{"paper_id":"c8b006d9-88e3-4516-9f28-068a2ba54f47","body_text":"During childhood and adolescence, heightened neuroplasticity allows the brain to reorganize and adapt to its environment. Disruptions in these malleable phases can result in permanent neurodevelopmental disorders. To identify pathological mechanisms that disrupt developmental neuroplasticity, we applied a systematic computational screen of hundreds of diseases for their impact on neuroplasticity. We discovered that inflammation would putatively disrupt neuroplasticity and validated this hypothesis in an  in vivo  experimental mouse model of developmental cortical plasticity. This work suggests that inflammation during the childhood period could have unrecognized negative consequences on the neurodevelopmental trajectory.\n\nDuring childhood and adolescence, the human brain undergoes tremendous reorganization during windows of heightened neuroplasticity. These windows of plasticity are critical periods that allow brain circuits to be refined by sensory and social experiences, which help to establish normal perception and higher cognitive function ( Johnson and Newport, 1989 ;  Nikolopoulos et al., 1999 ;  Lewis and Maurer, 2005 ;  Schorr et al., 2005 ;  Nelson et al., 2007 ;  Fox et al., 2010 ). Disruption of these critical periods can alter neural circuits that shape function and behavior, which may, in turn, contribute to a wide range of psychiatric and neurodevelopmental disorders, such as autism ( Weinberger, 1987 ;  LeBlanc and Fagiolini, 2011 ;  Takesian and Hensch, 2013 ;  Lee et al., 2014 ). Previous studies focused on several genes relevant to autism spectrum disorders ( MeCP2 ,  Ube3a ,  Fmr1 ) and identified marked disruptions in developmental cortical plasticity ( Tropea et al., 2009 ;  Yashiro et al., 2009 ;  Harlow et al., 2010 ). To our knowledge, no studies have conducted a systematic evaluation of pathological mechanisms that may disrupt developmental plasticity. The goal of this study was to leverage the growing repository of publically available transcriptome data from diverse disease states to identify pathological processes with the capacity to disrupt developmental cortical plasticity.\nTo identify pathological processes that disrupt developmental plasticity, we designed an integrative bioinformatics approach that identifies disruptive pathways through systematic evaluation of molecular profiles of disease states in humans and animals. Our approach adapts molecular matching algorithms from computational drug repurposing (for review, see  Hodos et al., 2016 ) to match transcriptional signatures of disease to those of neuroplasticity. To model plasticity, we leveraged the paradigmatic ocular dominance model of  in vivo  developmental plasticity ( Wiesel and Hubel, 1963 ) and generated transcriptional signatures from primary visual cortex (V1). We matched plasticity and disease signatures to produce a diverse list of diseases ranked by their likelihood to dysregulate developmental plasticity. Across this ranked list, we sought to identify shared pathophysiology, rather than generate hypotheses about individual disease matches. To quantify shared pathophysiology, we developed and applied a novel Disease Leverage Analysis (DLA) that identifies shared molecular patterns of disease signatures to reveal novel disruptors of developmental plasticity. By examining shared pathophysiology, DLA identified a strong relationship between the molecular signatures of inflammation and developmental plasticity. We tested the hypothesis that inflammation disrupts developmental plasticity in the ocular dominance model of developmental V1 plasticity and found that functional, experience-dependent plasticity  in vivo  was suppressed by systemic inflammation. Our study demonstrates the utility of an integrative bioinformatics approach for identifying disruptors of developmental neuroplasticity and suggests that inflammation may be an unrecognized risk factor for neurodevelopmental disorders.\n\nMale C57BL/6 mice (Charles River Laboratories) and  Lynx1 −/−  mice ( Miwa et al., 2006 ; a gift from Dr. Nathaniel Heintz, Rockefeller University, New York, NY) were group housed (three to five animals per cage) under a standard 12 h light/dark cycle (lights on at 7:00 A.M., lights off at 7:00 P.M.) with constant temperature (23°C) and  ad libitum  access to food and water. The Institutional Animal Care and Use Committee at the Icahn School of Medicine at Mount Sinai approved all procedures involving animals.\nLyophilized  Escherichia coli  lipopolysaccharide (LPS; 600,000 endotoxin units/mg; serotype 0127:B8, catalog #L5024, lot 073M4024V, Sigma-Aldrich) was reconstituted in sterile saline (0.9% NaCl) to yield a stock solution of 2 mg/ml, which was diluted with saline on the day of injection to yield a working concentration of 0.03 mg/ml.\nTranscriptomes were profiled with microarray to generate plasticity signatures. Experiment-naive juvenile C57BL/6 mice at postnatal day 29 (P29), adult  Lynx1 −/−  mice (>P60), and adult C57BL/6 mice (>P60;  n  = 3 each group) were anesthetized with isoflurane and cervically dislocated; bilateral V1 was removed, immediately frozen on dry ice, and stored at −80°C until processed. Total RNA was extracted from V1 using RNeasy Lipid Tissue Mini kit (Qiagen) and stored at −80°C. A total of 4.5 μg of cRNA was hybridized to Illumina WG-6 2.0 microarrays (750 ng/subarray). A juvenile plasticity signature was generated via differential expression analyses of juvenile versus adult V1 transcriptomes by first quantile normalizing probe-level data with limma ( Smyth, 2005 ) and then computing rank-based differential expression with RankProd ( Hong et al., 2006 ; both R packages available through the Bioconductor repository) to yield 193 unique mouse Entrez IDs. For downstream analysis, mouse Entrez IDs were mapped to human orthologs using the Mouse Genome Informatics homology reference to yield a 176 gene juvenile plasticity signature. We generated a  Lynx1 −/−  signature in an analogous manner to juvenile, to yield a  Lynx1 −/−  plasticity signature of 98 genes. Raw data for plasticity signatures is freely available at the Gene Expression Omnibus under accession number  GSE89757 .\nTo identify diseases that are predicted to dysregulate plasticity signature genes, we developed a molecular matching score, which is the sum of the absolute value of the rank-difference gene expression measure of disease signatures (for details of this expression measure, see  Dudley et al., 2009 ) that intersect with neuroplasticity signature genes. The absolute value was chosen to simplify downstream interpretation. This score is similar in spirit to the approach taken by  Zhang and Gant (2008) , except in our implementation high scores indicate significant overlap between disease and plasticity signatures, whereas low scores indicate little or no overlap. To compare match scores (M) across diseases, we normalized the scores with  n  = 10,000 permutations of scores using  M a c t u a l   −   M ¯ p e r m ∑ i = 1 n ( M p e r m i − M ¯ p e r m ) 2 n − 1 . \n p  Values were estimated using the Generalized Pareto Distribution ( Knijnenburg et al., 2009 ) on  n  permutations and were multiple test corrected using the method of  Benjamini and Hochberg (1995) .\nWe developed DLA to infer pathological processes that are shared across diseases and predicted to dysregulate plasticity signature genes. For pathological processes, we used the 50 “hallmark” gene sets (MSigDb;  Subramanian et al., 2005 ). We computed a pathology score for each hallmark gene set for each disease signature, for a total of 50 × 436 = 21,800 scores. A pathology score is the sum of the absolute value of the normalized disease signature gene expression that is shared with a hallmark gene set. The absolute value was chosen because the direction of effect for gene sets is not necessarily known. We next calculated a linear regression between the pathology scores for a specific gene set and the plasticity–disease molecular match scores. We estimated the  p  value for the association between the pathology scores and disease–plasticity scores (the β 1  coefficient) by computing 20,000 permutations of pathology scores using gene sets the same length as the input gene set and then calculating the regression on the permuted scores. If  ∑ i = 1 n [ 𝟙 [   |   β  1 p e r m i |   >   |   β 1 a c t u a l   |   ] ]   >   10 , where 𝟙 is the indicator function (i.e., the value is 1 when the conditional is satisfied and 0 otherwise) and  n  = the number of permutations, the  p  value was the empirical estimate:  ∑ i = 1 n |     β 1 p e r m i   |   >   |   β 1 a c t u a l   | n ; otherwise, the Generalized Pareto Distribution was used to estimate the  p  value ( Knijnenburg et al., 2009 ). Bonferroni’s method was used to correct for multiple hypothesis tests (denoted  p corrected  in the text). To account for the probability of a large coefficient by chance, actual β 1  coefficients were normalized by the permutated distribution of β 1  according to  β 1 a c t u a l   −   β1 ¯ p e r m ∑ i = 1 n ( β 1 p e r m i − β1 ¯ p e r m ) 2 n − 1 .  Positive β values are pathological processes associated with diseases that were predicted to disrupt plasticity signatures. Negative β values are pathological processes associated with diseases that are predicted to not disrupt plasticity signatures. To calculate enrichments of top DLA gene sets, we chose a conservative cutoff of  p corrected  < 5 × 10 −5  and then calculated the over-representation of inflammation gene sets among positive DLA associations. To do so, we used the hypergea (Bönn, 2016) R package, which uses a conditional maximum likelihood estimate to compute the odds ratio (OR) on adjusted cell counts (to avoid empty cells) and obtains two-sided  p  values from the hypergeometric distribution.\nExperiment-naive juvenile mice (P26;  n  = 5/group) were lightly anesthetized with isoflurane to avoid additional stressors and injected intraperitoneally before 12:00 noon eastern standard time with a dose of LPS that does not cross the blood–brain barrier ( Banks and Robinson, 2010 ; 300 μg/kg, ∼4.5 μg/mouse) or vehicle (150 μl of saline). Four hours later, mice were deeply anesthetized with isoflurane and decapitated, and bilateral V1 was removed under RNAse-free conditions, briefly rinsed in sterile saline (0.9% NaCl), immediately frozen on dry ice, and transferred to −80°C storage until processed. Total RNA was extracted from V1 using the RNeasy Lipid Tissue Mini Kit (Qiagen) and stored at −80°C. RNA yields ranged from 4.5 to 10 μg/sample and RNA integrity numbers ranged from 8.7 to 10 (mean, 9.8; SD, 0.32). Total V1 RNA was converted to cDNA using a High-Capacity cDNA Reverse Transcription Kit (Life Technologies). Quantitative PCR (qPCR) was performed by the Mount Sinai Quantitative PCR core facility using TaqMan probes (catalog numbers:  NogoR : 00445861,  Lynx1 : 01204957_g1,  S100a8 : 00496696_g1,  Lrg1 : 01278767_m1,  Lcn2 : 01324470_m1,  PirB : 01700366_m1,  Cldn5 : 00727012_s1,  Egr2 : 00456650_m1,  Npas4 : 01227866_g1, Il1: 00434228_m1,  Agmat : 01348862_m1,  Ch25h : 00515486_s1,  Alox12b : 01325300_gH,  Evpl : 01700609_m1,  Slc40a1 : 00489835_g1,  Arc : 01204954_g1,  S100a9 : 00656925_m1,  H2D1/H2K1 : 04208017_mH,  BDNF : 04230607_s1,  Nptx2 : 00479438_m1, and  Ppp3ca : 01317678_m1, Applied Biosystems). Quantification of the fold change was derived via the –ΔΔ CT method (equivalent to a log2 fold change) and significance was computed with parametric  t  tests of the Δ CTs, given the approximately normal distribution of Δ CTs. To prioritize qPCR validations, we identified the most differentially dysregulated juvenile neuroplasticity genes by an independent LPS brain study ( GSE3253 ) by subsetting with the LPS brain expression whose absolute expression statistic was ≥2 after conversion to  z -score.\nUnder light isoflurane anesthesia, the contralateral eye of experiment-naive P26 mice was sutured, and the animal was immediately injected intraperitoneally with LPS (300 μg/kg, ∼4.5 μg/mouse) or vehicle (150 μl saline). Three days later, single-unit electrophysiological recordings were taken in binocular zone of V1 in response to visual stimuli presented to each eye separately ( Gordon and Stryker, 1996 ). Briefly, recording was conducted under nembutal/chlorprothixene anesthesia. Visually evoked single-unit responses were recorded with 16-channel silicone probes (NeuroNexus) in response to a high-contrast single bar generated by visage system (Cambridge Research Systems). The signal was amplified and thresholded (OmniPlex, Plexon). To ensure single-unit isolation, the waveforms of recorded units were further examined off-line (Offline Sorter, Plexon). For each animal, approximately 3–10 single units were recorded in each of the four to six vertical penetrations spaced evenly (250 μm intervals) across the mediolateral extent of V1 to map the monocular and binocular zones and to avoid sampling bias. Monocular zone was identified when three consecutive units solely registered contralateral responses within a single penetration [ocular dominance score (ODS) of 1, see below for definition of ODS]. Secondary visual cortex was identified by the reversal of retinotopy seen as the electrode was moved into the secondary visual cortex ( Gordon and Stryker, 1996 ). Mice that experienced opening of the sutured eye or had poor recordings (<10 cells/mouse or <3 penetrations/mouse or lack of positive identification of monocular zone and secondary visual cortex) were excluded from further study. To analyze the electrophysiology data, normalized ocular dominance index (ODI) of single neurons was computed by a custom-made MATLAB code by peristimulus time histogram analysis of peak to baseline spiking activity in response to each eye: {[Peak(ipsi) − baseline(ipsi)] − [Peak(contra) − baseline(contra)]}/{[Peak(ipsi) − baseline(ipsi)] + [Peak(contra) − baseline(contra)]}, which produces a range of [−1,+1] where −1 is a completely contralateral dominated cell and +1 is a completely ipsilateral dominated cell. ODI is linearly transformed by assigning [−1.0, −0.5) =  1 , [−0.5, −0.3) =  2 , [−0.3, −0.1) =  3 , [−0.1, +0.1] =  4 , (+0.1, +0.3] =  5 , (+0.3, +0.5] =  6 , (+0.5, +1.0] =  7  to produce the ODS. Finally, the contralateral bias index (CBI), a monocular weighted, animal-level summary statistic, is computed from the ODS, as follows: [( n 1  −  n 7 ) + 2/3( n 2  −  n 6 ) + 1/3( n 3  −  n 5 ) +  N ]/2 N , where  N  = total number of cells and  n x  = number of cells corresponding to ODS of  x . Thus, a CBI value of 0.7 is contralateral dominant, and a CBI value of 0.4 is ipsilateral dominant. For statistical comparison of ocular dominance, ODSs of single neurons were plotted as a proportion histogram and compared via a nonparametric χ 2  test, and CBI values of single animals were compared via a  t  test. Saline-treated juvenile animals (age P26) were the comparison group. The experimenter was blind to the sample group. Sample sizes were statistically estimated prior to undertaking experimental work to be  n  = 6 per group, assuming the effect sizes seen in previous relevant work.\nAll statistical and computational analyses conducted with R (version 3.2.2) and Python (version 2.7.10). Parametric Welch  t  tests were two sided, unless otherwise noted. Sample sizes (denoted  n ) always indicate the number of mice. The influenza 95% confidence interval (CI) for the incidence rate ratio was estimated using the Katz log approach ( Fagerland et al., 2015 )  ( e l o g ( I R R ) ± 1 a + 1 b − 1 m − 1 n ) , where  a  and  b  are the successes and  m  and  n  are the totals (totals determined by dividing the successes by the published incidence rates. The detail of statistical analysis is shown in  Table 1 .\nStatistics\n\nTo enable molecular matching between plasticity and disease, we compared V1 transcriptomes of juvenile wild-type mice, during the critical period of elevated ocular dominance plasticity in V1 ( Gordon and Stryker, 1996 ), with adult wild-type mice with reduced plasticity, to identify a differential expression signature of 176 genes ( Fig. 1a ;  Table 2 ). We computationally matched this signature to 436 disease signatures derived from public microarray data using a previously described method ( Dudley et al., 2009 ;  Fig. 1a ). This systematic method applies a rank-based molecular matching algorithm to determine the molecular concordance between the plasticity signature and a given disease signature, where high scores indicate plasticity genes are significantly dysregulated by the disease and low scores indicate that the disease has no impact on plasticity genes (for details, see Materials and Methods;  Fig. 1b ). The molecular matching procedure produced a list of 436 diseases ordered by their prediction to disrupt the plasticity signature. Interestingly, highly ranked diseases included not only brain disorders known to disrupt plasticity, such as Huntington's disease ( Usdin et al., 1999 ;  Murphy et al., 2000 ;  Milnerwood et al., 2006 ), but also non-neurologic disorders (e.g., bacterial infections, inflammatory bowel disease, metabolic diseases), suggesting that a broad range of disease states may impact molecular pathways involved in plasticity.\nDiseases that dysregulate the juvenile plasticity signature are associated with inflammatory processes.  a ,  b , Juvenile plasticity signature of 176 genes (represented by green bars) generated by differential expression of P29 vs >P60 C57BL/6 male mice primary visual cortex was computationally matched to 436 disease signatures (represented by orange bars;  a ) using rank-based molecular matching where large scores indicate shared transcriptional phenotype ( b ).  c , DLA systematically identified processes associated to diseases that dysregulate the plasticity signature. Seven of the 14 largest associations were inflammation-related gene sets, and 7 of 7 of inflammation-related gene sets were strongly associated with plasticity (all at  p corrected  < 5 × 10 −5 : OR = 25.8, 95% CI = 1.4–491.2,  p  = 2.7 × 10 −3 , Fisher’s exact test). See  Figure 1-1  for source DLA data.\nJuvenile plasticity signature\nV1 transcriptomes were profiled with microarray from juvenile mice with naturally elevated experience-dependent plasticity to generate the juvenile plasticity differential expression signature. Using RankProd differential expression (DE) analysis, V1 of juvenile male mice C57BL/6 at P29 was compared to adult C57BL/6 mice (>P60;  n  = 3 each group) to identify 248 DE probes, which mapped to 193 unique mouse Entrez IDs. For downstream analysis, mouse Entrez IDs were mapped to human orthologs using the Mouse Genome Informatics homology reference to yield a 176 gene juvenile plasticity signature. FC, fold change. RP, rank product. pfp, percent false positive (i.e. false discovery rate).\nWe sought to identify shared pathophysiology across the diverse list of diseases predicted to dysregulate plasticity signatures. To do so, we developed DLA. This approach calculates the association between diseases that dysregulate the plasticity signature genes and the 50 gene sets in the hallmark library ( Subramanian et al., 2005 ) that represent well defined and distinct biological pathways. Specifically, DLA computes a linear regression between the molecular match score (a measure of strength of association between disease and plasticity signatures) and the pathology score (a measure of activity of the biological pathway in a given disease). Large regression coefficients indicate that a given biological pathway is highly active in diseases that dysregulate plasticity gene signatures and may be pathological to developmental plasticity. Using a multiple test-corrected, empirical  p  value threshold of  p  < 5 × 10 −5 , we found that 7 of 14 largest DLA associations were inflammation-related gene sets and that every inflammation-related gene set in the hallmark library was strongly associated with diseases that dysregulate the plasticity signature (7 of 7 inflammation gene sets at a threshold of  p corrected  < 5 × 10 −5 : OR = 25.8, 95% CI = 1.4–491.2,  p  = 2.7 × 10 −3 a , Fisher’s exact test;  Figs. 1c ,  1–1 ). Moreover, two of these gene sets, tumor necrosis factor-α (TNF-α) signaling via nuclear factor-κB (NF-κB) and interferon-γ (IFN-γ) response, reflect pathways involved in critical period plasticity ( Kaneko et al., 2008 ;  Nagakura et al., 2014 ).\nDLA identifies biological processes common to diseases that perturb the juvenile plasticity signature. To identify shared pathophysiology across the diverse list of diseases predicted to dysregulate the juvenile plasticity signature, we applied DLA. This approach calculates the association between diseases that dysregulate the plasticity signature genes and 50 well defined and distinct biological pathways. To do so, it computes a regression between the molecular match score (a measure that indicates the strength of association between the disease and plasticity signatures; see  Table 3  and Materials and Methods) and the pathology score (a measure of activity of the biological pathway in that disease; see Materials and Methods). Large regression coefficients indicate that the biological pathway may disrupt juvenile plasticity. Using a multiple test-corrected, empirical  p  < 5 × 10 −5 , seven of the 14 largest DLA associations were inflammation-related gene sets, and every inflammation-related gene set in the hallmark library was strongly associated with diseases that dysregulate plasticity genes (seven of seven inflammation gene sets at  p corrected  < 5 × 10 −5 : OR = 25.8, 95% CI = 1.4–491.2,  p  = 2.7 × 10 −3 , Fisher’s exact test). A total of 20,000 permutations of the gene sets was used to estimate  p  values and to normalize the regression coefficients to allow comparison between effect sizes for different biological pathways. Inflammation-related gene sets: TNF-α signaling via NF-κB, IFN-γ response, inflammatory response, complement, IL-2–Stat5 signaling, IFN-α response, IL-6–Jak–Stat3 signaling.\nTo control for nonplasticity aspects of age, we repeated the entire analysis using a  Lynx1 −/−  plasticity signature of 98 genes, which was identified by computing the differential expression between adult  Lynx1 −/−  and adult wild-type V1 ( Table 4 ). By releasing the  Lynx1  brake on plasticity,  Lynx1 −/−  mice have juvenile-like plasticity in adulthood ( Morishita et al., 2010 ;  Bukhari et al., 2015 ). Indeed, functional similarity is reflected in signature similarity, as juvenile and  Lynx1 −/−  plasticity signatures significantly overlap (35 genes shared, OR = 37.1, 95% CI = 23.8–58.0,  p  < 2.2 × 10 −16 b ). Applying DLA to the  Lynx1 −/−  molecular matches, we found a strong association between diseases predicted to disrupt  Lynx1 −/−  plasticity and inflammation-related gene sets (7 of 7 inflammation gene sets at a threshold of  p corrected  < 5 × 10 −5 : OR = 63.0, 95% CI = 3.2–1229.8,  p  = 7.6 × 10 −5 c  Fisher’s exact test;  Figs. 2 ,  2-1 ). Together, the bioinformatics analyses indicate that inflammation is a process central to diseases predicted to dysregulate plasticity gene expression, suggesting that inflammation may disrupt developmental plasticity.\nDiseases that dysregulate the adult  Lynx1 −/−  plasticity signature are associated with inflammatory processes. Using an adult  Lynx1 −/−  plasticity signature of 98 genes (generated by differential expression of primary visual cortex from >P60  Lynx1 −/−  vs >P60 C57BL/6 male mice), DLA systematically identified biological processes associated with diseases that dysregulate the adult  Lynx1 −/−  plasticity signature genes. Using adult  Lynx1 −/−  animals controls for age, as these adult animals have elevated plasticity similar to that of juvenile animals. Seven of the 11 largest associations were inflammation-related gene sets, and every inflammation-related gene set was strongly associated with plasticity (7 of 7 inflammation gene sets at  p corrected  < 5 × 10 −5 : OR = 63.0, 95% CI = 3.2–1229.8,  p  = 7.6 × 10 −5 , Fisher’s exact test). See  Figure 2-1  for source DLA data.\nDLA identifies biological processes common to diseases that perturb the  Lynx1 −/−  plasticity signature. To identify shared pathophysiology across the diverse list of diseases predicted to dysregulate the  Lynx1 −/−  plasticity signature, we applied DLA. Large regression coefficients indicate that the biological pathway may disrupt  Lynx1 −/−  plasticity. As with the juvenile plasticity signature, using a multiple test-corrected, empirical  p  < 5 × 10 −5 , we found that every inflammation-related gene set in the hallmark library was strongly associated with diseases that dysregulate plasticity genes (7 of 7 inflammation gene sets at  p corrected  < 5 × 10 −5 : OR = 63.0, 95% CI = 3.2–1229.8,  p  = 7.6 × 10 −5  Fisher’s exact test). A total of 20,000 permutations of the gene sets were used to estimate  p  values and to normalize the regression coefficients to allow comparison between effect sizes for different biological pathways. Inflammation-related gene sets: TNF-α signaling via NF-κB, IFN-γ response, inflammatory response, complement, IL-2–Stat5 signaling, IFN-α response, IL-6–Jak–Stat3 signaling.\nBased on the DLA findings, we hypothesized that inflammatory processes disrupt developmental cortical plasticity. To test this hypothesis, we induced a systemic inflammatory response via LPS and measured the impact on developmental plasticity and related gene expression. We injected a low dose of LPS (300 μg/kg, intraperitoneal) at the peak of juvenile ocular dominance plasticity at P26 and found a strong inflammatory response in V1, as indicated by a 2.4 log 2  fold increase of interleukin (IL)-1β compared with vehicle control ( p  = 3.4 × 10 −4 d ,  t  test of Δ CTs,  n  = 5 mice/group). To identify a focused subset of plasticity genes likely to be regulated by LPS (regardless of age or specific brain region) to test  in vivo , we investigated a highly significant molecular match between the juvenile plasticity signature and an adult brain-derived LPS transcriptome ( GSE3253 ; rank #14,  p  = 7.9 × 10 −4 e ; empirical  p  value calculated using molecular match algorithm;  Fig. 3a ;  Table 3 ). Next, we identified a subset of genes from  GSE3253  that is likely to play a larger role in the under-lying biology (the genes in the extremes using a  z -score cutoff) and intersected it with the plasticity signature to identify 16 shared genes ( Fig. 3b ). Notably, among these shared genes, we identified a negative correlation in their expression pattern (cor = −0.77,  p  = 0.0007 f , Spearman’s correlation). Among these 16 shared genes, the adult LPS data indicated the direction of expression of 13 genes would be reversed by LPS. Indeed, the majority (61.5%) of the 13 genes showed a complete reversal in their differential expression pattern in V1 after peripheral LPS administration during the critical period (qPCR, all reversed genes  p  < 5 × 10 −4 g ,  t  test of Δ CTs,  n  = 5 mice per group;  Fig. 3c ). These data indicate that genes in the plasticity signature that are also regulated by LPS act in an antagonistic fashion and naturally led us to the hypothesis that inflammation may suppress plasticity. Consistent with this logic, the established brakes of plasticity,  PirB  ( Syken et al., 2006 ) and  H2K1  and  H2D1  ( Datwani et al., 2009 ), showed increased expression compared with vehicle ( p  = 0.04 h  and  p  = 0.03 i , respectively,  t  test of Δ CTs,  n  = 5 mice per group;  Fig. 4a,b ). In addition, a trigger of plasticity that increases across development, BDNF ( Huang et al., 1999 ), which we predicted  in silico  to decrease after LPS ( Fig. 3b ), showed decreased expression compared with vehicle ( p  = 0.009 j ,  t  test of ΔCTs,  n  = 5 mice/group;  Fig. 4b ). In contrast, other known plasticity effectors ( Takesian and Hensch, 2013 ) were not changed relative to vehicle ( Nptx2 ,  Lynx1 ,  NogoR ,  Ppp3ca ;  p  > 0.1 k ,  t  test of Δ CTs,  n  = 5 mice per group), indicating that LPS may act through a specific subset of known and novel plasticity effectors ( Fig. 4b ).\nLPS reverses plasticity signature gene expression.  a , LPS disease signature shares plasticity signature genes  in silico  (molecular match rank #14,  p  = 7.9 × 10 −4 ;  Table 3 ; disease signature is from  GSE3253 : adult mouse whole-brain homogenate harvested 4 h after peripheral LPS injected intraperitoneally; genes with absolute value of the standardized expression ( z -score) ≥2 SDs from the mean were selected as the most differentially expressed by LPS).  b , The expression of the 16 genes shared between juvenile plasticity and the LPS disease signatures is anticorrelated (Spearman’s ρ = −0.77,  p  = 7.4 × 10 −4 ; LPS disease signature gene expression values fell in the range [−1,+1]; for plotting purposes, the plasticity gene expression fold change was linear transformed to [−1,+1]).  c , Of the 13 genes of the 16 predicted to be reversed by LPS, the majority (8 of 13; 61.5%) showed a complete reversal in their differential expression pattern in V1 after LPS administration (300 μg/kg LPS, intraperitoneally, at P26 during the peak of juvenile plasticity) relative to saline. LPS downregulated  Cldn5  and  Slc40a1  and upregulated  Alox12b ,  S100a9a ,  Ch25h ,  Lrg1 ,  S100a8 , and  Lcn2  ( n  = 5 mice per group). *** p  < 0.001, ** p  > 0.001 and ≤0.01, * p  > 0.01 and ≤0.05 (two-sided  t  tests of ΔCTs). Log2 fold change is −ΔΔ CT. Error bars indicate the SEM.\nMolecular matching among 436 disease signatures and the juvenile plasticity signature indicates diverse diseases may disrupt plasticity\nWe computationally matched the juvenile plasticity signature to 436 disease signatures derived from public microarray data. This systematic method applies a rank-based molecular matching algorithm to determine the molecular concordance between the plasticity signature and a given disease signature, where high scores indicate that plasticity genes are significantly dysregulated by the disease and low scores indicate that the disease has no impact on plasticity genes. Highly ranked diseases included not only brain disorders known to disrupt plasticity such as Huntington’s disease, but also non-neurologic disorders (e.g., bacterial infections, inflammatory bowel disease, metabolic diseases), suggesting that a broad range of disease states may impact molecular pathways involved in plasticity. ALS, Amyotrophic lateral sclerosis; AML, acute myeloid leukemia; APECED, autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy; CBCL, cutaneous B-cell lymphoma; CFS, chronic fatigue syndrome; CML, chronic myeloid leukemia; COPD, chronic obstructive pulmonary disease; DMD, Duchenne muscular dystrophy; FDIU, fetal death in utero; FDR, false discovery rate; GERD, gastroesophageal reflux disease; HIV, human immunodeficiency virus; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome; IUGR, intrauterine growth retardation; JRA, juvenile rheumatoid arthritis; LGLL, large granular lymphocytic leukemia; MCTD, mixed connective tissue disease; MDS, myelodysplastic syndrome; MGUS, Monoclonal gammopathy of undetermined significance; MMHCC, Mouse Models of Human Cancers Consortium; MODY, maturity onset diabetes in youth type 1; MS, multiple sclerosis; PBL, peripheral blood lymphocyte; PTSD, post-traumatic stress disorder; RA, rheumatoid arthritis; SARS, severe acute respiratory syndrome; SCID, severe combined immunodeficiency.\nInflammation induced by LPS suppresses experience-dependent plasticity in juvenile cortex.  a , Juvenile mice (P26) during the peak of ocular dominance plasticity were injected intraperitoneally with either LPS (300 μg/kg) or saline.  b ,  c , Mice were either subjected to qPCR analysis of plasticity effectors from V1 4 h after the injection ( b ) or underwent 3 d of MD followed by  in vivo  extracellular recordings to assess ocular dominance plasticity ( c ).  b , LPS increased known plasticity brakes  PirB  and  H2K1 / H2D1 , and decreased the plasticity trigger  BDNF . LPS had no effect on the plasticity effectors  Nptx2 ,  Lynx1 ,  NogoR , or  Ppp3ca . Log2 FC (fold change) is −ΔΔ CT ( n  = 5 mice per group). Error bar indicates the SEM. *** p  < 0.001, ** p  > 0.001 and ≤0.01, * p  > 0.01 and ≤0.05 ( t  test of Δ CTs).  c , Neurons from peripheral LPS-treated mice (purple histogram:  n  = 7 mice, 129 cells) showed decreased cortical responsivity to light in the ipsilateral vs contralateral eye, as quantified by a reduced right shift in the ODS distribution after 3 d of MD compared with control saline-treated juvenile mice with MD (gray histogram: 6 mice, 135 cells; χ 2  test of ODS distribution:  p  = 2.6 × 10 −6 ). Animal-level quantification of ocular dominance plasticity by CBI reflects the extent of the ocular dominance shift after 3 d of MD (right-side plot; low CBI indicates higher plasticity). CBI was strongly increased in the LPS-treated group (purple discs: CBI = 0.64 ± 0.02, 7 mice) compared with saline-treated group (gray discs: CBI = 0.49 ± 0.04, 6 mice), indicating that proinflammatory LPS had suppressed developmental plasticity (LPS vs saline, ** p  = 6 × 10 −3 , one-sided  t  test). qPCR data are reported as the mean ± SEM. Horizontal bars for the CBI plot indicate the mean.\nFinally, we tested whether inflammation suppresses experience-dependent developmental cortical plasticity  in vivo . We administered LPS (300 μg/kg, intraperitoneal) or saline on P26 immediately after suturing one eye to induce ocular dominance plasticity via monocular deprivation (MD;  Fig. 4a ). After 3 d of MD, we conducted  in vivo  single-unit recordings of activity-driven changes in the eye preference of single neurons (ocular dominance) in binocular V1 in response to light ( Gordon and Stryker, 1996 ). In mice treated with saline, we observed the expected shift in cortical responsivity to light stimulation from the deprived contralateral to nondeprived ipsilateral eye, as quantified by a decrease in the animal-level CBI (CBI = 0.49 ± 0.04; six mice; 135 cells), indicating the presence of developmental plasticity ( Fig. 4c , right-hand plot). In contrast, LPS significantly suppressed the shift in cortical responsivity from the deprived contralateral eye to the nondeprived ipsilateral eye, which was quantified by an increase in CBI and an elimination of the right shift in the distribution of ODSs of single neurons (CBI = 0.64 ± 0.02, seven mice, 129 cells; one-sided  t  test of CBIs:  p  = 0.006 l ; χ 2  test of ODS distribution:  p  = 2.6 × 10 −6 m ), indicating impaired plasticity during the critical period in V1 ( Fig. 4c , left-hand plot). Together, these data are consistent with our informatics-derived hypothesis by demonstrating that peripheral injection of LPS induces an inflammatory response in the brain and suppresses developmental cortical plasticity  in vivo.\n\nUsing an integrative bioinformatics approach, we found that inflammation disrupts developmental cortical plasticity. Our study demonstrates the utility of this approach for both identifying diseases that may disrupt plasticity and generating hypotheses on the molecular mechanisms underlying these disruptions. Moreover, our novel Disease Leverage Analysis facilitates novel hypothesis generation, because seemingly unrelated phenotypes, such as neuroplasticity and inflammation, can be connected based on apparently disparate tissues and diseases. Previous work indicating that the disease signal harmonizes across tissues ( Dudley et al., 2009 ) supports this approach and in the present study suggests that the molecular pathways underlying plasticity are shared in diverse tissues and are dysregulated in many disease states, including apparently non-neurological phenotypes (e.g., bacterial infections, inflammatory bowel disease, metabolic diseases). Importantly, the biological relevance of any given molecular match between plasticity and disease must be interpreted with care. In all cases, molecular matches indicate that plasticity and the disease phenotype share underlying molecular machinery. However, a given disease state in a specific tissue may or may not have an impact on functional plasticity or related gene expression if the disease state or tissue is sufficiently localized and segregated from neural tissue. Consequently, we developed Disease Leverage Analysis to use the information of all matches collectively to identify common disease processes and simultaneously shrink the hypothesis space to a manageable set of disease process-oriented hypotheses that bind together the diverse matches. This approach facilitated the unbiased and systematic use of apparently disparate disease signatures to generate novel hypotheses about shared disease mechanisms that may dysregulate plasticity. We find here that a common theme among these dysregulations is inflammation, a biological process that is well suited to communicate peripheral signals to the brain, disrupting plasticity.\nWe demonstrate several lines of evidence supporting a hypothesis that plasticity and inflammatory processes share components of underlying molecular networks. We computationally predicted associations between plasticity signature-perturbing diseases and TNF-α and IFN-γ pathways ( Figs. 1 ,  2 ) and also experimentally identified associations between systemic inflammation and increases in the plasticity brakes  PirB  and  MHC-1  in the brain ( Fig. 4 ). These predictions and observations confirm the known role of pathways involving TNF-α, IFN-γ, Pirb, and MHC-I on regulating developmental plasticity ( Syken et al., 2006 ;  Kaneko et al., 2008 ;  Datwani et al., 2009 ;  Nagakura et al., 2014 ) and extend them to the context of inflammation. We also showed that  BDNF , a neurotrophic factor essential to the opening of the critical period ( Huang et al., 1999 ), is decreased after LPS administration ( Fig. 4 ), which is consistent with the reported antagonistic relationship of peripheral LPS with BDNF ( Guan and Fang, 2006 ;  Schnydrig et al., 2007 ). Interestingly, we also found that the microglial activator Lcn2 ( Jang et al., 2013 ) is a member of both the juvenile and  Lynx1 −/−  plasticity signatures ( Tables 2 ,  4 ) and is dramatically increased after LPS in V1 during the critical period ( Fig. 3 ). Activation may inhibit microglia from carrying out their “resting-state” role in mediating experience-dependent plasticity ( Sipe et al., 2016 ), contributing to the dampening of plasticity by inflammation. Collectively, our work suggests a conflict between developmental cortical plasticity and immune-related molecular networks during inflammation, ultimately resulting in the suppression of plasticity during inflammation. Our study provides a novel subset of transcripts that can be used to guide future mechanistic studies into inflammation–plasticity interactions.\nLynx1 −/−  plasticity signature\nV1 transcriptomes were profiled with microarray from  Lynx1 −/−  mice (the plasticity brake Lynx1 is genetically deleted to allow experience-dependent plasticity even in adulthood) to generate the  Lynx1 −/−  plasticity differential expression signature. Using RankProd differential expression (DE) analysis, V1 of adult  Lynx1 −/−  male mice >P60 was compared with adult wild-type mice (> P60;  n  = 3 each group) to identify 132 DE probes, which mapped to 107 unique mouse Entrez IDs. For downstream analysis, mouse Entrez IDs were mapped to human orthologs using the Mouse Genome Informatics homology reference to yield a 98 gene  Lynx1 −/−  plasticity signature. FC, fold change. RP, rank product. pfp, percent false positive (i.e. false discovery rate).\nOur efforts to understand the molecular machinery involved in the suppression of developmental plasticity focused on immediate changes in gene expression in V1 via acute inflammation (qPCR 4 h after a single intraperitoneal injection of LPS). We expected this time point to be particularly sensitive to disruption because the earliest experience-dependent changes occur within hours to a day of MD at the level of firing rate of parvalbumin interneurons ( Aton et al., 2013 ;  Kuhlman et al., 2013 ;  Reh and Hensch, 2014 ) and protease ( Mataga et al., 2002 ) and microglia activity ( Sipe et al., 2016 ) as triggers for subsequent global ocular dominance plasticity, which takes a few days to be detected by single-unit recordings ( Gordon and Stryker, 1996 ). Importantly, such trigger events occur only during the juvenile critical period when our assay was performed, but not in the adult period ( Kuhlman et al., 2013 ). Thus, we reasoned that the baseline cortical expression signature at this early time point would be critical to gate global ocular dominance plasticity. In addition, peak acute inflammatory response, as measured by an increase in IL-1β in brain after peripheral LPS injection, is between 1 and 4 h ( Layé et al., 1994 ;  Eklind et al., 2006 ;  Richwine et al., 2009 ), a time course well suited to disrupt these earliest experience-dependent plasticity events. While we speculate that LPS disrupted these early trigger events unique to juvenile cortex, more work needs to be performed to understand the molecular events underlying the functional changes seen within hours of MD and to dissect the impact of inflammation on these events. In addition to the early phase of plasticity, inflammation may also impact plasticity mechanisms during the later phases of MD because TNF-α is essential to nondeprived eye potentiation via a homeostatic mechanism at 5-6 d of MD ( Kaneko et al., 2008 ). Ultimately, further work is necessary to tease out the interaction between inflammatory and plasticity mechanisms that contribute to the suppression of functional plasticity across multiple days of experience deprivation. Performing such work comparing acute versus chronic inflammatory models would provide fascinating insights into neuroimmune biology and would help to inform the important clinical question of the potential impact of acute and chronic inflammation on the neurodevelopmental trajectory in children.\nWhile our experimental efforts focused on the impact of acute inflammation on plasticity, our list of diseases predicted to impact plasticity include diseases that accompany chronic inflammation ( Table 3 ). Efforts to study human disease and animal models shed light on how acute versus chronic inflammation may affect plasticity. For example, components of plasticity and inflammation are dysregulated in epilepsy ( Vezzani and Granata, 2005 ). Acute inflammation from low to high doses (LPS) decreases the threshold for the induction of seizure ( Sayyah et al., 2003 ), and a single early life inflammatory insult increases susceptibility to seizure even into adulthood ( Galic et al., 2008 ). Chronic overexpression of inflammation-related genes in rodents causes an increased or decreased susceptibility to seizure, depending on gene dose ( Vezzani and Granata, 2005 ), and seizure itself appears to chronically induce inflammatory markers ( De Simoni et al., 2000 ). This evidence indicates a potential bidirectional effect of epilepsy and inflammation, wherein acute and chronic inflammation may have immediate and long-term effects on epilepsy-related plasticity mechanisms. In addition to epilepsy, cortical lesions and hypoxia-ischemia induce a robust inflammatory response that can endure chronically ( Bona et al., 1999 ;  Schroeter et al., 2002 ) and disrupt ocular dominance plasticity weeks after the injury ( Failor et al., 2010 ;  Greifzu et al., 2011 ). Interestingly, anti-inflammatory (ibuprofen) treatment rescues MD-induced sensory learning (increased visual acuity of the nondeprived eye) in adult (P70–P110) after cortical injury via photothrombosis in the nearby primary somatosensory; however, ibuprofen did not restore ocular dominance plasticity ( Greifzu et al., 2011 ). It is possible that the anti-inflammatory regimen or mechanism of action used was not sufficient to eliminate the inflammation and rescue ocular dominance plasticity, or it may reflect distinct mechanisms of plasticity and their modulation by inflammation in the juvenile cortex versus the adult. While it has been proposed that causes of plasticity disruption may be cortical deafferentiation in the case of cortical lesions or disruption of inhibitory interneurons in the case of hypoxia-ischemia, it is possible that inflammation downstream of injury disrupts plasticity and should be investigated further. In sum, there is evidence that chronic and acute inflammation go hand in hand with disrupted plasticity across a variety of brain disorders, on different time scales, and as a function of different underlying mechanisms. Going forward, work is necessary to understand the contribution of diverse inflammatory mechanisms in the disruption of various types of plasticity across a wide variety of neurological and neurodevelopmental conditions.\nOur finding that inflammation suppresses developmental cortical plasticity suggests a potential public health concern related to neurodevelopmental trajectory. During the height of the critical period for visual plasticity (peak is 0.5–2 years in humans;  Morishita and Hensch, 2008 ), children <5 years of age have the highest incidence of contracting LPS-carrying Gram-negative foodborne pathogens relative to other childhood or adult periods ( Centers for Disease Control and Prevention, 2013 ). Other infections that induce inflammation also show an increased incidence during the peak of developmental plasticity in humans; >80% of children <3 years of age experience otitis media (ear infection;  Marom et al., 2014 ) and children <5 years of age are hospitalized for influenza-related complications nearly an order of magnitude more often than children 5–17 years of age (incidence rate ratio = 8.1, 95% CI = 7.3–9.0; data are from  Dawood et al., 2010 ). Our work suggests that this increased incidence of infection during postnatal periods of developmental plasticity (relative to older ages) may be an unrecognized mechanism by which inflammation alters the neurodevelopmental trajectory. Most directly, the suppression of visual cortex plasticity could disrupt the development of binocular matching ( Wang et al., 2010 ), a process central to the development of normal vision that specifically depends on heightened plasticity during the critical period for visual development. In addition, higher-order cognitive processes could be disrupted, due to the hierarchical dependency of various critical periods of plasticity ( Takesian and Hensch, 2013 ). In addition, given that mechanisms of plasticity identified in the visual critical period have translated to other brain regions and functions ( Levelt and Hübener, 2012 ;  Nabel and Morishita, 2013 ;  Werker and Hensch, 2015 ), it is likely that inflammation could disrupt plasticity in other systems.\nOur work is a natural extension into the postnatal epoch of the growing body of research indicating deleterious brain and behavioral outcomes due to prenatal inflammatory exposure ( Choi et al., 2016 ;  Steullet et al., 2016 ;  Weber-Stadlbauer et al., 2017 ) and suggests that inflammation may have a more extensive impact on postnatal neurodevelopment and brain function than previously realized. In fact, childhood infections and inflammation are associated with subsequent diagnoses of autism, depression, and schizophrenia as well as declines in cognitive capacity ( Dalman et al., 2008 ;  Atladóttir et al., 2010 ;  Khandaker et al., 2014 ;  Benros et al., 2015 ). The elevated incidence rate of infections during childhood neurodevelopment (relative to older ages) and the association of childhood infection with subsequent neurodevelopmental disorder may indicate a partial explanation for the observed onset of psychiatric disorders in childhood and adolescence ( Lee et al., 2014 ). Our work showing that inflammation disrupts developmental cortical plasticity suggests an unrecognized risk factor for neuropsychiatric disorders and provides a starting point to investigate the underlying pathophysiology.\nWe show here that an integrative bioinformatics approach is well suited to interrogate the interactions between disease processes and disruptions in developmental plasticity. To extend this approach further, molecular matching could be expanded to the >71,000 experiments publically available (as of 4 August 2016, there were 71,885 Gene Expression Omnibus “Series”), and Disease Leverage Analysis could be expanded to the universe of biologically defined gene sets (as of 4 August 2016, MSigDb alone contained 13,311 sets), facilitating more comprehensive interrogation of the disease space and generation of more specific hypotheses about disease processes that disrupt plasticity. Moreover, this approach is not limited to interrogating neurodevelopment but can be extended to other neurological signatures beyond plasticity. We expect it will be useful for identifying connections between disease processes and other brain phenotypes that can be appropriately represented by a transcriptional signature.","source_license":"CC-BY-4.0","license_restricted":false}