Comprehensive site-specific whole genome profiling of stromal and epithelial colonic gene signatures in human sigmoid colon and rectal tissue

Abstract

The strength of associations between various exposures (e.g., diet, tobacco, chemopreventive agents) and colorectal cancer risk may partially depend on the complex interaction between epithelium and stroma across anatomic subsites. Currently, baseline data describing genome-wide coding and long noncoding gene expression profiles in the healthy colon specific to tissue type and location are lacking. Therefore, colonic mucosal biopsies from 10 healthy participants who were enrolled in a clinical study to evaluate effects of lignan supplementation on gut resiliency were used to characterize the site-specific global gene expression signatures associated with stromal vs. epithelial cells in the sigmoid colon and rectum. Using RNA-seq, we demonstrate that tissue type and location patterns of gene expression and upstream regulatory pathways are distinct. For example, consistent with a key role of stroma in the crypt niche, mRNAs associated with immunoregulatory and inflammatory processes (i.e., CXCL14, ANTXR1), smooth muscle contraction (CALD1), proliferation and apoptosis (GLP2R, IGFBP3), and modulation of extracellular matrix (MMP2, COL3A1, MFAP4) were all highly expressed in the stroma. In comparison, HOX genes (HOXA3, HOXD9, HOXD10, HOXD11, and HOXD-AS2, a HOXD cluster antisense RNA 2), and WNT5B expression were also significantly higher in sigmoid colon compared with the rectum. These findings provide strong impetus for considering colorectal tissue subtypes and location in future observational studies and clinical trials designed to evaluate the effects of exposures on colonic health.

most studies to date have contrasted gene expression signatures and mutations in proximal vs. distal colorectal cancer (CRC) (1, 3, 21); however, recent data suggest that a linear continuum of tumor molecular signatures may change gradually along colorectal subsites from the ascending colon to the rectum (60). This is consistent with clinical practice, where colon and rectal cancers are treated as distinct entities (47), and a growing body of evidence that associations between dietary patterns, chemoprotective nonsteroidal anti-inflammatory drugs (NSAIDs), and CRC risk differ across anatomic subsites (21, 37). In contrast, a recent comprehensive molecular characterization of human colon and rectal cancer demonstrated that after excluding hypermutated cancers, colon and rectal cancers appear to have similar patterns of genomic alteration (40). In addition, although several molecular profiles have been used to predict prognosis in CRC patients (4, 17, 36), it now appears that only minor differences at the gene expression level are detected in tumors arising in the proximal colon, distal colon and rectum (47).

The development of colon cancer is intimately regulated by cross talk between the complex milieu of the stroma, containing fibroblasts, inflammatory immune cells and vascular cells, and the intestinal stem cell niche (19, 28, 43, 55). Thus, colonic biopsies represent intestinal-specific transcriptional patterns consisting of signals originating from stromal and epithelial cells (7, 12, 24). To dissect the role of epithelial and stromal genes in intestinal homeostasis, we recently demonstrated in an NSAID clinical trial that the expression of genes involved in cell signaling, cellular movement, and cancer differed significantly [P < 0.01, false discovery rate (FDR) <0.001] in the epithelium and stroma (53). These differences are consistent with recent reports that stromal transcripts have unique predictive value (8, 24) and support the importance of separately evaluating gene expression in epithelium and stroma. Currently, however, baseline data describing genome-wide coding and long-noncoding gene expression profiles in the healthy colon specific to tissue type and location are lacking.

RNAs that lack protein-coding ability and contain open reading frames exceeding 200 nucleotides are designated long noncoding RNAs (lncRNAs) (16). LncRNAs are now considered to be important regulators of cell fate and are important in all levels of nuclear organization, e.g., reactivation of retrotransposons in cancer (59). In addition, other lncRNAs can also act as decoys, sequestering intracellular mediators and suppressing their ability to fulfill tumor suppressor functions (10). Recently, several studies have identified clinically relevant lncRNAs in CRC, e.g., serve as prognostic biomarkers (25, 30, 51, 57). To date, a genome-wide transcriptional profiling of the stromal and epithelial colonic coding and noncoding gene signatures in human sigmoid colon and rectal tissue has not been performed. Using this approach, we demonstrate that patterns of mRNA and lncRNA expression and upstream regulatory pathways are distinct with respect to tissue type and location within the large intestine.

MATERIALS AND METHODS

Research design.

Colon and rectal samples were obtained from participants in an ongoing randomized, double-blind crossover intervention of flaxseed lignan extract and placebo. Each intervention period lasted 60 days with at least a 60-day washout between the two periods, and sigmoidoscopy was performed at the end of each intervention period. All study procedures and materials were approved by the Fred Hutchinson Cancer Research Center (FHCRC) Human Research Protection Program, Institutional Review Board Committee B, and Texas A&M University, and informed, written consent was obtained from all participants prior to their starting the study. This trial is registered at http://www.clinicaltrials.gov/ as NCT01619020. Here we report on mucosal and colon segment differences in gene expression on a subset of participants.

Participants.

Healthy men and women, ages 20–45 yr, were recruited from the greater Seattle area. Exclusion criteria included tobacco use, consumption of >2 alcoholic beverages/day (equivalent to 720 ml beer, 240 ml wine, 90 ml hard liquor), regular use of prescription or over-the-counter medications, oral or IV antibiotic use within the past 3 mo, weight loss or gain of >4.5 kg in the past year, current or planned pregnancy, breastfeeding, chronic medical illness, history of gastrointestinal disorder (e.g., gastric or duodenal ulcer, ulcerative colitis, Crohn disease, celiac sprue, hereditary nonpolyposis colorectal cancer, familial adenomatous polyposis, pancreatic disease, previous gastrointestinal resection, radiation or chemotherapy) and cancer (other than nonmelanoma skin cancer), known allergy to nuts, seeds, and flaxseed, contraindications to sigmoidoscopy, inability to swallow pills, and dietary fiber intake ≥20 g/day as assessed by the Block Fruit/Vegetable/Fiber Screener (Nutrition-Quest, Berkeley, CA). Further screening before randomization included a medical history, measurement of blood pressure, and complete blood count, liver panel, chemistry panel, blood urea nitrogen, serum creatinine, urinalysis, and in women, a pregnancy test. Participants also completed the colon cleanse with oral magnesium citrate (60 mg/ml; 296 ml, Safeway, Pleasanton, CA) to assure that they could tolerate this aspect of the procedure and to provide continuity in gut perturbation across the timeline of the study. In addition, a physician interviewed and examined participants before each of their sigmoidoscopies.

Sigmoidoscopy, biopsy collection, and tissue separation.

Participants prepared for flexible sigmoidoscopy by following standard instructions, which included adherence to a clear-liquid diet for 24 h before the procedure and drinking oral magnesium citrate solution (296 ml) the evening before. The sigmoidoscopy procedure was conducted in the Colonoscopy Suite at the Seattle Cancer Care Alliance Outpatient Clinic, Seattle. At each sigmoidoscopy, a large-cup flexible biopsy forceps [maximum capacity with needle (3.7 mm), Boston Scientific, Natick, MA] was used to biopsy normal-appearing mucosa from the sigmoid colon and rectum. Within 1 min of removal from the colon, five biopsies were transferred to Hanks' buffer containing 20 mM EDTA and 40 mM dithiothreitol at 4°C and were held on ice for 10–15 min. The epithelial cells were separated from the stromal layer by vortexing (53). The stromal layer was removed with an autoclaved toothpick, and the epithelial fraction was then collected by centrifugation. Both fractions were separately stored in Ambion Denaturation solution (LifeTechnologies, Grand Island, NY) at −80°C until RNA extraction.

RNA extraction.

Total RNA was isolated from stromal and epithelial biopsies using Ambion RNAqueous (LifeTechnologies) following kit instructions. After isolation, RNA was treated with DNase (Ambion DNAFree) and assessed for quality on a Bioanalyzer 2100 using an RNA 6000 Nano chip (Agilent, Santa Clara, CA). The average RNA integrity number was 8.4. RNA concentration was assessed with a NanoDrop spectrophotometer (Thermo Scientific).

RNA sequencing and gene expression analysis.

Sequencing libraries were generated from 250 to 1,000 ng total RNA using the TruSeq RNA Sample Preparation kit (Illumina, San Diego, CA) following the manufacturer's instructions and quantified using a Library Quantification kit (Kapa Biosystems, Wilmington, MA) as per manufacturer's instructions. The libraries were pooled and sequenced on an Illumina HiSeq 2500 at SeqWright Genomic Services (Houston, TX). The RNA reads obtained from sequencing were mapped with the STAR aligner (15) using the default parameters to the Ensembl GRCh38 human reference. Reads were examined for quality control using FastQC (13) and quantified using HTSeq-count (2). Subsequently, the edgeR software package (45) was used for data normalization and differential gene expression testing. Standard edgeR model-based normalization was selected using the library sizes and a trimmed mean-M correction factor, and tagwise dispersions were shrunken toward the trended local-fit dispersion as recommended in the edgeR manual. Once dispersion estimates were obtained and negative binomial models were fitted to the data, edgeR performs testing procedures for determining differential gene expression. Given the specific experimental design, we used the generalized linear model (GLM) option in edgeR. GLMs are an extension of the classical linear models to nonnormally distributed response data and, in our case, allowed for simultaneous testing for gene expression differences in tissue type and location, at the same time adjusting for participant and dietary treatment within the crossover design. Benjamini-Hochberg FDR correction was subsequently used to account for multiple testing, and genes were considered to be differentially expressed (DE) if the corrected P values were < 0.05. In complementary analyses, the effect of tissue type and biopsy site location on gene expression was assessed. For this purpose, stromal-epithelial gene coexpression relationships were assessed using simple linear regression models, which provide a basis for comparison between our results and similar studies (41).

Pathway analyses.

“Functional enrichment” analysis was performed using Ingenuity Pathway Analysis (IPA) version 2.0 software (Ingenuity Systems, Redwood City, CA) (54). For this purpose, all differentially expressed genes (corrected P values < 0.05) for the two major comparisons with respect to tissue type and location were uploaded into three columns consisting of Illumina probe ID, t value (fold-change), and corrected P value data. By convention, genes that were upregulated in the first listed condition are shown in red and those that were downregulated are shown in green. By default, during IPA analysis, only molecules from the data set associated with the Ingenuity Knowledge Base repository (Ingenuity Systems) were considered. Functional Analysis identified the biological functions and/or diseases that were most significant to the data set. The significance of the association between the data set and the specific pathways of interest was determined in three ways: 1) ratio of the number of molecules from the data set that mapped to the pathway divided by the total number of molecules that mapped to the Ingenuity Knowledge Base pathway, 2) Fisher's exact test was used to calculate a P value determining the probability that the association between the genes in the data set and the pathway of interest could be explained by chance alone, and 3) activation state (“increased” or “decreased”) was inferred by the activation z-score. The derivations of the z-scores are based on relationships in the molecular network that represent experimentally observed “causal” associations between genes and those functions.

“Canonical pathway” analysis was used to identify networks from the IPA library that were most significantly modulated. Significance of the association between each data set and a canonical pathway was measured in two ways: 1) A ratio of the number of molecules from the data set that mapped to the pathway divided by the total number of molecules that mapped to the pathway, and 2) Fisher's exact test (see above).

“Upstream regulator” analysis was based on prior knowledge of expected associations between transcriptional regulators and their target genes stored in the Ingenuity Knowledge database, and significance for each transcriptional regulator was measured in two ways: Fisher's exact test (P value) was used to identify differentially expressed genes from the RNA-Seq data set that overlapped with genes known to be regulated by a transcriptional regulator. Since the regulation direction (“activating” or “inhibiting”) of an edge is not taken into account for the computation of overlap P values, the underlying network also included findings without associated directional attribute, such as protein-DNA (promoter) binding. In addition, the activation score (z-score) was used to infer the status of predicted transcriptional regulators by comparing the observed differential regulation of genes (“up” or “down”) in the data set relative to the literature-derived regulation direction, which can be either activating or inhibiting.

Noncoding RNA.

The Ensembl GRCh38.p3 release of the human genome containing annotations for 14,889 long noncoding genes in addition to the 20,296 coding genes and 14,426 pseudogenes was used to map the generated sequences. Reads [>200 bp in an intronic and intergenic (lincRNA) region] mapping against these additional regions were then included in our edgeR differential expression tests in addition to the coding genes. We considered all these results simultaneously when performing multiple testing correction as described previously (31, 48).

RESULTS

Our study utilized a GLM followed by gene network (IPA) analysis for the purpose of simultaneously testing for gene expression differences in tissue type and location, at the same time adjusting for dietary treatment within the crossover design. In this section we first report the results of the GLM analysis. Assessment of biopsy gene expression patterns revealed 5,428 DE genes between epithelial and stromal tissue types with 2,850 upregulated and 2,578 downregulated in the stromal vs. epithelial tissue comparison. Lists of the top DE protein-coding and noncoding genes between stromal and epithelial tissue are presented in Tables 1 and 2, respectively (for a full listing see Supplemental Table S1).¹ In contrast, testing for differences in sampling location revealed a total of 491 DE genes with 241 upregulated and 250 downregulated in the sigmoid colon vs. rectum comparison. The respective top DE coding and noncoding genes in the sigmoid colon and rectum are listed in Tables 3 and 4 (for a full listing see Supplemental Table S2). Using the DE lists of genes for the two major comparisons with respect to tissue type and location, we ran an IPA network analysis (33). The tissue-specific analysis revealed upstream regulators predicted to be activated and deactivated. The top 10 predicted activated and deactivated upstream regulators of stromal/epithelial comparison are presented by P value in Tables 5 and 6, respectively. Representative findings from this network analysis are illustrated in Fig. 1. Differences between sigmoid colon and rectum after pathway analysis resulted in the detection of homeobox genes of the HOX family of transcription factors (Fig. 2). This is noteworthy, because HOX-regulated genes are involved in cell differentiation, and their deregulation is associated with CRC (6, 27). Potential gene interaction between tissue location and tissue type was assessed, but no significant terms were detected after multiple testing correction. Interestingly, 200 genes (Supplementary Tables S1 and S2) were DE with respect to both stromal/epithelium and sigmoid/rectum comparisons (Fig. 3). The groups of DE genes with respect to only one of the two comparisons were used to form putative signature sets of genes.

Table 1.

Top 20 differentially expressed protein-coding genes between stromal and epithelial tissue

Gene Names	Fold Change (stromal/epithelial)	FDR (corrected P value)
Top 10 upregulated protein coding genes
GLP2R	37.80	7.81E-37
ANTXR1	14.87	3.40E-35
NES	8.83	1.40E-33
CXCL14	21.05	5.11E-33
CALD1	11.36	3.80E-31
IGFBP3	4.12	3.80E-31
MMP2	8.81	3.80E-31
COL3A1	16.57	4.63E-31
MFAP4	9.43	7.33E-31
MAP1B	6.68	7.44E-31
Top 10 downregulated protein coding genes
HBA2	0.07	6.41E-13
HBA1	0.07	2.51E-12
MATK	0.49	1.67E-11
THOC6	0.73	2.91E-11
PMF1	0.70	3.37E-10
C19orf48	0.58	4.75E-10
MCM7	0.72	7.17E-10
KLRD1	0.53	1.82E-09
HBB	0.11	1.90E-09
CCL5	0.47	2.02E-09

Table 2.

Top 20 differentially expressed long noncoding RNA genes between stromal and epithelial tissue

Gene Name	Gene Biotype	Fold Change	FDR
Top 10 upregulated noncoding genes
FENDRR	lincRNA	19.44	8.02E-31
MEG3	lincRNA	10.95	4.70E-20
RGS5	antisense	2.88	1.29E-16
C14orf132	lincRNA	4.15	1.40E-13
LINC01207	lincRNA	1.32	1.20E-05
FGD5-AS1	antisense	1.13	1.08E-03
FLJ22763	lincRNA	1.29	5.40E-03
MIR22HG	lincRNA	1.27	6.55E-03
KRTAP5-AS1	antisense	1.33	7.34E-03
DNMBP-AS1	antisense	1.22	9.67E-03
Top 10 downregulated noncoding genes
SNHG19	lincRNA	0.62	1.19E-05
SNHG8	lincRNA	0.73	1.97E-05
LINC01004	antisense	0.64	9.33E-05
EPB41L4A-AS1	lincRNA	0.74	2.30E-04
SNHG7	antisense	0.78	2.38E-04
SNHG15	lincRNA	0.72	2.56E-04
PRKAG2-AS1	antisense	0.70	3.86E-04
LINC00342	lincRNA	0.60	7.41E-04
FAM201A	antisense	0.70	8.73E-04
PKI55	lincRNA	0.79	1.53E-03

FDR, false discovery rate; lincRNA, long intergenic noncoding RNA.

Table 3.

Top 20 differentially expressed protein-coding genes between sigmoidal and rectal tissue

Gene Name	Fold Change (sigmoid/rectum)	FDR (corrected P value)
Top 10 upregulated protein-coding genes
HOXD10	6.03	2.69E-20
HOXD9	5.84	6.58E-20
OR51E2	4.21	8.86E-19
HOXD11	5.11	1.55E-18
NTRK2	3.03	1.36E-17
KIAA1715	1.44	1.80E-16
WNT5B	2.68	1.80E-16
HOXD12	5.32	2.80E-16
ATP12A	6.31	4.77E-16
PARP8	1.42	1.24E-15
Top 10 downregulated protein-coding genes
URAD	0.03	6.06E-32
HOXA3	0.54	2.29E-17
GPT2	0.64	2.05E-15
SERPINA6	0.36	6.37E-15
HOXA5	0.57	1.73E-14
ADRA2A	0.55	1.30E-13
FSIP2	0.48	2.15E-13
NPY1R	0.42	3.18E-13
FAM213A	0.62	8.63E-13
LUZP2	0.46	2.03E-11

Table 4.

Top 20 differentially expressed long noncoding RNA genes between sigmoidal and rectal tissue

Gene Name	Gene Biotype	Fold Change	FDR
Top 10 upregulated noncoding genes
HOXD-AS2	antisense	3.97	1.01E-16
LINC00094	antisense	1.24	1.44E-01
LINC00668	lincRNA	1.31	1.48E-01
CH17-360D5.1	lincRNA	1.39	1.87E-01
LINC01137	antisense	1.22	2.70E-01
CTBP1-AS2	antisense	1.10	2.89E-01
LINC00940	lincRNA	1.18	3.80E-01
LINC00607	lincRNA	1.48	4.06E-01
DIO3OS	lincRNA	1.29	4.48E-01
FGD5-AS1	antisense	1.06	4.80E-01
Top 10 downregulated noncoding genes
HOXA-AS3	antisense	0.49	5.18E-07
HOTAIRM1	antisense	0.61	9.48E-06
LINC00520	lincRNA	0.43	2.94E-03
CASC9	lincRNA	0.67	1.16E-02
HOXA-AS2	antisense	0.64	1.48E-02
LINC00543	lincRNA	0.80	6.45E-02
DNAJC3-AS1	lincRNA	0.83	7.88E-02
HAGLR	antisense	0.89	8.40E-02
COLCA1	antisense	0.79	9.48E-02
DPP10-AS1	antisense	0.83	1.13E-01

Table 5.

Predicted activated upstream regulators of stromal/epithelial comparison using IPA

Upstream Regulator	Log Ratio	Predicted Activation State	Activation z-Score	P Value of Overlap
TP53	−0.44	activated	5.19	1.52E-30
CEBPA	0.20	activated	4.09	6.27E-05
HTT	0.02	activated	3.84	9.28E-11
CTNNB1	0.18	activated	2.83	4.18E-20
MITF	0.59	activated	2.77	4.90E-05
POU5F1	−0.20	activated	2.77	2.84E-02
EGR1	0.41	activated	2.75	7.40E-04
JUNB	0.21	activated	2.64	8.64E-05
FOXF1	4.00	activated	2.63	3.79E-08
PAX6	−0.38	activated	2.57	1.07E-02

IPA, Ingenuity Pathway Analysis.

Table 6.

Predicted inhibited upstream regulators of stromal/epithelial comparison using IPA

Upstream Regulator	Log Ratio	Predicted Activation State	Activation z-Score	P Value of Overlap
AHR	0.33	inhibited	−4.38	3.28E-23
MYC	−0.54	inhibited	−3.47	1.27E-18
GLIS2	1.20	inhibited	−3.02	2.61E-08
GMNN	−0.49	inhibited	−3.00	1.47E-02
MED1	0.03	inhibited	−2.88	2.61E-07
FOXO1	0.19	inhibited	−2.39	1.01E-08
CCND1	0.07	inhibited	−2.32	4.44E-08
NR4A2	−0.15	inhibited	−2.23	4.88E-02
PAX5	−2.46	inhibited	−2.17	4.46E-04
TCF3	−0.21	inhibited	−2.16	7.04E-04

Fig. 1.

Genes downstream of TP53 and MYC are differentially expressed in stromal vs. epithelial tissue types. Potential upstream regulators as identified by Ingenuity Pathway Analysis (IPA) are listed. Upregulation (red) and downregulation (green) are noted with respect to the stromal/epithelial expression ratio. Predicted activation (orange ovals and arrows) and predicted inhibition (blue ovals and arrows) were estimated by IPA based on downstream gene expression patterns. See Supplementary Table S1 for a complete list of fold-changes and P values.

Fig. 2.

Homeobox genes of the HOX family are differentially expressed in the sigmoid colon and rectum. Upregulation (red) and downregulation (green) are noted with respect to the sigmoid/rectum expression ratio. See Supplementary Table S2 for a complete list of fold-changes and P values.

Fig. 3.

Differentially expressed genes across stroma/epithelium and sigmoid/rectum comparisons. The Venn diagram represents the relationship between the detected differentially expressed genes in each 1 of the 2 comparisons, i.e., stroma vs epithelium and sigmoid vs. rectal tissue.

Since a genome-wide catalogue of lncRNAs associated with stroma and epithelium from the sigmoid colon and rectum has not been previously generated, we compiled lists of differentially expressed annotated lncRNAs. A total of 205 lncRNAs (Supplemental Tables S3 and S4) were identified in the stroma/epithelial compartments associated with the sigmoid colon and rectum. The stroma was particularly enriched in lncRNAs linked to cancer and embryogenesis, e.g., FENDRR (17a, 58) and MEG3 (34, 38, 61). Interestingly, with respect to sigmoidal and rectal tissue distribution, HOTAIRM1, a potential prognostic marker of colorectal cancer (32, 51), was predominantly expressed in the rectum.

In the complementary stromal-epithelial gene coexpression analyses, we exhaustively computed 13,491 × 13,491 = 182,007,081 pairwise associations of genes expressed in both epithelial and stroma. We found 18,849,925 epithelial-stromal gene coexpression interactions to be significant (FDR < 0.05). The top 10 (smallest FDR values) interactions are reported in Table 7. Sets of the most highly correlated genes in the stroma and epithelium included immune inflammatory (CXCL9, HLA-DR, SAA2), reactive oxygen species (DUOXA2, NOS2), tryptophan metabolites (IDO1), barrier function (TGM2), and apoptosis (FAM3B)-related genes.

Table 7.

Top-ranked epithelial-stromal coexpression relationships in the colon

Epithelium	Stroma	T-Stat	P Value	FDR
CXCL9	IDO1	116.86	4.77E-48	8.69E-40
CXCL9	SAA1	75.57	2.93E-41	2.66E-33
CXCL9	NOS2	70.96	2.78E-40	1.69E-32
CXCL9	SAA2	66.49	2.84E-39	1.29E-31
CXCL9	C6orf223	53.06	8.82E-36	3.21E-28
HLA-DRA	SAA2	52.58	1.22E-35	3.69E-28
CXCL9	DUOXA2	52.09	1.70E-35	4.41E-28
TGM2	SAA2	49.96	7.48E-35	1.70E-27
CD164L2	SAA2	48.49	2.16E-34	3.31E-27
FAM3B	CXCL11	48.48	2.18E-34	3.31E-27

DISCUSSION

Using whole genome profiling, we showed substantial differences in patterns of gene expression and upstream regulatory pathways that are distinct with respect to tissue type (i.e., stroma vs. epithelium) and location within the large intestine (i.e., sigmoid colon vs. rectum). To our knowledge, this is the first study to globally contrast the expression signatures of coding and lncRNAs associated with stromal vs. epithelial tissue in the sigmoid colon and rectum.

Stroma vs. epithelium.

In the human colon, the gut mucosa consists of the epithelium, which is made up of a single sheet of columnar epithelial cells, and the lamina propria, a thin layer of loose connective tissue, which lies beneath the epithelium. The epithelial layer forms finger-like invaginations into the underlying connective tissue of the stroma. The base of these invaginations, or crypts, houses the multipotential stem cells, capable of generating all epithelial cell types (23). Three major terminally differentiated cell types are present in the colon: colonocytes or absorptive cells, mucus-secreting goblet cells, and peptide hormone-secreting endocrine cells. The lamina propria consists of a more heterogeneous collection of cell types, including fibroblasts, lymphocytes, macrophages, eosinophilic leukocytes, and mast cells (55). A punch biopsy also accesses part of the submucosa directly below the crypts, further contributing to the heterogeneity of the stromal fraction. With regard to the top 20 DE genes between stroma and epithelium, mRNAs associated with immunoregulatory and inflammatory processes (i.e., CXCL14, ANTXR1), smooth muscle contraction (i.e., CALD1), proliferation and apoptosis (i.e., GLP2R, IGFBP3), and modulation of extracellular matrix (i.e., MMP2, COL3A1, MFAP4) were all highly expressed in the stroma. Thomas et al. (53) also showed previously, using expression arrays, that sigmoid colon gene expression differed between stroma and epithelium in normal healthy colon. Using the same approach to separate stroma and epithelium as we describe here, they confirmed good separation of the two tissue components through analysis of genes known to be restricted to either stroma or epithelium by immunohistochemistry. Genes preferentially expressed in stroma included ones involved in inflammation, cellular adhesion, and extracellular matrix production. In comparison, genes preferentially expressed in epithelium were involved in metabolism of xenobiotics, fatty acids and lipids, and in apoptosis signaling, and ion transport.

As a first step to globally probe potential epithelial-stromal cross talk, we also examined stromal-epithelial gene coexpression relationships. This analysis revealed 18,849,925 significant (FDR < 0.05) epithelial-stromal coexpression interactions significant at FDR < 0.05 (Table 7). In comparison, Oh et al. (41) reported 929 significant epithelial-stromal coexpression interactions among 11,700 × 11,700 = 136,890,000 pairwise associations tested in normal breast tissue. Interestingly, all the top detected genes have been implicated in the link between inflammation and cancer of the colon (5, 18, 20, 35, 39, 44, 46, 49, 56, 62). Additional systems-level analyses on independent RNA-Seq data sets are needed to further explore how colonic stromal and epithelial interactions temporally evolve during malignant transformation.

Using IPA network analysis, we identified several activated and inhibited upstream regulators of the stromal/epithelial comparison. Predicted activated regulators (Table 5) included genes for proteins implicated in the regulation of a range of activities (e.g., cellular-stress responses, cell cycle regulation, cell adhesion, development, and stem cell pluripotency) supporting the heterogenous nature of stroma. Predicted inhibited upstream regulators (Table 6) represent transcription factors associated with stroma/epithelial cell development (e.g., regulation of xenobiotic metabolizing enzymes, cell fate and differentiation, and cell cycle control). These data provide insight into the gene expression landscape of the normal epithelium and stroma prior to the onset of intestinal tumorigenesis. This is noteworthy because the development of cancer is intimately linked to cross talk between cancer cells and the surrounding stromal cells (microenvironment) (19, 43, 55).

Sigmoid colon vs. rectum.

Based on clinical pathology and molecular genetic analysis, the rectum (15 cm from the anal verge) is often considered as part of the distal colon and not a separate entity, since from an embryonic perspective both are derived from the hindgut (40). Nonetheless, in relation to CRC, malignant predilection of adenoma has been reported to be site-dependent, with rectal polyps harboring a more aggressive phenotype than those of the sigmoid colon (50). Furthermore, we have detected differences in methylation patterns among different colon segments, particularly related to specific cancer-related genes (29). In the present study, we identified 491 genes that were DE in sigmoid colon compared with rectum. Of the top 20, five were HOX genes (HOXA3, HOXD9, HOXD10, HOXD11, and HOXD-AS2, a HOXD cluster antisense RNA) with all but HOXA3 expressed at higher levels in sigmoid colon. This is noteworthy, because several HOX genes have been associated with tumor location (47). In addition, HOX overexpression contributes to stem cell overpopulation that drives colon tumorigenesis (6).

WNT5B expression was also significantly higher in sigmoid colon than rectum. The secreted signaling proteins derived from the WNT genes have been implicated in epithelial and stromal cell development and differentiation, but also in oncogenesis. Wnt/β-catenin signaling has a central role in the development and progression of most colon cancers (11). Despite similar colon architecture between the two gut segments, the differences in methylation (14, 29) and our observed gene expression suggest possible differences in epigenetic regulation either through endogenous control or differential microbial metabolism and subsequent exposures from the lumen (9, 22, 26). Therefore, rectal mucosal biopsies may not be adequate to represent epigenetic effects in other parts of the distal colon.

In summary, we demonstrate that patterns of gene expression and upstream regulatory pathways are distinct with respect to tissue type and location within the large intestine. In addition, we were able to detect a range of lncRNAs, which include the enrichment of FENDRR and MEG3 in the stroma. With respect to the longitudinal axis of the colon, HOTAIRM1, a potential prognostic marker of colorectal cancer (32, 51), was predominantly expressed in the rectum. We propose that these gene sets can be used as a baseline for gene expression in the healthy colon specific to tissue type and location.

GRANTS

This work was funded in part National Institutes of Health Grants U01 CA-162077, R35CA-197707, and P30ES-023512.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

J.M.K., E.K., I.I., L.A.D., J.S.G., M.A.H., and T.W.R. analyzed data; J.M.K., E.K., I.I., L.A.D., J.S.G., M.A.H., T.W.R., J.L., and R.S.C. interpreted results of experiments; J.M.K., E.K., I.I., and J.S.G. prepared figures; J.M.K., E.K., I.I., L.A.D., J.L., and R.S.C. drafted manuscript; J.M.K., E.K., I.I., L.A.D., M.A.H., T.W.R., A.M.K., L.L., J.L., and R.S.C. edited and revised manuscript; J.M.K., E.K., I.I., L.A.D., J.S.G., M.A.H., T.W.R., A.M.K., L.L., J.L., and R.S.C. approved final version of manuscript; E.K., L.A.D., J.S.G., L.L., and J.L. performed experiments; I.I., L.A.D., M.A.H., T.W.R., A.M.K., L.L., J.L., and R.S.C. conception and design of research.