Sigmoid colon mucosal gene expression supports alterations of neuronal signaling in irritable bowel syndrome with constipation

Abstract

Peripheral factors likely play a role in at least a subset of irritable bowel syndrome (IBS) patients. Few studies have investigated mucosal gene expression using an unbiased approach. Here, we performed mucosal gene profiling in a sex-balanced sample to identify relevant signaling pathways and gene networks and compare with publicly available profiling data from additional cohorts. Twenty Rome III+ IBS patients [10 IBS with constipation (IBS-C), 10 IBS with diarrhea (IBS-D), 5 men/women each), and 10 age-/sex-matched healthy controls (HCs)] underwent sigmoidoscopy with biopsy for gene microarray analysis, including differential expression, weighted gene coexpression network analysis (WGCNA), gene set enrichment analysis, and comparison with publicly available data. Expression levels of 67 genes were validated in an expanded cohort, including the above samples and 18 additional participants (6 each of IBS-C, IBS-D, HCs) using NanoString nCounter technology. There were 1,270 differentially expressed genes (FDR < 0.05) in IBS-C vs. HCs but none in IBS or IBS-D vs. HCs. WGNCA analysis identified activation of the cAMP/protein kinase A signaling pathway. Nine of 67 genes were validated by the NanoString nCounter technology (FDR < 0.05) in the expanded sample. Comparison with publicly available microarray data from the Mayo Clinic and University of Nottingham supports the reproducibility of 17 genes from the microarray analysis and three of nine genes validated by nCounter in IBS-C vs. HCs. This study supports the involvement of peripheral mechanisms in IBS-C, particularly pathways mediating neuronal signaling.

NEW & NOTEWORTHY Peripheral factors play a role in the pathophysiology of irritable bowel syndrome (IBS), which, to date, has been mostly evident in IBS with diarrhea. Here, we show that sigmoid colon mucosal gene expression profiles differentiate IBS with constipation from healthy controls. These profiling data and analysis of additional cohorts also support the concept that peripheral neuronal pathways contribute to IBS pathophysiology.

INTRODUCTION

Irritable bowel syndrome (IBS) is a prevalent and costly condition with limited therapies and poorly understood pathophysiology. Affecting around 11% of the general population in the US (84), IBS is associated with a significant healthcare and economic burden, and a lack of treatment satisfaction (25, 49). Improved understanding of the molecular basis for the dysregulated gastrointestinal function and sensation in IBS may lead to development of targeted therapies and improved quality of life.

There is clear and growing evidence that peripheral mechanisms play a role in IBS pathophysiology for at least a subset of IBS patients (24). Peripheral mechanisms manifest in a range of phenotypes, including, but not limited to, alterations in intestinal permeability (41, 132), serotonin signaling (7, 36, 38, 45, 54, 63, 122), immune milieu (2, 4, 12, 13, 20, 30, 39, 94), bile acids (21), microbiota, and neuronal sensitivity (9, 16–18, 27, 57, 95, 118, 124). It is not known whether these phenotypes result from universally deregulated molecular pathways or whether IBS represents a syndrome caused by heterogeneous pathophysiological mechanisms. In order to identify molecular pathways associated with IBS, we conducted a microarray analysis of sigmoid mucosal gene expression in a pilot study of Rome III+ IBS patients and age- and sex-matched healthy controls (HCs) who were free of active psychiatric disease, and had limited use of medications.

The National Institutes of Health recently launched an initiative to improve experimental reproducibility (37). This has notoriously been a challenge in IBS. This likely results from the combination of heterogeneity in both population and study methodology, and pathophysiology resulting from subtle changes; a large sample is needed to detect small differences when standard deviations are large. Alternatively, IBS may not be caused by a universal pathophysiological mechanism. This difficulty with reproducibility may actually be underrepresented in the literature due to publication bias for positive results, since very few studies have used unbiased approaches (e.g., microarray or RNA sequencing evaluation of mucosal gene expression). In order to identify reproducible results within our own cohort, we compared our findings with publicly available data. Our comparison data come from two microarray studies comparing well-phenotyped IBS patients with HCs. The first study, referred to as “Mayo”, evaluated gene expression in sigmoid biopsies from 36 Rome II+ IBS patients (15 IBS with constipation, IBS-C) and 25 HCs (2). The second study, “Nottingham,” evaluated rectal biopsies in 56 Rome II+ (115) IBS patients (19 IBS-C) and 25 HCs in addition to a group of postinfection IBS patients (112). The only currently published RNA sequencing study in IBS (22), as well as the comprehensive follow-up studies (19, 23) have focused on IBS-D.

Here, we report on the 1) differential expression of transcripts found by microarray, 2) comparison with results from two prior microarray studies (Mayo and Nottingham), 3) weighted gene coexpression network analysis (WGCNA) and gene set enrichment analysis (GSEA) to identify pathways and networks relevant to IBS, and 4) validation of selected genes in a larger cohort. We have also created a website (http://uclacns.org/ibs-microarray-comparison/), which displays tabular and graphic results of gene expression data from this study, Mayo, and Nottingham for any gene present on the arrays.

MATERIALS AND METHODS

Participants and Protocol

Rome III+ (80) IBS patients and HCs (men and women, aged 18–55 yr) were recruited primarily by community advertisement. Bowel habit subtypes were based on Rome III criteria. The diagnosis was confirmed by a clinician with expertise in IBS. HCs had no personal or family history of IBS or other chronic pain conditions. Additional exclusion criteria for all subjects included infectious or inflammatory disorders, active psychiatric illness over the past 6 mo assessed by structured clinical interview for the DSM-IV (MINI) (105), use of corticosteroids in the past 6 mo, use of narcotics, antidepressants, or other medications that could affect neuroendocrine function in the past 2 mo, or current tobacco or alcohol abuse. Participants were compensated. The study was approved by the UCLA Institutional Review Board, and all subjects signed a written informed consent form prior to starting the study. Overall IBS symptom, abdominal pain, and bloating severity over the prior week were each assessed with a numeric rating scale (0–20) (90). Current anxiety and depression symptoms were measured with the Hospital Anxiety and Depression (HAD) scale (133). Sigmoid biopsies (30 cm) were obtained by flexible sigmoidoscopy following tap water enemas. Specimens were flash-frozen in liquid nitrogen, and RNA was extracted with TRIzol reagent (ThermoFisher Scientific, Waltham, MA). Participants for the microarray study as well as the additional participants included in the nCounter validation were selected from the same biorepository at UCLA. The only criteria used to select participants for inclusion in both the microarray and validation cohorts was optimal matching of sex and age within each bowel habit subtype.

Microarray

Gene expression was measured using the Arraystar Human LncRNA Microarray v. 3.0 (Arraystar, Rockville, MD), which profiles over 30,500 LncRNAs and 26,100 mRNAs. RNA quantity and quality were first measured using the NanoDrop ND-1000 (ThermoFisher Scientific), and RNA integrity was assessed by standard denaturing agarose gel electrophoresis. Sample labeling and array hybridization were performed according to the Agilent One-Color Microarray-Based Gene Expression Analysis protocol (Agilent Technologies, Santa Clara, CA), with minor modifications. Briefly, mRNA was purified from total RNA after removal of rRNA (mRNA-ONLY Eukaryotic mRNA Isolation Kit; Epicentre, Madison, WI). Then, each sample was amplified and transcribed into fluorescent cRNA along the entire length of the transcripts without 3′ bias, utilizing a mixture of oligo(dT) and random priming method (Flash RNA Labeling Kit, Arraystar). The labeled cRNAs were purified using the RNeasy Mini Kit (Qiagen, Valencia, CA). The concentration and specific activity of the labeled cRNAs (pmol Cy3/μg cRNA) were measured with the NanoDrop ND-1000. One microgram of each labeled cRNA was fragmented by adding 5 μl of 10× blocking agent and 1 μl of 25× fragmentation buffer, heated to 60°C for 30 min, and 25 μl of 2× GE hybridization buffer was added to dilute the labeled cRNA. Fifty microliters of hybridization solution was dispensed into the gasket slide and assembled to the LncRNA expression microarray slide. The slides were incubated for 17 h at 65°C in an Agilent hybridization oven. The hybridized arrays were washed, fixed, and scanned using the Agilent DNA Microarray Scanner (part no. G2505C). Agilent Feature Extraction software (v. 11.0.1.1) was used to analyze acquired array images. Quantile normalization and subsequent data processing were performed using the GeneSpring GX v. 12.0 software package (Agilent Technologies). After quantile normalization of the raw data, LncRNAs and mRNAs that had flags in present or marginal (“all targets value”) in at least 10 of 30 samples were chosen for further data analysis. Targets with mean raw signal intensity of less than 200 in both IBS or HCs were excluded. Data have been deposited in the ArrayExpress database at EMBL-EBI under accession no. E-MTAB-5811 (https://www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-5811.

Validation of Selected Targets

RNA (100 ng) from the 30 samples analyzed by microarray and 18 additional samples (identical recruitment and protocol) were analyzed using a custom nCounter panel including 67 targets and three reference genes (GAPDH, BACT, PPIA; NanoString Technologies, Seattle, WA).

Statistical Analysis

Code used for analyses in R is provided in the online Supplement.

Sample size.

Based on the proportion of differentially expressed transcripts (DETs) in the Mayo study (2), (23/47,000) and an approximate standard deviation of 0.5, we estimated that eight per group would provide 80% power to detect DETs with a fold change (FC) of ≥2 (78).

Microarray differential expression.

DETs were determined using the Linear Models for Microarray Analysis (limma) (107) package in R v. 3.3.1 (99) following log2 transformation. An adjusted P value threshold (Benjamini-Hochberg) of 0.05 determined statistical significance.

Acquisition and analysis of external data.

External data [Mayo:E-TABM-176 (2), Nottingham:E-GEOD-36701 (112)] were retrieved using the ArrayExpress package (61), and Affymetrix probe annotations were updated [annotate package (47)]. Data were normalized using the robust multiarray average function [affy package (46)]. The limma package was used to account for technical replicates (duplicateCorrelation function) and to determine DETs (107).

Determination of microarray probe transcript specificity.

The “ReAnnotator” pipeline (5) was used to obtain accurate alignments of Arraystar probe sequences from the current study to the exome. Updated annotations of Affymetrix probes (2015) were downloaded from ThermoFisher. Among nonspecific probes in the current study, the probe with the maximum variance across samples was retained. Common results were defined as probes targeting the same transcript or gene and differentially expressed in the same direction (i.e., up- or downregulated in both) with a threshold of P < 0.05.

WGCNA.

WGCNA (67) was performed in R on targets with mean raw signal intensity over 200 using a soft-thresholding power 14 and minimum module size 30. Highly correlated modules (r > 0.75) were merged.

Functional annotation.

Gene set enrichment analysis (GSEA) (111) was performed using the Java GSEA Desktop Application (v. 2.2.4) on all annotated mRNAs (log fold change and P value were entered) with mean signal intensity > 200 in either group (n = 14,775), using default parameters against the following gene sets obtained from Molecular Signatures Database (v. 6.0) (89): Reactome (43), KEGG (60), and Gene Ontology (GO) Biological Process (BP) (6). Overrepresented GO terms for Mayo/Nottingham comparison and WGCNA were determined using the hypergeometric function in the GOstats (44) package in R. These overrepresented terms as well as those yielded from analysis in DAVID (55, 56) were used to select descriptive module names. Pathway analysis was performed using Ingenuity Pathway Analysis (Qiagen, Valencia, CA) using experimentally observed direct relationships in the Ingenuity Knowledge Base.

NanoString nCounter expression and association with symptoms.

nCounter expression data was analyzed with the NanoStringNorm and limma packages in R (107, 120). Counts were normalized to the geometric mean of all samples to adjust for technical variation between the four cartridges. Due to high variation of the housekeeping genes included in the code set (GAPDH, ACTB, PPIA), RNA quantity was normalized to the geometric mean of genes with the lowest coefficients of variation (GABBR1, RBM5, MUC4, IL17RE, RNF19A, ELMOD3, USP34, CDC42EP5, OPLAH, SGSM2), which was determined using NanoStringNorm (results were unchanged when MUC4, which was differentially expressed, was removed from the normalization set). Three outliers (mean count z-scores > 3) were excluded from analysis (an additional with z-score 3.03 was retained) yielding 45 samples with nCounter data. Using the voom function (69) in the limma package, count data were prepared for analysis with linear models, and DETs were determined with limma. An adjusted-P value threshold of 0.05 was used to determine statistical significance. Association with symptoms was determined using linear models. If sex, age, or BMI was a predictor of the outcome variable, it was included in the model (sex for overall symptoms and all 3 for abdominal pain).

RESULTS

Study Population

Characteristics of the study participants are shown in Table 1. The total cohort with available NanoString nCounter gene expression data (excluding outliers, n = 3) included 31 IBS patients and 14 HCs (51.6 and 42.9% women, respectively). These included a total of 15 IBS-C patients (10 women, 5 men), 16 IBS-D patients (6 women, 10 men), and 14 HCs (6 women, 8 men).

Table 1.

Participant characteristics

Variable (range)	Microarray		nCounter
Group	IBS	HC	IBS	HC
n	20	10	31	14
%Women	50	50	51.6	42.9
Age	33.8 (10.8)	31.7 (8.5)	34 (12)	35.4 (8.4)
BMI	24.4 (6.4)	24.7 (2.5)	24.5 (5.5)	25.8 (3)
Ethnicity
Hispanic	1 (5%)	2 (20%)	4 (12.9%)	1 (7.1%)
Race
Asian	5 (25%)	3 (30%)	8 (25.8%)	3 (21.4%)
Black	5 (10%)	3 (10%)	8 (9.7%)	3 (7.1%)
White	5 (50%)	3 (40%)	8 (48.4%)	3 (57.1%)
Other/mixed	3 (15%)	2 (20%)	5 (16.1%)	2 (14.3%)
Bowel habit subtype
IBS-C	10 (50%)		15 (48.4%)
IBS-D	10 (50%)		16 (51.6%)
HAD anxiety score (0-21)	5.6 (6)	3.3 (2.8)	6.4 (6)*	2.9 (2.4)
HAD depression score (0-21)	3 (3.6)	1.9 (2.1)	3 (3.1)	1.4 (1.8)
Overall IBS severity (0-20)	10.4 (4.4)		10.1 (4)
Abdominal pain (0-20)	9.8 (4.6)		9.7 (4.2)

Values are means (SD) or frequency (%). nCounter, counter cohort is inclusive of microarray cohort. HC, healthy control; IBS, irritable bowel syndrome; BMI, body mass index; HAD, Hospital Anxiety and Depression; IBS-C, IBS with constipation; IBS-D, IBS with diarrhea.

^*P < 0.05 between IBS and HCs.

Only 24 participants (all IBS) reported any medication use, and most used laxatives/antidiarrheals and ibuprofen only on an as-needed basis. All participants were instructed to hold aspirin and nonsteroidal anti-inflammatory drugs (NSAIDs) for 72 h prior to the procedure. No participants in the microarray cohort reported use of antidepressants/neuromodulators. In the expanded cohort, one participant took sertraline (100 mg/day) and bupropion (150 mg/day), and another took trazodone (50 mg/day). Participants were instructed not to take unnecessary medications 24–48 h prior to the procedure.

Identification of IBS Gene Signature by Microarray Profiling Analysis

Differentially expressed transcripts.

For IBS vs. HCs and IBS-D vs. HCs, there were no DETs meeting the threshold of FDR < 0.05. However, there were 120 and 66 DETs using an unadjusted P value threshold of 0.001. For IBS-C, there were 1,270 DETs with FDR < 0.05 (790 upregulated and 480 downregulated), mapping to 1,246 specific transcripts (retaining the probe with highest variance across samples among nonspecific probes; Supplement 1). A heat map of the top 100 genes is shown in Fig. 1A. Since our sample was age and sex matched, our primary analysis did not include sex. There were six genes for which there was a significant effect of sex (FDR < 0.05), and the effect of IBS-C vs. HCs remained significant (FDR < 0.05) in all six after controlling for sex.

Fig. 1.

A: top 100 differentially expressed genes in irritable bowel syndrome with constipation (IBS-C) vs. healthy controls (HCs). B: differentially expressed genes in IBS-C vs. HC that are replicated in the Mayo Clinic cohort. Expression values are scaled. Descriptions are from gene ontology or KEGG pathways, except where referenced. Genes and samples (within bowel habit subtypes) are clustered by Euclidean distance. Sample clustering is based on top 500 probes after ordering by variance across all 30 sample. IBS-D, IBS with diarrhea; cAMP/cGMP, cyclic adenosine/guansosine monophosphate; GTP, guanosine triphosphate; GTPase, enzymatic function catalyzing formation of GDP from GTP; BMP, bone-morphogenic protein; SNP, single nucleotide polymorphism; MHC, major histocompatibility complex; ATP, adenosine triphosphate; GPCR, G protein-coupled receptor; PKA, protein kinase A.

Comparison with external gene expression data.

Demographics of the samples and study protocols from Mayo Clinic and the University of Nottingham are shown in Table 2. Probes from the Arraystar (Agilent) array used in our study and the Affymetrix array used in Mayo and Nottingham were matched using the following algorithm. First, we identified transcripts or genes matched by only a single probe in each array. There were 452 probes uniquely matched to a specific transcript, and one probe uniquely matched to a specific gene. There were 90 Affymetrix probes that were nonspecific and overlapped with more than one Arraystar probe (e.g., two Arraystar probes were transcript specific with multiple Affymetrix probes matching). In these cases, the affected Arraystar probes were reduced to retain the probe with the highest variance across samples (48 Arraystar probes excluded). This yielded a set of Affymetrix probes wherein each only matched a single Arraystar probe, although in many cases there were more than one Affymetrix probes matched with each Arraystar probe. The final matches were determined separately for Mayo and Nottingham by reducing the multiple matching Affymetrix probes to the one with the highest variance across samples. Among the 1,246 DETs (IBS-C vs. HCs) in the current study, 859 were matched with probes from the Mayo and Nottingham data. Among these, the number of DETs with the same direction of FC (P < 0.05) in IBS-C and HC samples from Mayo and Nottingham were 57 and 185, respectively, with 17 matched in all three data sets. To address a possible limitation of comparing rectal (Nottingham) expression with sigmoid, we focused further investigation on the 57 common genes between our study and the Mayo study (Fig. 1B and Table 3). Upregulated genes included reticulon 3, which is important in neurite outgrowth (29). Downregulated genes included corticotropin-releasing factor receptor 1 (CRHR1), SCO-spondin (SSPO), which we previously found to be hypermethylated in peripheral blood mononuclear cells from IBS vs. HCs (82), and the glial cell line-derived neurotrophic factor receptor GRFA4.

Table 2.

External sample and study characteristics

	Mayo			Nottingham
Sample	IBS-D	IBS-C	HC	IBS-D	IBS-C	HC
n	21	15	25	27	18	26
%Women	86%	100%	92%	67%	89%	62%
Recruitment	By mail from database, Rome II			Outpatient GI clinic, Rome II
Protocol (prep)	Sigmoidoscopy (Fleet’s enema)			Sigmoidoscopy (none)
Biopsy site	Sigmoid colon			Rectum
Microarray	Human Genome U133-Plus2.0 (Affymetrix)

Table 3.

Genes differentially expressed in IBS-C vs. HC concordant in Mayo

Name (Symbol)	AveExpr	FC (FDR)
Upregulated
Autophagy related 7 (ATG7)	10.99	1.2 (0.031)
Centriole, cilia and spindle associated protein (CCSAP)	8.90	1.4 (0.04)
Centromere protein J (CENPJ)*	7.63	1.4 (0.032)
Cofilin 2 (CFL2)	10.05	1.5 (0.015)
Family with sequence similarity 161 member A (FAM161A)	7.47	1.4 (0.022)
FERM, ARH/RhoGEF and pleckstrin domain protein 1 (FARP1)*	8.97	1.3 (0.026)
Folliculin (FLCN)*	9.81	1.3 (0.041)
IKBKB interacting protein (IKBIP)	8.38	1.5 (0.035)
Interleukin 17 receptor D (IL17RD)	8.26	1.4 (0.029)
IQ motif containing GTPase activating protein 2 (IQGAP2)*	12.94	1.3 (0.048)
Kelch repeat and BTB domain containing 3 (KBTBD3)	8.24	1.3 (0.025)
MAGE family member D4B (MAGED4/MAGED4B)	9.07	1.4 (0.024)
Mitochondrial calcium uptake 1 (MICU1)	8.00	1.3 (0.046)
NFU1 iron-sulfur cluster scaffold (NFU1)	9.91	1.4 (0.032)
N-acetylneuraminate pyruvate lyase (NPL)*	9.61	1.7 (0.022)
PostGPI attachment to proteins 1 (PGAP1)	7.54	1.6 (0.031)
Reticulon 3 (RTN3)	8.57	1.3 (0.049)
SRY-box 5 (SOX5)	8.02	1.4 (0.03)
Sperm flagellar 2 (SPEF2)	7.31	1.6 (0.033)
TIA1 cytotoxic granule associated RNA binding protein like 1 (TIAL1)*	9.36	1.3 (0.027)
Transmembrane protein 44 (TMEM44)	7.44	1.5 (0.003)
WD repeat domain 19 (WDR19)	8.66	1.3 (0.047)
Zinc finger protein, FOG family member 2 (ZFPM2)*	8.43	1.7 (0.01)
Zinc finger protein 789 (ZNF789)*	7.44	1.3 (0.032)
Downregulated
Acyl-CoA thioesterase 7 (ACOT7)*	8.58	0.6 (0.019)
Angiotensin II receptor type 2 (AGTR2)	7.35	0.6 (0.019)
Amidohydrolase domain containing 2 (AMDHD2)	9.54	0.8 (0.028)
Chromosome 20 open reading frame 141 (c20orf141)	11.70	0.8 (0.029)
Coiled-coil domain containing 27 (CCDC27)	7.81	0.8 (0.046)
Corticotropin releasing hormone receptor 1 (CRHR1)	7.53	0.7 (0.018)
DExH-box helicase 30 (DHX30)	8.16	0.8 (0.012)
Epsin 1 (EPN1)*	8.23	0.7 (0.041)
Family with sequence similarity 228 member A (FAM228A)	7.90	0.8 (0.038)
F-box and leucine rich repeat protein 18 (FBXL18)*	7.55	0.7 (0.013)
F-box protein 40 (FBXO40)	7.81	0.4 (0.032)
Fibronectin type III domain containing 11 (FNDC11/C20orf195)	9.56	0.8 (0.03)
GDNF family receptor alpha 4 (GFRA4)	11.93	0.7 (0.045)
Interleukin 17 receptor E (IL17RE)*	10.46	0.7 (0.025)
Inositol-3-phosphate synthase 1 (ISYNA1)*	8.84	0.8 (0.01)
Lipoxygenase homology domains 1 (LOXHD1)	9.48	0.7 (0.021)
Leucine rich repeat containing 27 (LRRC27)	8.87	0.8 (0.029)
Mitogen-activated protein kinase kinase 7 (MAP2K7)	11.66	0.7 (0.036)
Myosin light chain kinase 2 (MYLK2)	14.59	0.5 (0.008)
Nectin cell adhesion molecule 1 (NECTIN1)	8.27	0.6 (0.01)
Nectin cell adhesion molecule 4 (NECTIN4)	10.16	0.7 (0.029)
Neuropeptides B and W receptor 1 (NPBWR1)*	7.73	0.6 (0.017)
NUT family member 2D (NUTM2D)	9.59	0.8 (0.046)
Pregnancy specific beta-1-glycoprotein 6 (PSG6)	12.64	0.6 (0.01)
Purine rich element binding protein A (PURA)*	12.90	0.7 (0.026)
RNA exonuclease 1 homolog (REXO1)	9.43	0.7 (0.039)
SCO-spondin (SSPO)	9.15	0.8 (0.029)
Synaptogyrin 2 (SYNGR2)	11.12	0.8 (0.034)
TAP binding protein (TAPBP)*	9.20	0.7 (0.013)
Chromosome 14 open reading frame 80 (TEDC1)	13.37	0.5 (0.014)
Chromosome 16 open reading frame 59 (TEDC2)*	10.55	0.5 (0.008)
T-SNARE domain containing 1 (TSNARE1)	8.89	0.7 (0.008)
Ubiquitin associated protein 1 like (UBAP1L)	8.95	0.8 (0.041)

AveExpr, average expression (log2 transformed); FC, fold change, FDR, Benjamini-Hochberg adjusted P value.

^*Also differentially expressed in this direction in Nottingham.

Identification of IBS gene networks by performing WGCNA analysis.

WGCNA yielded 45 gene modules, of which 26 significantly correlated with IBS status (P < 0.05; Fig. 2). None of the modules was significantly associated with sex. One module, subsequently named “cAMP-PKA”, was associated with IBS, IBS-C, overall symptom, and bloating severity. Top GO terms associated with this module are shown in Table 4, and the other IBS-associated modules are described in Table 5. Results of the pathway analysis for the “cAMP-PKA” module are shown in Fig. 3, which is displayed as a subset of two networks. The IPA analysis highlights upregulation of cAMP/PKA signaling. Interestingly, following review of the “hub” genes, those with the highest “module membership” (correlations with module “eigengene”), revealed a role for cAMP signaling in other modules that were associated with IBS, and the function of the gene products was consistent with increased cAMP (Fig. 4 and Table 6). For example, phosphodiesterase (PDE)9A, which hydrolyzes cAMP, is in a module that is downregulated in IBS-C. IPA pathway analysis settings, main results, and references for the relationships in the network are provided as a supplementary file.

Fig. 2.

Weighted gene coexpression network analysis (WGCNA) modules associated with IBS. Values are correlations between module eigengene and IBS status or overall symptom severity (0–20) with associated P value. Red and blue indicate positive and negative correlations, respectively.

Table 4.

Gene ontology terms overrepresented among genes in the cAMP/PKA module

GO ID	P Value	Term	Genes
0035589	0.001	G-protein coupled purinergic nucleotide receptor signaling pathway	PTAFR, P2RY14
0051058	0.002	Negative regulation of small GTPase mediated signal transduction	ADRA1A, MYOC, RASAL2
1902692	0.004	Regulation of neuroblast proliferation	CX3CR1, SIX3
0045907	0.006	Positive regulation of vasoconstriction	ADRA1A, PTAFR
0048246	0.008	Macrophage chemotaxis	AZU1, CX3CR1
0071560	0.010	Cellular response to transforming growth factor beta stimulus	AMHR2, LDLRAD4, CDK9, CX3CR1, SOX5, HIPK2, CHST11
0042063	0.012	Gliogenesis	AZU1, ERCC2, ETV5, MBD1, MYOC, SOX5
0046578	0.013	Regulation of Ras protein signal transduction	ADRA1A, MYOC, RASAL2
0050808	0.016	Synapse organization	ACHE, ETV5, SNCG, PDLIM5
0061351	0.018	Neural precursor cell proliferation	CX3CR1, SIX3, SOX5, TACC2
0030182	0.020	Neuron differentiation	CNGA3, ERCC2, ETV5, FES, MBD1, MYOC, ROBO1, SIX3, SOX5, SPTAN1, LAMTOR3, CACNA1G, VAPA, RASAL2, PDLIM5, NSMF, HIPK2, MDGA2
1903307	0.020	Positive regulation of regulated secretory pathway	PTAFR, CACNA1G
0051591	0.022	Response to cAMP	CNGA3, CREM, HNF4A, PTAFR, PYGM, SLC26A6
0007155	0.024	Cell adhesion	ACHE, AZU1, DDR1, CX3CR1, FES, MYOC, PTAFR, ROBO1, ADAM12, EFS, FBLIM1, PCDHA11, ANTXR1
0007600	0.030	Sensory perception	AOC2, CNGA3, PTAFR, SCNN1D, SIX3, OR10H2, CNGB3
0019048	0.030	Modulation by virus of host morphology or physiology	VAPA, HIPK2
0034220	0.030	Ion transmembrane transport	AQP5, CNGA3, PTAFR, S100A1, SCNN1D, VDAC2, CACNA1G, ATP5J2, CNGB3, SLC26A6, SLC26A9
0099536	0.031	Synaptic signaling	ACHE, ADRA1A, CX3CR1, ETV5, SNCG, CACNA1G, NSMF, PANX2, SHISA7
0060395	0.042	SMAD protein signal transduction	HNF4A, HIPK2

Overrepresented terms were determined using the hypergeometric function in the GOstats package for R (44).

Table 5.

Description of IBS-associated modules

Module (size)	Description
Neuroactive ligand receptors 1 (758)	Hub genes enriched (0.01) for KEGG pathway “Neuroactive ligand-receptor interaction” and include 5-hydroxytryptamine receptor 4 (HTR4), cholinergic receptor muscarinic 1 (CHRM1), dopamine receptor D4 (DRD4), and glutamate metabotropic receptor 2 (GRM2). Full module also enriched for GO-BP “negative regulation of cytokine production” (0.002), and “negative regulation of tumor necrosis factor production” (0.0005)
Differentiation (435)	Hub genes enriched (FDR <0.05) for cell differentiation and nervous system development. Representative genes for the latter included glial cell derived neurotrophic factor (GDNF) and beta tubulin (TUBB2B). The module overall was enriched for RNA transcription (0.005).
Ion transmembrane transport (248)	Enriched for GO:BP “regulation of transmembrane ion transport” (0.005)
Cell cycle (273)	Very heavily enriched for cell-cycle genes (GO:BP, FDR = 1.5e-49)
Translation (2685)	Highly enriched for terms related to protein translation. Top two are GO:BP “RNA splicing via transesterification reactions” and “ribonucleoprotein complex biogenesis” (1.8e-07 for each)
Mitochondrial (1915)	Module and hub genes heavily enriched for terms and pathways related to cellular respiration and oxidative phosphorylation (FDR <1e-15 for several)
Hydrolase activity (165)	No terms or pathways enriched P < 0.005 or FDR <0.05. Nine of 45 hub genes map to GO:BP “regulation of hydrolase activity” (0.02)
cAMP/PKA (184)	Based on Ingenuity Pathway Analysis.
Development (234)	Hub genes marginally (0.006) enriched for “anatomical structure development” and “single organism developmental process.” Relevant genes include the transcriptional regulators GATA binding protein 2 (GATA2), Sp3 transcription factor (SP3) and homeobox B3 (HOXB3).
IgA production (127)	Heavily enriched for GO terms related to adaptive immune response (<0.005) as well as KEGG pathway “Intestinal immune network for IgA production” (FDR <0.0005). Hub genes include three HLA class molecules (DMA, DRA, DMB) integrin subunit alpha and beta (ITGA4, ITGB7).
Actin (77)	Enriched (<0.005) for several terms related to regulation of actin polymerization.
Cell junction (889)	Enriched (FDR <0.05) for GO:MF “cadherin-binding in cell adhesion” and CC “adherens junction.” Relevant genes include vinculin (VCL), tight junction protein 1 (TJP1) and catenin delta 1 (CTNND1)
GEF activity (224)	Hub genes enriched for GO:BP “guanyl-nucleotide exchange factor activity” (<0.005).
Transcription (143)	GO:BP “DNA-templated transcription” is most representative (19 genes, 0.04)
DNA repair (168)	Enriched for several terms related to DNA repair (i.e., GO:BP “double-strand break repair via nonhomologous end joining”, 0.001)
Neuronal (578)	GO:BP: “protein localization to synapse” is most enriched (0.0001) and there are 25 genes mapping to GO:BP “nervous system development” (0.03). Relevant genes include neuronal cell adhesion molecule (NRCAM) and neuropeptide Y (NPY).
Neuroimmune (1074)	Hub genes are enriched for sensory perception (FDR <0.05) and include 6 olfactory receptors. Hub genes also contain multiple genes related to the inflammatory response and innate immune system. These include tumor necrosis factor (TNF), CD70, CD80, macrophage expressed 1/Perforin-2 (MPEG1), toll like receptor 9 (TLR9), mucin 22 (MUC22), defensin beta 113 (DEFB113), cysteine rich secretory protein 3 (CRISP3), and the serine protease inhibitors serpin family E members 1&2 (SERPINE1/2).
Apoptosis (275)	Enriched for GO:BP “positive regulation of apoptotic process” (<0.005). Relevant genes include caspase 2 (CASP2) and slit guidance ligand 2 (SLIT2).
Negative regulation of JAK/STAT (139)	The top GO:BP term is “negative regulation of JAK-STAT cascade (0.003) based on caveolin 1 (CAV1) and inositol polyphosphate-5-phosphatase F (INPP5F).
GFRA4 (161)	No significantly enriched terms or pathways. GDNF family receptor alpha is one of top genes (MM = 0.93). Nonsignificantly enriched terms include GO:BP “regulation of GTPase activity” (0.05), “cellular response to BMP stimulus” (0.02), “glial cell-derived neurotrophic factor receptor signaling pathway” (0.04), and “catechol-containing compound metabolic process” (0.01).
Endothelial (107)	Enriched (0.001) for GO:BP “regulation of tube size” based on expression of endothelial nitric oxide synthase (eNOS, NOS3) and adrenoceptor alpha 1A (ADRA1A).
Neuroactive ligand receptors 2 (892)	Hub genes enriched for membrane proteins (GO:CC, FDR <0.05). Top 100 genes include receptors for NPY (NPY1R) and neuropeptides B and W (NPBWR1), melanocortin (MC5R) and GABA (GABRG1).
SSPO (186)	SCO-Spondin (SSPO) is among top two genes in module (MM = 0.95). GO:BP “ion transport” is also enriched (<0.005)
GPCRs (328)	Enriched (<0.005) for GO:BP “G-protein coupled receptor signaling pathway”. Relevant genes include 5-hydroxytryptamine receptors 6 and 3D(HTR6, HTR3D), urocortin 2 (UCN2), and four olfactory receptors.
Keratins (827)	Hub genes enriched for GO:CC “Intermediate filament cytoskeleton” (FDR = 0.05). The majority of these are keratins or keratin associated proteins. The module overall (but not hub genes exclusively) is enriched for GO:BP “response to steroid hormone” (0.009). Relevant genes include hepatocyte nuclear factor 4 alpha (HNF4A), KRAS proto-oncogene, GTPase (KRAS) and estrogen related receptor alpha (ESRRA).
OPRM1 (198)	Not significantly enriched for pathways or terms. Opioid receptor mu 1 (OPRM1) is a hub gene (MM = 0.84). Marginally enriched for GO:BP “sensory response to pain” (P = 0.02) due to OPRM1 and purinergic receptor P2X 4 (P2RX4).

Module membership (MM) is the correlation with module eigengene. Hub genes are those with MM > 0.8. MM P values < 1e-7 for all hub genes. Overrepresented terms were determined using the hypergeometric function in the GOstats package for R(44) or DAVID. (55, 56). Numbers in parentheses are unadjusted P values except where noted. GO, gene ontology; BP, biological process; MF, molecular function; CC, cellular component; cAMP, cyclic adenosine monophosphate; PKA, protein kinase A; GEF, guanine nucleotide exchange factor; GPCRs, G-protein coupled receptors.

Fig. 3.

IPA pathway analysis network for the cAMP/PKA module. Genes not in the cAMP/PKA module have shapes outlined in dashed lines. Fill is proportional to fold change from the microarray data (darker = greater fold change), and all are upregulated except for P2RY14. Figure generated using IPA (Qiagen, Valencia, CA).

Fig. 4.

cAMP signaling roles for hub genes in IBS-associated modules. Directionality of arrows indicates upstream/downstream of cAMP. Dotted lines indicate a negative correlation with the module eigengene. Note that CACNA1G in the guanine nucleotide exchange factor (GEF) activity module is a different isoform than in the cAMP/PKA (Fig. 3). Module memberships and P values are shown in Table 6.

Table 6.

Correlation of cAMP-related genes with module eigengenes

Gene	MM*	P	Module
Upregulated modules
ADRA1A	0.711	1.1E-05	cAMP/PKA
CREB3L1	−0.682	3.3E-05	Hydrolase activity
GABRB3	0.714	9.6E-06	Actin
GEM	0.772	5.9E-07	Actin
PANX2	0.830	1.4E-08	cAMP/PKA
P2RY14	−0.660	7.3E-05	cAMP/PKA
RGS20	0.500	4.9E-03	Actin
Downregulated modules
ARHGEF15	0.564	1.2E-03	GEF activity
CACNA1G**	0.934	4.6E-14	GEF activity
CACNB4	0.776	4.6E-07	GEF activity
CALHM2	0.818	3.3E-08	Transcription
KCNJ11	0.928	1.6E-13	Transcription
KCNMB3	0.896	2.3E-11	DNA repair
PDE9A	0.669	5.2E-05	DNA repair
RAPGEF6	0.918	8.9E-13	GEF activity

^*MM (module membership) is the correlation between the gene and the module eigengene;

^**a separate isoform from that in the orangered4 module.

Identification of IBS signaling pathways based on the gene expression data.

GSEA results are provided in a supplementary file. There were no gene sets enriched in IBS-C or HCs meeting a threshold of FDR < 0.25 (recommended GSEA threshold). Most gene sets that were enriched among HCs with nominal (unadjusted) P < 0.05 were thematically related to synaptic transmission or G protein-coupled receptor (GPCR) signaling. These included Reactome pathways “Activation of kainite receptors upon glutamate binding” (P = 0.024), “Gα(s) signaling” (P = 0.026), “GPCR downstream signaling” (P = 0.036), KEGG pathway “neuroactive ligand receptor interaction” (P = 0.014), and GO:BP gene-sets “neuronal synaptic plasticity” (P = 0.004), and “regulation of synaptic transmission glutaminergic” (P = 0.021). Despite the absence of DETs in IBS vs. HCs, the Reactome pathway “signaling by FGFR1 mutants” was enriched in HCs vs. IBS (p/FDR 0.012/0.19). The 20 genes in this gene set include growth factors and downstream effectors (JAK/STAT pathway). The top enriched Reactome gene set was smooth muscle contraction (p/FDR 0.004/0.29); however, these genes (actins, myosin kinases) likely represent expression at cell junctions rather than muscle. No gene sets were enriched at FDR < 0.25 in IBS-D vs. HC, IBS-C vs. IBS-D or in men vs. women. Within IBS, the following KEGG gene sets were enriched in men vs. women (p/FDR): “allograft rejection” (0.002/0.12), “graft vs. host disease” (0.006/.086), and “autoimmune thyroid disease” (0.002/0.139). These gene sets overlap and include human leukocyte antigen (HLA) complex proteins and tumor necrosis factor (TNF).

Validation of Selected Targets

Selection of genes to validate (n = 67) was based on the following: 1) visualization of dot plots to ensure consistent group difference, 2) programmatic search (annotate package in R) of abstracts linked to each Entrez ID for the words “pain” or “gastrointestinal” and manual search of retrieved abstracts, 3) comparison with external datasets, 4) comparison with results of IBS GWA studies [HTR3A, CDC42EP5 (125)], and 5) results of other studies.

Differential expression.

In the expanded cohort (15 IBS-C vs. 14 HCs), differential expression of 9/67 DETs was validated with nCounter (FDR < 0.05, same direction of FC as in microarray, Fig. 5 and Table 7). There were an additional six (ALDH3A1, ELMOD3, GH1, RAMP1, SSPO, TMEM80) with unadjusted P < 0.05 and five (ALDAH9A1, CLDN15, EZH2, RNF19A, YIF1B) with significant positive correlations (P < 0.05) between microarray and nCounter values.

Fig. 5:

nCounter expression of 9 validated genes in IBS-C vs. HC. Plotted results are mean (SD) for nCounter expression (n = 45). Short description of genes not described in paper: ITPKB is expressed in cells with a brush border (104) but is also important in immune cells (92) and neurons (110). Its molecular function is related to intracellular calcium regulation (113). SLC10A4 is a bile acid transporter, about which relatively little is known, including whether it is able to transport bile acids or neurotransmitters (1, 103, 109). Slc10a4 knockout mice show altered dopamine homeostasis (87, 128). MUC4 is major component of intestinal mucus. Roles include barrier function and host-microbe interactions. Expression of MUC22, a susceptibility locus for panbronchiolitis and childhood asthma, is upregulated in bronchial epithelial cells after exposure to a virus-like molecule or lipopolysaccharide (31, 52).

Table 7.

Results of nCounter validation

		IBS-C vs. HC	IBS-D vs. HC	IBS vs. HC
Gene	AveExpr (all n = 45)	FC (FDR)	FC (FDR)	FC (FDR)
AK9	11.7	1.1 (0.74)	1.1 (0.63)	1.1 (0.61)
ALDH3A1	11.1	0.7 (0.07)*	0.8 (0.3)	0.7 (0.06)*
ALDH9A1	14.8	1.1 (0.5)	1.0 (0.79)	1.0 (0.86)
AQP9	8.5	0.6 (0.17)	0.5 (0.21)	0.6 (0.06)*
CCDC134	11.9	1.0 (0.76)	0.9 (0.54)	0.9 (0.48)
CDC42EP5	16.7	1.1 (0.3)	1.0 (0.67)	1.1 (0.44)
CDK5	11.2	1.3 (0.05)	1.1 (0.57)	1.2 (0.09)*
CLDN15	12.5	1.0 (0.91)	0.9 (0.61)	1 (0.85)
CRHR1	6.4	0.5 (0.05)	0.6 (0.41)	0.5 (0.06)*
DBI	16.2	1.1 (0.33)	1.0 (1.0)	1.0 (0.59)
DHRS12	12.4	0.9 (0.48)	0.9 (0.61)	0.9 (0.48)
EFR3B	9.6	1.0 (0.87)	0.9 (0.79)	1.0 (0.93)
ELMOD3	12.3	0.8 (0.11)*	0.9 (0.61)	0.9 (0.18)
EZH2	13.0	1.0 (0.74)	1.0 (0.83)	1.0 (0.94)
FABP3	11.5	1.0 (0.94)	1.0 (0.96)	1.0 (0.97)
GABBR1	12.8	0.6 (0.01)	0.8 (0.54)	0.7 (0.04)
GABRA1	8.6	0.9 (0.76)	1.0 (1.0)	1.0 (0.88)
GABRG1	6.2	0.5 (0.05)	0.6 (0.21)	0.6 (0.04)
GH1	8.0	0.7 (0.12)*	0.6 (0.21)*	0.7 (0.06)*
HMGA1	14.3	1.0 (0.87)	1.1 (0.57)	1.1 (0.59)
HMGB1	17.0	1.2 (0.05)	1.1 (0.3)	1.2 (0.06)*
HTR3A	7.9	0.8 (0.33)	0.7 (0.21)	0.7 (0.14)*
IL17RD	9.9	0.8 (0.12)	0.9 (0.54)	0.8 (0.16)
IL17RE	14.9	0.9 (0.29)	1.0 (0.82)	0.9 (0.48)
ITPKB	12.6	0.8 (0.05)	0.8 (0.21)*	0.8 (0.04)
JRK	11.4	0.9 (0.63)	0.9 (0.61)	0.9 (0.48)
KAT5	14.2	1.1 (0.59)	1.1 (0.61)	1.1 (0.48)
MLN	6.5	1.1 (0.87)	0.7 (0.45)	0.9 (0.61)
MUC22	7.3	0.6 (0.05)	0.7 (0.21)	0.6 (0.04)
MUC4	15.9	0.7 (0.02)	0.8 (0.41)	0.8 (0.04)
MUCL1	6.4	0.8 (0.62)	0.8 (0.61)	0.8 (0.48)
NECTIN1	13.3	1.0 (0.76)	1.0 (0.83)	1.0 (0.97)
NKX2-1	7.1	0.6 (0.08)	0.6 (0.21)	0.6 (0.04)
NKX3-1	8.1	0.9 (0.48)	0.9 (0.61)	0.9 (0.48)
NOTCH4	10.8	0.9 (0.64)	1.0 (0.99)	1.0 (0.85)
NPBWR1	7.5	1.1 (0.74)	0.9 (0.82)	1.0 (0.97)
NRG3	8.3	0.9 (0.74)	1.0 (0.82)	0.9 (0.77)
OPHN1	12.5	1.1 (0.51)	1.1 (0.61)	1.1 (0.48)
OPLAH	12.2	0.7 (0.11)	0.8 (0.3)	0.8 (0.09)
OPRM1	6.7	0.8 (0.62)	0.7 (0.59)	0.8 (0.48)
PALLD	15.3	1.1 (0.62)	1.1 (0.61)	1.1 (0.51)
PCDHA3	8.5	0.6 (0.02)	0.8 (0.29)	0.7 (0.04)
PDE11A	9.5	0.8 (0.32)	0.9 (0.61)	0.9 (0.36)
PDE2A	10.5	0.7 (0.02)	0.9 (0.61)	0.8 (0.06)*
PPP1CA	17.0	1.1 (0.18)	1.2 (0.3)	1.2 (0.13)*
PPP1R1A	8.9	0.7 (0.16)	0.8 (0.54)	0.8 (0.18)
RAMP1	9.4	0.7 (0.09)*	0.7 (0.21)*	0.7 (0.04)
RBM5	14.9	0.9 (0.33)	1.0 (0.61)	0.9 (0.44)
RNF19A	14.9	1.0 (0.76)	1.0 (0.99)	1.0 (0.86)
S100A1	7.7	0.8 (0.37)	1.2 (0.61)	1.0 (0.86)
SGSM2	13.0	0.8 (0.05)	0.9 (0.62)	0.8 (0.15)
SIGMAR1	13.5	1.0 (0.76)	1.0 (0.82)	1.0 (0.78)
SLC10A4	7.2	0.5 (0.02)	0.6 (0.21)	0.6 (0.03)
SLC17A3	8.4	1.0 (0.91)	0.9 (0.61)	1.0 (0.85)
SMARCA4	14.7	1.1 (0.24)	1.1 (0.54)	1.1 (0.22)
SOX5	9.9	0.7 (0.01)	0.8 (0.21)	0.7 (0.03)
SPATS2	12.6	1.2 (0.16)	1.1 (0.57)	1.1 (0.21)
SSPO	6.2	0.5 (0.08)*	0.5 (0.21)	0.5 (0.04)
ST3GAL3	12.2	0.9 (0.32)	0.9 (0.56)	0.9 (0.29)
THBD	11.7	0.7 (0.12)	0.8 (0.54)	0.8 (0.15)
TMEM44	13.7	0.9 (0.76)	0.9 (0.72)	0.9 (0.73)
TMEM80	13.6	1.2 (0.12)*	1.0 (0.98)	1.1 (0.44)
TRIM37	13.4	1.1 (0.51)	1.1 (0.61)	1.1 (0.48)
TXLNG	13.3	0.9 (0.65)	0.9 (0.61)	0.9 (0.53)
UGT1A9	15.2	1.0 (0.78)	1.0 (0.98)	1.0 (0.87)
USP34	15.1	0.9 (0.09)	1.0 (0.61)	0.9 (0.2)
YIF1B	12.9	1.1 (0.33)	0.9 (0.61)	1.0 (0.87)

Boldface indicates FDR < 0.05. Unmarked values < 0.05 indicate direction did not match microarray. AveExpr, average expression (log2 transformed); FC, fold change, FDR, Benjamini-Hochberg adjusted P value.

^*Unadjusted P value < 0.05.

Associations between gene expression and IBS symptoms were found, although these did not retain significance after adjusting for multiple comparisons (n = 67 genes).

Discussion

The most important findings from this study are that 1) sigmoid mucosal gene expression profiles differentiate IBS-C from HCs and 2) results of multiple analyses and in several cohorts support the relevance of neurally mediated mechanisms in IBS overall and specifically in IBS-C.

Defining a Molecular Signature for IBS-C

Very few studies have reported DETs in IBS-C vs. HCs. A survey of the literature revealed ~90 DETs in IBS (those included in our microarray are shown in Table 8). Recently, Peters et al. (97) compared expression of tight junction proteins in sigmoid colonic mucosal biopsies from women with IBS-C vs. HC women, and none were reported as differentially expressed. As a demonstration of the heterogeneity in findings with different samples and methodology, some of the genes in this study were DETs in our study (FDR < 0.05; Table 9). An important strength of our study is that our microarray sample included no individuals using SSRIs, and there were only two with antidepressant use in the validation sample. Medication use was not reported in the Nottingham comparison study, but in the Mayo study, 42% of IBS participants took medications (SSRI, SNRI, TCA, or dopaminergic agent) vs. 1% of HCs. It is possible that this difference in medication use contributed to our findings. Divergent results by bowel habit subtype is supported by other studies (28, 57), but notably, there is evidence for the absence of subtype-specific effects as well (38, 93). We caution the reader against overinterpretation of these findings. Although gene networks may differ in IBS-C and IBS-D, one cannot conclude that these differences are causative and therefore represent different underlying pathophysiology. These differences are also not evidence of bowel habit differences at the phenotype level (i.e., nerve sensitivity). Furthermore, in accord with our sample size calculation, one can only conclude that there are unlikely to be genes with greater than a twofold difference between IBS-D and HC. We are underpowered to make a negative conclusion for smaller differences that could still be biologically relevant.

Table 8.

Comparison of findings from current study for genes found to be differentially expressed in the literature

		Fold Change, P Value
Symbol	Published Change & Subgroup (Ref.)	IBS	IBS-C	IBS-D	Product
ALDOB***	↑D (22)	0.9 (0.3)	0.8 (0.06)	1.0 (0.88)	Fructose-bisphosphate aldolase B
ALDOC***	↑D (22)	1.3 (0.05)	1.6 (0.01)	1.1 (0.45)	Fructose-bisphosphate aldolase C
AQP8	↓D (121)	1.3 (0.25)	1.7 (0.06)	1.0 (0.94)	Aquaporin-8
BIRC5	↑D (22)	1.1 (0.53)	1.5 (0.02)	0.8 (0.16)	Baculoviral IAP repeat-containing protein 5 isof. 3
C4BPA***	↑D (21)	0.9 (0.29)	0.7 (0.002)	1.2 (0.13)	Complement component 4 binding protein, alpha
CACNA1H***	↑ (102)	1.1 (0.57)	1.2 (0.07)	0.9 (0.36)	Voltage-dep. T-type calcium channel sub. Alpha-1h isof. A
CARD16***	↑ (2)	1.1 (0.44)	1.1 (0.25)	1.0 (0.86)	Caspase recruitment domain-containing protein 16 isof. 1
CASP1***	↓ (2)	0.9 (0.55)	1.0 (0.7)	0.8 (0.16)	Caspase-1 isof. Gamma
CCL20***	↑D (21, 22)	0.8 (0.12)	0.6 (0.004)	1.1 (0.7)	C-C motif chemokine 20 isof.1 precursor
CD163L1***	↑ (2)	1.2 (0.25)	1.6 (0.03)	1.0 (0.84)	Scavenger recep. Cysteine-rich type 1 protein m160 precursor
CKB	↑ (2)	0.9 (0.33)	0.9 (0.41)	0.9 (0.38)	Creatine kinase B-type
DTL	↑ (2)	0.8 (0.03)	0.9 (0.12)	0.8 (0.03)	Denticleless protein homolog
DUOX2**	↑ (2)	1.3 (0.06)	1.3 (0.05)	1.2 (0.18)	Dual oxidase 2 precursor
DUSP27	↓D (22)	1.0 (0.95)	0.9 (0.81)	1.0 (0.9)	Dual specificity phosphatase 27 (putative)
ERAP2	↑ (2)	0.9 (0.8)	1.4 (0.5)	0.6 (0.27)	Endoplasmic reticulum aminopeptidase 2
F2RL1	↑ (75)	0.9 (0.63)	0.9 (0.29)	1.0 (0.82)	Proteinase-activated receptor 2 precursor
F2RL3***	↓ (130)	1.0 (0.58)	1.0 (0.99)	0.9 (0.34)	PAR4 coagulation factor II (thrombin) receptor-like 3
FAAH	↓C (31)	1.1 (0.09)	1.2 (0.05)	1.1 (0.31)	Fatty-acid amide hydrolase 1
FCGR2A	↑ (2)	0.5 (0.09)	0.7 (0.3)	0.5 (0.05)	Low affinity Ig gamma Fc region recep, II-a isof.1 precursor
FGFR4*	↓D (21)	0.9 (0.12)	1.0 (0.86)	0.8 (0.01)	Fibroblast growth factor receptor 4 isof. 2 precursor
FN1	↓D (21, 22)	0.8 (0.07)	0.8 (0.12)	0.8 (0.12)	Fibronectin isof. 5 preproprotein
FOXP3	↓ (13)	0.6 (0.001)	0.5 (0.001)	0.7 (0.02)	Forkhead box P3
GPBAR1	↑C (21)	0.9 (0.2)	1.0 (0.46)	0.9 (0.14)	G protein-coupled bile acid receptor 1
GPER1	↑D (98)	1.3 (0.01)	1.5 (0.01)	1.2 (0.11)	G protein-coupled estrogen receptor 1
GUCA2B***	↑D (22)	1.2 (0.4)	1.4 (0.25)	1.1 (0.75)	Guanylate cyclase activator 2B (uroguanylin)
HELLS***	↑ (2)	1.2 (0.4)	1.6 (0.01)	1.1 (0.75)	Helicase, lymphoid-specific
HLA-DRB5	↓D (22)	1.2 (0.55)	1.0 (0.73)	1.3 (0.49)	MHC, class II, DR beta 5 precursor
HSPA1A***	↓D (22)	0.9 (0.48)	0.9 (0.57)	0.9 (0.51)	Heat shock 70 kda protein 1A/1B
HSPA1B	↓D (22)	1.0 (0.92)	1.0 (0.77)	1.0 (0.9)	Heat shock 70 kda protein 1B
IFIT3**	↓D (21, 22)	0.9 (0.07)	0.9 (0.19)	0.9 (0.07)	Interferon-induced protein with tetratricopeptide repeats 3
IGFBP2***	↑D (22)	1.6 (0.07)	2.1 (0.01)	1.2 (0.61)	Insulin-like growth factor-binding protein 2 precursor
IL10***	↓ (13, 30)	0.8 (0.28)	0.9 (0.62)	0.7 (0.18)	Interleukin-10 precursor
IL13	↑ (125)	0.9 (0.14)	0.8 (0.02)	1.0 (0.85)	Interleukin-13 precursor
IL1B	↑ (4)	0.8 (0.31)	0.8 (0.31)	0.8 (0.45)	Interleukin-1 beta proprotein
KCNS3	↑ (2)	1.0 (0.81)	0.8 (0.26)	1.1 (0.48)	Potassium voltage-gated channel subfamily S member 3
KDELR2	↑ (42)	1.1 (0.45)	1.2 (0.1)	1.0 (0.73)	ER lumen protein retaining receptor 2 isof. 1
LCN15***	↑D (22)	1.0 (0.9)	0.8 (0.16)	1.2 (0.23)	Lipocalin 15
LYZ	↓ (2)	1.0 (0.96)	1.3 (0.32)	0.8 (0.36)	Lysozyme C precursor
MCM5***	↑ (2)	1.0 (0.7)	1.2 (0.1)	0.9 (0.32)	DNA replication licensing factor MCM5
MS4A4A**	↑ (2)	1.4 (0.07)	1.8 (0.01)	1.1 (0.71)	Membrane-spanning 4-domains subfam. A member 4a isof. 1
MUC20*	↑ (2)	1.2 (0.04)	1.3 (0.01)	1.1 (0.49)	Mucin-20 isof. S
MX1***	↓D (22)	1.0 (0.94)	1.1 (0.76)	0.9 (0.67)	Interferon-induced GTP-binding protein Mx1
NCF4*	↑ (2)	1.3 (0.04)	1.6 (0.001)	1.0 (0.99)	Neutrophil cytosol factor 4 isof. 2
NOS2	↑ (4)	0.9 (0.07)	1.0 (0.48)	0.8 (0.02)	Nitric oxide synthase, inducible
NR1H4	↑D (21)	1.1 (0.88)	1.9 (0.11)	0.6 (0.18)	Bile acid receptor isof. 2
OCLN**	↑C (21)	1.2 (0.06)	1.2 (0.06)	1.1 (0.17)	Occludin isof. B precursor
P2RY1***	↑ (81)	1.1 (0.51)	1.1 (0.37)	1.0 (0.79)	P2Y purinoceptor 1
P2RY2	↑ (81)	1.0 (0.83)	1.0 (0.31)	0.9 (0.17)	P2Y purinoceptor 2
P2RY4***	↑D (21, 22)	1.0 (0.89)	0.9 (0.13)	1.1 (0.2)	Pyrimidinergic receptor P2Y, G-protein coupled, 4
PDZD3***	↑D (21, 22)	1.1 (0.25)	1.2 (0.25)	1.1 (0.4)	Na(+)/H(+) exchange regulatory cofactor NHE-RF4 isof. 2
PSME2***	↑ (2)	1.0 (0.62)	1.2 (0.01)	0.9 (0.09)	Proteasome activator complex subunit 2
RBP2	↑D (22)	1.2 (0.33)	1.4 (0.08)	1.0 (0.89)	Retinol-binding protein 2
RFC4***	↑ (2)	1.1 (0.27)	1.3 (0.04)	1.0 (0.81)	Replication factor C subunit 4
S100A10	↑ (20, 106)	0.9 (0.11)	0.9 (0.34)	0.8 (0.07)	Protein S100-A10
SLC10A2	↑D (21)	1.0 (0.86)	1.1 (0.33)	0.9 (0.51)	Ileal sodium/bile acid cotransporter
SLC6A4	↓ (36, 45, 63, 76, 122)	1.9 (0.01)	1.3 (0.38)	2.9 (0.001)	Sodium-dependent serotonin transporter
TAP2***	↑ (2)	1.1 (0.79)	1.9 (0.05)	0.6 (0.12)	Antigen peptide transporter 2 isof. 2
TFF1	↑D (22)	1.0 (0.92)	1.1 (0.62)	0.9 (0.76)	Trefoil factor 1 precursor
TH	↑D (129)	0.9 (0.25)	0.9 (0.37)	0.9 (0.26)	Tyrosine 3-monooxygenase isof. B
TJP1***	↓ (21, 119)	0.9 (0.21)	0.9 (0.5)	0.9 (0.13)	Tight junction protein ZO-1 isof. A
TLR2***	↑ (12)	1.0 (0.86)	1.1 (0.64)	1.0 (0.86)	Toll-like receptor 2 precursor
TLR4***	↑ (12)	1.0 (0.63)	1.0 (0.92)	1.1 (0.36)	Toll-like receptor 4 isoform A precursor
TNFSF15	↓ (21, 112)	1.0 (0.88)	1.0 (0.68)	1.0 (0.88)	TNF ligand superfam. member 15 isof. VEGI-251 precursor
TPH1	↓ (63)	1.3 (0.17)	1.5 (0.12)	1.2 (0.4)	Tryptophan 5-hydroxylase 1
TPSAB1	↑ (4, 75)	0.8 (0.03)	0.9 (0.17)	0.8 (0.02)	Tryptase alpha/beta-1 precursor
TRPV1	↑ (64, 81, 131)	0.8 (0.05)	0.7 (0.08)	0.8 (0.1)	TRP cation channel subfamily V member 1
VIP*	↑D (21, 22)	0.7 (0.06)	0.6 (0.03)	0.8 (0.25)	VIP peptides isoform 2 preproprotein
VSIG2	↑ (2)	0.8 (0.07)	0.8 (0.16)	0.8 (0.08)	V-set and Ig domain-containing protein 2 precursor
VSIG4*	↑ (2)	1.4 (0.03)	1.7 (0)	1.1 (0.5)	V-set and Ig domain-containing protein 4 isof. 1 precursor
WDR72***	↓D (14)	0.8 (0.02)	0.8 (0.01)	0.9 (0.15)	WD repeat-containing protein 72

↑, increased vs. healthy controls (HCs); ↓ decreased vs. HCs; D, IBS with diarrhea; C, IBS with constipation; isof., isoform; dep., dependent; sub., subunit; Ig, immunoglobulin; recep., receptor; subfam., subfamily; MHC, major histocompatibility complex; TNF, tumor necrosis factor; TRP, transient receptor potential.

^*Our results agree with reported findings at P < 0.05,

^**P < 0.1 and

^***not significant but same direction, respectively.

Table 9.

Differentially expressed tight junction genes

	Peters	UCLA	Mayo/Aerssens
Symbol	FC (P)	FC (FDR)	FC (P, FDR)
MARK2	1.2 (0.02)	1.4 (0.03)	0.9 (0.49, 0.78)
LLGL2	1.2 (0.03)	1.3 (0.02)	1.0 (0.34, 0.69)
TJAP1	1.2 (0.10)	0.6 (0.01)	1.0 (0.35, 0.69)
PRKCZ	1.2 (0.11)	1.8 (0.01)	1.0 (1.00, 1.00)
CLDN15	1.2 (0.22)	1.4 (0.02)	0.9 (0.01, 0.32)
MLLT4	1.2 (0.36)	1.5 (0.01)	1.1 (0.02, 0.34)
PRKCI	1.1 (0.46)	1.3 (0.04)	1.1 (0.10, 0.47)
CSNK2A1	2.0 (0.50)	1.7 (0.00)	0.9 (0.01, 0.32)
CLDN6	1.1 (0.56)	0.7 (0.01)	0.9 (0.08, 0.45)
SPTA1	0.9 (0.69)	1.4 (0.03)	1.0 (0.20, 0.58)
CSDA	1.0 (0.75)	0.7 (0.02)	0.8 (0.00, 0.30)
MPDZ	1.1 (0.77)	0.7 (0.02)	1.1 (0.15, 0.52)
B2M	3.1 (0.83)	1.4 (0.03)	1.0 (0.04, 0.39)

Tight junction genes measured in IBS-C vs. HC by Peters et al. (97); 13 of 88 were differentially expressed (FDR < 0.05). FC, fold change, FDR, Benjamini-Hochberg adjusted P value.

Evidence for Neurally Mediated Mechanisms in IBS

cAMP-PKA signaling.

WGCNA and IPA analysis identified upregulation of cAMP-PKA signaling in IBS and IBS-C biopsies compared with HCs. The downregulation of PDE2A (validated by nCounter), which hydrolyzes cAMP, is also supportive. PKA signaling activates T-type calcium channels, and in our dataset, expression of the T-type calcium channel Cav3.1 (CACNA1G) was increased in IBS (both IBS-C and IBS-D). The potential importance of this channel in IBS is supported by a study in rats, in which knockdown of Cav3.1 attenuated butyrate-induced visceral hypersensitivity (83). In addition, Scanzi et al. (127) found increased expression of an alternate isoform with similar properties (Cav3.2, CACNA1H) in biopsies from IBS patients vs. HCs and in a mouse model of visceral pain.

While cAMP-PKA signaling is important in nociceptive pain signaling (11), this pathway regulates many aspects of cellular physiology including chloride secretion (15) and serotonin release by enterochromaffin cells (34). Furthermore, while evidence exists for altered neuronal sensitivity in IBS (9, 16–18, 27, 38, 95, 100, 118, 124), both desensitization (95) and sensitization (124) may occur. For these reasons, as well as the increased relative abundance of mucosal projections of enteric neurons compared with extrinsic afferents in the colonic mucosa, our findings are more likely to represent signaling mechanisms unrelated to pain. It is also possible that the differences in cAMP-PKA signaling may involve other cell types. Interestingly one of the hub genes in the cAMP-PKA module is an epithelial cell apical sodium channel (SCNN1D, correlated with module eigengene: 0.91), which may influence fluid transport and contribute to the constipation phenotype (62).

Neurotransmitter and neuropeptide signaling.

CRHR1 was downregulated (Mayo study, current study microarray, and nCounter), perhaps involving a response to increased stress-induced release of CRH. In a previous study (30), CRHR1 mRNA was not detectable in 81% of samples by quantitative real-time PCR, but microarray and nCounter are both more sensitive detection methods.

The γ-aminobutyric acid type A (GABA A) receptor γ1 subunit (GABRG1) was also downregulated in IBS. In a genomewide association study, a variant in this gene was associated with IBS within the discovery (but not replication) sample (42). In the enteric nervous system, activation of GABA A receptors exerts an excitatory effect (33). These receptors are also located on epithelial cells, and agonists stimulate luminal chloride secretion (73). Thus, the potential relevance of GABA receptor downregulation in IBS-C could be decreased peristalsis and/or secretion. An additional mechanism may be related to GABA receptor subunit stoichiometry. There are 19 GABA receptor subunit genes in the human genome, and several dozen subunit combinations for the pentameric receptor have been described (35). Differential composition of GABA receptors can affect membrane targeting and ligand sensitivity, which may be relevant in IBS (35, 65). It is also possible that the relevance to IBS is related to neuroimmune mechanisms, as GABA receptors are present on immune cells (8).

Neurite outgrowth.

Several genes related to neurite outgrowth were dysregulated in IBS-C vs. HCs both in our cohort and in Mayo (reticulon 3, SSPO, GRFA4). SSPO, which encodes the secreted glycoprotein SCO-spondin, which may participate in neurite outgrowth and neuron/glial interaction (85), was previously found to be hypermethylated in peripheral blood mononuclear cells in IBS (82). Interestingly, changes in neuron density have been reported in full-thickness samples from animal models of IBS (72, 116) and in patients with slow-transit constipation (123).

Additional Notable Findings

We also found that high-mobility group box 1 (HMGB1) was upregulated in IBS-C. HMGB1 is a chromatin-binding protein as well as a proinflammatory cytokine capable of increasing permeability of intestinal epithelial cell monolayers and ileal permeability and bacterial translocation in mice. While HMGB1 has well-defined roles in inflammation, it has recently emerged as a potential mediator of nociceptive pain signaling (3, 40, 66). For example, in a mouse model of cyclophosphamide-induced cystitis, it mediated mechanical bladder sensitivity induced by the protease-activated receptor PAR4 (66). HMGB1 is a target of cyclin-dependent kinase-5 (CDK5) (117), which was also upregulated and validated in our cohort. In a colon cancer-derived cell line, CDK5 phosphorylated HMGB1 at Ser¹⁸¹, a posttranslational modification that has been associated with increased secretion of HMGB1 (71). CDK5 activation is also pronociceptive through interactions with TRPV and purinergic receptors (91, 126).

We would also like to note the enrichment of a few gene sets that met the threshold of FDR < 0.25. Specifically, enrichment of “signaling by fibroblast growth factor receptor 1 (FGFR1) mutants” in HCs vs. IBS overall may warrant further investigation into this pathway. In addition, enrichment of the gene sets “allograft rejection”, “graft vs. host disease”, and “autoimmune thyroid disease” in men vs. women with IBS suggests that immune regulation may play a larger role in IBS among men.

Limitations

The most important limitation to consider in the interpretation of these results is that the results are not cell type specific; furthermore, there may be group differences in cell type composition of the biopsies with no reliable methods to measure or account for these differences. The upregulated gene sets in HCs vs. IBS-C suggest that there may be more neurons and/or increased transcriptional activity in neurons. However, decreased enteric neurons in the submucosal plexus has previously been associated with constipation (10). We attempted to address this question by looking at expression of traditional neuronal and astrocyte/glial markers (Table 10), which did not reveal significant differences. Additionally, there are limitations to drawing conclusions on neuronal physiology based on findings from mucosal biopsies. Although the mucosa is innervated by sensory nerve fibers, and biopsies frequently include some submucosa, an alternative explanation is that the expression changes may reflect alterations in transcriptional activity of glial cells or enteroendocrine cells.

Table 10.

Expression of neuronal and glial/astrocytic markers

		IBS-C vs. HC	IBS vs. HC	IBS-D vs. HC
Gene Name, Symbol (alias, Entrez ID)	Mean	FC (p)	FC (p)	FC (p)
Neuronal
Enolase2, ENO2 (neuron speific enolase, 2026)	6.8	1.0 (0.62)	1.0 (0.74)	1.1 (0.29)
Ubiquitin C-terminal hydrolase L1, UCHL1 (PGP9.5, 7345)	9.0	0.8 (0.2)	1.0 (0.9)	1.3 (0.1)
Microtubule associated protein 2, MAP2 (4133)	5.0	1.2 (0.12)	1.1 (0.42)	1.2 (0.31)
	8.2	0.9 (0.35)	1.0 (0.96)	1.0 (0.84)
Synaptophysin, SYP (6855)	6.5	1.1 (0.15)	1.1 (0.43)	1.0 (0.93)
Neurofilament, heavy, NEFH (4744)	15.2	1.0 (0.74)	1.0 (0.39)	0.9 (0.24)
Calbindin 2, CALB2 (calretinin, 794)	9.2	0.8 (0.14)	0.9 (0.3)	1.0 (0.77)
	7.5	0.6 (0.04)	0.7 (0.04)	0.7 (0.13)
Astrocyte/glial
Proteolipid protein 1, PLP1 (5354)	7.3	0.8 (0.08)	0.9 (0.49)	1.1 (0.54)
	6.3	0.9 (0.46)	1.0 (0.72)	1.0 (0.9)
S100 calcium binding protein B, S100B (6285)	6.1	0.8 (0.14)	1.0 (0.8)	0.9 (0.28)
Glial fibrillary acidic protein, GFAP (2670)	7.9	0.9 (0.56)	0.9 (0.4)	0.9 (0.38)

Mean is mean log2 normalized signal; FC, fold change.

Another significant limitation is the small sample size of the microarray study. However, the composition (equal number of men and women, medication use) and rigor of phenotyping increases the probability that significant findings are not due to sampling bias. Despite the larger sample size included in the IBS vs. HC comparison (n = 20 vs. n = 10 for IBS-C), we did not find DETs meeting a FDR < 0.05 threshold. This may be have been due to bowel habit-specific differences in peripheral mechanisms, as has been suggested by others (57). Importantly, our study was powered to detect only larger changes (FC ≥ 2) at an FDR ≤0.05. Therefore, caution should be used in the interpretation of negative results (absence of changes).

Reproducibility between IBS studies remains a challenge. There were 85 DETs in the literature that were present on our microarray, and we found the same direction of change (up or down in IBS vs. HCs) for 44 of these (Table 8). We did not find differences in SERT, which has been reported in several (36, 63, 122) but not all (20, 22) studies. Although we have attempted to highlight findings that are reproducible though comparison with publicly available data, this approach has computational limitations as well as limitations due to differences in protocol (e.g., different bowel preparation) and patient populations. In particular, the current study used Rome III (80) criteria, whereas Mayo and Nottingham used Rome II (115), which are less specific and may identify a less severe population. Given that inclusion of IBS patients in all three cohorts was supervised by experts in the field, the differences in diagnostic criteria are less likely to have resulted in sample differences than for survey-based studies. We recommend pooling resources and increasing collaboration with the goal of larger samples that are uniformly phenotyped and processed in order to yield robust findings.

Conclusions

In conclusion, using a small but well-phenotyped and age- and sex-balanced sample, we have found additional support for peripheral mechanisms in IBS and demonstrate that these findings were more significant in IBS-C patients within our cohort, suggesting that pathophysiological processes differ between IBS-C and IBS-D. Based on some of the observed changes, future investigations should focus on alterations in neuronal physiology.

GRANTS

This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases Grants P50 DK-64539 (E. A. Mayer, L. Chang) and T32 DK-07180 (E. J. Videlock) and Barbara and Joel Marcus GI Fellowship Seed Grant (E. J. Videlock). Study personnel: P30 DK-41301 (PI, J. E. Rozengurt). Core services: NCATS UL1TR000124.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

E.J.V. and L.C. conceived and designed research; E.J.V., S.M.-J., and L.C. performed experiments; E.J.V. and S.M.-J. analyzed data; E.J.V., S.M.-J., J.M.H., D.I., C.P., E.A.M., and L.C. interpreted results of experiments; E.J.V. prepared figures; E.J.V. drafted manuscript; E.J.V., S.M.-J., J.M.H., D.I., C.P., E.A.M., and L.C. edited and revised manuscript; L.C. approved final version of manuscript.