Inflammation-associated DNA methylation patterns in epithelium of ulcerative colitis

ABSTRACT

Aberrant DNA methylation patterns have been reported in inflamed tissues and may play a role in disease. We studied DNA methylation and gene expression profiles of purified intestinal epithelial cells from ulcerative colitis patients, comparing inflamed and non-inflamed areas of the colon. We identified 577 differentially methylated sites (false discovery rate <0.2) mapping to 210 genes. From gene expression data from the same epithelial cells, we identified 62 differentially expressed genes with increased expression in the presence of inflammation at prostate cancer susceptibility genes PRAC1 and PRAC2. Four genes showed inverse correlation between methylation and gene expression; ROR1, GXYLT2, FOXA2, and, notably, RARB, a gene previously identified as a tumor suppressor in colorectal adenocarcinoma as well as breast, lung and prostate cancer. We highlight targeted and specific patterns of DNA methylation and gene expression in epithelial cells from inflamed colon, while challenging the importance of epithelial cells in the pathogenesis of chronic inflammation.

Introduction

Ulcerative colitis (UC) is one of the major subtypes of inflammatory bowel disease (IBD). Its phenotype is characterized by chronic inflammation of the intestinal colonic mucosa.¹ Individuals affected by long-standing UC experience multiple cycles of inflammation, damage to the colonic epithelium and repair.^1,2 The factors that control the pathogenesis of ulcerative colitis are being elucidated but, to date, our knowledge of this process is incomplete. Current models of UC pathogenesis invoke exaggerated and prolonged dysregulated immune responses to normally innocuous intestinal microbial antigens that develop in genetically pre-disposed individuals.^3–5 In the inflamed mucosa of UC patients, there is a greatly expanded acute and chronic immune infiltrate of the lamina propria, accompanied by an abundance of secreted factors from those cells, notably cytokines, such as TNF-α, IL-1, IL-6, and many others.^6–9 The consequences of sometimes many years of exposure to those factors on the intestinal epithelial cells (IECs) of the colon are not known. In vitro studies of IECs have indicated that they are capable of immune functions, notably the secretion of cytokines and chemokines that could influence the immune infiltrate of the lamina propria.^10–13 However, little is known about the specific role played by epithelial cells in the pathogenesis of UC from in vivo studies of patients.

In previous work, we showed that exposure of IECs to the pro-inflammatory cytokine IL-6 results in increased methylation of DNA, via stabilization of the DNA methylation enzyme DNMT1.¹⁴ IL-6-induced DNA methylation was accompanied by an altered phenotype of the IECs, including enhanced migration and ability to form foci in soft agar, 2 processes that are associated with neoplasia.¹⁴ However, whether the prolonged exposure of the colonic epithelium to inflammation in the disease setting of UC is also associated with altered DNA methylation of epithelial cells is not known. Variable DNA methylation patterns have been observed in colitis-associated cancer,^15,16 and have been shown to contribute to aberrant epigenetic gene silencing in sporadic colorectal cancer.^17,18 It is therefore plausible that aberrant DNA methylation might link chronic inflammation with carcinogenesis.

Prior studies have used genome-wide approaches to highlight distinct epigenetic patterns between affected diseased samples in IBD and non-affected controls.^19–24 Interpretation of results from those studies, in which whole mucosal biopsies were used to extract DNA, must consider the cellular heterogeneity of the samples. Whole colonic biopsies consist of a mixture of different cell types, including epithelial cells, stromal cells (such as fibroblasts), immune cells (such as macrophages and lymphocytes), and endothelial cells. It is known that different cell types have different patterns of epigenetic and transcriptional regulation.²⁵ Furthermore, the relative proportions of cell subtypes in the samples from which DNA and RNA are extracted can profoundly affect the overall DNA methylation and transcriptome pattern observed.²⁶ One other important factor in the design of experiments to assess epigenetic profiles is the inter-individual variation that exists in epigenetic signatures,²⁷ specifically in the presence of disease.²⁸

To account for both of these considerations, we chose to isolate and purify intestinal epithelium from whole colonic biopsies obtained from human subjects with sub-total UC. We aimed to utilize pure epithelial cells to generate within-patient, genome-wide DNA methylation and gene expression maps of affected (i.e., inflamed) and matched unaffected (i.e., non-inflamed) areas of the large intestine. We aimed to use these intra-individual maps to reflect the potential epigenetic variation at the intestinal origin of UC pathogenesis. Moreover, we wish to utilize this data to gain insight into the molecular mechanism underlying the progression of IBD to colitis-associated cancer.

Results

We report DNA methylation analysis and transcriptome analysis comparing distal (inflamed) and proximal (non-inflamed) colonic regions in purified epithelial cells in human individuals with sub-total UC (n = 13 sample-pairs).

Establishment of colonic cell suspensions enriched in epithelial cells

Our method is a modification of 2 previously developed techniques^29,30 that allows the detachment of whole epithelial crypts from mucosal biopsies of the colon. Flow cytometry with markers specific for IECs and bone marrow-derived cells were used to assess the cellular make-up of the suspensions resulting from the chelation procedure. It was found that more than 90% of the cell suspension comprised EpCAM positive cells, indicative of epithelial cells (Fig. 1). In this case, approximately 5% of the suspension cells in the inflamed samples were stained with the CD45 antibody, indicating bone marrow origin (Fig. 1D). The double negative cells in this analysis could be stromal cells, such as fibroblasts, or endothelial cells, which are neither epithelial nor of bone marrow origin. These data indicated a successful enrichment of epithelial cells in the suspension. We identified no significant difference between the proportion of CD45 positive cells between inflamed and non-inflamed regions of the colon. However, the absolute yield of DNA and RNA evaluated in each cell isolate in the distal and proximal region of diseased and non-diseased patients was also evaluated (Fig. S1). The mean yield of DNA in non-diseased specimens (7.83 ± 0.9 μg) was greater than that of diseased specimens (5.81 ± 0.6 μg). The total yield of RNA in each state also follows a similar trend (normal: 4.66 ± 0.64 μg; ulcerative colitis: 1.95 ± 0.31 μg), thus suggesting differences in epithelial cell yield between inflamed and non-inflamed states.

Figure 1.

Classification of cellular proportions in non-inflamed and inflamed colonic regions. Intestinal epithelial cells (IECs) isolated from mucosal pinch biopsies are illustrated (A-D). IECs were labeled with fluorescent antibodies EpCAM and CD45 to distinguish cell populations. Representative histograms of EpCAM positive labeled cells (red) and its isotype control (blue) are illustrated in non-inflamed (A) and inflamed (C) regions. Quantification of the percentage of epithelial cells in the IEC isolate was then performed. Representative scatterplots of epithelial positive cells (upper left), CD45 positive cells (lower right) and double negative cells (lower left and upper right) are illustrated in non-inflamed (B) and inflamed (D) regions.

Genome-wide DNA methylation: Sequence data and coverage

DNA methylation was assayed in inflamed and non-inflamed samples using the HELP-tagging assay. Multiplexing of TruSeq HELP-tag library samples was done with 6 libraries per lane. DNA methylation was measured at ∼1.9 million CCGG sites, and ∼1.6 million sites remained when sites with fewer than 5 MspI reads were removed. DNA methylation levels were measured using a modified version of the angle methylation score; this ranged from 0 (no DNA methylation) to 100 (complete DNA methylation). The average number of reads for all samples varied from 13.5 to 24.6 million HpaII reads per sample with an average depth of coverage of CCGG sites between 11.6 and 19.4x (Table S1).

Genome wide patterns of DNA methylation

The majority of CCGG sites in the genome were methylated (DNA methylation score ≥70) (Fig. S2A). Consistent with previous reports^31–34 a higher proportion of non-methylated CCGG sites (DNA methylation score ≤30) fell within the vicinity of the transcription start sites (TSS) of genes (±2 kb). The distribution in this region was bimodal, whereas other genomic locations such as the gene body and intergenic regions were predominantly methylated (Fig. S2A).

Gene regulation by epigenetic modification takes place at promoters and distally located regulatory elements.³⁵ To characterize the DNA methylation patterns at promoter regions, the mammalian expression atlas generated by the FANTOM consortium³⁶ was used to map CCGG sites that fall within the vicinity of site (TSS) ±2 kb. Results illustrated the relationship between CCGG rich and depleted regions and the relative DNA methylation state at these regions. As reported previously,³⁴ the proportion of CCGG sites was higher at the TSS, becoming relatively depleted up and downstream of the TSS. However, DNA methylation decreased in close proximity to TSS (mean score = 21.0 at the TSS) and then increased both up- and down-stream from each TSS peak (mean score = 74.1 ±2 kb from TSS) (Fig. S2B). This characterized the typically unbalanced nature of DNA methylation in a normal state, with 70–80% of the genome being methylated, whereas non-methylated loci generally tended to cluster in groups around the TSS of protein-coding genes.³⁷

Identification of differential methylation between inflamed and non-inflamed colonic regions in purified epithelial cell samples

During the initial analysis, a suspected mislabeling of paired samples was discovered. This was recognized during differential methylation analysis, as one pair was consistently methylated in the opposite direction of its grouped counterparts at each CCGG site (Fig. S3). This pattern was also present for the same sample in the transcriptome data set. Due to the manner in which samples are obtained, this consistent pattern observed in both DNA and RNA may therefore be attributable to upstream sample mishandling, perhaps during sample collection or cell isolation. For this reason, this pair of samples was excluded from any further analysis. Therefore, DNA methylation analysis comparing distal (inflamed) and proximal (non-inflamed) colonic regions in purified epithelial cell populations in individuals with sub-total UC (n = 12 pairs) is reported.

Using unsupervised hierarchal clustering, we identified no global differences in patterns of DNA methylation between inflamed and non-inflamed colonic regions, but numerous specific loci at which differential methylation occurred (n = 577). Of these differentially methylated sites (DMS), 371 (64%) showed higher methylation in inflamed regions compared with non-inflamed regions of the colon (Fig. 2A). A substantial number of genes contained multiple DMS mapping; of the 577 DMS, 324 mapped to 210 unique genes (Table S2). However, some variation may be attributable to the distinct epigenomic signatures observed in proximal and distal colonic regions; as identified previously.³⁴ This is reflected by 41 intersecting DMS mapping to 28 genes identified in normal epithelia and in the presence of disease (Table S3).

Figure 2.

Site-specific differential methylation between non-inflamed and inflamed colonic regions. DNA methylation values of DMS using a color scale from red (high DNA methylation) to yellow (low DNA methylation). Columns represent samples (n = 12 pairs) and rows represent all differentially methylated CCGG sites (n = 577) in inflamed (red) and non-inflamed samples (blue) (A). DMS mapping to active TSS regions (red) and enhancer regions (yellow) of HOXB3 (B), HOXB4 (C), and HOXB5 (D), as well as the ChIP-seq profiles of the set of histone marks assayed in colonic mucosa is shown. Below the histone marks, DMS (red) and all CCGG sites (green) mapping to each gene segment are also illustrated.

Generally, genes containing multiple DMS had similar patterns of DNA methylation, as has been documented in previous studies.^38,39 This was the case, for example, for HOXB3, HOXB4, HOXB5, HOXB6, HOXB7 HOXC4, FOXA2, TNS3, and OSR2, as these protein-coding genes showed similar patterns of differential methylation at multiple CCGG sites (Table S2). Among the DMS, 104 mapped to TSS peaks associated with 55 unique protein-coding genes (Table 2).

Table 2.

. List of significantly (Benjamini-Hochberg adjusted cutoff of 0.2) differentially methylated sites between inflamed and non-inflamed samples mapping to TSS peaks (±2 kb).

DMS	Gene Symbol	Description	ΔMethylation	FDR
chr17-46683559	HOXB7	homeobox B7	84.6260971	2.45E-10
chr17-46685296	HOXB7	homeobox B7	75.1675856	7.28E-06
chr20-22563052	FOXA2	forkhead box A2	-61.6184768	7.28E-06
chr20-22561829	FOXA2	forkhead box A2	-68.4510447	4.08E-05
chr20-22562626	FOXA2	forkhead box A2	-66.1910553	9.44E-05
chr3-25469915	RARB	retinoic acid receptor, beta	67.3695157	0.000201385
chr20-22562885	FOXA2	forkhead box A2	-60.7635967	0.000287983
chr12-54428544	HOXC5	homeobox C5	-55.7262250	0.000536599
chr2-234744505	HJURP	holliday junction recognition protein	75.5210131	0.000536599
chr12-54423547	HOXC6	homeobox C5	-65.0768474	0.00059452
chr20-22563181	FOXA2	forkhead box A2	-62.7972483	0.000957054
chr20-22562585	FOXA2	forkhead box A2	-61.3931252	0.001049545
chr3-72939150	GXYLT2	glycosyltransferase 8 domain containing 4	65.3541771	0.001169168
chr20-34147309	FER1L4	fer-1-like 4 (C. elegans)	58.3370116	0.001579846
chr12-54428644	HOXC5	homeobox C5	-71.9161826	0.001873619
chr2-95941784	PROM2	prominin 2	55.5342957	0.001873619
chr12-54428892	HOXC4	homeobox C4	-54.9707617	0.00190397
chr17-46683018	HOXB7	homeobox B7	65.8624452	0.002011668
chr4-6578275	MAN2B2	mannosidase, alpha, class 2B, member 2	50.7396690	0.002340171
chr12-54423460	HOXC6	homeobox C6	-60.5860321	0.003735155
chr17-46671931	HOXB6	homeobox B6	55.5863964	0.004275325
chr2-234744532	HJURP	holliday junction recognition protein	36.8117832	0.004425678
chr2-232259527	B3GNT7	UDP-GlcNAc:betaGal beta-1,3-N-acetylglucosaminyltransferase 7	-58.8405311	0.004575139
chr17-46674011	HOXB6	homeobox B6	61.5163110	0.00460977
chr12-54428601	HOXC5	homeobox C5	-62.2630015	0.005523152
chr16-89688808	DPEP1	dipeptidase 1 (renal)	-66.8219711	0.005523152
chr20-22562467	FOXA2	forkhead box A2	-63.0298815	0.010785706
chr12-54428668	HOXC5	homeobox C5	-60.969772	0.013669362
chr20-34147353	FER1L4	fer-1-like 4 (C. elegans)	52.260453	0.014286322
chr10-86005070	RGR	retinal G protein coupled receptor	65.313282	0.014380736
chr17-46654203	HOXB4	homeobox B4	49.255953	0.014380736
chr15-41708401	RTF1	Rtf1, Paf1/RNA polymerase II complex component, homolog (S. cerevisiae)	48.316353	0.015961603
chr19-46274874	DMPK	dystrophia myotonica-protein kinase	59.4463585	0.017703077
chr20-22563028	FOXA2	forkhead box A2	-49.9576549	0.017722042
chr11-126301918	KIRREL3	kin of IRRE like 3 (Drosophila)	-47.4455129	0.018426331
chr17-46683607	HOXB7	homeobox B7	54.1100473	0.026995683
chr20-61867695	BIRC7	baculoviral IAP repeat-containing 7	38.2431118	0.028259131
chr4-184828474	STOX2	storkhead box 2	-46.00453	0.03148721
chr6-111983139	FYN	FYN oncogene related to SRC, FGR, YES	20.9468069	0.03487875
chr12-5604367	NTF3	neurotrophin 3	-49.2315213	0.036133471
chr12-54424932	HOXC5	homeobox C5	-58.4822544	0.036861493
chrX-152710726	TREX2	three prime repair exonuclease 2; HAUS augmin-like complex, subunit 7	51.7780679	0.037115174
chr2-20648535	RHOB	ras homolog gene family, member B	39.8321014	0.039985426
chr17-46673605	HOXB6	homeobox B6	59.3137178	0.040288154
chr17-46670094	HOXB5	homeobox B5	53.006026	0.041816188
chr12-54425323	HOXC5	homeobox C5	-57.0819331	0.05446436
chr17-46672209	HOXB6	homeobox B6	59.4703343	0.055281525
chr4-6578378	MAN2B2	mannosidase, alpha, class 2B, member 2	54.0144389	0.055439078
chr17-46669859	HOXB5	homeobox B5	48.9420761	0.057845718
chr20-22562795	FOXA2	forkhead box A2	-55.262905	0.058748934
chr17-46672379	HOXB6	homeobox B6	37.9220212	0.059719627
chr17-46669811	HOXB5	homeobox B5	56.5792304	0.062007652
chr16-774464	CCDC78	coiled-coil domain containing 78	50.2311308	0.067168718
chr17-46651823	HOXB4	homeobox B4	54.0130147	0.070165337
chr12-54408697	HOXC6	homeobox C6	-44.2333111	0.072588767
chr12-53496682	SOAT2	sterol O-acyltransferase 2	-37.0650460	0.072799241
chr8-67835854	SNORD87	small nucleolar RNA, C/D box 87	43.5438862	0.076412069
chr4-13542884	NKX3-2	NK3 homeobox 2	57.5579881	0.088515052
chr20-22562765	FOXA2	forkhead box A2	-61.071834	0.088843008
chr19-2716283	DIRAS1	DIRAS family, GTP-binding RAS-like 1	33.1088621	0.093550807
chr2-1637565	PXDN	peroxidasin homolog (Drosophila)	51.3279908	0.102884331
chr6-168417047	KIF25	kinesin family member 25	36.4742119	0.102884331
chr1-9099850	SLC2A5	solute carrier family 2 (facilitated glucose/fructose transporter), member 5	-55.1186093	0.104285117
chr10-86005238	RGR	retinal G protein coupled receptor	42.1907947	0.104842784
chr22-41075645	MCHR1	melanin-concentrating hormone receptor 1	-40.1809805	0.108161774
chr3-128199762	GATA2	GATA binding protein 2	48.3977178	0.110171566
chr7-45146901	SNORA5B	small nucleolar RNA, H/ACA box 5C	36.5600523	0.110303096
chr12-54447039	HOXC4	homeobox C4	-50.1145258	0.111036454
chr11-33277701	HIPK3	homeodomain interacting protein kinase 3	53.68370846	0.11209042
chr5-92924000	NR2F1	nuclear receptor subfamily 2, group F, member 1	21.18993145	0.115121578
chr20-61471911	TCFL5	transcription factor-like 5 (basic helix-loop-helix)	-44.30269566	0.116800473
chr17-46669720	HOXB5	homeobox B5	22.19809798	0.118981641
chr12-54423616	HOXC6	homeobox C6	-53.3502713	0.12386958
chr20-61471704	TCFL5	transcription factor-like 5 (basic helix-loop-helix)	-49.7622302	0.128802549
chr12-54447386	HOXC4	homeobox C4	-38.0307095	0.134526374
chr17-46654113	HOXB4	homeobox B4	44.0713268	0.138722789
chr6-101847583	GRIK2	glutamate receptor, ionotropic, kainate 2	-54.4371579	0.14095978
chr12-54428396	HOXC5	homeobox C5	-47.9659770	0.142463163
chr17-46651361	HOXB4	homeobox B4	59.0212241	0.148051962
chr19-43858215	CD177	CD177 molecule	-52.8713803	0.151540502
chr5-162931219	MAT2B	methionine adenosyltransferase II	-49.2651751	0.153204861
chr17-79992546	DCXR	dicarbonyl/L-xylulose reductase	52.1129033	0.15397301
chr11-119180123	MCAM	melanoma cell adhesion molecule	48.8176834	0.164940401
chr22-31032999	SLC35E4	solute carrier family 35, member E4	49.9240692	0.165230457
chr17-46670995	HOXB5	homeobox B5	44.9380625	0.168400237
chr13-107141057	EFNB2	ephrin-B2	44.95724607	0.172094399
chr8-7306344	SPAG11B	sperm associated antigen 11A; sperm associated antigen 11B	-29.7742479	0.172388783
chr13-43597174	DNAJC15	DnaJ (Hsp40) homolog, subfamily C, member 15	49.5033384	0.176080002
chr17-46669789	HOXB5	homeobox B5	57.8205074	0.178166434
chr3-160474510	PPM1L	protein phosphatase 1 (formerly 2C)-like	-27.9110318	0.178166434
chr12-117580996	FBXO21	F-box protein 21	47.9564349	0.178915625
chr20-22562989	FOXA2	forkhead box A2	-48.453299	0.179191406
chr20-22562221	FOXA2	forkhead box A2	-43.4140179	0.179490127
chr9-138553850	LCN9	lipocalin 9	-36.5115718	0.185296539
chr17-46671518	HOXB5	homeobox B5	49.4577331	0.186225094
chr20-34148339	FER1L4	fer-1-like 4 (C. elegans)	32.0504781	0.187717624
chr2-95942146	PROM2	prominin 2	42.11374	0.19028988
chr19-35781244	MAG	myelin associated glycoprotein	-42.1237977	0.192763364
chr17-46654269	HOXB4	homeobox B4	39.0674004	0.193710002
chr1-204100320	ETNK2	ethanolamine kinase 2	-36.2391408	0.19718686
chr2-234744728	HJURP	holliday junction recognition protein	32.8961068	0.19718686
chr20-22562511	FOXA2	forkhead box A2	-55.6104561	0.19718686
chr12-117580915	FBXO21	F-box protein 21	21.1321240	0.199050637
chrX-152086803	ZNF185	zinc finger protein 185 (LIM domain)	43.1109175	0.199122705

DMS: Differentially Methylated site; FDR: False Discovery Rate.

Gene ontology (GO) analysis was then applied to protein-coding genes with differential DNA methylation events using the R package Goseq. Following bias correction, 21 GO biologic process (GOBP) categories were significantly enriched [False discovery rate (FDR) <0.2; Table 3]. The enriched GO terms were associated with skeletal system morphogenesis (GO:0048705, BH adjusted P = 9e-03), embryological development and embryological morphogenesis (GO:0048704, BH adjusted P = 0.04). These included a considerable number of genes from the HOXB family (HOXB3, HOXB4, HOXB5, HOXB6, and HOXB7) (Fig. 2B-D), the HOXC family (HOXC4, HOXC5, and HOXC6), as well as genes FOXA2, RARB, TBX3, and SERP1. Other GO terms identified were associated with haematopoietic (GO:0048534, BH adjusted P = 0.06), tissue (GO:0009888, BH adjusted P = 0.09) and organ (GO:0048513, BH adjusted P = 0.05) development as well as development of the immune system (GO:0002520, BH adjusted P = 0.09) (Table 3). Genes that were associated with DMS involved in immune system development included RORA, GLI3, NKX3–2, HOXB3, NFATC1, LMO2, KIRREL3, RUNX1, LRP5, TAL1, PITX2, RHOH, GATA2, CACNB4, SATB1, HOXB4, CARD11, and HOXB7.

Table 3.

. Gene Ontology Biological processes (GOBP) categories of differentially methylated evens at protein-coding genes with FDR <0.2 from a gene set analysis (GSA) using GOseq.

GOBPID	Count	P-Value	FDR	Term
GO:0048705	15	1.09E-06	0.009173675	skeletal system morphogenesis
GO:0001501	22	1.46E-06	0.009173675	skeletal system development
GO:0048568	19	1.28E-05	0.044713028	embryonic organ development
GO:0048598	22	1.43E-05	0.044713028	embryonic morphogenesis
GO:0060216	5	1.81E-05	0.044713028	definitive hemopoieosis
GO:0060348	11	2.27E-05	0.044713028	bone development
GO:0048562	15	2.49E-05	0.044713028	embryonic organ morphogenesis
GO:0048706	10	2.87E-05	0.045099637	embryonic skeletal system development
GO:0048513	61	3.31E-05	0.04625517	organ development
GO:0048704	8	4.50E-05	0.056693684	embryonic skeletal system morphogenesis
GO:0048534	18	5.75E-05	0.065834809	haematopoietic or lymphoid organ development
GO:0002520	18	8.40E-05	0.088121986	immune system development
GO:0061448	13	0.0001003	0.096219373	connective tissue development
GO:0009888	38	0.0001070	0.096219373	tissue development
GO:0009887	28	0.00012346	0.103615701	organ morphogenesis
GO:0051216	11	0.000207756	0.163464742	cartilage development
GO:0007389	17	0.000229614	0.170035874	pattern specification process
GO:0043009	21	0.000256876	0.17171298	chordate embryonic development
GO:0009790	30	0.000259159	0.17171298	embryo development
GO:0009792	21	0.000274174	0.172578713	embryo development ending in birth
GO:0003002	14	0.000299586	0.179594414	regionalization

GO analysis was then performed, removing the overlapping DMS mapping to genes previously identified in proximal and distal epithelia,³⁴ thus identifying the enrichment of pathways, potentially more specific to inflammation. The enriched GO term that was identified was associated with bone morphogenesis (GO: 0060349, BH adjusted P = 0.09). These genes included FGFR3, GL13, BMPR1B, LRP5, AXI2, and RARB.

To further explore the epigenetic context of these DMS, we obtained publically available ChromHMM data specific to transcriptionally active and repressed histone marks (H3K4me3, H3K4me1, H3K36me3, H3K27me3, H3K9me3, and H3K27ac) in normal colonic mucosa (Fig. S4).^40,41 These data were used to calculate enrichment of differential methylation at several genomic states including promoter, enhancer, and transcribed and repressed regions.

The majority of DMS and, indeed, the majority of all CCGGs mapped to regions in quiescent/low activity states (Fig. 3B), as was documented in previous studies (Fig. S4).^41,42 We also tested for enrichment and depletion of DMS, conditional on the CCGG density, within each genomic state (Fig. 3B). Enrichment was identified upstream from active TSS (P = 2.6e-04, Fisher exact test), downstream from active TSS (P = 6e-03, Fisher exact test), at enhancer states (Active enhancer 1: P = 1e-03; Active enhancer 2: P = 4.7e-06; Weak enhancer: P = 4.8e-09; Bivalent enhancer: P = 2.3e-06; Fisher exact test), as well as at transcriptionally repressed polycomb regions (P = 7.8e-15, Fisher exact test) (Fig. 3B). Significant depletion was only identified at a quiescent/low activity state (P = 2.1e-14, Fisher exact test). DMS at the active TSS region of HOXB3, HOXB4, and HOXB5, as well as the ChIP-seq profiles of the chromatin marks assayed in colonic mucosa are shown in fig. 2B-D. These figures illustrate elevated levels of H3K27ac at these regions containing hypermethylated loci, a histone mark associated with active regulatory elements that differentiates active from inactive enhancers and promoters.

Figure 3.

Annotation of CCGG sites to genomic regions and ChromHMM states. The percentage of CCGG sites (gray bars) and differentially methylated sites (DMS) (red bars) annotating to TSS, intragenic and intergenic regions is shown (A). A greater number of DMS mapped to intragenic regions as opposed to TSS (promoter) regions. The total percentage of probes is greater than 100% as several probes were classified as belonging to more than one class of genomic region. The percentage of CCGG sites (gray bars) and DMS (red bars) mapping to 18 active and repressed genomic states in colonic mucosa using the ChromHMM model is shown (B). Using the genomic states defined by the ChromHMM model, significant enrichment of DMS, conditioned on the CCGG density at each genomic state was tested. P-values were determined using a Fisher exact test (*P<0.05, ** P<0.005, *** P<0.0005). Results indicate significant enrichment at enhancers, TSS regions, as well as repressed polycomb regions.

Genome-wide differential expression analysis

mRNA expression was assayed in inflamed and non-inflamed samples using Affymetrix Human Transcriptome Array 2.0 (HTA 2.0) oligonucleotide arrays (see Materials and Methods). Genome-wide gene expression data were generated from purified epithelial cell populations from a subset of the same individuals that DNA methylation patterns were assessed. We report transcriptome analysis comparing inflamed and non-inflamed colonic regions from the same individuals affected by sub-total UC (n = 5 pairs).

We identified partial clustering between inflamed and non-inflamed samples (Fig. 4A) in a hierarchical clustering analysis. However, it is difficult to ascertain distinct global dissimilarity in gene expression patterns. Using the R package limma, we identified 73 transcripts that were differentially expressed between inflamed and non-inflamed IECs, corresponding to 62 unique known protein-coding genes (Fig. 4B, Table 4). A higher proportion of transcripts showed higher expression in inflamed regions (n = 46, 63%) compared with non-inflamed regions (n = 27, 37%). Differential expression at transcripts corresponding to cytokines, chemokines, and immune active soluble factors such as α and β defensins was relatively subtle however, and was not considered to be statistically significant. Relative expression levels of these genes are displayed in Table S4. Two related GOBP categories were significantly enriched (FDR <0.2) for differentially expressed genes. These biologic processes were ethanol oxidation (GO:0006069, BH adjusted P = 0.04) and ethanol metabolic processes (GO:0006067, BH adjusted P = 0.05). These genes included ADH1A, ADH1B, and ADH1C. Other genes that were differentially expressed included PRAC1, PRAC2 and HOXB13 (Fig. 4C-E). These 3 genes are located on chromosome 17q12.3 in relatively close proximity (∼4000 bp). It has previously been documented that these genes are highly expressed in the prostate (normal and cancerous) and distal parts of the colon and rectum in both human and mouse in a normal state.^43–45

Figure 4.

Genome-wide and site-specific gene expression patterns in non-inflamed and inflamed colonic regions. Unsupervised hierarchal clustering of global gene expression profiles (A). Each column represents individual samples [I: Inflamed; NI: Non-inflamed, (n = 5 pairs)] and each row represents individual transcripts. The relative gene expression differences are expressed by a color gradient intensity scale ranging from yellow (low expression) to red (high expression). Volcano plot of differential expression analysis (B): Log fold change (logFC) between inflamed and non-inflamed samples (X-axis), –Log¹⁰ for identified non-significant (black) and significant (red) (Benjamini-Hochberg adjusted P of < 0.2) transcripts (Y-axis). Boxplot with overlaying stripchart representing the relative gene expression differences between inflamed (red) and non-inflamed (blue) samples for genes PRAC1 (C), HOXB13 (D), and PRAC2 (E).

Table 4.

List of significantly (Benjamini-Hochberg adjusted P-value cutoff of 0.2) differentially expressed transcripts between inflamed and non-inflamed samples.

Transcript ID	Gene Symbol	Description	LogFC	FDR
TC17001649.hg.1	PRAC1	prostate cancer susceptibility candidate	3.8149	0.00012235
TC15000405.hg.1	GLDN	gliomedin	1.2632	0.00830502
TC15001837.hg.1	ANPEP	alanyl (membrane) aminopeptidase	-2.2182	0.00830502
TC17001651.hg.1	HOXB13	homeobox B13	1.4778	0.00830502
TC09002921.hg.1	ST6GALNAC6	ST6 (α-N-acetyl-neuraminyl-2,3-β-galactosyl-1,3)-N-acetylgalactosaminide α-2,6-sialyltransferase 6	1.8220	0.01424986
TC13000741.hg.1	KCTD12	potassium channel tetramerisation domain containing 12	2.3843	0.01424986
TC21000989.hg.1	B3GALT5	UDP-Gal:βGlcNAc β 1,3 galactosyltransferase, polypeptide 5	1.5520	0.01514616
TC08001294.hg.1	CPA6	carboxypeptidase A6	1.7881	0.01576238
TC10002938.hg.1	C10orf116	chromosome 10 open reading frame 116 adipogenesis regulatory factor	-0.6264	0.01576238
TC12001901.hg.1	NT5DC3	5-nucleotidase domain containing 3	1.0920	0.02110773
TC21000464.hg.1	C21orf88	chromosome 21 open reading frame 88	2.5530	0.02342417
TC01002752.hg.1	INSL5	insulin-like 5	2.8999	0.02794573
TC04001471.hg.1	PITX2	paired-like homeodomain 2	-1.5796	0.02842635
TC11000211.hg.1	SPON1	spondin 1, extracellular matrix protein	1.4101	0.02842635
TC04001410.hg.1	ADH1B	alcohol dehydrogenase 1B (class I), β polypeptide	-0.8644	0.03071734
TC04001409.hg.1	ADH1A	alcohol dehydrogenase 1A (class I), α polypeptide	-0.9172	0.03323028
TC07002589.hg.1	LINC-PINT	long intergenic non-protein coding RNA,p53 induced transcript	0.5743	0.03323028
TC03003022.hg.1	MYH15	myosin, heavy chain 15	1.3573	0.03857138
TC17002262.hg.1	B4GALNT2	β-1,4-N-acetyl-galactosaminyl transferase 2 (B4GALNT2)	-1.7076	0.03941676
TC04001411.hg.1	ADH1C	alcohol dehydrogenase 1C (class I), gamma polypeptide	-1.0960	0.03993123
TC06001299.hg.1	KIF13A	kinesin family member 13A	0.5311	0.04292623
TC05002796.hg.1	FLJ00157	Homo sapiens mRNA for FLJ00157 protein	-1.8563	0.04372244
TC20000349.hg.1	WFDC2	WAP 4-disulfide core domain 2	1.0036	0.04455290
TC0Y000341.hg.1	***	*** no description***	0.9024	0.05041901
TC17000638.hg.1	B4GALNT2	β-1,4-N-acetyl-galactosaminyl transferase 2	-1.5293	0.05129017
TC01000723.hg.1	ROR1	receptor tyrosine kinase-like orphan receptor 1	0.5806	0.05660819
TC0Y000275.hg.1	***	*** no description***	1.4056	0.05660819
TC01003752.hg.1	NUCKS1	nuclear casein kinase and cyclin-dependent kinase substrate 1	0.6628	0.05878908
TC03003359.hg.1	GXYLT2	glucoside xylosyltransferase 2	-0.6263	0.06076798
TC12001155.hg.1	LPCAT3	lysophosphatidylcholine acyltransferase 3	-0.5652	0.07070229
TC06003630.hg.1	DAAM2	disheveled associated activator of morphogenesis 2	-0.8143	0.07543406
TC05001095.hg.1	PP7080	uncharacterized LOC25845	-1.4284	0.07636208
TC04000168.hg.1	GBA3	glucosidase, β, acid 3	-1.4168	0.07717089
TC01004068.hg.1	MIR3916	microRNA 3916	0.5370	0.08218670
TC06004132.hg.1	MOCS1	molybdenum cofactor synthesis 1	-0.5321	0.08218670
TC10000593.hg.1	CDHR1	cadherin-related family member 1	0.8484	0.08218670
TC12000203.hg.1	PTPRO	protein tyrosine phosphatase, receptor type, O	0.8631	0.08218670
TC03000131.hg.1	RARB	retinoic acid receptor, β	-0.7255	0.08301883
TC06000926.hg.1	RFX6	regulatory factor X, 6	0.5280	0.08301883
TC15002698.hg.1	***	*** no description***	-0.4019	0.08318291
TC20001482.hg.1	LINC00261	long intergenic non-protein coding RNA 261	1.4472	0.08677308
TC01002882.hg.1	GCLM	glutamate-cysteine ligase, modifier subunit	-0.7496	0.09208306
TC07000931.hg.1	TMEM139	transmembrane protein 139	-0.6448	0.09320002
TC02002479.hg.1	GCG	glucagon	2.1546	0.0972245
TC05002797.hg.1	SLC9A3	solute carrier family 9, subfamily A (NHE3, cation proton antiporter 3), member 3	-2.1590	0.09722454
TC16000522.hg.1	CA7	carbonic anhydrase VII	1.4832	0.09722453
TC21000363.hg.1	CLDN8	claudin 8	2.8735	0.09722453
TC06000173.hg.1	HIST1H2AE	histone cluster 1, H2ae	-0.5886	0.09832936
TC10001089.hg.1	NEBL	nebulette	0.5596	0.10402139
TC14001475.hg.1	LINC00341	long intergenic non-protein coding RNA 341	0.5282	0.10402139
TC01005141.hg.1	***	*** no description***	0.9118	0.10531854
TC07003096.hg.1	LHFPL3	lipoma HMGIC fusion partner-like 3	-1.6865	0.10531854
TC12000329.hg.1	ANO6	anoctamin 6	1.6249	0.10531854
TC21000729.hg.1	C21orf88	chromosome 21 open reading frame 88	0.9782	0.10531854
TC04001299.hg.1	CDKL2	cyclin-dependent kinase-like 2 (CDC2-related kinase)	0.6730	0.10721497
TC16000969.hg.1	CACNG3	calcium channel, voltage-dependent, gamma subunit 3	0.5635	0.10608053
TC16002090.hg.1	CHST5	carbohydrate (N-acetylglucosamine 6-O) sulfotransferase 5	1.6146	0.12788495
TC20000726.hg.1	APMAP	adipocyte plasma membrane associated protein	-0.5785	0.12831669
TC05001096.hg.1	SLC9A3	solute carrier family 9, subfamily A (NHE3, cation proton antiporter 3), member 3	-1.7126	0.12952132
TC15000406.hg.1	GLDN	gliomedin	0.5266	0.12952132
TC17001568.hg.1	PYY	peptide YY	1.1306	0.12987431
TC17002257.hg.1	PRAC2	prostate cancer susceptibility candidate 2	0.5566	0.14674671
TC17000317.hg.1	PYY2	peptide YY, 2 (pseudogene)	0.5121	0.15335219
TC04000895.hg.1	***	*** no description***	-0.9907	0.15383132
TC16002091.hg.1	TMEM231	transmembrane protein 231	0.4550	0.15516000
TC04002356.hg.1	***	*** no description***	-0.9819	0.17219695
TC05003353.hg.1	RANBP17	RAN binding protein 17	0.5107	0.17219695
TC16001964.hg.1	TMEM231	transmembrane protein 231	1.1312	0.18186527
TC20000698.hg.1	FOXA2	forkhead box A2	0.5397	0.18186527
TC02001391.hg.1	B3GNT7	UDP-GlcNAc:βGal β-1,3-N-acetylglucosaminyltransferase 7	1.1406	0.18308688
TC07002442.hg.1	CROT	carnitine O-octanoyltransferase	-0.7765	0.18626684
TC09001528.hg.1	RP11–388N2.1	putative novel transcript	0.5826	0.18626684
TC03001525.hg.1	PRICKLE2	prickle homolog 2 (Drosophila)	0.3886	0.19988562

Integrative analysis of DNA methylation and gene expression data of inflamed and non-inflamed colonic regions of intestinal epithelial cells

Differentially methylated genes were significantly enriched among genes that were differentially expressed (P = 3e-04, Fisher exact test). Gene ontology analysis also revealed that gene sets that were significantly differentially expressed or differentially methylated were not assigned to any GO categories associated with immune response or other disease or cancer relevant processes. Integrative analysis was performed using both the methylation data and transcriptome data generated from the same 10 pure epithelial cell samples (n = 5 pairs). Five genes were both differentially expressed and differentially methylated in inflamed and non-inflamed samples (PITX2, ROR1, GXYLT2, RARB, and FOXA2), several of which are related to Wnt signaling or to embryonic, cell or organism development. For 4 out of the 5 differentially expressed and differentially methylated genes (ROR1, GXYLT2, RARB, and FOXA2) DNA methylation and gene expression were significantly inversely correlated (Fig. 5), as expected. For example, we identified hypermethylation (ΔMethylation = 67.4, FDR = 2e-04) and downregulation (logFC = - 0.72, FDR = 0.08) at the active TSS of RARB (Fig. 5A,B). Hypomethylation at multiple CCGG sites (Fig. 5C,D, Table S5) and upregulation (logFC = 0.54, FDR = 0.18) at a bivalent enhancer region of FOXA2 was also identified.

Figure 5.

Integrative analysis of differential DNA methylation and gene expression patterns. DNA methylation patterns, gene expression patterns and spearman correlations for genes RARB (A) and FOXA2 (C) are illustrated. Left panel: bar plot and overlaying strip chart of DNA methylation levels between inflamed (red) and non-inflamed (blue) samples; middle panel: bar plot and overlaying strip chart of relative gene expression levels between inflamed (red) and non-inflamed (blue) samples; right panel: spearman correlation between DNA methylation levels (X-axis) and relative gene expression levels (Y-axis); P: P-value; rho: Spearman correlation value. DMS mapping to the active TSS (red) of RARB (B), bivalent enhancer (tan) and repressed polycomb regions (gray) of FOXA2 (D), as well as the ChIP-seq profiles of the set of 6 histone marks assayed in colonic mucosa is shown. Below the histone marks, DMS (red) and all CCGG sites (green) mapping to the given segment of RARB and FOXA2 are also illustrated.

Discussion

In this study, we aimed to generate an integrative epigenome data set, combining genome-wide DNA methylation data and transcriptome data. Our aim was for this data to illustrate the epigenetic variation induced by chronic inflammation of the colon. The objective was to gain insight into the molecular mechanisms of chronic colitis and potentially into its progression to colitis-associated cancer.

Using genome-wide DNA methylation analysis, 577 DMS mapping to 210 unique protein-coding genes were identified. Significant hypermethylation in the presence of inflammation at promoter regions of genes associated with embryonic development and regionalization was observed, most notably, members of the homeotic HOXB gene family. The enrichment of differential methylation at gene sets associated with haematopoietic, tissue, organ, and immune response development was also identified. However, one unanticipated limitation of this aspect of the study was the choice of comparing distal (inflamed) and proximal (non-inflamed) colonic regions from within the same patient. Our hypothesis was that variation in DNA methylation and gene expression patterns could be a product of inflammation. While this strategy allowed us to reduce the major confounding effect of DNA sequence variability on DNA methylation variation,^46–48 some of the variability may also be attributable to the distinct epigenomic signatures observed in proximal and distal colonic regions, as we have recently reported.³⁴ In fact, 41 of the 125 (32.8%) of the DMS identified in normal epithelia were also differentially methylated in inflamed epithelia (Table S3). However, although there were a high number of intersecting DMS between both studies, loci mapping to RARB were not differentially methylated in normal IECs.³⁴ We can therefore identify genes with UC-associated DNA methylation changes that cannot be attributed to colonic location, loci that could potentially act as biomarkers to distinguish normal from inflamed colon epithelium.

Using genome-wide gene expression data generated from a subset of the same pure epithelial cell isolates, 73 differentially expressed transcripts, corresponding to 62 known unique protein-coding genes, were identified. GO analysis identified the enrichment of differentially expressed genes associated with ethanol oxidation (ADH1A, ADH1B, and ADH1C). Additionally, increased gene expression in the presence of inflammation at HOXB13 was identified, as well as prostate cancer susceptibility candidates PRAC1 and PRAC2. These genes have previously been identified to have higher expression levels in the distal colon in both humans and mouse in a normal state.^43–45 However, very little is known about PRAC1 and PRAC2 in the context of UC. Interestingly, the majority of identified variably methylated or expressed gene sets were not directly or indirectly associated with immune response or other cancer-relevant processes. However, some of the variability may also be attributable to the distinct epigenomic signatures observed in proximal and distal colonic regions.³⁴

IECs have several diverse functions, one of which involves acting as innate immune cells, controlling the interface between a potentially hostile colonic luminal environment and the host.^49–51 Many studies, mostly using cultured intestinal epithelial cell lines, have demonstrated the secretion of cytokines and chemokines by those cells.^10–13 Those results have supported the notion that intestinal epithelial cells promote chronic inflammation in IBD patients. In our DNA methylation and transcriptome analyses, we noted that several gene sets were enriched for processes associated with immune development. However, enrichment was greater for gene sets associated with skeletal system morphogenesis and development (Table 3). Our findings from purified epithelial cells are inconsistent with previous studies in UC that highlighted differential expression^52–55 and differential methylation^19,20,22 of key cytokines and inflammatory mediators (Table S4). However, those studies did not use purified epithelial cells, thus indicating that the cytokines/inflammatory mediators implicated in prior studies are likely to have been the route of confounding DNA methylation noise influenced by cell type heterogeneity; most probable representative of autoimmune cell phenotypes. Therefore, our results suggest that the role of colon epithelial cells in the pathogenesis of UC in vivo through cytokine and chemokine secretion may be less than previously estimated.

In a purified IECs isolate, 4 of the 5 genes that were both differentially expressed and differentially methylated had negative correlation between DNA methylation and expression levels (ROR1, GXYLT2, RARB, and FOXA2). RARB encodes the protein retinoic acid receptor β, a member of the thyroid-steroid hormone receptor family of nuclear transcription regulators. This receptor binds retinoic acid, which mediates cellular signaling in embryonic morphogenesis, cell growth, and differentiation. RARB has previously been identified as a tumor suppressor gene and found to be hypermethylated at promoter regions in several cancer phenotypes including breast cancer,⁵⁶ lung cancer,^57,58 and prostate cancer.^59,60 However, it was first established as a tumor suppressor gene in colon cancer cell lines⁶¹ and, subsequently, in cancerous colonic mucosa.⁶² In this case, we demonstrated promoter hypermethylation and downregulation of RARB in an inflamed state. Although RARB has been previously highlighted as a tumor suppressor gene in colon cancer, this is the first report of RARB potentially being implicated in non-cancerous UC in purified IECs. Therefore, the interplay between RARB promoter methylation and downregulation may play an important role in the link between UC and colon cancer. Further research into epigenomic dysregulation of this gene in neoplastic IECs would be required to determine the potential of this gene as a clinical marker for colitis-associated cancer risk.

FOXA2 is a member of the forkhead box gene family. It acts as a transcriptional activator, essential for effective development of endoderm derived organs and tissue and thus considered a master regulator of that lineage.⁶³ It has been suggested that FOXA2 may play a role in suppressing tumor development; for example, Zhu et al. identified FOXA2 as a tumor suppressor, demonstrating that it is downregulated in gastric cancer cell lines, and that inducing FOXA2 inhibits gastric cancer growth in vivo.⁶⁴ We have identified hypomethylation at multiple CCGG sites and upregulation of FOXA2 in an inflamed state but any functional significance of this dysregulation in carcinogenesis remains speculative without subsequent mechanistic studies. Increased sample size and further technical validation are also required to support the biologic significance of these results.

This epigenomic study has reported distinct patterns of DNA methylation and gene expression in the inflamed epithelial cells of UC. These results have shown that differential methylation occurs at promoter regions of tumor suppressor genes, as well as a wide spectrum of both active and repressed coding regions. The relative paucity of variably methylated loci or differentially expressed transcripts corresponding to cytokines and inflammatory mediators also raises questions over the degree to which epithelial cell contribute to chronic colitis. The lack of differential methylation and differential expression at genes associated with colorectal cancer makes interpretation of biologic significance somewhat unclear. Differential methylation and expression may be a genome-wide effect that's not targeted to specific genes that are implicated already in colon cancer risk; effects may be subtle or act through previously unknown pathways. The clinical implications of these reported findings, and the interplay between genetic and epigenetic signatures in the pathway of colitis-associated carcinogenesis needs to be explored in further studies. However, this integrative epigenomic data set will enhance our understanding of UC pathophysiology, potentially bridging the gap between genetic predisposition to UC related disease and UC pathogenesis.

Materials and Methods

Patient recruitment and collection of biopsies

All patient recruitment and sample collection was performed under human subjects protocol approval from the Galway University Hospitals Research Ethics Committee. Patients enrolled in the study were attending University Hospital Galway undergoing colonoscopy for the evaluation of colitis symptoms. Biopsies were collected from 13 patients having subtotal UC where both the colonoscopic appearance and histological evaluation of biopsies confirmed inflammation of the distal colon and the absence of inflammation in the proximal colon. All patient information, including duration of colitis, medication, smoking history, and extent of colitis were recorded in our questionnaire and stored in an encrypted database on the day of recruitment. Clinical characteristics of the patients are included in Table 1. None of the patients was taking any medication known to alter the DNA methylome (folic acid, sulfasalazine, or valproic acid).

Table 1.

. Clinical characteristics of UC patients (n = 13). Current treatments include 5-aminosalicyclic acid, azathioprine, 6-mercaptopurine, methotrexate, infliximab, and adalimumab.

Characteristic	n
Age
<40	6
40–50	2
50–60	3
>60	2
Sex
Male	6
Female	7
Duration of Colitis (years)
<5	3
5–10	3
10–15	1
15–20	4
>20	2
Severity (UCDAI scoring)
<5	6
5–10	6
>10	1
Smoking History
Non-smoker	9
Ex-smoker	4
Current smoker	0

Isolation of epithelial cells from pinch biopsies

Previous techniques for isolating purified IECs relied on isolation of cells at room temperature or 37°C.^65,66 Modifications to those techniques were developed that allow for epithelia to be obtained at 4°C to limit the detrimental effects to membrane integrity, cellular viability, and molecular degradation that can occur at higher temperatures.^34,67

Colonoscopic pinch biopsies (n = 10) were taken from the proximal and distal areas of the colon from healthy patients and stored in ice-chilled PBS. Biopsies then underwent washing (3x) with 5 ml of ice-chilled PBS and centrifuged at 250 x g for 5 min at 4°C. After the third wash, the ice-chilled PBS was replaced with 25 ml of chelation buffer (1 mM EDTA, 1 mM EGTA, 0.5 M DTT, 55 mM D-Sorbitol, 44 mM Sucrose with distilled water at pH 7.3) and stored for 2 h at 4°C on a rocker. After chelation, the samples were then shaken by hand for 30 sec. The cell suspension consisting mostly of intact colon crypts was transferred to a new centrifuge tube and this step was repeated until no more visible cells were liberated. Finally, the cell suspension was centrifuged at 250 x g for 10 min at 4°C, the supernatant was discarded, and the pellet of cells was resuspended in 2 ml of 0.5% BSA in PBS. The sample was aliquoted (200 μl) for cell staining. The remainder of the cells was centrifuged at 250 x g at 4°C and the resulting cell pellet was used for DNA & RNA extraction. All reagents were purchased from Sigma-Aldrich.

Flow cytometry

Each sample was incubated with 5 μl of blocking IgG goat serum (Sigma-Aldrich) and stored for 20 min at 4°C. Samples were then double stained with an epithelial specific marker [FITC Anti-Human CD326 (EpCAM), Biolegend], an immune cell specific maker (APC Anti-Human CD45, Biolegend) or each marker's isotype control (FITC Mouse IgG2b, κ, APC Mouse IgG, κ, both Biolegend). Samples were then stored for a further 20 min in the dark at 4°C, centrifuged at 320 x g at 4°C, and washed twice with 200 μl of 0.5% BSA in PBS. Disaggregation of crypts into a single cell suspension was achieved through pipetting of the cell isolate. Subsequently, cells were examined by flow cytometry using FACSCanto and data analysis was performed using WinMDI (Version 2.8) software.

HELP-tagging assay library preparation

Extraction of genomic DNA from IECs was performed using the method derived from the Albert Einstein University online resource (http://wasp.einstein.yu.edu/index.php/Protocol). All steps outlined in the protocol were followed exactly. Please refer to Supplementary Materials and Methods for a detailed protocol description with no modifications.

Genomic DNA (500 ng) was digested in a 50 µl reaction (containing 5 µl NEB1, 1 µl HpaII, and water) overnight at 37°C. The digest (2 µl) was run on a 1% agarose gel. TE buffer (450 µl) was added to the rest of the digest as well as 500 µl of saturated phenol: chloroform (1:1) and mixed well. The sample was then spun in a micro centrifuge at top speed for 20 min. The aqueous phase of the sample was then transferred to a new tube and precipitated with 1 µl of Ethachinamate (supplied by Wako-chemicals) and 50 µl of 3 M Sodium Acetate. Isopropyl alcohol (800 µl) was added and the sample was incubated at -20°C for 2 h and then spun at top speed for 20 min. The supernatant was then removed and the pellet of DNA was washed with 70% ethanol. The sample was then air-dried and resuspended in 5 µl of TE buffer. Adapter EcoP15I side (TS_AE adaptor) ligation was performed in a 13 μl reaction containing 2x Quick ligase buffer, 0.5 μl of TS_AE adaptor (0.1 μM), digested DNA and 1 μl of Quick Ligase for 15 min at room temperature. All subsequent protocol steps up to polymerase chain reaction (PCR) amplification were performed using the HELP-tagging library preparation protocol developed by Suzuki et al.⁶⁸ Please refer to Supplementary Materials and Methods for the exact protocol description.

The PCR product was extracted from a 3.5% low molecular weight agarose gel electrophoresis and purified by Mini-Elute gel Extraction Kit (Qiagen). Purified products were analyzed by Bioanalyzer to ensure integrity and purity followed by Illumina sequencing (end library size ∼160 bp). All enzymes used for the HELP-tagging assay were purchased from New England Biosciences unless otherwise stated. All adapters and primers were purchased from the WASP system at the Albert Einstein University. For a full list of adapters and primers used, refer to Table S6.

Processing of sequence data

Paired samples (n = 13) from UC patients met the pre-determined quality control standards and were analyzed. Sequencing was performed on an Illumina HiSeq 2000 at the Epigenomics Shared Facility of Albert Einstein University. For this assay, single-end 36–50 bp sequencing was required. The images generated by the Illumina sequencer were analyzed using Illumina Pipeline Software (version 1.4). Default read length of 36 bp was used for initial data pre-processing. Sequences for which adaptor sequence on the 3’end was found were then isolated. The adaptor sequence was replaced with a ploy(N) sequence of the same length and the Illumina ELAND pipeline was run on these sequences with the sequence length set to 27 bp. Data generated by the ELAND pipeline was used to count the number of aligned sequences overlapping each CCGG site in the hg19 build of the human genome. During the alignment process, a maximum of 2 mismatches in each sequence was accepted. For all non-unique alignments, sequences were assigned a partial count for each alignment location amounting to 1/n, where ‘n’ represented the total number of alignments. The number of sequences associated with each HpaII site was then divided by the total number of sequences (including partial counts) aligning to all HpaII sites in the same sample to normalize the data between experiments.⁶⁸

To measure the level of DNA methylation at each CCGG site, the normalized accumulative proportion (NAP) count for the HpaII digested sample was compared with the reference NAP count for the MspI digest. The DNA methylation angle score was calculated using the arctangent of the ratio of HpaII NAP count and MspI NAP count as described previously.⁶⁹ This allows normalization of HpaII counts in terms of variability of the MspI representation. DNA methylation levels reported here were calculated as one minus the DNA methylation angle score, for ease of interpretability, and range from 0 (no DNA methylation) to 100 (complete DNA methylation).³⁴

DNA methylation analysis

CCGG sites with fewer than 5 MspI reads were excluded from all analyses to improve DNA methylation estimation accuracy. Quantile normalization was then performed using the preprocessCore package in R.⁷⁰ The package limma⁷¹ was used to identify individual differentially methylated sites (DMS). A paired sample comparison was used, applying paired effects as a blocking factor in the design model, to compare inflamed and non-inflamed states within subjects. This linear model also incorporated covariates, including age, duration of colitis, severity of disease and total HpaII counts per sample. A false discovery rate analysis (Benjamini-Hochberg) was used to correct for multiple testing. A false discovery threshold of 0.2 was used to decide statistical significance. Differential methylation is expressed as delta methylation (ΔMethylation). ΔMethylation reflects the geometric mean methylation score of one sample group relative to another, a ΔMethylation value >0 reflects hypermethylation in the inflamed region (distal colon) and a ΔMethylation value <0 reflects hypermethylation in the non-inflamed region (proximal colon).

We tested for enrichment of gene ontology categories among genes for which at least one differentially methylated site was found close (±2 kb) to the transcription start site (TSS) as well as the rest of the gene body. The HELP-tagging assay profiles the DNA methylation status at CCGG sites. However, different genes may be associated with very different numbers of such sites, with genes associated with larger numbers of CCGG sites having a greater chance of being associated with at least one differentially methylated site. This can result in severe bias in gene set analysis.⁷² The R package Goseq was used to correct this bias. The Wallenius approximation was used to calculate the over and under representation of Gene Ontology (GO) categories among differentially methylated genes. A false discovery rate analysis (Benjamini-Hochberg) was used to correct for multiple testing. A false discovery threshold of 0.2 was used to decide statistical significance.

The UCSC table browser⁷³ was used to obtain coordinates of genomic regions including gene bodies, intergenic and intragenic regions. The mammalian expression atlas and enhancer peaks were obtained from the FANTOM consortium.^36,74 The epigenomics roadmap consortium was used to obtain coordinates for genomic regions for ChromHMM states in colonic mucosa.^40,41 Enrichment of differential methylation was conditioned on the CCGG density at each state. Significant enrichment was measured using a fisher exact test. These states, as well as states for other cell lines and tissues can be downloaded from

http://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/core_K27ac/jointModel/final/

Annotation of all CCGG sties and DMS to candidate genomic locations such as gene body, intergenic, intragenic, enhancers, TSS regions, and ChromHMM states was performed using customized python scripts.

Array hybridization: Human Transcriptome Array 2.0 (HTA 2.0)

Extraction of total RNA from IECs was performed using the method derived from the Albert Einstein University online resource (http://waspeinstein.yu.edu/index.php/Protocol). All steps outlined in the protocol were followed exactly. Please refer to supplementary materials and methods for a detailed protocol description with no modifications.

Total RNA sample processing and array hybridization took place at Core Unit Systems Medicine (SysMed) at the University of Wurzburg, Germany. The assessment of quality, integrity and quantity of total RNA, in vitro transcription for linear amplification, fragmentation, and biotin labeling was performed as outlined in the GeneChip WT Plus Reagent Kit user manual (Affymetrix). Preparation of buffers and staining for array hybridization was performed according to the GeneChip Hybridization Wash and Stain kit (Affymetrix) with no modifications. Samples were hybridized for 16 h at 45°C and 60 rpm to GeneChip Human Transcriptome Arrays 2.0; washing and staining was performed with a Fluidics Station FS450 using the fluidics script FS450_0001. Arrays were scanned with a GeneChip Scanner 3000 7G (Affymetrix) and processed by the Affymetrix GeneChip command console software (AGCC).

Transcriptome analysis

Raw signals of the arrays were processed using Affymetrix Power Tools,⁷⁵ applying Robust Multi-array Average (RMA) for background correction, quantile normalization and median polish summarization of probe sets.⁷⁶ Log transformed (log2) relative expression values for each probe were then annotated to each gene transcript. Identification of differentially methylated transcripts was performed using limma.⁷¹ A paired sample comparison was used, applying paired effects as a blocking factor in the design model, to compare inflamed and non-inflamed states within subjects. This model also incorporated covariates, including age, duration of colitis, and severity of disease. A false discovery rate analysis (Benjamini-Hochberg) was used to correct for multiple testing. A false discovery threshold of 0.2 was used to decide statistical significance.

Unsupervised hierarchical clustering of global DNA methylation and gene expression patterns was performed using complete linkage and Euclidean distance. Relationships between DNA methylation and gene expression data were examined using Spearman correlations.

Declarations

Ethical approval

All patient recruitment and sample collection was performed under human subjects protocol approval from the Galway University Hospitals Research Ethics Committee with written informed consent regarding participation. Consent for publication was also obtained under human subjects protocol approval from the Galway University Hospitals Research Ethics Committee.

Availability of supporting data

Methylome and transcriptome data sets generated and analyzed during the current study are available from the corresponding author on reasonable request. Coordinates for genomic regions for ChromHMM states in colonic mucosa can be downloaded from

http://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/core_K27ac/jointModel/final/

The UCSC table browser⁷³ was used to obtain coordinates of genomic regions. The Mammalian expression atlas and enhancer peaks were obtained from the FANTOM consortium.^36,74

Funding Statement

This work was supported by the Irish Research Council and further funding was provided by AbbVie Ireland.

Disclosure of potential conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgments

We thank Coralie Mureau for her assistance during sample preparation. We also thank staff from Einstein's Center of Epigenomics including the Epigenomics Shared Facility and Computational Epigenomics group who provided technical support with next generation sequencing.

Author Contributions

AB performed the experiments and generated and analyzed the data. AB, AG, JG, CS and LE interpreted the data and drafted the manuscript. LE, CS and AG devised the study concept. AG, CS and LE obtained funding and supervised the study. All authors critically reviewed and approved the final version of the manuscript.