene expression was calculated by applying 2^−ΔΔCt^ method^[@CR67]^. PPI network analysis {#Sec18} -------------------- Further, to understand the impact and interaction of CYPs and UGTs with proteins involved in SGs biosynthesis, the TAIR annotations of all CYPs, UGTs, and genes involved in SGs and GAs biosynthesis were used for PPI network analysis. For this, a predetermined PPI network of *Arabidopsis thaliana* (AtPIN, <http://ftp.arabidopsis.org/home/tair/Proteins/Protein_interaction_data/Interactome2.0/>)^[@CR68]^, was used as a template due to lack of reference genome of Stevia. Cytoscape software (version 2.8)^[@CR28]^ was used for visualization of PPI network and identification of crucial modules (putative master regulators) after considering the first neighbour of mapped TAIR IDs. It is suggested that if two selected Stevia proteins corresponded to two homologous proteins in the template *Arabidopsis* network, the encoded proteins were also considered to interact with each other in predicted Stevia network^[@CR69],[@CR70]^. The degree of the predicted network was defined as the number of neighbors of each node to identify putative hub proteins^[@CR49]^. Further, the integration of protein interactions and mRNA expression profiles of selected genes were analysed^[@CR17],[@CR49]^ to predict their putative role in different metabolic processes during developmental phase transitions. Electronic supplementary material ================================= {#Sec19} Supplementary information Table S2 Table S3 Table S4 Table S5 Table S6 Table S7 Table S9 **Electronic supplementary material** **Supplementary information** accompanies this paper at 10.1038/s41598-017-12025-y. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The financial support for this work was provided by Council of Scientific & Industrial Research, New Delhi in terms of CSIR-PLOMICS (BSC 301) project. GS acknowledge UGC, New Delhi for Junior Research Fellowship. This is CSIR-IHBT publication no. 4100. R.K.S. conceived and designed the study. G.S., N.G., R.V. performed experiments. G.S., G.D.S., P.S., R.P., analyzed data. S.S., A.K. provided plant materials. M.K.S., A.K.S. provided assistance in Illumina sequencing. G.S. & R.K.S. wrote the manuscript. S.K. helped in data interpretation and improvement of manuscript content. R.K.S. approved the final version of the manuscript. All authors read and approved the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

1.. Introduction ================ An accepted definition of electronic tongue \[[@b1-sensors-06-00019]\] entails an analytical system comprising an array of nonspecific, poorly selective, chemical sensors with cross-sensitivity to different compounds in a solution, and an appropriate chemometric tool for the data processing. From an instrumental point of view, the e-tongue comprises the array of chemical sensors, conditioning and measurement electronic circuits and calibrating and processing tools \[[@b2-sensors-06-00019]\]. Artificial Neural Networks (ANNs) are widely used for the data processing stage. For the analysis with liquid samples, there are two main kinds of electronic tongues, those employing potentiometric sensors \[[@b3-sensors-06-00019]\] and those employing voltammetric sensors \[[@b4-sensors-06-00019]\]. Recently in our laboratory, we have started a research line dealing with the use of e-tongues, mainly for quantitative multidetermination applications \[[@b5-sensors-06-00019]\], although e-tongues can also be used for identification or qualitative purposes \[[@b6-sensors-06-00019]\]. For quantitative multidetermination applications, the followed variant of ANN is the feedforward, backpropagation multilayer perceptron. In this line, soon we realized the high experimental effort involved in the generation of the departure information, i.e. appropriate standards and their corresponding measured responses. This information, needed for the building of the quantitative model, is normally the cumbersome stage in the use of ANNs for quantitative analysis. From this scope, and taking benefit of the automation features, the use of automated analytical systems can be an aid when working with e-tongues. The Flow Injection Analysis (FIA) technique has been scarcely used in this field, both with voltammetric \[[@b7-sensors-06-00019]\] and also with potentiometric sensors \[8\], although it does not bring any added advantage in the preparation of the training solutions. To improve this point, we have recently proposed the use of a Sequential Injection (SIA) system for the work with e-tongues. The SIA technique is an evolution of the FIA technique, in which selected portions of liquids can be retained and mixed by employing a bidirectional pumping device \[[@b9-sensors-06-00019]\]. The developed SIA system has been successfully used with voltammetric \[[@b10-sensors-06-00019]\] and potentiometric \[[@b11-sensors-06-00019]\] e-tongues. Just a single comparable work has been located in the literature, which describes an automated flow system that can be used to prepare mixtures of standards to be used for calibration of a voltammetric e-tongue \[[@b12-sensors-06-00019]\]. Our design is illustrated in [Figure 1](#f1-sensors-06-00019){ref-type="fig"}. The use of the SIA technique permits automated operation of the e-tongue for the preparation of mixed analyte standards, the handling of liquid samples and for the data acquisition during their measurement. In this way, the experimental effort involved in the calibration and operation of e-tongues can be clearly facilitated. The developed SIA e-tongue, when adequately controlled and synchronized, can be used for data acquisition of the sensor signals in steady state conditions and during transient sensor response, thus, adding new dimensions in the information used. Finally, the generated data is processed using a properly trained ANN to obtain the sought application. Different hardware, software, and also processing and programming tools can be applied to e-tongues design \[[@b13-sensors-06-00019]\]. "Virtual instrumentation" technique allows design of custom-made intelligent PC-based instruments called Virtual Instruments (VIs), which manages the PC hardware and others devices or analytical instruments (i.e DMM, oscilloscopes, pH-meters, multifunction DAQ cards) attached by means of the computer bus, serial, GPIB, parallel, or USB ports, etc. The use of this technique under the graphical LabVIEW development environment is an integrated choice to be considered for the e-tongue design just like a single, easy of use, reliable and expandable instrument \[[@b14-sensors-06-00019]\]. 2.. Experimental ================ 2.1. SIA System --------------- The SIA system ([Figure 2](#f2-sensors-06-00019){ref-type="fig"}) consists of two main parts, the flow system and the measurement system (i.e. acquisition system). The flow system comprises a PC controlled Crison 2030 microburette (Crison Instruments, Spain) with a 5 ml syringe (Hamilton, Switzerland), a MVP (Hamilton, Switzerland) 6-way multi-port valve (Ref HVXM R36760), a holding coil between the pumping system and the valve, a mixing cell and a PC controlled magnetic stirrer. The microburette assures the accuracy in the management of the standard solutions and diluting carrier. The multi-port valve allows access to the different ports (solutions). A 7 mL methacrylate mixing cell is employed for the homogenisation of the solutions, which are pumped into the sensor array. The PTFE holding coil tube (Bioblock, France) has an internal diameter of 1 mm and its volume is 5 ml. The system combines standards by performing specific dilutions from a number of appropriate stocks, and conveys the mixture to the sensor array. 2.2. Reagents ------------- All chemicals used for the preparation of the solutions were of analytical grade, and deionised water was used throughout. Potassium chloride and calcium chloride were purchased from Fluka (Switzerland). A 0.1 M potassium chloride (Panreac, Barcelona, Spain) solution was used as carrier. All the solutions were freshly prepared prior to use. Ion-selective electrode (ISE) membranes were prepared dissolving the ionophore, the plasticizer and the polymeric matrix in an organic solvent (Fluka, Switzerland). Two different ionophores for calcium were employed: bis\[4-(1,1,3,3-tetramethylbutyl) phenyl\] phosphate calcium salt (BTMBPPC) and ETH1001 (both from Fluka, Switzerland). The other ionophores had generic response and were sodium salt of the antibiotic tetronasin (TetronasinNa), provided by the University of Cambridge (UK) \[[@b15-sensors-06-00019]\], monensin sodium salt (Acros, Belgium) and lasalocid (Fluka, Switzerland). Plasticizers used were dioctyl phenylphosphonate (DOPP), 2-nitrophenyloctylether (o-NPOE) and dibutyl sebacate (DBS) (all from Fluka, Switzerland). The polymeric matrix was poly(vinyl) chloride (PVC) (Fluka, Switzerland) and the volatile solvent was tetrahydrofuran (THF) (Merk, Germany). Potassium tetrakis-4-chlorophenylborate (PTCPB) (Fluka, Switzerland) was used as additive to formulate certain membranes. Materials used for the preparation of the inner solid contact for the potentiometric sensors were the epoxy resin components Araldite M and HR hardener (Uneco, Barcelona, Spain), and graphite powder (50 μm, BDH Laboratory Supplies, UK) as the conducting filler. 2.3. Control and acquisition system ----------------------------------- The measurement system comprised the detection system (sensor array was incorporated into the system in series) and a Ag/AgCl double-junction reference electrode (Orion 900200). Sensors used in the sensor array were were all solid-state, tubular, flow-through electrodes of normal use in our laboratories when working with flow systems \[[@b16-sensors-06-00019]\]. The scheme of the flow-through tubular sensors is sketched in [Figure 3](#f3-sensors-06-00019){ref-type="fig"}, Their particular design facilitates their placement in series in the flow system. For proper functioning, the inner wall of the axial hole drilled through the inner solid contact is carefully covered with a PVC membrane, which is covering an ohmic solid contact made from a graphite-epoxy resin composite. Once formed, membranes were conditioned in a solution of their primary ion for 24 h. For a proper design of an electronic tongue application, several sensors are needed, and an appropriate selection for the ISE array must be done in order to generate differentiated and cross-term response. [Table 1](#t1-sensors-06-00019){ref-type="table"} summarizes the formulation of the different membranes used in the presented case, which is devoted to alkaline-earth ions. For this reason, membranes with rather selective response to calcium ion plus membranes with generic response to cations have been selected. Reference and ISE signals are passed to conditioning amplifiers (see [Figure 2](#f2-sensors-06-00019){ref-type="fig"}) based on the Burr-Brown INA116 instrumentation amplifier, to ensure that minimal current is drawn during voltage measurement. A low-pass filter was used afterwards for each channel to reduce system noise. A solution grounded electrode was added by the application circuit for ion measurement system \[[@b17-sensors-06-00019]\]. The interface card was a PCL-812PG multifunction I/O DAQ card (Advantech) featuring 16 single--ended gain programmable analog inputs multiplexed to a 12-bit/ 62.5Ksample/sec ADC, and two 16 digital I/O ports. The ADC was triggered by SW via the "measure" command. The system stores voltage samples every 200 ms for each input channel and the obtained voltage value for each measurement corresponds to the mean of 100 acquired samples. The card was interfaced to a 233 MHz Pentium II PC running LabVIEW6.1 under Windows 98. The PC also sends commands to the valve and microburette by RS-232 via COM1&2 ports and to the stirrer (on/off signal) by a digital output line from the multifunction card. The integration of this DAQ card in the LabVIEW environment is accomplished by means of specific drivers made available by the manufacturer. 2.4. Software ------------- The programmed VI commands the whole system (valve, microburette, DAQ card and stirrer). The main program executes a sequence structure which depending on the decoded command, sends appropriate orders to a particular device (each one has an ID number) of the system. For the serial port communications, VISA (Virtual Instrument Standard Architecture) sub-VIs from the LabVIEW libraries were employed. Another sub-VIs from Advantech library were used for handling the DAQ Card (SW trigger multichannel acquisition and digital output). The SW operation is based on reading and decoding user-programmed ASCII text files containing the commands instructions into three hierarchical levels: First Level: Experiments: Tasks for complete a job, i.e. preparing and measuring dilutions from the stocks during a calibration process.Second Level: Tasks: Instructions comprising commands to carry out basic or unitary tasks (system set-up, cleaning the system, preparing a single dilution).Third Level: Commands: Low-level orders sent to the devices (voltage measure, stirrer on/off, delay, open/close valve, etc) which form a specific second level task. Each order is preceded by the device ID number followed by a comma \<n, order\>. In this syntax, n refers to the controlled device in the SIA set-up, 1 for the valve, 2 for the burette, 3 for a wait delay, 4 for data acquisition and 5 for the stirrer. Specific orders for the valve and burette are exactly those given by the manufacturers of each device. The rest of options just indicate the time to remain activated. The scheme of a running experiment is showed in [Figure 4](#f4-sensors-06-00019){ref-type="fig"}. In this example, the system prepares a dilution from the stock solutions, conveys it to the sensor array in a liquid pulse and acquires the potentials from the array as a transient matrix. The final information is processed employing an ANN. The programming and operation of ANN models is to be performed in a separate, subsequent stage, by specific code written in MATLAB 6.5 environment. 3.. Results and discussion ========================== The front panel of the VI is divided in two sections ([Figure 5](#f5-sensors-06-00019){ref-type="fig"}). On the right side, the experimental data from each sensor is graphically displayed. On the left side, the instrument includes a "READY" button and a group of LEDs and string indicators that show the task and command currently on execution. It also displays the file name of the last recording stored. For each measurement, the data collected and the time stamp are stored in spreadsheet file format and are easily exported to other Windows applications like Notepad, EXCEL or SIGMAPLOT. In order to test the system, two procedures were followed. The first one was related to the accuracy of the acquisition system and the second one was related to the reproducibility when preparing liquid samples. Potentiometric direct calibrations and transient recordings were performed using the system for measuring different automatically prepared dilutions. [Figure 6](#f6-sensors-06-00019){ref-type="fig"} shows an example of a calibration run using the five electrodes prepared versus different solutions automatically prepared diluting a calcium ion mother solution. For each one the calibration was performed according to Nernst. In order to extract some validation data of the functioning of the system, the information on [figure 5](#f5-sensors-06-00019){ref-type="fig"} is processed by separating the experimental points in two groups: one group for calibrating the system (regression) and the other group for validating the procedure, by computing the resulting error in a leave-k-out scheme. The results are presented in the [Table 2](#t2-sensors-06-00019){ref-type="table"}, which shows the interpolation error for each sensor, corresponding to 2 standards solutions of different concentration. Observe that the limited errors are below 6%. To check the reproducibility of the system, an arbitrary 2mM calcium standard was prepared in the laboratory, and the same concentration was asked to the system to be prepared from the stock solution. After interpolation in a recent calibration, deviations from the expected value were lower than 5% in both cases, varying on the individual sensor considered. Repetitivity of the SIA system was estimated from 10 replicated dilutions and measurements of the 2mM sample, this one prepared manually and introduced in the system for being presented 10 times to the sensors. For the automatically prepared sample, repetitivity ranged from 1.1-4.2% RSD depending on the sensor, while the repeated measurement of the manually prepared sample showed an interval between 1.8-5.3% RSD. As observed, the precision on the preparation and operation is a main feature of the developed system. More interestingly, as a second experimental possibility, the system is capable of acquiring the transient response after the introduction of a sample plug (step response). [Figure 7](#f7-sensors-06-00019){ref-type="fig"} illustrates a transient recording obtained after the introduction of different Ca^2+^ solutions in a step response experiment, starting from the background electrolyte until the specified concentration level. The recording corresponds to a calcium ion selective electrode, which employs the neutral carrier ETH1001 (Fluka, Switzerland) as the electroactive element in the potentiometric membrane. The processing of the dynamic information contained in the transient response will add a further dimension in the complexity of the processed information, which will improve the ability of the e-tongue to discriminate species in a sample. This concept has been scarcely used in the field of potentiometric e-tongues, although it is of common use in the field of e-noses. This strategy has clear advantages, as an additional dimension of the response is added. In this way, the representation ability of the ANN model is improved with the kinetic features of the response, which can be an aid to solve the more difficult cases. Certain contributions in the chemical literature are precursors of this idea, as the work with biosensor response employing FIA \[[@b18-sensors-06-00019]\]. 4.. Conclusions =============== An electronic tongue based on potentiometric sensors and Sequential Injection Analysis (SIA) with a Virtual Instrument implemented in LabVIEW6.1™ has been developed and successfully tested. The functioning of the system has been improved using a virtual instrument, which just manages text instruction files, previously programmed by the user in order to command the whole system. These instructions can be reutilised in other different experiments. Using this automated approach, the time needed for the generation of the training information of an electronic tongue employing potentiometric sensors can be reduced from weeks to hours, with an added improvement of reproducibility and ease of use. A further advantage is that an additional dimension can be gained to the departure information, the kinetic profile or transient recorded for the sensors after the arrival of the sample that can be used to differentiate better the studied cases. The development of another independent processing VI is now in progress, which performs data compression necessary for training the ANN, where the significant features from the transient response will be extracted by the use of the Fourier or Wavelet transforms. Financial support for this work was provided by MECD (Madrid, Spain) through project CTQ 2004-08134, and by the Department of Universities and the Information Society (DURSI) from the Generalitat de Catalunya. {#f1-sensors-06-00019} {#f2-sensors-06-00019} {#f3-sensors-06-00019} {#f4-sensors-06-00019} {#f5-sensors-06-00019} {#f6-sensors-06-00019} {#f7-sensors-06-00019} ###### Formulation of the ion-selective membranes employed in the construction of the potentiometric sensor array. Selective membrane to Ionophore Additive Plasticizer PVC (%) ----------------------- ------------------- -------------------------------------------------------------- ---------------- --------- Calcium I BTMBPPC (5%) \- DOPP (65%) 30 Calcium II ETH1001 (1%) PTCPB (0.5%) o-NPOE (65.2%) 33.3 Generic I TetronasinNa (1%) PTCPB (0.1)[\*](#tfn1-sensors-06-00019){ref-type="table-fn"} o-NPOE (66%) 33 Generic II Monensin (3%) \- DBS (70%) 27 Generic III Lasalocid (3%) \- DBS (70%) 27 Molar ratio ###### Processing of experimental points in [figure 5](#f5-sensors-06-00019){ref-type="fig"} to validate the system operation. Expected value Sensor 1 Sensor 2 Sensor 3 Sensor 4 Sensor 5 -------------------------------- ---------------- ---------- ---------- ---------- ---------- ---------- **log a~Ca(1st\ Point)~** -2.602 -2.709 -2.723 -2.690 -2.743 -2.754 **Error(%)~(1st\ Point)~** \- -4.1% -4.6% -3.4% -5.4% -5.9% **log a~Ca\ (2nd\ Point)~** -3.204 -3.279 -3.293 -3.300 -3.320 -3.328 **Error**(**%)~(2nd\ Point)~** \- -2.3% -2.8% -3.0% -3.6% -3.8%

Introduction {#s1} ============ Secondhand tobacco smoke (SHS) consists of the side-stream smoke from the burning end of the cigarette, which contains the highest concentration of particulate matter, and the exhaled mainstream smoke [@pone.0034393-First1], [@pone.0034393-Spengler1]. Exposure to SHS is associated with diverse health effects in nonsmokers including heart disease, lung cancer, asthma flares, chronic obstructive pulmonary disease (COPD), and upper airway problems such as sinusitis [@pone.0034393-Eisner1], [@pone.0034393-Kawachi1], [@pone.0034393-Lam1], [@pone.0034393-Taylor1], [@pone.0034393-Eisner2], [@pone.0034393-Eisner3], [@pone.0034393-Pitsavos1], [@pone.0034393-Pitsavos2]. Occupational exposure to SHS presents a substantial health risk to workers [@pone.0034393-Hammond1], [@pone.0034393-Hammond2]. Flight attendants who worked on commercial aircraft before the ban on cigarette smoking (pre-ban FAs) experienced poor air quality and high levels of SHS in aircraft regardless of their class or cabin section [@pone.0034393-Lindgren1], [@pone.0034393-Neilsen1]. A pre-ban era chemical analysis of post-flight urine samples from these FAs showed elevated levels of urinary cotinine (a major metabolite of nicotine) close to levels currently observed in light or experimental smokers [@pone.0034393-Benowitz1], [@pone.0034393-Goniewicz1], which signifies that the FAs had been exposed to substantial levels of tobacco smoke on these aircraft [@pone.0034393-Repace1], [@pone.0034393-Samet1]. We previously showed a cohort of healthy never-smoking pre-ban FAs with significant history of exposure to SHS had abnormal lung function [@pone.0034393-Arjomandi1]. This cohort had curvilinear flow-volume curves (concave to the volume axis) and reduced airflow at mid and low lung volumes. More impressively, over half of the cohort had abnormal single breath carbon monoxide diffusing capacity (DcoSB) below the lower 95% prediction limit based on Crapo\'s reference equations [@pone.0034393-Crapo1]. To further characterize the pulmonary function abnormalities in this cohort of pre-ban FAs, we performed 1-min interval, progressive incremental, symptom-limited cardiopulmonary exercise testing in the supine posture and then measured diffusing capacity at incremental workloads during exercise. Our hypothesis was that since cardiopulmonary exercise is a more sensitive tool for detecting lung function abnormalities, all of pre-ban FAs would have an abnormal exercise response. In particular, we hypothesized that the pre-ban FAs with abnormal resting DcoSB (lower than 95% prediction limit) would have lower pulmonary capillary recruitment with exercise compared to those with normal levels of resting DcoSB (above the lower 95% prediction limit), and that at least some of the FAs with normal levels of resting DcoSB would also show lower exercise-induced increase in their diffusing capacity, indicating that even the FAs with "normal" resting DcoSB might have reduced pulmonary capillary recruitment. Methods {#s2} ======= Ethics Statement {#s2a} ---------------- The UCSF Institutional Review Board (IRB), the Committee on Human Research, approved this study. Written IRB-approved informed consent was obtained from all study participants. Study Design {#s2b} ------------ This was an observational cross-sectional study with convenience sampling of pre-ban FAs. Full details of the study methods are available in ([Methods S1](#pone.0034393.s005){ref-type="supplementary-material"}). Study Population {#s2c} ---------------- Between July 2003 and December 2010, we recruited pre-ban female FAs as part of a clinical investigation of the health effects of the cabin environment on flight attendants employed before and after the ban on smoking on commercial aircraft. Flight attendants were eligible to participate in the study if they had worked for at least five years on aircraft before the airline ban on cigarette smoking, were never-smokers (smoked less than 100 cigarettes lifetime), and had no previous clinical diagnosis of cardiac, pulmonary, or other diseases that could have adversely affected their pulmonary function. All subjects completed health and SHS exposure questionnaires [@pone.0034393-Arjomandi1], had a physical examination, and underwent pulmonary function testing and cardiopulmonary exercise testing. Full details of our methods are available in ([Methods S1](#pone.0034393.s005){ref-type="supplementary-material"}). Cardiopulmonary Exercise testing {#s2d} -------------------------------- Following an explanation of the exercise studies, the subject performed a physician-supervised, symptom-limited progressively increasing exercise test in the supine position on an electromagnetically braked, supine cycle ergometer (Medical Positioning Inc. Kansas City, MO). Subjects were advised to do their best, but otherwise were not encouraged; they could stop voluntarily at any time they believed they could not continue. We continuously monitored heart rate, blood pressure (BP), electrocardiogram (ECG), and breath-by-breath gas exchange. The protocol consisted of 3-min rest, 1-min unloaded (freewheeling) cycling at 60 rpm, followed by increasing work rate of 20--30 Watts to a maximum tolerated, and 5-min of recovery. Twelve lead ECGs were monitored continuously and were recorded along with BP every 2 min. Oxyhemoglobin saturation (O~2~sat) determined by pulse oximetry was recorded continuously. Subjects were rested for 30 min and a repeat exercise study was performed. In this second exercise study, incremental exercise in the supine posture on the same supine cycle ergometer was conducted using 6-min stages at 20, 40, 60, and 80% of maximum observed work, measuring within breath diffusing capacity (DcoWB), pulmonary blood flow (), and HR in duplicate at each stage. Measurements of DcoWB and were performed using a rapid infrared analyzer system via breath-by-breath metabolic measurement as described extensively previously [@pone.0034393-Newth1], [@pone.0034393-Huang1], [@pone.0034393-Martonen1], [@pone.0034393-Wilson1], [@pone.0034393-Ramage1]. Data Management and Analysis {#s2e} ---------------------------- Distributions of subjects\' characteristics (i.e., age, pulmonary function) were computed for all subjects. Measures of pulmonary function at rest as well cardiac and respiratory responses to exercise, based on percent predicted of normal, were calculated and examined. Differences in characteristics, pulmonary function, and exercise responses between the two groups of FAs with resting DcoSB below or above the lower 95% prediction limit were examined using Student\'s t-test. The percent predicted values for DcoWB at 40% maximum observed exercise were calculated using reference equations from Huang et al measured at 40% maximum observed exercise (DcoWB at rest = −0.057\*age+0.221\*height-11.525; DcoWB at 40% exercise = −0.023\*age+0.324\*height-25.273; reference equations for women) [@pone.0034393-Huang1], and Charloux et al (DcoWB = 1.77\*+12.16; reference equation not stratified by gender) [@pone.0034393-Goniewicz2]. Generalized estimating equations were used to compute regression lines for changes in DcoWB and with increasing exercise as well as changes in DcoWB with . The differences in exercise-induced changes in DcoWb and between the groups of FAs were examined using an interaction term in the regression models. Linear regression models were used to examine the association between resting DcoSB, DcoWB at baseline (0% work), and DcoWB at 40% maximum observed work and years of aircraft cabin SHS exposure. To account for other potential cabin factors [@pone.0034393-Lindgren1], [@pone.0034393-Rayman1], the years of SHS exposure (pre-ban years of employment) was adjusted for total years of employment in regression models. The association between lung function and years of cabin SHS exposure were modeled using LOWESS smoother (locally weighted scatterplot smoothing) analysis. Based on these models, a subgroup of pre-ban FAs with history of more than 10 years of cabin SHS exposure was identified and used for further analysis of the associations. All analyses were conducted in STATA (version 12.0). Results {#s3} ======= Subjects characteristics {#s3a} ------------------------ Characteristics of the pre-ban FAs are shown in [Table 1](#pone-0034393-t001){ref-type="table"}. Overall, 80 FAs were recruited for the study. Subjects were all healthy women between the ages of 41 and 76 who were never-smokers as defined by a lifetime history of less than 100 cigarettes use. None of the subjects were obese as defined by a body mass index (BMI)≥30, and none of them had any history of cardiopulmonary diseases or systemic diseases that may affect the cardiopulmonary function. All subjects exercised regularly (3 to 5 times weekly). 10.1371/journal.pone.0034393.t001 ###### Characteristics of pre-ban flight attendants. {#pone-0034393-t001-1} Subject Characteristics All FAs FA with normal Dco FA with abnormal Dco p-value -------------------------------------------------------- ------------ -------------------- ---------------------- -------------- Number 80 40 40 \- Age (years) 60.3±6.7 61.2±6.0 59.4±7.3 0.231 Height (cm) 164±5 164±6 165±5 0.321 BMI (kg/m^2^) 23.7±3.1 23.6±2.8 23.8±3.3 0.794 Hemoglobin level (g/dl) 14.1±1.1 14.2±1.1 13.9±1.1 0.213 DcoSB adjusted for Hgb (ml/min/mmHg) 20.0±3.0 21.7±2.9 18.3±2.1 **\<0.0001** DcoSB adjusted for Hgb (% predicted) 77.0±1.1 84.6±7.6 69.5±5.1 **\<0.0001** DcoSB adjusted for Hgb & alveolar volume (ml/min/mmHg) 4.3±0.5 4.4±0.5 4.1±0.5 **0.004** DcoSB adjusted for Hgb & alveolar volume (% predicted) 81.3±9.5 84.9±9.2 77.8±8.6 **0.0006** FEV~1~ (L) 2.56±0.38 2.60±0.31 2.51±0.43 0.289 FEV~1~ (% predicted) 102.5±13.7 105.8±13.8 99.3±12.6 **0.032** FVC (L) 3.38±0.52 3.46±0.48 3.30±0.55 0.152 FVC (% predicted) 106.8±14.2 110.4±14.1 103.3±13.6 **0.023** FEV~1~/FVC 0.76±0.04 0.75±0.04 0.76±0.05 0.635 TLC (L) 5.14±0.65 5.23±0.66 5.05±0.63 0.270 V~A~ (L) 4.70±0.61 4.91±0.53 4.49±0.61 **0.001** TLC (% predicted) 100.5±10.0 102.3±9.5 98.6±10.3 0.141 V~A~ (% predicted) 91.4±10.3 95.8±8.7 87.1±10.1 **0.0001** Air trapping \[TLC -- V~A~\] (L) 0.46±0.34 0.34±0.39 0.58±0.24 **0.004** Total length of employment (years) 26.8±10.0 26.5±10.6 27.1±9.5 0.840 Pre-ban length of employment (years) 18.4±8.2 18.4±7.6 18.5±8.8 0.940 Data is shown in mean ± standard deviation. Subjects were all female. Abbreviations: BMI: body mass index; DcoSB: single breath diffusing capacity; Hgb: hemoglobin; FEV~1~: forced expiratory volume in 1 second; FVC: forced vital capacity; TLC: total lung capacity measured by body plethysmography; V~A~: alveolar volume measured by single breath helium dilution. Total years of airline employment varied in length between 6 and 40. The FAs\' estimated length of exposure to aircraft cabin SHS (pre-smoking ban employment) was between 3 and 33 years representing a range of 15% to 100% of the total length of their active duty employment. As we had shown previously in a smaller cohort [@pone.0034393-Arjomandi2], half of the subjects had DcoSB at rest below the 95% prediction limit of their normal values for their sex, age, and height, based on Crapo\'s reference equations [@pone.0034393-Crapo1] (40 FAs had "abnormal Dco" and 40 had "normal" Dco). While all FAs had normal FEV~1~, FVC, and FEV~1~ to FVC ratio (FEV~1~/FVC), those with abnormal resting DcoSB had a slightly lower FEV~1~ and FVC but similar FEV~1~/FVC ([Table 1](#pone-0034393-t001){ref-type="table"}). Exercise Capacity {#s3b} ----------------- All FAs achieved a normal maximum exercise level based on maximum work and oxygen uptake (greater than 83% predicted value by Wasserman et al. [@pone.0034393-Wasserman1]). However, the FAs with abnormal resting DcoSB achieved lower levels of maximum exercise as indicated by lower maximum predicted work rate (mean±SD: 123.9±31.6 vs. 132.6±24.2%predicted watts, p = 0.020), and lower maximum oxygen uptake (max) (1.24±0.30 vs. 1.39±0.26 L/min, p = 0.019 and 88.7±2.9 vs. 102.5±3.1%predicted max, p = 0.001) ([Table 2](#pone-0034393-t002){ref-type="table"} & [Figure S1](#pone.0034393.s001){ref-type="supplementary-material"}). In regression models, the max was linearly associated with resting DcoSB (parameter estimate (PE)±SEM: 0.59±0.22; r = 0.29; p = 0.009) ([Figure 1A](#pone-0034393-g001){ref-type="fig"}) and with FEV~1~ (PE±SEM: 0.54±0.18; r = 0.33; p = 0.003) ([Figure 1B](#pone-0034393-g001){ref-type="fig"}) across all subjects. The Borg score for shortness of breath fatigue, and effort at maximum observed work [@pone.0034393-Borg1], [@pone.0034393-Belman1] was not significantly different between the two groups of FAs. {#pone-0034393-g001} 10.1371/journal.pone.0034393.t002 ###### Exercise Capacity. {#pone-0034393-t002-2} Subject Characteristics All FAs FA with normal Dco FA with abnormal Dco p-value ------------------------------------------- ------------ -------------------- ---------------------- ----------- Maximum work achieved (watts) 128.3±28.3 132.6±24.2 123.9±31.6 0.169 Maximum work achieved (% predicted watts) 128.9±33.2 137.4±30.2 120.3±34.3 **0.020** Maximum (L/min) 1.32±0.29 1.39±0.26 1.24±0.30 **0.019** Maximum (% predicted ) 95.6±20.2 102.5±19.7 88.7±18.6 **0.001** Maximum (L/min) 1.60±0.42 1.67±0.38 1.52±0.45 0.154 R at maximum work 1.23±0.14 1.23±0.12 1.24±0.16 0.706 Shortness of Breath (Borg Scale) 4.2±2.0 4.5±2.1 3.8±1.9 0.158 Effort (Borg Scale) 5.0±1.9 5.2±2.0 4.7±1.8 0.278 Fatigue (Borg Scale) 4.6±2.0 4.9±2.1 4.1±1.8 0.103 Data is shown in mean ± standard deviation. N = 80; abbreviations: : Oxygen uptake; : carbon dioxide output; R: respiratory gas exchange ratio. Cardiac Response to Exercise {#s3c} ---------------------------- The reduction in exercise capacity was associated with a significantly decreased maximum O~2~ pulse (oxygen uptake per heart beat) and decreased anaerobic threshold (AT) in FAs with lower resting DcoSB ([Table S1](#pone.0034393.s003){ref-type="supplementary-material"}). The stroke volume increased as expected with exercise in all FAs from a baseline of 0.067±0.022 (L/beat) measured in the supine posture at rest to 0.078±0.021 (L/beat) at 20% maximum observed exercise (p\<0.0001), and then plateaued with increasing levels of exercise. The stroke volume and its pattern of change with exercise was not significantly different between the two groups of FAs, which suggests that the decreased aerobic capacity observed in pre-ban FAs with lower resting DcoSB was due to a smaller arteriovenous (A-V) oxygen difference in this group [@pone.0034393-Wasserman1]. There was no difference in other cardiovascular measurements ([Table S1](#pone.0034393.s003){ref-type="supplementary-material"}). Respiratory Response to Exercise {#s3d} -------------------------------- The respiratory response to exercise in both groups was normal as reflected by the maximum levels of minute ventilation, respiratory gas exchange ratio (R), respiratory rate, tidal volume, total inspiratory time as a fraction of total respiratory cycle (Ti/Tot), and ventilatory equivalent of CO~2~ (/) ([Table S2](#pone.0034393.s004){ref-type="supplementary-material"}). Diffusing Capacity during Exercise {#s3e} ---------------------------------- The mean absolute difference between DcoWB and DcoSB values measured at rest in all FAs was 1.54±1.96 ml/min/mmHg (p\<0.0001). There was a significant correlation between the DcoWB and DcoSB values measured at rest in all FAs (Pearson\'s correlation coefficient = 0.67, p\<0.0001) ([Figure S2](#pone.0034393.s002){ref-type="supplementary-material"}). The DcoWB during exercise increased linearly with increasing work rate for all FAs. However the DcoWB in FAs with abnormal resting DcoSB increased less rapidly with increased work rate compared to the FAs with normal resting DcoSB (PE±SEM: 1.36±0.16 vs. 1.90±0.16 ml/min/mmHg per 20% increase in predicted watts; p = 0.020) ([Figure 2A](#pone-0034393-g002){ref-type="fig"}). The absolute and percent predicted (based on available reference equation measured at 40% maximum observed work [@pone.0034393-Huang1]) differences between DcoWB values measured at 40% of maximum observed exercise between the two groups of FAs were 4.9±1.0 ml/min/mmHg (p\<0.0001) and 13.8±3.0%predicted (p\<0.0001), respectively ([Table 3](#pone-0034393-t003){ref-type="table"}). Stratification of the FAs into tertiles or quartiles based on increasing resting DcoSB (15% or 10% increase in DcoSB, respectively) showed a similar pattern of lower increase in DcoWB with lower tertile or quartile of DcoSB ([Figures 2B and 2C](#pone-0034393-g002){ref-type="fig"}), suggesting that the effect of lower DcoSB on exercise-induced increase of DcoWB is incremental. {#pone-0034393-g002} 10.1371/journal.pone.0034393.t003 ###### Within-Breath Diffusing Capacity (DcoWB) at rest and with exercise. {#pone-0034393-t003-3} Subject Characteristics All FAs FA with normal Dco FA with abnormal Dco p-value ---------------------------------------------------------------------------------------------- ----------- -------------------- ---------------------- -------------- DcoWB at rest (ml/min/mmHg) 20.1±3.5 21.4±3.6 18.7±2.8 **0.0004** DcoWB at rest (% predicted Huang[\*](#nt104){ref-type="table-fn"}) 93.8±14.9 100.8±14.7 86.5±11.2 **\<0.0001** DcoWB at rest (% predicted Charloux[\*\*](#nt104){ref-type="table-fn"}) 96.3±21.4 103.8±20.3 88.7±20.0 **0.001** DcoWB at 40% maximum observed work (ml/min/mmHg) 24.8±4.9 27.1±4.9 22.3±3.6 **\<0.0001** DcoWB at 40% maximum observed work (% predicted Huang[\*](#nt104){ref-type="table-fn"}) 91.5±14.9 98.4±14.4 84.5±12.0 **\<0.0001** DcoWB at 40% maximum observed work (% predicted Charloux[\*\*](#nt104){ref-type="table-fn"}) 91.7±18.0 98.8±18.2 84.0±14.3 **0.0002** . Data is shown in mean ± standard deviation. Subjects were all female (N = 80). : based on predicted values from Huang and Charloux, respectively [@pone.0034393-Crapo1], [@pone.0034393-Wilson1]. Abbreviation: DcoWB: within-breath diffusing capacity. The pulmonary flow () increased with increasing work rate in both groups of FAs and there were no significant differences in the rate of increase between the FAs with normal and abnormal resting DcoSB (PE±SEM: 0.053±0.007 vs. 0.040±0.004 L/min/%predicted watts; p = 0.097) ([Figure 3A](#pone-0034393-g003){ref-type="fig"}). However, the DcoWB increased less rapidly with increasing in FAs with abnormal resting DcoSB than in those with normal resting DcoSB: (PE±SEM: 0.093±0.06 vs. 1.47±0.09 ml/min/mmHg/L/min; p = 0.0001) ([Figure 3B](#pone-0034393-g003){ref-type="fig"}). {#pone-0034393-g003} Association with Exposure to SHS {#s3f} -------------------------------- The pre-ban length of employment (our proxy for aircraft cabin SHS exposure) was similar between the two groups of FAs with normal and abnormal DcoSB at rest. While neither DcoSB nor DcoWB at rest were associated with years of SHS exposure, DcoWb during exercise showed a trend towards inverse association with years of cabin SHS exposure. Using percent predicted equation by Huang et al [@pone.0034393-Huang1], the pre-ban FAs, DcoWB at 40% maximum observed workload showed a trend inverse association with years of cabin SHS exposure (PE±SEM: −0.64%±0.37%, p = 0.095). In a subgroup of FAs with more than 10 years of cabin SHS exposure, the inverse association was statistically significant (PE±SEM: −0.99%±0.48%, r = 0.32 p = 0.032), suggesting a decrease of 0.99% in DcoWB for every year of SHS exposure beyond the first 10 years ([Figure 4](#pone-0034393-g004){ref-type="fig"}). Using percent prediction equations by Charloux et al [@pone.0034393-Goniewicz2] produced similar results (PE±SEM: −1.25%±0.55%; p = 0.029 at 40% workload and −1.19%±0.59%; p = 0.050 at 60% workload in the subgroup with history of more than 10 years of SHS exposure). {#pone-0034393-g004} Discussion {#s4} ========== In this study of 80 never-smoking pre-ban FAs with significant history of exposure to SHS, we confirmed our previous findings in a smaller cohort that about half of the pre-ban FAs had decreased resting diffusing capacity (DcoSB) below the 95% prediction limit as well as curvilinear flow-volume curves (concave to the volume axis) and decreased air flow at mid and low lung volumes. More importantly, we found that those FAs with reduced resting DcoSB had lower exercise capacity as indicated by their maximum predicted work (watts) and maximum oxygen uptake (). This lower exercise capacity was associated with decreased oxygen uptake per breath (O~2~ pulse) and decreased anaerobic threshold (anaerobic threshold was reached at lower absolute and lower % predicted value of maximum ). In addition, the exercise capacity was directly associated with their FEV~1~ at baseline. Furthermore, we found that while the two groups of FAs had similar pulmonary blood flow (), those with reduced DcoSB at rest increased their diffusing capacity with exercise (DcoWB) less with increasing workload and pulmonary blood flow, which indicates that despite exercise-induced increase in pulmonary blood flow, they had reduced pulmonary capillary recruitment. Finally, although no association between the resting diffusing capacity and SHS exposure was observed in our cohort, the diffusing capacity during exercise showed a trend association with the years of SHS exposure with the association reaching a significant level in those FAs with more than 10 years of exposure. Our finding of decreased exercise capacity in the group of FAs with reduced resting diffusing capacity is interesting as none of FAs had any history of cardiopulmonary diseases and all reported exercising regularly 3 to 5 times weekly. Although the average maximum watts and average peak achieved by all the FAs were within the normal reference range, the FAs with reduced resting diffusing capacity achieved significantly lower levels compared to those with higher resting diffusing capacity, and many within the former group achieved maximum work levels well below the normal reference range (25% only achieved levels below the 77% maximum predicted ). The lower exercise capacity of this group of FAs is likely due to decreased aerobic capacity as reflected by decreased at AT and decreased O~2~ pulse. Since the stroke volume was similar between the two groups of FAs, the decreased aerobic capacity thus appears to be due to decreased arteriovenous (A-V) oxygen difference, which in turn suggests an abnormal distribution of oxygenated blood to non-essential tissues and reduced distribution of oxygenated blood to exercising muscle [@pone.0034393-Wasserman1]. We also demonstrated that FEV~1~ was a predictor of exercise capacity in a linear fashion. The correlation coefficient of 0.33 between exercise capacity and FEV~1~ reflects a medium effect size according to the criteria of Cohen [@pone.0034393-Cohen1], and thus is consistent with a physiologically meaningful association. Given the presence of curvilinearity in flow-volume loops and decreased flow at mid and low lung volumes in the cohort, this association suggests that dynamic airflow obstruction may play a role in decreased exercise capacity of the FAs. The decreased exercise capacity and lower exercise-induced increase in diffusing capacity in the FAs with reduced resting diffusing capacity further substantiate our previous findings of abnormal resting pulmonary function tests (i.e. decreased flow in low- and mid-lung volumes, air trapping, and reduced resting diffusing capacity) in this cohort with significant history of SHS exposure. The lower increase in exercise diffusing capacity, its incremental nature based on resting diffusing capacity, and lower ratio of diffusing capacity to pulmonary blood flow during exercise suggest that these FAs have reduced pulmonary capillary bed recruitment. Further anatomic or physiologic studies such as CT imaging or assessment of the inspiratory flow-volume loop assessment during exercise could help elucidate the underlying cause. The finding that the FAs with reduced resting diffusing capacity had a lesser exercise-induced increase in their diffusing capacity with various stratification confirms our hypothesis that even some of the FAs with resting diffusing capacity within the 95% "normal" prediction limit had reduced pulmonary capillary recruitment. In addition, it shows that the diffusing capacity during exercise is a more sensitive measure of underlying abnormalities in pulmonary capillary bed than resting diffusing capacity. Together, these results show that the observation of a normal diffusing capacity at rest may not indicate that the pulmonary capillary recruitment during exercise will be normal. The association between diffusing capacity during exercise and cabin years of SHS exposure had a two-tailed p-value of 0.095; however, if we assume that the adverse event on lung function would only be seen with increased SHS exposure (a one-tail assumption), the association then becomes significant (one-tail p-value of 0.048). In addition, the LOWESS smoother analysis showed a significant lung function and SHS exposure association (two-tailed p-value of 0.032) in the subgroup of FAs who had a longer SHS exposure history, a finding that is supported by the plausible biologic mechanism that higher exposure could cause higher adverse events. The correlation coefficient (r) of 0.32 between diffusing capacity during exercise and cabin years of SHS exposure represents a medium effect size according to the criteria of Cohen [@pone.0034393-Cohen1], and thus is not only statistically significant, but also is consistent with a meaningful association. Although most FAs did report additional SHS exposure apart from cabin exposure, it is expected that their non-cabin related SHS exposure was relatively insignificant compared to the intensity of their exposure while aboard aircraft [@pone.0034393-Repace1]. In a previous study of this cohort, we investigated potential contribution of non-cabin related SHS exposure to subjects\' lung function via comparison of the subjects with only cabin SHS exposure and those reported additional non-airline SHS exposure (i.e. exposure during childhood, adulthood, and/or non-airline employments), and did not find any significant differences in lung function between the two groups [@pone.0034393-Arjomandi1], which also suggest that non-cabin SHS exposure in FAs was dwarfed by their cabin SHS exposure. Overall, our findings provide strong physiologic evidence consistent with presence of emphysema and/or COPD in this cohort of never-smoking FAs who were exposed to high concentrations of SHS for extended periods of time in the aircraft cabin. It remains unclear whether the pulmonary function abnormalities seen in this cohort are stable or progressive. Our study has several possible limitations. First, and from a technical standpoint, performing the within breath diffusing capacity (DcoWB) measurement during exercise requires excellent self-control of breathing by the subject to maintain a low and constant expiratory flow, even if the subject is able to see and monitor flow on a screen during the maneuver. The measurement becomes increasingly difficult to perform at high workloads. To improve distribution of blood flow and ventilation-perfusion matching, we elected to perform our studies in the supine posture [@pone.0034393-Stokes1]. We aimed to attain maximum recruitment of the pulmonary capillary bed and diffusing capacity at a lower work rate during supine exercise rather than in the erect posture, thus easing the demands on the subject to control breathing during the test maneuver. The FAs who participated in our study were able to perform the DcoWB technique reliably during exercise from rest to 60% maximum observed work rate, but few were able to reach 80% let alone near maximum work rates. However, as shown in the [results](#s3){ref-type="sec"} section, DcoWB increased with workloads and in a linear manner with no evidence of a plateau up to the maximum levels measured in FAs. Reproducibility of the DcoWB measurements was within 5%, which was considered satisfactory, and linear relationships were observed between DcoWB and and were similar to those reported previously by other investigators [@pone.0034393-Huang1], [@pone.0034393-Goniewicz2]. Second, in contrast to previous studies, we felt carbon monoxide (CO) backpressure on DcoWB measurement had to be considered. Stokes [@pone.0034393-Stokes2] and Huang [@pone.0034393-Huang2] concluded the effect was negligible in their studies. Charloux et al found carboxyhemoglobin levels increased from 0.3% to 5.2% after 12 measurements in 4 subjects. In our study, carboxyhemoglobin levels increased from 0.3% to 10% after 12 measurements, and thus, we adjusted DcoWB for changes in hemoglobin and carboxyhemoglobin. Third, our study, which was a cross-sectional study of a group of FAs with a remote and unique high SHS exposure, did not include a similar control population for comparison. However, we used the available established prediction formulas to determine whether the obtained physiologic measurements were within the "normal" range. The appropriate control groups for our pre-ban FAs would be post-smoking ban flight attendants, who have been exposed to aircraft cabin [@pone.0034393-Lindgren1], [@pone.0034393-Rayman1] except for SHS, and "ground-level" controls with no history of significant SHS exposure, both of which are challenging control groups as the post-ban flight attendants are in general younger and the "ground-level" controls do not have the specific exposure to non-SHS cabin factors that may contribute to lung function abnormalities. Of note, the normal predicted values include a wide range, and if one could show a longitudinal decline for an individual with time, even a value within the normal range may be considered to be abnormal. While our study was a cross-sectional one with measurements at a single time point, we were able to show that some FAs with normal resting diffusing capacity had abnormal physiologic responses to exercise, which suggest that despite having resting values within the "normal" range, they had abnormal pulmonary function. Fourth, the pulmonary function tests abnormalities of the FAs in this study do not meet the definition of COPD by GOLD criteria as the FAs had FEV~1~ within the normal predicted range and as they did not have any respiratory complaint at baseline. Traditionally, both COPD diagnosis and severity evaluation have been based on spirometry [@pone.0034393-Rabe1], [@pone.0034393-Celli1], and change in FEV~1~ over time is still the most widely accepted measure of disease progression. However, FEV~1~ has limitations as it measures only one aspect of the disease and is not predictive of disease progression, especially in early disease [@pone.0034393-Nishimura1], [@pone.0034393-Gelb1], [@pone.0034393-Franciosi1]. In addition, COPD patients with similar FEV~1~ may show very different underlying pathologies, for example predominantly airspace disease (i.e. emphysema) or disease of the airways, as manifested by increased airway wall thickness [@pone.0034393-Gelb1]. Thus, we believe that despite the "normal" GOLD classification of the FAs in our cohort, the physiologic abnormalities that we have observed in them are best explained and most consistent with presence of COPD and emphysema [@pone.0034393-Wan1]. Finally, we estimated the SHS exposure through the proxy of years of pre-ban employment using our occupational/employment questionnaire [@pone.0034393-Arjomandi2]. This method of SHS assessment may not be adequately sensitive for true amount of SHS exposure and may also be prone to recall bias. However, the cabin SHS exposure, which is relatively readily quantified using employment history, was much higher than levels experienced outside aircraft cabin in most circumstances [@pone.0034393-Repace1], and dwarfs the non-cabin SHS exposure for these pre-ban FAs. In conclusion, in this cohort of never-smoking pre-ban FAs with remote but significant history of cabin SHS exposure, we found physiologic abnormalities consistent with presence of emphysema and/or COPD. The FAs with lower diffusing capacity had exercise limitation and lower increase in their diffusing capacity with increasing workload and increasing pulmonary blood flow suggesting that those with lower resting diffusing capacity had reduced pulmonary capillary bed recruitment. Exposure to SHS in the aircraft cabin seemed to be a predictor for lower diffusing capacity during exercise in those with higher history of SHS exposure. Supporting Information {#s5} ====================== ###### **Maximum work, maximum oxygen uptake (** **), and maximum ventilatory equivalent of carbon dioxide (** **/** **) for never smoking pre-smoking ban flight attendants.** Black and white bars represent flight attendants with normal and abnormal resting diffusing capacity, respectively. (TIF) ###### Click here for additional data file. ###### **Correlation between single breath carbon monoxide diffusing capacity at rest (sitting position) and within breath diffusing capacity in supine position.** (TIF) ###### Click here for additional data file. ###### **Cardiovascular response to exercise.** Data is shown in mean ± standard deviation. \* N = 80; subjects were all female. Abbreviations: SBP: systolic blood pressure; DBP: diastolic blood pressure; : Oxygen uptake; AT; anaerobic threshold. (DOC) ###### Click here for additional data file. ###### **Respiratory response to exercise.** Data is shown in mean ± standard deviation. \* N = 80; subjects were all female. Abbreviations: SBP: systolic blood pressure; DBP: V~T~: tidal volume; diastolic blood pressure; : minute ventilation; : Oxygen uptake; /: ventilatory equivalent of CO~2~; AT: anaerobic threshold; R: Respiratory gas exchange ratio. (DOC) ###### Click here for additional data file. ###### **Supplemental methods.** (DOC) ###### Click here for additional data file. We would like to thank Kizza Chadiha, Cecilia Yu, and Susan Beaulien for assistance with subject recruitment, Oliver Beech, Patricia Vollmer, and Emily Ghio for their help with performance of cardiopulmonary exercise testing, and Drs. Jay Nadel and Paul Blanc for their consultation. We also would like to appreciate the contribution of the flight attendants who took time out of their busy schedules to participate as research subjects in this study. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**Funding provided by The Flight Attendants Medical Research Institute, The National Institutes of Health (K23 HL083099), Northern California Institute for Research and Education and the University of California San Francisco Cardiovascular Research Institute Faculty Development Funds. 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: WMG MA. Performed the experiments: WMG MA. Analyzed the data: MA TH WMG NS. Contributed reagents/materials/analysis tools: WMG MA TH. Wrote the paper: MA WMG. Obtained funding: WMG MA RR. Recruited Subjects: RR MA WMG.

{#sp1 .204}

The authors wish to add Ka Fai Leung as the third author to the manuscript. The updated byline is: Kanwal Abbasi1, Kelly N. DuBois1, Ka Fai Leung1, Joel B. Dacks2, Mark C. Field1\* The updated citation is: Abbasi K, DuBois KN, Leung KF, Dacks JB, Field MC (2011) A Novel Rho-Like Protein TbRHP Is Involved in Spindle Formation and Mitosis in Trypanosomes. PLoS ONE 6(11): e26890. doi:10.1371/journal.pone.0026890 Ka Fai Leung Performed the experiments, Contributed reagents/materials/analysis tools, and Wrote the Manuscript. **Competing Interests:**No competing interests declared.

Why was the cohort set up? ========================== East London Genes & Health (ELGH) is a community based, long-term study of health and disease in British Bangladeshi and British Pakistani people in east London. ELGH has a population-based design incorporating cutting edge genomics with electronic health record (EHR) data linkage and targeted recall-by-genotype (RbG) studies. ELGH currently has 38 899 volunteers and is actively recruiting with funding to expand to 100 000 volunteers by 2023. ELGH is an open access data resource, and its research will impact upon a population at high need and will redress the poor representation of non-White ethnic groups in existing population genomic cohorts.[@dyz174-B1] Almost a quarter of the world's population, and 5% of the UK population, are of South Asian origin.[@dyz174-B2] The risk of coronary heart disease is three to four times higher, and of type 2 diabetes (T2D) two to four times higher, in British South Asians compared with White British people.[@dyz174-B3]^,^[@dyz174-B4] East London incorporates one of the UK's largest South Asian communities (29% of 1.95 million people), of which 70% are British Bangladeshi and British Pakistani, and its population live in high levels of deprivation (Tower Hamlets, Hackney, Barking and Dagenham are the 9th, 10th and 11th most deprived local authorities in England).[@dyz174-B5] Compared with White British people, British South Asians living in east London have a 2-fold greater risk of developing T2D,[@dyz174-B6] nearly double the risk of non-alcoholic liver disease[@dyz174-B7](many volunteers are practising Muslims and do not drink alcohol) and over double the risk of multimorbidity,[@dyz174-B8] with the onset of cardiovascular disease occurring 8 years earlier in men.[@dyz174-B8] Determinants of poor cardiometabolic health start early in the life course, and east London rates of overweight and obese children are among the highest in the UK. Recent genomic advances offer exciting potential to better understand the genetic causation of disease,[@dyz174-B9] including rare loss-of-function gene variants.[@dyz174-B10] Genetic variation relevant to British Bangladeshi and British Pakistani populations, such as autozygosity arising from parental relatedness, is under-researched with regards to potential effects on complex adult phenotypes at a population level.[@dyz174-B11]^,^[@dyz174-B12] ELGH fosters authentic, inclusive, long-term engagement in its research, to deliver future health benefits to the population it represents. Community involvement in ELGH helps prioritize areas for research, including T2D, cardiovascular disease, dementia and mental health. ELGH undertakes a range of public engagement work, including collaboration with the award-winning Centre of the Cell.[@dyz174-B13] Who is in the cohort? ===================== ELGH (see [Figure 1](#dyz174-F1){ref-type="fig"}) incorporates population-wide recruitment to Stage 1 studies, and targeted recruitment to Stage 2 recall-by-genotype (RbG) studies. Stage 3 and 4 studies are planned. {#dyz174-F1} During Stage 1, ELGH invites voluntary participation of all British Bangladeshi and British Pakistani individuals aged 16 and over, living in, working in or within reach of east London. Recruitment is largely undertaken by bilingual health researchers, and takes place in: (i) community settings, e.g. mosques, markets and libraries, supported by a third-sector partner organization (Social Action for Health); and (ii) health care settings, e.g. GP surgeries, outpatient clinics. Stage 1 volunteers complete a brief questionnaire, give consent to lifelong EHR linkage and donate a saliva sample for DNA extraction and genetic tests. Between April 2015 and mid-June 2019, ELGH recruited 38 899 volunteers to Stage 1. At the most recent data linkage (November 2018), 97% of 31 646 had valid NHS numbers: 61% had linked primary care health record data available; 84% had linked secondary care data. By 2020, near-complete (\>95%) linkage to primary care health records is expected with improved data connectivity, supported by Health Data Research UK. Recruitment into outer London regions, and a new study site in Bradford, are planned for 2019/20, areas with similar ethnic populations and comparable health needs. Summary data from the Stage 1 volunteer questionnaire and EHR data linkage are presented in [Table 1](#dyz174-T1){ref-type="table"}, including both baseline and longitudinal health data. Basic demographics of ELGH volunteers are compared with population-wide data in [Figure 2](#dyz174-F2){ref-type="fig"}, and highlight that the convenience sampling approach in Stage 1 recruitment has achieved a sample broadly representative of the background population with regard to age and sex, but which modestly favours recruitment of women over men in those aged \<45 years. ELGH volunteers live in areas of high deprivation (97% in the most deprived two quintiles of the Index of Multiple Deprivation). Parental relatedness is reported by 19% of ELGH volunteers. {#dyz174-F2} ###### Baseline characteristics of ELGH volunteers from self-reported questionnaire and electronic health record data ---------------------------------------------------------------------------------------------- ------------------------------------------------------- **Self-reported questionnaire data (*n* = 31 646)** Year of birth *n* = 31 634 Median 1977 Interquartile range 1967-85; range 1915-2002 Sex *n* = 31 639 Male *n* = 13 928 (44%) Female *n* = 17 710 (56%) Ethnicity *n* = 31 508 Bangladeshi or British Bangladeshi *n* = 21 083 (67%) Pakistani or British Pakistani *n* = 10 319 (33%) Parental relatedness Yes = 5959 (18.8%) No = 24 601 (77.8%) Don\'t know = 1013 (3.2%) Not documented = 77 (0.2%) **Linked electronic health record data (*n* = 21 514)** Index of Multiple Deprivation (2015) Quintile 1 (most deprived), *n* = 12 551 (58%) Quintile 2, *n* = 8400 (39%) Quintile 3, *n* = 411 (2%) Quintile 4, *n* = 131 (0.6%) Quintile 5 (least deprived), *n* = 11 (0%) Number of ELGH volunteers with common conditions: Type 2 diabetes 4769 (22%) Hypertension 3956 (18%) Ischaemic heart disease 1048 (4.9%) Dementia 47 (0.2%) Asthma 2297 (10.7%) COPD 255 (1.2%) Number of ELGH volunteers with commonly recorded clinical data measured in the past 5 years: Body mass index 18 654 (86.7%) Mean = 27.16kg/m^2^ Standard deviation = 4.87kg/m^2^ Range = 14--69kg/m^2^ Smoking status 18 938 (88%) Never smoked = 14 353 (76%) Ex-smoker = 2146 (11%) Current smoker = 2276 (12%) Uninformative coding = 194 (1%) ---------------------------------------------------------------------------------------------- ------------------------------------------------------- ELGH operates under ethical approval, 14/LO/1240, from London South East NRES Committee of the Health Research Authority, dated 16 September 2014. How often have they been followed up? ===================================== ELGH contains real-world EHR data, its collection triggered by a broad range of clinical encounters including routine and emergency care. East London has an extensive track record of using routine clinical health care data (predominantly from primary care) in research studies.[@dyz174-B6]^,^[@dyz174-B7]^,^[@dyz174-B14] Electronic performance dashboards are embedded in clinical practice, facilitating high quality and equitable disease screening and clinical care.[@dyz174-B15]^,^[@dyz174-B16] Primary care health records were digitized around 2000 and offer a rich source of data on clinical encounters since then, but also include pre-digitization dates of diagnoses and summarized clinical events (e.g. type 2 diabetes, diagnosed in 1992). Health data linkage and extraction takes place 3-monthly and ELGH volunteers have consented for lifelong EHR access, facilitating longitudinal follow-up. ELGH can invite volunteers to Stage 2 studies up to four times per year for more detailed study visits, e.g. recall by genotype (RbG) and/or phenotype, for clinical assessment and collection of biological samples, subject to ethics approval, volunteer acceptability and community advisory group approval. As at August 2019, around 60 ELGH volunteers have participated in Stage 2 RbG studies. What has been measured? ======================= Available data are summarized in [Table 2](#dyz174-T2){ref-type="table"}. ###### Summary of all data types currently available in ELGH for Stage 1 volunteers, and planned for late 2019 onwards Data source Data fields Volunteers Duration of data collection -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------- Volunteer questionnaire Basic details: name, date of birth, ethnicity, address, GP, NHS number All Cross-sectional at study entry Self-reported diabetes status Self-reported parental relatedness Self-assessment of overall health and well-being Electronic health records Primary care (GP) records (currently coded in READ2 or CTV3 clinical terminologies): All volunteers where data linkage is possible (currently 61%, due to increase to \>95% in 2019 with linkage to a wider GP practice network) Real-world data with access to all available historical (retrospective) data and lifelong (prospective) data Sociodemographic data Diagnoses Prescribing data Clinical measurements, e.g. height, weight, body mass index, blood pressure Laboratory tests (e.g. blood tests) Care processes, e.g. referrals Quality and Outcomes Framework indicators: including care processes and outcomes for common diseases (e.g. diabetes, asthma, depression), public health concerns (smoking, obesity) and preventative measures (e.g. blood pressure checks) NHS Health Checks: screening for diabetes, heart disease, kidney disease, stroke and dementia offered to 40--74 year olds Secondary care (hospital) records: All volunteers in contact with secondary (hospital) care Real-world data, includes retrospective data since 2012 and lifelong (prospective) data Diagnoses (ICD10) and procedures (OPCS-4) Chronic problem listing (SNOMED) Laboratory tests (e.g. blood tests, histopathology, microbiology) Maternity records External health record data sets and registries Hospital Episode Statistics All volunteers (planned) Real-world data, includes retrospective data since 2003, and lifelong (prospective) data Office for National Statistics mortality data National Cancer Registration and Analysis Service, NCRAS National Cardiovascular Outcomes Research: national audits Genetic investigations Whole exome sequencing All volunteers reporting parental relatedness (currently 19%) Not applicable Illumina Infinium Global Screening Array v3.0+MD All volunteers up to 50 000 (to be undertaken in early 2019) Not applicable Recall studies Bespoke clinical phenotyping and sample collection according to genotype of interest. Core samples taken on all for methylation assays, transcriptomics, lipidomics and metabolomics 60 volunteers to date, with approval for all volunteers to be approached for recall up to four times per year Dependent on protocol - **Volunteer questionnaire** ([Supplementary File 1](#sup1){ref-type="supplementary-material"}, available as [Supplementary data](#sup1){ref-type="supplementary-material"} at *IJE* online). This self-report questionnaire collects brief data including: name, date of birth, sex, ethnicity, contact details, diabetes status, parental relatedness and overall assessment of general health and well-being. The questionnaire has been designed to facilitate high throughput recruitment and volunteer inclusivity where language and cultural differences may exist, and to be used with or without researcher assistance. The questionnaire does not capture environmental factors (e.g. no self-reported data on smoking, alcohol, diet, physical activity---although smoking and alcohol use are available from other data sources, discussed below). Completion of the volunteer questionnaire triggers health record data linkage via NHS number. - **NHS primary care health record data linkage.** We first design data extraction in human-readable form ([Supplementary File 2](#sup1){ref-type="supplementary-material"}, available as [Supplementary data](#sup1){ref-type="supplementary-material"} at *IJE* online) and then code this in Structured Query Language ([Supplementary File 4](#sup1){ref-type="supplementary-material"}, available as [Supplementary data](#sup1){ref-type="supplementary-material"} at *IJE* online). Coded fields are extracted from EHR systems, curated to research phenotypes of interest and developed both incrementally and on demand. Search terms are used, including READ2 diagnostic codes, prescribing data, laboratory test results and clinical measurements and processes. Data concordance was checked between volunteer questionnaires and their EHR, with \>99% concordance for gender and year of birth. Almost all cases of data discordance were due to technical errors with questionnaire optical character recognition or user data completion, and were resolved with manual checking. Data outside clinically plausible ranges, or with clear data entry errors, are removed. A detailed description of our data processing is in [Supplementary File 3](#sup1){ref-type="supplementary-material"}, available as [Supplementary data](#sup1){ref-type="supplementary-material"} at *IJE* online. Missing data exist, but at relatively low frequency in routinely collected and incentivized clinical measures, e.g. smoking status is recorded in the EHR of 88% of volunteers in the 5 years preceding the most recent data linkage. Repeated measures of routinely collected data and cross-validation across information sources can mitigate the impact of missing data where it exists, as can statistical techniques, such as sensitivity analysis and multiple imputation.[@dyz174-B17] - **NHS local secondary care health record data linkage.** Linkage to Barts Health NHS Trust data provides secondary care data for all ELGH volunteers who have attended this hospital system (24 852 volunteers at the latest linkage). Available data include clinician-coded SNOMED-CT acute and chronic problem lists, laboratory and imaging results. OPCS-4 (Office for Population, Censuses and Surveys) and ICD-10 (World Health Organization International Classification of Diseases and Related Health Problems) codes are available for every finished episode of care. For example, maternity data linkage within Barts Health identified 4172 female ELGH volunteers with maternity records available for one or more pregnancies. Linkage to other local hospitals (including those providing mental and community health care) is planned in 2019/2020. - **Planned linkage to national health record and other datasets.** ELGH will link to further datasets in 2019/2020, including national NHS Hospital Episode Statistics (HES) and NHS Mortality Data,[@dyz174-B18]^,^[@dyz174-B19] to include admissions and discharge, diagnosis and operation codes, maternity, psychiatric and critical care from 1997, and accident and emergency data, ICD-10 and OPCS-4 codes from 2008. NHS mortality data provide data on cause of death. Other planned data linkages to national registries include the National Cancer Registration and Analysis Service and National Cardiovascular Outcomes Research. - **Genomics.** DNA is extracted from the Oragene (DNA Genotek) saliva system and stored from all Stage 1 volunteers. To date, 20-40X depth exome sequencing has been performed (*n* = 3781) or is in progress (*n* = 1492) on volunteers reporting parental relatedness. By late 2019 (funding secured), 50 000 samples from stage 1 volunteers will be genotyped on the Illumina Infinium Global Screening Array v3.0 (with an additional 46 662 Multi-Disease variants).[@dyz174-B20] Array content includes rare disease-associated mutations (e.g. all pathogenic and likely pathogenic variants in ClinVar), pharmacogenetic associations and genome-wide coverage for association studies (based on the 26 populations present in Phase III of 1000 Genomes Project, optimized for imputation accuracy), polygenic risk score and Mendelian randomization studies. In 2019/2020, if support is secured from an evolving Life Sciences Industry Consortium or elsewhere, high-depth exome sequencing will be performed on up to 50 000 volunteer samples. The intention is for genotyping and high-depth exome sequencing to be performed on up to 100 000 volunteer samples by 2023. - **Samples for other --omics.** Core study samples are taken from all volunteers recalled in stage 2 studies, including a blood cell pellet (for repeat DNA analyses), plasma aliquots and blood cell RNA preservation (Paxgene), for studies including methylation assays, transcriptomics, proteomics, lipidomics and metabolomics. What has it found? Key findings and publications ================================================ ELGH is a new resource that continues to grow in size and content and, to date, has been used for three main areas of work, as follows. Characterization of common phenotypes ------------------------------------- Using Type 2 diabetes (T2D) as an exemplar, we show the ability for detailed phenotypic characterization of ELGH volunteers using EHRs ([Table 3](#dyz174-T3){ref-type="table"}). Of 21 514 volunteers in ELGH with available linked EHR data, 4769 (22%) have a diagnosis of T2D in their primary care record. Basic sociodemographic data (age, gender, ethnicity) of volunteers were recorded in 100%, and smoking status had been obtained within 2 years of the most recent data linkage in 94%. In over 97% of volunteers with T2D, body mass index, markers of glucose control (HbA1c) and serum cholesterol were measured and available in the 2 years preceding ELGH participation. Hypertension, ischaemic heart disease and chronic kidney disease were observed in 47%, 15% and 11%, respectively, of the 4769, and erectile dysfunction was present in 26% of men. Retinal complications of T2D are recorded and graded, with 82% of volunteers having undergone screening within the past 2 years. Prescribing data show recent insulin prescriptions in 16%, and the use of single or multiple non-insulin agents, as well as use of cardiovascular drugs (e.g. lipid-lowering therapy). These data show the potential to perform cross-sectional analyses in ELGH from EHR data. ###### Example of a specific disease phenotype: characteristics of ELGH volunteers with type 2 diabetes. Data are presented in summary and descriptive formats as indicated. Missing data are estimated where available, e.g. for clinical care processes and measurements, but not diagnostic coding where the absence of a code is taken to indicate the absence of a diagnosis Volunteers with type 2 diabetes = 4769 (22%) Missing data ------------------------------------------------------------------------------ ------------------------------------------------------- ------------------------------------------- ----------- ---- Sociodemographic data Age Mean years (sd) 46 (11) 0% Sex Male, *n* (%) 2445 (51) Female, *n* (%) 2324 (49) Ethnicity British Bangladeshi and Bangladeshi, *n* (%) 3860 (81) British Pakistani and Pakistani, *n* (%) 822 (17) Other, *n* (%) 87 (2) Index of Multiple Deprivation (2015) Quintile 1 (most deprived), *n* (%) 12551 (58) 0% Quintile 2, *n* (%) 8400 (39) Quintile 3, *n* (%) 411 (2) Quintile 4, *n* (%) 131 (1) Quintile 5 (least deprived), *n* (%) 11 (0) Smoking status (recorded in the past 2 years) Data available, *n* 4712 6% Never smoked, *n* (%) 3348 (71) Ex-smoker, *n* (%) 810 (17) Current smoker, *n* (%) 554 (12) Coding uninformative, *n* (%) 57 (1) Country of birth Data available, *n* 2585 46% Born in Bangladesh, *n* (%) 2130 (82) Born in Pakistan, *n* (%) 347 (13) Born in England, *n* (%) 54 (2) Born elsewhere, *n* (%) 54 (2) Historic T2D data Age at T2D onset Data available, *n* 4769 0% Mean years (sd) 46 (11) Duration of T2D Data available, *n* 4769 Years (range) 7 (0--51) Diabetes risk state before T2D Pre-diabetes, *n* (%) 1241 (26) NA Gestational diabetes (females), *n* (%) 370 (16) Body mass index (BMI) at T2D diagnosis Data available, *n* 3507 5% Mean kg/m^2^ (sd) 28.8 (4.9) HbA1c at T2D diagnosis Data available, *n* 3176 33% Mean HbA1c mmol/mol (sd) 61.9 (18.7) Total cholesterol at T2D diagnosis Data available, *n* 3427 28% Mean total cholesterol, mmol/l 5.0 (1.2) Current T2D data (recorded within the past 2 years) Body mass index Data available, *n* 4630 3% Mean kg/m^2^ (sd) 27.9 (4.8) HbA1c Data available, n 4656 2% Mean mmol/mol (sd) 59.2 (15.6) Total cholesterol Data available, *n* 4640 3% Mean total cholesterol, mmol/l 3.8 (1.0) Retinal screening Data available 4769 NA Not in the screening programme, *n* (%) 241 (5) In the screening programme, *n* (%) 4528 (95) Screened in past 2 years, *n* (%) 3926 (82) Diabetes complications and multimorbidity Other diagnoses Hypertension, *n* (%) 2245 (47) NA Chronic kidney disease, *n* (%) 516 (11) Neuropathy, *n* (%) 164 (3) Ischaemic heart disease, *n* (%) 719 (15) Peripheral vascular disease, *n* (%) 174 (4) Erectile dysfunction (males), *n* (%) 1310 (27) Stroke, *n* (%) 180 (4) Atrial fibrillation, *n* (%) 54 (1) Heart failure, *n* (%) 126 (3) Number of cardiovascular multimorbidities in the presence of type 2 diabetes One or more conditions, *n* (%) 3804 (80) Two or more conditions, *n* (%) 1291 (27) Three or more conditions, *n* (%) 468 (10) Four or moreconditions, *n* (%) 198 (4) Five or more conditions, *n* (%) 85 (2) Drug prescribing Insulin Prescribed in the past 12 months, *n* (%) 756 (16) NA Mean years on insulin, (sd) 8.8 (5) Non-insulin diabetes therapies Metformin prescribed in the past 12 months, *n* (%) 3695 (77) Sulphonylurea prescribed in past 12 months, *n* (%) 1321 (28) Prescribing regimens Prescribed no non-insulin diabetes therapies, *n* (%) 845 (18) Prescribed one non-insulin diabetes therapy, *n* (%) 2057 (43) Prescribed two non-insulin diabetes therapies, *n* (%) 1127 (24) Prescribed three or more non-insulin diabetes therapies, *n* (%) 740 (16) Lipid-lowering treatment Prescribed in the past 12 months, *n* (%) 4256 (89) NA, not available; sd, standard deviation. EHR data also give the potential to study longitudinal phenotypic traits, retrospectively and prospectively. Median duration of T2D in ELGH volunteers was 7 years (range 0--51 years). For all volunteers with T2D, year of onset was recorded, and prescribing data and clinical measurements (including body mass index, HbA1c and cholesterol) at the time of diagnosis (+/- 6 months) were available for nearly two-thirds of volunteers. Before T2D onset, 26% (993) had had a diagnosis of pre-diabetes and 16% (370) of women had had a diagnosis of gestational diabetes, allowing study of progression from at-risk to disease states. Multimorbidity is an increasing problem in ageing populations with high rates of chronic long-term disease; in the ELGH population we identified that of the 4769 ELGH volunteers with T2D, 80% had at least one, and 27% had two or more cardiovascular multimorbidities ([Table 3](#dyz174-T3){ref-type="table"}). Rare allele frequency gene variants occurring as homozygotes, including predicted loss-of-function knockouts ------------------------------------------------------------------------------------------------------------ All ELGH volunteers self-reporting parental relatedness (19%) have been selected for exome sequencing. Genomic autozygosity (homozygous regions of the genome identical by descent from a recent common ancestor) means that rare allele frequency (minor allele frequency \<0.5%) variants normally only seen as heterozygotes are enriched for homozygote genotypes. ELGH expands existing, smaller studies of autozygosity to investigate the health and population effects of such variants, with a focus on loss of function variants.[@dyz174-B11]^,^[@dyz174-B21]^,^[@dyz174-B22] The accuracy of self-reported parental relatedness to actual autozygosity measured at the DNA level by exome sequencing ([Figure 3](#dyz174-F3){ref-type="fig"}) is a modest predictor of actual autozygosity, e.g. we find 8.2% of individuals who declare that their parents are not related in fact have \>2.5% genomic autozygosity. For British Bangladeshi volunteers, mean autozygosity is slightly lower than expected given the reported parental relationship (possibly due to confusion over the meaning of e.g. 'first cousin' versus 'second cousin'), whereas for British Pakistani volunteers, mean autozygosity is slightly higher than expected (possibly due to historical parental relatedness). With an ELGH sample size of 100 000 we estimate we will identify rare variant-predicted loss-of-function homozygotes in \>5000 human genes. ELGH plans to work with other studies on an international human knockout variant browser. {#dyz174-F3} Recall by genotype (and/or phenotype) studies --------------------------------------------- RbG studies, applied to population cohorts with genomic data, are of increasing research interest[@dyz174-B23] and use the random allocation of alleles at conception (Mendelian randomization) to aid causal inference in population studies, reduce biases seen with observational studies and develop functional studies. RbG studies can target specific single variants (or an allelic series for a gene) and polygenic variants (e.g. extremes of polygenic risk scores). ELGH is undertaking RbG studies in Stage 2 using bespoke clinical phenotyping tailored to the genotype or phenotype of interest. To date, three research consortia have undertaken ELGH RbG studies, one recalling volunteers with loss-of-function gene variants relevant to immune phenotypes, another phenotyping individuals with rare variants in genes implicated in T2D and obesity and a third involving an industrial partnership to aid therapeutic development for a rare autosomal recessive metabolic disorder.[@dyz174-B24] Successful recall completion rates to these RbG studies are between 30% and 40%. What are the main strengths and weaknesses? =========================================== ELGH has multiple strengths as a large, population-based study, and its novel, pragmatic design offers opportunities to combine genomic investigation with longitudinal and cross-sectional description of health and disease as determined from EHR data.[@dyz174-B25] ELGH reaches a British Bangladeshi and British Pakistani population with a high burden of disease, generalizable to a wider global population and building on existing genetic studies that have been criticized for focusing on White populations and substantially under-recruiting from minority ethnic groups.[@dyz174-B26] High rates of autozygosity in ELGH volunteers lead to homozygous genotypes at variants with rare allele frequencies that will aid gene discovery, and RbG studies will generate novel translational impact.[@dyz174-B11]^,^[@dyz174-B24] Future studies on autozygosity will inform novel population level insights into the impact of genetic variation on health. The ability to invite all volunteers to Stage 2 studies offers the possibility to develop subcohorts and trials within cohorts in the future. Our community-based recruitment approach offers broad reach into the target population. However, to date, ELGH has modestly over-recruited British Bangladeshi versus British Pakistani volunteers. To support increased recruitment of British Pakistani volunteers, recruitment is expanding into outer London boroughs and a new Bradford Genes & Health. The use of real-world EHR data is both a strength and weakness of ELGH. Strengths include the ability to obtain longitudinal data available on multiple diseases and disease risks via primary care, in large numbers of volunteers in a feasible and cost-effective manner. Data linkage is not yet complete, but will improve in 2019 with improved infrastructure and linkage to national registries and databases. Weaknesses are that EHR data may be inferior to observational epidemiological studies in ascertaining some phenotypes, e.g. recent diseases of minor severity (which do not necessarily require health care access) or subclinical disease. Additionally, although outcomes can be studied relatively well, EHR data have limited opportunity to study certain exposures, e.g. health behaviours, physical activity, diet and some other environmental influences. Can I get hold of the data? Where can I find out more? ====================================================== ELGH offers an open access resource to international, academic and industrial researchers to drive high-impact, world-class science. Data access is managed at several levels, as follows. - Level 1. Fully open data: summary data are distributed via our [website]{.ul}, e.g. genotype counts and annotation of knockout variants from exome sequencing, and prevalence of phenotype and traits data. - Level 2. Genotype data (SNP chip genotyping, or high-throughput sequencing) are (or will be) available under data access agreements granted by the independent Wellcome Sanger Institute Data Access Committee. Individual sequencing (e.g. cram) and genotype files (e.g. vcf) are available within 6 months on the European Genome-phenome Archive[@dyz174-B27] (EGA). - Level 3. Individual-level phenotype data are held in an ISO27001 compliant data safe haven environment under data access agreement, currently hosted by the UK Secure e-Research Platform.[@dyz174-B28]^,^[@dyz174-B29] The data safe haven contains the latest genetic data linked to the questionnaire and health record phenotypes, and data export is tightly controlled. This 'bring researchers to the data' model allows us to share regular data updates, maintain complex data linkages and avoid large file data transfers. This model provides robust reassurance to volunteers that their health data will be carefully looked after, with maximum security against data breaches. External researchers can to apply to undertake research with ELGH via a formal application process(details are available on the [website]{.ul}), and most will be required to have their own research ethics approval to work with ELGH. Applications are assessed by both the executive board and community advisory group, according to community prioritization, acceptability and scientific merit. Profile in a nutshellEast London Genes & Health (ELGH) is a large-scale, community genomics and health study (to date 38 899 volunteers; target 100 000 volunteers).ELGH was set up in 2015 to gain deeper understanding of health and disease, and underlying genetic influences, in British Bangladeshi and British Pakistani people living in east London.ELGH prioritizes studies in areas important to, and identified by, the community it represents. Current priorities include cardiometabolic diseases (high prevalence of early onset) and mental illness. However, studies in any scientific area are possible, subject to community advisory group and ethical approval.ELGH combines health data science \[using linked UK National Health Service (NHS) electronic health record data\] with exome sequencing and SNP array genotyping to elucidate the genetic influence on health and disease, including the contribution from high rates of parental relatedness to rare genetic variation and homozygosity (autozygosity), in two understudied ethnic groups. Linkage to longitudinal health record data enables both retrospective and prospective analyses.Through Stage 2 studies, ELGH offers researchers the opportunity to undertake recall-by-genotype and/or recall-by-phenotype studies on volunteers. Subcohort studies, trials within cohort and other study designs are possible.ELGH is a fully collaborative, open access resource, open to academic and life sciences industry scientific research partners. Funding ======= We acknowledge funding from the Wellcome Trust (102627, 210561), the Medical Research Council (M009017), Higher Education Funding Council for England Catalyst, Barts Charity (845/1796), Health Data Research UK and the NHS National Institute for Health Research Clinical Research Network. Supplementary Material ====================== ###### Click here for additional data file. We acknowledge, with gratitude, the substantial contribution to ELGH from all of its volunteers who have provided samples and health record access, and have consented for recall. We thank the supporting recruitment teams and community organizations, including Social Action for Health. Invaluable support has also been received from the general practitioners in the local health system, who have made routinely collected health data available for research with support of the Queen Mary University of London Clinical Effectiveness Group. We are also grateful to the NHS National Institute for Health Research (NIHR) North-Thames Clinical Research Network (CRN), NHS Clinical Commissioning Groups (CCGs) (Tower Hamlets, City and Hackney, Newham, Waltham Forest, Barking and Dagenham), and Barts Health NHS Trust. ELGH looks forward to its expansion in 2019 in Bradford, Luton and Watford. **Conflict of interest:** None declared.

{#sp1 .441} {#sp2 .442} {#sp3 .443}

Introduction {#sec1} ============ Alzheimer's disease (AD) is a neurodegenerative disorder characterized by the presence of Aβ~1--42~ amyloid plaques in the brain and tangles of tau protein inside the neuron cells, which lead to progressive dementia and finally death. Among various environmental risk factors, persistent infection of brain cells by bacteria and virus, especially herpes simplex virus type-1 (HSV-1), has been observed to play a role in AD.^[@ref1]−[@ref5]^ The presence of HSV-1 in association with amyloid deposition in the cerebral cortex region of the brain was established more than a decade ago in patients with familial AD.^[@ref6]^ HSV-1 was also shown to upregulate Aβ generation in cultured neuronal and glial cells,^[@ref7]^ and later its genomic DNA was shown to colocalize with amyloid plaques found in the brain of patients with AD.^[@ref8]^ A few other studies have implicated this virus for increased risk of dementia, disruption of genetic and molecular networks,^[@ref9]^ seeding β-amyloidosis in brain cells,^[@ref10]^ and AD-specific phosphorylation of tau protein.^[@ref11]^ The data discussed so far have correlated HSV-1 infection and AD comprehensively; however, mechanistic events still remain poorly understood. Even though a peptide derived from glycoprotein B of HSV-1 has been reported earlier to form β amyloid-like aggregates,^[@ref12]^ the aggregation potential of HSV-1 proteome has not been analyzed to date. The current study adopts a hypothetical cellular proteasomal activity as the basis for the generation of HSV-1 peptides, as the proteasomes, in addition to maintaining cellular protein turnover, are also involved in generating peptides from viral proteins, which are then presented on the cell surface of T-cells^[@ref13]^ and demonstrate the aggregation properties of one such peptide *in silico* and *in vitro*. The sequences of HSV-1 proteins were retrieved from online databanks and screened for identifying aggregation-prone proteins using software TANGO and AGGRESCAN. Further, the 20S proteasome cleavage sites within the selected proteins were predicted using an *in silico* support vector machine (SVM) method using an online server. The aggregation-prone peptides were identified using the data of 20S proteasome cleavage sites and analysis of the hydrophobic regions present along the entire length of the selected proteins (gM and gK). These studies led to the identification of a peptide (~208~LYHRPAIGVIVGCELMLRFVAVGLIVGT~235~) derived from HSV-1 glycoprotein K (HSV-1 gK~208--235~), which showed the aggregation score and hydrophobicity value equivalent to those of the Aβ~1--42~ peptide. Additionally, the sequence comparison of HSV-1 gK~208--235~ and Aβ~1--42~ peptide revealed homology at their C-termini. Thereafter, the synthetic peptide was subjected to *in vitro* solubilization and the resultant aggregates were characterized using Congo red and Thioflavin T (ThT) fluorescence assays and Fourier transform infrared (FTIR) spectroscopy. All of these studies suggested the formation of amyloid-like aggregates. The overall shape of the peptide aggregates was found to be spheroid instead of classical fibrillar structures, as revealed by atomic force microscopy (AFM). The cytotoxicity of the spheroid aggregates was assessed via cell viability assay performed using primary cells---mouse splenocytes, wherein the dose-dependent toxicity was observed. Results {#sec2} ======= *In Silico* Screening for Selection of Aggregation-Prone HSV Proteins {#sec2.1} --------------------------------------------------------------------- The average aggregation scores of 70 proteins of HSV-1 \[Supporting Information (SI), [Table S1](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf)\], FMRP-1 (Fragile-X-Mental Retardation-1 Protein), low aggregation-prone protein (negative control), and a well-established amyloidogenic peptide, Aβ~1--42~ (positive control) were calculated using software TANGO and AGGRESCAN. FMRP-1 was chosen as a negative control protein as it is highly abundant in neuron cells and plays a role in synapse, cell-to-cell communication,^[@ref14]^ and is not known to have amyloidogenic properties. On the other hand, the Aβ~1--42~ peptide was chosen as the positive control peptide as its amyloidogenic aggregation properties are well characterized in the literature and is known to form amyloid fibrils.^[@ref15],[@ref16]^ Therefore, the protein/peptide, which showed an aggregation score close to that of the Aβ~1--42~ peptide, could be considered as aggregation-prone candidates. The average aggregation score of each protein is calculated by summing up the aggregation scores of each residue (total aggregation score) of the protein and dividing it by the total number of residues in the protein or peptide. The TANGO/AGGRESCAN average aggregation scores of all proteins were compared to the score of negative control protein FMRP-1 (score of 2.59/0.6) and the positive control peptide Aβ~1--42~ (score of 36.48/3.6). Software TANGO and AGGRESCAN both identified glycoprotein M and K as the most aggregation-prone proteins, as suggested by their highest average scores (SI, [Table S1](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf)). As shown in [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}A, the gM and gK showed TANGO average aggregation scores comparable to that of the peptide Aβ~1--42~, whereas in the case of AGGRESCAN scores, the values of gM and gK were found to be more than 3 times higher than those of the Aβ~1--42~ peptide. Hence, these two proteins were chosen for further analysis. It is important to note here that the shorter length of the Aβ~1--42~ peptide may influence its score; therefore, to neutralize this bias and compare the aggregation properties of control and test subjects more explicitly, the aggregation scores, which are calculated per residue across the entire length of the proteins, were determined using TANGO and AGGRESCAN. As shown in [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}B, both gM and gK demonstrated several intermittent segments displaying aggregation scores equivalent to (TANGO) or more (AGGRESCAN) than Aβ~1--42~ peptide, whereas the negative control, FMRP-1, exhibited a minimum number of such regions, which were of smaller lengths and aggregation scores. This analysis further reaffirms the high inherent aggregation propensity of the selected proteins. {#fig1} Prediction of 20S Proteasome Cleavage Sites and Identification of Aggregation-Prone Peptide(s) {#sec2.2} ---------------------------------------------------------------------------------------------- The 20S proteasome cleavage sites present in the glycoprotein M and K were predicted *in silico* by the Pcleavage, an SVM-based method, and NetChop algorithms. According to the obtained data, the majority of the predicted cleavage sites in the proteins were positioned within the regions identified as aggregation-prone (SI, [Table S2](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf)), leaving a few such regions intact. The aggregation-prone regions in gM and gK (identified using TANGO) that were located between two consecutive 20S cleavage sites were considered for peptide selection. Across the entire length of gK and gM, seven and eight such regions were observed, respectively, as shown in SI [Table S2](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf). Analysis of sequences of all of the 15 aggregation-prone regions of gM and gK combined and the positions of the cleavage sites (predicted by Pcleavage) revealed only one continuous peptide, a 28-residue-long stretch of gK (HSV-1 gK~208--235~). The 20S cleavage sites, which were predicted using the NetChop algorithm, had placed two more sites in the middle of the HSV-1 gK~208--235~ peptide region, producing two peptides out of the same regions, but of much shorter length. Between the two prediction tools, Pcleavage and NetChop, the former has been reported to predict cleavage sites with a higher accuracy;^[@ref17]−[@ref19]^ therefore, peptide HSV-1 gK~208--235~, generated using the Pcleavage algorithm, was chosen for further analysis. The average aggregation score and hydrophobicity values of the Aβ~1--42~ and HSV-1 gK~208--235~ peptides were recalculated using the TANGO/AGGRESCAN algorithms and peptide analyzing tool (Thermo Fisher), respectively, and compared. The hydrophobicity value of the HSV-1 gK~208--235~ peptide (58.68) was found to be slightly higher than that of Aβ~1--42~ peptide (54.77), as shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}A. The TANGO average aggregation score of HSV-1 gK~208--235~ peptide (32.58) was found to be equivalent to that of the Aβ~1--42~ peptide (36.48), with the latter having a slightly higher value ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}B). In contrast to this, the AGGRESCAN average aggregation score of HSV-1 gK~208--235~ peptide (12.33) was found to be slightly higher than 3 times the value of the Aβ~1--42~ peptide (3.61), as shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}B, indicating a high propensity of the test peptide for aggregation. A sequence analysis of these peptides revealed that the C-terminal sequences of both HSV-1 gK~208--235~ (~22~VGLIVG~27~) and Aβ~1--42~ (~32~IGLMVG~37~) peptides were homologous to each other ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}C). Additionally, the presence of a cysteine residue in the middle of the HSV-1 gK~208--235~ peptide may form disulfide bonds among aggregated species, further stabilizing the aggregates. The Aβ~1--42~ and HSV-1 gK~208--235~ peptides were also analyzed using AMYLPRED2, a web tool that uses a consensus of different methods, which are developed to predict the amyloidogenic features of protein/peptide. The output of the program revealed that, in the case of the HSV-1 gK~208--235~ peptide, 23 out of 28 residues showed amyloidogenic propensity, a tally higher than the positive control Aβ~1--42~ peptide (SI, [Figure S1](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf)), suggesting the potential of the test peptide to form amyloidogenic aggregates. The glycoprotein K was also examined for the presence of glycosylation sites across its primary sequence using UniProt database (P68333). The only two residues at positions 48 (Asparagine) and 58 (Asparagine) were reported to be glycosylated, which did not fall within the sequence of the HSV-1 gK~208--235~ peptide, further suggesting that the behavior of its aggregation will not be affected by glycosylation. These *in silico* analyses of the HSV-1 gK~208--235~ peptide suggested that it has amyloidogenic properties and therefore it was selected for further *in vitro* studies. ![(A) Graphical comparison of hydrophobicity of the Aβ~1--42~ and HSV-1 gK~208--235~ peptides. (B) The predicted TANGO and AGGRESCAN average aggregation scores of Aβ~1--42~ and HSV-1 gK~208--235~ peptides. The HSV-1 gK~208--235~ peptide demonstrates comparable aggregation (TANGO and AGGRESCAN) values to that of the Aβ~1--42~ peptide. (C) Graphical representation of TANGO and AGGRESCAN aggregation scores per residue of Aβ~1--42~ and HSV-1 gK~208--235~ peptides. The C-termini of the two peptides showed equivalent aggregation scores, and the homologous sequences toward the C-terminal of Aβ~1--42~ (~32~[IGLMVG]{.ul}~37~) and HSV-1 gK~208--235~ (~22~[VGLIVG]{.ul}~27~) peptides have been enlarged and underlined.](ao0c00730_0002){#fig2} *In Vitro* Aggregation Analysis {#sec2.3} ------------------------------- The peptide HSV-1 gK~208--235~ was solubilized in the buffer and allowed to aggregate during *in vitro* experiments. Samples of the peptide were prepared at 55, 100, 200, 300, and 600 μM concentrations in phosphate-buffered saline (PBS) for the aggregation analysis, as described in the [Materials and Methods](#sec4){ref-type="other"} section. The lowest concentration of 55 μM was used in the assay because lower than 50 μM peptide concentrations did not produce enough detectable signals.^[@ref20]^ Second, since the aggregation samples were required to be diluted to a final concentration of 50 μM for fluorescence emission recording, a slightly higher concentration of 55 μM was used as the minimum peptide concentration to maintain the uniformity among the tested samples. The solutions of the peptide were observed to turn turbid instantaneously upon dilution with PBS at all of the five tested concentrations, suggesting thereby a strong propensity of the HSV-1 gK~208--235~ peptide to aggregate. The sample prepared at a 100 μM peptide concentration was analyzed for the presence of amyloid aggregates by Congo red assay. As shown in [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}, the sample produced a characteristic red bathochromic shift in the absorption maxima of Congo red dye, from 483 to 507 nm. This 24 nm shift in Congo red absorption maxima indicated the possibility of the presence of amyloid-like aggregates. To confirm this nature of the aggregates, the samples were further subjected to amyloid-specific ThT fluorescence assay. {#fig3} The peptide samples prepared at 55, 100, 200, 300, and 600 μM concentrations were mixed with the ThT dye, as explained in the [Materials and Methods](#sec4){ref-type="other"} section. The resultant samples were excited at 450 nm, and the emission scan was recorded at 470--700 nm with the peak intensity at 485 nm. As shown in [Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}A, a concentration-dependent increase in the fluorescence signal was recorded with peptide samples prepared at 55, 100, and 200 μM; however, the increment in the fluorescence signal, obtained with the sample prepared at 200 μM concentration, was not similar to what was observed with the sample prepared at the 100 μM peptide concentration. This suggested the occurrence of a phenomenon of self-association among the initially formed aggregates, wherein self-association reduces the number of sites available for ThT dye binding and hence results in decreased fluorescence emission. As shown in [Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}A, the fluorescence signals remained almost similar for samples prepared at 100, 200, and 300 μM, suggesting the self-association events to be the reason behind the observation. Likewise, the fluorescence emission of the sample prepared at 600 μM concentration reduced further, demonstrating the increased association of aggregates at such a high concentration ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}A). Such behavior of self-association has been reported earlier with the amyloid fibrils at high peptide concentrations.^[@ref21]^ These observations indicate the possible presence of amyloid-like aggregates in the samples. Further, the kinetics of the peptide aggregation was studied over a period of 48 h using samples prepared at 100 μM peptide concentration. After setting up the aggregation reaction, the samples were withdrawn at 0 h and thereafter after every 4 h, up to 16 h, and the latter samples were withdrawn at 24 and 48 h. The diluted samples were mixed with the ThT dye, and an emission scan was recorded as described earlier. Since the peptide showed instant aggregation, the maximum fluorescence intensity was recorded at 0 h. Thereafter a decrease in fluorescence intensity was recorded up to 24 h. The fluorescence intensity was observed to have stabilized thereafter as the emission graph of the 48 h sample almost overlapped with the 24 h sample ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}B). A similar behavior of instant aggregation and a subsequent decrease in fluorescence intensities have been reported earlier. Cloe et al. (2011) showed that a shorter mutant peptide ΔE22-Aβ~1--39~, generated from a mutant ΔE693 (Japanese) β-amyloid precursor protein, aggregates instantly and also forms amyloid fibrils.^[@ref21]^ In another recently reported study, Adler et al. (2017) enhanced the hydrophobicity of the Aβ~1-40~ peptides by chemically modifying their N-terminus with saturated octanoyl or palmitoyl lipid chains. The lipid modification increased the local hydrophobicity of the peptide, which led to the acceleration in fibrillation kinetics.^[@ref22]^ {#fig4} The aggregation sample prepared at a 100 μM peptide concentration was further analyzed to determine the structural features of the peptide aggregates. The sample was prepared as described in the [Materials and Methods](#sec4){ref-type="other"} section and subjected to AFM. The peptide aggregates were observed to form spheroid oligomeric species of diameter ∼15 nm ([Figure [5](#fig5){ref-type="fig"}](#fig5){ref-type="fig"}A). Further, the *z* (axis) profile analysis revealed the height of the oligomer as ∼12 nm, as depicted in [Figure [5](#fig5){ref-type="fig"}](#fig5){ref-type="fig"}B. No classical amyloid fibrils were observed in the AFM analysis. {#fig5} Secondary Structure of Peptide Aggregate {#sec2.4} ---------------------------------------- To further affirm if the spheroid oligomeric aggregates contain amyloid-like structures, the aggregation sample was subjected to attenuated total reflection (ATR)-FTIR spectroscopy. The analysis of FTIR spectra allows the prediction of the protein secondary structure content. The infrared spectrum of a protein often contains peaks of multiple amide bonds due to the vibrational contributions from the amino acid side chains and protein backbone. The absorption band generated by C=O stretching in the peptide is designated as amide I, and the N--H banding pattern is denoted as amide II. The amide I band in the spectra is useful in predicting the secondary structure of protein/peptide.^[@ref23]^ The amide bands at 1634/1635 cm^--1^, 1645 cm^--1^, and 1651/1653/1655 cm^--1^ correspond to the β-sheet, random coil, and α-helix secondary structures, respectively.^[@ref24],[@ref25]^ The FTIR spectra of spheroid aggregate showed a sharp band at 1634.84 cm^--1^ ([Figure [6](#fig6){ref-type="fig"}](#fig6){ref-type="fig"}), indicating the presence of a high β-sheet content. The wide band between 3100 and 3500 cm^--1^ shows the presence of residual moisture (H~2~O) in the peptide aggregate. A highly symmetrical amide I band made it difficult to process spectra for deconvolution to estimate the random coil/α-helix in the spheroid aggregate. This symmetrical and prominent amide I band at ∼1634 cm^--1^ suggests the presence of a high β-sheet content in the spheroid oligomer, further indicating the formation of amyloid-like aggregates. {#fig6} Cytotoxic Properties of HSV-1 gK~208--235~ Peptide {#sec2.5} -------------------------------------------------- Accumulation of the neurodegenerative plaques, composed of Aβ~1--42~ peptide, is the most prominent of all pathological hallmarks observed in the brain of patients with AD. The toxic effects of Aβ~1--42~ amyloid aggregates have been established in neurons and many other cells when applied extracellularly.^[@ref26],[@ref27]^ To determine if HSV-1 gK~208--235~ amyloid-like spheroid aggregates were cytotoxic, a preliminary cell viability assay was performed using mouse primary cells. The splenocytes were isolated from mouse spleen and were grown in culture in the presence of viral peptide aggregates at concentrations 2.5, 5, 10, and 20 μM for 48 h. The estimation of viable cells was performed at 24 and 48 h after culture, as described in the [Materials and Methods](#sec4){ref-type="other"} section. Although untreated cells appeared healthy up to 48 h with a negligible reduction in viable cell percentage, dose-dependent cell death was observed in treated cells, wherein extensive cell death was observed at 10 μM and higher concentrations of aggregates after 48 h of culturing ([Figure [7](#fig7){ref-type="fig"}](#fig7){ref-type="fig"}A,B). The median toxic dose (TD~50~) of the HSV-1 gK~208--235~ peptide aggregates against the mouse primary splenocyte cells was observed at 7.11 and 4.35 μM for 24 and 48 h, respectively. Collectively, the *in silico* and *in vitro* results suggest that the HSV-1 gK~208--235~ peptide is capable of self-assembly into spheroid oligomeric amyloid-like aggregates that are toxic to primary cells. Hence, we hypothesize that primary or recurrent infections of HSV-1 may lead to the release of HSV-1 gK~208--235~ and/or other such peptides inside host cells, leading to the formation of amyloid-like aggregates, and prove toxic to neurons or other host cells. {#fig7} Discussion {#sec3} ========== Herpes simplex virus-1, a neurotropic virus, following its primary infection establishes a latent infection in sensory ganglia, especially the trigeminal, through retrograde axonal transport.^[@ref28]^ The recurrent activation of HSV-1 may allow it to enter the central nervous system (CNS) in either manifesting a disease condition, for example, herpetic encephalitis, or entering latency.^[@ref29],[@ref30]^ By employing CD8^+^ T cells, which recognize viral antigen peptides presented on the infected cell surface in complex with major histocompatibility complex-1 (MHC-1) molecules, the immune system effectively eliminates virus-infected cells and keeps viral spread under check. In relation to antigen presentation in the CNS, the expression of MHC-I molecules in the hippocampus area of the human brain, endothelium and microglia, was reported long before it was observed in patients with AD.^[@ref31]^ According to the earlier belief, neurons were devoid of major histocompatibility complex-1 (MHC-1) expression and were considered immune-privileged.^[@ref32]^ However, many studies have shown the expression of MHC-I molecules in neurons of the human brain.^[@ref33]−[@ref35]^ Additionally, the potential of the neuroinflammatory machinery is well studied in a neurodegenerative disease like Parkinson's disease (PD).^[@ref36]^ According to a recent study, the recurrent infection of HSV-1 in mice was shown to induce hallmarks of neurodegeneration as observed in AD, e.g., accumulation of Aβ protein, hyperphosphorylation of tau protein, and induction of neuroinflammation.^[@ref37]^ The periodic reactivation of HSV-1 suggests its effective evasion of the immune system at least in the initial stages of the infection. Since the presentation of peptides with MHC-I is the key step of antigen presentation and is linked to cytotoxic cell death,^[@ref38]^ blocking antigen presentation to cytotoxic T lymphocytes (CTL) helps the virus to remain hidden. In cells including neurons, peptides used for antigen presentation are primarily released by the 20S proteasome activity. The proteases and peptidases located in the cytoplasmic and endoplasmic reticulum (ER) then trim these larger peptides at the amino terminus for MHC-1 presentation.^[@ref39],[@ref40]^ These antigenic peptides are transported across the ER through a transporter associated with antigen processing (TAP). Inside the ER, these peptides are further trimmed into shorter peptides by peptidases and loaded on to MHC-1 molecules for surface display.^[@ref41]−[@ref45]^ Under normal physiological conditions, this mechanism leads to the recognition of virus-infected cells by the host immune system, especially for CTL responses. In 1995, two research groups demonstrated separately that HSV-1 evades host immune response by blocking the MHC-1 presentation of its antigenic peptides.^[@ref46],[@ref47]^ Both the groups demonstrated that infected cell protein-47 (ICP-47), encoded by HSV-1, inhibits the transport of peptides across the ER, leading to the inhibition of MHC-1-mediated antigen presentation. HSV-2 has also been reported to encode ICP-47 variant protein with the same function.^[@ref48]^ Recently, a report demonstrated that a conserved sequence, "~50~PLL~52~", present within the central region of ICP-47 is essential for its inhibitory activity.^[@ref48]^ These earlier findings and the observations made with the HSV-1 gK~208--235~ peptide in the current study led us to hypothesize that blocking of TAP by HSV-1-encoded ICP-47 may create a transient accumulation of HSV-1-derived peptides in addition to cellular peptides. This milieu of peptides of HSV-1 origin generated by 20S proteasome activity, which may include HSV-1 gK~208--235~ as well as other peptides of HSV-1, if not during every occasion of reactivation, at some point of time may contribute to intracellular aggregation events. The hypothesis is schematically depicted in [Figure [8](#fig8){ref-type="fig"}](#fig8){ref-type="fig"}. The HSV-1 gK~208--235~ peptide, which was identified in the current study, showed self-assembly into spheroid and cytotoxic aggregates. {#fig8} Peptide HSV-1 gK~208--235~ showed aggregation propensities equivalent to control Aβ~1--42~ peptide even though it is half the size of the latter. In addition to showing equivalent scores of aggregation and hydrophobicity, the two peptides also share a short homologous sequence of six amino acid residues at their respective carboxyl termini. The abundance of hydrophobic amino acid residues across the entire length of the HSV-1 gK~208--235~ peptide, as reflected in its aggregation parameters, makes it prone to instant aggregation. This behavior of the peptide was clearly reflected during *in vitro* solubilization studies aimed at analyzing the kinetics of its aggregation. The maximum aggregation, measured in terms of ThT fluorescence intensity, was obtained at 0 h, which suggests that the monomeric peptide, dissolved in dimethyl sulfoxide (DMSO), upon diluting with aqueous buffer aggregated instantly. The samples, collected up to 48 h, showed a decline in fluorescence up to 24 h and got stabilized thereafter. This observation is indicative of a dynamic aggregation process undergoing continuous peptide association and dissociation and stabilizes upon reaching an equilibrium-like state, as was observed with the stabilization of the ThT fluorescence. The C-terminus of the HSV-1 gK~208--235~ peptide is homologous to the C-terminus of the Aβ~1--42~ peptide ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}B). An N-terminal truncated Aβ~1--42~ peptide has been shown to form spherical oligomeric channels instead of amyloid fibrils, and this form of aggregates also has been found to be neurotoxic.^[@ref49]^ The truncated Aβ~1--42~ peptide is of the length equivalent to that of the HSV-1 gK~208--235~ peptide. The similarity between these two peptides at the level of length, sequence, and shape of their amyloid-like spheroid aggregates suggests a similar function for the HSV-1 peptide. For performing cytotoxicity experiments, aggregates of the HSV-1 gK~208--235~ peptide were generated at a 100 μM concentration, and thereafter, diluted samples were added to the splenocyte cultures. The fact that the peptide is highly prone to aggregation is shown by the kinetics of its aggregation, and its presence in free form is highly unlikely and it was assumed that the aggregation samples contained a negligible amount of free peptide. The AFM data had shown the structure of the aggregates to be nonfibrillar and of spheroid shape. A recent study, where the oligomeric species were found to be more toxic compared to the fibrils,^[@ref50]^ suggests that nonfibril structures are potent toxins. Additionally, the aggregates showed concentration-dependent cytotoxicity, even at aggregate concentrations as low as 2.5 μM, indicating the role of aggregates in the observed cytotoxicity. Although the actual mechanism of action of these aggregates remains to be elucidated and may be pursued in future studies, it may be noted that spherical aggregates were prepared at 100 μM (∼0.33 mg/mL); however, their cytotoxicity was observed at a much lower concentration (\<10 μM). The 10 μM peptide concentration is equivalent to 0.03 mg/mL, which is much lower than that of the cytoplasmic protein concentration, estimated to be ∼100 mg/mL,^[@ref51]^ and the total macromolecular concentration, including proteins, lipids, and sugar, of ∼400 mg/mL.^[@ref52]^ However, the crowded environment of the cell cytoplasm may provide a condition where the effective concentration of the freshly released peptides, after 20S proteasome cleavage, may remain high due to low diffusion amidst crowding and aided by their blocked transport into the ER lumen. These events may not be common under normal conditions, but may arise during HSV-1 infection and may lead to aggregation events at low peptide concentrations. Once the aggregates are formed, they were found toxic at very low concentrations, as shown in our study. Additionally, according to one report, proteins that are involved in neurodegenerative disease were found more likely to form amyloid fibrils under crowded conditions than in dilute solutions.^[@ref53]^ However, the effect of such molecular crowding on the HSV-1 gK~208--235~ peptide aggregation needs to be pursued further. We would like to link hypothetically another hallmark event that is observed in AD pathology, which is hyperphosphorylation of tau protein with HSV-1 recurrent infection.^[@ref54]^ The HSV-1 infection has been shown to contribute directly to the hyperphosphorylated forms of tau, which are extremely prone to aggregation.^[@ref11]^ Hence, HSV-1 during active infection is likely to express the protein ICP-47, which may block the TAP-facilitated transport of peptides across ER, and simultaneously HSV-1 kinase may phosphorylate tau protein making it prone to aggregation. The intraneuronal buildup of aggregation-prone viral peptides and hyperphosphorylated tau protein may trigger the aggregation process. With the background knowledge about established facts of HSV-1 infection and AD pathology and the observations made during the current study, we would like to propose a theory that HSV-1 peptides, which may be generated during HSV-1 infection due to proteasomal activity, may contribute to AD etiology and pathology in combination with other events triggered by the viral infection. This mechanistic hypothesis offers a fresh look at the events that follow an HSV-1 infection and may play a combined role in AD; although, the hypothesis still remains to be validated by more substantial *in vitro* and *in vivo* support. Materials and Methods {#sec4} ===================== Protein Sequences {#sec4.1} ----------------- Protein sequences of HSV-1 were retrieved from the NCBI protein database (<http://ncbi.nlm.nih.gov/protein/>) and UniProt (<http://uniprot.org/>) using keywords like human herpesvirus 1, envelope glycoproteins, or by the respective protein name. The protein sequence of Fragile-X-Mental Retardation-1 Protein (FMRP-1), an abundant and constitutively expressed protein in normal healthy brain neurons,^[@ref55]^ (UniProt, Q06787), and a well-established amyloidogenic peptide, Aβ~1--42~,^[@ref15]^ were used as negative and positive controls, respectively, for *in silico* aggregation prediction. Prediction Tools {#sec4.2} ---------------- Online software TANGO (<http://tango.crg.es/>), AGGRESCAN (<http://bioinf.uab.es/aggrescan/>), and AMYLPRED2 (<http://aias.biol.uoa.gr/AMYLPRED2/>), based on computer algorithms, were used for the prediction of aggregation-prone regions in unfolded polypeptide/peptide chains.^[@ref56]−[@ref59]^ For predicting the proteasome cleavage sites of proteins, Pcleavage, a support vector machine (SVM)-based method (<http://www.imtech.res.in/raghava/pcleavage/index.html>), and NetChop V3.0, based on artificial neural network (<http://www.cbs.dtu.dk/services/NetChop/>), were used.^[@ref17],[@ref18]^ The hydrophobicity of the peptide fragments was calculated using an online peptide analyzing tool (Thermo Fisher). Prediction of 20S Proteasome Cleavage Sites {#sec4.3} ------------------------------------------- The retrieved protein sequences were scrutinized for aggregation score using software TANGO and AGGRESCAN at default parameter values, viz., pH 7.4 and 310 K temperature. No sliding window was used for the TANGO and AGGRESCAN score calculation. Selected glycoproteins of HSV-1 were fed into the proteasome cleavage prediction tools Pcleavage and NetChop with threshold/cutoff values of 0.6 and 0.9, respectively. Aggregation Sample Preparation {#sec4.4} ------------------------------ A 28-residue-long peptide fragment (~208~LYHRPAIGVIVGCELMLRFVAVGLIVGT~235~) derived from HSV-1 glycoprotein K was chemically synthesized in the lyophilized form with \>95% purity ("S" BioChem, India). The 1 mg peptide was dissolved in a small amount (∼25 μL) of dimethyl sulfoxide (DMSO) and diluted with Milli-Q water to prepare a stock solution of 1 mg/mL such that the concentration of DMSO remains less than 3%. The stock solution was then centrifuged at 15 000*g* for 2 min. The final concentration of the dissolved peptide in the stock solution was estimated by Bradford's assay. Aggregation samples were prepared by mixing desired volumes of peptide and phosphate-buffered saline (10 × PBS) stock to obtain desired concentrations, viz., 55, 100, 200, 300, and 600 μM, of the peptide in 1 × PBS. The resultant peptide solutions were incubated on a thermomixer at 37 °C and stirred at 1200 rpm for the desired period. The samples were then analyzed using Congo red absorption and Thioflavin T fluorescence assays. Congo Red Absorption Assay {#sec4.5} -------------------------- The aggregation samples were mixed with Congo red dye (Sigma-Aldrich) for absorption assay. A working solution of 15 μM Congo red was prepared from a stock of (0.1%) 1435 μM stock solution. The aggregation sample (150 μL, 100 μM) was mixed with 3.13 μL of Congo red stock solution in such a way that the final prepared concentration of 50 μM of peptide aggregate and 15 μM Congo red is achieved in a total of 300 μL volume of 1 × PBS. The absorption spectra of the resultant samples were recorded at 400--600 nm wavelength at a resolution of 1 nm using an ELISA plate reader (Thermo Fisher). Thioflavin T (ThT) Fluorescence Assay {#sec4.6} ------------------------------------- The selective binding of mature amyloid fibrils with Thioflavin T emits fluorescence in the wavelength range of 485--500 nm.^[@ref60]^ Different volumes of samples of aggregation reactions, set at 55, 100, 200, 300, and 600 μM concentrations of peptide, were mixed with the ThT stock solution (10×) in 1 × PBS to obtain a final concentration of 20 μM ThT and a peptide concentration of 50 μM. The resultant samples were excited at 450 nm, and the fluorescence was recorded in the range of 470--700 nm on a fluorescence spectrophotometer (Cary Eclipse, Agilent Technologies). The slit width of excitation and emission was 5 nm. The peak intensity of fluorescence at 485 nm was used for statistical analysis (mean values and standard error mean). Atomic Force Microscopy (AFM) {#sec4.7} ----------------------------- A custom cut 10 × 10 mm glass slide of thickness 1 mm was treated with a saturated solution of potassium hydroxide in absolute ethanol, as described in ref ([@ref61]), to minimize the surface roughness. The peptide aggregate was washed three times with Milli-Q water by centrifugation (12 000*g* for 2 min) and resuspension cycles. The washed peptide aggregate was incubated on a glass slide and dried in a dust-free environment for 10 min, and the images were obtained by AFM scan. The AFM data were analyzed using WSxM software.^[@ref62]^ Atomic force microscopy (AFM) was performed at Material Research Centre, Malaviya National Institute of Technology (MNIT), Jaipur, Rajasthan, India. ATR-FTIR Spectroscopy {#sec4.8} --------------------- The peptide aggregate was analyzed by attenuated total reflection--Fourier transform infrared (ATR-FTIR) spectroscopy. The peptide aggregate sample washed with 10 μL of Milli-Q water was dried on a diamond crystal in an ATR cell. A PerkinElmer Spectrum Two FTIR spectrometer with an MCT detector was used to measure the spectrum in the spectral range of 4000--400 cm^--1^ at a resolution of 1 cm^--1^ and an average of 64 scans. FTIR spectroscopy was performed at Material Research Centre, Malaviya National Institute of Technology (MNIT), Jaipur, Rajasthan, India. Cell Viability Assay {#sec4.9} -------------------- The C57BL/6 mice were kept at the animal house facility of the University of Rajasthan, Jaipur. The splenocytes were gifted by Dr. A. S. Ansari's lab for the current experiment. The mice were sacrificed, and the spleen was removed aseptically. The isolated spleen was washed in cold PBS. The spleen was teased gently against a sterile frosted glass slide to dislodge the cell mass in RPMI 1640 medium. The resultant splenocyte suspension was washed in PBS and the cells were collected by centrifugation at 200*g* for 5 min at 4 °C. The pellet obtained was suspended in a red blood cell (RBC) lysis buffer (155 mM NH~4~Cl, 12 mM NaHCO~3,~ and 0.1 mM ethylenediaminetetraacetic acid, EDTA) and incubated for 10 min on ice, and the cells were collected by centrifugation at 200*g* for 5 min. The collected cell pellet was washed again in RPMI 1640 medium and finally resuspended in 2 mL of RPMI 1640 medium with 10% serum. The counting of viable cells was performed by mixing the diluted cell suspension with 0.4% trypan blue in a 1:1 ratio and loading the mixture onto a hemocytometer, wherein the cells were counted using an inverted microscope (Zeiss Primovert). The final splenocyte count was adjusted to 1 × 10^6^ cell/mL. The splenocyte (0.5 mL) in RPMI 1640 medium with 10% serum was added to the wells of 24-well culture plates with different concentrations (2.5, 5, 10, and 20 μM) of peptide aggregate and incubated in a 5% CO~2~ incubator at 37 °C up to 48 h. The peptide aggregates were prepared at a concentration of 100 μM, as described earlier. The images of the cells were captured by a Zeiss Primovert inverted microscope at a final magnification of 400×, and the counting of viable cells was performed by mixing the cell suspension with 0.4% trypan blue at a 1:1 ratio. Statistical Analysis {#sec4.10} -------------------- The statistical significance was accepted at *P* \< 0.05. Analysis of variance (ANOVA) test was used for statistical analysis. Statistical analyses were performed using GraphPad Prism 6.0 software. The Supporting Information is available free of charge at [https://pubs.acs.org/doi/10.1021/acsomega.0c00730](https://pubs.acs.org/doi/10.1021/acsomega.0c00730?goto=supporting-info).Average aggregation of scores (low to high) of protein sequences of HSV-1; depiction of 20S proteasome cleavage sites of gK and gM of HSV-1, predicted with Pcleavage and NetChop; analysis of HSV-1 gK~208--235~ and Aβ~1--42~ peptides using AMYLPRED2 algorithm ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acsomega.0c00730/suppl_file/ao0c00730_si_001.pdf)) Supplementary Material ====================== ###### ao0c00730_si_001.pdf The authors declare no competing financial interest. The authors have complied with all relevant guidelines and regulations of the Institutional Animal Ethics Committee (IAEC) of University of Rajasthan (CPCSEA registration no. 1678/GO/RE/S/12/CPCSEA) for animal testing and research. All of the experimental protocols performed were approved by IAEC for a study in collaboration with Dr. A. S. Ansari's lab (IAEC approval No: UDZ/IAEC/II/04; dated 10-05-2019) in C57BL/6 strain of mice. The splenocytes, from the above-mentioned lot of mice, were obtained from Dr. A. S. Ansari's lab, University of Rajasthan. The authors thank Dr. Jay Kant Yadav for suggestions on peptide stock preparation and Dr. Anuj K. Sharma for providing the fluorescence spectrophotometer facility at Central University of Rajasthan. The authors gratefully acknowledge the generous gift of splenocytes of C57BL/6 mice by Dr. A. S. Ansari, University of Rajasthan, Jaipur. They also thank the Material Research Centre, MNIT, Jaipur, for allowing them to perform AFM and FTIR spectroscopy at their facility. They gratefully acknowledge the funding received from SERB, India, in the form of a research grant (SERB Grant ref No. YSS/2015/000138). VKS and SK thank UGC for research fellowships. The authors thank Central University of Rajasthan for support.

Introduction {#s1} ============ Glaucoma is one of the leading causes of blindness in the world. At least 2 million people have primary open-angle glaucoma in the United States, and the number is rising.^[@i2164-2591-6-5-14-b01]^ Glaucoma is a progressive neurodegenerative disease, and patients need to be monitored so that clinicians can detect evidence of progression in a timely manner. By examining the optic nerve, clinicians can look for signs of progression or signs that may be associated with progression of the optic nerve, such as neuroretinal rim defects, optic disc hemorrhage, peripapillary atrophy, and retinal nerve fiber layer abnormalities, all of which are important for the identification of glaucomatous progression.^[@i2164-2591-6-5-14-b02]^ Optical coherence tomography (OCT) has been an important aid in monitoring glaucoma progression. To date, the retinal nerve fiber layer (RNFL) measurement concentric to the optic nerve has been the most commonly used parameter to achieve this goal.^[@i2164-2591-6-5-14-b03][@i2164-2591-6-5-14-b04][@i2164-2591-6-5-14-b05][@i2164-2591-6-5-14-b06]--[@i2164-2591-6-5-14-b07]^ Nevertheless, detection of glaucoma progression remains one of the most challenging tasks in the management of glaucoma. Review of stereoscopic disc photographs is one of the fundamental forms of image analysis that allows structural changes to be identified by clinicians.^[@i2164-2591-6-5-14-b08]^ Determination of structural progression with serial optic disc photographs suffers from poor interobserver agreement. Several studies such as the one conducted by Jampel and associates^[@i2164-2591-6-5-14-b09]^ have shown this poor interobserver reliability (*j* = 0.20) when glaucoma specialists compared optic disc photographs for signs of structural progression. This study was designed to compare the accuracy of optic disc photograph subtraction as an aid to detect glaucoma progression to that of the traditional method of examining serial stereoscopic disc photographs by glaucoma specialists. We propose an optic disc image subtraction method as part of a software tool that provides clinicians with supplemental information regarding signs of glaucoma deterioration as part of the electronic medical record and in real time, and we evaluate its performance with regard to detection of glaucoma progression. Methods {#s2} ======= Participants {#s2a} ------------ Ninety-two glaucomatous eyes of 65 patients were selected from the Glaucoma Division\'s database at the University of California, Los Angeles (UCLA). Diagnosis of glaucoma was based on presence of thinning or notching of the neuroretinal rim, RNFL defects or significant asymmetry in the neuroretinal rim between the two eyes, and visual field loss that was entirely consistent with the optic nerve damage, as measured with standard automated perimetry. Eyes with a minimum of 1-year follow-up were included. Exclusion criteria were (1) concomitant ocular disease other than glaucoma; (2) any media opacity that would interfere with imaging and lead to poor-quality optic disc photographs; (3) poor illumination of the photographs; and (4) improper stereoscopy (large vertical shift on the screen between the stereoscopic pairs. The quality of the images was assessed on the basis of good-quality stereoscopy, contrast, clarity, illumination, and absence of excessive vertical shift between the stereoscopic pair. The study was approved by the Institutional Review Board of UCLA and was performed in adherence with the Declaration of Helsinki. Disc Photograph Review {#s2b} ---------------------- Stereoscopic disc photographs taken at baseline and last follow-up were reviewed by three expert glaucoma specialists (JC, KNM, and RA) 2 years prior to initiation of the study and were identified as either progressive or stable. The experts were masked to clinical information and each other\'s decisions. Progression of glaucoma was defined as a loss of neuroretinal rim that could be accompanied by an increase in peripapillary atrophy, by worsening of visible RNFL defects, or by optic disc hemorrhages.^[@i2164-2591-6-5-14-b10]^ A decision was made if at least two observers agreed on the progression (or stability) of the eyes. Thirty-three eyes were identified as having progressive optic nerve damage. Fifty-nine eyes were identified as stable based on review of stereoscopic disc photographs. Imaging Methods {#s2c} --------------- All photographs were obtained by the same fundus camera (Model 60306; Carl Zeiss, Oberkochen, West Germany) at the flash 3 setting (240 watts). One baseline and follow-up image each from each eye were aligned with software (i2k Align Retina; DualAlign, LLC, Clifton Park, NY) based on the dual-bootstrap algorithm. Only the right image of each stereo pair was used. In a preprocessing step, the original stereo fundus photographs (4400 × 3600 pixels) were rescaled to 1100 × 900 pixels for computational efficiency. Images were stored in an uncompressed, lossless TIFF format. These images were then converted to gray scale and underwent histogram matching to enhance contrast and remove the influence of changes in global illumination from baseline to follow-up visit ([Fig. 1](#i2164-2591-6-5-14-f01){ref-type="fig"}). {#i2164-2591-6-5-14-f01} The intensity of the baseline image was subtracted from the follow-up image, and the difference was shown as a colormap superimposed on the grayscale follow-up image. Thresholding of the difference image was performed by the triangle method.^[@i2164-2591-6-5-14-b11][@i2164-2591-6-5-14-b12]--[@i2164-2591-6-5-14-b13]^ In the subtraction colormaps, red pixels within the optic disc area indicate a decrease in intensity from the baseline to the follow-up image, and green pixels indicate an increase ([Fig. 2](#i2164-2591-6-5-14-f02){ref-type="fig"}). All green (or red) pixels have the same color intensity. In case of progressive neural rim thinning, red regions, indicating an increase in brightness, were superimposed on the thinning part of the rim (both examples in [Fig. 2](#i2164-2591-6-5-14-f02){ref-type="fig"}). If the optic disc was stable, no large areas of red or green pixels were created in the colormap ([Fig. 3](#i2164-2591-6-5-14-f03){ref-type="fig"}). All the image-processing procedures were streamlined and performed in MATLAB (R2015b; The Mathworks, Natick, MA). A software tool with a user-friendly interface was developed that allows clinicians to choose optic disc photographs with multiple image file formats. {#i2164-2591-6-5-14-f02} {#i2164-2591-6-5-14-f03} Grading Method {#s2d} -------------- These subtraction colormap images were reviewed and graded by three masked glaucoma experts (RA, NP, and JC) solely based on these images. The grading was performed in two different scales: (1) a progression scale ranging from 1 (no progression) to 5 (significant progression) and (2) a binary scale where 1 represents glaucoma progression and 0 represents stability. Statistical Analysis {#s2e} -------------------- The decision of the expert panel after review of follow-up and baseline stereoscopic disc photographs was considered the gold standard for the calculation of sensitivity, specificity, and accuracy of the gradings based on the colormaps. Sensitivity was calculated as the proportion of progressive eyes that were correctly identified as such. Specificity measured the proportion of stable eyes that were correctly identified as stable. Accuracy was the ratio of truly progressive and truly stable eyes to the total number of eyes. κ Statistics (weighted κ) were used to estimate agreement of ordered categorical grading between observers. Agreement (κ) can be calculated by dividing the degree of agreement actually achieved beyond chance by the degree of agreement that is attainable beyond chance. Results {#s3} ======= Demographic and clinical characteristics of the study sample are shown in [Table 1](#i2164-2591-6-5-14-t01){ref-type="table"}. The median time interval between baseline and follow-up visits was 4.4 years (range: 1.0--16.8, interquartile range \[IQR\]: 3.2--5.9). ###### Demographic Data and Baseline/Follow-up Characteristics of the Study Cohort  In this cohort, 31 (out of 33) progressive eyes had progressive rim loss, and only two eyes experienced RNFL thinning as the only sign of progression. In the areas of RFNL loss, the light is absorbed by the RPE; thus, the affected areas appear darker. As such, they are superimposed with green pixels in the subtraction colormap (green areas in the in the peripapillary area in [Fig. 2C](#i2164-2591-6-5-14-f02){ref-type="fig"}). Three eyes had optic disc hemorrhages at baseline, which disappeared at the subsequent visit and were followed by progressive rim loss in the same location. The software was able to highlight the disappearance of all three disc hemorrhages indicated by red pixels in the corresponding area of the optic disc as shown in [Figure 4](#i2164-2591-6-5-14-f04){ref-type="fig"}. {#i2164-2591-6-5-14-f04} With the aid of subtraction maps, clinicians detected glaucoma progression in 25 to 27 (out of 33) of the progressive eyes and 8 to 10 (out of 59) of stable eyes. [Table 2](#i2164-2591-6-5-14-t02){ref-type="table"} summarizes sensitivity, specificity, and accuracy results. Classification accuracy for the examination performed by the three clinicians was 81.5%, 81.5%, and 84.8%. Sensitivities of the examination performed by the three clinicians were 0.76, 0.79, and 0.82, and the specificities were 0.85, 0.83, and 0.86, respectively. Weighted κ scores revealed good agreement between clinicians (weighted κ = 0.68; 95% confidence interval \[CI\]: 0.60--0.77) for progression grades (1--5 scales). Agreement among clinicians was substantial (weighted κ = 0.81; 95% CI: 0.74--0.85) for binary scores. ###### Analysis of Sensitivity, Specificity, and Accuracy of Individual Reviewers Using the Subtraction Colormaps Produced by the Software to Detect Glaucomatous Structural Changes  In order to validate the image subtraction algorithm and check for possible false-positive results, 10 pairs of disc photographs taken on the same day, as well as 10 pairs of identical disc photos, were fed to the software; as expected, no false-positive results were observed, and the resulting subtraction maps were all clear ([Fig. 5](#i2164-2591-6-5-14-f05){ref-type="fig"}). {#i2164-2591-6-5-14-f05} On review of clinically progressing eyes where the colormaps failed to identify progression, we found that longitudinal optic disc changes tended to be more subtle; hence, a false-negative result was produced ([Fig. 6](#i2164-2591-6-5-14-f06){ref-type="fig"}). Also, in the case of uneven illumination in disc photographs, the histogram-matching technique failed in removing the effect of inhomogeneous illumination. Accordingly, distortions in the form of large areas of red and green pixels were produced in the periphery of subtraction colormaps for three progressive eyes as well as four stable eyes ([Fig. 7](#i2164-2591-6-5-14-f07){ref-type="fig"}). As a result of such distortions, clinicians might mistake stable eyes for progressive eyes. [Figure 8](#i2164-2591-6-5-14-f08){ref-type="fig"} shows how the software could highlight fundus camera artifacts in the form of circle-shaped darker areas in the central (red pixels) and temporal (green pixels) regions of the optic disc. {#i2164-2591-6-5-14-f06} {#i2164-2591-6-5-14-f07} {#i2164-2591-6-5-14-f08} Discussion {#s4} ========== Detection of glaucoma progression involves monitoring changes in both functional (visual field) and structural (optic disc or RNFL) domains. Stereoscopic disc photographs have been used widely by ophthalmologists to document the optic disc appearance and detect changes over time. This approach has been demonstrated to be superior to subjective documentation and drawings of the disc.^[@i2164-2591-6-5-14-b14]^ This method has been used in numerous glaucoma studies such as the Ocular Hypertension Treatment Study,^[@i2164-2591-6-5-14-b15]^ Early Manifest Glaucoma Trial,^[@i2164-2591-6-5-14-b16]^ and The Collaborative Initial Glaucoma Treatment Study.^[@i2164-2591-6-5-14-b17]^ Evidence suggests that structural damage can often be detected before development of detectable visual field defects.^[@i2164-2591-6-5-14-b18][@i2164-2591-6-5-14-b19][@i2164-2591-6-5-14-b20][@i2164-2591-6-5-14-b21][@i2164-2591-6-5-14-b22][@i2164-2591-6-5-14-b23][@i2164-2591-6-5-14-b24]--[@i2164-2591-6-5-14-b25]^ Some signs of structural progression are not easily detectable. With the proliferation of electronic medical record systems, incorporation of optic disc image subtraction tools, and in general, change detection software, may help clinicians detect glaucoma deterioration. The idea of utilizing image subtraction to detect optic nerve changes and glaucomatous damage progression is not novel.^[@i2164-2591-6-5-14-b26][@i2164-2591-6-5-14-b27][@i2164-2591-6-5-14-b28][@i2164-2591-6-5-14-b29]--[@i2164-2591-6-5-14-b30]^ Reus and associates^[@i2164-2591-6-5-14-b20]^ found that the accuracy of diagnosing glaucoma by reviewing stereoscopic optic disc photographs to be at best 81%. New computer software such as matched flicker software programs^[@i2164-2591-6-5-14-b31],[@i2164-2591-6-5-14-b32]^ or optic disc photograph subtraction proposed in this study could assist clinicians in highlighting possible changes of the optic nerve head. With the subtraction colormaps, clinicians had an acceptable accuracy in finding glaucoma progression, which was consistent among three glaucoma experts, as measured by the weighted κ statistic. This interobserver agreement was good for progression grades, and it understandably improved when binary scores were used. Furthermore, fundus camera artifacts were highlighted by the software. In this study, for the sake of analysis, clinicians used only the subtraction colormaps to find progression. However, the identification of glaucomatous optic nerve change is only a part of the bigger clinical picture for each patient and should be used along with all the other modalities. To reach this goal, as stated in previous studies,^[@i2164-2591-6-5-14-b32]^ we believe that ancillary tests in glaucoma should have higher specificity (the percentage of negative designations that are truly negative) rather than higher sensitivity (the proportion of positive diagnoses that are truly positive) because of the low incidence of progression in eyes under treatment. In our study, the overall specificity was 85%. Distortions in the subtraction colormaps as a result of inhomogeneous illumination caused false-positive results. For instance, clinicians may perceive such distortions as RNFL thinning. Fundus camera artifacts are potentially another source of false positives. If these artifacts are located within the clinical disc margin, they could be mistaken for neural rim loss, although the pattern can usually be recognized as spurious. Parallax often appeared on subtraction maps, suggesting that the proposed optic disc image subtraction method has the potential to automatically detect parallax. A separate study is worthwhile to evaluate the accuracy of parallax detection. Newer fundus cameras ensure that fixed-angle photographs of the optic disc can be taken. In the absence of fixed-angle optic disc photographs, parallax will cause apparent shifts in vessel position that may be perceived as glaucoma-related change. Radcliffe et al.^[@i2164-2591-6-5-14-b31]^ have shown that retinal blood vessels shift locations in eyes with progressive neuroretinal rim loss due to glaucoma. In general, only global blood vessel positional shifts may result from parallax, which is a global image transform and cannot distort a small region of a blood vessel. Less experienced clinicians should use caution when distinguishing focal blood vessel shift from global shifts or expansion/contraction of arteries (without displacement). Our study demonstrated that the subtraction colormaps produced by the optic disc image subtraction software offered fair sensitivity for detecting glaucomatous progression and could also rule out its presence with a good specificity. Observation of subtraction colormaps yielded good to substantial interobserver agreement and provided additional information to the clinician, in a real time, for detecting glaucoma progression. Compared to viewing of stereoscopic optic disc photographs and matched flicker software, the advantages of the proposed image subtraction tool are the ease of use, independence from stereoscopic photographs, and its potential use in real time; the printout is static and can be placed either on physical charts or electronic medical records. It should be noted that the optic disc image subtraction tool is meant to complement existing methods for detection of progression, not to replace them. This study had a number of limitations. First, image quality is a concern for every imaging modality and study. We used only images with acceptable quality in this study, although in clinical settings, suboptimal optic disc photos (decentered, motion artifact, low illumination) are not infrequent. The diagnostic value of subtraction colormaps significantly depends on the quality of the baseline and follow-up optic disc photographs. Several qualitative aspects are important as far as stereoscopic disc photographs are involved, such as image resolution (governed by camera resolution and field of view), uniformity of retinal illumination (determined by the acquisition condition and the spherical geometry of the eye), and the contrast of objects (blood vessels, optic disc, nerve fibers, hemorrhages, etc.) on the retina. Enhancement of the last item through histogram matching was considered in this study. As expected, the histogram-matching method was not able to compensate for uneven illumination on baseline or follow-up images. This was apparent in seven eyes in our cohort. Uneven illumination due to nonideal acquisition conditions can introduce distortions into optic disc images. These distortions are commonly perceived as smooth intensity variations across the optic disc photographs^[@i2164-2591-6-5-14-b33],[@i2164-2591-6-5-14-b34]^ and must be eliminated in the future version of the proposed software. Presence of cataract or other media opacities could also compromise the results. In future work, we plan to enhance the histogram-matching method by subdividing the image into smaller segments and performing histogram-matching in an adaptive manner. These modifications will enable the software to tolerate uneven illumination and moderate levels of media opacity. Second, we used an external software tool to align the baseline and follow-up images. Incorporation of the image alignment step into the proposed software will produce a more clinically useful tool for use in routine care of patients. Third, confirmation of progression in glaucomatous eyes can be challenging. Therefore, there may be no consensus among clinicians regarding progression in some glaucomatous eyes; in this study, we used the decision of an expert panel on progression (or stability) of each eye. Such decisions were considered the gold standard in evaluation of the performance of the proposed software. There is a possibility that subtle signs of glaucoma progression might have been missed. The majority of progressing eyes in our study were identified based on changes in the neuroretinal rim. Since the number of eyes with disc hemorrhages or RNFL thinning was relatively low, a statistically meaningful subgroup analysis was not possible. We are planning to include more patients with disc hemorrhages or RNFL thinning in order to make this subgroup analysis statistically meaningful. Last, although our approach has the potential to improve the interobserver agreement on structural progression, evaluation of serial optic disc photographs that are augmented with subtraction colormaps was still qualitative rather than quantitative, and expert opinions may vary. In the context of glaucoma progression detection with confocal scanning laser tomography, and in particular the Heidelberg Retina Tomograph (HRT; Heidelberg Engineering, Heidelberg, Germany), Chauhan and coinvestigators^[@i2164-2591-6-5-14-b35]^ have quantified and evaluated the size and depth of clusters of pixels that have undergone significant change from the baseline. They indicated that HRT\'s topographic change analysis algorithm performs at least as well as observer classifications of disc photographs. Progressive RNFL thinning measured with OCT can be used to detect progressive disease. Leung and colleagues used the RNFL thickness map provided by event-based change analysis software, called the Guided Progression Analysis (GPA), for detecting and identifying patterns of RNFL progression.^[@i2164-2591-6-5-14-b03]^ The GPA software compares the RNFL thickness patterns of follow-up scans to a baseline scan and indicates thinning as possible loss (yellow) or likely loss (red). Even though we evaluated the performance of the proposed image subtraction technique by clinicians solely looking at the subtraction colormaps, in practice this technique must be used along with other routine structural (e.g., side-by-side viewing of stereoscopic optic disc photographs, circumpapillary RNFL thickness, and optic nerve head parameters) as well as functional tests. The correlation between the extent of change shown on subtraction colormaps and other structural/functional changes needs to be thoroughly explored. The proposed software tool and the underlying image subtraction method produce a single static image (subtraction colormap) that can highlight areas of possible concern. The performance of the proposed image subtraction method will need to be compared to other means of detecting structural glaucomatous changes, such as alternation flicker, in future studies.^[@i2164-2591-6-5-14-b36]^ Supplementary Material ====================== ###### Click here for additional data file. Supported by an unrestricted departmental grant from Research to Prevent Blindness. Presented as a poster at the annual meeting of the Association for Research in Vision and Ophthalmology (ARVO), Baltimore, Maryland, United States, May 7--11, 2017. Disclosure: **N. Amini**, None; **R. Alizadeh**, None; **N. Parivisutt**, None; **E. Kim**, None; **K. Nouri-Mahdavi**, None; **J. Caprioli**, None

Citation {#SECID0EIBAC} ======== Carvalho RF, Amaral-Silva PM, Spadeto MS, Nunes ACP, Carrijo TT, Carvalho CR, Clarindo WR (2017) First karyotype description and nuclear 2C value for *Myrsine* (Primulaceae): comparing three species. Comparative Cytogenetics 11(1): 163--177. [https://doi.org/10.3897/CompCytogen.v11i1.11601](10.3897/CompCytogen.v11i1.11601) Introduction {#SECID0E6BAC} ============ Previous studies regarding the chromosome number in Primulaceae (s. [@B6]) are available for some genera, as: *Cyclamen* Linnaeus, 1753 ([@B9], [@B24]), *Anagallis* Linnaeus, 1753 ([@B3], Bennett and Leitch 2012), *Lysimachia* Linnaeus, 1753 ([@B7], Bennett and Leitch 2012, [@B14]), *Androsace* Linnaeus, 1753 ([@B15]), *Elingamita* Baylis, 1951 ([@B17]), *Trientalis* Linnaeus, 1753 ([@B46]), *Ardisia* Swartz, 1788 ([@B26]), *Primula* Linnaeus, 1753 ([@B2], [@B13], [@B45]), and *Dodecatheon* Linnaeus, 1753 ([@B33]), and *Myrsine* Linnaeus, 1753 ([@B10], [@B17], [@B18], [@B22], [@B40]). Except the genus *Cyclamen* and *Myrsine*, these taxa comprise annual and biennial herbaceous species. The cosmopolitan *Myrsine* Linnaeus is one of the main genera of Primulaceae, considering species richness, represented by tree and shrub species ([@B23]). Its members are generally dioecious plants, characterized by ramiflorus and congested inflorescences, and flowers with oppositipetalous stamens. Despite *Myrsine* being one of the largest and most important genera of Primulaceae, only eighteen species, among the 300 estimated from this genus, have been studied regarding cytogenetic aspects. Fifteen of these species occur in the African, Asian and Oceania continents (*Myrsine coxii* Cochayne, 1902, *Myrsine divaricata* Cunningham, 1839, *Myrsine kermadecensis* Cheeseman, 1887, *Myrsine nummularia* (Hooker f.) Hooker f., 1867, *Myrsine salicina* (Hooker f.) Hooker f., 1864, *Myrsine argentea* Heenan et de Lange, 1998, *Myrsine oliveri* Allan, 1961, *Myrsine chathamica* Mueller, 1864; *Myrsine africana* Linnaeus, 1753; *Myrsine sandwicensis* Candolle, 1841, *Myrsine seguinii* Léveille, 1914, *Myrsine semiserrata* Wallich, 1824, *Myrsine australis* (A. Richard, 1832) Allan, 1947, *Myrsine capitellata* Wallich, 1824), and just three occurs in America continent (*Myrsine matensis* (Mez, 1902) Otegui, 1998; *Myrsine guianensis* (Aublet, 1775) Kuntze, 1891, *Myrsine coriacea* (Swartz, 1788) Brown ex Roemer et Schultes, 1819. The chromosome number (2n = 46 or 2n = 48) was the only karyotype data reported, without any images of the chromosomes. In addition, the evolutionary aspects that culminated in the karyotype diversification within the genus are poorly understood. One interesting ecological aspect observed in Neotropical species of *Myrsine* that occur in Brazil is that some of them occur in more than one biome, as Cerrado, Atlantic Forest, and Amazonian Forest, while others are restricted of one of these biomes, as Atlantic Forest (BFG 2015). Among species that occur in Atlantic Forest, for example, some are able to occupy different types of vegetation within this biome, including Restinga Vegetation, High Altitude Campos, Rocky Outcrops, Ombrophyllous and Mixed Ombrophyllous Forests, while others are able to occupy just one type of vegetation ([@B19]). Considering the distinct ecological aspects, cytogenetic studies are relevant to show other differences between these species. Studies combining cytogenetics and nuclear DNA content have offered data for understanding evolutionary processes in different species ([@B16], [@B25]). Measurement of the nuclear DNA content is complementary to cytogenetic information and is useful for detecting genome size variations between related species ([@B28], [@B25]). Fine adjustments in cytogenetic procedures, combining advances in microscopy and image analysis systems, can provide accurate karyotype characterization for *Myrsine* species. Here, we study three species of *Myrsine* that occur in contrasting types of vegetation of the Brazilian Atlantic Forest, aiming to determine the chromosome number, describe the karyotype and measure the nuclear DNA. Material and methods {#SECID0EXPAC} ==================== Plant samples {#SECID0E2PAC} ------------- Three species were selected for this study: 1. *Myrsine coriacea* (Voucher -- T.T. Carrijo 1458, VIES herbarium), which is a widespread species in Atlantic Forest found in all types of vegetation, including open areas within Ombrophyllous and Mixed Ombrophyllous Forests, Rock Outcrops, High Altitude Campos, and Restinga Vegetation; 2. *Myrsine umbellata* (Voucher -- T.T. Carrijo 1467, VIES herbarium), which is found in mostly all types of vegetation of *Myrsine coriacea*, except High Altitude Campos; and 3. *Myrsine parvifolia* (Voucher -- T.T. Carrijo 2232, VIES herbarium), a species restricted to Restinga vegetation (BFG 2015). Fruits and leaves of all species were collected. *Myrsine coriacea* and *Myrsine umbellata* were sampled from October 2012 to July 2015 in a forest remnant located in Iúna municipality, Espírito Santo (ES) State, Brazil (20°21\'6\"S -- 41°31\'58\"W), characterized as Rocky Outcrops, at 600 (*Myrsine coriacea*) and 1,100 m.s.m (*Myrsine umbellata*). *Myrsine parvifolia* was collected in a forest remnant located in Guarapari municipality, ES, Brazil (20°36\'15\"S -- 40°25\'27\"W), characterized as coastal sandy plains vegetation (Restinga) at sea level. Leaves and fruits of *Solanum lycopersicum* L. and *Pisum sativum* L. (internal standards for flow cytometry -- FCM, 2C = 2.00 pg and 2C = 9.16 pg, respectively, [@B38]) were supplied by Dr. Jaroslav Doležel (Experimental Institute of Botany, Czech Republic). In vitro plantlet recovering {#SECID0ECFAE} ---------------------------- Fruit pericarp was manually removed and the seeds were desinfested according to [@B34] and germinated in a medium composed of MS salts (Sigma) and vitamins ([@B31]), 30 g L^-1^ sucrose (Sigma), 7 g L^-1^ agar and 2.685 µM naphthaleneacetic acid (NAA, Sigma). *Solanum lycopersicum* and *Pisum sativum* seeds were subjected to the same disinfestation procedure and inoculated in medium without NAA. Germination was done at 25 °C under a 16/8 hours (light/dark) regime. Nuclear 2C value measurement {#SECID0E1GAE} ---------------------------- In order to adapt the FCM for *Myrsine*, the following procedures were done: (a) initially, from leaves collected in the field of male and female individuals (samples) and of the two standards; (b) afterward, replacing the dithiothreitol antioxidant by polyethylene glycol (PEG) in nuclei isolation buffer; and (c) from leaves of the samples and *Pisum sativum* plantlets in vitro cultivated. Nuclei suspensions were obtained by co-chopping ([@B20]) leaf fragments (1 cm^2^) cut from each sample (*Myrsine* species) and standard (*Solanum lycopersicum* or *Pisum sativum*). The suspensions were processed and stained following [@B35] and [@B38] and analyzed with the flow cytometer Partec PAS II/III (Partec GmbH). *Myrsine* genome size was measured by multiplying the 2C value of the internal standard using the fluorescence intensity corresponding to G~0~/G~1~ nuclei peak. Mean 2C values were compared by the *F* test at 5% probability. Cytogenetic analysis {#SECID0EUJAE} -------------------- Roots were cut from the in vitro plantlets, treated with 5.0 μM amiprofos-methyl (APM) (Agrochem KK Nihon Bayer) for 12, 15, 18 or 24 h at 4°C, rinsed in distilled water (dH~2~O) for 20 min and fixed in methanol:acetic acid (3:1) for 24 h. The fixative solution was changed three times and the material was stored at -20°C ([@B12]). The roots were washed, macerated in 1:5 pectinase solution (enzyme:dH~2~O) for 3 h at 34°C, or 1:20 enzymatic pool (4% cellulase -- Kinki Yakult MFG, 1% macerozyme -- Kinki Yakult MFG, and 0.4% hemicellulase -- Sigma) for 1 h 30 min or 1 h 45 min at 34°C, washed in dH~2~O, fixed, and stored at -20°C. Slides were prepared and stained according to [@B12] and analyzed on a Nikon eclipse Ci-S microscope (Nikon). Prometaphases and metaphases were captured using the 100× objective and a CCD camera (Nikon Evolution^TM^) coupled to a Nikon microscope 80i (Nikon). About 100 slides were analyzed for each *Myrsine* species. Chromosome morphometry was characterized and the class was determined as proposed by [@B27] and reviewed by [@B21]. Using chromosome morphometric data (total, short and long arm length), the standardized Euclidean Distance and Unweighted Pair-Group Method Average (UPGMA) was applied to each species. In addition, the value of the relative size (% size in relation to sum of the mean values of total length, Table [1](#T1){ref-type="table"}) of each chromosome was compared among species by the Scott-Knott test at 5% probability. Analyses were made using the software R 3.2.4 ([@B39]). ###### Morphometry and chromosome class performed at least 10 prometaphases/metaphases. In all species were found chromosomes morphologically indentical, similar and distinct. ---------------------- -------------------- --------------------- ------- ------ ------- ------------------- ------------- ------- ------- ------ ------- ------------------- ------------- ------- ------- ------ ------- ------------------- *Myrsine parvifolia* *Myrsine coriacea* *Myrsine umbellata* Chrom. Total ± SD Short Long r Class Relative size (%) Total ± SD Short Long r Class Relative size (%) Total ± SD Short Long r Class Relative size (%) 1 2.64 ± 0.29 1.01 1.63 1.61 SM 5.60 2.79 ± 0.09 1.24 1.55 1.25 M 6.14 2.72 ± 0.06 1.14 1.59 1.39 M 6.60 2 2.47 ± 0.23 1.09 1.37 1.25 M 5.24 2.45 ± 0.11 1.02 1.42 1.38 M 5.38 2.67 ± 0.06 1.14 1.54 1.35 M 6.48 3 2.45 ± 0.22 0.86 1.59 1.85 SM 5.19 2.35 ± 0.10 1.09 1.26 1.15 M 5.17 2.13 ± 0.16 0.94 1.19 1.26 M 5.16 4 2.44 ± 0.27 0.68 1.75 2.55 SM 5.17 2.30 ± 0.05 1.02 1.27 1.24 M 5.06 2.13 ± 0.08 0.84 1.29 1.53 SM 5.16 5 2.24 ± 0.18 0.71 1.53 2.13 SM 4.76 2.29 ± 0.08 0.99 1.30 1.30 M 5.04 2.08 ± 0.14 0.74 1.34 1.80 SM 5.04 6 2.21 ± 0.17 0.73 1.48 2.00 SM 4.69 2.22 ± 0.17 0.86 1.36 1.57 SM 4.88 1.88 ± 0.11 0.64 1.24 1.92 SM 4.56 7 2.18 ± 0.25 0.81 1.37 1.68 SM 4.62 2.17 ± 0.12 0.78 1.39 1.77 SM 4.77 1.83 ± 0.11 0.79 1.04 1.31 M 4.44 8 2.16 ± 0.27 0.61 1.55 2.51 SM 4.59 2.12 ± 0.11 0.78 1.34 1.71 SM 4.67 1.83 ± 0.09 0.59 1.24 2.08 SM 4.44 9 2.15 ± 0.29 0.86 1.29 1.49 M 4.55 2.04 ± 0.15 0.81 1.23 1.50 SM 4.49 1.83 ± 0.09 0.59 1.24 2.08 SM 4.44 10 2.13 ± 0.25 0.61 1.51 2.45 SM 4.51 2.00 ± 0.10 0.78 1.23 1.56 SM 4.41 1.78 ± 0.12 0.59 1.19 2.00 SM 4.32 11 2.09 ± 0.22 0.79 1.31 1.65 SM 4.44 2.00 ± 0.17 0.75 1.26 1.67 SM 4.41 1.68 ± 0.13 0.59 1.09 1.83 SM 4.08 12 1.99 ± 0.16 0.75 1.23 1.63 SM 4.22 1.89 ± 0.10 0.71 1.18 1.64 SM 4.16 1.68 ± 0.08 0.59 1.09 1.83 SM 4.08 13 1.97 ± 0.23 0.66 1.31 1.96 SM 4.19 1.89 ± 0.07 0.57 1.32 2.31 SM 4.16 1.68 ± 0.10 0.49 1.19 2.40 SM 4.08 14 1.95 ± 0.14 0.65 1.30 2.00 SM 4.14 1.84 ± 0.06 0.55 1.29 2.32 SM 4.06 1.68 ± 0.14 0.66 1.02 1.52 SM 4.08 15 1.93 ± 0.16 0.72 1.21 1.67 SM 4.11 1.81 ± 0.11 0.65 1.16 1.78 SM 3.98 1.58 ± 0.06 0.64 0.94 1.46 M 3.84 16 1.85 ± 0.13 0.65 1.20 1.82 SM 3.93 1.81 ± 0.08 0.57 1.24 2.17 SM 3.98 1.58 ± 0.06 0.69 0.89 1.29 M 3.84 17 1.85 ± 0.23 0.72 1.13 1.56 SM 3.93 1.78 ± 0.04 0.66 1.11 1.66 SM 3.91 1.58 ± 0.09 0.69 0.89 1.29 M 3.84 18 1.84 ± 0.19 0.70 1.15 1.63 SM 3.92 1.71 ± 0.13 0.65 1.06 1.63 SM 3.77 1.58 ± 0.11 0.59 0.99 1.67 SM 3.84 19 1.82 ± 0.22 0.63 1.20 1.89 SM 3.87 1.68 ± 0.11 0.55 1.13 2.03 SM 3.70 1.58 ± 0.08 0.59 0.99 1.67 SM 3.84 20 1.75 ± 0.18 0.68 1.06 1.55 SM 3.71 1.67 ± 0.09 0.58 1.09 1.85 SM 3.69 1.58 ± 0.13 0.49 1.09 2.20 SM 3.84 21 1.68 ± 0.14 0.79 0.89 1.13 M 3.56 1.55 ± 0.16 0.49 1.06 2.17 SM 3.42 1.43 ± 0.14 0.59 0.84 1.42 M 3.48 22 1.66 ± 0.16 0.58 1.08 1.85 SM 3.53 1.55 ± 0.04 0.35 1.20 3.33 A 3.42 1.38 ± 0.11 0.49 0.89 1.80 SM 3.36 23 1.66 ± 0.30 0.58 1.08 1.86 SM 3.53 1.52 ± 0.07 0.39 1.13 2.88 SM 3.34 1.28 ± 0.10 0.59 0.69 1.17 M 3.13 Sum 47.22 16.99 30.23 100.00 45.53 16.96 28.57 100.00 41.30 15.79 25.51 100.00 ---------------------- -------------------- --------------------- ------- ------ ------- ------------------- ------------- ------- ------- ------ ------- ------------------- ------------- ------- ------- ------ ------- ------------------- Chrom = chromosomes; Total = total length; SD = standard deviation; Long/Short = arm length; r = arm ratio -- long/short; M = metacentric; SM = submetacentric; A = acrocentric; Relative size = % size in relation to sum of the mean values of total length; Sum = sum of the mean values. Results {#SECID0EXHBG} ======= In vitro plantlet recovering {#SECID0E2HBG} ---------------------------- Approximately 60 days after in vitro inoculation, plantlets were obtained for the three *Myrsine* species. All plantlets exhibited sufficient and morphologically normal leaves and roots for FCM and cytogenetic analyses, respectively. Nuclear 2C value measurement {#SECID0EMIBG} ---------------------------- FCM analysis performed on leaves collected in the field did not result in histograms showing profile G~0~/G~1~ peaks. So, dithiothreitol antioxidant was replaced by PEG in the nuclei isolation buffer OTTO I. This change provided G~0~/G~1~ peaks, exhibiting a coefficient of variation (CV) less than 5% for *Myrsine umbellata* and the two internal standards. The channel of the *Pisum sativum* G~0~/G~1~ peak however was closer to *Myrsine umbellata* than *Solanum lycopersicum* Thus, based on linearity international criteria for FCM, *Pisum sativum* was the standard chosen for the next measurements. The mean 2C value of the male (2C = 6.65 pg ± 0.02) and female (2C = 6.67 pg ± 0.11) *Myrsine umbellata* individuals were statistically identical by the *F* test. Considering these previous results, the 2C value was measured from leaves of in vitro plantlets. The mean values were 2C = 4.81 pg ± 0.05 for *Myrsine parvifolia*, 2C = 6.60 pg ± 0.14 for *Myrsine coriacea* and 2C = 6.63 pg ± 0.13 for *Myrsine umbellata*. The mean value of the *Myrsine umbellata* in vitro plantlets was statistically identical to the males and females in the field. Therefore, the mean value adopted for this species was 2C = 6.65 pg, which was statistically equal to the *Myrsine coriacea*. Cytogenetic analysis {#SECID0EJNBG} -------------------- Roots exposed to a 12 h APM provided prometaphases, exhibiting chromosomes at a distinct chromatin compact level, and metaphases. Enzymatic maceration in 1:5 pectinase solution ensured the chromosomes remained inside the cell, allowing an accurate determination of 2n = 45 or 2n = 46. Chromosome number of 2n = 45 was found for 12.60% individuals of *Myrsine parvifolia* and 8.45% of *Myrsine coriacea*, with 2n = 46 for the three species. Based on these results, the next slides were made from roots of particular seedlings with 2n = 45 or 2n = 46. Root maceration with 1:20 enzymatic pool for 1h 30 min supplied chromosomes no damage to the chromatin structure, without overlapping, with well-defined centromeres and free of cytoplasm debris (Fig. [1](#F1){ref-type="fig"}). {#F1} Karyotype characterization was possible only after carefully testing the time and concentration of the APM antitubulin and cell wall enzymes. *Myrsine parvifolia* presented a greater total sum of the length of the chromosomes despite having less nuclear DNA content. For this species only, we found prometaphase chromosomes showing low level of chromatin compaction (Fig. [2a](#F2){ref-type="fig"}), resulting in a higher sum of the total length (Table [1](#T1){ref-type="table"}). *Myrsine coriacea* and *Myrsine umbellata* did not show pronounced variation in chromatin compaction, but the quality of the chromosomes allowed us to characterize the karyotype and to assemble the karyogram (Fig. [2b--c](#F2){ref-type="fig"}, Table [1](#T1){ref-type="table"}). {#F2} Morphometric analysis was used to classify the chromosomes and evidence similarities and differences among species karyotypes. *Myrsine parvifolia* presented three metacentric (2, 9 and 21) and 20 submetacentric (1, 3--8, 10--20, 22 and 23) chromosome pairs, *Myrsine coriacea* showed five metacentric (1--5), 17 submetacentric (6--21 and 23) and one acrocentric (22) chromosome pairs, and *Myrsine umbellata* displayed nine metacentric (1--3, 7, 15--17, 21 and 23) and 14 submetacentric (4--6, 8--14, 17, 18, 20 and 22) chromosome pairs (Fig. [2](#F2){ref-type="fig"}, Table [1](#T1){ref-type="table"}). Morphologically similar and identical chromosomes groups were found in all species. *Myrsine parvifolia* presented sets of two chromosome pairs (5--6, 13--14, 16--17 and 22--23), as did *Myrsine coriacea* (4--5, 10--11, 13--14, 15--16 and 19--20), and *Myrsine umbellata* presented three sets of two (11--12, 16--17 and 18--19) and one set of three chromosome pairs (8--10). The other chromosome pairs in each species were considered morphologically distinct (Fig. [2](#F2){ref-type="fig"}, Table [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}). Using morphometric data and applying the UPGMA statistical analysis, the chromosomes of each *Myrsine* species were grouped in three clusters in all species (Fig. [3a--c](#F3){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Chromosome groups formed by qualitative analysis of all species were clustered by UPGMA, supporting previous findings. {ref-type="table"}) of each chromosome of *Myrsine coriacea* and *Myrsine umbellata*. The chromosomes 1, 2, 6, 7, 11, 14, 19 and 23 (\*) between the species are statistically different in relation to mean relative size according to Scott Knott test at 5% of probability.](comparative_cytogenetics-11-163-g003){#F3} As the mean 2C values of *Myrsine coriacea* (6.60 pg) and *Myrsine umbellata* (6.65 pg) were statistically identical, the Scott-Knott test was used to compare the relative size (Table [1](#T1){ref-type="table"}) of each chromosome of these species. Chromosomes 1, 2, 6, 7, 11, 14, 19 and 23 differed between the species, while the others were statistically identical (Fig. [3d](#F3){ref-type="fig"}, Table [2](#T2){ref-type="table"}). ###### Chromosome groups of the *Myrsine* karyotype suggested from karyogram evaluation (Fig. [2](#F2){ref-type="fig"} and Table [1](#T1){ref-type="table"}) and confirmed by UPGMA clustering (Fig. [3a--c](#F3){ref-type="fig"}). ---------------------- ----------------------------------------- -------------------------------- ----------------------------------------- Species \*Karyogram evaluation \*\*UPGMA clustering \*\*\*Confirmed chromosome groups *Myrsine parvifolia* 5--6; 13--14; 16--17; and 22--23 1 and 2; 3--11; and 12--23 5--6; 13--14; 16--17; and 22--23 *Myrsine coriacea* 4--5; 9--10; 13--14; 15--16; and 19--20 1; 2--11; and 12--23 4--5; 9--10; 13--14; 15--16; and 19--20 *Myrsine umbellata* 8--10; 11--12; 16--17; and 18--19 1 and 2; 3--5, 7; and 6, 8--23 8--10; 11--12; 16--17; and 18--19 ---------------------- ----------------------------------------- -------------------------------- ----------------------------------------- \* Chromosome groups morphologically identical or similar defined from all morphometric data (total length, short and long arms, r = ratio long/short arm, chromosomal class; relative size) and observation of the karyogram. \*\* Chromosome groups formed by UPGMA clustering method using data about total, short and long arms length. \*\*\* Common chromosome groups evidenced by two analyses (qualitative *x* quantitative). Discussion {#SECID0EWCAI} ========== The first step in FCM was to define the best antioxidant and internal standard. The presence of secondary metabolites in the *Myrsine* leaves, such as tannins, saponins, flavonoids and steroids ([@B1]) made this challenging. These compounds probably prevented us from measuring the 2C value in individuals from the field when the OTTO I buffer ([@B35]) was supplemented with dithiothreitol. Cytosolic compounds can reduce or inhibit the interaction of the fluorochromes and DNA during the nuclei staining step ([@B32]). Antioxidants inhibit this interference, preserving the chromatin structure ([@B41]). Nevertheless, the dithiothreitol was not efficient at providing nuclei suspensions suitable for FCM. Thus, this compound, which is more specific for molecules that possess free sulfhydryl groups, was replaced by PEG because of its wide spectrum for antioxidant activities, an effect called PEGylation ([@B44]). Due to this effect, PEG was more efficient at inhibiting the action of cytosolic compounds, resulting in G~0~/G~1~ peaks for *Myrsine umbellata* and *Pisum sativum* with CV below 5%. Owing to the linearity parameter, *Pisum sativum* was a more adequate standard relative to *Solanum lycopersicum*, which reduced measurement errors. Secondary metabolite interference was completely resolved for other *Myrsine* species using in vitro plantlets propagated in a controlled environment. FCM measurements from leaves collected in the field may have been influenced by environmental conditions. Secondary metabolite production is influenced by humidity, temperature, light intensity and the availability of water and nutrients ([@B4]). Thus, the conditions at each elevation gradient can be associated with the FCM result, suggesting a differentiated production of secondary metabolic compounds for *Myrsine* at distinct altitudes. Genome size in *Myrsine* had only been reported for *Myrsine africana* as 2C = 2.46 pg ([@B22]), which was measured by Feulgen microdensitometry using *Vigna* sp. as standard. Levels of endoreduplication in cells of *Vigna radiata*, varying from 2C to 64C, were reported by [@B36]. Thus, the differences, which were about 200% between the values found for *Myrsine* species in this study and the value observed for *Myrsine africana*, can be related to the C value of *Vigna radiata* used as reference. Values close to *Myrsine umbellata* and *Myrsine coriacea* species were reported for *Cyclamen purpurascens* Mill. (2C = 6.60 pg) and *Dodecatheon meadia* L. (2C = 5.58 pg). Higher DNA contents were described for *Cyclamen coum* Mill. (2C = 13.56 pg), *Soldanella pusilla* Baumg. (2C = 12.36 pg), and lower values for *Soldanella hungarica* Simonk (2C = 3.16 pg) and *Primula vulgaris* Huds (2C = 0.47 pg) (Bennett and Leitch 2012). The interspecific variation for the 2C DNA value found in this study, as for other species of Primulaceae (Bennett and Leitch 2012), suggests the occurrence of karyotype changes. As well as for FCM, karyotype data about *Myrsine* species in the literature are very limited, with only the chromosome number reported ([@B10], [@B17], [@B18], [@B29], [@B30], [@B40]). In vitro *Myrsine* plantlets were fundamental for providing sufficient quantities of roots for the cytogenetic study independent of the reproductive period. Meticulous standardization of the antimitotic agent and enzymatic maceration were also essential for accurate chromosomal characterization. Chromosome number 2n = 46 ([@B10], [@B17], [@B18], [@B29], [@B40], present study) and 2n = 48 ([@B30]) had been reported, but this was the first record of 2n = 45. The odd chromosome number 2n = 45 was well-marked by absence of the homologue pair of the chromosome 23 (Fig. [2](#F2){ref-type="fig"}). So, other cytogenetic approaches should be performed from *Myrsine* individuals separately to know the cause of this aneuploidy. Some chromosome groups determined by statistical analysis are morphologically distinct, such as chromosomes 22 and 23 of *Myrsine coriacea*. Although clustered (Fig. [3b](#F3){ref-type="fig"}), these chromosomes are cytogenetically distinct, with 22 being acrocentric and 23 submetacentric (Fig. [2b](#F2){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Likewise, distinct chromosomes clustered in *Myrsine parvifolia* (Fig. [3a](#F3){ref-type="fig"}, Table [2](#T2){ref-type="table"}) and *Myrsine umbellata* (Fig. [3c](#F3){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Chromosome 1 of *Myrsine coriacea* and 1 and 2 of *Myrsine parvifolia* and *Myrsine umbellata* presented the highest contrast, considering the morphology and Euclidean distances (Fig. [2](#F2){ref-type="fig"}, Fig. [3](#F3){ref-type="fig"}). Similarities and differences regarding relative size (% size in relation to sum of the mean values of total length, Table [1](#T1){ref-type="table"}) were shown between *Myrsine coriacea* and *Myrsine umbellata* through the Scott-Knott test. The similarities, which were shown for some chromosomes, imply that these species could have originated from a common ancestor. The distinct chromosomes are likely to be attributed to karyotype changes that happened throughout their evolution, altering the chromosome relative size and contributing to taxa diversification. Comparative investigations of the karyotypes of related species have usually been applied to infer the evolutionary role of karyotypic modifications in different taxa and to describe the pattern and directions of chromosomal evolution within a group ([@B43], [@B42], [@B5]). Based on the constant chromosome number displayed by *Myrsine* species, interspecific variation of the nuclear 2C value between *Myrsine parvifolia* compared to *Myrsine coriacea* and *Myrsine umbellata* was also caused by karyotype alterations. The changes to the nuclear DNA content have also been attributed to structural rearrangements and/or heterochromatin polymorphisms ([@B37], [@B5]). In conclusion, the first karyotype description and data about nuclear 2C value were shown for three *Myrsine* species. Besides of the comparison between them, these data represent the basis to understand karyotype evolution in *Myrsine*. Author contribution statement {#SECID0EHVAI} ----------------------------- The authors Carvalho RF, Amaral-Silva PM, Spadeto MS and Clarindo WR conceived, designed and conducted the tissue culture, flow cytometry and cytogenetic approaches. Carvalho CR contributed the flow cytometry analysis. Amaral-Silva PP and Carrijo TT collected and identified the *Myrsine* species. Nunes ACP did the statistical analysis. All authors contributed equally to manuscript editing and revision and approved the final manuscript for submission. Conflict of interest {#SECID0EUVAI} -------------------- The authors declare they have no conflict of interest. We thank the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, Brasília -- DF, Brazil), the Fundação de Amparo à Pesquisa do Espírito Santo (FAPES, Vitória -- ES, Brazil, grant: 61860808/2013) and the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES, Brasília -- DF, Brazil) for financial support. [^1]: Academic editor: M. Guerra

Introduction {#s1} ============ Delaying older adults' transition from living in the community to institutionalization is of major public health importance. It is also important for older individuals themselves, most of whom would prefer to remain living in the community [@pone.0046061-Luppa1]. A number of studies, conducted across a range of settings (eg. population-based and dementia-based samples) have investigated risk factors for institutionalization in older adults [@pone.0046061-Luppa1], [@pone.0046061-Gaugler1]. Dementia, disability in Activities of Daily Living (ADL) and Instrumental Activities of Daily Living (IADL) are the most consistent risk factors for admission to a residential aged care facility (RACF) [@pone.0046061-Luppa1], [@pone.0046061-Gaugler1]. However, other studies have identified sociodemographic and socioeconomic factors as additional important predictors of institutionalization in older adults [@pone.0046061-Heyman1], [@pone.0046061-Lieberman1]. While most studies that have investigated predictors of institutionalization have looked at severe cognitive impairment, the contribution of mild cognitive impairment (MCI) to institutionalization in older adults is not clear. One study has looked at the association of cognitive impairment not including dementia with adverse outcomes including institutionalization in older adults [@pone.0046061-Tuokko1]. However, the diagnosis was based on clinical judgment rather than the use of specific diagnostic criteria. More research is also needed to understand factors contributing to institutionalization in an ethnically diverse population. There is growing ethnic diversity in older populations in many western countries including Australia, Canada, and the USA [@pone.0046061-Gibson1], [@pone.0046061-Turcotte1], [@pone.0046061-He1]. Cultural differences in values and expectations of family support as well as the availability of culturally appropriate residential aged care services could all contribute to different rates of institutionalization for minority elders. To our knowledge, no study has been conducted to investigate MCI as a risk factor of institutionalization in older adults. The objective of this study was to investigate a range of risk factors including demographics, socioeconomic status, health risk factors, health conditions including MCI, physical performance, medication use and service use as predictors of institutionalization in an ethnically diverse community-based cohort of older men, enrolled in the Concord Health and Ageing in Men Project (CHAMP). Investigating predictors of institutionalization in the CHAMP cohort represents a unique opportunity due to the availability of data including a range of clinical assessments, cognitive assessments, physical performance measures and use of community-based home care services. Methods {#s2} ======= Study Population {#s2a} ---------------- Participants were community-dwelling older men, participating in the CHAMP study, an ongoing cohort study in Sydney, Australia [@pone.0046061-Cumming1]. Eligible participants were aged ≥70 years at baseline and living in a specific study area (the Local Government Areas of Burwood, Canada Bay and Strathfield) near Concord Hospital. The only exclusion criterion was living in a RACF. The Electoral Roll was chosen as the sampling frame for the study. Registration on the Electoral Roll is compulsory and regularly updated, making it a suitable population-wide sampling frame. Invitation letters were sent to 3627 men and contact was made with 3005. Most of the 622 men who were not contacted did not have a listed telephone number. One hundred and ninety of the contacted men were not eligible for the study because they had moved out of the study area, moved into a nursing home, or had died. Of the 2815 eligible men contacted, 1511 (53.7%) participated in the study. An additional 194 (11.4%) men living in the study area heard about the study from friends or the local media and were recruited before receiving an invitation letter, giving a final sample of 1705 participants. All participants gave written informed consent. The study was approved by the Human Research Ethics Committee Concord RG Hospital. Data Collection {#s2b} --------------- Participants underwent baseline assessments that comprised self-completed study questionnaires and a clinical assessment that consisted of physical performance measures, neuropsychological testing and medication inventory. Following the initial baseline assessment, the men were contacted regularly at 4-monthly intervals to enable updating of data on institutionalization. Data collected during baseline assessments (2005--7), including self-reported questionnaire data and clinical information, were used in the current analysis along with longitudinal data on institutionalization. Ascertainment of Predictor Variables {#s2c} ------------------------------------ The main groups of predictor variables included demographic factors, socioeconomic status, health risk factors, health conditions, physical performance measures, medication use and service use. These predictors have been identified based on the clinical significance, and based on previous studies investigating risk factors for institutionalization [@pone.0046061-Luppa1], [@pone.0046061-Gaugler1]. ### Sociodemographic factors {#s2c1} Sociodemographic variables included age, marital status (married versus other) and living arrangements (live alone versus live with others). Social support was measured using the shortened Duke Social Support Index (DSSI) which measures both social support satisfaction and social interactions [@pone.0046061-Koenig1]. The first item in the DSSI was modified in the CHAMP which allowed the creation of two separate variables for the number of family and non-family supports. These variables were entered into models separately to the score for social interactions and subjective support. The men were also asked their country of birth which enabled grouping of the men into the categories of Australian-born, overseas-born from an English-speaking country (ESB), and overseas-born from a non-ESB. The CHAMP study area has a high proportion of immigrants and as a result, only 49.8% of men in the CHAMP study were born in Australia and 19.6% were born in Italy. ### Socioeconomic status {#s2c2} Socioeconomic status was measured using four separate variables: age at leaving school, main lifetime occupation (managers and professionals versus other), source of income (government pension only versus other) and house ownership. ### Health risk factors {#s2c3} Physical activity was assessed using the Physical Activity Scale for the Elderly (PASE) [@pone.0046061-Washburn1]. Participants were asked about whether they had ever consumed alcohol and whether they had consumed at least 12 alcoholic drinks in the past 12 months. This enabled categorization of current non-drinkers into lifelong abstainers and ex-drinkers. For those who consumed at least 12 drinks in the past year, the frequency and quantity of alcohol consumption was assessed, enabling categorization of drinkers as either safe drinkers (1--21 drinks per week) or harmful drinkers (\>21 drinks per week) [@pone.0046061-Australian1]. Smoking status (never smoker, ex-smoker, current smoker) was also assessed. ### Health conditions {#s2c4} Data on medical conditions were obtained from the self-reported questionnaires in which participants reported if they had any of the following diseases: diabetes, thyroid dysfunction, osteoporosis, Paget's disease, stroke, Parkinson's disease, epilepsy, hypertension, coronary artery disease or myocardial infarction, angina, congestive heart failure, intermittent claudication, chronic obstructive lung disease, liver disease, chronic kidney (renal) disease or kidney (renal) failure, cancer (excluding non-melanoma skin cancers), or arthritis. The number of reported comorbidities was dichotomized at the upper quartile (≤4 versus\>4). Participants also self-reported the presence of shortness of breath, and a history of having fallen in the past 12 months. Data on self-rated health were obtained and dichotomized into excellent/good versus fair/poor/very poor. Corrected visual acuity was assessed using a Bailey-Lovie chart [@pone.0046061-Bailey1] and poor vision was defined as those with \<6/19 visual acuity. The presence of incontinence was defined as leaking urine at least two or three times a week. Participants were asked about the presence of chronic pain (pain in the last six months that has lasted for ≥3 months and been experienced every day). Participants were also asked how much pain interfered in their normal activities in the past four weeks as part of the Short Form 12 [@pone.0046061-Ware1]. Participants were considered to have chronic intrusive pain if they reported the presence of chronic pain and pain that interfered with normal activities moderately, quite a bit or extremely. Depressive symptoms were assessed with the 15-item Geriatric Depression Scale (≥5 indicative of depressive symptoms) [@pone.0046061-Yesavage1]. Anxiety symptoms were measured using the Goldberg Anxiety Scale [@pone.0046061-Goldberg1], with \>5 considered as presence of anxiety. ### Diagnosis of cognitive impairment {#s2c5} Participants were screened for cognitive impairment using the Mini Mental State Examination (MMSE) [@pone.0046061-Folstein1] and the Informant Questionnaire on Cognitive Decline (IQCODE) [@pone.0046061-Jorm1] during the baseline clinic assessment. In addition to the cognitive screen participants also completed other cognitive assessments including Addenbrooke's Cognitive Examination [@pone.0046061-Dudas1], Trail Making Task B [@pone.0046061-Reitan1], Weigl-Colour Form Sorting test [@pone.0046061-Byrne1] and Logical Memory Recall test [@pone.0046061-Wechsler1]. Participants with a MMSE less than or equal to 26 and/or IQCODE greater than 3.6 were invited to have detailed clinical assessments by the study geriatrician. This assessment included a review of medical comorbidities and medications, a standardized neurological assessment, a more detailed informant interview [@pone.0046061-Waite1] and the Rowland Universal Dementia Assessment Scale (RUDAS) [@pone.0046061-Rowland1]. At a weekly consensus meeting two geriatricians, a neurologist and a neuropsychologist reviewed all medical, cognitive, informant and functional data and reached a final diagnosis of cognitive status for each participant. At the end of the screening and clinical assessments, participants were categorized as having dementia (n = 93), MCI (n = 120), unknown cognitive status (n = 164) or cognitively intact (n = 1328). Participants determined to be cognitively impaired but not demented were given the diagnosis of MCI, if they met the clinical criteria described by Petersen et al 2004 [@pone.0046061-Petersen1]. Although MCI was categorized according to the sub-types defined by Petersen et al, for the purposes of analyses and given small cell sizes, participants with all sub-types of MCI were grouped together. This is consistent with subjects fulfilling the general criteria for MCI [@pone.0046061-Winblad1]. Diagnosis and classification of dementia was based on the Diagnostic and Statistical Manual of Mental Disorders (4^th^ edition) revised criteria and well recognized criteria for dementia subtypes [@pone.0046061-Roman1], [@pone.0046061-McKeith1], [@pone.0046061-McKhann1]. All sub-types of dementia were grouped together for analyses given small cell sizes. ### Physical function and performance {#s2c6} Functional status was measured with ADL and IADL scales. Disability in ADL was defined as needing help with ≥1 activities included in the modified Katz ADL scale [@pone.0046061-Katz1]. Disability in IADL was defined as needing help with ≥1 activities included in the OARS IADL scale [@pone.0046061-Fillenbaum1]. Physical performance was assessed by administering a standard performance battery that included the following tasks: (i) walking speed (m/s) over a 6-m course, adjusted for height; (ii) chair stands test-time to successfully complete five chair stands was assessed and time dichotomized at the slowest quartile; (iii) muscle (grip) strength (kg), and (iv) dynamic balance test. Muscle strength was measured using a Jamar dynamometer (Promedics, Blackburn, UK). The score was calculated as the grip strength (kg) of the dominant hand (best of two trials). Dynamic balance was assessed with a coordinated stability task [@pone.0046061-Lord1]. Scores were dichotomized at the highest (worst) quartile. Participants, who did not complete the tests due to physical inability, were included in the worst quartile for the corresponding performance measures. ### Medication assessment {#s2c7} Medication data were coded using the Iowa Drug Information Service (IDIS) drug code numbers. Polypharmacy was defined as the use of ≥5 regular prescription medicines [@pone.0046061-Gnjidic1]. Psychotropic medication use was defined as exposure to the following drug classes: anticonvulsants (IDIS code level 28120000), antidepressants and antipsychotics (IDIS code level 28160000) and anxiolytics (IDIS code level 28240000). ### Service use variables {#s2c8} Participants were asked about their use of a number of community services during the past 12 months. These services included: spending at least one day in an aged care day centre, being visited by Home Care to help with personal or household duties, using services of the Community and Aged Care Packages (CACPs), or any service to deliver or prepare meals at home. Participants were categorized as using one or more of these services in the past year versus using none of these services. Participants were also asked whether they had spent at least one night in a hostel or nursing home in the past 12 months and this was entered into models as a separate variable. Ascertainment of Outcome Variable {#s2d} --------------------------------- Institutionalization was defined as entry into a nursing home facility or hostel at any time during follow-up to 6.58 years. In Australia, there are two main forms of residential aged-care facilities: low-level care facilities (hostels) and high-level care facilities (nursing homes). Self-care retirement villages are not considered to be RACF and so moving into one of these facilities was not considered "institutionalization". Data on institutionalization were ascertained through a regular phone contact with the participants or their nominated contact person at 4-monthly intervals. While our data does not enable us to discriminate between permanent and respite institutionalization, the majority of admission to aged-care facilities in Australia are permanent [@pone.0046061-Australian2]. Statistical Analysis {#s2e} -------------------- Data are summarized as means (standard deviation) or numbers (proportions). Differences between institutionalized and non-institutionalized participants were compared using the two-sided t-test or χ^2^-test where appropriate. Initial univariate analyses of the association between the various study measures and institutionalization were conducted using Log-rank tests and examination of survival curves. Tests for linear trends were performed for continuous variables to determine the linearity of their relationship with institutionalization and, hence, whether to enter these variables into models as continuous or categorical variables. Testing for co-linearity between the variables was performed. There was no evidence of correlation between the variables. The appropriate parameterization of continuous variables as either categorical or continuous was also confirmed in the final model by using Akaike's Information Criterion (AIC). Univariate Cox regressions were conducted to determine unadjusted hazard ratios for admission to an aged care facility for the various study measures. Variables that had a *p*\<0.25 in univariate analyses were included in the multivariate model with institutionalization as the outcome. Backward stepwise elimination was used to eliminate non-significant variables from the multivariate model. Backward stepwise elimination has an advantage over other methods (eg. forward) as it allows to examine a model with all independent variables as well as the joint predictive capability of all variables. Clinically significant interactions, and interactions identified in previous studies between dementia and urinary incontinence, dementia and falls, and arthritis and pain were examined by adding the interaction terms into the main effect models one at a time and including the significant interaction terms in the final model. None of the interaction terms remained significant in the model. In the final model, the proportional hazards assumption was assessed through use of a time-dependent covariate method, analysis of Schoenfeld residuals plots and graphical methods (eg. survival plots) for each variable. Upon the examination of the results of time-dependent covariate method, Schoenfeld residuals plots and survival curves, it was identified that the MCI covariate was violating the assumptions. To address this, the step function proportional hazards or piecewise Cox model was used to test the effect of MCI on institutionalization. Data were analyzed using SAS version 9.2 (SAS Institute Inc., Cary, North Carolina). The Kaplan-Meier survival curves were generated using SPSS software version 19.0 (SPPS Inc, Chicago, Illinois). Results {#s3} ======= Descriptive characteristics are provided in [Table 1](#pone-0046061-t001){ref-type="table"}. Of 1705 men studied at baseline, a total of 125 (7.3%) were institutionalized during a mean follow-up of 4.94 (range: 0.08--6.58) years. The mean age, social support satisfaction and social interactions were significantly different between institutionalized men compared with non-institutionalized men. The proportion of men institutionalized increased up to the age of 84 years, and then slightly dropped ([Figure 1](#pone-0046061-g001){ref-type="fig"}). In relation to health conditions, there were significant differences in all factors apart from the presence of anxiety symptoms between the two groups. {#pone-0046061-g001} 10.1371/journal.pone.0046061.t001 ###### Characteristics of the study population according to institutionalization status. {#pone-0046061-t001-1} Variable Total population (n = 1705)[a](#nt102){ref-type="table-fn"} Institutionalized(n = 124, 7.3%) Not institutionalized(n = 1581, 92.7%) P values --------------------------------------------------------- ------------------------------------------------------------- ---------------------------------- ---------------------------------------- ---------- ***Socio-demographic factors*** Age, mean (SD) 76.9 (5.5) 81.4 (5.7%) 76.6 (5.3%) \<0.001 Currently married 1278 (74.9%) 73 (58.9%) 1205 (76.2%) \<0.001 Live alone 318 (18.1%) 44 (35.5%) 274 (17.5%) \<0.001 Social support satisfaction, high (DSSS score ≥19) 1294 (76.9%) 69 (57.0%) 1225 (78.5%) \<0.001 Social interactions, high (DSSS score ≥9) 1019 (61.2%) 48 (40.3%) 971 (62.8%) \<0.001 Country of birth Australia 849 (49.8%) 82 (66.1%) 767 (48.5%) ESB immigrant 105 (6.2%) 9 (9.9%) 96 (11.2%) Non-ESB immigrant 751 (44.1%) 33 (28.7%) 718 (48.4%) 0.0002 ***Socio-economic factors*** Occupation, professional 505 (29.8%) 21 (16.9%) 484 (30.8%) 0.0011 Own house outright 1494 (88.9%) 104 (86.0%) 1390 (89.2%) 0.27 Years of education, ≥7 years 1429 (84.7%) 115 (94.3%) 1314 (83.9%) 0.002 Source of income, pension only 773 (45.9%) 66 (54.1%) 707 (45.3%) 0.06 ***Health risk factors*** Physical activity, normal/high (PASE score ≥80) 1263 (74.9%) 55 (45.5%) 1208 (77.2%) \<0.001 Alcohol consumption Lifelong non-drinker 147 (8.8%) 8 (6.5%) 139 (9.0%) Ex-drinker 250 (14.9%) 30 (24.4%) 220 (14.2%) Safe drinker (1--21 drinks per week) 1151 (68.7%) 77 (62.6%) 1074 (69.2%) Harmful drinker (\>21 drinks per week) 127 (7.6%) 8 (6.5%) 119 (7.7%) 0.02 Smoking status Never smoker 629 (37.3%) 51 (41.8%) 578 (37.0%) Previous smoker 956 (56.7%) 62 (50.8%) 894 (57.2%) Current smoker 101 (6.0%) 9 (7.4%) 92 (5.9%) 0.38 ***Health conditions*** Comorbidities, ≥5 237 (14.0%) 30 (24.6%) 207 (13.2%) 0.0005 Urinary incontinence 232 (14.0%) 27 (23.3%) 205 (13.3%) 0.003 Visual acuity, low (\<6/19) 74 (4.5%) 15 (12.5%) 59 (3.9%) \<0.001 Chronic intrusive pain 223 (13.4%) 24 (20.5%) 199 (12.8%) 0.02 Self-rated health, good or excellent 1176 (69.9%) 68 (57.1%) 1108 (70.9%) 0.002 Shortness of breath 210 (13.3%) 24 (19.4%) 186 (11.8%) 0.013 Hearing loss 1027 (61.1%) 85 (70.3%) 942 (60.4%) 0.03 History of falls 322 (19.1%) 50 (41.7%) 272 (17.4%) \<0.001 Depressive symptoms 246 (14.6%) 45 (37.5%) 201 (12.9%) \<0.001 Anxiety symptoms 123 (7.4%) 10 (8.7%) 113 (7.3%) 0.57 Cognitive status Normal 1492 (87.5%) 73 (58.9%) 1419 (89.8%) Mild cognitive impairment 120 (7.0%) 14 (11.3%) 106 (6.7%) Dementia 93 (5.5%) 37 (29.8%) 56 (3.5%) \<0.001 ***Physical function and performance*** ADL disability (needing help with ≥1 task) 141 (8.3%) 40 (32.3%) 101 (6.4%) \<0.001 IADL disability (needing help with ≥1 task) 697 (41.6%) 97 (80.8%) 600 (38.6%) \<0.001 Grip strength, poor (lowest quartile and unable) 486 (28.7%) 71 (58.2%) 415 (26.4%) \<0.001 Chair stands, slow (lowest quartile and unable) 462 (27.7%) 74 (63.3%) 388 (25.0%) \<0.001 Walking speed, slow (lowest quartile and unable) 242 (14.5%) 59 (49.6%) 183 (11.8%) \<0.001 Dynamic balance test, poor (lowest quartile and unable) 478 (29.1%) 70 (60.3%) 408 (26.7%) \<0.001 ***Medication use*** Polypharmacy (≥5 medicines) 639 (37.7%) 52 (42.3%) 587 (37.3%) 0.27 Psychotropic medications 211 (12.4%) 22 (17.9%) 189 (12.0%) 0.06 ***Service use*** Stay in nursing home in past year 29 (1.7%) 9 (7.4%) 20 (1.3%) \<0.001 Use of services in past year 193 (11.3%) 49 (39.5%) 144 (9.1%) \<0.001 ADL = Activities of Daily Living [@pone.0046061-Katz1]; DSSI = Duke Social Support Index [@pone.0046061-Koenig1]; IADL = Instrumental Activities of Daily Living [@pone.0046061-Fillenbaum1]; PASE = Physical Activity Scale for the Elderly [@pone.0046061-Washburn1]. Missing data not included in percentages. The multivariate Cox proportional model showed that age, marital status, social satisfaction, social interactions, country of birth, alcohol use, cognitive status, ADL disability, IADL disability, grip strength, and service use were significant predictors of institutionalization. However, the use of a time-dependent covariate and analysis of Schoenfeld residuals demonstrated that MCI violated the proportional hazards assumption (χ^2^ = 6.44, p = 0.01). Therefore, as the effect of MCI on institutionalization was not stable over entire follow-up time, and the proportional hazards assumption was not valid, the piecewise Cox proportional models were used to test the effect of MCI on institutionalization. The follow-up period was divided at 3.4 years (1250 days) based on examination of the survival curve for MCI and institutionalization ([Figure 2](#pone-0046061-g002){ref-type="fig"}). {#pone-0046061-g002} [Table 2](#pone-0046061-t002){ref-type="table"} shows the results of the Cox proportional hazards models for the first 3.4 years of follow-up (Model 1) and beyond 3.4 years (Model 2) of follow-up. In the Cox regression model up to 3.4 years of follow-up, age, high social interactions, country of birth, dementia, ADL disability, IADL disability, grip strength and service use were significant predictors of institutionalization. Dementia (HR = 5.43, 95%CI: 3.00--9.81), ADL disability (HR = 3.22, 95%CI: 1.80--5.77), IADL disability (HR = 3.01, 95%CI: 1.32--6.86) and use of services (HR = 2.61, 95%CI: 1.46--4.66) were the most significant predictors of institutionalization. Interestingly, MCI was not associated with institutionalization during this time interval (HR = 0.72, 95%CI: 0.22--2.36). Participants were less likely to be institutionalized if they had high social interactions (HR = 0.48, 95%CI: 0.27--0.85) and were a NESB immigrant (HR = 0.31, 95%CI: 0.16--0.60). 10.1371/journal.pone.0046061.t002 ###### Final multivariate models for predictors of institutionalization up to 3.4 years and beyond 3.4 years of follow-up. {#pone-0046061-t002-2} Model 1: Predictors up to 3.4 years (n = 1658) Model 2: Predictors beyond 3.4 years (n = 1428) --------------------------------------------------- ------------------------------------------------ ------------------------------------------------- -------------------- --------- Age group (80+ vs \<80) 1.90 (1.05--3.45) 0.04 2.37 (1.37--4.08) 0.002 Marital status (currently married vs not married) NA NA 0.42 (0.24--0.72) 0.002 Social Interactions (high vs low) 0.48 (0.27--0.85) 0.01 0.47 (0.26--0.87) 0.02 ESB immigrant vs Australian born 0.91 (0.35--2.37) 0.85 0.92 (0.32--2.61) 0.87 NESB immigrant vs Australian born 0.31 (0.16--0.60) 0.0005 0.41 (0.22--0.77) 0.005 MCI vs normal 0.72 (0.22--2.36) 0.58 4.39 (2.17--8.87) \<0.001 Dementia vs normal 5.43 (3.00--9.81) \<0.001 6.05 (2.95--12.44) \<0.001 ADL disability (yes vs no) 3.22 (1.80--5.77) \<0.001 2.72 (1.36--5.46) 0.005 IADL disability (yes vs no) 3.01 (1.32--6.86) 0.009 2.71 (1.43--5.13) 0.002 Grip strength (low vs high) 2.19 (1.22--3.93) 0.008 1.95 (1.14--5.13) 0.002 Service use (yes vs no) 2.61 (1.46--4.66) 0.001 NA NA ADL = Activities of Daily Living [@pone.0046061-Katz1]; ESB = English Speaking Background; IADL = Instrumental Activities of Daily Living [@pone.0046061-Fillenbaum1]; MCI = Mild Cognitive Impairment; NESB = Non-English Speaking Background. NA, not applicable = not a significant predictor during the time period. In the Cox regression model for the period beyond 3.4 years of follow-up, age, marital status, social interactions, country of birth, cognitive status, ADL disability, IADL disability, and grip strength were statistically significant predictors of institutionalization. Dementia (HR = 6.05, 95%CI: 2.95--12.44), MCI (HR = 4.39, 95%CI: 2.17--8.87), ADL disability (HR = 2.72, 95%CI: 1.36--5.46) and IADL disability (HR = 2.71, 95%CI: 1.43--5.13) were the most significant predictor factors for institutionalization in this later time period. Married participants (HR = 0.42, 95%CI: 0.24--0.70), NESB immigrants (HR = 0.41, 95%CI: 0.22--0.77), and those with high social interactions (HR = 0.47, 95%CI: 0.26--0.87) were less likely to be institutionalized. Discussion {#s4} ========== In this prospective population-based study, we identified a number of predictors of institutionalization. The strongest predictors were dementia, MCI, ADL and IADL disability. Older adults with dementia had approximately six times the risk of institutionalization compared with those who did not have dementia. The predictive value of MCI changed with the length of follow-up. MCI was a significant predictor of institutionalization beyond 3.4 years of follow-up. In this period, participants with MCI had approximately four times the risk of institutionalization compared with those who were cognitively intact. The rate of institutionalization of 7.3% in this study is slightly lower when compared to previous studies conducted in population-based settings in Australia, Europe and USA. An Australian study of community-dwelling adults aged ≥60 years, reported an 8.7% permanent nursing home placement over 14 years of follow-up [@pone.0046061-McCallum1]. In a study of adults aged ≥75 years living in Germany, 7.8% of participants were institutionalized during a mean follow-up of 7.6 years [@pone.0046061-Luppa2]. A USA study reported an institutionalization rate of 13.6% over 12 years of follow-up [@pone.0046061-Bharucha1]. The difference in the institutionalization rate across studies may be due to the greater ethnic diversity in the CHAMP study. Moreover, older men are generally less likely to be institutionalized as their wives commonly act as their caregiver at home. The findings of this study are consistent with meta-analyses that have highlighted dementia, ADL and IADL disability as the most important predictors of institutionalization in older adults [@pone.0046061-Luppa1], [@pone.0046061-Gaugler1], [@pone.0046061-Luppa2]. However, our study is the first study to show that MCI is an important predictor of institutionalization. Individuals with MCI are at an increased risk of developing dementia [@pone.0046061-Plassman1], [@pone.0046061-Manly1]. Therefore, our finding that MCI only contributes to an increased risk of institutionalization after more than three years suggests that this increased risk is associated with a progression of the MCI to dementia with an associated increased risk of institutionalization. In the CHAMP population, of the 120 men diagnosed with MCI at baseline, 82 men were re-assessed at Year 2 follow-up. Of these, 12% (n = 10) had progressed to dementia over two years and three of these had been institutionalized. This progression rate of MCI to dementia is similar to other community-based studies [@pone.0046061-Busse1]. However, other studies have reported higher conversion rates [@pone.0046061-Fischer1]. The prevalence of MCI has been found to be higher amongst older men than women in those living in nursing and veteran care homes [@pone.0046061-Guo1]. Cognitive impairment, excluding dementia, has been shown to predict adverse outcomes including institutionalization and mortality in older adults [@pone.0046061-Tuokko1]. Therefore, delaying the onset of dementia in individuals with MCI may reduce the risk of institutionalization, which is of major clinical and public health importance [@pone.0046061-Chukwujama1]. A recent study highlighted that not all predictors of institutionalization are robust with varying follow-up periods [@pone.0046061-CohenMansfield1]. Identification of risk factors that predict institutionalization over short versus a longer period of time may inform future interventions to delay institutionalization [@pone.0046061-CohenMansfield1]. We also found a strong relationship between country of birth, and risk of institutionalization. NESB immigrant men were about 70% less likely to be institutionalized compared with both Australian-born men. Different rates of institutionalization between ethnic groups has important implications for the planning of community and RACF services for older people, particularly as the proportion of older persons from an NESB in Australia is increasing [@pone.0046061-Gibson1]. Further research is required to confirm whether this difference in rates of admission is due to different cultural values about the role of family in supporting older persons to remain in the community or whether it is due to a relative lack of culturally and linguistically appropriate residential aged care services. Consistent with previous work [@pone.0046061-Grundy1], [@pone.0046061-McCann1], we also found that being married was associated with a reduced risk of institutionalization. Poor grip strength was also an important risk factor for institutionalization. One study has reported an association of weaker grip strength with increased risk of long-term nursing home stay in the unadjusted models only [@pone.0046061-Cooper1]. Interestingly, falls were not associated with an increased risk of institutionalization in our study, which is in contrast to previous studies [@pone.0046061-Tinetti1], [@pone.0046061-Dunn1]. Urinary incontinence was also not a significant predictor of institutionalization in this population, which is consistent with one study [@pone.0046061-HolroydLeduc1] but not with another [@pone.0046061-Thom1] conducted in community-dwelling older people. It may be that factors such as the physical performance measures and urinary incontinence are not significant predictors of institutionalization in models that already include ADL and IADL disability as these composite measures of function are determined in part by physical function and continence status. The major strengths of the CHAMP study include its representative sampling from the community, detailed assessment of cognitive status, comprehensive, objective and clinically validated physical performance measures, and availability of a range of important risk factors for institutionalization. However, there are some limitations to the present study. We were unable to investigate the association of caregiver characteristics with the onset of institutionalization. Some studies have shown that in addition to participant characteristics, caregiver characteristics are important determinants of nursing home placement for persons with dementia [@pone.0046061-Lieberman1], [@pone.0046061-Yaffe1] while another study found that compared to participant characteristics, caregiver characteristics may not play an important role in predicting institutionalization [@pone.0046061-Chan1]. This was a study of community-dwelling older men living in a defined geographical location, which may limit the study's generalizabilty to men living in other areas. It should be noted that the response rate in the CHAMP study is similar to other comparable cohort studies of this type [@pone.0046061-Cumming1]. Also the findings of this study may not be applicable to older women. Moreover, the validity of self-report data in participants with cognitive impairment may be questionable. In conclusion, in this population, the strongest predictors of institutionalization were dementia, MCI, ADL and IADL disability. Older adults with dementia had approximately a six times higher risk of institutionalization compared with those who did not have dementia. The contribution of MCI to institutionalization changed with time, with MCI being a significant predictor only beyond 3.4 years of follow-up. Participants with MCI had approximately a four times higher risk of institutionalization compared with those who were cognitively intact. Our findings suggest that in addition to other risk factors, MCI should also be considered when estimating the risks of long term institutionalization in older adults. Delaying the onset of dementia in individuals with MCI may reduce the risk of institutionalization in older adults, which is of major clinical and public health importance. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DG FS RC LW FB VN DH DGL. Analyzed the data: DG FS. Contributed reagents/materials/analysis tools: DG FS RC LW FB VN DH DGL. Wrote the paper: DG FS.

Introduction {#S0001} ============ The current prevalence of myopia in India is considerably higher than that reported in previous studies.[@CIT0001] Poor awareness, social taboo, and illiteracy are the most common causes of negligence and hesitation to receive visual acuity correction.[@CIT0002] Uncorrected visual acuity leads to unsatisfactory academic performance and constrained social interaction.[@CIT0003] Spectacle correction is a viable option, but it is underused owing to low social acceptance.[@CIT0004] Laser in-situ keratomileusis (LASIK) or laser vision correction is an effective method for correcting refractive errors; however, it is not feasible for high-power corrections. In patients who require high-power corrections, the phakic intraocular lens is a viable alternative. Phakic lenses have become increasingly popular in the current scenario of refractive surgery because they induce relatively few higher-order aberrations at the cornea level and preserve the natural accommodations of patients.[@CIT0005] STAAR surgical Visian ICL™ has been extensively studied by various surgeons globally, and its efficacy has been proven effectively.[@CIT0006]--[@CIT0008] However, the implantable phakic copolymer lens (IPCL), manufactured by Care Group, Inc., has not been previously studied. This study retrospectively analyzed the safety and efficacy of the IPCL. Materials and methods {#S0002} ===================== Seventy-five eyes of 50 patients who had undergone an IPCL implantation operation from March 2015 to February 2017 were analyzed. Written informed consent was obtained from all the patients in tenets with declaration of Helsinki. The study was conducted in Ruby Eye Hospital after acquiring approval from the institutional review board. Patients aged 18--35 years with stable refraction were included in this study. In addition to high myopia (\>−8 D), rejection for LASIK owing to thin corneas and stable post-LASIK regression were other indications for implantation. Exclusion criteria were as follows: advanced keratoconus, irregular corneal topography, an anterior chamber depth of \<3.0 mm, narrow angles on gonioscopy, and endothelial guttae. Preoperative examination {#S0002-S2001} ------------------------ A detailed ophthalmic examination was performed in anterior and posterior segments through slit lamp biomicroscopy and indirect ophthalmoscopy, respectively. Visual acuity assessment was performed using Snellen charts; however, assessment results were converted to logmar units for statistical analysis. Corneal topography examination was performed using a Pentacam (software; Oculus, Wetzlar, Germany)"; the intraocular pressure was measured using noncontact tonometry and Goldman applanation tonometry, whereas the macular thickness and posterior pole status were evaluated through optical coherence tomography (Stratus OCT, software version, Carl Zeiss Meditech, Jena, Germany). The anatomy of the corneal endothelium was assessed using a slit lamp, and its functionality was determined through serial ultrasonic pachymetry (specular microscopy was not available at our center). A corneal endothelium with stable serial pachymetry and without evidence of guttae was considered healthy. Gonioscopy was performed in all the patients to assess angle anatomy. The patients with no manifestation of any abnormality were evaluated the next day for their refractive status by performing cycloplegia refraction. The white-to-white (WTW) diameter was measured manually by using digital calipers in the supine position under topical anesthesia. The WTW diameter was also measured using optical biometry (IOL Master, Zeiss Inc.); however, the manual measurement was considered the final measurement. Routine yttrium aluminum garnet peripheral iridotomy (YAG-PI) was performed at least 1 week prior to surgery. After peripheral iridotomy (PI), the patients were treated with topical steroids. The IPCL is a customized lens that is manufactured after obtaining three preoperative parameters, namely subjective refraction, anterior chamber depth, and WTW diameter. Contact lens users were asked to discontinue usage for 15 days prior to the implantation procedure. Description of lens {#S0002-S2002} ------------------- The IPCL is made of a hybrid acrylic hydrophilic material ([Figure 1](#F0001){ref-type="fig"}). It is a rectangular lens with eight holes; two in each haptic, four along the transitional zone, and two along the periphery of the optic, which determines the orientation of the lens inside the eye. The peripheral optical holes should always be directed upward inside the eye. The haptics of the lens has three curves, and the central curve is smaller in diameter than the other two curves. It has a central vault that obviates contact with the anterior capsule of the lens (Surgical procedure: [[video S1]{.ul}](http://youtu.be/nNIdhzEZ-ng) and [[video S2]{.ul}](http://youtu.be/eZJxvgIiOg8))Figure 1Implantable phakic copolymer lens over the butterfly cartridge.. Under strict aseptic preparations, an IPCL was loaded in a butterfly cartridge containing balanced salt solution and a few drops of dispersive viscoelastic. The IPCL was placed in the inner groove of the cartridge, with the vault facing upward. The orientation of the IPCL inside the cartridge was identified using the peripheral holes provided in the optic. For right-eye implantations, the peripheral optical holes were on the left of the cartridge, and for left-eye implantations, the peripheral optical holes were on the right of the cartridge. Next, the haptics of the IPCL was taped to lock in the cartridge. Proper care was taken not to damage the optic portion. Then, the wings were folded and introduced into the groove of the handle. The plunger was pushed to visualize the smooth forward movement of the lens; any restriction and folding of the haptic inside the cartridge warranted reloading of the lens. This completed the loading of the lens. Then, a 2.8-mm temporal clear corneal incision was made in the patient's eye, and two side ports were made at 6 and 9 o'clock positions diagonally opposite to each other. Intracameral dispersive viscoelastic fluid (hydroxypropyl methylcellulose) was introduced to create a space between the crystalline lens and corneal endothelium. The open end of the cartridge was introduced into the corneal incision. Then, using a slow and controlled push technique, the IPCL could unfold in the intracameral space. The unfolding of the lens occurred with the vault facing up, and correct unfolding was ensured by confirming that the peripheral optic holes were in a superior position. After complete unfolding of the lens, the leading haptic was tucked behind the iris by using a lens guide, followed by tucking of the trailing haptic. Finally, the viscoelastic fluid was washed out using a simcoe cannula, which ensured complete removal of the fluid from the intracameral space, including the inter-lens face, to prevent postoperative inflammation and intraocular pressure spikes. Postoperative assessment {#S0002-S2003} ------------------------ Visual acuity was assessed using Snellen visual acuity chart; however, it was converted to logmar units using standard conversion table for statistical assessment. Refraction was assessed using an auto refractometer to determine the amount of residual refractive error. The vault status of the IPCL was assessed through anterior segment optical coherence tomography (Cirrhus HD-OCT 5000, Zeiss Inc., Jena). ASOCT was performed at 1-month and 6-month postoperative period. Vault height (VH) was assessed in each post-op visit. VH was determined in photic and scotopic conditions. The anterior chamber reaction was assessed through slit lamp biomicroscopy by placing a 5 m×2-mm slit beam obliquely, preferably under dark light ambience. Outcome assessment {#S0002-S2004} ------------------ Scheduled postoperative visits were conducted on postoperative days (PODs) 1, 7, 30, 90, and 180 s. During each visit, the patient was assessed for lens position ([Figure 2](#F0002){ref-type="fig"}), vault status, visual acuity, contrast sensitivity, intraocular pressure, and refractive status. On POD 30, the patients were asked to grade their satisfaction with the visual outcome on a scale of 1--5 provided on the satisfaction form ([Table 1](#T0001){ref-type="table"}).Table 1Satisfaction scoresGrades of satisfactionScoreNumber of patients (50);\ n (%)Unsatisfied10Acceptable22 (4%)Satisfied35 (10%)Very satisfied420 (40%)Extremely satisfied523 (46%) Figure 2(**A, B,** and **C**) Postoperative lens position on pupillary dilatation in slit lamp. Results {#S0003} ======= In total, 75 eyes of 50 patients were included in the study. The mean age of the patients at the time of surgery was 25.36 years (\[standard deviation \[SD\]: 3.64, min: 18 years and max: 34 years). Twenty-six were male patients and rest were female patients, respectively, were included in the study ([Table 2](#T0002){ref-type="table"}).Table 2Patient demographics and visual acuity (N-75)**Age (in years)**Mean with SD25.36±3.60Median25Range18, 34**Age group (in years)**18--22.910 (20%)23--27.925 (50%)28--32.912 (24%)33--383 (6%)**Anterior chamber depth**\ **(in mm)**Mean with SD3.521±0.82Median3.545Range2.89, 4.14**White-to-white diameter**\ **(in mm)**Mean with SD11.70±0.42Median11.68Range10.7, 12.6**Pre-op intraocular pressure**\ **(in mmHg)**Mean with SD14.3±2.7Median15.2Range10, 22**Best-corrected visual acuity**\ **(pre-op) (in logmars)**Mean with SD0.38±0.26Median0.4Range (Min, Max)0, 1.2**Uncorrected post-op visual acuity (in logmars)**Mean with SD0.24±0.16Median0.2Range (Min, Max)0, 0.7**Follow-up (in years)**Mean with SD1.8±0.5Median1.8Range0.5, 2.8**Post-op spherical equivalent**Mean with SD0.65±0.28Median0.67Range (Min, Max)0, 1.5 Visual acuity {#S0003-S2001} ------------- The mean preoperative best-corrected visual acuity (BCVA) was 0.367 logmar units (SD: 0.266, SEM- 0.031, min: 1.2 and max: 0.0). The average refractive error corrected was −19.57 D spherical equivalents (min: −5 D and max: −27.25 D). The mean cylindrical error corrected was −2.86 D (min: 1.5 D and max: −5.5 D). Approximately 64.4% of the recipients (48 eyes) received a spherical IPCL, whereas the remaining received a toric IPCL. The mode of insertion in both types of IPCLs was identical. The orientation of the toric IPCL was identical to that of the spherical IPCL. The rotation of the IPCL was not required along the steep axis of the cornea because the toricity was incorporated in the lens hence it needs to be placed along 0- and 180-degree meridian only. The mean unaided postoperative visual acuity in the POD 30 was 0.225 logmar units (SD: 0.172, SEM- 0.020). The mean uncorrected visual acuity in the POD 30 of follow-up was significantly superior to the preoperative BCVA (*P*≤0.0001) ([Figure 3A](#F0003){ref-type="fig"}). 89.33% of the patients attained either same or better visual acuity in comparison to preoperative BCVA ([Figure 3A](#F0003){ref-type="fig"}). Forty-four eyes achieved greater than 0.1 logmar improvement compared to their preoperative BCVA ([Figure 3C](#F0003){ref-type="fig"}). Eight eyes had comparatively less visual outcome. Five eyes of three patients exhibited poor outcomes owing to ametropic amblyopia, dull foveal reflex, and macular scar; these outcomes had been explained to the patients prior to the procedure. However, none of the patients experienced any deterioration of vision during the study period. No loss of lines occurred during the observation period ([Table 2](#T0002){ref-type="table"}). A one-line improvement in contrast sensitivity was observed in 78.6% of the operated eyes.Figure 3(**A**) Comparative analysis between preoperative best-corrected visual acuity and postoperative uncorrected visual acuity. (**B**) Postoperative residual refractive power in diopters. (**C**) Visual Gain in postoperative period compared to preoperative best-corrected visual acuity. (**D**) Scatter plot of vault height assessed by anterior segment OCT (Zeiss Inc., Jena). Intraocular pressure {#S0003-S2002} -------------------- The mean preoperative intraocular pressure was 14.3±2.7 mmHg ([Table 1](#T0001){ref-type="table"}) and 18.3±3.5 mmHg on POD 1. The intraocular pressure remained within the normal range during all the points of follow-up. Anterior chamber cells and flare {#S0003-S2003} -------------------------------- Approximately, 86% of the eyes exhibited a clinically nonsignificant inflammatory reaction (≤2 cells) on POD 1. One patient experienced a severe inflammatory reaction with hypopyon on POD 1. However, complete remission of the inflammation was observed by POD 7, which remained stable thereafter. None of the patients required topical immunosuppressive therapy beyond 3 weeks. Postoperative refraction assessment by using the subjective auto refractometer {#S0003-S2004} ------------------------------------------------------------------------------ The mean residual refractive power was 0.65 D (0--1.5 D, SD: 0.29). In total, 30 eyes (40%) had spherical equivalents between 0 and 0.5 D, 42 eyes (56%) of eyes had residual power of 0.5--1 D, and rest three eyes had\>1D residual power ([Figure 3B](#F0003){ref-type="fig"}). However none of the patients required any form of spectacle correction. Furthermore, 86% of the patients were high to extremely satisfied with visual outcomes ([Figure 4](#F0004){ref-type="fig"}), thus obviating the need for further intervention. None of the patients required an explanation, a replacement, or a rotation of the IPCL during the follow-up period.Figure 4Satisfaction scores of the patients. Postoperative VH {#S0003-S2005} ---------------- Mean VH in ambient light condition was 296 microns SD, 43.59 microns median, 289 microns and in dark conditions were 323 microns SD, 56 microns median, 312 microns, the difference was statistically significant (*p*-value\<0.001). Mean VH at 1 month and 6 months were 286 microns±35 microns and 285 microns±38 microns, respectively. There was no significant difference in the VH at 6-month interval. Scatter plot ([Figure 3D](#F0003){ref-type="fig"}) demonstrating VH in ambient light condition at 1-month follow-up. Follow-up {#S0003-S2006} --------- The mean follow-up of the patients was 1.8 years (SD: 0.56 years, min: 0.5 and max: 2.8). The follow-up period was calculated until the end of the study period. Approximately 100% of the patients attended the follow-up on POD 7 and POD 30. However, all the patients did not attend the follow-up on PODs 90 and 180. Complications [(Table 3](#T0003){ref-type="table"}) {#S0003-S2007} --------------------------------------------------- One eye in one patient exhibited a severe inflammatory reaction with hypopyon on POD 1. Ultra-sonogram evaluation showed acoustic free vitreous cavity with normal retinochoroidal scleral thickening; hence, endophthalmitis was ruled out. After a diagnosis of toxic anterior shock syndrome (TASS), aggressive topical immunosuppression (prednisolone eye drops) was administered to the patient. Within the next 2 days, an increase was observed in the reaction with a marginal increase in hypopyon, which subsequently started resolving and completely resolved over 2 weeks. As the inflammation decreased, the patient's vision improved completely and remained stable until the last follow-up.Table 3ComplicationsComplicationsNumber of eyes (Percentage)1. Iris adhesions02. Corneal edema5 (6.66%)3. Pupillary distortion1 (1.3%)4. Intraocular lens dislocation05. Halo vision2 (2.66%)6. Angle closure glaucoma1 (1.3%)7. Cataract1 (1.3%)8. Corneal pigmentation3 (4%)9. Iris atrophy1 (1.3%) One eye in one patient had pupillary block acute congestive glaucoma on POD 1 owing to non-patent PI and significant anterior vaulting of the lens. However, the YAG-PI procedure was re-performed on the same crater as that of the previous PI. With adequate medical management, normal intraocular pressure was restored. Cataract formation is a significant postoperative complication of posterior chamber phakic intraocular lenses. We observed an anterior capsular cataract along the para-central area of crystalline lens in one eye, which was observed on dilatation of pupil 1 year postoperatively. However, it remained stable for the next 1 year and did not show progression. The patient had mild complaints of decreased vision but was not provided with any further intervention because of satisfactory binocular vision. Slit lamp examination of this patient revealed that the vault was normal, and no evidence of any contact with the anterior capsule of the lens was observed. We presumed that the cataract formation could be attributed to either manipulation during the intraocular procedure or to extremely high myopia. The patient was 22 years old and had received a −28 D IPCL intraocular lens. None of the patients had cystoid macular edema or retinal detachment during the observation period. Discussion {#S0004} ========== The posterior chamber phakic intraocular lens has become the only type of intraocular lens for the correction of many refractive errors.[@CIT0009] The reasons for its considerable success are the biocompatibility of the materials used, preservation of pupillary activity, and far from the corneal endothelium.[@CIT0010] The safety and efficacy of the Visian ICL™, with central holes, have been demonstrated in multiple centers and have been extensively reported in the literature.[@CIT0011],[@CIT0012] However, to our knowledge, the IPCL with peripheral optic holes has not been described in the literature. Similar to the ICL, the IPCL can be implanted through a 2.8-mm incision, regardless of the amount of refractive correction and without any effect on the biomechanics of the central part of the cornea. Spectacle correction of high myopia results in unsatisfactory vision correction because of higher-order aberrations.[@CIT0013] An intraocular lens at the focal point of the eye not only reduces the higher-order aberrations but also increases the field of vision.[@CIT0014] Thus, we presume that the phakic intraocular lens provides vision of a higher quality than spectacle correction (before surgery). In our study, the preoperative mean BCVA with spectacles was 0.38 logmar units, and the post-IPCL implantation mean unaided visual acuity was 0.24 logmar units. This difference was statistically significant (*P*=0.001). With the IPCL, the refractive results are predictable and stable, unlike those obtained using LASIK for high myopic correction, because the implantation of the IPCL does not involve the risk of flap-related complications and myopic regression.[@CIT0015] However, in patients older than 45 years, the risks of cataract development and refractive shift increase.[@CIT0016] Our study results showed a predictable visual outcome in 82% of the patients and 98% of the patients if we exclude eyes with refractive correction of \>−20 D. Additionally, in these eyes, uncorrected visual acuity exhibited two lines of improvement compared with the preoperative BCVA in 45% of the eyes. Our safety results were comparable to those of corneal refractive surgery in our center. Three eyes had complications, one eye with each of the following: pupillary block, cataract, and TASS . Future pupillary blocks were obviated by ensuring PI through retro-illumination. Cataracts are potential complications of phakic intraocular lenses. The reported incidence of cataracts is 1.1--5% according to a meta-analysis. However, only 0--1.8% of the cases are clinically significant and require an explanation of the phakic lens and cataract surgery.[@CIT0017] The incidence of cataracts in our study was 1.5% (one case), which was similar to the incidence reported in a previous study on ICL. However, because it was far from the pupillary axis and caused mild visual defects, no further intervention was sought. Conclusion {#S0005} ========== Thus, IPCL with peripheral optic holes is associated with highly satisfactory visual outcomes for patients with moderate to high myopia. Furthermore, it provides optimum long-term stability of vision. The follow-up period of our study was long, which enabled satisfactory assessment of postoperative stability. We believe that with foreseeable long-term results, IPCL can be considered an effective alternative to ICL in developing countries, thus adding a crucial component to the refractive surgery armamentarium. Disclosure {#S0006} ========== The authors report no conflicts of interest in this work.

Introduction {#Sec1} ============ Plants employ a wide range of induced defences in response to herbivore attack. These defences result in morphological changes and synthesis of secondary metabolites, which cause a decrease in herbivore performance (Karban and Baldwin [@CR26]; Walling [@CR54]; Howe and Jander [@CR17]; Alba et al. [@CR3]) or enhance the performance of natural enemies of the herbivores (Turlings et al. [@CR52]; Sabelis et al. [@CR43]; Rasmann et al. [@CR38]; Dicke and Baldwin [@CR16]; Abe et al. [@CR1]). The induction of these plant defences depends on the ability of the plant to identify and recognize its attackers (Baldwin and Preston [@CR7]; Walling [@CR54]; de Vos et al. [@CR15]; Wu and Baldwin [@CR56]), and varies with the herbivore species (Stout et al. [@CR49]; de Vos et al. [@CR15]; Rodriguez-Saona et al. [@CR40]) and time since attack (Kant et al. [@CR23]). Different herbivore species on the same plant can thus affect each other through the defences that they induce (Viswanathan et al. [@CR53]; Kessler and Halitschke [@CR29]; Kaplan et al. [@CR25]). When the herbivores affect each other negatively, this interaction is sometimes misleadingly referred to as "indirect competition" to distinguish it from resource competition, which is, however, also an indirect interaction. Attacks by one herbivore species may reduce or increase plant defence against other herbivore species (Karban and Carey [@CR27]; Karban and Baldwin [@CR26]; Viswanathan et al. [@CR53]; Rodriguez-Saona et al. [@CR39]; Poelman et al. [@CR36]; Bruessow et al. [@CR11]). It has been suggested that species with comparable feeding modes induce similar plant defences and they are therefore expected to have a negative effect on each other's performance (Rodriguez-Saona et al. [@CR39]; Howe and Jander [@CR17]; Soler et al. [@CR48]). Indeed, early studies of plant defences showed that two closely related mite species induced resistance with similar effects on the performance of one of these species (Karban and Carey [@CR27]). To date, studies on the effects of simultaneous plant attacks by various herbivore species have mainly focused on herbivores of different feeding guilds, which are thought to induce different defensive pathways (Rodriguez-Saona et al. [@CR39], [@CR40]; Howe and Jander [@CR17]; Soler et al. [@CR48]). We investigated the effects of simultaneous attacks of tomato plants by two herbivores with similar feeding modes, but with opposite effects on plant defence responses. The spider mite *Tetranychus urticae* is well known for inducing defences in various plant species, including tomato (Li et al. [@CR32]; Kant et al. [@CR23], [@CR24]; Ament et al. [@CR5]), although there is substantial variation in induction among strains, with some strains even suppressing plant defences (Kant et al. [@CR23], [@CR24]; Alba et al. [@CR4]). It feeds on plant tissue by piercing parenchyma cells and sucking out their contents. This feeding induces direct plant defences such as proteinase inhibitor activity within one day (Kant et al. [@CR23]). Earlier feeding by defence-inducing *T. urticae* results in lower performance of later-arriving herbivores (Karban and Carey [@CR27]; Karban et al. [@CR28]; Sarmento et al. [@CR44]). Although *T. evansi* has the same feeding mode, it performed better on plants that had previously been attacked by conspecifics (Sarmento et al. [@CR44]). This increased performance coincided with reduced levels of defence-related plant constituents such as proteinase inhibitors, which were below the levels in plants that had not been attacked (Sarmento et al. [@CR44]). These inhibitors hamper the action of digestive proteinases present in the herbivore gut (Ryan [@CR41]; Koiwa et al. [@CR30]), including those of spider mites (Li et al. [@CR32]; Kant et al. [@CR23]), and are normally induced by herbivore attacks. The low activity of proteinase inhibitors in leaves previously attacked by *T. evansi* coincided with a lack of upregulation of the proteinase inhibitor gene *WIPI*-*II*, which is dependent on the jasmonic acid pathway. *PR*-*P6*, a marker gene of the salicylic acid pathway, was also not upregulated by attacks from *T. evansi*, suggesting that the lower defence in plants that had previously been attacked by *T. evansi* was not caused by negative crosstalk between the two pathways. This was recently confirmed using several marker genes for both pathways (Alba et al. [@CR4]). The reduction of defences in tomato plants by *T. evansi* also resulted in better performance of *T. urticae* (Sarmento et al. [@CR44], [@CR45]). It is not known yet how simultaneous attacks by these two herbivores affect tomato plant defence. Here, we investigated the effect of simultaneous attacks of the same leaf by both spider mite species on locally induced plant defences. Besides *T. evansi*, there are several other examples of herbivores that interfere with plant defence responses (Musser et al. [@CR34]; Bede et al. [@CR9]; Lawrence et al. [@CR31]). However, most of these studies did not quantify the effects of defence suppression on insect performance, leaving open the possibility that defence suppression could benefit the natural enemies of the herbivores and thus the plant (Kahl et al. [@CR20]). Several recent studies have specifically shown effects of defence suppression on herbivore performance (Kant et al. [@CR24]; Sarmento et al. [@CR44], [@CR45]; Consales et al. [@CR12]; Dafoe et al. [@CR14]; Alba et al. [@CR4]). Here, we use a similar approach and quantify plant defences through herbivore performance (oviposition) and by measuring the activity levels of proteinase inhibitors in plant tissue to investigate the effect of simultaneous attack by "inducer" (i.e., *T. urticae*) and "reducer" (i.e., *T. evansi*) herbivores on plant defence. To evaluate the effects of simultaneous attacks on plant defence, it is essential to know the timing of plant responses to herbivore attacks. Whereas it is known that *T. urticae* induces direct plant defences in tomato within 1 day, there is little information on the timing of the effects of *T. evansi* on plant defences. Sarmento et al. ([@CR44]) found increased oviposition of *T. evansi* on tomato plants 7 days after attack by conspecific mites, but it is possible that a shorter period of attack results in similar downregulation of plant defences. Therefore, we compared the timing of the reduction of plant defences by *T. evansi* with the timing of induction by *T. urticae* to subsequently investigate the effects of simultaneous attacks. Materials and methods {#Sec2} ===================== Plant material {#Sec3} -------------- Tomato seeds (*Solanum lycopersicum* var Santa Clara I-5300) were sown in a commercial plant substrate (Bioplant^®^, Bioplant Misturadora Agrícola LTDA) in a polystyrene tray (8 × 16 cells), and kept inside a fine-meshed cage in a greenhouse to avoid infestation with herbivores. After 21 days, the plants were transferred to plastic pots (2 L) that contained a mixture of soil plus bovine manure (3:1) and fertilizer (4--14--8 N--P--K). Tomato plants were further grown in mite-proof screen cages in a greenhouse until they were 45 days old and had at least four completely developed leaves. Subsequently, they were used either for the experiments or for spider mite rearing. Mites {#Sec4} ----- *Tetranychus evansi* and *T. urticae* were obtained in 2002 from naturally infested tomato plants of the same variety as above in a greenhouse at the Federal University of Viçosa, Brazil (Sarmento et al. [@CR44]). Both species were cultured on detached tomato leaves, with the petiole inserted into a PVC tube containing water to maintain leaf turgor. Tubes with infested leaves were kept in PVC trays filled with detergent and water (1:25, v/v), which served to prevent the escape of mites and invasions by mites and other non-flying arthropods. The mass culture was maintained in a room at 25 ± 3 °C and 70--90 % relative humidity and 12 h of light per day. Timing of induction of direct plant defences {#Sec5} -------------------------------------------- The third leaf down of randomly selected tomato plants that were 45 days old (four fully developed leaves) was infested for 0 (no infestation, control), 1, 2, 3 or 4 days with 100 adult females of *T. urticae* or *T. evansi*, while the other leaves were kept clean. Four plants were used for each treatment, so a total of 40 plants were used for this experiment. Insect glue (Cola Entomológica; Bio-Controle, São Paulo, Brazil) was applied to the petioles of leaves on which mites were released to prevent them from moving to another leaf. Leaves of control plants from the same batch that were the same age were also treated with glue. Plants were kept inside mite-proof screen cages in a greenhouse during the experiment. After infestation for 1--4 days, 20 leaf discs (∅ 12 mm) were made per plant from all leaflets of the leaves damaged by *T. evansi* or *T. urticae* and from corresponding leaves of uninfested control plants using a cork borer (Huffaker et al. [@CR18]). The mites as well as their web and eggs were carefully removed from the discs with a fine brush under a stereoscopic microscope, taking care not to damage the leaf discs any further. Discs were subsequently kept in Petri dishes (Ø 8 cm) containing wet cotton wool. Two leaflets of the same leaf were used to assess proteinase activity (see below). We used oviposition rates of *T. evansi* and *T. urticae* as stand-in measures of herbivore performance. The oviposition rate of spider mites is closely correlated to the population growth rate (Sabelis [@CR42]; Janssen and Sabelis [@CR19]). The oviposition rates of *T. evansi* and *T. urticae* were measured on the discs. Because the oviposition rate of spider mites decreases with age (Sabelis [@CR42]), female mites of a similar age were used in the oviposition experiments. To obtain such cohorts, several adult females were allowed to lay eggs on detached tomato leaves on wet cotton wool. The adults were removed after 24 h and the eggs were reared to adulthood. One randomly selected adult female of *T. evansi* or *T. urticae* was placed on each disc 2 days after it had turned adult. The oviposition rate was measured after 4 days (28 ± 2 °C; 70 ± 10 % RH 12 h light). Oviposition rates were averaged per spider mite species and per plant. Spider mites that did not survive the entire period of oviposition were discarded (the average per plant therefore consisted of the average of up to ten mites of each species). The experiments with plants infested by *T. evansi* and *T. urticae* could not be carried out at the same time for logistical reasons. Treatments can therefore only be compared to controls within the same experiment. Simultaneous attack by *T. evansi* and *T. urticae* {#Sec6} --------------------------------------------------- A preliminary experiment was performed to investigate the effect of different numbers of mites damaging the plants on the subsequent performance of both mite species. The third leaf of randomly selected tomato plants was infested with either 100 or 200 adult females of either species for 4 days, and leaf discs were made from the infested leaves as above. Four plants were used per treatment, i.e., 16 in total. Subsequently, the oviposition rates of individual females of both species (ten females of each per plant) were assessed after 4 days as described above. To study the effects of simultaneous attack, plants were either concurrently infested with adult females of *T. evansi* and *T. urticae*, with either of the two species separately, or they were not infested. Based on the results of the preliminary experiment, we decided to infest plants with 100 adult mites of each species in the case of co-infested plants (200 mites in total), whereas plants with only one of the two species received 100 mites. The third leaf from below of randomly selected 45-day-old tomato plants was infested for 1 day as described above while the other leaves were kept clean. There were four plants per treatment; 16 plants in total. Subsequently, the damaged leaves were cleaned and leaf discs were made as above. One adult female of *T. evansi* or *T. urticae* was released per leaf disc as above and the oviposition rate was evaluated after 4 days. Proteinase inhibitor assays {#Sec7} --------------------------- The proteinase inhibitor (PI) activity was measured in the same leaves as used for the oviposition experiments. Two leaflets per infested leaf (above) and from the corresponding control leaf were frozen in liquid nitrogen and stored at −80 °C. Subsequently, each sample was ground with mortar and pestle and a crude protein extract was obtained as described by Otha et al. ([@CR35]). Essentially, the leaves were homogenized in extraction buffer (0.1 M Tris--HCl buffer, pH 8.2 and 20 mM CaCl~2~; 1:3 w/v); the homogenate was then centrifuged at 17,200×*g* for 30 min at 4 °C and the supernatant was collected, which was used to determine the protein content and all other assays. Protein concentration was determined by the method described by Bradford ([@CR10]), using a solution of 0.2 mg/ml bovine serum albumin as standard. A standard spectrophotometric assay was used to measure trypsin inhibitory activity in the supernatant. A 100-μL aliquot of trypsin (4.7 × 10^−5^ M) was mixed with 100 μL of the supernatant and 500 μL extraction buffer (0.1 M Tris--HCl buffer, pH 8.2 and 20 mM CaCl~2~). The mixture was incubated at room temperature for 5 min. Controls consisted of 600 μL extraction buffer and 100 μL trypsin (4.7 × 10^−5^ M). A 700-μL aliquot of the mixture (tests and controls) was added to 500 μL extraction buffer and 500 μL Na-benzoyl-[d]{.smallcaps},[l]{.smallcaps}-arginine-4-nitroanilide hydrochloride ([d,l]{.smallcaps}-BApNA, 1.2 mM). Trypsin activity was monitored for 150 s at intervals of 30 s at 410 nm absorbance on a spectrophotometer. The difference between the absorbances measured at 150 s and 60 s was used to determine the trypsin activity. Measurements were performed in triplicate per sample. The results obtained were converted to milligrams trypsin inhibited per gram of protein according to the following equation: mg trypsin inhibited per gram of protein = *AB*/1000*PC*, where *A* = enzyme control − absorbance at 410 nm of the extract, *B* = sample dilution, *P* = protein concentration of the extracts (g/mL), and *C* = trypsin factor, the result of the activity of 1 μg of trypsin on the substrate [d,l]{.smallcaps}-BApNA measured at 410 nm; for the combination of trypsin and [d,l]{.smallcaps}-BApNA, the result is 0.019 (Kakade et al. [@CR21]). Statistics {#Sec8} ---------- Differences in mean oviposition rates per plant among treatments were tested with a generalized linear model (GLM) with a Gaussian error distribution (R Development Core Team [@CR37]). Contrasts among treatments were assessed by aggregating non-significant treatment levels in an a posteriori stepwise procedure (Crawley [@CR13]). Differences in PI activity were analyzed with a GLM with a Gaussian error distribution. We correlated the oviposition rates of the mites with PI activity measured in the same leaf. Because we expected oviposition not to depend linearly on proteinase inhibitor activity but to follow a dose--response curve, we also fitted such a curve (a three-parameter logistic model) of the form$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{Ovip}} = a + \frac{b}{{1 + {\text{e}}^{{c - {\text{PI}}}} }},$$\end{document}$$where Ovip = oviposition rate of the spider mites, PI = the proteinase activity, and *a*, *b*, and *c* are parameters that were estimated with the nls function in R (R Development Core Team [@CR37]). Models were compared with the "anova" function in R (R Development Core Team [@CR37]) and with Akaike's information criterion (AIC), and nonsignificant parameters were removed from the model. We also used piecewise regression to identify the correlation of oviposition rate within various ranges of proteinase inhibitor activity and to assess the approximate value of the inflection point of the dose--response curve. In short, piecewise regression consists of fitting different linear regressions to various ranges of the data, choosing the ranges that result in the lowest residual standard error (Crawley [@CR13]). Results {#Sec9} ======= Timing of induction of direct plant defences {#Sec10} -------------------------------------------- The oviposition rates of *T. urticae* and *T. evansi* were significantly affected by previous attack of the plants by *T. urticae* (GLM, *T. urticae*: *F*~4,15~ = 57.4, *P* \< 0.0001; *T. evansi*: *F*~4,15~ = 25.7, *P* \< 0.0001). Oviposition on leaves of clean plants (0 days of previous infestation) was significantly higher than on leaves previously attacked by *T. urticae*, and the oviposition after 1 day of previous infestation by *T. urticae* was lower than that after several days of previous infestation (Fig. [1](#Fig1){ref-type="fig"}a). These data confirm that *T. urticae* induces direct plant defences in tomato plants within 1 day (Kant et al. [@CR23]).Fig. 1Average oviposition rates (eggs/female/4 days + SE) of *Tetranychus evansi* (*white bars*) and *T. urticae* (*gray bars*) on discs made from leaves that were previously attacked by *T. urticae* (**a**) or *T. evansi* (**b**) for 1--4 days or that were not previously attacked (0 days). Oviposition rates were averaged over a maximum of 10 adult females per plant, and each treatment was repeated on 4 plants. For each panel and each species,*bars with the same letter* are not significantly different (contrast among treatments after a GLM). For logistical reasons, the experiments corresponding to panels **a** and **b** were not carried out at the same time. Therefore, treatments can only be compared within the same experiment The oviposition rates of both species was also significantly affected by previous attacks by *T. evansi* (GLM, *T. urticae*: *F*~4,15~ = 9.5, *P* = 0.0005; *T. evansi*: *F*~4,15~ = 95.5, *P* \< 0.0001) (Fig. [1](#Fig1){ref-type="fig"}b). Oviposition on leaves that were previously attacked by *T. evansi* for 1 or 2 days was significantly higher than that on leaves of clean plants (0 days of previous infestation) (Fig. [1](#Fig1){ref-type="fig"}b). Oviposition by *T. evansi* on plants that were previously attacked by *T. evansi* for 3 and 4 days was lower than that on plants attacked for 1 or 2 days, but higher than that on clean plants (Fig. [1](#Fig1){ref-type="fig"}b). Oviposition by *T. urticae* on plants previously attacked by *T. evansi* for 3 and 4 days was not significantly higher than on plants that were not attacked before (Fig. [1](#Fig1){ref-type="fig"}b). There was a significant effect of attacks by both species on proteinase inhibitor (PI) activity in the attacked leaves (Fig. [2](#Fig2){ref-type="fig"}a, GLM, *T. urticae*: *F*~4,15~ = 8.6, *P* = 0.0008; *T. evansi*: *F*~4,15~ = 3.19, *P* = 0.044). Levels of PI activity were significantly lower in leaves of unattacked plants (0 days of previous infestation) than in leaves previously attacked by *T. urticae* for 1--4 days (Fig. [2](#Fig2){ref-type="fig"}a). In contrast, PI activity was significantly lower in leaves attacked by *T. evansi* than in clean leaves (Fig. [2](#Fig2){ref-type="fig"}b). The PI activity showed a negative relation with the oviposition rates: when activity levels were high, oviposition rates were low, and vice versa (cf. Figs [1](#Fig1){ref-type="fig"}, [2](#Fig2){ref-type="fig"}).Fig. 2Average proteinase inhibitor (PI) activity (in mg trypsin/total protein + SE) in leaves that were previously attacked by *T. urticae* (**a**) or *T. evansi* (**b**) for 1--4 days, or that were not attacked (0 days). Oviposition rates (Fig. [1](#Fig1){ref-type="fig"}) were measured on the same leaves. Within each panel,*bars with the same letter* are not significantly different (contrast among treatments after GLM). See the legend to Fig. [1](#Fig1){ref-type="fig"} for further explanation In conclusion, both the oviposition data and the PI activity levels show that the two herbivores affect plant defences within 1 day: whereas *T. urticae* upregulates defences, *T. evansi* downregulates them. We therefore decided to study the effects of simultaneous attack by both species after 1 day of infestation. Simultaneous attack by *T. evansi* and *T. urticae* {#Sec11} --------------------------------------------------- The oviposition rates of the two spider mite species did not differ significantly on leaves that were previously attacked by 100 or 200 mites of either species (Fig. [3](#Fig3){ref-type="fig"}). We therefore decided to use 100 mites of each species to infest the leaves of the plants, resulting in 200 mites on leaves that were attacked simultaneously and 100 mites on leaves that were attacked by one of the two species.Fig. 3Average oviposition rates (number of eggs per female per 4 days + SE) of *T. evansi* and *T. urticae* on leaves that had previously been infested for 4 days. **a** Previous infestation with 100 (*white bars*) or 200 *T. urticae* (*light gray bars*). There was no effect of the number of mites used for the infestation on the oviposition rate of *T. evansi* (GLM with gamma error distribution: *df* = 1.6, deviance = 0.004, *P* = 0.118) or *T. urticae* (*df* = 1.6, deviance = 0.0005, *P* = 0.769). **b** Previous infestation with 100 (*dark gray bars*) or 200 *T. evansi* (*black bars*). Again, there was no effect of the number of mites on the oviposition of *T. evansi* (*df* = 1.6, deviance = 0.0007, *P* = 0.63) or *T. urticae* (*df* = 1.6, deviance = 0.00014, *P* = 0.93) The oviposition rates of the two species were significantly affected by the plant treatments (GLM, *T. urticae*: *F*~3,12~ = 61.5, *P* \< 0.0001; *T. evansi*: *F*~3,12~ = 9.84, *P* = 0.0015) (Fig. [4](#Fig4){ref-type="fig"}). Previous infestation by *T. evansi* for 1 day resulted in higher oviposition rates than on previously uninfested plants for both species, confirming our earlier findings (Sarmento et al. [@CR44], [@CR45]). As expected, a previous infestation by *T. urticae* resulted in lower oviposition rates for both species. Simultaneous infestation resulted in intermediate oviposition rates, which were not significantly different from that on clean plants for *T. evansi*, but it was somewhat lower than that observed on clean plants for *T. urticae* (Fig. [4](#Fig4){ref-type="fig"}).Fig. 4Average oviposition rates (number of eggs per female per 4 days + SE) of *T. evansi* (*white bars*) and *T. urticae* (*gray bars*) on leaves that were previously infested for 1 day by *T. evansi* or *T. urticae*, by both, or were not infested (clean leaves). For each species,*bars with the same letters* do not differ significantly (contrast among treatments after GLM) PI activity was significantly affected by the infestation (Fig. [5](#Fig5){ref-type="fig"}, *F*~3,12~ = 3.87, *P* = 0.038). The activities in leaves previously attacked by *T. evansi* and in clean leaves were significantly lower than in leaves that were previously attacked by *T. urticae* and by both mite species (Fig. [5](#Fig5){ref-type="fig"}). Comparison of Figs. [4](#Fig4){ref-type="fig"} and [5](#Fig5){ref-type="fig"} show a less clear negative relation of oviposition to level of PI activity than above (Figs. [1](#Fig1){ref-type="fig"}, [2](#Fig2){ref-type="fig"}).Fig. 5Average proteinase inhibitor (PI) activity (in mg trypsin/total protein, + SE) in leaves previously attacked for 1 day by *T. urticae,* by *T. evansi,* by both, and in uninfested leaves (clean). Oviposition rates (Fig. 5) were measured on the same leaves.*Bars with the same letter* are not significantly different (contrast among treatments after GLM) Discussion {#Sec12} ========== Our results confirm earlier findings that *T. evansi* downregulates plant defences (Sarmento et al. [@CR44], [@CR45]; Alba et al. [@CR4]). In particular, both *T. evansi* and *T. urticae* had higher oviposition rates on leaves previously attacked by *T. evansi*. Our results also show that the *T. urticae* used here induces defences in tomato plants (Li et al. [@CR32]; Kant et al. [@CR23], [@CR24]; Ament et al. [@CR5]), and that *T. evansi* is sensitive to these defences (Sarmento et al. [@CR44]). The effects of induction as well as reduction of defences on oviposition occurred within 1 day. Indeed, the oviposition rates of the herbivore species on leaves previously attacked by *T. evansi* were highest after 1 day of previous infestation and decreased subsequently (Fig. [1](#Fig1){ref-type="fig"}), which could be due to an increase of plant defences because of a longer period of attack and consequently higher damage levels, or because of a decrease in the quality of the leaf discs due to depletion by the herbivores. The fact that the proteinase inhibitor (PI) activity did not increase with the period of infestation by *T. evansi* (Fig. [2](#Fig2){ref-type="fig"}b) seems to point to the latter explanation. Whereas the line of *T. urticae* used here induced plant defences, resulting in lower oviposition rates, *T. evansi* reduced plant defences, causing higher oviposition rates. Surprisingly, the oviposition rates of both species on leaves that were previously attacked by both species simultaneously were intermediate between the oviposition rates observed on leaves previously attacked by either of the two species separately (Fig. [4](#Fig4){ref-type="fig"}), suggesting that the effects of both species on the effective plant defences roughly cancel out. Hence, *T. evansi* can reduce plant defences to levels lower than those present in clean plants (Sarmento et al. [@CR44]) but cannot reduce defences induced by *T. urticae* to those low levels. Likewise, Alba et al. ([@CR4]) found that *T. evansi* did not suppress the accumulation of phytohormones involved in plant defence in leaves co-infested with *T. urticae*, but did suppress the expression of downstream defence marker genes. This suggests that the compounds that are possibly involved in the reduction of plant defences by *T. evansi* are not capable of completely circumventing defences, and that elicitors involved in the induction of plant defences by *T. urticae* can partially rescue the defences reduced by such compounds. Possibly, plants cannot cope with these compounds and elicitors simultaneously, but the higher activity of PI observed in leaves attacked by both mites shows that there is at least some defence response in the doubly infested leaves (Fig. [5](#Fig5){ref-type="fig"}). This is further confirmed by the oviposition rate of *T. urticae*, which was slightly, but significantly, different on leaf discs from co-infested plants than on leaf discs from clean plants. The high activity of PI in co-infested leaves (Fig. [5](#Fig5){ref-type="fig"}) and the intermediate oviposition rates on these leaves (Fig. [4](#Fig4){ref-type="fig"}) suggest that the activity of this defensive compound does not correlate well with the level of plant defences as reflected in herbivore performance. However, the PI levels were measured at the start of the oviposition tests and the activity levels in the leaf discs may have changed during the 4 days of the oviposition assay. We therefore used the oviposition data of the first day of the experiment on simultaneous attack to investigate the correlation between PI activity level and oviposition rates (oviposition data from the experiment on the timing of induction were collected once after 4 days, so they could not be used for this). As with many toxic and defensive compounds, one would expect that low and very low activity levels have no effect on performance, whereas high and very high activity levels would have the maximum effect. We therefore fitted dose--response curves as well as linear models to the data. The correlation between PI activity level and oviposition of *T. urticae* was bordering on significant (Fig. [6](#Fig6){ref-type="fig"},*F*~1,14~ = 4.0, *P* = 0.065). A piecewise regression model did not give a significantly better fit than a linear model, but a three-parameter logistic model gave a significant better fit than the linear model (Fig. [6](#Fig6){ref-type="fig"}, *F*~1,13~ = 7.44, *P* = 0.017, AIC of the linear model: 45.3, AIC of the logistic model: 40.1). Neither of the models was significant for *T. evansi*. The seven points of lowest PI activity corresponding to the plateau of high oviposition in *T. urticae* (gray points in Fig. [6](#Fig6){ref-type="fig"}) are from the 4 plants that were previously attacked by *T. evansi* and 3 of the clean plants. It is clear that an increase of proteinase inhibitor activity to levels above \~40 does not further decrease the oviposition rate of *T. urticae* (Fig. [6](#Fig6){ref-type="fig"}), and that the oviposition rate of *T. evansi* does not correlate with proteinase inhibitor activity. This absence of a linear correlation between PI activity and mite performance shows that it is potentially problematic to use PI activity to quantify the levels of plant defence experienced by the herbivores. This is hardly surprising given the many and varied changes that occur in plants upon attacks by herbivores (Baldwin and Preston [@CR7]; Baldwin et al. [@CR8]; Kant and Baldwin [@CR22]; Alba et al. [@CR4]), but the repercussion of this is that plant defences can only be assessed in a comprehensive way through measurements of herbivore performance because this integrates the impacts of all defensive actions of the plant (Kahl et al. [@CR20]; Consales et al. [@CR12]). Possibly, there is spatial variation in the concentrations of defensive compounds within single leaves (Shroff et al. [@CR47]), and the mites preferentially feed on tissues with low defence levels. In this case, the levels of PI activity measured in this study may not be representative of the tissue on which the mites were feeding. We suggest that it is informative to test other defensive traits for correlation with herbivore performance in a similar fashion.Fig. 6Relationships between proteinase inhibitor (PI) activity (data from Fig. [5](#Fig5){ref-type="fig"}) and the oviposition rates (eggs/female/day) of *T. urticae* (*circles*) and *T. evansi* (*diamonds*) on discs from the same leaf (data included in Fig. [4](#Fig4){ref-type="fig"}). Proteinase activity levels were assessed at the onset of the oviposition assay; oviposition was measured 1 day later. The seven *gray symbols* correspond to the lowest seven PI activity levels, which were from the four plants previously attacked by *T. evansi* and three of the four clean plants (Fig. [5](#Fig5){ref-type="fig"}). The fitted curve is a 3-parameter dose--response curve \[Ovip = 6.81 − 1.40/(1 + e^30.08 − *PI*^)\] to the oviposition data of *T. urticae* We show here and elsewhere (Sarmento et al. [@CR44]) that *T. urticae* can profit from the decrease of plant defences caused by *T. evansi*. Even in doubly infested leaves, *T. urticae* has a higher oviposition rate than on leaves with conspecifics only. In contrast, the performance of *T. evansi* decreases as a consequence of defences induced by *T. urticae*. This suggests that *T. urticae* should preferentially attack plants previously infested by *T. evansi*, and the latter should prefer plants attacked by conspecifics to plants attacked by *T. urticae*. This remains to be tested. Meanwhile, it is clear that the two herbivore species affect each other through induced plant responses, and this can affect the course of within-plant competition between them. However, when populations of the two species were allowed to grow on the same plants, populations of *T. urticae* showed low population growth rates and were outcompeted by *T. evansi*. In contrast, *T. evansi* was not significantly affected by the presence of *T. urticae* (Sarmento et al. [@CR45]). The profuse web produced by *T. evansi* was probably one of the causes, because it hinders *T. urticae* (Sarmento et al. [@CR45]), but reproductive interference between the two species may also have played a role (Sato et al. [@CR46]). This shows that assessment of defence-mediated indirect effects among herbivores cannot serve as a prediction for the outcome of competition, and competition experiments are essential to assess the net effect of simultaneous attacks on the population dynamics of the herbivores. In general, it is thought that chewing insects such as caterpillars induce the jasmonic acid (JA) defence pathway, whereas phloem-sucking insects such as aphids and whiteflies induce the salicylic acid (SA) pathway (Walling [@CR54], [@CR55]; de Vos et al. [@CR15]; Zarate et al. [@CR57]). However, it is known that several species of spider mites induce both pathways (Kant et al. [@CR23]; Ament et al. [@CR5]; Matsushima et al. [@CR33]), and there is accumulating evidence that *T. evansi* induces neither of the two (Sarmento et al. [@CR44], [@CR45]; Alba et al. [@CR4]). Interactions between these two pathways have often been shown (Thaler et al. [@CR50], [@CR51]; Arimura et al. [@CR6]; Bruessow et al. [@CR11]). This suggests that herbivores that induce different defensive pathways may increase each other's performance on co-attacked plants (Rodriguez-Saona et al. [@CR39], [@CR40]; Bruessow et al. [@CR11]; Soler et al. [@CR48]) because they are differentially susceptible to defences mediated by different signaling pathways that interact with each other (Thaler et al. [@CR51]). We show that the performance of *T. urticae* was improved on plants previously attacked by *T. evansi*, but the opposite was not the case. The increased performance of *T. urticae* shows that positive indirect effects through plant defences are not necessarily restricted to insects with different feeding modes that induce different defensive pathways: *T. urticae* can induce both pathways (Kant et al. [@CR23]) but *T. evansi* appears to induce neither of them (Sarmento et al. [@CR44]). The two herbivore species studied here are closely related, and both feed on the contents of leaf parenchyma cells. Yet they cause contrasting effects on plant defences and affect each other's performance on plants through induced defences. In fact, earlier studies have shown considerable variation in the induction of and sensitivity to induced plant defences within one species (Kant et al. [@CR24]). We therefore suggest that it is better to focus on the actual effects of herbivores on plant defences rather than generalizing across feeding modes (Agrawal [@CR2]). **Author contribution statement** {#d30e1620} --------------------------------- EFdO and AJ conceived and designed the experiments. EFdO performed the experiments. AJ and EFdO analyzed the data. AJ, EFdO, and AP wrote the manuscript. We dedicate this paper to our late dear friend and mentor Maurice Sabelis, who commented on an earlier version of the manuscript. The comments of the anonymous reviewers resulted in substantial improvement of the manuscript. We thank Renato Sarmento, Felipe Lemos, Martijn Egas, Merijn Kant, Dan Li, Martijn Egas, Juanma Alba, Bart Schimmel, Livia Ataide, Hans Breeuwer, Fabricio Ribeiro, Ana Bernardo, Cleide Dias, Eraldo Lima, Carlos Villarroel, Marcela Siveira, Bram Knegt, Madelaine Venzon, and two anonymous reviewers for discussions, support, and comments. Fabrício Ribeiro and Camila Rocha Silva helped with the proteinase inhibitor assays. EFdO received a scholarship from CAPES, AJ received a scholarship from FAPEMIG (CBB-30003/09), and AP was supported by CNPq. The authors declare no conflict of interest. [^1]: Communicated by Diethart Matthies.

1. Introduction {#sec1-ijerph-17-00884} =============== Adolescence is the phase of life between late childhood and adulthood \[[@B1-ijerph-17-00884]\]. It is a unique developmental stage during which an individual is constantly being shaped and influenced by their environment \[[@B2-ijerph-17-00884]\]. The effects of the environment are sometimes irreversible on mental and emotional development as well as physical maturation \[[@B3-ijerph-17-00884],[@B4-ijerph-17-00884]\]. Positive youth development can lead to a healthy and successful adulthood \[[@B5-ijerph-17-00884]\]. Providing an environment that supports positive youth development is thus beneficial for not only adolescents, but also for individuals of all ages \[[@B6-ijerph-17-00884]\]. Pet ownership may be an important environmental factor for mental well-being among adolescents as several studies have suggested an association between pet ownership and mental well-being among adolescents \[[@B7-ijerph-17-00884],[@B8-ijerph-17-00884],[@B9-ijerph-17-00884]\]. However, this association is not limited to adolescence. Companionship with pets may be important for positive mental health and well-being \[[@B10-ijerph-17-00884]\]. Connection with pets provides benefits to those with mental health problems by offering emotional support \[[@B11-ijerph-17-00884]\]. Moreover, pet ownership is a modifiable environmental factor \[[@B12-ijerph-17-00884],[@B13-ijerph-17-00884]\] because we can choose whether we own pets or not. However, the results of previous studies have been controversial, and a recent systematic review showed there is a shortage of high-quality and longitudinal studies that consider probable differences among different species \[[@B14-ijerph-17-00884]\]. Although dogs and cats are among the most popular companion animals in the world, they may have different effects on the mental well-being of humans, which are activated through different underlying mechanisms depending on the type of ownership. A recent study showed that the human--dog interaction through dogs' human-like gazing behavior increased human oxytocin \[[@B15-ijerph-17-00884]\], which has received increasing attention for its role in promoting positive social behavior and stress regulation, and its potential as a therapeutic intervention for addressing various aspects of psychiatric disorders \[[@B16-ijerph-17-00884],[@B17-ijerph-17-00884]\]. Conversely, a recent meta-analysis revealed that the presence of the parasite *Toxoplasma gondii* in the human body, which may be transmitted from cats to humans, is significantly associated with increased risk of traffic accidents and suicide attempts among those infected \[[@B18-ijerph-17-00884]\]. *T. gondii* alters host's behavior \[[@B19-ijerph-17-00884]\], and the oocytes seem to be a risk factor for developing schizophrenia \[[@B20-ijerph-17-00884]\]. Thus, cat ownership in childhood may be related to later schizophrenia risk \[[@B21-ijerph-17-00884]\]. On the other hand, the ALSPAC cohort study in the UK showed that cat ownership in pregnancy and childhood did not increase the risk of adolescent psychotic experiences \[[@B22-ijerph-17-00884]\]. Given these results, it is possible that the impact of dog and cat ownership on adolescents' mental well-being may be more complex than it seems. This study aimed to examine the effect of dog and cat ownership on the longitudinal trajectory of the mental well-being of adolescents using data from a population-based birth cohort study (The Tokyo Teen Cohort), while taking into account a wide range of confounding variables and considering the differences among the two analyzed species. To the best of our knowledge, this is the first study that longitudinally analyzes the effect of pet ownership, taking into account species difference while also analyzing data from a prospective and population-based birth cohort study with a large sample. 2. Materials and Methods {#sec2-ijerph-17-00884} ======================== 2.1. Data and Samples {#sec2dot1-ijerph-17-00884} --------------------- This study was part of the Tokyo TEEN Cohort (TTC) project (for the protocol, see \[[@B23-ijerph-17-00884]\]), an ongoing, prospective, and population-based birth cohort study on adolescents and their primary caregivers. Briefly, the TTC aimed to investigate the health and development of adolescents, and its details are described elsewhere \[[@B24-ijerph-17-00884],[@B25-ijerph-17-00884],[@B26-ijerph-17-00884]\]. In the first time point of the study, a sample of 3171 households with adolescents aged 10 years (i.e., born between September 2002 and August 2004) was obtained from 3 municipalities (Chofu, Mitaka, and Setagaya) in Tokyo, Japan, by random sampling from the basic resident register. When children were aged 12 years, 3007 households participated in the second time point of the study (follow-up rate: 94.8%). Trained interviewers obtained written informed consent from the adolescents' primary caregivers, asked adolescents and their caregivers to complete a set of questionnaires, conducted a semi-structured interview, and measured anthropometric data (height, weight, and grip). The study protocol of the TTC was approved by the institutional review boards from the Tokyo Metropolitan Institute of Medical Science (Approval number: 12--35), SOKENDAI (Graduate University for Advanced Studies (2012002)) and the University of Tokyo (10057). 2.2. Variables {#sec2dot2-ijerph-17-00884} -------------- Adolescents were interviewed to determine whether they have any pets in their home, "Do you have any pets?" at age 10. Their responses were coded in 2 dichotomized variables as 1) own (1) or not own (0) for a dog or 2) own (1) or not own (0) for a cat. 2.3. Outcome {#sec2dot3-ijerph-17-00884} ------------ Mental well-being was assessed at ages 10 and 12, using the self-report questionnaire which was a 5-item World Health Organization Well-Being Index (WHO5) \[[@B27-ijerph-17-00884],[@B28-ijerph-17-00884]\]. Each item assessed the degree of well-being over the past 2 weeks on a 6-point Likert-type scale ranging from 0 (at no time) to 5 (all of the time). Total scores derived from the WHO5 ranged from 0 to 25, with higher scores indicating better psychological well-being. The total raw score, ranging from 0 to 25, was multiplied by 4 to obtain the final score, with 0 representing the worst imaginable well-being and 100 representing the best imaginable well-being. The WHO5 scale was used in the original format without modifications. All existing language versions of this questionnaire are available on the website \[[@B29-ijerph-17-00884]\]. 2.4. Covariates {#sec2dot4-ijerph-17-00884} --------------- The covariates included were sex, age, parental age, parental educational level, annual household income, and the number of siblings. This information was collected at age 10. We adjusted for multiple confounders which were applied in the previous studies, including sex \[[@B30-ijerph-17-00884],[@B31-ijerph-17-00884],[@B32-ijerph-17-00884]\], age \[[@B31-ijerph-17-00884]\], parental age \[[@B22-ijerph-17-00884],[@B32-ijerph-17-00884]\], parental educational level \[[@B22-ijerph-17-00884],[@B30-ijerph-17-00884],[@B32-ijerph-17-00884]\], annual household income (in relevance to social class and work status) \[[@B22-ijerph-17-00884],[@B30-ijerph-17-00884],[@B31-ijerph-17-00884],[@B32-ijerph-17-00884]\], and number of siblings (in relevance to presence of older siblings, number of people in the household, and household crowding) \[[@B22-ijerph-17-00884],[@B31-ijerph-17-00884],[@B32-ijerph-17-00884]\]. 2.5. Statistical Analysis {#sec2dot5-ijerph-17-00884} ------------------------- Linear regression analysis was performed to estimate the associations between pet ownership at age 10 and well-being at age 12. We calculated the non-standardized *B*s (multiple regression coefficients) of dog-ownership and cat-ownership. We then adjusted for the covariates. Two-factor mixed-design analysis of variance (ANOVA) for pet-owner types (non-dog/cat owners, dog owners (owned no cats), cat owners (owned no dogs)) and 2 time points (ages 10 and 12) was performed. The group who owned both dogs and cats was excluded because their number was too low, and the sample was not representative (*n* = 9). The significance level (*α*) was set to 0.05 for a 2-sided test. All statistical analyses were conducted using IBM SPSS Statistics for Windows, version 24.0.0.1 (IBM Corp., Armonk, NY, USA). 3. Results {#sec3-ijerph-17-00884} ========== Of the 3171 initially enrolled households, 2584 (81.5%) were included in our final analytic sample. Participants who had missing data on pet ownership, well-being at age 10 and 12, sex, age (months), parental age, parental educational level, annual household income, and the number of siblings were excluded from the present analyses. There were no differences among the excluded and included subjects in terms of dog ownership (*χ*^2^ = 0.20, *p* = 0.650), well-being at age 10 (*χ*^2^ = 37.36, *p* = 0.167), well-being at age 12 (*χ*^2^ = 19.67, *p* = 0.943), sex (*χ*^2^ = 0.33, *p* = 0.566), age in months (*χ*^2^ = 15.46, *p* = 0.563), mother's age (*χ*^2^ = 38.34, *p* = 0.115), father's age (*χ*^2^ = 31.07, *p* = 0.843), mother's educational level (*χ*^2^ = 10.15, *p* = 0.071), and father's educational level (*χ*^2^ = 5.54, *p* = 0.354), although there were differences in terms of cat ownership (*χ*^2^ = 5.55, *p* = 0.018), annual household income (*χ*^2^ = 228.80, *p* \< 0.001), and number of siblings (*χ*^2^ = 16.45, *p* = 0.012). Cat ownership was higher in excluded subjects, and annual household income and number of siblings were higher in included subjects. Demographic characteristics of participants are shown in [Table 1](#ijerph-17-00884-t001){ref-type="table"}. Approximately 10% of adolescents owned dogs, and 4% owned cats. The results of linear regression analysis showed that dog ownership at age 10 predicted better well-being at age 12 compared to no dog ownership (*B* = 2.61, 95% CI: 0.17--5.05, *p* = 0.036), while cat ownership at age 10 predicted worse well-being at age 12 compared to no cat ownership (*B* = −5,65. 95% CI: −9.26--−2.03, *p* = 0.002). The effect remained significant after adjusting for covariates (dog: *B* = 2.45, 95% CI: 0.19--4.71, *p* = 0.033; cat: *B* = 6.14, 95% CI: −9.49--−2.79, *p* \< 0.001). These results are shown in [Table 2](#ijerph-17-00884-t002){ref-type="table"}. Among boys, dog ownership at age 10 predicted better well-being at age 12 compared to no dog ownership (*B* = 3.32, 95%CI: 1.00--7.86, *p* = 0.011), while cat ownership at age 10 predicted worse well−being at age 12 compared to no cat ownership (*B* = −6.55, 95%CI: −11.60--−2.45, *p* = 0.010). The effect remained significant after adjusting for covariates (dog: *B* = 4.04, 95%CI: 0.90--7.18, *p* = 0.012,; cat: *B* = −7.03, 95%CI: −11.60--−2.45, *p* = 0.003). Among girls, neither dog ownership nor cat ownership at age 10 predicted better nor worse well-being at age 10 (dog: *B* = 0.64, 95%CI: −2.84--4.12, *p* = 0.719; cat: *B* = −4.63, 95%CI: −9.89--0.62, *p* = 0.084). Adjusting for covariates did not change the results for dog ownership (*B* = 0.65, 95%CI: −2.61--3.91, *p* = 0.696) but did for cat ownership (*B* = −5.34, 95%CI: −10.27--−0.41, *p* = 0.034). Two-way mixed-design ANOVA showed significant interaction of time points and owner types (*F* (2, 2572) = 6.78, *p* = 0.001). Simple main effect of owner types was not significant at age 10 (*F* (2, 2572) = 0.18, *p* = 0.835), but it was significant at age 12 (*F* (2, 2572) = 6.61, *p* = 0.001). Bonferroni adjustments were administered for multiple comparisons and found significant pairs at age 12 as follows: cat owners (owned no dogs) and non-dog/cat owners (*p* = 0.017) and cat owners (owned no dogs) and dog owners (owned no cats) (*p* = 0.001). Other pairs were not significant. The simple main effect of time points was significant in non-dog/cat owners (*F* = 85.55, *p* \< 0.001) and cat owners (owned no dogs) (*F* = 26.21, *p* \< 0.001) but not significant in dog owners (owned no cats) (*F* = 1.38, *p* = 0.240). The related result below is shown in [Figure 1](#ijerph-17-00884-f001){ref-type="fig"}. Among boys, a significant interaction of time points and owner types was also found (*F* (2, 1357) = 6.202, *p* = 0.002). Simple main effect of owner types was not significant at age 10 (*F* (2, 1357) = 0.189, *p* = 0.828), but it was significant at age 12 (*F* (2, 1357) = 6.284, *p* = 0.002). Bonferroni adjustments were administered for multiple comparisons and found significant pairs at age 12 as follows: dog owners (owned no cats) and non-dog/cat owners (*p* = 0.020), and dog owners (owned no cats) and cat owners (owned no dogs) (*p* = 0.002). Other pairs were not significant. The simple main effect of time points was significant in no dog or cat owners (*F* = 33.08, *p* \< 0.001) and cat owners (owned no dogs) (*F* = 14.21, *p* \< 0.011) but was not significant in dog owners (owned no cats) (*F* = 0.30, *p* = 0.581). Among girls, the interaction of time points and owner types was not significant (*F* (2, 1212) = 1.810, *p* = 0.164). Simple main effects of owner types was neither significant at age 10 (*F* (2, 1212) = 0.029, *p* = 0.972) nor at age 12 (*F* (2, 1212) = 1,704, *p* = 0.182). The simple main effect of time points was significant in all groups (non-dog/cat owners: *F* = 54.63, *p* \< 0.001; cat owners: *F* = 12.04, *p* = 0.001; dog owners: *F* = 5.28, *p* = 0.022). 4. Discussion {#sec4-ijerph-17-00884} ============= This is the first study to investigate the different effects of dog ownership and cat ownership on adolescents' well-being, adjusting for various demographic and socioeconomic variables using a large-sample, longitudinal, population-based study. The prevalence of dog/cat ownership in this study was consistent with a previous large-scale Japanese study \[[@B33-ijerph-17-00884]\]. Dog ownership at age 10 was associated with increased well-being at age 12 compared to no dog ownership, and cat ownership at age 10 was associated with decreased well-being at age 12 compared to no dog ownership. These results were also the same after adjusting for covariates, including socio-demographic factors. Previous studies have shown that the mental well-being of an individual in their adolescence has a long-lasting impact on the individual's later life \[[@B2-ijerph-17-00884],[@B34-ijerph-17-00884]\], even though mental well-being was shown to generally decline throughout adolescence \[[@B30-ijerph-17-00884]\]. Previous studies have revealed that the life-long trajectory of well-being is U-shaped; it declines through the teen years to young adulthood, hits the bottom around the 40s or 50s, and increases thereafter \[[@B30-ijerph-17-00884],[@B35-ijerph-17-00884]\]. In our study, we confirmed that well-being declined from age 10 to age 12 and identified the preventive effect of dog ownership on the decline of well-being. On the other hand, we also identified that the well-being of the cat owners' group significantly declined compared with the 2 other groups (dog owner group and non-dog/cat owner group). Our results suggested that the effect of dog and cat ownership on adolescent well-being may have different underlying mechanisms. One factor may be the owner's physical activity with their pet. Dog owners often go walking with their pet \[[@B36-ijerph-17-00884]\]. Dog walking brought adolescents 7--8% more physical active minutes per day \[[@B37-ijerph-17-00884]\] and has a benefit for children's overweight or obesity \[[@B38-ijerph-17-00884]\]. However, cat owners may not go for a walk with their pets but may play with pets indoors. This could cause the difference in time length and intensity of owner's physical activity and lead to higher/lower well-being. As previous research shows, some parents own a pet because they want to teach their children responsibility and kindness \[[@B39-ijerph-17-00884]\]. Children from household with pets may learn responsibilities that benefit the development of their well-being. This point should be also considered in future studies. Deepening this discussion with a biological view, we know that oxytocin is a neuropeptide, receptors for which are distributed in one's brain. Oxytocin relates to trust in other humans \[[@B40-ijerph-17-00884]\] and modulates our sociality \[[@B41-ijerph-17-00884]\]. Further, through social interaction, oxytocin suppresses cortisol concentration, which is a response to stress \[[@B42-ijerph-17-00884]\]. Dog gaze has been shown to increase the oxytocin level in owner's urine \[[@B15-ijerph-17-00884]\], which means the dog may increase oxytocin in its owner, promote social bonding, and decrease stress levels. However, a similar testing method was not used for cats, therefore, we should be careful how we interpret the results of cat ownership in this study. Future studies should aim to investigate the effects of cats on humans based on useful methods in previous studies \[[@B43-ijerph-17-00884],[@B44-ijerph-17-00884]\]. With respect to cats, a previous study about the risk of childhood cat ownership on schizophrenia in later life, mentioned the possibility of *T. gondii* infection from cats \[[@B21-ijerph-17-00884]\]. Moreover, we know that *T. gondii* is a parasite which affects warm-blooded animals \[[@B18-ijerph-17-00884]\]. Their primary hosts are cats \[[@B22-ijerph-17-00884]\]. *T. gondii* affects the brain through inflammation or changes in the microbiome \[[@B45-ijerph-17-00884]\]. As several experiments have shown, *T. gondii* alters the behavior of rodents, making them easier prey for cats \[[@B19-ijerph-17-00884]\]. Mice infected by *T. gondii* once lose innate aversion for cat's urine permanently even after the infection has been removed \[[@B46-ijerph-17-00884]\]. In humans, *T. gondii* may be associated with an elevated risk for mental health issues, such as psychosis-like symptoms, bipolar disorder, violence, suicide attempts, anxiety disorder, and obsessive disorder \[[@B45-ijerph-17-00884]\]. In summary, a cat can be an infection source of *T. gondii* in its owner, causing brain dysfunction by inflammation or alteration in the microbiome, and leading to psychiatric symptoms. A future TTC study will attempt to answer these remaining questions of psychological, physiological, and biological mechanisms. Our results showed that not all types of pet ownership enhance adolescents' well-being, and that there can be substantial differences based on the species owned. Compared with cat ownership, dog ownership seemed to be more beneficial for maintaining well-being across adolescence. A previous experimental study reported that interaction with a therapy dog during 20 minutes improved well-being among students \[[@B47-ijerph-17-00884]\]. Considering the fact that a very small duration of interaction promoted well-being, constant interactions with a dog in one's home may provide even greater benefits. Further, interactions between adolescents and dogs are not limited to therapy and could be promoted in school and other various situations where pets are allowed. On the other hand, when living together with cats, people should be aware of the risk of *T. gondii*. Based on previous studies, we believe that in order to prevent this disease, it would be helpful for the respective owners to keep his/her cat indoors, limit its hunting, clean litter pan daily, and dispose of feces in the toilet while wearing disposal gloves \[[@B48-ijerph-17-00884]\]. A recent study attempted to develop a vaccine for *T. gondii* \[[@B49-ijerph-17-00884]\]. These challenges may lead to a better life for cats and humans. One of the strengths in our study is overcoming the methodological limitations which were suggested in the systematic review \[[@B14-ijerph-17-00884]\]. Firstly, the sample size of TTC was large (*N* = 2584), while those in the previous studies were small (*N*: 15--1541). Secondly, samples in most of the previous studies were homogeneous and self-selected; however, the TTC sample was population-based. Thirdly, the main study design of previous studies was cross-sectional (13/22), while our TTC study was longitudinal with a prospective design, which enabled us to analyze temporal direction. Fourth, TTC had several demographic and socioeconomic variables to adjust covariates. We adjusted for multiple confounders which were applied in the previous studies. Fifth, dogs and cats were separately addressed in this study to investigate the varying effects of different pet types. There are nevertheless some limitations. Firstly, TTC was not initially designed to study pet ownership, so we missed the data of pet ownership before and after age 10. Adolescents who lost their pets before age 10 or who owned pets after age 10 might be coded as non-owners in this study. We also did not have the information on the age of pets, duration of pet ownership, amount of time spent with the pet, which family member takes most care of the pet, and the strength of the attachment to the pet. Involvement with pets may be reflected in the strength of the ownership effect. These factors should be considered in future studies. Secondly, pets are not only limited to dogs and cats. Other pet species may also have a different effect. Thirdly, we adjusted for all confounders mentioned in the previous studies, however, the possibility of another confounding factor exists. Fourth, we did not examine whether the prediction of well-being from dog ownership and cat ownership is limited to adolescents in Tokyo, which is an affluent urban area in a developed Asian country. Geographical or cultural differences in ownership may exist. Fifth, we did not consider the subjective or general health of pet owners. Subjective health was controlled in the previous wellbeing study \[[@B30-ijerph-17-00884]\]. In the systematic review of pets and adolescents, the possibility exists that families with children experiencing difficulties in health or development may tend to have more or fewer pets \[[@B14-ijerph-17-00884]\]. Future studies should examine problems in health and/or developmental difficulties. Sixth, we did not differentiate whether the pet owner was the child, another family member, or the family as a whole in this study. Future studies can consider whether this difference could also have an effect. To the best of our knowledge, this was the first large-sampled, longitudinal, population-based study which has investigated the different effects of dog and cat ownership on adolescent's well-being while adjusting for covariates and analyzing the differences among these two species. When compared to the well-being of non-dog/cat owners, dog ownership predicted positive adolescents' well-being, and cat ownership predicted negative adolescents' well-being. The well-being of dog owners was maintained through the study period from 10--12 years, whereas the well-being of cat owners seemed to decline through the study period from 10--12 years. 5. Conclusions {#sec5-ijerph-17-00884} ============== Dog ownership and cat ownership differently predicted adolescents' well-being. Dog ownership had a positive effect on adolescents' well-being compared to no dogs, however, cat ownership had a negative effect compared to no cats. The well-being trajectory of dog owners was maintained through adolescence, while that of cat owners declined. We would like to sincerely thank all of the adolescents and their primary caregivers who participated in TTC. We acknowledge the work of all the interviewers who conducted the data collection. conceptualization, A.N., T.K., K.E., and S.Y.; methodology, K.E., S.A., K.M., M.N. (Miho Nagasawa), and J.I.; formal analysis, S.Y. and K.E.; data curation, S.A., K.K., M.H.-H., and I.K.; writing---original draft preparation, K.E., S.Y., A.N., S.A., and M.N. (Miharu Nakanishi); writing---review and editing, K.K., M.H.-H., T.K., K.M., M.N. (Miho Nagasawa), I.K., J.I., and S.U.; project administration, A.N.; funding acquisition, A.N., S.Y., K.K., and T.K. All authors have read and agreed to the published version of the manuscript. This work was supported by a Grant-in-Aid for Scientific Research on Innovative Areas (23118002; Adolescent Mind & Self-Regulation) from the Ministry of Education, Culture, Sports, Science, and Technology of Japan. This study was also supported by JSPS KAKENHI (grant numbers JP16H06395, 16H06398, 16H06399, 16K21720, 16K15566, 16H03745, and 17H05931). This work was supported in part by the UTokyo Center for Integrative Science of Human Behavior (CiSHuB), the International Research Center for Neurointelligence (WPI-IRCN) at the University of Tokyo Institutes for Advanced Study (UTIAS), and the MEXT\*-Supported Program for the Private University Research Branding Project (2016-2019). The authors declare no conflict of interest. {#ijerph-17-00884-f001} ijerph-17-00884-t001_Table 1 ###### Demographic characteristics of participants (N = 2584). All Dog owners Cat owners Non-dog/cat owners -------------------------------------- --------------------- ------- ------------ ------------ -------------------- ------- ------- ------- ------- Sex Female 1218 47.1% 119 46.9% 49 45.0% 1053 47.2% Male 1366 52.9% 135 53.1% 60 55.0% 1177 52.8% Age (months) 122 3.28 122 3.21 121 3.26 122 3.29 Well-being at age 10 79.06 16.61 79.42 16.83 80.04 15.65 78.98 16.63 at age 12 75.14 18.88 77.53 17.60 69.69 21.06 75.11 18.87 Parental age Mother 41.97 4.15 41.48 4.27 42.40 4.11 42.00 4.14 Father 44.12 5.17 43.40 5.12 45.26 5.00 44.16 5.19 Educational level of mother High school or less 411 15.9% 44 17.3% 15 13.8% 352 15.8% 2-year college 1150 44.5% 129 50.8% 53 48.6% 973 43.6% 4-year university 932 36.1% 71 28.0% 38 34.9% 826 37.0% Graduate university 91 3.5% 10 3.94% 3 2.8% 79 3.5% Educational level of father High school or less 470 18.2% 59 23.2% 28 25.7% 385 17.3% 2-year college 360 13.9% 33 13.0% 23 21.1% 305 13.7% 4-year university 1444 55.9% 137 53.9% 44 40.4% 1268 56.9% Graduate university 310 12.0% 25 9.8% 14 12.8% 272 12.2% Annual household income (10,000 yen) 0--299 64 2.5% 7 2.8% 4 3.7% 53 2.4% 300--599 636 24.6% 65 25.6% 39 35.8% 535 24.0% 600--999 1095 42.4% 96 37.8% 38 34.9% 964 43.2% 1000+ 789 30.5% 86 33.9% 28 25.7% 678 30.4% Number of siblings 1.16 0.79 1.12 0.78 1.07 0.84 1.16 0.78 ijerph-17-00884-t002_Table 2 ###### Multiple linear regression analysis for well-being at age 12. Unadjusted Adjusted ^1^ --------------- ------------ -------------- ---- ------- ------- ------- ------- ---- ------- ------- Dog ownership 2.61 0.17 \- 5.05 0.036 2.45 0.19 \- 4.71 0.033 Cat ownership −5.65 −9.26 \- −2.03 0.002 −6.14 −9.49 \- −2.79 0.000 ^1^ Adjusted for well-being at age 10, sex, age (months), parental age, parental educational level, annual household income, and the number of siblings.

Introduction {#s1} ============ Extensive characterization of cancer genomes has begun to change the classification of neoplasms and the choice of therapies ([@bib20]). The genetic profiles of most cancers are notoriously heterogeneous, often including thousands of mutations affecting genes with a wide range of credentials\-\--from those well-known to drive oncogenic behavior to those not known to have a role in pathogenesis. Moreover, cancers continue to accumulate mutations during carcinogenesis, producing tumor subclones with selectable features such as drug resistance or enhanced growth potential ([@bib39]). Despite this heterogeneity, consistent patterns have been observed, such as the high frequency of gain-of-function or loss-of-function mutations affecting specific proto-oncogenes or tumor suppressor genes in cancers that arise in certain cell lineages. Conversely, coincident mutations in certain genes are rare, even when those genes are frequently mutated individually in specific types of cancer ([@bib28]). Examples of these 'mutually exclusive' pairs of mutations have been reported in a variety of cancers ([@bib72]; [@bib65]; [@bib52]; [@bib57]; [@bib67]); the mutual exclusivity has usually been attributed either to a loss of a selective advantage of a mutation in one gene after a change in the other has occurred ('functional redundancy') or to the toxicity (including 'synthetic lethality') conferred by the coexistence of both mutations in the same cells. We recently reported that the mutual exclusivity of gain-of-function mutations of *EGFR* and *KRAS*, two proto-oncogenes often individually mutated in lung adenocarcinomas (LUADs), can be explained by such synthetic toxicity, despite the fact that products of these two genes operate in overlapping signaling pathways and might have been mutually exclusive because of functional redundancies ([@bib65]). Support for the idea that the mutual exclusivity of *KRAS* and *EGFR* mutations is synthetically toxic in LUAD cells was based largely on experiments in which we used doxycycline (dox) to induce expression of mutant *EGFR* or *KRAS* alleles controlled by a tetracycline (tet)-responsive regulatory apparatus in LUAD cell lines containing endogenous mutations in the other gene ([@bib65]). When we forced mutual expression of the pair of mutant proteins, the cells exhibited signs of RAS-induced toxicity, such as macropinocytosis and cell death. In addition, we observed increased phosphorylation of several proteins known to operate in the extensive signaling network downstream of RAS, implying that excessive signaling, driven by the conjunction of hyperactive EGFR and KRAS proteins, might be responsible for the observed toxicity. Recognizing that such synthetic toxicities might be exploited for therapeutic purposes, we have extended our studies of signaling via the EGFR-RAS axis, with the goal of better understanding the biochemical events that are responsible for the previously observed toxicity in LUAD cell lines. In the work reported here, we have used a variety of genetic and pharmacological approaches to seek evidence that identifies critical mediators of the previously observed toxicities. Based on several concordant findings, we argue that activation of extracellular signal-regulated kinases (ERK1 and ERK2), serine/threonine kinases in the EGFR-RAS-RAF-MEK-ERK pathway, is a critical event in the generation of toxicity, and we show that at least one feedback inhibitor of the pathway, the dual specificity phosphatase, DUSP6, is a potential target for therapeutic inhibitors that could mimic the synthetic toxicity that we previously reported. Results {#s2} ======= Synthetic lethality induced by co-expression of mutant KRAS and EGFR is mediated through increased ERK signaling {#s2-1} ---------------------------------------------------------------------------------------------------------------- In previous work, we established that mutant EGFR and mutant KRAS are not tolerated in the same cell (synthetic lethality), by placing one of these two oncogenes under the control of an inducible promoter in cell lines carrying a mutant allele of the other oncogene. These experiments provided a likely explanation for the pattern of mutual exclusivity in LUAD ([@bib65]). While we documented several changes in cellular signaling upon induction of the second oncogene to produce toxicity, we did not establish if there is a node (or nodes) in the signaling network sensed by the cell as intolerable when both oncoproteins are produced. If such a node exists, we might be able to prevent toxicity by down-modulating the levels of activity; conversely, we might be able to exploit identification of that node to compromise or kill cancer cells. To seek critical nodes in the RAS signaling pathway, we extended our previous study using the LUAD cell line we previously characterized (PC9, bearing the EGFR mutation, E746_A750del) and two additional LUAD lines, H358 and H1975. H358 cells express mutant KRAS (G12C), and H1975 cells express mutant EGFR (L858R/T790M). As in our earlier work, we introduced tet-regulated, mutant *KRAS* (G12V) into these lines to regulate mutant KRAS in an inducible manner and used the same vector encoding GFP rather than KRAS as a control. This single-vector system includes rtTA constitutively expressed from a ubiquitin promoter, allowing us to induce KRAS with the addition of dox ([@bib40]). KRAS or GFP were appropriately induced after adding dox to the growth medium used for these cell lines ([Figure 1A](#fig1){ref-type="fig"}). To establish whether induction of a mutant *KRAS* transgene is detrimental to H358 cells producing endogenous mutant KRAS or H1975 cells producing mutant EGFR proteins, we cultured cell lines in dox for 7 days and measured the relative numbers of viable cells with Alamar blue. As we previously showed, the number of viable PC9 cells is reduced by inducing mutant KRAS ([Figure 1A](#fig1){ref-type="fig"}). Similarly, when mutant KRAS was induced in either H358 or H1975 cells for seven days, we observed fewer viable cells compared to cells grown without dox or to cells in which GFP was induced ([Figure 1A](#fig1){ref-type="fig"}). These results indicate that increased activity of the RAS pathway, either in LUAD cells with an endogenous *KRAS* mutation (H358 cells) or with an endogenous *EGFR* mutation (PC9 and H1975 cells) is toxic to these cell lines. {#fig1} We previously documented increases in phosphorylated forms of the stress kinases, phospho-JNK (P-JNK) and phospho-p38 (P-p38), as well as in phospho-ERK (P-ERK or P-p44/42), in one of these cell lines (PC9) 72 hr after treatment with dox ([@bib65]; [@bib67]). We used a phospho-protein array to assess the status of protein activation more broadly after KRAS induction, using PC9-tetO-KRAS cells after 1 and 5 days of dox treatment ([Figure 1B](#fig1){ref-type="fig"}, [Figure 1---figure supplement 1A](#fig1s1){ref-type="fig"}). After 5 days, we again observed increases in P-JNK, P-p38, and P-ERK ([Figure 1---figure supplement 1A](#fig1s1){ref-type="fig"}), suggesting that three major branches of the MAPK pathway are activated after extended induction of mutant KRAS. In addition, several other proteins show enhanced phosphorylation at this time. At 24 hr after addition of dox, however, only P-ERK and P-AKT show a pronounced increase ([Figure 1B](#fig1){ref-type="fig"}). Specifically, the stress kinases, JNK and p38, were not detected as phosphorylated proteins with the protein array. A possible interpretation of these findings is that ERK may be phosphorylated relatively soon after induction of mutant KRAS, with subsequent phosphorylation (and activation) of stress kinases and several other proteins. We also observed increased phosphorylation of ERK 24 hr after induction of mutant KRAS by western blot in all three LUAD cell lines ([Figure 1C](#fig1){ref-type="fig"}). In H358 and in H1975-based cell systems we observed persistently increased levels of P-ERK and, ultimately, the presence of cleaved PARP ([Figure 1---figure supplement 1B](#fig1s1){ref-type="fig"}). We previously reported multiple mechanisms of RAS-induced toxicity in PC9-tetO-KRAS cells ([@bib65]). Based on the cleavage of PARP in the studies shown here, apoptosis appears to be at least one of the mechanisms of reduced viability in H358 and H1975 cell lines. The results shown in [Figure 1](#fig1){ref-type="fig"} suggest that ERK itself could be the signaling node that causes a loss of viable cells when inappropriately activated. As one test of this hypothesis, we used trametinib ([@bib21]), an inhibitor of MEK, the kinase that phosphorylates ERK, to ask whether reduced levels of P-ERK would protect cells from the toxicity caused by induction of mutant KRAS. In all three LUAD cell lines, trametinib completely or partially rescued the loss of viable cells caused by induction of mutant KRAS by dox ([Figure 1D](#fig1){ref-type="fig"}, [Figure 1---figure supplement 1C](#fig1s1){ref-type="fig"}). We confirmed that doses of trametinib that protected cells from the toxic effects of seven days of treatment with dox were associated with reduced levels of P-ERK after 24 hr of induction of mutant KRAS ([Figure 1D](#fig1){ref-type="fig"}). A PI3K inhibitor, buparlisib, did not rescue mutant KRAS-induced lethality in H358-tetO-KRAS cells ([Figure 1---figure supplement 1D](#fig1s1){ref-type="fig"}), implying that the toxic effects of KRAS are not mediated by enhanced signaling via PI3K. To extend these findings and further challenge the hypothesis that P-ERK is an important node in the cell signaling network downstream of KRAS that confers cell toxicity, we transduced LUAD cell lines with retroviral vectors encoding shRNAs that 'knock down' expression of ERK1 or ERK2. Using two different shRNAs for each gene, as well as a non-targeted shRNA vector as control, we stably reduced the levels of ERK1 or ERK2 in the three LUAD cell lines ([Figure 1E](#fig1){ref-type="fig"}). When PC9 and H358 lines were treated with dox to assess the effects of ERK1 or ERK2 knockdowns on the loss of viable cells, we found that depletion of ERK2, but not ERK1, rescued cells from KRAS toxicity after 7 days in dox ([Figure 1E](#fig1){ref-type="fig"}). In H1975 cells, however, neither knockdown of ERK1 nor of ERK2 prevented KRAS-induced cell toxicity. Since trametinib rescues the number of viable cells after induction of KRAS in H1975 cells ([Figure 1D](#fig1){ref-type="fig"}), it seemed possible that either ERK1 or ERK2 might be sufficient to mediate RAS-induced toxicity in this line. In that case, it would be necessary to reduce the levels or the activity of both ERK proteins to rescue H1975 cells from toxicity. We tested this idea by treating dox-induced H1975-tetO-KRAS cells with SCH772984 ([@bib43]), a drug that inhibits the kinase activity of both ERK1 and ERK2 ([Figure 1---figure supplement 1E](#fig1s1){ref-type="fig"}). As we observed with the MEK inhibitor, trametinib, in other lines ([Figure 1D](#fig1){ref-type="fig"}, far right), the ERK inhibitor reduces KRAS-associated toxicity in H1975 cells with concomitant reductions of P-ERK1 and P-ERK2 ([Figure 1---figure supplement 1E](#fig1s1){ref-type="fig"}). To examine this issue in a different way, we performed a genome-wide CRISPR-Cas9 screen to evaluate mechanisms of mutant KRAS-induced toxicity in an unbiased manner. After growing H358-tetO-KRAS cells for 7 days following introduction of the appropriate vectors carrying Cas9 and a library of DNA encoding gene-targeted RNAs (see Materials and methods), guide RNA (sgRNA) targeting ERK2 (MAPK1) was highly enriched in cells grown in the presence of doxycycline ([Figure 1---figure supplement 1F](#fig1s1){ref-type="fig"}, [Supplementary file 1](#supp1){ref-type="supplementary-material"}). Guide RNA targeting RAF1 (CRAF) was also significantly enriched. Data from this CRISPR-Cas9 genome-wide screen strongly suggests that depletion of critical proteins in the RTK-RAS pathway can mitigate the toxicity induced by excess RAS activation. Collectively, our data suggest that LUAD cell lines are sensitive to inappropriate hyperactivation of the ERK signaling node and that toxicity mediated by activation of the RAS pathway is ERK-dependent. DUSP6 is a major regulator of negative feedback, expressed in LUAD cells, and associated with KRAS and EGFR mutations and with high P-ERK levels {#s2-2} ------------------------------------------------------------------------------------------------------------------------------------------------ The evidence that hyperactive ERK signaling has toxic effects on LUAD cells raises the possibility that cancers driven by mutations in the RAS pathway may have a mechanism to 'buffer' P-ERK levels and thereby avoid reaching a lethal signaling threshold. Genes encoding negative feedback regulators are typically activated at the transcriptional level by the EGFR-KRAS-ERK pathway to place a restraint on signaling ([@bib4]). Such feedback regulators previously implicated in the control of EGFR-KRAS-ERK signaling include the six dual specificity phosphatases (DUSP1-6), the four sprouty proteins (SPRY1-4) and the three sprouty-related, EVH1 domain-containing proteins (SPRED1-3) ([@bib4]; [@bib34]). To begin a search for possible negative regulators of RAS-mediated signaling in LUAD cells driven by mutations in either *KRAS* or *EGFR*, we asked whether mutations in either proto-oncogene would up-regulate one or multiple members of these families of regulators, based on the assumption that such proteins might constrain P-ERK levels, leading to optimal growth without cytotoxic effects. To search for potential negative regulators specifically involved in LUAD, we compared amounts of RNAs from *DUSP, SPRY* and *SPRED* gene families in tumors with and without mutations in either *KRAS* or *EGFR*, using RNA-seq data from The Cancer Genome Atlas (TCGA) ([@bib8]) ([Figure 2A,B](#fig2){ref-type="fig"} and [Figure 2---figure supplement 1A,B](#fig2s1){ref-type="fig"}). *DUSP6* was the only negative-feedback regulatory gene with significantly different levels of expression when we compared tumors with mutations in either *KRAS* or *EGFR* with tumors without such mutations (Bonferoni corrected p \< 0.01, two-tailed t-test with Welch's correction). Further, *DUSP6* mRNA was significantly up-regulated in LUAD tumors with mutations in common RTK-RAS pathway components compared to those without, consistent with a role of DUSP6 in regulating EGFR-KRAS-ERK signaling ([Figure 2---figure supplement 1C](#fig2s1){ref-type="fig"}) ([@bib4]; [@bib44]; [@bib45]; [@bib22]; [@bib30]; [@bib73]). *DUSP6* RNA was also present at higher levels in LUADs with *EGFR* or *KRAS* mutations than in tumors without such mutations in an independent collection of 83 tumors collected at the British Columbia Cancer Agency (BCCA, p = 0.004), confirming the findings derived from the TCGA dataset ([Figure 2C](#fig2){ref-type="fig"} and [Figure 2---figure supplement 1D](#fig2s1){ref-type="fig"}). Furthermore, *DUSP6* RNA was more abundant in EGFR/KRAS mutant LUADs than in normal lung tissue (p\<0.0001) whereas no significant differences in *DUSP6* levels were observed between normal lung tissue and tumors without mutations in either of these two genes (p = 0.64) ([Figure 2C](#fig2){ref-type="fig"} and [Figure 2---figure supplement 1D](#fig2s1){ref-type="fig"}). {#fig2} To ascertain whether *DUSP6* is up-regulated specifically in tumors driven by mutant KRAS or mutant EGFR signaling rather than in tumors associated with activation of other oncogenic pathways, we measured *DUSP6* RNA in experimental systems driven by the activation of various oncogenes. In transgenic mouse models of lung cancer, *Dusp6* RNA was present at significantly higher levels in the lungs of mice bearing tumors driven by mutant *EGFR* or *KRAS* transgenes than in normal mouse lung epithelium ([Figure 2D](#fig2){ref-type="fig"}) ([@bib17]; [@bib18]; [@bib54]). In contrast, *Dusp6* RNA levels were not significantly different in lungs from mice with tumors driven by MYC and in normal mouse lung tissue ([Figure 2D](#fig2){ref-type="fig"}). Similarly, increased levels of *DUSP6* RNA were observed in primary human epithelial cells only when the cells were also transduced with mutant *RAS* genes, but not with a variety of other oncogenes or with plasmids encoding GFP (p \< 0.0001) ([Figure 2E](#fig2){ref-type="fig"}) ([@bib6]). Lastly, our LUAD cell lines engineered to produce KRAS^G12V^ in response to dox showed an increase in *DUSP6* RNA that correlated with augmented phosphorylation of ERK and cell toxicity ([Figure 2F](#fig2){ref-type="fig"}). It is unclear why increased levels of *DUSP6* RNA are not sufficient to decrease P-ERK in these inducible systems; this may reflect the localization of P-ERK, which we have not explored here. Together, these findings suggest that DUSP6 is a critical negative feedback regulator activated in response to oncogenic signaling by mutant RAS or EGFR proteins in LUAD. In our previous study ([@bib65]) (see also [Figure 1---figure supplement 1A](#fig1s1){ref-type="fig"}), we found that co-induction of oncogenic KRAS and EGFR activated not only ERK, but also JNK and p38 MAPK pathways, albeit at later times. To investigate whether *DUSP6* is up-regulated solely in response to phosphorylation of ERK or also in response to phosphorylation of JNK and p38, we assessed the relationship of amounts of *DUSP6* RNA in tumors with levels of P-ERK, P-JNK and P-p38 proteins as determined for TCGA ([@bib8]), using the Reverse Phase Protein Array (RPPA). LUADs with a *KRAS* or an *EGFR* mutation contained significantly higher levels of P-ERK -- but not of P-JNK or P-p38 -- than did tumors without those mutations, consistent with a role for these oncogenes in ERK activation ([Figure 2G](#fig2){ref-type="fig"}). Furthermore, tumors with high *DUSP6* RNA have relatively high amounts of P-ERK but not of P-JNK or P-p38 ([Figure 2H](#fig2){ref-type="fig"}). Lastly, there is a positive correlation between P-ERK levels and *DUSP6* RNA in LUAD ([Figure 2I](#fig2){ref-type="fig"}), whereas no such association was observed between *DUSP6* RNA and P-JNK or P-p38 ([Figure 2---figure supplement 1E,F](#fig2s1){ref-type="fig"}). Together, these observations support the proposal that *DUSP6* is expressed in response to activation of ERK and that it serves as a major negative feedback regulator of ERK signaling in LUAD, buffering the potentially toxic effects of ERK hyperactivation. Knockdown of DUSP6 elevates P-ERK and reduces viability of LUAD cells with either KRAS or EGFR oncogenic mutations {#s2-3} ------------------------------------------------------------------------------------------------------------------ If DUSP6 is a negative feedback regulator of RAS signaling through ERK, then inhibiting the function of DUSP6 in LUAD cell lines driven by oncogenic KRAS or EGFR should cause hyperphosphorylation and hyperactivity of ERK, possibly producing a signaling intensity that causes cell toxicity, as observed when we co-express mutant KRAS and EGFR. Consistent with this prediction, introduction of *DUSP6*-specific siRNA pools into PC9 cells decreased DUSP6 levels and reduced the number of viable cells to levels similar to those observed when mutant *EGFR*, the driver oncogene, was itself knocked down ([Figure 3A](#fig3){ref-type="fig"}). siRNA pools for either *DUSP6* or *EGFR* decreased DUSP6 protein levels. A decrease in DUSP6 protein levels with siRNA against *EGFR* RNA can be explained by a reduction in EGFR protein levels causing a decrease in ERK activation ([Figure 3A](#fig3){ref-type="fig"}) and subsequently diminishing expression of *DUSP6*, a direct negative feedback regulator of ERK activity. Importantly, almost complete knockdown of DUSP6 was required to elicit toxic effects in PC9 cells. {#fig3} The pool of Dharmacon-synthesized siRNAs we used is composed of 4 individual siRNAs (labeled DUSP6-6,7,8 and 9, [Figure 3](#fig3){ref-type="fig"} and [Figure 3---figure supplement 1A,B](#fig3s1){ref-type="fig"}). We tested the individual siRNAs to confirm knockdown of DUSP6 protein and assess cell viability after siRNA treatment ([Figure 3---figure supplement 1A,B](#fig3s1){ref-type="fig"}). Treatment of PC9 cells with any one of three particular siRNAs resulted in a significant decrease in DUSP6 levels (particularly DUSP6-6 and DUSP6-7), however, the number of viable cells on day 5 was greater than in cells treated with the non-targeting control siRNA ([Figure 3---figure supplement 1A,B](#fig3s1){ref-type="fig"}). This observation was in contrast to the loss of cell viability we documented with the siRNA pool against DUSP6 ([Figure 3](#fig3){ref-type="fig"}). However, treatment with one other siRNA in the pool, DUSP6-8, resulted in the greatest depletion in DUSP6 protein and also a striking loss of cell viability ([Figure 3---figure supplement 1A,B](#fig3s1){ref-type="fig"}), consistent with the results from the siRNA pool. This suggests that DUSP6 protein levels need to be substantially depleted to exert an effect in PC9 cells. Because only one siRNA in the pool (DUSP6-8) had a deleterious effect on PC9 cells, we confirmed the effects of this siRNA by utilizing another siRNA that targets a different region of DUSP6 mRNA (A 5' coding sequence is targeted by DUSP6-Qiagen, whereas a 3' coding sequence is targeted by DUSP6-8). DUSP6-Qiagen suppresses DUSP6 protein to a level similar to what we observed with the siRNA pool ([Figure 3B,C](#fig3){ref-type="fig"}). We also observed a loss of cell viability in PC9s cells treated with DUSP6-Qiagen siRNA comparable to that of the siRNA pool, suggesting these effects are not off-target ([Figure 3B,C](#fig3){ref-type="fig"}). While it was anticipated that knockdown of mutant EGFR would diminish the numbers of viable cells by reducing levels of P-ERK and its growth-promoting signal, cells in which DUSP6 was knocked down with siRNAs also displayed reduced P-ERK levels five days after transfection, not the expected increase in phosphorylation of ERK ([Figure 3A](#fig3){ref-type="fig"}). One way to reconcile this apparent discrepancy is to examine the kinetics of phosphorylation and dephosphorylation of ERK after manipulation of the abundance of DUSP6 and its resulting effects on RAS signaling. To determine whether an initial, transient increase in P-ERK occurred after nearly complete knockdown of DUSP6, preceding the observed reduction in viable cells, we measured P-ERK in two cell lines with mutations in *EGFR* (PC9 and H1975 cells), one cell line with a mutation in *KRAS* (A549 cells) and a lung squamous cell carcinoma with wildtype *EGFR* and *KRAS* (HCC95 cells) 24 hr after addition of DUSP6 siRNA. In the three cell lines assessed with mutant *EGFR* or *KRAS*, there was a small but consistent increase (\~1.5 fold) in P-ERK 24 hr after receiving DUSP6 siRNA, compared to non-targeting siRNA controls ([Figure 3D](#fig3){ref-type="fig"}). Within 5 days, knockdown of DUSP6 reduced the numbers of viable cells in the LUAD lines with activating *KRAS* or *EGFR* mutations (PC9, H1975 and A549 cells), but not in a cell line with no known activating mutations affecting the EGFR-KRAS-ERK pathway (HCC95 cells) ([Figure 3E](#fig3){ref-type="fig"}). Mirroring the decrease in viability, cleaved PARP was also induced five days after DUSP6 knockdown in EGFR/KRAS mutant, but not EGFR/KRAS wildtype cells ([Figure 3---figure supplement 1C](#fig3s1){ref-type="fig"}). While there was no correlation between sensitivity to DUSP6 knockdown and basal DUSP6 protein levels, KRAS or EGFR mutant cell lines demonstrate higher P-ERK levels and/or a high P-ERK to DUSP6 protein ratio that could contribute to P-ERK hyperactivity and the subsequent decrease in cell viability after inhibition of DUSP6 ([Figure 3](#fig3){ref-type="fig"}-figure supplement D,E,F). Lastly, as described above, reduction of ERK1 or ERK2 levels with shRNAs in EGFR-mutant PC9 cells partially rescued the decreased cell viability caused by DUSP6 knockdown, suggesting that ERK -- at least in part - mediates the toxic effects of DUSP6 inhibition ([Figure 3---figure supplement 1G,H,I](#fig3s1){ref-type="fig"}). These data suggest that knockdown of DUSP6 or potentially other negative feedback regulators that can increase P-ERK would reduce cell viability in cells containing an oncogenic *KRAS* or *EGFR* mutation. Pharmacological inhibition of DUSP6 reduces the number of viable LUAD cells bearing mutations that activate the ERK pathway {#s2-4} --------------------------------------------------------------------------------------------------------------------------- The results presented thus far suggest that LUAD cells with mutations in *KRAS* or *EGFR* depend on negative regulators like DUSP6 to attenuate P-ERK for survival, offering a potentially exploitable vulnerability that could be useful therapeutically. However, blocking synthesis of DUSP6 efficiently with siRNA is difficult, in part because reduced levels of DUSP6 lead to increased levels of phosphorylated ERK, stimulating a subsequent increase in *DUSP6* mRNA. As *DUSP6* mRNA rises, more siRNA may be required to sustain the reduction of DUSP6. Based on this negative feedback cycle, we reasoned that pharmacological inhibition of the enzymatic activity of DUSP6 would be more effective. A small molecule inhibitor of DUSP6, (E)−2-benzylidene-3-(cyclohexylamino)−2,3-dihydro-1H-inden-1-one (BCI), was identified through an in vivo chemical screen for activators of fibroblast growth factor signaling in zebrafish ([@bib41]; [@bib33]). BCI inhibits DUSP6 allosterically, binding near the active site of the phosphatase, inhibiting activation of the catalytic site after binding to its substrate, ERK ([@bib41]). BCI also selectively inhibits DUSP1, which, like DUSP6, has catalytic activity dependent on substrate binding. However, as demonstrated in [Figure 2A](#fig2){ref-type="fig"}, *DUSP1* is not significantly up-regulated in LUADs with *EGFR* or *KRAS* mutations. Furthermore, siRNA-mediated knockdown of DUSP1, as opposed to knockdown of DUSP6, has no effect on viability of EGFR-mutant H1975 cells, suggesting that DUSP6 should be the main target of BCI ([Figure 4---figure supplement 1A,B](#fig4s1){ref-type="fig"}). We tested 11 lung cancer cell lines - 8 with a *KRAS* or *EGFR* mutation and 3 with no known activating mutations in these genes -- with a dosing strategy covering the previously determined active range of the drug ([@bib61]). We predicted that cancer lines with mutations in *KRAS* or *EGFR* would be more sensitive to the potential effects of BCI treatment on numbers of viable cells, since DUSP6 would be required to restrain the toxic effects of P-ERK in these cells. Our findings are consistent with this prediction ([Figure 4A,B](#fig4){ref-type="fig"}). The cell lines fell into three categories of sensitivity: 1) the most sensitive lines, with IC50s between 1--3 uM and with \> 90% loss of viable cells at 3.2 uM, all harbored *KRAS* or *EGFR* mutations; 2) the one line with intermediate sensitivity, H1437 (IC50 \> 4 uM), contains an activating mutation in *MEK* (Q56P); and 3) the relatively insensitive lines (IC50s ≥ 5 uM) lack known mutations affecting the EGFR-KRAS-ERK signaling pathway. The insensitive cell lines did not demonstrate the marked (\> 90%) reduction in numbers of viable cells observed with the sensitive cell lines and only sensitive cell lines showed induction of cleaved PARP after BCI treatment ([Figure 4---figure supplement 1C](#fig4s1){ref-type="fig"}). Together, these data suggest that pharmacological inhibition of DUSP6 specifically kills cells with *EGFR* or *KRAS*-mutations. ![Treatment with the DUSP6 inhibitor BCI selectively kills LUAD cell lines with KRAS or EGFR mutation, implying a dependence on ERK-mediated signaling.\ (**A--B**) BCI induces toxicity specifically in lung cancer cell lines with mutations in genes encoding components in the EGFR-KRAS-ERK pathway. (**A**) Eleven lung cancer cell lines were treated with increasing doses of BCI for 72 hr based on the reported effective activity of the drug ([@bib61]). Cell lines could be assigned to three distinct groups: sensitive (red), intermediate (green) and insensitive (black). All sensitive cell lines contained either *EGFR* or *KRAS* mutations; the intermediate and insensitive cell lines were wild-type for genes encoding components of the EGFR-KRAS-ERK signaling pathway (as determined by the Sanger Cell Line Project and the Cancer Cell Line Encyclopedia \[[@bib5]\]). Experiments were done in biological duplicate with the average values presented ±SEM. (**B**) Crystal Violet stain of cells plated in the indicated doses of BCI or control (0 = 0.1% DMSO) for 72 hr. Sensitive cells with a *KRAS* mutation (H358 cells; denoted with red underlining) show a more pronounced decrease in cell number than do cells without oncogenic mutations in genes encoding components of the EGFR-KRAS-ERK pathway (H1648 cells; black underlining). Experiments were done in biological duplicate with a representative image shown. (**C**) BCI increases P-ERK levels specifically in BCI-sensitive cell lines. Sensitive lines (H358, PC9, H1975 and A549; red underlining) and insensitive lines (HCC95 and H1648; black underlining) were treated with the indicated doses of BCI or vehicle control (0.1% DMSO) for 30 min, and the levels of ERK (p44/p42) and P-ERK (P-p44/42 T202/Y204) assessed by Western blot. P-ERK appeared in the sensitive cells at low doses of BCI, but P-ERK levels did not increase in the insensitive cells at the tested doses of BCI. (**D**) Dosimetry plots from the experiment shown in panel. (**C**) (**E--F**) Cell lines sensitive to BCI are also dependent on P-ERK for survival. BCI-sensitive cells with oncogenic mutations in *EGFR* or *KRAS* (PC9 and H358, respectively; red underlining) and BCI-insensitive cells (H1648 and HCC95; black underlining) were treated with the indicated doses of the MEK inhibitor trametinib for 72 hr; viable cells were measured with Alamar blue and compared to cells receiving the vehicle control (0 = 0.1% DMSO). (**E**) Treatment with trametinib decreased P-ERK levels as determined by western blot. (**F**) The reduction in P-ERK corresponded to a greater decrease in viable cells in BCI-sensitive lines (red coloring), compared to BCI-insensitive cell lines (black coloring).](elife-33718-fig4){#fig4} P-ERK levels increase in LUAD cells after inhibition of DUSP6 by BCI, and P-ERK is required for BCI- mediated toxicity {#s2-5} ---------------------------------------------------------------------------------------------------------------------- Based on findings in the preceding section, we predicted that BCI-mediated inhibition of DUSP6 would increase P-ERK to toxic levels, similar to the effects of co-expressing mutant *KRAS* and *EGFR*. To test this proposal, we measured total ERK and P-ERK after BCI treatment in sensitive and insensitive cell lines. A subset of the most sensitive cell lines, H358 (KRAS mutant) and PC9 and H1975 (EGFR mutants), demonstrated a large, dose-dependent increase in P-ERK in response to BCI treatment, with appreciable increases observed even at the lowest doses tested (1 uM) ([Figure 4C,D](#fig4){ref-type="fig"}). This induction of P-ERK precedes the appearance of cleaved PARP and cell death, as indicated by a time course of observations after BCI treatment in KRAS-mutant H358 cells ([Figure 4---figure supplement 1D](#fig4s1){ref-type="fig"}). Likewise, another sensitive cell line, A549 (KRAS mutant), demonstrated an increase in P-ERK, albeit at higher BCI concentrations, consistent with a less acute BCI sensitivity ([Figures 3C](#fig3){ref-type="fig"} and [4C,D](#fig4){ref-type="fig"}). Conversely, BCI did not induce increases in P-ERK in the insensitive cell lines HCC95 and H1648, even at the highest levels of BCI (10 uM) ([Figure 4C,D](#fig4){ref-type="fig"}). Importantly, cell lines sensitive to BCI were also dependent on sustained P-ERK signaling for survival, as the MEK inhibitor trametinib, while effectively reducing P-ERK in all cell lines, reduced cell viability to a greater degree in BCI- sensitive lines (H358 and PC9) compared to BCI-insensitive lines (H1648 and HCC95; [Figure 4E,F](#fig4){ref-type="fig"}). Thus, the oncogenic mutation profile and dependency on activation of the EGFR-RAS-ERK pathway correlates with dependence on DUSP6 activity. These correlations are likely to reflect the central significance of P-ERK as a determinant of cell growth and viability. To confirm whether P-ERK is involved in regulation of BCI-mediated cell death, we treated KRAS mutant H358 cells with a combination of BCI and the ERK1/2 inhibitor VX-11E, predicting that simultaneous inhibition of DUSP6 and ERK would mitigate the toxic effects of BCI treatment. Unlike other ERK inhibitors such as SCH772984, VX-11E does not block ERK phosphorylation, but instead limits ERK activity following phosphorylation ([@bib10]). Consistent with this, while no difference in P-ERK induction was observed, VX-11E treatment limited BCI- induced phosphorylation of the downstream ERK target RSK ([Figure 4---figure supplement 1F](#fig4s1){ref-type="fig"}). In addition, treatment with VX-11E lead to a relative increase in the number of viable cells after BCI treatment in a dose-dependent manner, with higher VX-11E concentrations demonstrating less decline in viability in response to BCI compared to lower doses ([Figure 4---figure supplement 1E](#fig4s1){ref-type="fig"}). Together, these data suggest that ERK activation plays a vital role in mediating the inhibitory effects of BCI treatment in KRAS or EGFR mutant lung cancer cells. To further understand BCI-mediated toxicity, we searched for potential resistance mechanisms through an unbiased, genome-wide CRISPR screen of the type described earlier ([Figure 1---figure supplement 1F](#fig1s1){ref-type="fig"}). If loss of genes targeted by guide RNA confers resistance, that can reveal the nature of the pathway being targeted, since inhibited expression of the gene mitigates the effects of the drug. We performed this screen in H460 cells that are mutant (Q61H) for KRAS and sensitive to BCI ([Figure 4A](#fig4){ref-type="fig"}). In the screen, we found that sgRNAs targeting KRAS were significantly enriched in *KRAS*-mutated H460 cells upon treatment with BCI compared to untreated controls ([Figure 4---figure supplement 1G](#fig4s1){ref-type="fig"}, [Supplementary file 1](#supp1){ref-type="supplementary-material"}). Guide RNA targeting *KRAS* were depleted in the absence of drug suggesting a dependence on mutant KRAS in this cell line. These results suggest that KRAS pathway activity is a major determinant of sensitivity to BCI ([Figure 4---figure supplement 1G](#fig4s1){ref-type="fig"}). To validate these results, we cloned two individual sgRNAs targeting *KRAS* and transduced H460 cells. After 7 days of puromycin selection, the polyclonal population was evaluated for KRAS depletion ([Figure 4---figure supplement 1H](#fig4s1){ref-type="fig"}). The KRAS-targeted and control H460 cells were treated at this time point with a dose response of BCI for 72 hr. Cells that contained sgRNAs against *KRAS* were less sensitive to BCI than cells containing control sgRNA and un-manipulated cells ([Figure 4---figure supplement 1I](#fig4s1){ref-type="fig"}). We also generated two clones of DUSP6-deficient H358 cells using CRISPR-Cas9 and independent guide RNAs ([Figure 4---figure supplement 1J](#fig4s1){ref-type="fig"}). Unexpectedly, both clones remained responsive to BCI's cell killing activity ([Figure 4---figure supplement 1K](#fig4s1){ref-type="fig"}). These results may be explained by the presence of DUSP1 ([Figure 4---figure supplement 1J](#fig4s1){ref-type="fig"}) and the reported activity of BCI against DUSP1 in addition to DUSP6. Further studies will be required to ascertain if these cells are still dependent on P-ERK for BCI-mediated sensitivity through DUSP1 or through another mechanism. While BCI sensitivity may not be solely due to DUSP6, our genome-wide screen for resistance to BCI suggests activation of the RAS pathway is at least partly required. To further test RAS pathway dependency and its relation to BCI sensitivity, we predicted that stimulating the EGFR-RAS-ERK pathway in a BCI-insensitive cell line would make the cells more dependent on DUSP6 activity and more sensitive to BCI. Using HCC95 lung squamous carcinoma cells, which express relatively high levels of wild-type EGFR ([Figure 5A](#fig5){ref-type="fig"}), we showed that EGF increased the levels of both P-EGFR and P-ERK, confirming activation of the relevant signaling pathway ([Figure 5A,B](#fig5){ref-type="fig"}, [Figure 5---figure supplement 1](#fig5s1){ref-type="fig"}). In addition, BCI further enhanced the levels of P-ERK, especially in the EGF-treated cells, with dose-dependent increases; these findings are similar to those observed in cell lines with *EGFR* or *KRAS* mutations ([Figure 4C,D](#fig4){ref-type="fig"}). After pretreatment with EGF (100 ng/mL) for ten days and treating the cells with increasing doses of BCI to inhibit DUSP6, 3 uM BCI reduced the number of viable HCC95 cells by approximately 40% compared to the control culture that did not receive EGF ([Figure 5C](#fig5){ref-type="fig"}). This outcome implies that prolonged EGF treatment and subsequent activation of P-ERK signaling makes HCC95 cells dependent on DUSP6 activity, as also observed in cell lines with *EGFR* or *KRAS* mutations ([Figure 4A](#fig4){ref-type="fig"}). Taken together, these findings suggest that LUAD cells with *KRAS* or *EGFR* mutations are sensitive to BCI because the drug acutely increases P-ERK beyond a tolerable threshold in a manner analogous to the synthetic lethality we previously described in LUAD lines after co-expression of mutant KRAS and EGFR ([@bib65]). {#fig5} Discussion {#s3} ========== The pattern of mutual exclusivity observed with mutant *EGFR* and mutant *KRAS* genes in LUAD is a consequence of synthetic lethality, not pathway redundancy; co-expression of these oncogenes is toxic, resulting in loss of viable cells ([@bib65]; [@bib67]). There are reports of exceptions to this mutual exclusivity but these arise in conditions that include inhibition of EGFR ([@bib7]; [@bib55]). This is to be expected, as cells treated with kinase inhibitors are not experiencing the effects of both oncogenes (i.e. mutant EGFR and mutant KRAS). A cancer cell that has not been exposed to inhibitors (e.g. against mutant EGFR) could arise, particularly at an advanced stage of disease, with activating mutations in both EGFR and KRAS; but we would anticipate that other events---like decreased RAS-GTP levels\-\--might prevent P-ERK from reaching toxic levels. Despite the possible exceptions, it remains critical to understand why, based on the pattern of mutual exclusion, cells are generally unable to tolerate the combination of these two oncogenes more readily. And what are the biochemical mechanisms by which the toxicity is mediated, might be modulated to avoid lethality, or could be exploited therapeutically? To address these questions, we began by regulating the expression of mutant *KRAS* in LUAD cell lines carrying mutant *RAS* or *EGFR* alleles. The levels of RAS activation in these cells are not expected to mirror what is found in tumors; these levels presumably will exceed what tumors can tolerate. We suggest that tumor cells could experience this state during progression, particularly when co-mutations in the RAS pathway have occurred. Understanding how the toxicity arises provides insight into mutual exclusivity and how limits for RAS activation may be set and exploited in cancer cells. Our efforts to answer these questions have led to the conclusions that the toxicity is mediated through the hyperactivity of phosphorylated ERK1/2 and that inhibition of DUSP6 may re-create the toxicity through the role of this phosphatase as a negative regulator of ERK1/2. Several results reported here support these conclusions: (i) the previously reported toxicity that results from co-expression of mutant *EGFR* and mutant *KRAS* is accompanied by an early increase in the phosphorylation of ERK1/2, and the effects can be attenuated by inhibiting MEK (which phosphorylates and activates ERK1/2) or by reducing ERK levels with inhibitory RNAs; (ii) DUSP6, a phosphatase known to be a feedback inhibitor of ERK activity, is present at relatively high levels in LUADs with *EGFR* and *KRAS* mutations; and (iii) inhibition of DUSP6, either by introduction of siRNAs or by treatment with the drug BCI, reduces the number of viable LUAD cells with *EGFR* or *KRAS* mutations or of BCI-resistant cells exposed to EGF. Taken in concert, these findings support a general hypothesis about cell signaling. Activation of a biochemical signal from a critical node, such as ERK, in a signaling pathway must rise to a certain level to drive neoplastic changes in cell behavior; if signal intensity falls below that level, the cells may revert to a normal phenotype or initiate cell death as a manifestation of what is often called 'oncogene addiction" ([@bib47]; [@bib69]; [@bib15]; [@bib66]; [@bib59]). Conversely, if the intensity of signaling rises to exceed a higher threshold, the cells may display a variety of toxic effects, including senescence, vacuolization, or apoptosis ([@bib65]; [@bib11]; [@bib58]; [@bib27]; [@bib48]; [@bib75]). In this model, two approaches to cancer therapy can be envisioned: (i) blocks to signaling that reverse the oncogenic phenotype or induce the apoptosis associated with oncogene addiction, or (ii) enhancements of signaling that cause selective toxicity in cells with pre-existing oncogenic mutations, a form of synthetic lethality that depends on changes that produce a gain rather than a loss of function. The former is exemplified by using inhibitors of EGFR kinase activity to induce remissions in LUAD with EGFR mutations ([@bib37]; [@bib49]; [@bib50]). Based on the findings presented here, the latter strategy might be pursued by using inhibitors of DUSP6 or other negative feedback regulators to block its usual attenuation of signals emanating from activated ERK1/2. Several factors are likely to determine the threshold for producing the cell toxicity driven by hyperactive signaling nodes, such as ERKs, in cancer cells. These factors are likely to include allele-specific attributes of oncogenic mutations in genes such as *KRAS* ([@bib26]) and *BRAF* ([@bib26]; [@bib71]; [@bib46]); the cell lineage in which the cancer has arisen ([@bib61]; [@bib71]; [@bib74]); the levels of expression of mutant cancer genes ([@bib75]; [@bib46]; [@bib12]; [@bib2]); the co-existence of certain additional mutations ([@bib5]); and the multiple proteins that negatively regulate oncogenic proteins through feedback loops, such as MIG6 on EGFR ([@bib2]; [@bib38]; [@bib3]), GAPs on RAS proteins ([@bib13]; [@bib68]), or SPROUTYs and DUSPs on kinases downstream of RAS ([@bib30]; [@bib61]; [@bib74]). All such factors would need to be considered in the design of therapeutic strategies to generate signal intensities that are intolerable specifically in cancer cells. DUSP6 is a well-established negative regulator of ERK activation in a normal cellular context (reviewed in [@bib29], and [@bib64]), so it is perhaps not surprising that this protein appears to have a critical role in persistently limiting ERK activation, even in a pathological context such as cancer. The findings presented here, as well as recent results from others ([@bib61]; [@bib35]; [@bib70]), support several underlying features of a therapeutic strategy based on inordinate signaling activity involving RAS proteins: that the activity of ERK needs to be actively controlled in cancer cells of diverse tissue origins; that hyperactivation of ERK can be deleterious to cells; and that inhibition of negative regulators like DUSP6 can create a toxic cellular state. This leads to the hypothesis that cancer cells *dependent* on ERK signaling have an active RTK-RAS-RAF-MEK pathway that produces levels of activated (phosphorylated) ERK1/2 that *require* attenuation. In other words, ERK-dependent tumor cells, including cancers driven by mutant RTK, RAS, BRAF, or MEK proteins, will have a vulnerability to hyperactivated ERK and that vulnerability can potentially be exploited by inhibition of feedback regulators like DUSP6. Relevant to this concept are recent studies that address 'drug addiction' whereby cells lose viability when the inhibitor (e.g. vemurafenib) is removed ([@bib25]; [@bib32]; [@bib14]; [@bib42]; [@bib63]). These scenarios, in which an additional mutation can arise in the RTK-RAS-RAF-MEK pathway, create conditions similar to those we have modeled, once the inhibitor is removed. Additionally, Hata et al. have shown that mutations can arise while cells are exposed to a drug; as mentioned above, such mutations might appear to violate patterns of mutual exclusivity but the pattern only arose because of pathway down-modulation ([@bib24]) Recently, Leung et al. have found a similar dependency on ERK activation limits in mutant BRAF-driven melanoma ([@bib35]). The mechanisms of cell toxicity that arise from hyper-activation of ERK are likely to be diverse. We previously documented autophagy, apoptosis and macropinocytosis in cells expressing mutant EGFR and mutant KRAS, and others have described parthanatos and pseudosenescence as mechanisms for cell death from hyper-activation of ERK ([@bib25]). ERK-dependent processes may differ from cell type to cell type based on mutation profiles and cellular state at the time of ERK activation. This same dependence on ERK (ERK2 specifically) has been documented for senescence when mutant RAS is introduced into normal cells ([@bib60]). The hypothesis that DUSP6 regulates ERK activity in the presence of signaling through the RAS pathway is particularly attractive in view of the frequency of *RAS* gene mutations in human cancers and the difficulties of targeting mutant RAS proteins ([@bib62]; [@bib51]; [@bib16]). Because DUSP6 directly controls the activities of ERK1 and ERK2, rather than proteins further upstream in the signaling pathway, it appears to be well-situated for controlling both the signal delivered to ERK through the activation of RAS and the signal emitted by phosphorylated ERK. Recently, Wittig-Blaich et al. have also found that inhibition of DUSP6 by siRNA was toxic in melanoma cells carrying mutant BRAF ([@bib70]). Inhibition of other DUSPs, like DUSP5, that regulate ERK1 and ERK2 may create similar vulnerabilities and should be explored ([@bib30]; [@bib31]). These ideas should provoke searches for inhibitors of DUSPs and other feedback inhibitors of this signaling pathway, as well as experiments that better define the downstream mediators and the consequences of non-attenuated ERK signaling. Materials and methods {#s4} ===================== Cell lines and culture conditions {#s4-1} --------------------------------- PC9 (PC-9), H358 (NCI-H358), H1975 (NCI-H1975), H1648 (NCI-H1648), A549, H460 (NCI-H460), H23 (NCI-H23), H2122 (NCI-H2122), H1650 (NCI-H1650), H2009 (NCI-H2009), H2030 (NCI-H2030), H1437 (NCI-H1437) and HCC95 cells were obtained from American Type Tissue Culture (ATCC) or were a kind gift from Dr. Adi Gazdar (UTSW) or Dr. Romel Somwar (MSKCC). Cell lines were periodically checked for mycoplasm contamination and found to be negative. Cells have been validated by STR profiling. For experiments involving doxycycline inducible constructs, cells were maintained in RPMI-1640 medium (Lonza) supplemented with 10% Tetracycline-free FBS (Clontech) or FBS that was tested to be Tet-free (VWR Life Science Seradigm), 10 mM HEPES (Gibco) and 1 mM Sodium pyruvate (Gibco). For other experiments, cells were grown in RPMI-1640 medium (Thermo Fisher) supplemented with 10% FBS (Sigma), 1% Glutamax (Thermo Fisher) and Pen/Strep (Thermo Fisher). Cells were cultured at 37°; air; 95%; CO2, 5%. Where indicated, doxycycline hyclate (Sigma-Aldrich) was added at the time of cell seeding at 100 ng/ml. Trametinib (Selleckchem), Buparlisib (Selleckchem), SCH772984 (Selleckchem), Dual Specificity protein phosphatase 1/6 inhibitor (BCI) (Calbiochem), and EGF recombinant human protein solution (Thermo Fisher) were added at the time of cell seeding at the indicated doses. Plasmids and generation of stable cell lines {#s4-2} -------------------------------------------- Plasmids used were identical to those described in a prior publication ([@bib65]). In brief, DNAs encoding mutant KRAS or GFP were cloned into pInducer20, a vector that carries a tetracycline response element for dox-dependent gene control and encodes rtTA, driven from the UbC promoter ([@bib40]). Lentivirus was generated using 293 T cells (ATCC), psPAX2 \#12260 (Addgene, Cambridge, MA) and pMD2.G (Addgene plasmid\#12259). Polyclonal cell lines (H358-tetO-GFP, H358-tetO-KRAS^G12V^, PC9-tetO-GFP, H1975-tetO-GFP) and single cell-derived clonal cell lines (PC9-tetO-KRAS^G12V^, H1975-tetO-KRAS^G12V^) were used. pLKO.1-based lentiviral vectors were used to establish cells stably expressing shRNAs for the indicated genes. Knockdown was achieved using two independent shRNAs targeting *ERK1* (noted in text as A4 or ERK1-4 and A5 or ERK1-5) or *ERK2* (noted in text as G6 or ERK2-6 and G7 or ERK2-7) RNAs. shRNA-GFP: GCAAGCTGACCCTGAAGTTCAT shRNA-ERK1 (A4): CGACCTTAAGATTTGTGATTT shRNA-ERK1 (A5): CTATACCAAGTCCATCGACAT shRNA-ERK2 (G6): TATTACGACCCGAGTGACGAG shRNA-ERK2 (G7): TGGAATTGGATGACTTGCCTA shRNAs targeting GFP or a scramble sequence were used as controls. shRNA constructs were kindly provided by J. Blenis, Weill Cornell Medicine. Lentivirus was generated using 293 T cells as above. After transduction, polyclonal cells were selected with puromycin and maintained as a stable cell line. Measurements of protein levels {#s4-3} ------------------------------ Cells were lysed in RIPA buffer (Boston Bioproducts) containing Halt protease and phosphatase inhibitor cocktail (Thermo Fisher). For experiments involving dox-inducible constructs, lysates were cleared by centrifugation, and protein concentration determined by Pierce BCA protein assay kit (Thermo Fisher). Samples were denatured by boiling in loading buffer (Cell Signaling). 20 μg of lysates were loaded on 10% MiniProtean TGX gels (Bio-Rad), transferred to Immun-Blot PVDF membranes (Bio-Rad), blocked in TBST (0.1% Tween-20) and 5% milk. For all other experiments, samples were denatured by boiling in loading buffer (BioRad) and 25 μg of lysates were loaded on 4--12% Bis-Tris gradient gels (Thermo Fisher), run using MOPS buffer, transferred to Immobilon-P PVDF membranes (Millipore) and blocked in TBST (0.1% Tween-20)/5% BSA (Sigma). Primary incubation with antibodies was performed overnight at 4° in 5% BSA, followed by appropriate HRP-conjugated secondary antisera (Santa Cruz Biotechnology) and detected using ECL (Thermo Fisher). Antibodies were obtained from Cell Signaling and raised against the following proteins: phospho p-38 (4511), p38 (8690), p-p44/p42 (ERK1/2) (9101), p44/p42 (ERK1/2) (4695), p-SAPK/JNK (4668), SAPK/JNK (9252), P-EGFR (3777, 2234), EGFR (2232), KRAS (8955), PARP (9542), cleaved-PARP (5625), α-Tubulin (3873) and β-Actin (3700, 4970). Additionally, we used an antibody against GFP (A-21311, Thermo Fisher), DUSP1 (ab1351, abcam) and DUSP6 (ab76310, abcam and SC-377070, SC-137426, Santa Cruz).. For 24 hr time course experiments, 100,000 cells (PC9, H1975) or 500,000 cells (H358) per well were seeded in a 6-well plate and stimulated with dox or dox and drug. For 5 day experiments, 25,000 cells were seeded in 6-well format. For 7 day time course experiments, 300,000 cells (H358) or 30,000 cells (H1975) were seeded into 10 cM plates and media was changed every day. For proteome profiler array, 200 ug of total lysate was incubated on membranes in the A/B set (ARY003B, R and D Systems) and processed according to protocol (R and D Systems). Film exposures were scanned and spot density quantified using Image Studio Lite (Licor). Data were plotted in Microsoft Excel. For western blots with BCI and Trametinib, cells were seeded to achieve 80% confluency 18 hr post seeding. Medium was aspirated and replaced with antibiotic-free medium containing drug at indicated concentrations and incubated for 30 min. Cells were lysed and protein levels assessed as stated above. Quantification of western blot images was performed using ImageJ software. Scanned files were saved in TIFF format, and background was subtracted from all images. Rectangle tool was used to fully encompass each separate band. Rectangles and bands were assigned lanes and histogram plots were generated based on each lane. Each histogram was enclosed using a straight line across the bottom and the 'magic wand' tool generated a value for area of histogram. These values were exported to and assessed using Excel and Graphpad Prism software. Measurements of viable cells {#s4-4} ---------------------------- For experiments with dox-inducible constructs, cells were seeded into media containing doxycycline (100 ng/ml) and/or drug (Trametinib, SCH772984). Media (with or without doxycycline or drug) were replenished every 3 days during the 7 days. At indicated time points, medium was aspirated and replaced with medium containing Alamar Blue (Thermo Fisher). Fluorescence intensities from each well were read in duplicate on a FLUOstar Omega instrument (BMG Labtech), and data plotted in Microsoft Excel. Cells were seeded in triplicate in 24-well format at 1,000 cells/well (PC9 or H1975 derivatives) or 5,000 cells/well (H358 derivatives). For other experiments, cells were grown in 6-well plates, Alamar Blue added, and intensities measured for each well in quadruplicate using a Cytation 3 Multi Modal Reader with Gen5 software (BioTek). For crystal violet assays, cells were seeded to achieve 80--90% confluency at the end point in the absence of drug treatment. 18 hr later, medium was aspirated and replaced with medium containing drug. Cells were incubated for 72 hr, washed with PBS and Crystal Violet solution (Sigma) was added and incubated for 2 min before washing again with PBS and imaging. Genomic datasets and analyses {#s4-5} ----------------------------- RNA-Seq (RSEM) data for EGFR-KRAS-ERK pathway phosphatases (DUSP1-6, SPRED1-3, SPRY1-4) along with corresponding mutational data for *EGFR*, *KRAS, MET, ERBB2, BRAF, NF1, NRAS* and *HRAS* for 230 lung adenocarcinoma tumors from The Cancer Genome Atlas ([@bib8]) were downloaded from cBioPortal (<http://www.cbioportal.org/>) ([@bib9]; [@bib19]). Expression of each gene was compared between tumors with *KRAS* or *EGFR* mutations and those without, using an unpaired T-Test. Resulting p-values were adjusted for multiple comparisons using a Bonferroni correction and the --Log~2~ value plotted as an indication of significance. Normalized expression values (sample gene value -- median gene expression across all samples/row median absolute deviation) for each gene were also plotted using MORPHEUS software (<https://software.broadinstitute.org/morpheus>, Broad Institute) as a heat map. Expression of DUSP6 was also individually compared for tumors with EGFR mutation only, KRAS mutation only, or any RTK-RAS-ERK pathway mutation (EGFR, KRAS, MET, BRAF, ERBB2, NRAS, HRAS or NF1) vs those wild-type for the in each instance using a two-tailed Mann-Whitney U-Test in Prism 7 (Graphpad). Reverse phase protein array (RPPA) data (replicate-base normalized \[[@bib1]\]) for 182/230 tumors were downloaded from the UCSC Cancer Genomics Browser. Levels of MAPKPT202Y204, P38PT180Y18 and JNKPT183Y185 were compared between samples with a *KRAS* or *EGFR* mutation and those without, using the Mann-Whitney U-Test.. Likewise, samples were separated into groups with high and low DUSP6 expression levels, based on the highest and lowest *DUSP6* expression quartiles; MAPKPT202Y204, P38PT180Y18 and JNKPT183Y185 levels were compared between the groups as above. Lastly, MAPKPT202Y204 levels from RPPA (RBN values) were correlated with *DUSP6* expression (Log~2~ RSEM values), and the Pearson correlation coefficient and p-value determined. As phospho-protein levels were predicted to be higher in samples with KRAS or EGFR mutation or high DUSP6, one-tailed p-values were calculated. *DUSP6* expression was also compared between tumors with and without *EGFR* or *KRAS* mutations in 83 tumors and matched normal lung tissues from the BC Cancer Agency (BCCA) and deposited in the Gene Expression Omnibus (GSE75037) as described above. Similarly, *DUSP6* expression was compared between human epithelial cells expressing various oncogenes or GFP control (GSE3151) ([@bib6]). Lastly, Affymetrix Mouse Genome 430 2.0 Arrays were used to profile the lung from genetically engineered mouse models of lung cancer with and without the expression of different driver oncogenes (EGFR-DEL, EGFR-L858R, KRAS-G12D and MYC) ([@bib18]; [@bib54]; [@bib53]) and levels of DUSP6 compared using a two-tailed Mann-Whitney U-Test in Prism seven software (Graphpad). siRNA transfections {#s4-6} ------------------- For the time course experiments, 50,000 cells (PC9) per well were seeded in a 6-well plate. For the endpoint experiments, 50,000 cells (PC9, PC9-shERK1-5, PC9-shERK2-7, PC9-shScramble) or 75,000 cells (1975, A549, HCC95) per well were seeded. Cells were then transfected with ON-TARGETplus siRNA pools (Dharmacon) against the following targets as previously described ([@bib36])\-\-- EGFR (L-003114-00-0010), KIF11 (L-003317-00-0010), KRAS (L-005069-00-0010), DUSP6 (L-003964-00-0010)---as well as a non-targeting control (D-001810-10-20). In addition, to test specificity for DUSP6, siRNAs comprising the pool (J-003964-06-0005, J-003964-07-0005, J-003964-08-0005 and J-003964-09-0005) were also tested individually. An additional siRNA (Hs_DUSP6_6 FlexiTube siRNA SI03106404, Qiagen) targeting a different region of DUSP6 coding sequence than J-003964-08-0005 was tested to establish that the decreased viability was not due to off target effects. DUSP6-8 (Dharmacon) Target Sequence: GGCATTAGCCGCTCAGTCA DUSP6-Qiagen (Qiagen) Target Sequence: GTCGGAAATGGCGATCAGCAA For consistent transfection efficiency across experiments, 10 uL of 20 uM siRNA pool was added in 190 uL of OptiMEM (Life Technologies) and 5 uL of Dharmafect was added in 195 uL of OptiMEM (Life Technologies) at room temperature. The siRNA and Dharmafect suspensions were mixed and incubated for 20 min prior to transfection. Media was changed 24 hr after transfection. For sustained knockdown of targets, transfections were conducted on Day 0 and again on Day 3. Viable cells were measured using Alamar Blue as described above. For the time course experiment, cell viability was determined on Day 1, Day 3 (prior to second transfection) and Day 5 or only on Day 5. Results were compared between each siRNA and non-targeting control using a one-sample t-test as previously described ([@bib36]). BCI dose-response treatments {#s4-7} ---------------------------- Dose-response curves for BCI were established using a modified version of the protocol previously described ([@bib36]). Briefly, cells were seeded in quadruplicate at optimal densities into 96-well plates containing media with and without BCI at indicated doses in 0.1% DMSO. Viable cells were measured 72 hr later with Alamar Blue as described above. All experiments were performed in at least biological duplicate and plotted ±SEM. For HCC95 sensitization assays, cells were cultured with or without 100 ng/mL of EGF Recombinant Human Protein Solution (Life Technologies) for 10 days prior to seeding in 96-well plates for BCI dose response assays with or without EGF. The cells were allowed to adhere for 24 hr before treatment with 17 different concentrations of BCI, ranging from 0 to 8 uM, with 0.5 uM increment doses at 0.1% DMSO concentration. Additionally, 100 uM of Etoposide (0.1% DMSO) was added as a positive control for cell death. Cell viability was determined after 72 hr of drug exposure using Alamar Blue. Graphpad Prism software was used to create dose response curves. For BCI rescue experiments, 75,000 H358 cells were seeded in 6-well plates and adhered for 24 hr. After attachment, the cells were treated with varying combinations of VX-11e and BCI with the final DMSO concentration at 0.1% in each well. Cells were treated for 72 hr and then the media was switched with fresh media containing Alamar blue for viability assessment. Resulting values for each BCI + VX-11e containing well were normalized to well containing corresponding concentration of VX-11e only. Experiments were performed in biological triplicate and the average ±SEM plotted. Quantitative RT-PCR {#s4-8} ------------------- Cells were homogenized and RNA extracted using the RNeasy Mini kit (Qiagen) according to the manufacturer's instructions. cDNA was prepared using the High-Capacity cDNA Reverse Transcription kit (Thermo Fisher). RT--PCR reactions were carried out using the TaqMan Gene Expression Master Mix (Thermo Fisher) and TaqMan Gene Expression Assays (Thermo Fischer) for *DUSP6* (Hs00169257_m1) and *GAPDH* (Hs99999905_m1). Reactions were run on a QuantStudio6 Real Time PCR system (Thermo Fisher). The ΔΔCt method was used for relative expression quantification using the average cycle thresholds. Genome-wide CRISPR screens {#s4-9} -------------------------- Genome-wide screens were performed with the Toronto Knockout version 3 (TKOv3) library ([@bib23]). Lentivirus was generated from the TKOv3 library in low passage (\<10) 293FT cells (Thermo Fisher) using Lipofectamine 3000 (Thermo Fisher). Approximately 120 million target cells were then infected with the TKOv3 library virus at an MOI of 0.3, in order to achieve an average 500-fold representation of the sgRNAs after selection. Cells were selected on puromycin for 7 days and then 35 million cells were seeded in culture. For the depletion screens, cells were passaged every 3 days, and after 14 population doublings, 35 million cells were harvested for genomic DNA extraction. For the enrichment screens, media (containing BCI or doxycycline) was changed every 3 days until cell death was no longer observed, at which point the remaining cells were harvested for genomic DNA extraction. sgRNA inserts were amplified with NEBNext High-Fidelity 2X PCR Master Mix (New England BioLabs). Samples were then purified and sequenced on a NextSeq 500 kit (Illumina). For validation of the screen, two separate guides targeting KRAS were cloned into lentiCRISPR v2^75^, lentivirus generated and H460 cells were transduced. Seven days after puromycin selection cells were harvested for protein analysis and seeded in the presence of BCI. A guide against LacZ was used as a control. sgRNA_Lacz: GAGCGAACGCGTAACGCGAA sgRNA_KRAS-1: GGACCAGTACATGAGGACTG sgRNA_KRAS-2: GTAGTTGGAGCTGGTGGCGT For targeting of *DUSP6*, two separate guides were cloned into lentiCRISPR v2, lentivirus generated, and H358 cells were transduced. A clonal population of cells were expanded and screened by western blotting and by DNA sequencing of the *DUSP6* locus. sgRNA_DUSP6-1: GTGCGCGCGCTCTTCACGCG sgRNA_DUSP6-2: ACTCGTATAGCTCCTGCGGC Analysis of CRISPR screen {#s4-10} ------------------------- Sequencing reads were aligned to the reference library to determine the abundance of each sgRNA. sgRNAs with less than 30 raw read counts were excluded from further analysis. The read counts were then normalized to the total number of reads obtained from the respective sample. The log2 fold-change of each sgRNA was calculated by adding a pseudocount of 1 and comparing the abundance of the sgRNAs in the final cell population to their respective abundance in the TKOv3 plasmid library. Finally, genes were ranked according to the second-most enriched or second-most depleted sgRNA. Funding Information =================== This paper was supported by the following grants: - http://dx.doi.org/10.13039/501100000024Canadian Institutes of Health Research PJT-148725 to William W Lockwood. - http://dx.doi.org/10.13039/501100004376Terry Fox Research Institute to William W Lockwood. - http://dx.doi.org/10.13039/501100000245Michael Smith Foundation for Health Research Scholar Award to William W Lockwood. - http://dx.doi.org/10.13039/100000002National Institutes of Health to Harold Varmus. - Meyer Cancer Center at Weill Cornell Medicine to Harold Varmus. - BC Cancer Foundation to William W Lockwood. We would like to thank Katerina Politi (Yale University) for providing gene expression data from her transgenic mice. We would like to thank members of the Varmus lab for useful discussions and Oksana Mashadova, in particular, for experimental help. Additional information {#s5} ====================== No competing interests declared. Conceptualization, Data curation, Formal analysis, Supervision, Investigation, Methodology, Writing---original draft, Project administration, Writing---review and editing. Data curation, Formal analysis, Investigation, Methodology. Data curation, Formal analysis, Investigation, Methodology. Data curation, Formal analysis. Resources, Data curation, Formal analysis, Investigation. Conceptualization, Data curation, Formal analysis, Supervision, Investigation, Methodology, Writing---original draft, Project administration, Writing---review and editing. Conceptualization, Supervision, Writing---original draft, Project administration, Writing---review and editing. Additional files {#s6} ================ 10.7554/eLife.33718.012 ###### Table containing the log2 fold change values for all sgRNAs from CRISPR-Cas9 screens. 10.7554/eLife.33718.013 Data availability {#s7} ----------------- All data generated or analysed during this study are included in the manuscript and supporting files. Source data files have been provided for Figures 2 and Figure 2-supplemental figure 1 in the Methods section and/or in the text. The following previously published datasets were used: CancerGenome Atlas Research Network2014TCGA LUADcBioPortalluad_tcga_pub GazdarAGirardLStephenLWanLZhangW2017Expression profiling of 83 matched pairs of lung adenocarcinomas and non-malignant adjacent tissueNCBI Gene Expression OmnibusGSE75037 NevinsJR2005Oncogene Signature DatasetNCBI Gene Expression OmnibusGSE3151 10.7554/eLife.33718.021 Decision letter Cooper Jonathan A Reviewing Editor Fred Hutchinson Cancer Research Center United States 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. \[Editors' note: formal revisions were requested, following approval of the authors' plan of action.\] Thank you for submitting your article \"Hyperactivation of ERK by mutation-driven RAS signaling or by inhibition of DUSP6 is toxic to lung adenocarcinoma cells\" for consideration by *eLife*. Your article has been reviewed by three peer reviewers, one of whom (Thomas Look) is a member of our Board of Reviewing Editors, and Jonathan Cooper as the Senior Editor. The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this letter to express the many issues that we feel must be addressed if this work is to advance to publication. Summary: In this manuscript Unni et al., describe their work on understanding the mechanism leading to cytotoxicity following forced overexpression of KRAS in LUAD cell lines harboring EGFR or KRAS mutations. They conclude that such toxicity owes to hyperactivation of pERK, which occurs through DUSP6. They show that MEK/ERK inhibitor treatment rescues the effect of forced overexpression while reducing the levels of pERK and that DUSP6 mRNA levels are increased in LUAD harboring EGFR or KRAS mutations relative to those with a wild type status of these proteins. Genetic or pharmacologic targeting of DUSP was found to induce pERK (transiently) and lead to diminished cell proliferation in EGFR/KRAS mutant cell lines but not in those with a wild type status. While overall the findings are intriguing, there are gaps in the experimental evidence provided which bias their conclusions. If the authors can address the issues noted below however this could be a very interesting article. We have the following suggestions to improve the manuscript. For clarity, we address below the essential revisions related to the results presented in each figure. 1\) Figure 1 a) Figure 1A: The expression level of tet-driven ectopic KRAS^G12V^ is very high in each of the three lung adenocarcinoma lines. The authors should demonstrate by western blotting whether these levels are comparable to endogenous KRAS levels found in lung adenocarcinoma lines. Studies from Tyler Jacks on mutant KRAS in knockin mice over 10 years ago emphasized the importance of expression levels of mutant KRAS to mediate transformation in lung adenocarcinoma. Massive overexpression of KRAS^G12V^ can induce senescence rather than transformation in most cell types. b\) Figure 1D: The trametinib/SCH984 experiments in Figure 1D require additional appropriate controls; the authors should show the effect of treatment in the absence of dox stimulation. Also, why are different concentrations of trametinib used? c\) In addition to the biochemistry and cell count data shown in each panel of Figure 1, straightforward studies need to be done to establish the cellular mechanism underlying the cell growth phenotype. Before and after induction of KRAS^G12V^ using dox, studies of cell cycle progression by DNA flow cytometry, for β-GAL expression assays should be done to assess senescence, and cleaved caspase 3, TUNEL and PARP cleavage assays should be done to assess apoptosis. The reduction of cell number 7-days after dox induction of KRAS^G12V^ of \~ 20-40 percent of control could be due to growth arrest, senescence or apoptosis. This will be important to establish with regard to the author\'s interpretation that over activation of ERK leads to synthetic lethality. The need for studies of cellular phenotype in the context of the experiments shown in Figure 1 apply broadly to experiments throughout the paper. d\) Does overexpression of KRASG12C in EGFR or KRAS WT cells also induce toxicity which can be mitigated by co-treatment with trametinib in cells with no activating mutations in EGFR-RAS-ERK pathway? The levels of overexpressed KRASG12C are so high that such levels could induce toxicity by exceeding the upper threshold of RAS-mediated signaling even in cells without hyperactive EGFR-KRAS-ERK signaling. e\) Since the main claim is that hyperactivation of ERK leads to toxicity, the authors ought to replicate their dox-inducible experiments with active ERK to show that these also lead to cytotoxicity. Also, just because inhibiting MEK/ERK reverses the some of the phenotype in plastic, this doesn\'t exclude that other RAS effectors are also involved (see induction of pAKT in addition to pERK in Figure 1B). The authors should thus carry out similar siRNA or inhibitor studies to demonstrate that the cytotoxic effects of KRAS overexpression are not due to pAKT. 2\) Figure 2 a) Figure 2A: There should be symbols under the heat map denoting which tumors are KRAS mutant and which are EFGR mutant. Are there any tumors with ALK/ROS1/RET activation by translocation? Are there tumors with NF1, MET and BRAF mutations? These should be addressed since most of lung adenocarcinoma have activation of RTK/RAS/RAF pathway (TCGA, 2014) and increased dusp6 is among the five gene signatures to predict poor outcome (Chen et al., 2007). b\) After describing the emergence of DUSP6 as a key gene responding to activated ERK based on their own studies, the authors should put their findings of DUSP6 elevated expression in the context of previous work. It has been known for over 10 years that DUSP6 is transcriptionally regulated by ERK-responsive ETS transcription factors downstream of MAPK activation and that DUSP6 serves as a major negative feedback regulator of ERK signaling (Reffas et al., 2000; Kawakami et al., 2003; Eblaghie et al., 2003; Li et al., 2007; Ekerot et al., 2008; Furukawa et al., 2008; Jurek et al., 2009). c\) Figure 2B and C: For these panels, please show separate data categories for KRAS mutant and EGFR mutant tumors. The box plots shown do not really look statistically significantly different. The medians are close, and the quartiles are largely overlapping. Which statistical test was used to show significant differences? Has a biostatistician been consulted about whether parametric or non-parametric tests would be better for these comparisons? A detailed statistical section is needed in the Materials and methods section, outlining the statistical tests used for the data shown in each figure. d\) In Figure 2 A-C, the authors must also include the correlation between DUSP6 and P-p38 and p-JNK to show the readers that no correlation exists between them. e\) Figure 2D: Please provide information or citations about the mouse models used in this figure. f\) Figure 2E: The Material/Methods description about this experiment seems to be missing. Expression levels need to be shown of the transduced onco-proteins by Western blot to verify overexpression. g\) Figure 2G and H: Same comments as 2B and C. 3\) Figure 3 a) Figure 3A: Only one siRNA was used for DUSP6 or EGFR and there is no rescue experiment to prove that the effect is due to knockdown of the intended target. DUSP6 de-phosphorylates ERK so ERK phosphorylation is expected to be increased with DUSP6 knockdown. Please explain why Figure 3A shows the opposite with reduced p-ERK after siDUSP6 knockdown. b\) Figure 3A: Can the toxicity mediated by DUSP6 knockdown in PC9 cells be reduced by co-treatment with trametinib (Figure 3A)? This would help confirm that toxicity caused by depletion of DUSP6 in these cells is due to increased pERK levels. c\) Figure 3B: PC9 cells in this panel show increased p-ERK after DUSP6 kd. Please explain why this is the opposite result compared to Panel A. d\) Figure 3B and C: Did the authors determine if the levels of DUSP6 in their EGFR and KRAS wild-type cell lines, HCC95 and H1648, are not as high as in the cell lines with EGFR or KRAS activating mutations? Lower levels of DUSP6 would indicate that these cells do not need to have similar buffering ability as the EGFR or KRAS mutant cell lines and would be consistent with their findings in Figure 2. In Figure 3B, it seems that DUSP6 is also buffering the pERK levels in HCC95 because knockdown of DUSP6 increases the levels of pERK in these cells by over 1.5 fold, similar to the cells with EGFR or KRAS activating mutations; but this increase in pERK has no impact on the cell survival (Figure 3C). This needs to be reconciled with the authors\' main claim. e\) There seems to be a disconnect between the first part of the manuscript, where the authors describe the mechanism of toxicity caused by forced overexpression of KRAS and the second part, where they describe the effect of DUSP inhibition in cell lines with endogenous KRAS/EGFR mutations. To close this gap, the authors need to determine the levels of DUSP6 protein in their dox-inducible KRAS models and how they correlate with pERK. They should determine if the induction of pERK is transient, as that observed with DUSP6 siRNA, or sustained. Finally, they should use DUSP6 siRNA or BCI to determine if this reverses the effect of forced KRAS expression on pERK or proliferation. Another question that has not been addressed is why doesn\'t forced overexpression of KRAS induce sufficient DUSP6 to override the induction of pERK, if DUSP expression is under the control of EGFR/KRAS? Could there be other factors involved? If the authors can experimentally address these it would be a significant improvement. f\) More evidence is needed to show that the increase in DUSP mRNA is associated with an increased in DUSP6 protein levels or increased DUSP activity. The data in Figure 2I, where the authors compare the relationship between pERK (measured by RPPA) and DUSP6 mRNA is the only such evidence provided. This is underwhelming because the correlation is not great (r=0.1) and because the pERK does not seem to have been controlled for total ERK. g\) Finding that DUSP6 siRNA caused only a transient elevation in pERK (followed by inhibition of pERK at longer intervals), while mimicking the antiproliferative effect observed with forced overexpression of KRAS needs to be clarified with additional experiments. As it stands, it is difficult to agree with the authors conclusion that DUSP6 is the main mediator of the proliferative effect or that the effect of DUSP6 is through pERK. One consideration may be to use constitutively active ERK (i.e. DD phosphomimetic mutants) and attempt to reverse the effect of DUSP6. This is another point that, if addressed experimentally, could add value to the manuscript. h\) If indeed it is true that a transient induction in pERK leads to cytotoxicity several days later, then does EGF stimulation (which also causes a transient induction in pERK) have the same effect in EGFR WT/KRAS MT cells? i\) All three reviewers noted that HCC95 (RRID:CVCL_5137) is a squamous cell carcinoma line. This not comparable to lung adenocarcinoma because this tumor type has different pathway dependencies than LUAD (TCGA, Nature 2012). A wild type LUAD line should be tested instead. j\) Figure 3C: Same comments as 3A. k\) The investigators might consider inactivating DUSP6 with CRISPR-Cas9 in these cell lines to show genetic dependence. 4\) Figure 4 a) Figure 4A: 11 lung cancer cell lines were treated with BCI and viable cells were measured after 72-hour treatment. Most of these cell lines are fast growing cells with doubling time around 20\~30 hours. Reduction of viable cells by 72-hours does not necessarily mean cell killing. TUNEL or caspase-3/PARP western blot needs to be performed to detect the levels of apoptosis. Senescence and cell cycle arrest should be examined too. b\) Figure 4C and D: p-ERK levels at a single time point was measured in multiple cell lines (30 min). In order to explain the loss of viable cells at 72hours in Figure 4A, p-ERK should be measured at serial time points such as 1, 6, 12, 24, 48, 72 hours. c\) The data on BCI are very interesting but it\'s not clear how have the authors established that BCI is selective for DUSP? Sensitive and insensitive cell lines all have IC50s in the 1-5μM range (12/13 cells lines tested). What is the IC50 of the inhibitor in DUSP6 KD or KO cells (or in DUSP1/6 double KD/KO cells)? How do the authors know that the antiproliferative effect of BCI is due to its ability to induce pERK? These questions need to be addressed experimentally. The authors should also attempt to show that the inhibitor has an effect in vivo, given their claim that such an approach could be therapeutically beneficial in patients. d\) Figure 4E and F: Four cell lines were treated with 125nM trametinib for 72 hours. 125nM is a very high dose for trametinib-sensitive cell lines such as H358 (reported IC50 \~50nM). Apoptosis levels should be measured to document any cell death. 5\) Figure 5. See comments for Figure 3B. HCC95 is a SQCC cell line that is quite different from the LUAD cells used in the rest of this manuscript. a\) In figures where the authors make a comparison of protein levels under different conditions the uncut blot with all the lanes on the same blot must be shown. It is not ideal to make a comparison between levels of proteins on two different strips of blots. For example, in Figure 5B, authors claim that EGF increase levels of pEGFR and pERK in HCC95 cells when the pEGFR and pERK levels with and without EGF are on two different strips of blots. This is important to confirm that BCI treatment increases pERK levels in EGFR treated cells. Quantifications should be shown in addition. Overall comment: The authors should soften earlier claims that EGFR-mutations and KRAS-mutations are synthetically lethal in the adenocarcinoma subtype of NSCLC (Unni et al., 2015). Recent genetic studies of EGFR-mutant lung cancer (Blakely et al., 2017) have shown that 2.5%\~4.7% EGFR-mutant lung cancer also harbor KRAS copy number gain or activating mutations. Tumor genomic analyses have indicated that bona fide driver mutations causing lung adenocarcinoma are not as mutually exclusive as previously thought (e.g. PMIDs: 25301630, 28498782, 28445112, 29106415, and other recent publications), particularly in metastatic lung adenocarcinoma instead of early-stage disease and/or during the evolution of treatment resistance in metastatic disease. Further, preclinical studies have shown acquired KRAS gain/activation in response to EGFRi in EGFR-mutant NSCLC, indicating that this can be a mechanism of drug resistance (Politi et al., 2000; Eberlein et al., 2015). The authors should acknowledge these recent works and temper their claim of absolute mutually exclusivity in this disease, at least under these more advanced-stage disease contexts and in the light of the emerging literature. \[Editors\' note: further revisions were requested prior to acceptance, as described below.\] Thank you for resubmitting your work entitled \"Hyperactivation of ERK by mutation-driven RAS signaling or by inhibition of DUSP6 is toxic to lung adenocarcinoma cells\" for further consideration at *eLife*. First, let me apologize profusely for the very long time this has taken. The problem is basically that the reviewers are at an impasse and could not decide on a course of action. Briefly, this is a follow up to a previous paper, that reported anti-proliferative effects of activation of Ras and EGFR in the same cancer cells. This new paper has provides evidence that the anti-proliferative effect of over-expressed, active Ras is due to ERK hyperactivation, and that DUSP6 is critical to prevent adverse effects of active ERK in lung cancer cells. In the revision, the authors have added results using ERK shRNA, ERKi, and a CRISPR screen that together provide convincing evidence that ERK mediates the antiproliferative effect of oncogene overexpression. The finding that sgKRAS are enriched in the BCI screen also indirectly supports that conclusion. The authors have also ruled out that the PI3K pathway is not involved by adding PI3K inhibitor experiments. These additions greatly strengthen the conclusion that hyperactivation of ERK is detrimental. However, the evidence that DUSP6 plays a key role is not convincing. The DUSP6 knockdown was done with a single siRNA with no rescue experiment. (Other siRNAs did not effectively inhibit DUSP6 expression). Attempts to recapitulate the result with a DUSP6 CRISPR knockout were unsuccessful. Therefore, the conclusion that DUSP6 is necessary relies in large part on the specificity of the chemical BCI. The DUSP6 KO cells have the same IC50 to BCI, while DUSP1 siRNA did not affect proliferation in their system. Based on these results, the authors cannot claim that the BCI effect is DUSP6 specific nor that the BCI effect in DUSP6KO cells is driven by DUSP1 inhibition. It probably isn\'t, and by inference, the observed phenotype is not dependent on DUSP6 alone but on other ERK specific DUSPs as well. The reviewers disagreed whether these weaknesses undermined the impact to the point where the paper was unsuitable for publication, or whether lengthy additional experiments would be needed. The *eLife* approach is not to ask for multiple rounds of revision. In this spirit, I suggest two possibilities: Either \- Provide convincing evidence that validates DUSP6 as the key enzyme that downregulates ERK in lung cancer cells (e.g. try more siRNAs to find another that gives strong knockdown, and/or rescue DUSP6 siRNA with DUSP6 from mouse, or with silent mutations in the siRNA target sequence). Or, \- Modify the title and abstract to allow for the possibility that other DUSPs are involved and be more open about the shortcomings of the results. 10.7554/eLife.33718.022 Author response \[Editors' notes: the authors' response after being formally invited to submit a revised submission follows.\] > Summary: > > In this manuscript Unni et al., describe their work on understanding the mechanism leading to cytotoxicity following forced overexpression of KRAS in LUAD cell lines harboring EGFR or KRAS mutations. They conclude that such toxicity owes to hyperactivation of pERK, which occurs through DUSP6. They show that MEK/ERK inhibitor treatment rescues the effect of forced overexpression while reducing the levels of pERK and that DUSP6 mRNA levels are increased in LUAD harboring EGFR or KRAS mutations relative to those with a wild type status of these proteins. Genetic or pharmacologic targeting of DUSP was found to induce pERK (transiently) and lead to diminished cell proliferation in EGFR/KRAS mutant cell lines but not in those with a wild type status. While overall the findings are intriguing, there are gaps in the experimental evidence provided which bias their conclusions. If the authors can address the issues noted below however this could be a very interesting article. The main message of our paper is that p-ERK hyperactivation is intolerable in cancer cells and that this property---the toxic consequences of exceeding a certain level of activated (phosphorylated) ERK---creates a therapeutic target in RAS pathway-mutated cancers: DUSP6, an ERK phosphatase that plays a major role in modulating the activity of ERK in lung cancer cells. We arrived at these findings by studying the mutually exclusive pattern of EGFR and KRAS mutations in lung adenocarcinoma. As your review points out, there may be exceptions to this mutual exclusivity, but we do not believe they undermine our arguments. Nevertheless, we will make changes to the Discussion section to include the observations that the reviewers cite and also note their shortcomings (such as the difficulty of knowing whether normally excluded combinations of mutations have occurred in the same cell or separately in tumor subclones). Of particular relevance, p-ERK intolerance has also been documented in 'drug addicted' cells when inhibitors are removed (Kong et al., 2017; Hong et al., 2018, Das Thakur et al., 2013, Moriceau et al., 2015, Sun et al., 2014). Cells in these conditions appear to violate the mutual exclusivity pattern, however, we would argue that these mutations could have arisen while cells were on drug (Hata et al., 2016). > We have the following suggestions to improve the manuscript. For clarity, we address below the essential revisions related to the results presented in each figure. > > 1\) Figure 1 a) Figure 1A: The expression level of tet-driven ectopic KRAS^G12V^ is very high in each of the three lung adenocarcinoma lines. The authors should demonstrate by western blotting whether these levels are comparable to endogenous KRAS levels found in lung adenocarcinoma lines. Studies from Tyler Jacks on mutant KRAS in knockin mice over 10 years ago emphasized the importance of expression levels of mutant KRAS to mediate transformation in lung adenocarcinoma. Massive overexpression of KRAS^G12V^ can induce senescence rather than transformation in most cell types. Our intention was to force excess RAS pathway activation (beyond what is present in tumor cells) and determine if this is tolerated. The purpose of doing this was to model what might be happening when co-mutations arise in the RAS pathway. We recognize that these levels of RAS may not be commonly experienced by tumor cells and will state this explicitly in the text. Even though our system, like many others, is artificial, it provides an experimental platform for understanding why hyper activation of ERK occurs and allowed us to define a vulnerability (DUSP6 inhibition) that we could exploit. We have now stated this in the text (Results section). b\) Figure 1D: The trametinib/SCH984 experiments in Figure 1D require additional appropriate controls; the authors should show the effect of treatment in the absence of dox stimulation. Also, why are different concentrations of trametinib used? Different concentrations were required to rescue the phenotype in different cells. We will now provide a plot to show dose response (plus/minus dox and plus/minus drug). The degree to which p-ERK is induced in each cell line appeared to require different concentrations of a MEK inhibitor to reset p-ERK back to an acceptable level. A dose response curve for doxycycline plus trametinib is now plotted and shown in Figure 1---figure supplement 1C. > c\) In addition to the biochemistry and cell count data shown in each panel of Figure 1, straightforward studies need to be done to establish the cellular mechanism underlying the cell growth phenotype. Before and after induction of KRAS^G12V^ using dox, studies of cell cycle progression by DNA flow cytometry, for β-GAL expression assays should be done to assess senescence, and cleaved caspase 3, TUNEL and PARP cleavage assays should be done to assess apoptosis. The reduction of cell number 7-days after dox induction of KRAS^G12V^ of \~ 20-40 percent of control could be due to growth arrest, senescence or apoptosis. This will be important to establish with regard to the author\'s interpretation that over activation of ERK leads to synthetic lethality. The need for studies of cellular phenotype in the context of the experiments shown in Figure 1 apply broadly to experiments throughout the paper. We have previously published some of the effects of co-induction of mutant EGFR and mutant KRAS. In that study, we documented apoptosis, autophagy, vacuolization and macropinocytosis in cell lines similar to those we now use (Unni et al., 2015). A recent study (Hong et al., 2018) found that cancer cells that were 'drug addicted' die by apoptosis, parthanatos and pseudosenescence when inhibitors were removed (similar p-ERK overload principle). For these reasons there are likely to be several distinct mechanisms that result in a loss of cell viability. Our goal here has been to focus on the factors that generate the cytotoxic signal, rather than on a cell's response to the signal. Nevertheless, in response to the reasonable concerns raised by this comment, we will assess the extent of apoptosis by measuring cleaved PARP, CASP3 activity, or Annexin V levels to help clarify our statements about cell toxicity. We have measured cleaved PARP in the H358 and H1975 experimental systems described in this manuscript and some assays are shown in Figure 1---figure supplement 1. We characterized mechanisms of cell toxicity in PC9 cells that express both mutant EGFR and mutant KRAS in our earlier paper in *eLife*. d\) Does overexpression of KRASG12C in EGFR or KRAS WT cells also induce toxicity which can be mitigated by co-treatment with trametinib in cells with no activating mutations in EGFR-RAS-ERK pathway? The levels of overexpressed KRASG12C are so high that such levels could induce toxicity by exceeding the upper threshold of RAS-mediated signaling even in cells without hyperactive EGFR-KRAS-ERK signaling. Overexpression of KRAS G12V is likely to result in cell death or senescence in a variety of cell lines, as others have shown in the past. However, the motivation for experiments in Figure 1 was to test the limits of *cancer* cells to activation of the RAS pathway in a reproducible way that could allow us to study the mechanism by which the toxic signals arise. We have added a comment on RAS-mediated senescence in the text (Discussion section). > e\) Since the main claim is that hyperactivation of ERK leads to toxicity, the authors ought to replicate their dox-inducible experiments with active ERK to show that these also lead to cytotoxicity. Also, just because inhibiting MEK/ERK reverses the some of the phenotype in plastic, this doesn\'t exclude that other RAS effectors are also involved (see induction of pAKT in addition to pERK in Figure 1B). The authors should thus carry out similar siRNA or inhibitor studies to demonstrate that the cytotoxic effects of KRAS overexpression are not due to pAKT. Including data that an inducible active ERK2 allele is toxic would be valuable. However, creating and characterizing these lines will take a significant amount of time. We have prioritized other experiments that the reviewers advise that will strengthen our paper. We did show increases in p-AKT at an early time point and it is possible that effectors of RAS other than RAF proteins can also be toxic to cells. To address this point, tetO KRAS G12V cells will be treated with a PI3K inhibitor (to inhibit AKT phosphorylation) and placed on dox. We will document the effects on cell viability, and on p-AKT and p-ERK signaling. We have included data with a PI3K inhibitor (buparilisib) in H358-tetO-KRAS cells (Figure---figure supplement 1D). Using the same cell line, a genome wide CRISPR-Cas9 screen did not reveal an enrichment of guide RNA targeting PIK3CA (Figure 1---figure supplement 1F and Supplementary file 1). > 2\) Figure 2a) Figure 2A: There should be symbols under the heat map denoting which tumors are KRAS mutant and which are EFGR mutant. Are there any tumors with ALK/ROS1/RET activation by translocation? Are there tumors with NF1, MET and BRAF mutations? These should be addressed since most of lung adenocarcinoma have activation of RTK/RAS/RAF pathway (TCGA, Nature 2014) and increased dusp6 is among the five gene signatures to predict poor outcome (Chen et al., 2007). Symbols will be added for mutant KRAS, EGFR, BRAF, NF1, MET, ERBB2 etc. Translocations were not assessed in this data set. The Chen et al., five gene signatures is interesting (*DUSP6, MMD, STAT1, ERBB3* and *LCK*). Perhaps this signature is most common in KRAS mutant tumors, something Chen et al., do not address. We will comment on these observations in the revised manuscript. We have now indicated which tumors are KRAS mutants and which are EGFR mutants in Figure 2A. In addition, we have added another heat map with NF1, BRAF, MET, ERBB2, NRAS and HRAS status indicated as Figure 2---figure supplement 1A. Further, we have compared levels of DUSP6 mRNA among all tumors with RTK-RAS-RAF pathway mutations vs those with only wild type components in this pathway; see Figure---figure supplement 1C. > b\) After describing the emergence of DUSP6 as a key gene responding to activated ERK based on their own studies, the authors should put their findings of DUSP6 elevated expression in the context of previous work. It has been known for over 10 years that DUSP6 is transcriptionally regulated by ERK-responsive ETS transcription factors downstream of MAPK activation and that DUSP6 serves as a major negative feedback regulator of ERK signaling (Reffas et al., 2000; Kawakami et al., 2003; Eblaghie et al., 2003; Li et al., 2007; Ekerot et al., 2008; Furukawa et al., 2008; Jurek et al., 2009). The reviewers correctly point to many papers that highlight the significance of DUSP6 in controlling ERK activity. Our main point in this analysis was to discover which of the prominent negative regulators (DUSPs, SPRYs and SPREDs) have been significantly modulated in lung adenocarcinoma. This revealed that lung adenocarcinoma with mutations in KRAS or EGFR appear to rely on DUSP6 to actively restrain the RTK-RAS-RAF-MEK-ERK pathway. Our work also emphasizes that tumor cells have a level of ERK activation that is *still* subject to negative feedback regulation and that this reliance is a vulnerability based on the data we show in Figure 1. We will mention several of the papers the reviewers cite to properly document the previous work with DUSP6. The papers suggested by the reviewers are cited through two reviews in the text (Discussion section). > c\) Figure 2B and C: For these panels, please show separate data categories for KRAS mutant and EGFR mutant tumors. The box plots shown do not really look statistically significantly different. The medians are close, and the quartiles are largely overlapping. Which statistical test was used to show significant differences? Has a biostatistician been consulted about whether parametric or non-parametric tests would be better for these comparisons? A detailed statistical section is needed in the Materials and methods section, outlining the statistical tests used for the data shown in each figure. We will include a detailed analysis of the statistical tests used and the rationale. We will also separate KRAS and EGFR mutant cases. However, it should be noted that assessing KRAS and EGFR mutant tumors in separate groups will limit sample numbers. As a result, statistical power will be reduced, especially in RPPA assessment where the number of samples available is already limiting. It was for this reason that samples with KRAS and EGFR mutations were pooled. Of note, the box plots in Figure 2B and C are on a log scale, suggesting that the differences seen between the medians are quite large. We have now included more details about the statistical tests used in the methods section. We have also separated EGFR and KRAS mutant tumors and compared each to KRAS/EGFR WT groups and included these data in Figure 2---figure supplement 1B,D). d\) In Figure 2 A-C, the authors must also include the correlation between DUSP6 and P-p38 and p-JNK to show the readers that no correlation exists between them. These data are part of 2H. No correlations were observed. We will also provide plots similar to 2I for p-JNK and p-p38. Correlation plots for P-p38 and P-JNK have been included as Figure 2---figure supplement 1E,F. > e\) Figure 2D: Please provide information or citations about the mouse models used in this figure. We will provide the proper citations of the mouse models used (Politi et al., 2006; Fisher et al., 2001; Felsher and Bishop, 1999). The appropriate citations for the mouse models are now included (subsection "DUSP6 is a major regulator of negative feedback, expressed in LUAD cells, and associated with KRAS and EGFR mutations and with high P-ERK levels"). > f\) Figure 2E: The Material/Methods description about this experiment seems to be missing. Expression levels need to be shown of the transduced onco-proteins by Western blot to verify overexpression. These data were retrieved from a previous publication and not adequately cited (GSE3151, Bild et al., 2006; Kim et al., 2010). We will correct this in the text to clearly state that it is from publicly available data. This citation has been added, and the origin of the data is now clearly stated in the text (subsection "DUSP6 is a major regulator of negative feedback, expressed in LUAD cells, and associated with KRAS and EGFR mutations and with high P-ERK levels"). > g\) Figure 2G and H: Same comments as 2B and C EGFR and KRAS mutant tumors will be assessed separately. However, as mentioned above, combining these genotypes provides greater statistical power. Addressed above and in Figure 2---figure supplement 1. > 3\) Figure 3 a) Figure 3A: Only one siRNA was used for DUSP6 or EGFR and there is no rescue experiment to prove that the effect is due to knockdown of the intended target. DUSP6 de-phosphorylates ERK so ERK phosphorylation is expected to be increased with DUSP6 knockdown. Please explain why Figure 3A shows the opposite with reduced p-ERK after siDUSP6 knockdown. We have used pooled siRNA in the knockdown experiments. We now have PC9 cells expressing wildtype or catalytically inactive DUSP6. These cells will be used to verify that our DUSP6 siRNA is on-target (using siRNA against the 3'UTR of the endogenous *DUSP6* which is not represented in the transgenes). Unfortunately, despite repeated efforts, no siRNAs targeting the 3'UTR proved to be effective at knocking down DUSP6. In addition to using the original siRNA pool, we have now transfected PC9 cells with each individual siRNA comprising the pool, four in total, all of which target the coding region. This revealed a dose-dependent effect of knockdown on growth inhibition: suppression but incomplete knockdown stimulated cell growth, whereas more complete inhibition was toxic to cells (Figure 3---figure supplement 1A,B). This conforms with our hypothesis and suggests the siRNAs for DUSP6 are on target. The challenging aspect of this study was that knockdown of DUSP6 will result in increased p-ERK leading to increased DUSP6 mRNA, which is being targeted by the siRNA ('technical' feedback loop). p-ERK is reduced in this figure probably because the measurement was made on day 5 (see 3B) when the cells are dying. We will include a measure of apoptosis (cleaved PARP) at this time point. We will also provide a longer exposure of the western blot showing the efficiency of DUSP6 knockdown as there could be remaining DUSP6 protein. This may contribute to the decreased p-ERK on day 5. We have assessed cleaved PARP on day 5 after DUSP6 knockdown and shown that it is indeed induced in EGFR mutant H1975 cells but not EGFR/KRAS wild-type HCC95 cells (Figure 3---figure supplement 1C). We conclude that P-ERK levels are low at day 5 after DUSP6 knockdown because cells with increased P-ERK have already become non-viable by this time. This point was further investigated in BCI experiments below. > b\) Figure 3A: Can the toxicity mediated by DUSP6 knockdown in PC9 cells be reduced by co-treatment with trametinib (Figure 3A)? This would help confirm that toxicity caused by depletion of DUSP6 in these cells is due to increased pERK levels. This experiment was tried several times, but we could not find a dose of Trametinib that rescues lethality. MEK and ERK inhibitors are lethal in this line (PC9) and this presents a technical problem: both ERK inhibition and ERK hyperactivation are not tolerated. However, H1975 cells are much more tolerant to ERK inhibition using SCH772984. Experiments are underway to treat H1975 cells that have received siRNA against DUSP6 with an ERK inhibitor like SCH772984 to try and rescue the loss of cell viability. Addition of drug (SCH772984) to cells transfected with siRNA against DUSP6 led to indiscriminate toxicity; cells could not withstand the stress of transfection coupled with an ERK inhibitor. As an alternative strategy, we knocked down DUSP6 with siRNA in PC9 cells co-transduced with shRNAs targeting either ERK1 or ERK2 as used in Figure 1. Stable, viable cells were established with reduced ERK levels. These cells displayed increased relative viability after DUSP6 knockdown compared to shScramble control cells, further suggesting that ERK plays a role in mediating the toxic effects of DUSP6 inhibition (now shown in Figure 3---figure supplement 1G,H,I). > c\) Figure 3B: PC9 cells in this panel show increased p-ERK after DUSP6 kd. Please explain why this is the opposite result compared to Panel A. These data are from 24hr samples, not 5 day samples (Figure 3A). We expect that acute loss of DUSP6 should increase levels of p-ERK that will then be part of a feedback loop of *DUSP6* activation, followed by de-phosphorylation of ERK. As mentioned above, 24 hours after DUSP6 knockdown, cells are still viable and P-ERK is induced whereas, at day 5, cells have induced cleaved PARP and demonstrate substantially decreased viability. We postulate that cells transfected with siRNA for DUSP6 develop high levels of P-ERK and subsequent cell death, leaving only cells with lower P-ERK at day 5. > d\) Figure 3B and C: Did the authors determine if the levels of DUSP6 in their EGFR and KRAS wild-type cell lines, HCC95 and H1648, are not as high as in the cell lines with EGFR or KRAS activating mutations? Lower levels of DUSP6 would indicate that these cells do not need to have similar buffering ability as the EGFR or KRAS mutant cell lines and would be consistent with their findings in Figure 2. In Figure 3B, it seems that DUSP6 is also buffering the pERK levels in HCC95 because knockdown of DUSP6 increases the levels of pERK in these cells by over 1.5 fold, similar to the cells with EGFR or KRAS activating mutations; but this increase in pERK has no impact on the cell survival (Figure 3C). This needs to be reconciled with the authors\' main claim. We will provide the basal levels of DUSP6 across the lines in one figure. Additionally, we will include the quantitation of p-ERK changes from our independent experiments to help establish the fold changes. We have included immunoblots indicating the basal levels of DUSP6 and P-ERK across the panel of cell lines (Figure 3---figure supplement 1D). Due to the variability associated with transfection-based experiments, we have now compiled dosimetry for three independent western blots in Figure 3B and plotted the results. This revealed that EGFR or KRAS mutant, but not wild type, cells consistently demonstrate increased P-ERK upon DUSP6 knockdown. All these results and their potential implications are now described in the text (subsection "Knockdown of DUSP6 elevates P-ERK and reduces viability of LUAD cells with either KRAS or EGFR oncogenic mutations"). e\) There seems to be a disconnect between the first part of the manuscript, where the authors describe the mechanism of toxicity caused by forced overexpression of KRAS and the second part, where they describe the effect of DUSP inhibition in cell lines with endogenous KRAS/EGFR mutations. To close this gap, the authors need to determine the levels of DUSP6 protein in their dox-inducible KRAS models and how they correlate with pERK. They should determine if the induction of pERK is transient, as that observed with DUSP6 siRNA, or sustained. Finally, they should use DUSP6 siRNA or BCI to determine if this reverses the effect of forced KRAS expression on pERK or proliferation. Another question that has not been addressed is why doesn\'t forced overexpression of KRAS induce sufficient DUSP6 to override the induction of pERK, if DUSP expression is under the control of EGFR/KRAS? Could there be other factors involved? If the authors can experimentally address these it would be a significant improvement. We don't understand why the reviewers believe there is a "disconnect" between the early and late phases of the manuscript. In fact, we believe that there is a logical flow from identification of p-ERK as the locus that transmits a toxic signal to the implication of DUSP6 as a critical regulator of the activity of ERK. So the notion of an informational gap is not clear to us. Nevertheless, we agree that the situation is complicated by the kinetics of activation and de-activation of the components of the signaling system, and we will follow the request to obtain more kinetic data. We will perform time course experiments in tetO-RAS lines, measuring p-ERK induction and DUSP6 protein levels at 1,3,5 and 7 days. Contrary to the reviewers' speculation, we would not anticipate that DUSP6 siRNA or BCI would reverse the effects of forced KRAS expression; they should potentiate the effects of KRAS. On the other hand, it is unclear why DUSP6 cannot 'override' the induction of p-ERK. It is possible that p-ERK has fully localized to the nucleus and is no longer accessible to DUSP6. We will consider these possibilities in the revised manuscript. The kinetics of p-ERK induction have been provided for H358 (days 1, 3, 5 and 7) and H1975 (day 7) cells in Figure 1---figure supplement 1B). The time course of induction of p-ERK upon treatment of H358 cells with BCI (at 1, 6, 12, 24, 48, 72 hours) is provided in Figure 4---figure supplement 1D. Experiments with BCI provide initial assessment of the kinetics of induction of p-ERK upon DUSP6 inactivation. The kinetics suggest that p-ERK induction for at least 24 hours is required before markers of apoptosis (PARP cleavage) are detected. Similar kinetics of p-ERK induction (at least 24-48 hours) before PARP cleavage detection were observed for H358-tetO-KRAS cells (subsection "Synthetic lethality induced by co-expression of mutant KRAS and EGFR is mediated through increased ERK Signalling", subsection "P-ERK levels increase in LUAD cells after inhibition of DUSP6 by BCI, and P-ERK is required for BCI-mediated toxicity."). > f\) More evidence is needed to show that the increase in DUSP mRNA is associated with an increased in DUSP6 protein levels or increased DUSP activity. The data in Figure 2I, where the authors compare the relationship between pERK (measured by RPPA) and DUSP6 mRNA is the only such evidence provided. This is underwhelming because the correlation is not great (r=0.1) and because the pERK does not seem to have been controlled for total ERK. The RPPA assays (from TCGA) are controlled for total protein. The time course studies in tetO lines and BCI-treated cell lines may help address the issue of 'sufficient' levels of DUSP6. We have performed and included in the manuscript time course experiments for both dox treatment in TetO cell lines and BCI treated cells as described above. > g\) Finding that DUSP6 siRNA caused only a transient elevation in pERK (followed by inhibition of pERK at longer intervals), while mimicking the antiproliferative effect observed with forced overexpression of KRAS needs to be clarified with additional experiments. As it stands, it is difficult to agree with the authors conclusion that DUSP6 is the main mediator of the proliferative effect or that the effect of DUSP6 is through pERK. One consideration may be to use constitutively active ERK (i.e. DD phosphomimetic mutants) and attempt to reverse the effect of DUSP6. This is another point that, if addressed experimentally, could add value to the manuscript. The 'transient elevation' in p-ERK with siRNA against DUSP6 may be a technical limitation of this assay as previously described. The time course studies in tetO lines and with BCI will help establish this. We will try and rescue the effects of siRNA against DUSP6 and BCI in H1975 cells---a cell line that is tolerant to ERK inhibition and provides an experimental system to 'dial' back appropriate p-ERK levels. A constitutively active ERK mutant is likely to be lethal in the presence of tet-induced mutant KRAS; for instance, ERK mutants have been documented to have a lethal effect in melanoma cells (Goetz et al., 2014). We will focus our attention on H1975 cells to rescue the effects of DUSP6 siRNA and BCI w/ MEK or ERK inhibition. As mentioned above, we have used shRNA to inhibit ERK1 or ERK2 in PC9 cells and inhibition of ERK1 or ERK2 limited the toxic effects of knocking down DUSP6 (Figure 3---figure supplement 1G,H,I). In addition, we observed similar reductions in the toxic effects of BCI when H358 cells were co-treated with an ERK inhibitor (Figure 4---figure supplement 1E). These experiments further reinforce the conclusion that inhibition of DUSP6 (genetically or pharmacologically) decreases viability of EGFR or KRAS mutant cells -- at least partially -- through ERK induction. > h). If indeed it is true that a transient induction in pERK leads to cytotoxicity several days later, then does EGF stimulation (which also causes a transient induction in pERK) have the same effect in EGFR WT/KRAS MT cells? As previously mentioned, the effects of transient vs. prolonged p-ERK will be addressed by studying the time course of response to BCI in cells and to mutant KRAS induced by dox. Transient treatment of cells with EGF is not expected to cause cytotoxicity based on our earlier experiments. In fact, transient administration of EGF to HCC95 cells failed to shift the IC50 for BCI; only prolonged exposure to EGF did that (Figure 5A). We will consider inclusion of this information during revision of the manuscript. We have provided time course experiments of dox-mediated induction in TetO-KRAS cells (Figure 1---figure supplement 1B) and of BCI treatment in H358 cells (Figure 4---figure supplement 1D). Both approaches caused an initial increase in P-ERK levels coupled with later induction of cleaved-PARP and subsequent decrease in P-ERK (Figure 4---figure supplement 1D). > i\) All three reviewers noted that HCC95 (RRID:CVCL_5137) is a squamous cell carcinoma line. This not comparable to lung adenocarcinoma because this tumor type has different pathway dependencies than LUAD (TCGA, 2012). A wild type LUAD line should be tested instead. HCC95 cells were used because they are a cancer cell line (lung) that does not have mutations in EGFR or other examined components of the RAS pathway. We will state in the text that this is a squamous lung cancer cell line. Based on our data, cells with a mutation in the RAS pathway are likely be vulnerable to DUSP6 inhibition, regardless of cell lineage. Cells without these mutations (like HCC95) illustrate cancers (of any origin) that we would predict to be un-responsive to DUSP6 inhibition. We have noted the nature of HCC95 cells in the text (subsection "Knockdown of DUSP6 elevates P-ERK and reduces viability of LUAD cells with either KRAS or EGFR oncogenic mutation"). > j\) Figure 3C: Same comments as 3A. Addressed above. > k\) The investigators might consider inactivating DUSP6 with CRISPR-Cas9 in these cell lines to show genetic dependence. We have now created PC9 and H358 cells in which *DUSP6* has been deleted or damaged using CRISPR-Cas9. We anticipate that the cells selected for this loss may have developed other mechanisms to maintain p-ERK or may have other pathways activated to bypass the need for p-ERK to mediate survival. Thus, they may or may not be uniquely vulnerable to inhibition of MEK or ERK. We will analyze the recently identified mutant cells for p-ERK levels, growth rates, and sensitivities to BCI and to inhibition of MEK and ERK. We created H358 cells deficient in DUSP6 (Figure 4---figure supplement 1J,K). These cells were equally sensitive to BCI as cells that were targeted with a control (lacZ) sgRNA. We suspect that DUSP1 may be controlling p-ERK levels in the absence of DUSP6, and BCI has known specificity towards DUSP1 in addition to DUSP6. Additionally, these cell lines were derived from clones, so it is possible that new mutations or pathway re-wirings have taken place and that they continue to control p-ERK. > 4\) Figure 4 a) Figure 4A: 11 lung cancer cell lines were treated with BCI and viable cells were measured after 72-hour treatment. Most of these cell lines are fast growing cells with doubling time around 20\~30 hours. Reduction of viable cells by 72-hours does not necessarily mean cell killing. TUNEL or caspase-3/PARP western blot needs to be performed to detect the levels of apoptosis. Senescence and cell cycle arrest should be examined too. We agree that the reduction in numbers of viable cells does not reveal the mechanism by which the numbers were reduced, and we have been careful to avoid any suggestion that it does (e.g. by labeling our charts "number of viable cells"). As explained earlier, our focus has been on the generation of the toxic signal, not the response to it. Nevertheless, in response to the reviewers' concerns on this point, we will examine cells for apoptosis by measuring PARP cleavage. We have assessed cleaved-PARP after BCI treatment in a panel of seven sensitive and insensitive cell lines and in a time course experiment in H358s (Figure 4---figure supplement 1). Only sensitive cell lines induced cleaved PARP after treatment. > b\) Figure 4C and D: p-ERK levels at a single time point was measured in multiple cell lines (30 min). In order to explain the loss of viable cells at 72hours in Figure 4A, p-ERK should be measured at serial time points such as 1, 6, 12, 24, 48, 72 hours. We will provide a time course of our measurements of p-ERK and DUSP6 in H1975 and H358 cells during treatment with BCI. We provide a time course for H358 cells treated with BCI at the time points suggested by the reviewer in Figure 4---figure supplement 1. > c\) The data on BCI are very interesting but it\'s not clear how have the authors established that BCI is selective for DUSP? Sensitive and insensitive cell lines all have IC50s in the 1-5μM range (12/13 cells lines tested). What is the IC50 of the inhibitor in DUSP6 KD or KO cells (or in DUSP1/6 double KD/KO cells)? How do the authors know that the antiproliferative effect of BCI is due to its ability to induce pERK? These questions need to be addressed experimentally. The authors should also attempt to show that the inhibitor has an effect in vivo, given their claim that such an approach could be therapeutically beneficial in patients. A potential target of BCI is DUSP1, but we have data showing that siRNA against DUSP1 is not lethal in H1975 cells while siRNA to DUSP6 is. We will now also try to rescue the effects of BCI in H1975 cells by inhibiting ERK, using this cell line for the reasons described above (3b,g). In addition to PC9 CRISPR-Cas9 DUSP6 knockout cells described above (Figure 3, comment k), we have generated PC9 cells expressing wild type and catalytically inactive DUSP6. We will use these cell models to determine if there is a shift in the BCI IC50 with manipulation of DUSP6 to further evaluate its role as the biological target of BCI. Importantly, we have now completed a genome-wide CRISPR screen in H460 cells (an KRAS mutant cell line sensitive to BCI), looking for loss of function mutations that confer resistance to BCI. The most highly enriched guide RNA in this screen was specific for KRAS, suggesting that BCI is 'on-target' with respect to its proposed role in causing excessive activation of the RAS pathway. We will confirm these findings by measuring BCI sensitivity in H460 cells treated with siRNA against KRAS and expect to include a version of the results in the revised paper to further support our interpretation of our findings with BCI. We have included data showing that DUSP1 knockdown, as opposed to DUSP6 knockdown, is not lethal in H1975 cells (Figure 4---figure supplement 1A,B). Further, we have demonstrated that co-treatment with an ERK inhibitor decreases the toxic effects of BCI in H358 cells (Figure 4---figure supplement 1E,F). Lastly, we added the genome-wide CRISPR screen data in H460 cells showing that sgRNAs for KRAS are enriched upon BCI treatment, which we have subsequently confirmed using individual sgRNAs and BCI dose response experiments. Together, these results confirm that BCI works mainly through DUSP6 in the context described and mediates its toxic effects through ERK. > d\) Figure 4E and F: Four cell lines were treated with 125nM trametinib for 72 hours. 125nM is a very high dose for trametinib-sensitive cell lines such as H358 (reported IC50 \~50nM). Apoptosis levels should be measured to document any cell death. The purpose of these experiments was to address whether there was a correlation between the sensitivity of cells to ERK inhibition *and* their sensitivity to ERK hyperactivation. This correlation appears to hold (PC9 and H358 vs. HCC95 and H1648). > 5\) Figure 5. > > See comments for Figure 3B. HCC95 is a SQCC cell line that is quite different from the LUAD cells used in the rest of this manuscript. > > a\) In figures where the authors make a comparison of protein levels under different conditions the uncut blot with all the lanes on the same blot must be shown. It is not ideal to make a comparison between levels of proteins on two different strips of blots. For example, in Figure 5B, authors claim that EGF increase levels of pEGFR and pERK in HCC95 cells when the pEGFR and pERK levels with and without EGF are on two different strips of blots. This is important to confirm that BCI treatment increases pERK levels in EGFR treated cells. Quantifications should be shown in addition. Data in the western blots are normalized to total ERK and actin to make the values comparable. This will be explicitly stated in methods. Additionally, samples grown with and without EGF will be run in the same gel for the highest dose of BCI to provide a visual comparison. We have included a western blot using extracts of cells treated and not treated with EGF, run on the same gel in Figure 5---figure supplement 1. > Overall comment: > > The authors should soften earlier claims that EGFR-mutations and KRAS-mutations are synthetically lethal in the adenocarcinoma subtype of NSCLC (Unni et al., 2015). Recent genetic studies of EGFR-mutant lung cancer (Blakely et al., Nature Genetics 2017) have shown that 2.5%\~4.7% EGFR-mutant lung cancer also harbor KRAS copy number gain or activating mutations. Tumor genomic analyses have indicated that bona fide driver mutations causing lung adenocarcinoma are not as mutually exclusive as previously thought (e.g. PMIDs: 25301630, 28498782, 28445112, 29106415, and other recent publications), particularly in metastatic lung adenocarcinoma instead of early-stage disease and/or during the evolution of treatment resistance in metastatic disease. Further, preclinical studies have shown acquired KRAS gain/activation in response to EGFRi in EGFR-mutant NSCLC, indicating that this can be a mechanism of drug resistance (Politi et al., 2000; Eberlein et al., 2015). The authors should acknowledge these recent works and temper their claim of absolute mutually exclusivity in this disease, at least under these more advanced-stage disease contexts and in the light of the emerging literature. As noted earlier, we will provide a more thorough discussion of these points in the manuscript. The additional discussion of these issues has been added to the text (Discussion section). We hope that with these changes and additions to the manuscript, the revised version will suitable for re-submission and consideration for publication at *eLife*. \[Editors\' note: further revisions were requested prior to acceptance, as described below.\] > Thank you for resubmitting your work entitled \"Hyperactivation of ERK by mutation-driven RAS signaling or by inhibition of DUSP6 is toxic to lung adenocarcinoma cells\" for further consideration at eLife. First, let me apologize profusely for the very long time this has taken. The problem is basically that the reviewers are at an impasse and could not decide on a course of action. > > Briefly, this is a follow up to a previous paper, that reported anti-proliferative effects of activation of Ras and EGFR in the same cancer cells. This new paper has provides evidence that the anti-proliferative effect of over-expressed, active Ras is due to ERK hyperactivation, and that DUSP6 is critical to prevent adverse effects of active ERK in lung cancer cells. In the revision, the authors have added results using ERK shRNA, ERKi, and a CRISPR screen that together provide convincing evidence that ERK mediates the antiproliferative effect of oncogene overexpression. The finding that sgKRAS are enriched in the BCI screen also indirectly supports that conclusion. The authors have also ruled out that the PI3K pathway is not involved by adding PI3K inhibitor experiments. These additions greatly strengthen the conclusion that hyperactivation of ERK is detrimental. However, the evidence that DUSP6 plays a key role is not convincing. > > The DUSP6 knockdown was done with a single siRNA with no rescue experiment. (Other siRNAs did not effectively inhibit DUSP6 expression). Attempts to recapitulate the result with a DUSP6 CRISPR knockout were unsuccessful. Therefore, the conclusion that DUSP6 is necessary relies in large part on the specificity of the chemical BCI. The DUSP6 KO cells have the same IC50 to BCI, while DUSP1 siRNA did not affect proliferation in their system. Based on these results, the authors cannot claim that the BCI effect is DUSP6 specific nor that the BCI effect in DUSP6KO cells is driven by DUSP1 inhibition. It probably isn\'t, and by inference, the observed phenotype is not dependent on DUSP6 alone but on other ERK specific DUSPs as well. > > The reviewers disagreed whether these weaknesses undermined the impact to the point where the paper was unsuitable for publication, or whether lengthy additional experiments would be needed. The eLife approach is not to ask for multiple rounds of revision. In this spirit, I suggest two possibilities: > > Either > > \- Provide convincing evidence that validates DUSP6 as the key enzyme that downregulates ERK in lung cancer cells (e.g. try more siRNAs to find another that gives strong knockdown, and/or rescue DUSP6 siRNA with DUSP6 from mouse, or with silent mutations in the siRNA target sequence). > > Or, > > \- Modify the title and abstract to allow for the possibility that other DUSPs are involved and be more open about the shortcomings of the results. 1\) To test additional DUSP6 siRNAs for their effects on protein abundance and cell fitness, we obtained a DUSP6-specific siRNA from Qiagen (the previous siRNAs were prepared by Dharmacon). In contrast to the DUSP6-8 siRNA that targeted a sequence in the 3' domain of DUSP6 mRNA, the new species (called DUSP6-Qiagen in Figure 3B,C) targeted a sequence in the 5' coding region of DUSP6 mRNA and reduced levels of DUSP6 protein in PC9 cells to levels similar to those achieved with one of the previously tested Dharmacon siRNAs (DUSP6-8 in Figure 3B and Figure 3---figure supplement 1A) and with the pool of four Dharmacon siRNAs (DUSP6-pool in Figure 3B and Figure 3---figure supplement 1A). Furthermore, DUSP6-Qiagen reduced the number of viable PC9 cells to a level similar to that observed with the pooled Dharmacon siRNAs (Figure 3C). We have described the effects of this second effective inhibitory RNA in the text and conclude that it strengthens the case for a central role of DUSP6 in regulation of ERK activity in RTK-RAS-driven LUAD. We also point out that this conclusion is supported by the correlation between the effects of one Dharmacon siRNA (DUSP6-8) on both DUSP6 protein levels and cell fitness (Figure 3---figure supplement 1A,B). 2\) We also attempted, unsuccessfully, to rescue the effects of DUSP6 siRNA by generating a plasmid encoding *DUSP6* mRNA with several synonymous mutations in the coding sequence to render the mRNA target sequence resistant to the siRNA without changing the protein sequence. For a variety of technical reasons related to transfection procedures, we have not been able to perform these experiments in a reproducible manner. We are convinced that the work required to carry out a satisfying rescue experiment would take an unreasonable amount of time and inappropriately delay publication, when we have provided the requested data with a second effective siRNA. 3\) Despite our positive findings with the Qiagen siRNA, we recognize that our conclusions about the role of DUSP6 in regulation of the activity of ERK kinases should be cautious. (DUSP6 may not be the only important regulator and we cannot fully exclude some off-target effects of our siRNAs.) We have therefore removed specific mention of DUSP6 in the title of the manuscript, and we have modulated the description of the results in the abstract, along the lines suggested in your letter. One other relevant item: two recent papers confirm the significance of the level of ERK kinase activity in another cancer type, melanoma, and address the possible role of DUSP6. Leung et al., over-express *ERK2* in melanoma cell lines and show that high levels of ERK2 protein are toxic specifically in lines that carry BRAF V600E. Wittig-Blaich et al., use a complex screening method to identify genes that produce a synthetic lethality when disrupted in melanoma cell lines carrying the BRAF V600E mutation; one of the five implicated genes is *DUSP6*, allowing the authors to draw conclusions similar to our own. We mention and cite these papers (Leung et al., 2018 and Wittig-Blaich et al., 2017) in the Discussion section. In addition to the changes that address your main concerns, we have found a few places in the text that lacked clarity upon careful re-reading of our previously submitted revision. [^1]: These authors contributed equally to this work. [^2]: These authors also contributed equally to this work.

Introduction ============ Detection and adequate response to nonself is essential for survival and development in all multicellular organisms. An important part of the innate immune detection in plants and animal lineages is ensured by a class of signal transducing proteins known as NB-LRR proteins in plants and nucleotide-binding oligomerization domain (NOD)-like receptors (NLRs) in animals ([@evu251-B61]). Plant NB-LRR proteins sense the presence of fungal, oomycete, nematode, bacterial, or viral pathogens and trigger an immune response in the form of a localized cell death reaction termed the hypersensitive response ([@evu251-B45]; [@evu251-B42]). NB-LRR proteins represent the resistance proteins involved in effector-triggered immunity as they sense strain-specific pathogen effectors or the modification of self, induced by these effectors. Plant genomes encode large repertoires of NB-LRR proteins with up to several hundred members. NB-LRR genes are typically highly polymorphic between individuals and subject to positive diversifying selection resulting from the host-pathogen arms race. Animal NLRs, in turn, are activated by relatively invariant MAMPs (microbe-associated molecular patterns) and at least in mammals, the number of NLRs is more limited than in plant genomes ([@evu251-B49]; [@evu251-B82]). Animal NLRs and plant NB-LRR receptors are collectively designated NLRs and are members of the family of STAND proteins (signal-transducing ATPase with numerous domains), ([@evu251-B58]; [@evu251-B23]). These proteins typically comprise a central nucleotide binding and oligomerization domain (NOD) linked to an N-terminal effector domain and a C-terminal domain composed of superstructure-forming repeats such as LRR, WD, HEAT, ANK, or TPR motifs. One can distinguish two main classes of NOD domains: The NACHT (named after the NAIP, CIITA, HET-E and TP-1 proteins) and the NB-ARC domain. In general, plant NB-LRR proteins display an NB-ARC NOD domain whereas animal NLRs display a NACHT domain, although many instances of NB-ARC STAND proteins are described also in animal lineages. In most cases, the C-terminal domain of plant and animal NLRs corresponds to a LRR domain, but other types of repeat domains have been reported for instance in fish and marine invertebrates such as *Hydra* and the coral *Acropora digitifera* ([@evu251-B95]; [@evu251-B56]; [@evu251-B37]). The N-terminal effector domains are variable and either correspond to coiled-coil or Toll/interleukin-1 receptor (TIR) domains in plants ([@evu251-B42]), whereas CARD, BIR, PYD, death domain (DD), and DED are found in animals ([@evu251-B64]). In addition to these domains, a variety of other N-terminal domains, sometimes restricted to a given phylum, has been reported ([@evu251-B37]; [@evu251-B107]). In spite of the remarkable overall resemblance between these immune receptors in plant and animal lineages, it is unclear if this similarity is the result of evolutionary conservation ([@evu251-B5]; [@evu251-B61]; [@evu251-B106]). It has been proposed that build-up of NLRs is the result of convergent evolution by association of a limited set of preexisting domains such as NOD and LRR domains. Remarkably, NLRs appear not only to be involved in the immune response to pathogenic nonself, but an emerging trend reveals that these receptors may also control other forms of biotic interactions, for instance between animal hosts and their symbiotic microbiome ([@evu251-B21]). With an estimated 5.1 million species, the fungal kingdom represents a major eukaryotic lineage and a sister group of the holozoa ([@evu251-B8]; [@evu251-B38]). Because of their overall organization, most cells in fungal organisms are in direct contact with their biotic environment. In addition to a variety of pathogenic and symbiotic interactions, fungi are also exposed to diverse adverse biotic interactions as hosts of a variety of pathogens and parasites such as mycoviruses, mycophagic bacteria, mycoparasitic fungi, and grazing nematodes ([@evu251-B59]; [@evu251-B74]; [@evu251-B9]; [@evu251-B26]; [@evu251-B81]). In the recent years, the awareness for the existence and importance of fungal nonself recognition and defense systems is gradually increasing. Based on the common central role for STAND proteins as intracellular innate immune receptors in plant and animals, it is not unreasonable to suppose that STAND proteins may play similar roles in fungi. And indeed, there is evidence for the involvement of STAND proteins in the detection of nonself and the control of programmed cell death in fungi, thus stressing the analogy between animal and plant NLRs. The HET-E protein of *Podospora anserina*, one of the founding members defining the NACHT domain, is involved in a fungal nonself recognition and programmed cell death process termed heterokaryon incompatibility ([@evu251-B83]; [@evu251-B53]). Incompatibility is triggered when genetically distinct individuals belonging to the same fungal species undergo cell fusion and corresponds to a pleiotropic cellular response culminating in the programmed cell death of the fusion cell ([@evu251-B76]; [@evu251-B7]). HET-E has a tripartite domain organization typical of STAND proteins, with a central NACHT domain, a C-terminal WD40 repeat domain and an N-terminal HET domain. The HET domain is found in different proteins involved in fungal incompatibility and corresponds to a death effector domain ([@evu251-B91]; [@evu251-B69]). HET-E is part of a larger gene family comprising ten members, termed NWD genes. Five of these proteins also comprise an N-terminal HET domain and two of those correspond to genetically identified incompatibility genes (HET-D and HET-R) ([@evu251-B68], [@evu251-B71]; [@evu251-B17]). The five other members display different N-terminal domains. The WD repeat regions of the members of the gene family are hypervariable. The repeats show a high level of internal repeat conservation, and are undergoing concerted evolution, meaning that repeat shuffling and exchanges occur both within and between members of the gene family ([@evu251-B71]; [@evu251-B18]). In addition, the repeat region is subjected to positive diversifying selection operating specifically on four amino acid positions of each individual repeat, which map to the protein--protein interaction surface of the WD-repeat β-propeller structure. Another member of the gene family, termed NWD2, shows an N-terminal domain homologous to the prion-forming domain of the HET-s prion protein of *P.anserina*. It is proposed that NWD2 acts as an activator of the HET-S pore forming toxin by triggering transconformation of its prion-forming domain and subsequent activity of the HeLo toxicity domain ([@evu251-B34]; [@evu251-B24]; [@evu251-B84]; [@evu251-B88]). This mode of signal transduction between a STAND protein and an trans-acting effector domain was proposed to be widespread in fungi and in addition to the \[Het-s\] prion-forming motif, two additional prion-like motifs (termed σ and PP) have been described ([@evu251-B24]). These motifs were found as N-terminal domains of STAND proteins of various types such as NACHT-WD, NACHT-ANK, or NB-ARC-TPR proteins. It was recently shown that the \[Het-s\] prion domain, and the N-terminal prion motif of NWD2 can functionally replace the PYD region in NLRP3-mediated CARD activation ([@evu251-B14]). Involvement of STAND proteins in incompatibility is not restricted to *P. anserina*, as the *vic 2* and *vic 4* loci of the chestnut blight fungus *Cryphonectria parasitica* were found to encode STAND proteins ([@evu251-B20]). Although fungal STANDs have been initially identified in the context of heterokaryon incompatibility (conspecific nonself recognition), it appears that the role of fungal STAND proteins is not limited to heterokaryon incompatibility, as the number of STAND-encoding genes greatly exceed the number of incompatibility genes. There are several reports indicating that STAND proteins are polymorphic and rapidly evolving and subject to extensive expansion in paralogous gene families in a variety of fungal species ([@evu251-B30]; [@evu251-B62]; [@evu251-B12]; [@evu251-B54]; [@evu251-B108]; [@evu251-B41]; [@evu251-B101]). In *Tuber melanosporum*, an expanded *nank* (NACHT ANK) family is, in addition, characterized by a remarkable diversification mechanism based on alternative splicing of multiple codon-sized microexons ([@evu251-B41]). Based on the similarity between fungal STAND proteins and plant and animal NLRs and their involvement in nonself recognition and programmed cell death, we have proposed that STAND protein may also correspond to general nonself receptors in fungi ([@evu251-B70]). This proposed function could account for their high level of polymorphism and rapid diversification, and their expansion in certain species critically depends on interorganismal interactions. Although the genomics of NLRs in plant and animal species and lineages has been the subject of many studies, the overall distribution and organization of NLR-related genes in the fungal phylum has not been investigated systematically to date. The fungal phylum offers the advantage of an extensive genomic coverage with several hundred completed genomes currently available ([@evu251-B35]). Herein, we have analyzed 198 complete fungal genomes (corresponding to 164 different species) for the presence of NLR related proteins. We report on the NLR domain architecture, variability and repertoire size in these 164 fungal species. We find evidence of extensive variation of NLR copy numbers both within and between species. Several NLR domain architectures appear presently restricted to the fungal phylum, whereas others also exist in animal or plant lineages. NLRs appear restricted to filamentous species and are missing from yeast genomes, suggesting that presence of NLRs is associated with multicellularity. Our data suggest an extensive modularity of domain associations, with recurring inventions of domain architectures. Finally, a proportion of the C-terminal domains of NLRs show strong internal conservation, as described for the rapidly evolving HNWD family of *P.anserina*. We find evidence for positive diversifying selection acting on C-terminal domains of the TPR and ANK type, as previously reported for the WD repeats. This overall picture of NLR protein repertoire in fungal genomes now highlights similarities and differences between nonself recognition strategies in different eukaryotic lineages and sheds new light on the evolutionary history of this type of receptors. Materials and Methods ===================== Identification -------------- IR and functionally validated (FV) queries were obtained by extraction of NACHT and NB-ARC domains from the full-length sequences according to PfamA PF05729.7 and PF00931.17 profile matches ([@evu251-B31]). PSI-BLAST searches ([@evu251-B2]) with three iterations and an *E* value cut-off of 10^−5^ were carried independently for each query sequence on the NCBI "nr" database (June 27, 2013), and then combined. The candidate set was pruned from sequences with multiple disjoint matches to the queries and from very short sequences (below 100 amino acids). Then it was limited to sequences from complete or draft whole-genome sequencing and resequencing projects, according to Genome OnLine Database (\[[@evu251-B66]\], as of September 18, 2013), for which at least 2,000 sequences were available in the nr database. Intrastrain identical copies of sequences were removed, whereas interstrain identical sequences were kept. Boundaries of the NB domain were determined as the longest stretch of matches from all NACHT and NB-ARC queries in the PSI-BLAST search. Proportional Venn diagrams were generated using BioVenn ([@evu251-B40]). Noncanonical P-loop variants were detected by inspecting single-residue changes at the four conserved positions of the motif in multiple sequence alignments of NB domains generated by Clustal Omega 1.1.0 ([@evu251-B89]), with two iterations separately for nonredundant sets of 4596 NACHT and 1174 NB-ARC STANDs found in the entire nr database (not limited to whole-genome projects). Annotation ---------- In-house signatures were generated using HMMER 3.0 ([@evu251-B28]) for the HET-s, PP, and σ prion-forming domains, and the NAD1, Goodbye, HeLo-like, sesA, and sesB domains. Representative sequences of prion-forming domains were aligned using several tools: ClustalW 2.1 ([@evu251-B57]), ClustalOmega 1.2.0 ([@evu251-B89]), Mafft 7.029b ([@evu251-B51]), and Muscle 3.8.31 ([@evu251-B29]). The best alignments in terms of the normalized Median Distance (norMD, \[[@evu251-B99]\]) were used for the HMM training with default parameters ([@evu251-B27]). A single representative sequence for each nonprionic domain was submitted to the HHsenser web tool (PSI-BLAST parameters: *E* value cut-off of 10^−3^, coverage of hits at least 50% \[[@evu251-B94]\]) to build a data set including at least 500 sequences in the "permissive" alignment. "Strict" alignments were retrieved and used in iterative HMM training. After each round of the training, sequences with the score below 25.0 or the score/bias ratio below ten were excluded from the alignments; the procedure was repeated until convergence. Finally, the in-house HMMs were included in a PfamA-style repository with their sequences and domain thresholds set to 25.0. STAND sequences were scanned using PfamA and in-house signatures. Particular annotation was attributed to a given domain if the HMM profile match was entirely contained within domain boundaries extended by a 20 residue-wide envelope. In the case of overlapping annotations from the same PfamA clan, the hit with the lower *E* value was chosen, except for the P-loop NTP-ase clan (CL0023), where NACHT or NB-ARC annotations were always preferred (if above the PfamA threshold). Annotations from repeat-containing clans: Ankyrin (CL0465), Beta propeller (CL0186), and TPR (CL0020, includes HEAT repeats) were merged to three main categories: ANK, WD40, and TPR, respectively. Highly overlapping N-terminal annotations, as well as three prion-forming domain annotations were merged (see Results). Conflicting annotations from PfamA HeLo and in-house HeLo-like signatures were resolved in the favor of the former. Numerical suffixes of signature names were truncated and sequential occurrences of identical annotations were squeezed. Domain associations were visualized using the graphviz package ([@evu251-B32]). Distribution of domain architectures was quantified by means of paralog and ortholog hits. Ortholog index counted number of species (distinguished by binomial name) in which a given architecture was found. Paralog index summed the number of sequences with a given architecture in all species (the average number was added if several strains were sequenced for the particular species). Phylogeny --------- All phylogenetic trees were calculated through the maximum likelihood estimation based on alignments of NB or N-terminal domains extracted from nonidentical sequences. In each case, the best alignment was selected according to the norMD score out of alignments generated by the same MSA tools as above. Then, the alignment was pruned from columns with more than 50% of gaps (using trimAl \[[@evu251-B15]\]) and submitted to PhyML 3.0 ([@evu251-B36]) with default options (model LG, tree topology search NNI). Interstrain identical sequences were added to the trees after estimation. Phylogenetic trees were drawn using the R project with the "ape" (version 3.0-8, \[[@evu251-B72]\]) and "phangorn" (version 1.7-4, \[[@evu251-B86]\]) packages, and the TreeDyn editor ([@evu251-B19]; [@evu251-B25]). Genes with no clear ortholog in all ("orphans") or in some other strains ("semiorphans") from the same species were identified according to co-phenetic distances between leaves in multistrain phylogenetic trees (using the R package ape). To detect highly homologous pairs of NB domain sequences associated with different N-terminal domains, BLASTP scores ([@evu251-B2]) were calculated in the all-against-all manner for the entire data set of NB domains. A target was counted as highly similar to the query if the match scored at least 99% of the maximum score obtained by the query. To avoid false positives, only matches with at least 80% identity over 100 or more amino acids were counted; sequences with unknown N-terminal annotation were also excluded. Repeat Domain Analysis ---------------------- Highly internally conserved repeats were detected using T-reks ([@evu251-B46]) with customized parameters (PSIM = 0.85, kmeans = 10, overlapfilter on, external MSA: ClustalW 2.1, and Muscle 3.6); repeat regions shorter than 100 amino acids were filtered out. Sequences from dikarya, metazoan, and viridiplantae belonging to ANK, WD40, and TPR clans were extracted according to designation in the Pfam repository (27.0) and availability in the nr database (June 27, 2013). The content of highly internally conserved repeats was calculated as above. "Skipredundancy" from the EMBOSS package ([@evu251-B80]) was used to obtain the nonredundant count of the highly conserved repeats. For analysis of *P. anserina* ANK and TPR motifs, genes-encoding STAND proteins with individual repeats displaying over 85% internal conservation were analyzed, excluding *hnwd* family members, resulting in a set of ten genes. They code for TPR, ankyrin, or HEAT repeats. For each gene, the repeat-encoding DNA was polymerase chain reaction (PCR)-amplified from five wild isolates from the Wageningen collection (Wa94, Wa 96, Wa97, Wa99, and Wa100) ([@evu251-B100]), gel-purified, cloned in the XL-PCR-TOPO plasmid (Invitrogen, Life Technologies) and sequenced. Sequences were manually assembled before further analysis. In addition, sequences from the S strain were extracted from the *P. anserina* genome sequence and were added to the data set. Protein repeats were identified using RADAR ([@evu251-B39]). Individual repeats were then aligned using ClustalW and a neighbour-joining tree constructed using MEGA5 ([@evu251-B97]). Sequences clustering together with a high bootstrap support were analysed further, and the other were discarded from the data set. For each data set, sequences in duplicates were then discarded, so that a single copy of each repeat sequence was maintained. To detect signs of positive selection, five analyses (SLAC, FEL, REL, MEME, FUBAR) were conducted for each data set using the HYPHY suite ([@evu251-B77]; [@evu251-B78]). The cut-off was set at 95% confidence interval for SLAC, FEL, MEME, and FUBAR analyses, and over 100 for REL analysis. We considered codons as being submitted to positive selection when they were detected as such by at least three of these approaches. As recombination can lead to false-positive identification by these methods, we also ran PARRIS, which account for the possibility of recombination and is proven to be more robust in these conditions ([@evu251-B85]). Also, only positions where three or more codons were identified were considered to be under positive diversifying selection. Homology modeling of TPR, ANK, and HEAT repeats was performed using HHPred ([@evu251-B94]), and protein structure graphics were obtained using Polyview ([@evu251-B79]). Results ======= Identification of the Fungal STAND NLR Repertoires -------------------------------------------------- To identify NLR-like proteins in the different complete fungal genomes, we have used NACHT and NB-ARC NOD domains from previously identified STAND proteins as queries. We have defined three different query sets. The first set comprised a list of fungal STAND proteins previously identified in the context of the study of fungal incompatibility. This query set we termed IR (incompatibility-related) includes the *P.anserina* HET-E, HET-D, and HET-R incompatibility genes and the fungal STAND proteins comprising a putative prion-forming domain ([@evu251-B71]; [@evu251-B24]). A second set, termed FV, was constituted of plant and animal proteins that have been validated as bona fide NLRs in functional studies and include, for instance, human NOD1 and NOD2, the NALP receptors and *Arabidopsis* RPP1, 8 and 13 and RPS 2, 4, and 5. A third set, termed PD (phylogenetically diverse), comprised an ensemble of STAND proteins with NACHT and NB-ARC Pfam-A annotations with a large phylogenetic distribution, ranging from bacteria to plants and animals and included sequences from different major lineages ([supplementary file S1](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The NB-ARC and NACHT sequences were extracted from the different query set and used in PSI-BLAST searches with three iterations and an *E* value cut-off of 10^−5^ on the complete annotated genome sequences of 198 strains of 164 fungal species (corresponding to the complete fungal genomes deposited at NCBI at the time of the study, [supplementary file S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The IR, FV, and PD query sets recovered 5,571, 1,053, and 4,657 hits, respectively ([supplementary fig. S1](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The IR recovered the most hits, whereas the FV set led to the lowest number of hits, but FV hits were almost entirely included in the IR and PD sets. The FV query did recover only a very limited number of NACHT domain STAND proteins, but was more efficient in the identification of NB-ARC STAND proteins ([supplementary fig. S1](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). We included all hits in our candidate set, which thus adds up to 5,616 sequences ([supplementary file S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online) corresponding to 4,476 (79.7%) and 1,144 (20.4%) NACHT and NB-ARC hits, respectively (four sequences were hit by both NACHT and NB-ARC queries). Hits were found in 122(101) of the 198(164) strains (species). In these 122 strains, there is mean number of STANDs per genome of 46, with a median of 37. Fungal NLR Domain Annotation ---------------------------- Next, we have annotated the hit sequences using Pfam and in-house annotation tools. Among the NOD domains, NACHT were more frequent than NB-ARC domains, but both categories were abundant ([fig. 1](#evu251-F1){ref-type="fig"} and [supplementary fig. S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The NACHT to NB-ARC ratio is 5:1. This contrasts with the situation observed in *viridiplantae*, where NB-ARC largely predominates (NACHT to NB-ARC ratio based on Pfam annotations is 1:180). In bacterial STAND proteins, both types are common; however, NB-ARC domains are also more abundant than NACHT domains (approximately in a 2:1 ratio). The higher occurrence of NACHT versus NB-ARC makes the fungal NLR candidate set more animal-like, because in metazoans NACHT domains are more frequent than NB-ARC (in a 17:1 ratio). A remaining 28% of the NOD regions picked up in the BLAST searches were neither annotated as NACHT nor NB-ARC by Pfam. Among the candidates were also a number of sequences showing noncanonical P-loop motifs with recurrent variations around the canonical GXXXXGKT/S motif ([supplementary table S1](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online), a situation also described in plant NLRs ([@evu251-B10]). F[ig]{.smallcaps}. 1.---Domain annotation in the fungal NLR set. Pie charts show the distribution of domain annotation in the N-terminal, NOD, and C-terminal domains, respectively. In each pie chart, the light gray corresponds to the fraction of domains with no annotation. NLRs have a typical tripartite domain organization, with a central NOD flanked N-terminally by an effector domain and C-terminally by an autoinhibitory/ligand-binding domain, often composed of superstructure-forming repeats ([@evu251-B58]). For annotation of the N-terminal domains in addition to the Pfam annotation, we have generated HMM signatures for a series of additional domains that have been found previously as N-terminal domains of fungal STAND proteins ([@evu251-B71]; [@evu251-B24]). Signatures were generated for the HET-s, PP, and σ prion-forming domains, the NAD1, Goodbye, HeLo-like, sesA, and sesB domains ([@evu251-B24]). HMM signatures were generated starting from a relevant individual sequence or a sequence alignment in the case of the short prion-forming motifs (see Materials and Methods). After annotation of the hit sequences, a strong overlap between the sesA and HeLo-like annotated set as well as the NAD1 and Goodbye annotated set was noticed, indicating that these domains are in fact related. For the sake of simplicity, we chose to merge these annotations using the Goodbye and HeLo-like designation for the NAD1/Goodbye group and sesA/HeLo-like group, respectively. It was noted previously that the sesB domain is related to lipases with α/β hydrolase fold ([@evu251-B33]; [@evu251-B24]), and not surprisingly, there was also some level of overlap between the sesB annotation and Pfam annotation related to α/β hydrolases. In this case also, we chose to merge the sesB and the α/β hydrolase Pfam annotations into a single category. We also merged the three PFD signature (HET-s, PP, and σ) into a single category. These motifs are unrelated in primary structure but have similar presumed functions. Among the Pfam annotations, we retained for these analyses only annotations that occur at least ten times in the set. A variety of other N-terminal and C-terminal annotations occur in a very limited number of NLR candidates ([supplementary file S3](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Materia](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1)l online) and were not analyzed further. We end up this way with 12 annotation categories for the N-terminal domains ([fig. 1](#evu251-F1){ref-type="fig"} and [table 1](#evu251-T1){ref-type="table"}). Among the annotated domains, the most frequent domains encountered as N-terminal effector domains are the Goodbye-like, HeLo-like, sesB-like, and PNP_UDP domains (each in the range of 20%). Then, the HET, Patatin, HeLo, and PFD domains are still relatively common (in the 4--1% range), while the other domains represent less than 1% of the annotations ([fig. 1](#evu251-F1){ref-type="fig"} and [supplementary fig. S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The PNP_UDP domain has been previously identified as an N-terminal effector domain in NLR proteins from the coral *A.digitifera* ([@evu251-B37]), and a sesB-related α/β hydrolase fold was found in a putative NLR in a bryophyte ([@evu251-B105]). Globally, roughly half of the sequences show no annotation in the region N-terminal to the NOD domain. In particular, in the basidiomycota, our annotation of the N-terminal domain is very limited with about only 15% of the sequences receiving an annotation ([supplementary fig. S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). Table 1List of the 12 Annotations Classes Retained for the N-Terminal Domains of Fungal NLRsDesignationPutative FunctionReference and/or PFAM ID.C2Membrane targetingPF00168Goodbye-likeUnknown[@evu251-B24], this studyHeLoPore formation[@evu251-B88]/PF14479HeLo-likeUnknown[@evu251-B33], this studyHETUnknown[@evu251-B91]/PF06985PatatinPhospholipasePF01734Peptidase S8Serine proteasePF00082PFDSignal transduction[@evu251-B24]/PF11558PKinaseProtein kinase domainPF00069PNP_UDPPhosphorylasePF01048RelA_SpoTppGpp synthesisPF04607sesB-likeLipase, esterase[@evu251-B33], this study In the domain C-terminal of the NOD domains, again only 52% of the sequences matched a Pfam A annotation. Ankyrin, WD-40 and TPR motifs corresponded to, respectively, 42, 29, and 25 % of the annotated sequences ([fig. 1](#evu251-F1){ref-type="fig"} and [supplementary fig. S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). In ascomycota, ANK repeats were more abundant whereas WD40 repeats prevailed in basidiomycota. No LRR motifs were found in agreement with a previous study ([@evu251-B93]). We conclude that fungal genomes encode a variety of NLR-like proteins with a great diversity of N-terminal and C-terminal repeat domains. Whereas the NACHT and NB-ARC, and ANK, WD, and TPR domains have been previously found in plant and animal STANDs, only a fraction of the N-terminal domains (like the PNP_UDP) have also been found in NLRs from other phyla. A large fraction (roughly 50%) of the N-terminal and C-terminal domains do not respond to known annotations. Diversity and Plasticity in Domain Architectures ------------------------------------------------ Next, we analyzed the domain architectures of the fungal NLR candidate set. Globally, there is a great diversity of domain architectures. To illustrate this aspect, we focused our analysis on the 1,228 sequences for which all three domains (N-, NOD, C-) have an annotation. The 12 annotated effector domains and NACHT and NB-ARC NOD domains can in principle lead to 24 (12 × 2) domain associations, and of those, 21 occur in our candidate set. Similarly, all six combinations of NACHT and NB-ARC with WD, TPR, and ANK motifs are found in the set. Globally, of the 72 possible tripartite domain architectures (12 effector domains × 2 NOD domains × 3 repeat domain), 32 are actually found in the set ([fig. 2](#evu251-F2){ref-type="fig"}). In general, for a given N-terminal domain, a type of architecture for the NOD and C-terminal domain predominates. Some domains show a strong bias in association, for instance HeLo-like and Patatin are almost invariably associated with NACHT and NB-ARC, respectively. Others like HET have a more equilibrated association with either NACHT or NB-ARC. This preferential combinatorial domain association is presented for the 12 N-terminal effector domain types ([fig. 3](#evu251-F3){ref-type="fig"}). There is also a preferential association between NOD types and C-terminal repeat type; NACHT is preferentially followed by ANK or WD whereas NB-ARC preferentially by TPR ([supplementary fig. S3](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). These preferential association trends always suffer exceptions, as a small fraction of the NB-ARC domains are associated with ANK or WD, and a small fraction of the NACHTs is followed by TPRs. The fact that in our sequence set some domain architectures are encountered only once suggests that some of the missing architectures might be identified by analyzing additional species. F[ig]{.smallcaps}. 2.---Domain architectures of fungal NLRs. The figures list the domain architectures found in 1,228 NLR candidates with tripartite annotation. For each of the architectures, the total count and percentages are given. F[ig]{.smallcaps}. 3.---Diagram of preferential domain associations in fungal NLRs. For each of the 12 annotation classes for the N-terminal domains of the fungal NLRs, the type of NOD, and C-terminal domain that are found associated with it are shown. The size of the disk is proportional to the abundance of a given architecture. For the NOD domains, "UNK" denotes unknown (nonannotated) domains. For the C-terminal domains, "REST" denotes unknown (nonannotated) domains and other annotations (distinct from WD, TPR, ANK). When inspecting the distribution of annotated N-terminal domains in phylogenetic trees based on the NOD domains, it appears that phylogeny of the N-terminal domains is frequently distinct from that of the NODs. This is apparent in two ways. First, in the global candidate set, the phylogenic trees based on the N-terminal domains are not congruent with the phylogenies of the NODs ([supplementary fig. S4](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). Then, when generating phylogenetic trees from the NLR complement from a given species based on the NOD sequences, domain architectures based on N-terminal domains do no form monophyletic groups but rather are to some extent scattered in different branches of the tree. For instance, in the phylogenetic tree based on the NOD domain of the NLR complement of the species *Bipolaris maydis*, the HET domain is found in different branches of the tree. The same is true for PNP_UDP, Goodbye, and HeLo-like domains ([supplementary fig. S5](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). This distribution and the observed combinatorial domain association suggest that de novo generation of specific domain architectures can occur by domain fusion events between N-terminal domains and a different lineage of NODs. In order to explore this aspect, we analyzed our NLR candidate set for situations in which a given NOD is highly similar to a NOD embedded in a distinct domain architecture. [Table 2](#evu251-T2){ref-type="table"} lists such situations in which highly similar NODs (between 80% and 99% identity) are associated with totally distinct N-terminal domains. Such situations can be explained by envisioning relatively recent domain fusion events, in which an N-terminal domain was swapped for another. Table 2Pairs of NLRs with Highly Homologous NOD Domains and Distinct N-Terminal DomainsGi Ident 1Tax Name 1N-Term 1NOD 1C-Term 1Gi Ident 2Tax Name 2N-Term 2NOD 2C-Term 2ScoreIdentity \[%\]156035777*S. sclerotiorum* 1980 UF-70**HELO-LIKE**NACHTWD40156060563*S.sclerotiorum* 1980 UF-70**SESB-LIKE**NACHTWD4024097.5156044028*S. sclerotiorum* 1980 UF-70**HELO-LIKE**NACHTWD40156060563*S.sclerotiorum* 1980 UF-70**SESB-LIKE**NACHTWD4026799.2156050803*S. sclerotiorum* 1980 UF-70**HELO-LIKE**NACHTUNK156060563*S. sclerotiorum* 1980 UF-70**SESB-LIKE**NACHTWD4023892.4451851214*B. sorokiniana* ND90Pr**HET**NACHTWD40189211806*P. tritici-repentis* Pt-1C-BFP**HELO-LIKE**NACHTWD4035488.3189209021*P. tritici-repentis* Pt-1C-BFP**HET**NACHTWD40189211806*P. tritici-repentis* Pt-1C-BFP**HELO-LIKE**NACHTWD4035086.8189209021*P. tritici-repentis* Pt-1C-BFP**HET**NACHTWD40482814165*S. turcica* Et28A**HELO-LIKE**NACHTWD4034984.8225559733*A. capsulatus* G186AR**PNP_UDP**NACHTWD40159124379*Aspergillus fumigatus* A1163**HELO-LIKE**NACHTWD4030682.3242760112*Talaromyces stipitatus* ATCC 10500**HELO-LIKE**UNKUNK212547165*T. marneffei* ATCC 18224**PNP_UDP**NACHTTPR21989.3242760112*T. stipitatus* ATCC 10500**HELO-LIKE**UNKUNK212547167*T. marneffei* ATCC 18224**PNP_UDP**NACHTTPR21989.3322704939*M. anisopliae* ARSEF 23**SESB-LIKE**NACHTANK342868671*Fusarium oxysporum* Fo5176**PNP_UDP**NACHTANK31780.8322704939*M. anisopliae* ARSEF 23**SESB-LIKE**NACHTANK475672654*F. oxysporum* f. sp. cub. race 4**PNP_UDP**UNKANK31980.8347826932*B. fuckeliana* T4**HET**NB-ARCTPR472238659*B. fuckeliana* BcDW1**SESB-LIKE**NB-ARCTPR56390.0353243899*Piriformospora indica* DSM 11827**HELO-LIKE**NACHTUNK353245097*Pi. indica* DSM 11827**SESB-LIKE**NACHTWD4030980.8402073505*G. graminis* var. *tritici* R3**HELO-LIKE**NACHTUNK402073554*G. graminis* var. tritici R3**RELA_SPOT**NACHTUNK33983.6402073505*G. graminis* var. *tritici* R3**HELO-LIKE**NACHTUNK402073555*G. graminis* var. tritici R3**RELA_SPOT**NACHTWD4033983.6402073505*G. graminis* var. *tritici* R3**HELO-LIKE**NACHTUNK402085097*G. graminis* var. tritici R3**RELA_SPOT**NACHTWD4033883.6429861644*C. gloeosporioides* Nara gc5**GOODBYE-LIKE**NACHTUNK429850945*C. gloeosporioides* Nara gc5**PNP_UDP**NACHTUNK22482.1429861644*C. gloeosporioides* Nara gc5**GOODBYE-LIKE**NACHTUNK429853607*C. gloeosporioides* Nara gc5**PNP_UDP**NACHTWD4022280.646130696*F. graminearum* PH-1**HELO-LIKE**NACHTWD4046138235*F. graminearum* PH-1**PNP_UDP**NACHTUNK33885.1[^2] Together, these observations suggest the existence of a combinatorial assortment of the N-terminal, NOD, and C-terminal repeat domains in fungal STAND proteins that resulted in a large diversity of domain architectures. The fact that domain architecture types do not represent a monophyletic group and the existence of highly similar NODs associated with distinct N-terminal domains, suggest that domain architecture invention events are not limited to a ancestral founding events but may reoccur frequently. Highly Conserved WD, ANK, and TPR Domains Are Enriched in Fungal NLRs --------------------------------------------------------------------- The analysis of STAND protein evolution in *Podospora* has revealed the existence of a NACHT-WD gene family (*nwd*), characterized by WD-repeats showing a high level of internal repeat conservation, meaning that the individual WD-repeats of a given gene are highly similar to each other (with about 85% identity at the amino acid level) ([@evu251-B83]; [@evu251-B71]; [@evu251-B18]). This internal repeat conservation is associated with a concerted evolution of the repeats, caused by constant reshuffling and exchanges of repeats both within a given gene or between different members of the gene family, which allows for rapid diversification ([@evu251-B71]; [@evu251-B18]). To determine if the presence of highly conserved repeats is a more general occurrence in fungal NLR proteins, we analyzed the NLR set for the presence of internally conserved repeats. Globally, 16% of the annotated repeats were found to show high internal conservation (over 85% identity over a minimum total length of 100 amino acids); respectively, 10%, 21.2%, and 21.6 % of ANK, TPR, and WD-repeats showed high internal conservation (the proportions varied somewhat between ascomycetes and basidiomycetes), ([fig. 4](#evu251-F4){ref-type="fig"}*A*). These observations indicate that the internal repeat conservation noted for WD repeats in *P.anserina* is a common property of a significant proportion of the NLR-like proteins and that this phenomenon is also encountered with ANK and TPR motifs both in ascomycetes and basidiomycetes. We have analyzed the occurrence of such highly conserved repeats in ANK, TPR, and WD-type repeats in plants, metazoan, and fungi ([supplementary table S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). We found that the fraction of repeats with high internal conservation is globally very low (0.4%, 0.8%, and 1.2% in viridiplantae, metazoan, and dikarya, respectively). There is thus a specific enrichment for highly conserved repeats in fungal NLR proteins. In dikarya, occurrence of highly conserved ANK, TPR, and WD repeats occurs mainly in NLR-like proteins, which globally account for 60-70% of the occurrence of highly conserved repeats. We conclude that highly conserved ANK, TPR, and WD repeats are highly enriched in fungal NLRs, as compared with their global occurrence. F[ig]{.smallcaps}. 4.---Superstructure-forming repeat domains of fungal NLRs. (*A*) Pie chart of repeat type found in ascomycete (top) and basiodiomycete (bottom) NLR candidates. For each repeat type, the fraction of repeats showing high internal conservation (HiC, 85% identity over at least 100 amino acids) is shown. (*B*) Distribution of the number of repeats in fungal NLR candidates for ANK, TPR, and WD repeats (Pfam signatures PF00023, PF13374, and PF00400, respectively). (*C*) Repeat length distribution in fungal NLR candidates for highly conserved ANK, TPR, and WD repeats. The distribution of the number of repeats per gene was different in ANK and TPR, compared with WD repeats. There was a gradual decrease in the class size with increasing number of repeats per protein in the case of ANK and TPR, while in the case of WD, class sizes were relatively constant from 1 to 14 repeats but then dropped sharply above 14 repeats (enough for the formation of two β-propellers; [fig. 4](#evu251-F4){ref-type="fig"}*B*). This difference might be related to the fact that ANK and TPR motifs form open-ended superstructures ([@evu251-B43]) rather than closed circular structures (β-propellers) in the case of WD-repeats ([@evu251-B96]). In the case of the TPR motifs, there is also apparently a preference for an even number of repeats. The maximum number of WD repeats was 21, which corresponds to the highest number of WD-repeats identified so far in a WD β-propeller domain and could allow for formation of a triple β-propeller. The occurrence of a low number (\<6--7) WD repeats, which a priori do not allow for formation of a closed β-propeller, might be due to the presence of cryptic repeats too degenerate to match Pfam signatures. The size distribution of the repeats corresponded to a very narrow range, typically 33-34 and 42-43 for ANK and WD repeats, respectively. Most TPR motifs were 42 amino acids in length, with only a minor fraction corresponding to in the canonical 34 amino acid length ([fig. 4](#evu251-F4){ref-type="fig"}*C*). Next, we analyzed whether or not highly conserved repeats are randomly associated with the different N-terminal effector domains. All frequent N-terminal domains can be found associated with highly conserved repeats, but it appears that certain N-terminal domains are preferentially associated with highly conserved repeats, as for instance the HET domain but also the prion-forming domains, whereas others like the Goodbye domains are very seldom associated with this type of repeats ([supplementary table S3](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). Phylogenetic Distribution ------------------------- Next, we analyzed the phylogenetic distribution of NLRs in fungi ([fig. 5](#evu251-F5){ref-type="fig"} and [supplementary file S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). NLRs were absent from certain lineages; in particular, no hits were found in any of the 38 analyzed Saccharomycotina genomes, or in the Schizosaccharomycetes. Similarly, we found no hits in early branching lineages of the microsporidia, chytrids, and mucorales. In contrast, hits were abundant in major basidiomycetes (agaricomycetes, 1,589 hits in 22 species, 72 hits per species) and ascomycetes lineages (pezizomycotina, 3,955 hits in 98 species, 40 hits per species). When comparing the annotation of the ascomycota and basidiomycota ensembles, three main trends are apparent. The ratio of NACHT to NB-ARC is slightly different in both lineages, with NB-ARC being rarer in basidiomycota (with a 1:8 ratio of NB-ARC to NACHT, compared with 1:4 in ascomycota). The abundance of the different types of repeat motifs also differs in both lineages: WD, ANK, and TPR account for 27, 9 and 8% of the C-terminal domain annotations in basidiomycota, compared with 10, 27, and 14% in ascomycota. The higher abundance of NB-ARC and TPR motifs in ascomycotina is expected, considering the preferential association of NB-ARC with TPR motifs ([supplementary fig. S3](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). The level of annotation of the N-terminal domains is very different in both lineages, with only 13% of the sequences receiving an annotation in the basidiomycota, compared with 63% for the ascomycota. This difference is probably related to the fact that our in-house annotations derive from ascomycete sequences. F[ig]{.smallcaps}. 5.---Phylogenic distribution of fungal NLRs. The list of species and strains in which NLR candidates were identified is shown together with their phylogenetic position. For each strain/species, the total count of NLR candidates and of the different N-terminal domains, NOD domains, and C-terminal repeat domains is given, as well as the count and the fraction of the repeat domains that show high internal conservation (HiC). Variation in the number of NLRs per genome is extreme, ranging from 1 (or 0) to 274 in the endophytic basidiomycetes species *Piriformospora indica.* In that species, NLR-like proteins correspond to 2.3% of the total proteins. Fifteen species show more than 100 NLR genes. There can be strong variations in the number of hits even between related species. For instance, within the *Aspergillus* genus, NLR numbers range from 12 to 99. The same is true even between strains belonging to the same species, as discussed below. Yet, in certain lineages of the pezizomycotina, there appears to be some group-specific increase or decrease in the number of hits. In particular, the hypocreales containing several Trichoderma species have significantly higher numbers of NLRs than the rest of the pezizomycotina (78 genes per species as opposed to 40, *P* = 0.006). The onygenales group containing several dermatophytes shows less hits than the rest of the pezizomycotina (16 genes per species, *P* = 0.018). We compared the occurrence of the 12 different N-terminal domains in the different species and there again the diversity between species is considerable. None of the 12 annotated domains has a universal distribution in all species displaying NLRs but some are found in a large fraction of species like the Goodbye-like, HeLo-like, and sesB-like domains found in NLRs of 88, 73, and 75 species, respectively. Other domains are found in a narrow species range, like the C2 domain found only in a few basidiomycota. As already noted for the total NLRs numbers, there is a high variability in the number of domain occurrences, even for closely related species, with for instance the number of PNP_UDP NLRs ranging from 2 to 23 in different species of the *Aspergillus* genus. Some domains show a strong tendency for marked expansions, while other are usually found as a single occurrence. We calculated a paralog-to-ortholog index, corresponding to the ratio of number of occurrences of the domain to the number of species in which the domain is encountered. The domains showing the highest number of occurrences per species were PNP_UDP and Goodbye, with a mean occurrence of 7.5 and 7.4 per species, respectively, while in contrast HeLo and Patatin domains showed the lowest occurrence (1,4 and 1,7) ([supplementary table S4](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). These two domains are most generally found as one or two occurrences per species, but some rare exceptions of marked expansion occur as for instance for the HeLo domain in the *fusaria*. When considering the C-terminal repeat domains, the fraction of repeats with high internal conservation varies dramatically between species from 0 to up to 58% in *Laccaria bicolor*. 72 strains, among the 122 displaying NLRs proteins, have at least one gene with internally conserved WD, TPR, or ANK repeats ([fig. 5](#evu251-F5){ref-type="fig"} and [supplementary file S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). Species in which such NLR-like proteins with high conserved repeats are particularly abundant are *L.bicolor,B.maydis*, and *Talaromyces stipitatus.* HSP90 and its co-chaperones SGT1 and RAR1 play important roles in NLR function both in plants and animals ([@evu251-B48]). We analyzed the complete fungal genomes for presence of putative SGT1 and RAR1 homologs and found SGT1 matches in all analyzed complete genomes and RAR1 matches in 111 out of 122 strains displaying NLR matches. Intraspecific Variation Reveals Extensive Polymorphism of the Fungal NLR Repertoire ----------------------------------------------------------------------------------- Previous reports suggest that fungal STAND proteins show high level of intraspecific variation ([@evu251-B71]; [@evu251-B30]; [@evu251-B12]; [@evu251-B41]). In addition, the extensive variation in STAND copy numbers in different species and the specific expansion of certain domain architectures in certain lineages suggest a death-and-birth evolution of these genes in fungi ([fig. 5](#evu251-F5){ref-type="fig"}), ([@evu251-B62]; [@evu251-B54]; [@evu251-B108]; [@evu251-B101]). In order to document this aspect, we chose to assess intraspecific variability in NLR proteins in our candidate set. We have thus specifically compared the NLR complement in 15 species for which the sequences of several strains are available. To compare the gene complement in each strain, phylogenetic trees were constructed based on the NOD domains only, and the trees were inspected for conservation of orthologous pairs (or triplets) between strains ([supplementary fig. S6](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). In all 15 analyzed species, some level of polymorphism in the NLR complement is observed. A variable fraction of the NLR sequences lack a clear ortholog in the other analyzed strain(s). [Table 3](#evu251-T3){ref-type="table"} presents, for each of the 15 species, the number of NLR proteins that are polymorphic, including the number of NLRs that are orphans (defined as a gene that does not show a clear ortholog in other strains from the same species) or semiorphans (when a pair of orthologous genes are found in two strains but not in a third). We find that NLR proteins are polymorphic between strains of the same species, and the fraction of polymorphic NLRs is systematically higher than for the total proteome (of note however is the fact that the level of polymorphism in the total proteome varies dramatically in different species, as the fraction of polymorphic proteins varies from 9.6% in *Penicillium digitatum* to 100% in *Rhizoctonia solani*). In addition, in many species, a significant proportion of the NLR candidates do not have a conserved ortholog in the other strain(s) (i.e., orphans or semiorphans). For instance, in *Aspergillus niger* CBS 513.88, ten sequence show no ortholog in the other strain (ATCC 1015) with a cut-off distance value of 1, which corresponds to about 50% identity. Inspection of the phylogenetic trees reveals the existence of numerous such orphans as well as strain-specific expansion in certain branches of the tree, arguing for a birth-and-death evolution of these sequences ([supplementary fig. S6](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). This strain-specific expansion is for instance evident *Fusarium oxysporum* Fo5176. Table 3Polymorphism of NLR-Like Proteins in Different Strains from the Same Species[^3] Relation of HET Domain to TIR Domains ------------------------------------- The HET domain acquired this designation because it was found in different proteins involved in nonself recognition in the form of heterokaryon incompatibility in fungi ([@evu251-B91]). In particular, this domain constitutes the N-terminal effector domain of the HNWD family members, which includes the *het-e*, *het-d*, and *het-r* incompatibility genes. Functional studies have identified this domain as being a cell death and incompatibility effector domain in *P. anserina* ([@evu251-B69]). We now find that the HET domain is relatively frequent as N-terminal domain of fungal NLR-like proteins and that it is often found associated with highly conserved repeats, potentially capable of rapid diversification. Because this study shows that many species display HET domain NLR-like proteins, we analyzed this domain further. We first conducted PSI-BLAST searches in the nr ("nonredundant") database with the HET-e1 HET domain by excluding fungal sequences and found that homologs of this domain are also found outside of the fungal kingdom in Stramenopiles, Haptophyceae, Choanoflagellates, green algae, and bryophytes. Next, we used Hidden Markov model searches to identify remote homologs of the HET-e1 HET domain. Both algorithms that we used (HHpred and JackHHmer) identified similarity to TIR domains. In particular, the two best hits in HHpred were to structure-based profiles constructed from the TIR domain of PdTIR from *Paracoccus denitrificans* ([@evu251-B16]) and of the TcpB protein from *Brucella melitensis* ([@evu251-B50]; [@evu251-B1]; [@evu251-B92]). [Figure 6](#evu251-F6){ref-type="fig"} presents an alignment of fungal HET domain proteins with bacterial and human TIR domains of known structure and related HET domains from phylogenetically diverse origins. The region of similarity of roughly 100 amino acids encompasses three main conserved blocks. These blocks of similarity map to the elements of secondary structure of the TIR domain α/β fold; the alignment, however, does not include the entire TIR domain, as similarity drops off after the region corresponding to helix αC. TIR domains function as adaptor domains in cell death and immune defense signaling cascades and function by interacting with partner TIR domains ([@evu251-B65]). This potential homology between HET and TIR domains suggests that HET domains may function by recruiting HET domain proteins and signaling downstream. F[ig]{.smallcaps}. 6.---Alignment of fungal HET domains with TIR domain proteins. The TIR domains of two bacterial proteins of known structure and of the human TLR1 TIR domain (boxed in red) are aligned with the HET domains of *P. anserina* HET-e1 and *Neurospora crassa* TOL (boxed in blue) together with related sequence of diverse phylogenetic origin annotated as HET domains in Pfam. On top of the alignment, the elements of secondary structure of *Brucella* TcpB are shown. Sequence designations are as follows: Paracoccus, *Paracoccus denitrificans,* gi\|500070302; Brucella, *Brucella melitensis*, gi\|516360271; Human Tlr1, *Homo sapiens*, gi\|194068387; Candidatus, *Candidatus Accumulibacter*, gi\|589611804; Emiliania, *Emiliania huxleyi*, gi\|551574256; Ectocarpus, *Ectocarpus siliculosus,* gi\|298709304; Thalassiosira, *Thalassiosira pseudonana*, gi\|224000455; Salpingoeca, *Salpingoeca rosetta*, gi\|514691135, Physcomitrella, *Physcomitrella patens,* gi\|168042266; *Podospora*, *P. anserina*, gi\|3023956 (HET-e1); Neurospora*, Neurospora crassa*, gi\|553134703 (TOL). ANK and TPR Motifs of NLR Proteins of P. *anserina* Show Repeat Length Polymorphism and Positive Diversifying Selection ----------------------------------------------------------------------------------------------------------------------- Superstructure-forming repeats with high internal conservation are enriched in fungal NLRs. These repeats belong to three types of superstructure-forming repeats, WD, ANK, and TPR motifs. We have previously shown that WD repeats of NLR-like proteins show extensive repeat size polymorphism in *Podospora* and are subject to concerted evolution and positive diversifying selection ([@evu251-B71]; [@evu251-B18]). We extended this analysis to ANK and TPR motif NLR proteins of *Podospora,* in order to determine whether repeat size polymorphism and diversifying selection was a common property of such repeat domains. We selected eight *P. anserina* NLR-encoding genes showing highly conserved ANK and TPR motifs, and PCR-amplified the repeat region from genomic DNA from five different wild isolates. For each locus, sequence analysis revealed repeat number polymorphism (RNP) ([table 4](#evu251-T4){ref-type="table"}). ANK repeat numbers ranged from 7 to above 16, whereas TPR motif numbers ranged from 2 to above 14. The RNPs observed suggest frequent recombination between repeats within a locus, and possibly between loci encoding the same type of repeats, as previously reported for WD-repeats ([@evu251-B71]; [@evu251-B18]). Table 4Repeat number polymorphism in ANK and TPR Repeat Domains of NLR Proteins from *Podospora anserina*Pa_2\_8180 PNP-UDP/ NACHT/ ANKPa_3\_8560 PNP-UDP/ NACHT/ ANKPa_2\_10340 sesB-like/ NB-ARC/ TPRPa_3\_9910 PFD/ NB6ARC/ TPRPa_5\_8060 PFD/ NB-ARC/ TPRPa_6\_7270 sesB-Like/ NACHT/ TPR(HEAT)Pa_6\_7950 sesB-like/ NACHT/ TPRPa_7\_3550 UNK/ NB-ARC/ TPRS107114410211Wa94ND8\>1597313NDWa96812\>1498611\>13Wa9713\>14ND119129\>13Wa99\>16112107NDND\>13Wa10010107NDND1010\>13Total\>57\>62\>494335414551Unique2236391921192929[^4] Next, we selected one ANK repeat locus and one TPR motif locus for which we had sequenced the highest number of repeats (Pa_3\_8560 and Pa_2\_10340, respectively) and analysed the variability of the repeats from individual loci. For each locus, individual repeat sequences were aligned and analysed for position under positive selection (see Materials and Methods) ([fig. 7](#evu251-F7){ref-type="fig"}). Five positions showed signs of positive selection in the ANK repeats and three in the TPR motifs. To locate the positive selection and polymorphic sites on the repeat domain structure, the repeats were homology-modeled to ANK and TPR domains of known structure. The TPR motif domain of Pa_2\_10340 was modeled using the human kinesin light chain 2 structure (Protein Data Bank \[PDB\] ID 3EDT) as template. In the TPR motifs, all positive selection sites as well as the other polymorphic position mapped to the concave side of the TPR structure in the α-helical regions. The ANK repeat domain of Pa_3\_8560 was modeled using the structure of artificial ANK repeat domain of engineered protein OR264 (PDB ID 4GPM) as template. In the ANK repeats, with one exception, the positive selection and polymorphic site also mapped on the concave surface of the ANK repeat domain in the inner helices and the β-hairpin/loop region, which correspond to the binding interface of ANK repeats based on cocrystal structures ([@evu251-B43]). F[ig]{.smallcaps}. 7.---Hypervariable sites in *P. anserina* TPR and ANK repeats of NLRs. (*A*) Alignment of individual TPR motif sequences found in different alleles of *Pa_2\_10340* (sesB-like/NB-ARC/TPR) is shown. Positions under positive selection are marked with a red dot; other highly variable positions are marked with a yellow dot. The TPR domain of Pa_2\_10340 was modeled using the human kinesin light chain 2 structure as (PDB ID 3EDT) as the template. Color coding of the positive selection and variable sites is as above. (*B*) Alignment of individual ANK repeat sequences found in different alleles of *Pa_3\_8560* (PNP_UDP/NACHT/ANK) is shown. Positions under positive selection are marked with a red dot, other highly variable positions are marked with a yellow dot. The ANK repeat domain of *Pa_3\_8560* was modeled using the structure of the artificial ANK repeat domain of the engineered protein OR264 (PDB ID 4GPM) as the template. Color coding of the positive selection and variable sites is as above. We also analysed two putative proteins from different species to determine whether this localization of the polymorphisms might be common to other ANK and TPR motifs. We chose the ANK and TPR proteins with the highest number of highly conserved ANK and TPR motifs, gi116208038 from *Chaetomium globosum* (PNP_UDP/NACHT/ANK) and gi255934897 from *Penicillium chrysogenum* (UNK/AAA/TPR), with, respectively, 21 ANK repeats and 21 TPR motifs. By comparing the repeats and mapping the variable positions onto a homology model (PDB ID 4GPM for ANK and 3EDT for TPR), we found that polymorphisms map to the same positions in the α-helices of the concave surface of the TPR domain and to the inner helices and β-hairpin/loop region of the concave interface of the ANK domain ([supplementary fig. S7](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). Based on the localization of these polymorphic sites, it can be inferred that if repeat contraction/expansion/shuffling occurs in these genes, these events will lead to ANK and TPR arrays with modified binding interfaces. Collectively, these analyses suggest that the evolution of ANK and TPR motifs of *Podospora* NLR candidates is analogous to the evolution of highly conserved WD repeats of NLR-like proteins, which has been described previously ([@evu251-B71]; [@evu251-B18]). Discussion ========== In plants and animals, NLRs are essential components of innate immunity. Work on fungal incompatibility revealed the existence of NLR homologs in fungi with functions in the detection and response to nonself. Herein, we have analysed close to 200 fungal genomes for the presence of NLR candidates and describe the identified sequences. We find that multicellular pezizomycetes and agaricomycete generally encode large and diverse repertoires of NLR-like genes. Diversity of N-Terminal Effector Domains ---------------------------------------- Many of the N-terminal effector domains of fungal NLRs remain completely uncharacterized, in particular in basidiomycotina. We have nevertheless defined 12 main annotation classes for these N-terminal domains that roughly accounts for 50% of the sequence set. For some of these domain classes, functional information is available, although it is in most cases fragmentary. One of the domains, that was previously identified as an effector domain in animal NLRs is the PNP_UDP domain. Indeed, this domain was found as N-terminal domain of NLRs in the coral *A.digitifera* ([@evu251-B37]), and also as an effector domain associated with a DD in sponge ([@evu251-B107])*.* In addition, we reveal a remote similarity between the HET domain and the TIR domain, originally identified in Toll-like receptors in mammals and also found as the N-terminal domains of a large fraction of plant NLRs. Considering this similarity, it might be hypothesized that HET domain fungal NLRs are functionally analogous to plant TIR-NB-LRR proteins. TIR domains regulate immune responses by homo and heterodimerization; HET-domain containing NLRs like the *P.anserina* HET-e, HET-d, and HET-r proteins may therefore mediate the incompatibility response by interaction with downstream HET domain proteins acting as adaptor domains. A large fraction of the N-terminal domains is related to the HeLo domain identified in the HET-s prion protein of *P. anserina* ([@evu251-B34]; [@evu251-B88]). This domain is a cell death execution domain that can be activated following prion transconformation of the PFD region of HET-S. The HeLo domain is then translocated to the cell membrane, where it functions as a pore-forming toxin ([@evu251-B63]; [@evu251-B88]). The HeLo domain is found as the N-terminal domain of NLRs in many different species, but even more frequent is a variant form of this domain that we term HeLo-like, which could potentially play a similar role in cell death execution. Another abundant class is the sesB-like domain, which corresponds to a predicted lipase domain ([@evu251-B33]; [@evu251-B24]). This lipase domain is found in the human SERAC1 protein, which was found to be involved in a metabolic disease ([@evu251-B103]). Human SERAC1 displays phospholipid esterase activity and is able to modify lipid composition of the plasma membrane. It might be that sesB-like domains induce specific plasma membrane modification in response to nonself. Our annotation list contains another lipase domain, namely the Patatin domain. Interestingly, the Patatin lipase domain was involved in the control of PCD and defense in plants ([@evu251-B13]; [@evu251-B55]; [@evu251-B52]). Based on the fact that one of the incompatibility genes of the fungus *C.parasitica* encodes an NLR with a Patatin domain, it can be reasonably inferred that Patatin-like domains might also function in the control of cell death in fungi. Considering that the C2 domain, found as N-terminal effector domain in basidiomycete NLR candidates, is a lipid-binding domain ([@evu251-B22]), it appears that a significant fraction of the identified N-terminal domain of fungal NLRs target membranes or lipids. The RelA_SpoT domain was so far only described in bacterial and plant chloroplast proteins ([@evu251-B4]); we now identify it as the N-terminal effector domain of fungal NLRs. This enzymatic activity-carrying domain is responsible for the synthesis of the ppGpp bacterial alarmone, which mediates the stringent response in bacteria ([@evu251-B11]). One possible explanation of the presence of this domain as an N-terminal domain of fungal NLRs would be that fungi exploit the prokaryotic signaling ppGpp cascade to manipulate bacterial pathogens, competitors, or symbionts. The same might be true for the PNP_UDP class. Based on the analysis of the PNP_UDP domain in NLRs from *P.anserina*, these domains are predicted to be methylthioadenosine/*S*-adenosylhomocysteine nucleosidases, which are involved in the synthesis of quorum-sensing molecules like Al-2 ([@evu251-B73]). Maybe these effector domains manipulate prokaryotic signaling in the context of adverse or beneficial interactions. Globally, when one considers domains found N-terminal to the NOD domain in this NLR collection, two main categories emerge. Class 1 domains correspond to domains that have a proposed enzymatic function or a potential direct role in cell death induction. In this first class, one finds the proposed PNP_UDP, RelA_SpoT, sesB-like, and Patatin lipase domains and also the HeLo and HeLo-like proposed pore-forming toxin domains. In this class, the N-terminal domain is believed to represent a direct effector of the NLR activation. Class 2 domains correspond to domains that more likely have an adaptor function, a situation more typical of plant and animal NLRs, where domains such as CARD, PYD, and TIR recruit effectors by homotypic domain interactions to signal downstream, rather than representing the terminal effector of the immune cascade. CARD and PYD mediate NLR signaling by a prion-like mechanism, involving formation of higher-order complexes ([@evu251-B104]; [@evu251-B14]; [@evu251-B60]). The three prion-forming domains (HET-s, PP and σ) associated with fungal NLRs correspond to this second class ([@evu251-B24]). The HET domain, possibly homologous to the TIR domain, also likely corresponds to this second class. Many of the fungal NLR-related proteins fall into class 1, while apparently in plant and animal lineages this situation is less frequent, although as mentioned previously PNP_UDP NLRs have been described in corals ([@evu251-B37]). In complex (multicellular) bacterial STAND proteins, the presence of N-terminal domains with predicted enzymatic activity such as a metacaspase domain is common, in particular in cyanobacteria ([@evu251-B58]; [@evu251-B3]). It might be that at the base of STAND protein evolution the all-in-one architecture is ancestral and that incorporation of adaptor domains between the NLR receptor and the effector represents a further sophistication of the signaling process. Superstructure-Forming Repeat Domains in Fungal NLR-Related Proteins -------------------------------------------------------------------- We failed to identify NLR candidates with LRR motifs, a situation already reported in a study specifically tailored to the identification of LRR pattern-recognition receptors in fungal genomes ([@evu251-B93]). Instead, STAND proteins displayed ANK, TPR, and WD motifs. ANK, TPR, and WD motifs were found associated with NLRs in the coral *A.digitifera* ([@evu251-B37])*.* Similarly, in their analysis of the repertoire of NLRs in the sponge *Amphimedon queenslandica,* [@evu251-B107] also reported that in several holozoan and nonholozoan genomes NACHT domain proteins are associated with ANK, TPR, and WD repeats, but no LRR motifs were found ([@evu251-B107]). Based on the unified NLR nomenclature proposed in 2008, these authors stated that such non-LRR STAND proteins should not be designated NLR and that this designation should be restricted to proteins encompassing LRR motifs. In that sense, all candidates identified in the fungal genomes would not represent bona fide NLRs. We do however consider that given the combinatorial association of different repeat domains with NACHT or NB-ARC domains in fungi and other lineages, it is reasonable to assume that regardless of the type of superstructure-forming repeats they harbor, these NLR-related genes display related functions. Although a restrictive nomenclature offers the advantage of simplicity, because it is based on domain architecture, it may not be optimal for a global understanding of the role and evolution of NLR-related genes across phyla. As another illustration of this principle, the DEATH-NACHT domain proteins are found in the cnidarian *Hydra magnipalpillata* that lack LRR motifs but cluster with vertebrate NLRs ([@evu251-B107]). Similarly, [@evu251-B37] also favor the notion that different NOD/WD, ANK, TPR, and LRR associations are ancestral and that in certain lineages, NOD/LRR architectures have flourished whereas other architectures were lost ([@evu251-B37]). Following this plausible model, it might be proposed that the NOD/LRR architecture was specifically lost in the fungal lineage while NOD/TPR, ANK, and WD architecture were expanded. NLR loss in certain lineages is not uncommon; nematodes and arthropods are apparently devoid of NLRs ([@evu251-B61]; [@evu251-B37]) and TIR-NB-LRRs have been reduced or lost in monocotyledon plants ([@evu251-B47]). A significant fraction of the superstructure-forming repeat domains in fungal NLRs show strong internal conservation, a situation we have previously described for the WD-repeat domains of the *nwd* gene family of *Podospora* ([@evu251-B83]; [@evu251-B71]; [@evu251-B18]). We have found that this internal conservation corresponded to the concerted evolution of the repeats both within and between members of the gene family, and was typically associated with repeat number polymorphism. In addition, these WD-repeats show positive diversifying selection at specific codon positions, corresponding to amino acid positions defining the ligand-binding interface of the WD β-propeller structure ([@evu251-B71]). Due to the high conservation of the repeats, these sequences are prone to RIP (repeat induced point mutation), a genomic defense mechanism that mutates and methylates repeated sequences premeiotically in fungi ([@evu251-B87]). At least in *Podospora*, the effect of RIP on these repeat regions might represent a mechanism of hypermutation, allowing a rapid diversification of these sequences. We have proposed that the combination of these evolutionary mechanisms constitutes a process for generating extensive polymorphism at loci that require rapid diversification. This study now suggests that this regimen of concerted evolution and positive diversifying selection might be of general relevance to the evolution of a fraction of fungal NLRs. We find that many superstructure-forming repeat domains in fungal NLR show strong internal repeat conservation and that in *Podospora,* ANK and TPR motifs also show RNP and signs of positive selection at positions predicted to be located in the interaction surfaces in the ANK and TPR structures. In the context of nonself recognition, rapid diversification of the receptors might be particularly critical; it appears that the modularity and plasticity properties of superstructure-forming repeats might have been exploited in many fungal species, to allow diversification of their NLR repertoires. Among the three superstructure-forming repeat types, ANK repeats were the most common in fungal NLRs candidates. The involvement of ANK repeats in host-symbiont or host--parasite interaction was highlighted by previous studies, showing that ANK repeat proteins are enriched in symbiotic and obligate intracellular bacterial species, as compared with free-living species ([@evu251-B44]). Similarly, a rapidly evolving family of ANK repeat proteins was found to control host--parasite interaction in *Wolbachia* ([@evu251-B90]). ANK repeat domain proteins were also found to be specifically enriched during expression changes associated with nonself recognition in *Podospora* ([@evu251-B7]). Thus, across phyla, ANK repeat domains appear often to be involved in the regulation of inter-organismal interactions. Architectural Diversity of Building Blocks in NLRs -------------------------------------------------- One of the marked characteristics of the fungal NLRs is the extensive domain architecture diversity. Studies of the NLR repertoires in lower animals already hinted at this diversity in domain architecture ([@evu251-B56]; [@evu251-B37]; [@evu251-B107]). The description of the fungal NLRs further illustrates this diversity. Even with the very partial annotation, we establish a great variety of architectures, revealing a combinatorial association of different N-terminal, NOD and repeat domains. This diversity is evident both in the phylum and within a given species, which can display tens of different NLR domain architectures. Importantly, in many cases a given domain architecture does not have a monophyletic ancestry. Rather, it appears that reoccurring domain fusion events lead to multiple independent inventions of the same architectures. These domain associations appear not to be limited to ancestral events, as suggested by the fact that NOD with 99% identity can be found associated with totally distinct N-terminal domains. These observations, as well as the species or strain-specific expansions of paralogs, are compatible with the notion that fungal NLRs evolve by a birth-and-death regimen. Others have previously documented the role of birth-and-death evolution in fungal *het* gene homologs in the basidiomycetes ([@evu251-B101]). This apparent plasticity of the NLR repertoire, based on the combinatorial association of a variety of effector, NOD and C-terminal receptor domains, might represent a mechanism that allows a rapid adaptation of the NLR repertoire in the arms-race with the variable biotic environment. The combinatorial build-up of an immune repertoire from a limited set of elementary domain is also a general characteristic of the immune-related proteins in plants and animals ([@evu251-B67]). Phylogenetic Distribution of NLRs and Possible Functions in Immunity and Beyond ------------------------------------------------------------------------------- Our analysis of the phylogenetic distribution of NLR homologs in fungi indicates that their presence is apparently restricted to filamentous multicellular fungi. We found no NLR homologs in yeast species. The simplest interpretation of this lack of NLR homologs in yeast species is that this gene family was lost in unicellular fungi, because the constraints on the management of biotic interactions are fundamentally different for multi and unicellular organisms. *Soma* and *germen* are essentially one and the same thing in the latter organisms, therefore the maintenance of a machinery aimed at protection of the *soma* against parasitism may not be required in yeasts, in particular when considering that one common outcome of NLR-controlled defense in animals, plants, and fungi is programmed cell death. We also failed to identify NLR-related genes in early branching non-dikarya fungal lineages of the chytrids, microsporidia, and mucorales and also in some dikarya basidiomycete lineages such as the tremellomycetes and the pucciniomycotina, in agreement with previous studies ([@evu251-B101]). This could indicate that NLR-like genes were lost in these lineages or that the level of divergence of the NACHT and NB-ARC domains used in our search prevented their detection. Within the filamentous agaricomycotina and pezizomycotina, the number of NLR homologs varies dramatically between species. One may attend to establish a relationship between the species ecology and the constitution of the NLR homolog repertoire ([supplementary file S2](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [Supplementary Material](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) online). This can only be made with extreme caution, because in many cases, the information available on the species ecology is at best fragmentary and many species have multiple habitats and life-styles. In some groups, there is a significant enrichment or scarcity of NLRs. For instance, animal dermatophytes of the onygenales have in general few NLR genes. But it is difficult to determine whether this is related to the phylogenetic position or to ecology. If the function of NLR homologs in fungi is related to innate immunity, the prediction might be that fungi potentially in relation to diverse pathogens or competitors or hosts should be particularly enriched in terms of NLR repertoire, and reciprocally, that in fungi living in less populated niches, smaller repertoires could be sufficient. This prediction might be verified in some instances as, for example, in the case of the highly versatile pathogens like *Fusarium* species or mycoparasitic Trichoderma species, in which the repertoire is large. In the thermophile *Chaetomium thermophilum,* the citrus fruit pathogen *Pe. digitatum* or the "whisky fungus" *Baudonia compniacensis* have small repertoires and inhabit restrictive niches. Similarly, specialized pathogens, such as *Claviceps purpurea*, might be protected against microbial competitors by their host immune system, which could explain the low number of NLRs. The current view of the role of the NLRs in the animal lineage is expanding. Initially viewed as immune receptors whose role is to detect and respond to pathogenic nonself, it is becoming apparent that these receptors are also critical for the management of other nonpathogenic biotic interactions, notably with the symbiotic microbiome ([@evu251-B21]). For instance, the human NOD2 NLR is required for the establishment of a commensal microbiome in the intestine ([@evu251-B75]). Similarly, it has been proposed that the expanded NLR repertoires in the coral *A.digitifera* could be devoted to the interaction with an obligate dinoflagellate endosymbiont ([@evu251-B37]). In the fungal kingdom, it has been emphasized that pathogenesis and symbiotic interaction are based on similar mechanisms ([@evu251-B102]). It might thus be proposed that part of the NLR repertoires found in fungi might function in the control of a variety of biotic interactions and not be strictly devoted to an immune function per se (understood as the response to pathogenic nonself). These proteins might be involved in the control of nonself recognition in the context of fungal pathogenicity, or symbiosis in the form of ECM formation, endophytic growth, lychen formation, or interaction with symbiotic endobacteria. As already mentioned, NACHT domain proteins are specifically expressed during mycorhizal symbiosis in *L.bicolor,* and in *T.melanosporum,* an expanded family of NACHT-ANK proteins is characterized by a remarkable mechanism of diversification based on alternative splicing of codon-sized mini exons ([@evu251-B62]; [@evu251-B41]). In this study, the species showing the highest number of NLRs is *Pi. indica,* which is an endophytic fungus ([@evu251-B108]). Conclusion ========== Fungal NLR homologs have been shown, in two species, to be involved in nonself recognition and in the control of PCD ([@evu251-B83]; [@evu251-B20]). We now report that filamentous fungi possess variable repertoires of NLR homologs, which show similarities and differences with NLRs in plant and animal lineages. This glimpse of fungal NLR diversity represents a further opportunity in comparative immunology for a more complete understanding of the build-up and evolution of immunity in eukaryotes. Although viridiplantae NLR repertoires are characterized by their considerable size (NLR repertoires with several hundreds of NB-LRR genes are not uncommon), mammalian NLR repertoires are fixed and reduced, most likely due of the presence of an adaptive immune system ([@evu251-B61]). In lower animals, NLR repertoires appear more extended, with again up to several hundred NLR genes in certain species ([@evu251-B56]; [@evu251-B37]; [@evu251-B107]). The fungal NLR repertoires similarly appear highly variable, but only exceptionally reach the complexity found in lower animals and land plants. The common occurrence of rapidly evolving ANK, TPR, and WD nonetheless may entail these repertoires with the plasticity required to cope with a complex and changing biotic environments. Animal and plant NLRs employ mechanistically distinct strategies for defense, in the form of intracellular PAMP detection in animals and ETI (effector-triggered immunity) in plants ([@evu251-B61]). It will be of interest to determine which strategies have been adopted in the fungal lineage. The involvement of NLR-like proteins in incompatibility, in which cell death is triggered by the recognition of an allelic variant of an endogenous protein by an NLR is compatible with a model of effector-triggered immunity ([@evu251-B70]; [@evu251-B20]; [@evu251-B6]). Fungi possess extremely diverse lifestyles involving a variety of obligate or facultative biotic interactions; further functional studies are now required to understand which role these fungal NLR homologs play in the management of these diverse interorganismal interactions and which mechanistic strategies underlie NLR function in fungi. Supplementary Material ====================== [Supplementary files S1--S3](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), [figures S1--S7](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1), and [tables S1--S4](http://gbe.oxfordjournals.org/lookup/suppl/doi:10.1093/gbe/evu251/-/DC1) are available at *Genome Biology and Evolution* online (<http://www.gbe.oxfordjournals.org/>). ###### Supplementary Data This work was funded by a grant from the Agence National de la Recherche (ANR Blanc "Mykimun"), by the Australian Research Council and by the National Health and Medical Research Council (NHMRC). B.K. is an NHMRC Research Fellow. W.D. is on leave from Faculty of Fundamental Problems of Technology, Wroclaw University of Technology, Poland. [^1]: **Associate editor**: Kenneth Wolfe [^2]: N[ote]{.smallcaps}.---Gi Ident, GenBank identification. [^3]: N[ote]{.smallcaps}.---Orph, orphan; Ident, identification; Semiorph, Semiorphan. [^4]: N[ote]{.smallcaps}.---ND, not determined.

INTRODUCTION {#s1} ============ Oral cancer is the tenth most prevalent cancer accounting for almost 300,000 new cases annually worldwide, with two thirds occurring in developing countries \[[@R1], [@R2]\]. Although tobacco smoking and alcohol drinking are major risk factors for oral cancer, there are still 15-20% non-smokers and non-drinkers developing this disease \[[@R3]\], indicating that genetic factors, alone or interaction with environmental factors may also be important in the development of this cancer \[[@R4], [@R5]\]. Many epidemiologic studies demonstrated that some single nucleotide polymorphisms (SNPs) were correlated with oral cancer susceptibility \[[@R6], [@R7]\]. Recently, a genome-wide association study (GWAS) identified a new susceptibility loci at fatty acid desaturase 1 (*FADS1*) gene (rs174549) associated with laryngeal squamous cell carcinoma (LSCC) in Chinese population \[[@R8]\]. *FADS1* encodes delta-5 desaturases and is involved in the metabolism of polyunsaturated fatty acids (PUFAs). Several SNPs at the *FADS1* gene influence the concentration of long-chain PUFAs in plasma \[[@R9]\]. Previous experimental studies found that PUFAs and its metabolites could inhibit tumor proliferation and invasion in head and neck cancer cell \[[@R10], [@R11]\]. However, so far, there is limited epidemiologic research on the role of *FADS1* gene polymorphism in oral cancer risk. Fish, the major dietary sources of long chain PUFAs, is a favorite food for residents of Fujian (a province located on the southeast coast of China). The protective effect of fish intake was showed in several cancers (such as esophagus, gastric, liver, etc.)\[[@R12]--[@R14]\]. Edefonti et al.\[[@R15]\] found unsaturated fats dietary pattern could reduce the risk of oral and pharyngeal cancer. However, to date, relatively few studies have reported the fish intake related to oral cancer. Moreover, it is still unclear whether *FADS1* gene polymorphism and its interaction with fish intake could contribute to the prevention of oral cancer risk. Therefore, the present case-control study was to assess the independent and combined effects of the new susceptibility loci (rs174549) and fish consumption on oral cancer in southeast China. RESULTS {#s2} ======= The distribution of all subjects on demographic variables and potential confounding factors are described in Table [1](#T1){ref-type="table"}. There were no significant differences between cases and controls in demographic characteristics (except that greater proportions of cases were lower educated and rural settings). As expected, smoking, drinking, vegetables and fruits intake were associated with oral cancer risk (*P*\<0.05). The main histology types were squamous cell carcinoma (85.43%) and adenocarcinoma (8.94 %). ###### Distribution of selected characteristics among case and control subjects Variables Case (%) n = 302 Control (%) n = 574 *P* value -------------------------- ------------------ --------------------- ----------- Age (years) 0.149  ≤44 46(15.23) 117(20.38)  45-59 143(47.35) 279(48.61)  60-74 87(28.81) 137(23.87)  ≥75 26(8.61) 41(7.14) Gender 0.104  Male 200(66.23) 348(60.63)  Female 102(33.77) 226(39.37) Education Level \<0.001  Illiteracy 40(13.25) 71(12.37)  Primary-Middle school 189(62.58) 256(44.60)  High school or above 73(24.17) 247(43.03) Marital status 0.113  Married 268(88.74) 528(91.99)  Others 34(11.26) 46(8.01) Residence \<0.001  Rural 154(50.99) 161(28.05)  Urban 148(49.01) 413(71.95) Family history of cancer 0.660  No 235(77.81) 454(79.09)  Yes 67(22.19) 120(20.91) Tobacco smoking \<0.001  No 143(47.35) 408(71.08)  Yes 159(52.65) 166(28.92) Alcohol drinking \<0.001  No 179(59.27) 459(79.97)  Yes 123(40.73) 115(20.03) Vegetables \<0.001  ≤1 time/day 139(46.03) 168(29.27)  \>1times/day 163(53.97) 406(70.73) Fruits \<0.001  ≤3 times/week 233(77.15) 276(48.08)  \>3 times/week 69(22.85) 298(51.92) Table [2](#T2){ref-type="table"} shows the effects of rs174549 polymorphism and fish consumption on oral cancer. Genotype distribution of *FADS1* (rs174549) among controls was in agreement with Hardy-Weinberg equilibrium (*P*\>0.05). After adjustment for potential confounders, *FADS1* A variant allele was associated with a significantly decreased risk of oral cancer: the ORs were 0.65 (95% CI: 0.42-0.99) for codominant model and 0.67 (95% CI: 0.46-0.98) for recessive model. Moreover, when stratified by demographic characteristics, the statistically significant reverse associations were only emerged in men and those age ≤ 60 years. When stratified by main lifestyle factors, the protective effect of AA genotype was especially evident in smokers and alcohol drinkers (Figure [1](#F1){ref-type="fig"}). ###### Effects of *FADS1* rs174549 polymorphism and fish intake on oral cancer Variable Case (%) (n = 302) Control (%) (n = 574) Unadjusted odds ratios (95% CI) Adjusted odds ratios^a^ (95% CI) ------------------------- -------------------- ----------------------- --------------------------------- ---------------------------------- rs1745496 (P~HWE~=0.32) Codominant model  GG 106(35.10) 169(29.44) 1.00 1.00  AG 147(48.67) 274(47.74) 0.86(0.62-1.17) 0.86(0.62-1.20)  AA 49(16.23) 131(22.82) 0.60(0.40-0.90) 0.65(0.42-0.99) Dominant model  GG 106(35.10) 169(29.44) 1.00 1.00  AG+AA 196(64.90) 405(70.56) 1.30(0.96-1.74) 1.26(0.92-1.72) Recessive model  GG+AG 253(83.77) 443(77.18) 1.00 1.00  AA 49(16.23) 131(22.82) 0.65(0.46-0.94) 0.67(0.46-0.98) Fish intake  0-2 times/week 138(45.69) 147(25.61) 1.00 1.00  3-6 times/week 112(37.09) 178(31.01) 0.67(0.48-0.93) 0.85(0.58-1.25)  ≥7 times/week 52(17.22) 249(43.38) 0.22(0.15-0.32) 0.27(0.18-0.42)  P for trend \<0.001 \<0.001 ^a^ Adjusted for age, gender, education, marital status, residence, family cancer history, smoking, drinking, vegetables and fruits. {#F1} Additionally, with regard to fish consumption, the frequency of fish intake was categorized into three groups according to the tertiles of controls (0-2 times/week, 3-6 times/week, ≥7 times/week). Fish intake ≥7 times/week showed a 73% reduction in risk for oral cancer compared to those who ate fish less than 2 times/week (OR: 0.27, 95% CI: 0.18-0.42). Moreover, there was a tendency of decreased risk with the increasing frequency of fish consumption (all P for trend \<0.001). We further evaluated the joint effects of rs174549 polymorphism in recessive model and fish consumption on the risk of oral cancer (Table [3](#T3){ref-type="table"}). A significantly lower OR was observed in individuals who carrying AA genotype and consumed fish ≥7times/week compared with GG+AG carriers who ate fish less than 2 times/week (OR: 0.30, 95% CI: 0.14-0.63). Moreover, a positive multiplicative interaction between *FADS1* gene and fish intake for oral cancer was found (OR~multiplicative~ = 0.70, 95% CI: 0.51-0.96, *P*=0.028; data not shown). ###### Interactions between *FADS1* rs174549 polymorphism and fish intake in oral cancer Variables Cases (%)N =302 Controls (%)N = 574 OR (95% CI) OR^a^ (95% CI) ----------- ---------------- ----------------- --------------------- ----------------- ----------------- FADS1 Fish intake GG+AG 0-2 times/week 120(39.74) 120(20.91) 1.00 1.00 GG+AG 3-6 times/week 94(31.13) 124(21.60) 0.76(0.52-1.10) 0.93(0.60-1.42) GG+AG ≥7times/week 39(12.91) 199(34.67) 0.20(0.13-0.30) 0.25(0.15-0.40)  AA 0-2 times/week 18(5.96) 27(4.70) 0.67(0.35-1.27) 0.74(0.36-1.52)  AA 3-6 times/week 18(5.96) 54(9.41) 0.33(0.18-0.60) 0.50(0.25-0.97)  AA ≥7times/week 13(4.30) 50(8.71) 0.26(0.13-0.50) 0.30(0.14-0.63) ^a^ Adjusted for age, gender, education, marital status, residence, family cancer history, smoking, drinking, vegetables and fruits. DISCUSSION {#s3} ========== To our knowledge, this case-control study is the first to report independent and joined effects of the new susceptibility loci in *FADS1* (rs174549) and fish consumption on oral cancer in southeast China. We found AA genotype was associated with a decreased risk of oral cancer compared to the GG genotype. Moreover, fish intake ≥7times/week also reduced the risk of oral cancer. Furthermore, there was a positive multiplicative interaction between *FADS1* gene and fish intake for oral cancer. To date, only a GWAS study revealed that *FADS1* polymorphism (rs174549) showed a protective effect on LSCC \[[@R8]\]. This finding is consistent with the present results. Although the mechanism of rs174549 polymorphism on oral cancer is not clear, previous study found there were 49 SNPs in high linkage disequilibrium with rs174549 in the same chromosome, and these SNPs might influence the expression of *FADS1* through their effects on host genes \[[@R8]\]. Moreover, *FADS1* variation could suppress inflammatory response through influencing the metabolism of PUFAs. The main metabolites of PUFAs include arachidonic acid (AA; a pro-inflammatory factor), eicosapentaenoic acid (EPA), and docosahexaenoic acid (DHA) (EPA and DHA are anti-inflammatory factors). Horiguchi et al.\[[@R16]\] found *FADS1* polymorphism (rs174547) was correlated with lower AA, but unchanged for EPA or DHA. Yao et al.\[[@R17]\] showed *FADS1-FADS2* gene cluster variation could inhibit the conversion of α-linolenic acid (ALA) to AA. Tanaka et al.\[[@R18]\] demonstrated rs174537 polymorphism in *FADS1* could increase the expression of EPA. Therefore, we speculated that imbalance between AA and EPA and DHA might be a mechanism of *FADS1* rs174549 polymorphism on oral cancer. Our study revealed that fish intake might be a benefit factor for oral cancer risk, which is consistent with previous studies \[[@R19]\]. Higher intakes of fish which contain key anti-inflammatory nutrients (LC-PUFA, EPA, DHA, etc.)\[[@R20]\], have been reported to suppress inflammation, oxidative stress and cancer risk in animal study \[[@R21]\] and observational study \[[@R22]\]. Additionally, Actis et al.\[[@R23]\] found dietary lipids (especially for n-3 fatty acids) also reduced cell proliferation and differentiation of murine oral squamous epithelium. Therefore, these have been reasonably hypothesized that fish consumption may lower the risk of oral cancer. Interestingly, our results demonstrated a significant gene-diet interaction between *FADS1* gene and fish intake for oral cancer risk. Yeates et al.\[[@R24]\] showed that *FADS1* gene variant was associated with decreased level of AA in serum and increased ALA to DHA in high fish intakes population. An explanation for our finding might be that the genetic variation in *FADS1* could interact with dietary fish oil to increase delta-5 activity and change LC-PUFA proportions \[[@R25]\]. There are some limitations in this study. First, the present study only evaluated the effect of total fish consumption on oral cancer, and did not further to analyze the effects of different fish species and the preparation methods. Hence, these factors should be taken into account in future studies. Second, only one SNP was chosen based on a recent GWAS which may not represent a comprehensive view of *FADS1* gene variation. Further studies on variation in susceptible regions of *FADS1* gene are needed. Third, since this is a very preliminary study, further animal or cell experiments with more rigorous design are also required to explore the mechanisms. In conclusion, our results suggest that *FADS1* rs174549 polymorphism showed a protective role in etiology of oral cancer. Moreover, fish intake may be an interacting factor that decreases oral cancer risk in individuals with the mutant genotype of rs174549. Further research on gene-diet interaction in oral cancer is warranted to obtain more conclusive outcomes. MATERIALS AND METHODS {#s4} ===================== Study design and population {#s4_1} --------------------------- This hospital-based case-control study was conducted from September 2005 to September 2010 in Fujian, China. 305 oral cancer patients were recruited from the First Affiliated Hospital of Fujian Medical University. As reported previously \[[@R26]\], inclusion criteria of cases were as follows: (1) all cases were newly diagnosed and histologically confirmed primary oral cancer; (2) all cases are Chinese Han population and live in Fujian Province; (3) all cases aged 20 to 80 years. Patients with second primary, recurrent oral cancer, previous radiotherapy or chemotherapy, were excluded from this study. 579 cancer-free control subjects were selected from the physical examination population in the same hospital and frequency-matched to the cases group by gender and age (±3 years). The cancer-free status was ascertained according to the results of physical examination. Those who were direct relatives to the cases or had a previous history of cancer were excluded. The recruiting rate for oral cancer patients was 98.3% and the rate for control subjects was 96.9%. The present study was approved by the Institutional Review Board (IRB) of Fujian Medical University (Fuzhou, China). All participants agreed to this study and signed a consent form. Data and sample collections {#s4_2} --------------------------- All epidemiological data were collected by face-to-face interview using a standardized questionnaire, including information on demographic characteristics, smoking, drinking, diet factors, residential history, and family history of cancer. The subjects were considered smokers if they had smoked at least 100 cigarettes during their lifetime. Alcohol consumers were defined as those who had consumed at least 1 drink/week continuously for at least 6 months. A 5-10 ml blood sample was collected from each subjects with an EDTA-coated vacuum tube and stored at −80°C. Selection of SNPs and genotyping {#s4_3} -------------------------------- Genomic DNA was extracted from whole-blood samples using the Qiagen Blood Kit (Qiagen, Chatsworth, CA). All samples were genotyped by the 50-nuclease TaqMan assay, using the ABI PRISM 7900HT Sequence Detection System (ABI, Foster City, CA). TaqMan primers and FAM- or VIC-labeled probes were designed using the Primer Express Oligo Design software v2.0 (ABI PRISM). The PCR primers were as follows: forward 5'-CCAGCCTGTCTACTTTCCCA-3' and reverse 5'-TCTACGTCCGCT TCTTCCTCAC-3'. Amplification conditions were as follows: 95°C for 10 min, then followed by 40 cycles of 95°C for 5 sec, and 60°C for 30 sec. Allelic Discrimination Sequence Detector Software was used to read the completed PCR plates with an ABI 7900HT Sequence Detector in the end point mode. For the assessment of genotyping results, laboratory personnel were blinded to the case-control status. All assays were carried out in 384-well arrays with 8 no-template controls and 8 duplicated samples in each plate for quality control. Approximately 5% of the samples were randomly repeated for quality control purposes. Genotyping call rates were over 99.0% and the concordance rate reached 100%. Due to genotyping failure of some DNA samples, only 302 cases and 574 control subjects with complete genotyping data can be used for further analysis. Statistical analysis {#s4_4} -------------------- Statistical software R (version 3.1.1) was used for our data analyses. The χ^2^ test was performed for the socio-demographic covariates of the case and control subjects. Hardy--Weinberg equilibrium was conducted using a goodness-of-fit χ^2^ test with linkage disequilibrium analyzer (LDA) software v.1.0 for SNP among the controls. Unconditional logistic regression models were used to estimate odds ratios (ORs) and their 95% confidence intervals (CIs). Interaction between the SNP and fish intake was evaluated using unconditional logistic regression model. Statistical significance was considered at the *P*\<0.05 level. This work was supported by Scientific Research Program of Education Department of Fujian Province (No.JAT160207), Joint Funds for the Innovation of Science and Technology of Fujian province (No.2016Y9033) and Natural Science Foundation of Fujian Province (No.2015J01304). We are great thank to Prof. Dongxin Lin and all staff of State Key Laboratory of Molecular Oncology in Chinese Academy of Medical Sciences for great help with genotyping. **CONFLICTS OF INTEREST** The authors declare no conflicts of interest.

1. Introduction {#sec1-geriatrics-04-00007} =============== NHS is facing an ageing population and older people have a higher prevalence of mental health problems in the general hospital settings as compared to the community. More than a third of patients older than 70 years requiring acute admission have dementia, but only half of these patients have been diagnosed \[[@B1-geriatrics-04-00007],[@B2-geriatrics-04-00007]\]. Patients with dementia often have other associated medical co-morbidities which directly or indirectly could result in poorer outcomes including inpatient falls, longer hospital stays, emergency readmission and institutionalisation \[[@B2-geriatrics-04-00007],[@B3-geriatrics-04-00007],[@B4-geriatrics-04-00007],[@B5-geriatrics-04-00007],[@B6-geriatrics-04-00007]\]. Admission to an acute hospital can be distressing and disorientating for a person with dementia and is associated with a decline in cognitive and functional ability. The National Audit of Dementia (NAD) in the UK showed a wide variation in the quality and approach of care for acutely unwell patients with dementia admitted to the hospital \[[@B7-geriatrics-04-00007]\]. Dementia care relating to multidisciplinary assessment, staffing levels, staff education and training; rehabilitation and discharge planning; and access to certain specialist services including Psychiatry Liaison service can be variable and sub-optimal \[[@B8-geriatrics-04-00007]\]. This could potentially increase the risk of adverse clinical outcomes. The objective of this study is to record the demographics, and patient characteristics including medical co-morbidities and polypharmacy, to understand and benchmark the clinical outcomes of those patients with dementia admitted acutely across three acute sites within Aneurin Bevan University Health Board (ABUHB). This would help us to explore quality initiatives to improve our patient care and plan our future services accordingly. The secondary objective is to study the predictors of poor outcome including 30-day readmission rate, inpatient mortality and discharge to a new care home. 2. Methods {#sec2-geriatrics-04-00007} ========== 2.1. Study Design {#sec2dot1-geriatrics-04-00007} ----------------- This was a retrospective observational cohort study based on analysis of the existing data for all the patients with dementia admitted acutely to ABUHB. 2.2. Setting {#sec2dot2-geriatrics-04-00007} ------------ All patients with dementia admitted acutely to one of the three acute hospital sites (Royal Gwent Hospital, Nevill Hall Hospital and Ysbyty Ystrad Fawr) within ABUHB were included. 2.3. Data and Statistical Analysis {#sec2dot3-geriatrics-04-00007} ---------------------------------- Information on demographics, medical co-morbidities, and clinical outcomes including inpatient fractures, discharge to a new care home, readmission and mortality was extracted electronically from the clinical workstation, clinic letters and coding from 1st January 2016 to 30th June 2016. Readmission and Mortality data was reviewed for 2 years until 30th June 2018. Data on inpatient falls was extracted from DATIX, which is a web-based patient safety software for healthcare risk management which includes incidents of inpatient falls. The description of the study cohort was completed and baseline characteristics of all patients have presented as means ± standard deviation. The *t*-tests were used to compare the mean baseline characteristics of patients, and chi-square tests were used to compare categorical variables. Mortality rates were compared using the difference between proportions test. Sub-analysis for predictors of poor outcome (inpatient mortality, discharge to a new care home and readmission within 30 days) were explored using multivariate analysis for age, gender, Charlson Comorbidity Index (CCI), number of drugs, number of ward transfers, length of stay any fragility fracture (including hip, wrist, spine, humerus and pelvis) were studied using a Binomial Linear Model with logit link function. We have considered 7 separate factors that we believe may be predictors of adverse outcomes. These can act independently or may interact with other factors in determining the outcomes, so the multivariate analysis is somewhat complex. The multivariate analysis results in likelihood scores for each factor or combination of factors. All statistics were conducted using STATISTICA StatSoft data analysis software system, version 9.1 (StatSoft, Inc., 2010, Tulsa, OK, USA). The generalised linear and non-linear models building module of the Statistica statistics package used binomial modelling with logit linking. *p*-values ≤ 0.05 were taken to be statistically significant. Ethical approval was not required for this service evaluation as this work does not constitute a research study according to the Health Research Authority decision tool. In addition, this service evaluation was completed based on the recommendations of the NAD, which is a clinical audit programme, looking at the quality of care received by people with dementia in general hospitals. However, all questions and forms required to carry out the study were sent to the ABUHB Research and Development (R&D) Department, to assess risks to patient identification and the Health Board. The study was approved by the R&D as a service evaluation as patients were not directly interviewed and no identifiable patient data was recorded. The study was compliant with the personal data protection regulation in the U.K. 3. Results {#sec3-geriatrics-04-00007} ========== 3.1. Patient Characteristics {#sec3dot1-geriatrics-04-00007} ---------------------------- The total number of emergency admissions above 18 years and above 65 years across three acute sites in the whole year 2016 were 37378 and 21437 respectively. A total of 2474 acute admission were recorded in the year 2016 from the 1770 acute dementia patients. We studied 953 consecutive dementia patients from 01st January 2016 to 30th June 2016 who had a total of 1167 episodes of acute admissions. The mean age on admission was 84.5 ± 7.8 years. The proportion of females was 63.5% and the mean age of females (85.2 ± 7.6 years) was significantly higher in comparison to the mean age of males (82.0 ± 7.7 years) (*p* \< 0.001). About 70% (n = 670/953) were previously living in their own homes and 26.34% (n = 251/953) were admitted from care homes. Mean CCI and the number of drugs was 6.05 ± 1.5 and 5.1 ± 2.1 respectively. 15.4% (147/953) patients were on antipsychotics. 6% of patients had 3 more or transfer during an index admission. 3.2. Clinical Outcomes {#sec3dot2-geriatrics-04-00007} ---------------------- Overall mean hospital stay was 19.4 ± 27.2 days (median 9 days; range 0--326 days). 3.4% (32/953) patients were excluded from analysis for discharge destination due to coding errors noted in the Clinical Work Station. 59.5% patients admitted from their own homes (n = 399/670) were discharged back to the original residence, 21.6% (145/670) were discharged to a new care home and 18.1% (121/670) died. Among those admitted from care homes, 79.7% (200/251) patients were discharged back to original care home and 20.3% patients (51/251) died. The rate of discharge to a new care home was an approximately 1.68 times higher rate than the patients being originally admitted from a care home (251 referred from a care home but 51 died, 145 discharged to a new care home, therefore total CH discharges = 345). Overall 1.6% patients (15/953) died within one day, 2.2% (21/953) died within 2 days, 2.4% patients died (23/953) within 3 days and 51 (5.3%) died within 7 days. However, 4.4% patients (11/251) admitted from care home died within 3 days as compared to 1.7% (12/690) admitted from the community, which is significantly higher (*p* = 0.017, a difference in proportion test). Overall inpatient mortality was 16.0% (153/953). The 30-days, 90-days and one-year mortality were 22.3% (213/953), 29.6% (283/953) and 49.2% (469/953) respectively ([Figure 1](#geriatrics-04-00007-f001){ref-type="fig"}). The mortality rate of those admitted form care homes as inpatient and at one year was 20.3% (51/251) and 66.1% (166/251) respectively. In comparison, the mortality rate of those admitted from own home as inpatient and at one year was 18.1% (12/670) and 43.6% (292/670) respectively. The observed mortality rates between patients admitted from home or from a care home were not significant for inpatient mortality (*p* = 0.45) but were highly significant for one-year mortality (*p* \< 0.001), both using a difference in proportions test. Kaplan--Meier survival curve is shown in [Figure 2](#geriatrics-04-00007-f002){ref-type="fig"} to show the proportion of patients with acute dementia living over two years or more. Overall 30-day readmission rate was 17.2% (138/800) with a mean hospital stay of 14.6 ± 17.9 days. Nearly half of dementia patients (49.4%, 395/800) were readmitted once over one year. Patients admitted two and three times over two years follow up were 32% (259/800) and 17% (138/800) respectively. The total number of inpatient falls in ABUHB across three acute sites including rehabilitation community hospital beds between 1st January 2016 and 30th June 2016 were 1799. The total occupied bed days (OBD) were 225812, giving the rate of inpatient falls of 7.96/1000 OBD. The rate of falls in general medical beds under unscheduled care during this period were 9.0/1000 OBD and in the mental health division were 9.9/1000 OBD. The total number of falls for acute dementia patients during the same period of 6 months was 267, affecting 109 patients. The range of falls were 1 to 17 and mean inpatient falls among those with dementia was 2.44. The inpatient falls rate for dementia patients, based on the 18,515 bed days was 14.4/1000 OBD which is significantly higher (1.8 times) as compared to overall general medical beds inpatient falls rate (*p* \< 0.0001, the difference in proportions test). 5% dementia patients (48/953) sustained a fracture following an inpatient fall during an index admission. Half (24/48) of inpatient fractures were the hip fracture. The incidence of inpatient hip fracture was 2.5% (n = 24/953). Other fractures were pelvis = 5; wrist = 2; humerus = 2; vertebrae = 1; ankle = 2; others = 12. 3.3. Predictors of Poor Outcomes {#sec3dot3-geriatrics-04-00007} -------------------------------- Sub-analysis were completed to appraise predictors of adverse outcomes. For the risk of inpatient mortality, the best prediction required all factors plus the interaction between age and CCI, with a likelihood score of 23.6. The most predictive single factor was age, with a likelihood score of 9.6. The best two predicting factors were age and the length of stay with a likelihood score of 15.26, and the best three factors were age, length of stay and presence of fragility fracture with a likelihood score of 19.0. These three factors accounted for 80% of the maximum likelihood of the risk. For the risk of discharge to a new care home, the best prediction required all factors plus the interaction between age and CCI, with a likelihood score of 68.98. The most predictive single factor was the length of stay, with a likelihood score of 55.9. The best two predicting factors were age and the length of stay with a likelihood score of 67.8, and the best three factors were age, length of stay and number of ward transfers with a likelihood score of 68.5. These three factors accounted for 99% of the maximum likelihood of the risk. For the risk of readmission within 30 days, the best prediction required all factors plus the interaction between age and CCI, with a likelihood score of 15.8. The most predictive single factor was the number of drugs, with a likelihood score of 11.5. The best two predicting factors were the number of drugs and the length of stay with a likelihood score of 13.5, and the best three factors were the number of drugs, length of stay and CCI with a likelihood score of 14.8. These three factors accounted for almost 94% of the maximum likelihood of the risk. 4. Discussion {#sec4-geriatrics-04-00007} ============= Worldwide, the numbers of people living with dementia will increase from 50 million in 2018 to 152 million in 2050, a 204% increase \[[@B9-geriatrics-04-00007]\]. More than 850,000 people are affected in the UK and the number of patients with dementia will increase rapidly with the ageing population \[[@B10-geriatrics-04-00007]\]. This rising prevalence of dementia will have a significant impact on health and social care costs. Over 70% of inpatients are over the age of 65 years and caring for acutely unwell older patients is challenging for healthcare systems \[[@B1-geriatrics-04-00007]\]. 30% of all inpatients in general hospitals could have dementia, who are particularly more challenging due to multiple co-morbidities, inability to communicate their care needs and atypical presentation \[[@B1-geriatrics-04-00007]\]. People with dementia have the right to fair and equitable treatment but care needs may not be easily recognised and are often misinterpreted to be manifestations of dementia itself \[[@B11-geriatrics-04-00007]\]. Therefore, caring for people with dementia inevitably places pressure on hospitals to provide safe high-quality care. The risk of inpatient death in patients with dementia is higher (11.8%) as compared to those without dementia (6.6%) \[[@B5-geriatrics-04-00007]\]. Delayed or under-recognition of dementia among patients by the medical staff is one the key reason for poorer clinical outcomes. In this study, inpatient mortality was 16%. A similar prospective observational study carried out in an urban tertiary Irish referral centre has also reported inpatient mortality of 15% \[[@B2-geriatrics-04-00007]\]. We have observed that one-year mortality was higher for care home patients (66.1%) as compared to those form their own homes (43.6%). One-year mortality in older care home residents in England and Wales has been reported as 26.2% but we are not aware of any study which has reported mortality for hospitalised acute dementia patients from care home \[[@B12-geriatrics-04-00007]\]. In this study, overall 2% patients died within the first two days of the hospital admission. The proportion of patients dying within the first 3 days was significantly higher for those admitted from a care home. This does not only warrants further studies but suggest an enhanced integrated partnership working among old age psychiatry, care of the elderly and community teams to consider advanced care planning and supporting last days of life in the preferred place of residence. Dementia is associated with impaired mobility and people with dementia are at two times higher risk of inpatient fall and adverse outcomes including discharge to a new care home and prolonged length of stay when compared to those with no cognitive impairment \[[@B3-geriatrics-04-00007],[@B13-geriatrics-04-00007]\]. In this study, inpatient falls rate for acute dementia patients was 1.8 times as compared to all patients admitted during 6 months. Patients with dementia are at a 3-fold increase risk of preventable readmissions and require more healthcare services than those without dementia \[[@B14-geriatrics-04-00007]\]. The 30-day readmission rate in this study was 17.2% but half of the patients were readmitted over one-year. The point prevalence of acute delirium among inpatients has been reported between 17.6% and 20.7% using different diagnostic criteria in the same population, with a prevalence of 34.8% above 80 years \[[@B15-geriatrics-04-00007]\]. A meta-analysis based on pooled findings from multiple small studies has reported a prevalence of delirium on admission in 10--31% of medical inpatients and 3--29% during hospital stay could develop acute delirium \[[@B16-geriatrics-04-00007]\]. There is a dearth of definite and accurate data from the hospitals or Health Board over a longer period. In this study, documented delirium was mentioned in the 9.6% of e-discharges. Based on NAD report 2016--2017, it is likely that delirium in this study has been under recognised and under-reported. Nevertheless, the prescription of antipsychotic drugs in 15.6% of dementia patients might suggest that these patients were receiving treatment for delirium. Therefore, a definitive and accurate determination of in-hospital delirium prevalence, using standardized delirium instruments, is needed. Previous research found that smartphone application can enhance doctors' awareness of delirium and enhance diagnostic accuracy \[[@B17-geriatrics-04-00007]\]. This will augment treating delirium based on the current evidence and recommendations \[[@B18-geriatrics-04-00007]\]. The best predicting factors for poor outcomes including inpatient mortality, discharge to a new care home or readmission within 30 days were age, length of stay, co-morbidity burden and number of drugs. The most predictive single factor for inpatient mortality was age as shown previously \[[@B19-geriatrics-04-00007]\]. This study has several strengths. It is in line with the recommendations of NAD and would support health care organisations to establish a pragmatic approach to record such clinical outcomes as part of clinical governance. We have explored wider clinical outcomes for acutely unwell dementia patients including readmission and long-term mortality. Impact of hospitalisation was explored with respect to inpatient falls. This study has also analysed predictors of poor outcomes which could help clinical staff to streamline care plan towards the end of life priorities and supporting families and carers to formulate advance care plans, maintaining dignity during last days of life. We learnt that the Health Board does not keep separate complaint data for dementia patients and it was not pragmatic to analyse complaints data. An effort has been taken to benchmark the existing clinical outcome to streamline the processes for reporting dynamic data to Dementia Board and Executive team on regular basis. This could lead to an introduction of quality initiatives to not only measure but minimise the impact of hospitalisation, particularly inpatient falls, dehydration and pressure ulcers. We acknowledge the low power of this study as a weakness of the study. We also acknowledge that the data is only from one Health Board, therefore it cannot be generalised and we have not examined the comprehensive impact of hospitalisation including delirium, dehydration and pressure sores. This study has not completed an in-depth analysis for reasons for readmission or cause of death. We also acknowledge that this is a retrospective study, therefore data on type of dementia, severity of dementia, caregiver burden, carers' views, health-related quality of life, work-related staff stress and the number of patients requiring one-to-one care is not being reported. Furthermore, this study did not report the prevalence of stroke \[[@B20-geriatrics-04-00007]\] and depression \[[@B21-geriatrics-04-00007]\] which are commonly associated with dementia and lead to adverse outcomes. Further studies to overcome these limitations is warranted. With the growing number of people with dementia and associated adverse clinical outcomes, training of our future healthcare professionals in the care of older people is needed. Lack of appropriate training to complete dementia assessment and discharge planning could be a factor for sub-optimal care and poor clinical outcomes \[[@B8-geriatrics-04-00007]\]. Nursing staff training on geriatric giants have shown to reduce work-related stress and similarly, dementia teaching to medical students and foundation doctors have shown to improve competencies but the impact of such interventions on clinical outcomes has not been studied yet \[[@B22-geriatrics-04-00007],[@B23-geriatrics-04-00007],[@B24-geriatrics-04-00007]\]. Psychiatry Liaison services like Rapid Assessment Interface and Discharge (RAID) have shown improved clinical outcomes for acutely unwell patients with mental health problems \[[@B25-geriatrics-04-00007],[@B26-geriatrics-04-00007],[@B27-geriatrics-04-00007]\]. Further investment and developing more links with the community services could possibly avoid admitting very frail patients to the hospital, particularly those with advanced dementia. The standard recommendations are to commission dynamic research using a fixed methodology to estimate changes in dementia incidence, prevalence and particularly mortality over time \[[@B28-geriatrics-04-00007]\]. However, there is a dearth of research that tracks all three parameters. Regular evaluation of clinical outcome data to benchmark services for acute older people has been proposed by the NAD \[[@B7-geriatrics-04-00007]\]. Such data could be used to introduce quality improvement initiatives to improve patient-related clinical outcomes in this vulnerable group \[[@B29-geriatrics-04-00007]\]. 5. Conclusions {#sec5-geriatrics-04-00007} ============== Acute patients with dementia have a higher risk of adverse outcomes and the impact of hospitalisation. Improving safety and quality of care for patients with dementia in acute hospitals will benefit all patients and is a high priority for the NHS. Further similar studies will improve organizational understanding of clinical outcomes for acute dementia patients. This would also facilitate quality improvement initiatives to improve patient care and modernisation of community service. The authors are grateful to all members of the Department of Geriatric Medicine, Ysbyty Ystrad Fawr for their continued support for research activities. The authors would especially like to thank all staff and therapists working in ABUHB. The authors would like to thank Jane Power for her administrative support and Salma Zabaneh/Migle Turauskaite for her contributions with references and proofreading. Authors are very grateful to Research and Development, Aneurin Bevan University Health Board for their support and guidance. Authors would like to thank Kate Hooton and Beverly Lewis for providing inpatient falls data. No external funding was applied. I.S. was responsible for the study concept, designed the service evaluation and wrote the first draft. All authors contributed towards data collection, data analysis, drafting and revising the paper and agree to be accountable for all aspects of the work. No external funding was applied. The authors report no other conflict in this work. {#geriatrics-04-00007-f001} {#geriatrics-04-00007-f002}

Many enzymes make use of covalent catalysis to achieve substantial rate enhancements, often by recruiting cofactors such as PLP^[@R1]^, TPP^[@R2]^ and flavins^[@R3]^. To ensure high turnover, these enzymes are inherently required to ensure both rapid and reversible cofactor-ligand adduct formation. In the case of the UbiD enzyme family, reversible decarboxylation has been suggested to occur via a 1,3-dipolar cycloaddition process between the substrate and the UbiD-cofactor, prenylated FMN (prFMN)^[@R4]^, enabled by the azomethine ylide character of the latter. While 1,3-dipolar cycloaddition is commonly used in organic synthesis^[@R5]^, here we describe this particular process in enzymatic catalysis. Diels-Alder reactions have also recently been found to be catalysed by dedicated enzymes^[@R6]^, but these are not reported to be reversible. Similarly, 1,3-dipolar cycloaddition reactions such as biocompatible 'click'-chemistry are not considered reversible^[@R7]^, although the possibility of force-induced cycloreversion is a recent and controversial topic^[@R8]--[@R11]^. In the case of the fungal ferulic acid decarboxylase **(**Fdc) UbiD-enzyme^[@R12]^, 1,3-dipolar cycloaddition between prFMN^iminium^ and the cinnamic acid substrate ([Fig 1](#F1){ref-type="fig"}) is proposed to lead to an initial pyrrolidine cycloadduct (**Int1**) that supports decarboxylation concomitant with ring opening, resulting in formation of a distinct prFMN-alkene adduct (**Int2**). A conserved Glu residue is proposed to donate a proton to the alkene moiety, leading to a second pyrrolidine cycloadduct (**Int3**). Finally, the reaction cycle concludes with cycloelimination of **Int3**, leading to alkene product formation and release. The reaction is readily reversible at elevated \[CO~2~\] ^[@R12]^. There is no direct experimental evidence that any of the proposed intermediates exist, although indirect evidence supporting this mechanism has been reported^[@R13]^. Alternative mechanisms proposed to occur with distinct forms of the cofactor have been discounted^[@R14]^. However, it remains unclear how the enzyme is able to ensure reversibility of the 1,3-dipolar cycloaddition steps. We aimed to determine whether detailed structural insights could be gained into catalysis by *A. niger* Fdc1, an enzyme used as a model prFMN decarboxylase by virtue of the fact it yields atomic resolution crystal structures^[@R12]^. Results {#S1} ======= Structure of the *An*Fdc1 substrate complex {#S2} ------------------------------------------- Catalytically competent *An*Fdc1 crystals were made by avoiding illumination of the protein during sample preparation and crystallisation, preventing photo-induced cofactor isomerisation and inactivation^[@R14]^. Incubation of crystals with cinnamic acid for brief periods of time (typically 10-15 seconds) led to rapid sample deterioration, likely due to styrene accumulation. Electron density obtained with diffraction data from cinnamic acid soaked crystals (prior to crystal dissolution) unambiguously revealed the location of the substrate benzyl moiety above the prFMN^iminium^ isoalloxazine ring system, but revealed density corresponding to multiple conformations of the substrate acrylic acid moiety (*vide infra*). In contrast, rapid soaking and flash-cooling of crystals with alpha-fluorocinnamic acid revealed clear density for the substrate positioned directly above the prFMN^iminium^ azomethine ylide functionality (*i.e.* **Sub** complex, to 1.1 Å resolution; [Fig 2a](#F2){ref-type="fig"}). We could confirm that cinnamic acid binds to the protein in the same conformation, by using inactive *An*Fdc1 crystals that contained FMN rather than prFMN^iminium^ (to 1.26 Å resolution; [Supplementary Fig 1a](#SD1){ref-type="supplementary-material"}). In the case of the alpha-fluorocinnamic acid prFMN complex, the substrate Cα and Cβ carbons are located directly above the prFMN^iminium^ C1' and C4a ([Fig 1](#F1){ref-type="fig"}), at distances of 3.0 Å and 3.4 Å respectively, in a reactive conformation compatible with the proposed cycloaddition. The E282Q *An*Fdc1 variant allows structure determination of the **Int2** covalent adduct {#S3} ----------------------------------------------------------------------------------------- In order to determine whether covalent adducts (either **Int1** or **Int2**) might accumulate in the absence of the Glu282 mediated protonation step, we sought to determine the crystal structure of the inactive *An*Fdc1 E282Q variant^[@R14]^ in complex with cinnamic acid and cinnamic acid derivatives. Indeed, UV-Vis spectroscopy of *An*Fdc1 E282Q incubated with cinnamic acid leads to formation of a distinct spectral species, indicative of adduct formation ([Supplementary Fig 2a](#SD1){ref-type="supplementary-material"}). The E282Q crystal structure obtained by co-crystallisation with cinnamic acid reveals a covalent prFMN-substrate complex, corresponding to the proposed **Int2** species following decarboxylation (to 1.18 Å resolution; [Fig 2b](#F2){ref-type="fig"}). The substrate alkene bond retains the *trans* configuration but substantially deviates from ideal planar geometry, with a torsional angle of τ = \~112° (as opposed to 180° for a planar *trans* configuration) and an elongated Cα=Cβ bond length of 1.37 Å. Exchange of the bound cinnamic acid for alpha-fluorocinnamic acid or pentafluorocinnamic acid readily occurs *in crystallo*, demonstrating a **Sub** ⇌ **Int2 + CO~2~** equilibrium exists in E282Q ([Supplementary Fig 1b,c](#SD1){ref-type="supplementary-material"}). The alpha-fluorocinnamic acid **Int2** structure (to 1.13 Å resolution) directly demonstrates *syn*-pyramidalisation^[@R15]^ of the substrate alkene Cα (pyramidalisation angle of ϕ \~20°), which is likely to be a result of the torsional strain imposed. The observed alkene distortion is a likely consequence of the firm hold the enzyme active site exerts on the substrate phenyl moiety, held in place by I187, I327 and F437 at a position directly above and stacked with the prFMN plane. For all E282Q **Int2** structures, the prFMN C1' is located further out of the prFMN plane by \~0.4 Å when compared to other prFMN single bond adduct species (*i.e*. PDB 4ZA9; [Fig 2c](#F2){ref-type="fig"}). In the case of cinnamic acid E282Q **Int2**, this places the Cβ directly above the prFMN C4a at a distance of 3.04 Å, while the Cα atom is in close proximity (\~3.2 Å) of the location occupied by the Glu282 carboxylate moiety in the wild-type enzyme structure. Structure of cycloadducts **Int1'** and **Int3'** formed with alkyne compounds {#S4} ------------------------------------------------------------------------------ The substrate and product alkyne analogues phenylpropiolic acid and phenylacetylene were used to determine whether cycloadducts **Int1** and/or **Int3**, respectively can be observed in the wild-type enzyme. These compounds can also act as dipolarophiles, but cycloaddition will lead to the double bond-containing 3-pyrroline **Int 1'** rather than the pyrrolidine cycloadducts ([Fig 1](#F1){ref-type="fig"}). Upon addition of either alkyne compound, a clear shift is observed in the enzyme UV-Vis spectrum indicating modification of the prFMN^iminium^ cofactor, with the formation of a species distinct from the E282Q **Int2** adduct ([Supplementary Fig 2b](#SD1){ref-type="supplementary-material"}). Stopped-flow experiments reveal the formation of this alkyne intermediate occurs with *k~obs~* = 101.5 ± 0.4 s^-1^ ([Supplementary Fig 3](#SD1){ref-type="supplementary-material"}), but following prolonged incubation no product formation could be observed ([Supplementary Fig 2c](#SD1){ref-type="supplementary-material"}). To investigate the phenylpropiolic acid complex prior to reaction (i.e. **Inhib** complex), inactive crystals of *An*Fdc1 (with prFMN in a semiquinone radical form) were used, revealing the inhibitor (to 1.29 Å resolution) binds in a similar manner to the alpha-fluorocinnamic acid:prFMN or cinnamic acid substrate:FMN complexes obtained. This positions the alkyne Cα and Cβ directly above the prFMN^radical^ C1' and C4a at distances of 3.3 Å and 3.7 Å ([Supplementary Fig 1d](#SD1){ref-type="supplementary-material"}). In contrast, crystals of active *An*Fdc1 co-crystallised with phenylpropiolic acid (to 1.01 Å resolution) or soaked with phenylacetylene (to 1.10 Å resolution) revealed clear density for a 3-pyrroline cycloadduct (**Int3'**, analogous to the proposed **Int3**) formed between the alkyne moiety and the prFMN^iminium^ ([Fig 3a](#F3){ref-type="fig"}). Hence, surprisingly, the phenylpropiolic acid derived adduct lacked density for the carboxylate moiety, and was highly similar to that obtained with phenylacetylene ([Supplementary Fig 1e](#SD1){ref-type="supplementary-material"}). This suggests decarboxylation has occurred over the prolonged time frame required for co-crystallisation (typically exceeding 24 hours). Indeed, when using a rapid soaking procedure as an alternative approach, a distinct cycloadduct (**Int1'**; to 1.21 Å resolution) is observed that retains the carboxyl moiety attached to the prFMN C1' group (and thus resembles the proposed **Int1**; [Fig 3b](#F3){ref-type="fig"}). We used a mass spectrometry-based detection method to demonstrate slow decarboxylation occurs within a 24 hour time period in solution, confirming the conversion of **Int1'** to **Int3'** mimicking the natural activity of the enzyme with cinnamic acid occurs ([Supplementary Fig 4a](#SD1){ref-type="supplementary-material"}). We then took advantage of the distinct spectral change associated with cycloadduct formation to determine if cycloaddition with natural substrates can also be monitored in solution. Indeed, rapid mixing of the enzyme with cinnamic or sorbic acid substrates revealed spectral changes in the UV-Vis spectrum of prFMN similar to those observed with the alkyne inhibitors ([Supplementary Fig 3b,c](#SD1){ref-type="supplementary-material"}). This suggest that, for these substrates, cycloadduct(s) accumulate under multiple turnover conditions. The *An*Fdc1 active site constrains cycloadduct conformation {#S5} ------------------------------------------------------------ While both **Int1'** and **Int3'** structures confirm 1,3-dipolar cycloaddition can occur with the prFMN^iminium^ cofactor, several features indicate a significant degree of strain in these cycloadduct(s). The phenylpropiolic acid **Int1'** structure deviates from ideal bond parameters both in the position of the inhibitor phenyl group (Cβ out-of-phenyl plane angle is \~24° as opposed to ideal 0°), the deviation of the 3-pyrroline ring (C1'-Cα-Cβ bond angle of \~99°, as compared to ideal 108°), prFMN tetrahedral C4a bond angles (C2-C4a-Cβ bond angle of \~103° *versus* ideal 109.5°of *sp^3^* C4α) as well as the elongated C4a-Cβ bond length of 1.7 Å ([Fig 4a](#F4){ref-type="fig"}). Following decarboxylation, the position of the **Int3'** 3-pyrroline ring shifts ([Fig 3c](#F3){ref-type="fig"}) to alleviate some of the phenyl group strain (Cβ out-of-plane angle reduced to \~16°) as well as the deviation of the 3-pyrroline ring (C1'-Cα-Cβ bond angle of \~111°; Cβ-C4a bond length of 1.65 Å). The protein does not alter in conformation when comparing the **Inhib/Int1'** and **Int3'** structures, with the notable exception of Glu282, which reorients to occupy the substrate/inhibitor carboxylate binding pocket in **Int3'** (a motion similar to the CO~2~ for Glu282 exchange proposed to occur during catalysis, [Fig 1](#F1){ref-type="fig"}). In the latter, Glu282 forms an interaction with the 3-pyrroline substrate-derived Cα (distance \~3.2 Å to either E282 oxygen; [Fig 3a](#F3){ref-type="fig"}). Glu282 mediated proton donation or abstraction would interconvert **Int3'** with the putative prFMN C1' -- alkyne adduct **Int2'** (the complex analogous to **Int2**; [Fig 1](#F1){ref-type="fig"}). However, an 'unstrained' linear C1'-alkyne adduct --with a C1'-Cα≡Cβ angle of 180°-- is sterically incompatible with the *An*Fdc1 active site, suggesting a highly strained **Int2'** (similar to strained cyloalkyne species^[@R15]^) connects the **Int1'** and **Int3'** species, providing a likely explanation for the extremely slow rate of **Int1'** decarboxylation. Mutagenesis alleviates cycloadduct strain and impacts catalysis {#S6} --------------------------------------------------------------- As observed for the **Int2** E282Q structures, the strain observed in the **Int1'** and **Int3'** structures appears to be a direct consequence of the inherent restrictions imposed by the rigid *An*Fdc1 active site. The position of the substrate aromatic group appears constrained to remain π-stacked with the prFMN isoalloxazine ring, by virtue of steric interactions with the Phe437 sidechain located adjacent to the conserved Leu439. This in turn imposes restrictions on the extent by which the prFMN isoalloxazine ring system can deform upon formation of a C4a-Cβ bond with substrate, as the flavin uracil ring is sandwiched between Ala172 on the *si* face and the substrate aromatic group at the *re* face. We constructed the F437L and L439G variants and determined the crystal structures in complex with phenylpropiolic acid, as well as the activity with cinnamic acid. We found both mutations severely compromised enzyme activity as judged from steady state turnover. In the case of L439G, a substantial increase of *K*~M~ was observed, while the F437L variant was only affected in *k~cat~* confirming a key role for Phe437 in catalysis ([Supplementary Fig 5b,d](#SD1){ref-type="supplementary-material"}). In the corresponding phenylpropiolic acid adduct crystal structures for both variants, **Int1'** decarboxylation *in crystallo* was not observed, even following co-crystallisation and prolonged incubation (up to several weeks) prior to X-ray exposure. This lack of decarboxylation activity with **Int1'** was confirmed in solution ([Supplementary Fig 4b](#SD1){ref-type="supplementary-material"}). In the case of the L439G **Int1'** crystal structure (to 1.12 Å resolution), a modest reorientation of the substrate carboxyl group along with a minor reduction of the Cβ out-of-plane angle is observed, associated with a corresponding reorientation of the F437 phenyl ring ([Fig 3d](#F3){ref-type="fig"}). The void created by the L439G mutation is filled with a water molecule located within hydrogen bonding distance of the **Int1'** carboxylate, reducing the hydrophobicity of the carboxylate binding pocket. In addition, the L439G void was partially filled by the phenyl group of a second phenylpropiolic acid molecule, bound at the entrance of the active site. This observation offers a possible explanation for the drastic increase of the observed *K*~M~ for cinnamic acid in the L439G (as opposed to F437L) mutant, as similar binding of a second cinnamic acid molecule could be required for decarboxylation by L439G. The hydrophobic nature of the carboxylic acid binding pocket has been implicated in the mechanism of other decarboxylases^[@R16],[@R17]^. In marked contrast, the F437L mutation does not affect the carboxylate binding or conformation, but subtly alters the position of the substrate phenyl moiety (structure obtained to 1.08 Å resolution; [Fig 3e,f](#F3){ref-type="fig"}). This reorientation, while retaining a strained conformation of the Cβ out-of-plane angle of \~25°, is associated with elongation of the Cβ-C4a bond to 1.8 Å. It thus appears the Phe437 side chain counteracts the forces exerted on the prFMN 3-pyrroline ring through enzyme interactions with the **Int1'** carboxylate. Smaller dipolarophile compounds form dead-end cycloadducts {#S7} ---------------------------------------------------------- Unfortunately, other mutations at the Phe437 position that could lead to further reduction of strain imposed by the 437 side-chain led to variants that did not bind the prFMN cofactor. We therefore co-crystallised the WT enzyme with 2-butynoic acid, avoiding the enzyme imposed adduct strain by interactions with the missing substrate phenyl moiety. In this case, an **Int1'^butynoic^** structure was obtained to 1.02 Å resolution, revealing no decarboxylation had occurred ([Fig 3g](#F3){ref-type="fig"}). Furthermore, the **Int1'^butynoic^** is markedly different from the strained phenylpropiolic acid derived **Int1'** structure ([Fig 3h](#F3){ref-type="fig"}). Little to no features associated with a strained **Int1'** configuration can be observed in the case of **Int1'^butynoic^**, a consequence of repositioning of the prFMN^iminium^ C4a and associated uracil plane. This repositioning leads to a near ideal conformation of the **Int1'^butynoic^** 3-pyrroline ring, with a reduced Cβ-C4a bond length of 1.56 Å and a C1'-Cα-Cβ bond angle of \~107°, but is incompatible with the presence of a substrate phenyl group held within the context of either the WT or F437L variant structures. As a consequence of the uracil plane repositioning, the prFMN O1 has moved out of the oxyanion binding-pocket formed by Q191 and two water molecules. Decarboxylation of **Int1'**, and by similarity **Int1**, is likely assisted by the relatively hydrophobic nature of the carboxylate binding pocket, and ring opening of the strained cycloadduct to **Int2'/Int2**. It is however unclear what drives the cycloelimination step from **Int3** to product release, a process not observed for **Int3'**. Careful analysis of electron density associated with rapidly flash-cooled WT crystals incubated with cinnamic acid reveals this can be modeled as a mixture of 30% **Int3** and a 40% product complex (the remainder corresponding to the inactive hydroxylated prFMN species), suggesting cycloelimination is the rate limiting step under these conditions, as has previously been suggested^[@R13]^ (to 1.24 Å resolution; [Fig 2d](#F2){ref-type="fig"}). As observed with the **Int3'** structure, the **Int3** Cβ has an out-of-plane angle of \~13°, with additional strain observed for the Cα-Cβ-Cγ angle of \~122°, both associated with the tight hold the enzyme exerts on the substrate phenyl moiety and resulting in an C1' *endo* envelope conformation of the pyrrolidine ring (as opposed to the N *endo* conformation of the **Int3'** enforced by the Cα-Cβ double bond ([Fig 4a](#F4){ref-type="fig"}). Hence, to determine how **Int3** strain might assist cycloelimination, we determined the structure of an **Int3^crotonic^** species (to 1.39 Å resolution), obtained by co-crystallisation with crotonic acid (2-butenoic acid; [Fig 2e](#F2){ref-type="fig"}). A comparison of **Int3^crotonic^** with the cinnamic acid **Int3** ([Fig 2f](#F2){ref-type="fig"}) reveals the **Int3^crotonic^** pyrrolidine ring adopts a distinct envelope conformation with the N in *endo* position, that allows the prFMN^iminium^ C1' to adopt a more in-plane position, resembling the **Int3'** geometry. The **Int3^crotonic^** conformation cannot be adopted by the cinnamic acid **Int3**, as severe clashes between the substrate phenyl ring and the active site occur. We suggest the lack of strain in the **Int3^crotonic^** species is coupled to the fact no cycloelimination can be observed for this species, while a strained conformation such as that observed for cinnamic acid **Int3** is required for progression along the catalytic cycle. Combining the various crystal structures of **Sub/Inhib**, **Int1'/Int2** and **Int3'** complexes with our solution data leads to a comprehensive overview of the *An*Fdc1 catalytic cycle ([Fig 4](#F4){ref-type="fig"}). The rigid *An*Fdc1 active site brings the substrate into close proximity of the prFMN^iminium^ cofactor, aligning the Cα-Cβ double bond (or triple bond in case of alkyne inhibitors) with the C1'-N5-C4a azomethine ylide to poise the complex for cycloaddition. Formation of a pyrrolidine adduct between a similar putative encounter complex of cinnamic acid and free prFMN^iminium^ in solution would be accompanied by repositioning of both the Cα-Cβ substituents (*i.e.* carboxylate/benzyl group), as well as the prFMN uracil moiety. This is a direct consequence of the ring formation and the accompanying *sp^2^* to *sp^3^* change for substrate Cα/Cβ and prFMN C1'/C4a carbon atoms. However, within the confines of the enzyme active site, the extent to which reorientation can occur is limited. In fact, the enzyme active site appears evolved to have maximum complementarity with the substrate and product complexes, and hence constrains both the substrate carboxylate and phenyl moieties to adopt a position resembling the substrate complex throughout catalysis. In the case of the **Int1'^butynoic^** and **Int3^crotonic^** adducts, the lack of a phenyl moiety allows for a more relaxed conformation of the substrate/prFMN^iminium^ adduct within the confines of the active site. Crucially, neither **Int1'^butynoic^** decarboxylation nor **Int3^crotonic^** cycloelimination are observed suggesting that strain contributes to catalysis for both these steps. Computational studies provide further insight into the *An*Fdc1 catalytic cycle {#S8} ------------------------------------------------------------------------------- An active 'cluster' site model comprising \~ 230 atoms was built from the E282Q **Int2** X-ray crystal structure. The positions of peripheral atoms were fixed to maintain a crystal structure-like geometry, in addition to constraining the position of the F437 and L439 side chains in most calculations to allow their contribution to strain in the adducts to be investigated. Stable models of each species as identified in [Fig 1](#F1){ref-type="fig"} were found for cinnamic acid, phenylpropiolic acid and crotonic acid. In the case of cinnamic acid, cycloaddition (**Sub** → **Int1**) was found to occur as a 2-step reaction, via a ring-open covalent adduct (**Int1^open^**), which possesses an unusual Cβ--C4a interaction (2.60 Å) in addition to a Cα--C1' bond (1.58 Å). The barrier to formation of the ring-closed **Int1** intermediate is small ([Fig 5](#F5){ref-type="fig"}), so the reaction may effectively occur as one asynchronous step. In contrast, no stable **Int1^open^**-type intermediates were observed for phenylpropiolic or crotonic acid cycloaddition, which also leads to significantly more stable **Int1** cycloadducts. Cinnamic acid decarboxylation occurs from the ring-open **Int1^open^** species and the **Int1^open^** → **Int2^CO2^** potential barrier is relatively low at \~ 16 kJ mol^-1^ ([Fig 5](#F5){ref-type="fig"}). The barriers for phenylpropiolic and crotonic acid decarboxylation were not determined, but may be much higher than for cinnamic acid and/or occur via a different mechanism, as these reactions cannot proceed via an **Int1^open^** intermediate and the decarboxylation of these is then significantly endothermic. In all cases, the **Int2^CO2^**, which contain noncovalently bound CO~2~, remain ring-open with broken (\~3 Å) Cβ-C4a bonds. The crotonic acid derived **Int2^CO2^** is most stable as it can adopt an unstrained ring-open conformation with near-ideal Cα--Cβ geometry. In contrast, both cinnamic acid and phenylpropiolic **Int2^CO2^** intermediates adopt strained conformations with non-linear Cα--Cβ--Phenyl angles. Relaxation of the side chain constraints of Phe437 and Leu439, as a proxy of the F437L and L439G variants, relieves \~14 kJ mol^-1^ of strain from the cinnamic acid derived **Int2^CO2^** adduct consistent with the role of these residues in constraining the geometry of the phenyl moiety of the adducts. Once the CO~2~ has vacated the active site, Glu282 mediated protonation of Cβ leads to the formation of a ring-closed **Int3** species for all compounds. The release of CO~2~ and Cβ protonation were not studied in detail, as these reactions require substantial rearrangement of Glu282, which would require much larger models. Cycloelimination (**Int3** → **Prod**) to form non-covalently bound product complexes is exothermic and likely to be rate limiting in all three cases ([Fig 5](#F5){ref-type="fig"}). Formation of styrene is the least exothermic (38 kJ mol^-1^), while cycloelimination of phenylacetylene is highly unfavorable, being 81 kJ mol^-1^ uphill and likely proceeding via a barrier \> 100 kJ mol^-1^ (not determined). Similarly, the barrier to propylene cycloelimination is estimated at \~100 kJ mol^-1^. This is in agreement with our observation that turnover does not occur with either crotonic or phenylpropiolic acid, with the reaction stalling at the **Int3** stage. Crucially, in the case of cinnamic acid, the cycloelimination barrier was determined to be feasible, at 64 kJ mol^-1^ and 78.5 kJ mol^-1^ in the constrained and F437/L439-relaxed models respectively. The 14.4 kJ mol^-1^ difference is at least qualitatively consistent with the \~30 fold decrease in *k~cat~* observed in the F439L variant (which retains some element of strain). Discussion {#S9} ========== In the case of the Fdc branch of the UbiD family, our data unambiguously establishes that the enzyme relies on covalent catalysis by the unusual prFMN cofactor through a reversible 1,3-dipolar cycloaddition process. The fact the active site is highly complementary in shape to the substrate/prFMN^iminium^ complex prior to cycloaddition has the serendipitous consequence it does not yield to allow pyrrolidine ring substituents to adopt the ideal conformation. This results in the formation of strained intermediates, ensuring rapid progression through the catalytic cycle. While substrate distortion/ground state destabilization has long been suggested to contribute to enzyme catalysis^[@R18],[@R19]^, it has only been conclusively demonstrated in a limited number of cases^[@R20],[@R21]^, and rarely includes atomic resolution insights^[@R22]--[@R25]^. Our study reveals in *An*Fdc1 distortion occurs specifically for covalent substrate-cofactor adducts only, and the strain is directly linked to progression along the reaction coordinate. Hence, by targeted destabilization of intermediate species only, *An*Fdc1 harnesses 1,3-dipolar cycloaddition as a readily reversible reaction. This aligns with the proposals from Albery and Knowles^[@R26]^ for evolution of an efficient enzyme by control of the internal thermodynamics of the bound states. Interestingly, many members of the UbiD family act directly on aromatic substrates, and can act in the carboxylative direction as suggested for the anaerobic degradation of benzene^[@R27],[@R28]^. (De)carboxylation of aromatic substrates presents a large inherent barrier to cycloaddition due to transient substrate dearomatisation, as well as scope for formation of a highly stable **Int2** adduct by re-aromatisation^[@R29]^. How these problems are overcome, particularly in view of whether the prFMN-dependent catalysis in these enzymes is similar to the Fdc branch, awaits a more detailed analysis of additional members of the UbiD family. Finally, our data suggests mechanochemical reversion of 1,3-dipolar cycloaddition reactions is feasible, providing further support to the possibility of developing mechanoresponsive materials through reversible click-chemistry^[@R8]--[@R11]^. Our data show Nature is able to harness and control 1,3-dipolar cycloaddition chemistry by combination of a unique cofactor with strict mechanochemical control. Methods {#S10} ======= See [Supplementary Information](#SD1){ref-type="supplementary-material"} for methods. Supplementary Material {#SM} ====================== This work was supported by the grants BBSRC BB/K017802 and ERC pre-FAB 695013. The authors acknowledge the assistance given by the use of the Manchester Protein Structure Facility. We thank Diamond Light Source for access (proposal numbers MX12788 and MX17773) that contributed to the results presented here. D.L. is a Royal Society Wolfson Merit Award holder. **Data availability** The data generated and analysed in this study, including the computational modeling data associated with all figures, are available from the corresponding authors upon reasonable request. The diffraction data and corresponding atomic models have been deposited in the PDB under accession codes 6R3G, 6R3F, 6R3I, 6R2Z, 6R30, 6R3O, 6R2T, 6R2R, 6R3N, 6R3L, 6R3J, 6R2P, 6R34, 6R33 and 6R32. **Author contributions** SSB cloned, expressed and purified *An*Fdc1 and various variants. SSB determined all crystal structures with guidance from DL. KAPP generated various *An*Fdc1 variants, and collected solution data with cinnamic acid. AS assisted with crystallisation of the **Int3^crotonic^** state. SAM assisted with *An*Fdc1 reconstitution. IG assisted with crystallisation of FMN substitued AnFdc1 and the **Int3^butynoic^** state (with help from IK). KF assisted with protein purification and solution data collection and analysis. SH performed all computational studies. All authors discussed the results and participated in writing of the manuscript. DL initiated and directed this research. **Competing interests** The authors declare no competing interests. {#F1} {ref-type="supplementary-material"} shows the product complex for comparison). **e.** Omit electron density contoured at 4 sigma corresponding to **Int3^crotonic^**. **f.** Overlay between the **Int3** and **Int3^crotonic^** acid covalent adducts formed.](EMS83924-f002){#F2} {#F3} {ref-type="fig"} and [2f](#F2){ref-type="fig"}).](EMS83924-f004){#F4} {ref-type="supplementary-material"}), with potential energy normalised to the **Sub/Inhib** and **Int3** energies. The relaxed cinnamic acid energies were determined for models where restraints on the side chains of Phe437 and Leu439 were removed, which alleviates some of the strain on the Cβ--Ph bond ([Supplementary Fig. 8](#SD1){ref-type="supplementary-material"}). The two phenylpropiolic acid **Int1** species interconvert via rotation of the Cα--CO~2~ bond and the conformation with the carboxyl group in the phenyl plane is the higher energy species and is not stable in the DFT model used here. The crotonic acid **Int3 → Prod** transition state actually represents a minimum energy crossing point ([Supplementary Figs. 9, 10](#SD1){ref-type="supplementary-material"}).](EMS83924-f005){#F5}

"Case Report: Potential Arsenic Toxicosis Secondary to Herbal Kelp Supplement" by [@b1-ehp0115-a00574] is fundamentally flawed, both scientifically and with regard to the regulation of dietary supplements. [@b1-ehp0115-a00574] claimed to have found "detectable levels of arsenic in eight of the nine kelp herbal supplements, ranging from 1.59 ppm to 65.5 ppm by dry weight (1.59, 2.28, 9.55, 9.97, 10.5, 24.1, 34.8, and 65.5 ppm)," with a median of 10.23 ppm. In this instance, concentrations are irrelevant without disclosing the mass of the capsules. This would allow for calculation of the potential exposure to arsenic; any valid scientific argument on toxicity has to be based on exposure levels and daily intake, not on concentration. For example, if we applied a 50-mg mass as the capsule mass, this would equate to arsenic concentrations of 0.0795, 0.114, 0.478, 0.499, 0.525, 1.21, 1.74, and 3.275 μg/capsule, respectively. With a serving of one capsule per day, this is well below the normal daily intake cited by [@b1-ehp0115-a00574]: > Nonoccupationally exposed individuals \[had\] an average total (inorganic and methylated) arsenic intake of 40 μg/day. U.S. dietary intake of inorganic arsenic has been estimated to range from 1 to 20 μg/day (Schoof et al. 1999). A 500-mg capsule, in effect multiplying the daily intake 10-fold, would still result in all the products below the average daily total intake of 40 μg as cited by [@b1-ehp0115-a00574]. The glaring omission of the mass of the capsules and the subsequent presentation of the data as a concentration allowed [@b1-ehp0115-a00574] to provide a provocative story and headline. Once the real-world metrics are applied, however, the fog is dispersed and these numbers are obviously well within the numbers the authors cited as daily intake values. These data are no longer provocative and make it impossible for kelp supplements to be painted as "unsafe" as the authors suggested. [@b1-ehp0115-a00574] also were not diligent in researching kelp supplements, and they overlooked key references. One important oversight is the [@b4-ehp0115-a00574], which contains a monograph on kelp supplements, with guidelines on arsenic concentration in kelp. The European Pharmacopoeia sets a limit of 90 ppm total arsenic in kelp. Pharmacopoeial monographs are developed over the course of years by experts in the field. The sheer fact that this reference was not cited by [@b1-ehp0115-a00574] further reveals their ignorance on the subject. The application of the Food and Drug Administration (FDA) tolerance level for arsenic as residue in muscle meat of chicken and turkey, and in eggs ([@b7-ehp0115-a00574]) in terms of concentration used by [@b1-ehp0115-a00574] is also not applicable just on product mass alone. For instance, a 4-oz serving of turkey is converted to 113398.0924 mg. Therefore, a 2-ppm limit is applicable and logical based on the mass of the product. This serving of 4 oz turkey at a 2-ppm concentration would obviously result in exponentially greater exposure to arsenic (approximately 2,300 times greater) than a 50-mg kelp dietary supplement capsule at a 2-ppm concentration. To imply that there is toxicity associated with anything---be it a food, pharmaceutical, or dietary supplement---without applying the appropriate metrics is irresponsible and potentially damaging, as well as confusing, to the consumer who may benefit from that product. In addition to the inappropriate use of metrics, [@b1-ehp0115-a00574] did not differentiate between the different species of arsenic present in the kelp samples. This differentiation is significant since as the authors themselves present, "In most cases the toxic moiety is presumably trivalent arsenic in the form of inorganic arsenious acid (arsenite)." In fact, the California Clean Drinking Water Act of 1986 ([@b2-ehp0115-a00574]), commonly referred to as "Proposition 65," sets limits only on inorganic arsenic compounds (oxides). This limit is set at 10 mg/day. The California Proposition 65 limit was determined by taking the no observed effect level, which is defined as "the highest level at which a chemical can be administered to an organism without any adverse effect (for example upon health, growth, development, reproductive capacity or lifetime) being observed" ([@b2-ehp0115-a00574]), and then dividing by 1,000. In addition, the [@b8-ehp0115-a00574] has set a limit of 3 ppm inorganic arsenic. There is no limit for total or organic arsenic compounds. The absence of blood arsenic at the time of poisoning from the study ([@b1-ehp0115-a00574]) is also a relevant and questionable deficiency. With respect to supplement regulation, supplements must be accurately labeled as mandated by the [@b3-ehp0115-a00574] ([@b3-ehp0115-a00574]) and actively monitored by the FDA. Regulatory action is taken when and where appropriate. The FDA, on a number of occasions, has stated that the DSHEA provides all the legislative authority needed to regulate dietary supplements. In testimony before the House of Representatives Committee on Government Reform, Robert E. Brackett (Center for Food Safety and Applied Nutrition, FDA) stated that the > FDA regulates the safety, manufacturing, and labeling of dietary supplements, while \[the Federal Trade Commission\] has primary responsibility for regulating the advertising of these products." ([@b6-ehp0115-a00574]) In conclusion, contrary to the viewpoint of [@b1-ehp0115-a00574], dietary supplements are in fact regulated, have a well-established history of safety, and are essential to the health of the nation. [^1]: The author is employed by a trade association representing the natural products industry.

{#sp1 .136} {#sp2 .137} {#sp3 .138} {#sp4 .139} {#sp5 .140} {#sp6 .141} {#sp7 .142} {#sp8 .143} {#sp9 .144}

Background ========== Lung cancer is a malignant carcinoma with high morbidity and mortality in Chinese population. Non-small cell lung cancer (NSCLC) accounts for approximately 80% of all lung cancers. The synthetical therapy has been developed remarkably, however the efficacy on locally advanced or metastatic NSCLC is still poor. Recently, the molecular-targeted therapy with gefitinib shows favorable performance. Gefitinib is a tyrosine kinase (TK) inhibitor of epidermal growth factor receptor (EGFR). It blocks signal pathways involved in proliferation and survival of cancer cells \[[@B1]\], and displays activity against malignant tumors. Two large randomised phase II studies (IDEAL1 and 2) in patients with locally advanced or metastatic NSCLC after failure of platinum-based chemotherapy showed a higher response rate of gefitinib (12%-18%) \[[@B2],[@B3]\]. Compared to docetaxel, gefitinib showed superior progression-free survival (PFS), objective response rate (ORR), better tolerability, and similar quality of life (QOL) improvement rates in pretreated NSCLC \[[@B4]\]. Gefitinib was also effective and safe in Chinese patients with recurrent advanced NSCLC \[[@B5]\]. In 2006, Niho et al. reported response rate of 30%, median survival time (MST) of 13.9 months and 1-year survival rate of 55% in advanced NSCLC after first-line single agent treatment with gefitinib\[[@B6]\]. Some other groups also reported that first-line single agent treatment with gefitinib may have better effect in patients with advanced NSCLC than standard first-line chemotherapy \[[@B7]-[@B10]\]. Gefitinib showed clinical benefits for EGFR mutation NSCLC patients with extremely poor performance status (PS)\[[@B11],[@B12]\]. The large randomized trial (IPASS research) which compared gefitinib with carboplatin/paclitaxel in patients with advanced NSCLC demonstrated superiority of gefitinib relative to carboplatin/paclitaxel in terms of PFS, ORR, tolerability, and QOL improvement rates. However, the overall survival (OS) and disease-related symptom improvement rates were similar \[[@B13]\]. In 2009, Kim et al. demonstrated that compared to pre-gefitinib eras, the survival of advanced NSCLC patients was significantly improved in post-gefitinib eras in Korea \[[@B14]\]. However, the present data regarding first-line treatment with single agent gefitinib against NSCLC in Chinese population are not sufficient. Here, we conducted a study of single agent treatment with gefitinib in 45 patients with advanced NSCLC in order to assess its efficacy and toxicity in Chinese patients. Materials and methods ===================== Patients -------- 45 patients with histologically or cytologically confirmed stage IIIB or IV NSCLC received gefitinib as first-line treatment between July 2006 and Oct 2008 at the First Affiliated Hospital of Nanjing Medical University. All of these patients were treated initially and had at least one measurable focus according to standard Response Evaluation Criteria in Solid Tumors (RECIST) \[[@B15]\]. These 45 patients consisted of 19 males and 26 females with median age around 61.8 years (range: 30-78). 17 patients had smoking history. In terms of tumor histologic types, the patients included 26 adenocarcinomas, 4 bronchioloalveolar carcinomas, 10 squamous cell carcinomas and 5 adenosquamous carcinomas. According to American Joint Committee on Cancer (AJCC) staging manual, 14 patients were in stage IIIB and 31 patients in stage IV. The Eastern Cooperative Oncology Group Performance Status (ECOG-PS) value was less than 2 in 32 patients, and 3 - 4 in 13 patients (Table [1](#T1){ref-type="table"}). All patients provided written informed consent before enrollment. This protocol was approved by the Institutional Review Boards of the participating centers. ###### Clinical material and efficacy of the 45 patients Characters NO. CR, n (%) PR, n (%) SD, n(%) PD, n (%) ------------------ ----- ----------- ----------- ---------- ----------- Gender  Male 19 0 15.8(3) 36.8(7) 47.4(9)  Female 26 0 46.1(12) 38.5(10) 15.4(4) Age(year)  \< 70 35 0 34.3(12) 37.1(13) 28.6(10)  ≥70 10 0 30.0(3) 40.0(4) 30.0(3) Smoking status  Smokers 17 0 17.6(3) 41.2(7) 41.2(7)  Non-smokers 28 0 42.9(12) 35.7(10) 21.4(6) Tumor histology  Adeno. 26 0 38.5(10) 42.3(11) 19.2(5)  BAC 4 0 75.0(3) 25.0(1) 0.0(0) Squamous 10 0 10.0(1) 30.0(3) 60.0(6)  Adenosquamous 5 0 20.0(1) 40.0(2) 40.0(2) Stage  IIIb 14 0 28.6(4) 50.0(7) 21.4(3)  IV 31 0 35.4(11) 32.3(10) 32.3(10) Brain metastasis 4 0 75.0(3) 25.0(1) 0.0(0) PS value  ≤ 2 32 0 37.5(12) 37.5(12) 25.0(8)  3\~4 13 0 23.0(3) 38.5(5) 38.5(5) Therapy ------- Gefitinib (AstraZeneca Company) was administered orally 250 mg daily, 28 days as a cycle. The treatment was continued until disease progression or intolerable toxicity. Observation index ----------------- We conducted a thorough physical examination on each patient to acquaint with the health status (PS method). Blood routine, hepatic and renal function, electrocardiogram, PET/CT or CT were examined. These indexes were reexamined regularly during the trial, and the image examination was performed after the first one cycle. After that, the image examination was conducted once two cycles. The follow-up of patients by telephone or outpatient service for 1 year was performed. Evaluative standards -------------------- Tumor response was assessed as complete response (CR), partial response (PR), stable disease (SD), or progression disease (PD) in accordance with the standard of RECIST \[[@B15]\]. A CR was defined as the complete disappearance of all clinically detectable tumors for at least 4 weeks. A PR was defined as an at least 30% decrease in the sum of the longest diameters of the target lesions for more than 4 weeks without new area of malignant disease. PD indicated an at least 20% increase in the sum of the longest diameter of the target lesions or a new malignant lesion. Stable disease was defined as insufficient shrinkage to qualify for PR and insufficient increase to qualify for PD. An objective response rate (ORR) indicated the proportion of patients achieved CR and PR, while a disease control rate (DCR) indicated the proportion of patients achieved CR, PR and SD. Progression-free survival (PFS) was measured from Day 1 of treatment until the first objective or clinical sign of disease progression. Overall survival (OS) was measured from Day 1 of treatment until the date of death. The alteration of patients\' symptoms including appetite, fatigue, cough, dyspnea, hemoptysis and pain referencing to Lung Cancer Symptom Scale (LCSS) \[[@B16]\] was observed. Symptomatic remission was considered if the score over 25 points. Symptom remission time means the span from initial administration to symptom remission. Adverse effects including 5 degrees (0-IV) were evaluated following the standard enacted by the World Health Organization in 1981. Statistical considerations -------------------------- The data was analyzed by SPSS11.5. Intergroup comparison was conducted by X2 checking. Survival analyses were performed by Kaplan-Meier method. Survival deviation was calculated by Log-Rank test. All P-values were considered significant if P ≤ 0.05. Results ======= Clinical efficacy ----------------- All of these patients were eligible. None of the patients achieved CR. 15 patients (33.3%) achieved PR and 17 patients (37.8%) had stable disease (SD). 13 patients (28.9%) developed progressive disease (PD). ORR and DCR was 33.3% and 71.1% respectively. Subset analysis according to basic traits of the patients was shown in Table [1](#T1){ref-type="table"}. Table [2](#T2){ref-type="table"} showed that the efficacy of gefitinib therapy correlated with gender, tumor histology (P \< 0.05). However, other factors such as age, smoking status, disease stage, and ECOG-PS didn\'t correlate with the efficacy of gefitinib therapy. ###### Gradational analysis of ORR and DCR Characters ORR(%) P value DCR(%) P value ----------------- -------- --------- -------- --------- Gender  Male 13.3 0.033 52.6 0.019  Female 40.0 84.6 Age(year)  \< 70 34.3 1.000 71.4 1.000  ≥70 30.0 70.0 Smoking status  Smokers 17.6 0.082 58.8 0.281  Nonsmokers 42.9 78.6 Tumor histology  Adeno. And BAC 43.3 0.044 83.3 0.027  Non-adeno.  13.3 46.7 Stage  IIIb 28.6 0.909 78.6 0.699  IV 35.5 67.7 PS value  ≤ 2 37.5 0.561 75.5 0.589  3\~4 23.1 61.5 It is notable that there were 4 patients with brain metastasis in this trial, including 3 cases of PR and 1 case of SD. Brain metastatic focuses disappeared in 2 patients of PR, and their primary tumor reduced. One of them expressed headache palliative at the day 1. The primary and metastatic tumors of one patient reduced two weeks later. Remission of symptoms --------------------- In this trial, except 5 patients whose PS = 0, 29 of the other 40 patients (72.5%) achieved palliative symptoms such as fatigue, cough, pain, etc. Remission time arranged from 1 to 14 days, median remission time was 8 days. Overall survival ---------------- MST of the 45 patients was 15.3 months by Oct 15, 2008, (95% CI 11.22-19.38). OS arrange from 7.4 to 23 months, and the patient who had the longest OS was still alive at the most recent follow-up. The 1-year survival rate was 50%. The Kaplan-Meier survival curve was showed in Figure [1](#F1){ref-type="fig"}. The MST of patients with adenocarcinoma and non-adenocarcinoma was 17.1 months (95%CI 14.79-19.41) and 11.2 months (95%CI 8.67-13.73), respectively. The MST of patients with adenocarcinoma was remarkably longer than that of non-adenocarcinoma (P = 0.0149) (Figure [2](#F2){ref-type="fig"}). Other factors such as gender, smoking status, etc., had no obvious effects on survival (Smokers indicated current or former smokers, and nonsmokers was defined as persons who had never smoked.). {#F1} {#F2} Progression-free survival time ------------------------------ The median PFS was 6.0 months, (95% CI 4.36-7.64). Kaplan-Meier curve of PFS was showed in Figure [3](#F3){ref-type="fig"}. {#F3} Toxicity and adverse effects ---------------------------- As shown in Table [3](#T3){ref-type="table"}, the most common toxicities of gefitinib treatment were rash (53.3%) and diarrhea (33%). In addition, 26.7% and 22.2% of the patients showed dehydration and pruritus of skin. 6.7% of the patients showed Grade 2 or 3 hepatic toxicity. 4.4% of the patients (2 persons) showed oral ulcer. No patients developed interstitial lung disease (ILD). Most of the toxicity was grade 1 to 2, and remitted after treatment. Grade 3 rash of one patient was remitted by reducing the dose of gefitinib. The relationship between rash and OS is showed in Figure [4](#F4){ref-type="fig"}. ###### Assessment of toxicity (case, %) Toxicity Grade(WHO) -------------------------------- ------------ ---------- -------- -------- ------ Rash 21(46.7) 19(42.2) 4(8.9) 1(2.2) 0(0) Pruritus 35(77.8) 10(22.2) 0 0 0 Dry skin 33(73.3) 11(24.4) 1(2.2) 0 0 Diarrhea 30(66.7) 13(28.9) 2(4.4) 0 0 Oral ulcer 43(95.6) 2(4.4) 0 0 0 Nausea/vomit 37(82.2) 8(17.8) 0 0 0 Hepatic toxicity 42(93.3) 1(2.2) 2(4.4) 0 0 Interstitial lung Disease(ILD) 45(100.0) 0 0 0 0 {#F4} Discussion ========== Because of high morbidity and mortality, investigators pay more attentions to the therapy of lung cancer in recent years. Platinum-based combination chemotherapy has been the standard first-line therapy for advanced NSCLC. However, it brings about severe adverse effects such as vomiting, renal toxicity, cytopenia, etc.. Recently, molecular-targeted agents have been introduced in the treatment of NSCLC. Gefitinib, a tyrosine kinase inhibitor of EGFR, has been allowed to treat NSCLC clinically. The second-line treatment with gefitinib has response rate, survival benefit and safety not inferior to chemotherapy. Two trials in patients who previously failed platinum-based chemotherapy, IDEAL-1 and 2, revealed a favorable ORR (12-18%), a DCR of 50%, and good tolerability of gefitinib treatment \[[@B2],[@B3]\]. Gefitinib have been suggested to have better efficacy in patients of females or non-smokers, patients with adenocarcinoma (particularly with bronchioloalveolar carcinoma), patients with previous immune/endocrine therapy, and patients with a PS of 0 or 1\[[@B2]\]. A trial about the treatment of NSCLC patients from Asia with gefitinib resulted in an ORR more than 25% and a DCR more than 60% \[[@B17]\]. Recently, Lee et al. \[[@B5]\] demonstrated that, as second-line therapy, gefitinib has superior PFS, better tolerability, and similar QOL improvement rates compared to docetaxel. Nowadays, more and more clinical investigations have been carried out to evaluate the efficacy of gefitinib as first-line treatment of advanced NSCLC. Niho et al.\[[@B6]\] reported a response rate of 27% with gefitinib treatment in 40 patients with advanced NSCLC. Yang et al.\[[@B18]\] from Taiwan reported that first-line treatment with gefitinib in 196 patients with NSCLC achieved an ORR of 42%, a DCR of 61%, and a 1-year survival rate of 47.5%. A large phase III trial IPASS, which was designed to compare gefitinib as first-line treatment of NSCLC patients with standard chemotherapy, demonstrated superiority of gefitinib in terms of 12-month rates of PFS (24.9% vs. 6.7%, P \< 0.05), ORR (43.0% vs. 32.2%, P = 0.0001), and tolerability profile compared with carboplatin plus paclitaxel. Recently, Maemondo et al.\[[@B9]\] reported that the gefitinib group had a significantly longer median PFS (10.8 months vs. 5.4 months; P \< 0.001), as well as a higher response rate (73.7% vs. 30.7%, P \< 0.001) than the standard chemotherapy group. A study conducted in Japan also showed a longer PFS in gefitinib group than the cisplatin plus docetaxel group (9.2 months vs. 6.3 months, P \< 0.0001) \[[@B10]\]. In our study of first-line treatment with gefitinib in Chinese patients with advanced NSCLC, we obtained an ORR of 33.3%, a DCR of 71.1%, a median PFS of 6.0 months, and a median OS of 15.3 months. These results were compatible with the reports aforementioned. The IPASS study suggested that gefitinib would be efficacious in first-line treatment of locally advanced or metastatic NSCLC patients with adenocarcinoma who have never or seldom smoked \[[@B13]\]. Consistent with this result, we found that females and patients with adenocarcinoma (including bronchioloalveolar caicinoma) were more sensitive to gefitinib. Although the response rate of gefitinib in non-smokers seemed higher than that in smokers, the result had no statistical significance due to the small sample size. The OS of patients with adenocarcinoma was longer than that of patients with non-adenocarcinoma (17.1 months vs. 11.2 months, P = 0.0149). However, other factors such as gender and smoking status have no obvious correlation to OS. In addition, we found that the OS of patients with rash was longer than that of patients without rash, and a longer OS was coupled with greater rash. Because there were few cases with grade 2 or more serious rash, this result needs to be verified further. Moreover, our study showed favorable efficacy of gefitinib in patients with brain metastasis. Gefitinib is well tolerated in advanced NSCLC. The common adverse effects of gefitinib were skin rash, diarrhea, anorexia, elevated aminotransferase lever, and interstitial lung disease, etc \[[@B9]-[@B11],[@B19]\]. Similarly, mild toxicities including skin rash (53.3%), diarrhea (33%), Grade 2 or 3 hepatic toxicity (6.7%), and oral ulcer (4.4%) were observed in our study. No patients developed ILD. Since the tolerance of gefitinib in NSCLC is better than chemotherapy, and gefitinib could provide clinical benefits for patients with extremely poor PS \[[@B11],[@B12]\], it may be a better choice to treat patients who can\'t tolerate chemotherapy compared to best supportive care (BSC). It has been recently reported that the sensitivity and survival benefit of gefitinib treatment was higher in NSCLC patients with EGFR mutations than the patients without EGFR mutations \[[@B20]-[@B22]\]. Chinese patients of lung cancer have a higher frequency of EGFR mutations than American patients. As a result, Chinese patients were much more sensitive to gefitinib than Americans \[[@B23]\]. Besides mutations, gene copy number and polymorphism of EGFR were also related to the responsiveness of gefitinib in advanced NSCLC \[[@B24],[@B25]\]. EGFR mutations of NSCLC patients can be detected using plasma and pleural effusion samples, which provides a noinvasive method to predict the efficacy of gefitinib in advanced NSCLC \[[@B26]\]. Detecting the mutations of EGFR plays an important role in guiding the first-line treatment with gefitinib in patients with advanced NSCLC. Besides EGFR mutations, the favorable PFS after gefitinib treatment was also associated with high levels of serum surfactant protein D (SP-D) \[[@B27]\]. In future studies, we will investigate the molecules which affect and (or) can be used to predict the efficacy of gefitinib in NSCLC. Conclusions =========== Single agent treatment with gefitinib is effective in patients with advanced NSCLC, and well tolerated in Chinese patients. Gefitinib could be used as first-line treatment for specific subgroups of NSCLC such as females, non-smokers, and patients with adenocarcinoma. Abbreviations ============= NSCLC: non-small-cell lung cancer; TK: tyrosine kinase; EGFR: epidermal growth factor receptor; CR: complete response; PR: partial response; SD: stable disease; PD: progression disease; PFS: progression-free survival; ORR: objective response rate; QOL: quality of life; PS: performance status; ECOG-PS: Eastern Cooperative Oncology Group performance status; DCR: disease control rate; OS: overall survival; RECIST: Response Evaluation Criteria In Solid Tumors; LCSS: Lung Cancer Symptom Scale; MST: median survival time; ILD: interstitial lung disease; BSC: best supportive care; TTP: time to progression; SP-D: serum surfactant protein D Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= YQS contributed to conception and design, and gave final approval of the version to be published. ZXW contributed to conception and design. YMY acquired the data and revised the manuscript critically for important intellectual content. YTG acquired the data and drafted the manuscript. YFS acquired the data. XLH and WL contributed to statistic analysis. All authors have read and approved the final manuscript. Acknowledgements ================ This work was supported by grants from the Jiangsu Provincial Natural Science Foundation (NO. BK2008477), the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry 2009 (IA09), and the open project program of the Health Bureau of Jiangsu province (XK18 200904).