DNA Methylation and Transcription Patterns in Intestinal Epithelial Cells From Pediatric Patients With Inflammatory Bowel Diseases Differentiate Disease Subtypes and Associate With Outcome

Abstract

Background & aims: We analyzed DNA methylation patterns and transcriptomes of primary intestinal epithelial cells (IEC) of children newly diagnosed with inflammatory bowel diseases (IBD) to learn more about pathogenesis.

Methods: We obtained mucosal biopsies (N = 236) collected from terminal ileum and ascending and sigmoid colons of children (median age 13 years) newly diagnosed with IBD (43 with Crohn's disease [CD], 23 with ulcerative colitis [UC]), and 30 children without IBD (controls). Patients were recruited and managed at a hospital in the United Kingdom from 2013 through 2016. We also obtained biopsies collected at later stages from a subset of patients. IECs were purified and analyzed for genome-wide DNA methylation patterns and gene expression profiles. Adjacent microbiota were isolated from biopsies and analyzed by 16S gene sequencing. We generated intestinal organoid cultures from a subset of samples and genome-wide DNA methylation analysis was performed.

Results: We found gut segment-specific differences in DNA methylation and transcription profiles of IECs from children with IBD vs controls; some were independent of mucosal inflammation. Changes in gut microbiota between IBD and control groups were not as large and were difficult to assess because of large amounts of intra-individual variation. Only IECs from patients with CD had changes in DNA methylation and transcription patterns in terminal ileum epithelium, compared with controls. Colon epithelium from patients with CD and from patients with ulcerative colitis had distinct changes in DNA methylation and transcription patterns, compared with controls. In IECs from patients with IBD, changes in DNA methylation, compared with controls, were stable over time and were partially retained in ex-vivo organoid cultures. Statistical analyses of epithelial cell profiles allowed us to distinguish children with CD or UC from controls; profiles correlated with disease outcome parameters, such as the requirement for treatment with biologic agents.

Conclusions: We identified specific changes in DNA methylation and transcriptome patterns in IECs from pediatric patients with IBD compared with controls. These data indicate that IECs undergo changes during IBD development and could be involved in pathogenesis. Further analyses of primary IECs from patients with IBD could improve our understanding of the large variations in disease progression and outcomes.

Keywords: Epigenetics; Gut Microbiota; Human Intestinal Organoids; Intestinal Epithelium.

Graphical abstract

Figure 1

Overview of study design and multi-dimensional scaling (MDS) analysis of genome-wide datasets. (A) Outline of study design. (B) MDS plots for each dataset: (i) DNAm based on batch corrected M-values; (ii) r-log normalized RNAseq gene expression counts; (iii) gut microbiota 16S operational taxonomic units normalized counts. Samples are labelled according to diagnosis (CD, Crohn’s disease; UC, ulcerative colitis; control) and gut segment. Schematic in part A adapted from Tauschmann et al.

Figure 2

Contribution of diagnosis and inflammation to variance within each data layer. (A) Bar chart of the explained variance by diagnosis and inflammation across each dataset separated by gut segment. (B–E) Scatterplot of P values derived from differential DNAm (I and I) and gene expression (I and I) in sigmoid colon (SC) samples. For each CpG or gene, P values were generated for the comparison between Crohn’s disease (CD)/ulcerative colitis (UC) and control, and inflammation status (ie, inflamed vs non-inflamed). CpGs and genes with significant P values are plotted in purple for inflammation, in red for diagnosis, and in green if significant for both comparisons. Adjusted P < .01 was considered as significant.

Figure 3

Differential DNAm and gene expression analysis were performed separately for terminal ileum (TI) (A and B) and sigmoid colon (SC) (C and D), taking mucosal inflammation into account. (A and C) Venn diagrams of significant differentially methylated positions (DMPs), differentially expressed genes (DEGs), and regulatory DMRs (rDMRs). (B and D) Example of disease-specific rDMRs displaying DNA methylation levels expressed as Beta value on the y-axis in the left panel separately for TI and SC samples in the upper and lower panel, respectively. Beta value of 0 represents un-methylated, while 1 represents fully methylated CpG site. Genomic location is indicated on the x-axis. The middle panel displays identified rDMR (enlarged). The right panel displays a boxplot of the respective gene expression according to diagnosis. (B) rDMR within the APOA1 identified in TI-derived epithelium of children diagnosed with CD. (D) rDMR within the BACH2 gene identified in colonic IEC.

Figure 4

Pathway enrichment analysis of disease-specific regulatory DMRs (rDMRs). Pathway enrichment analysis was performed on identified rDMRs derived from the 3 comparisons between Crohn’s disease (CD) vs controls in terminal ileum (TI) and sigmoid colon (SC) samples (left and middle panel) and ulcerative colitis (UC) vs controls in SC samples (right panel). Analysis was performed using InnateDB and Reactome database and significant enrichment of individual pathways is displayed as the -log10 (adjusted P value).

Figure 5

Stability of disease-associated intestinal epithelial DNA methylation changes: (A) Correlation plot of DNA methylation (Beta values) of disease-associated differentially methylated positions (DMPs) at diagnosis and at repeat endoscopy for each patient at the 2 time points. Shown are Crohn’s disease (CD)-associated DMPs (left) and ulcerative colitis (UC)-associated DMPs (right) in sigmoid colon (SC) epithelium (adjusted P <. 01). (B) Brightfield microscopic images of fully grown intestinal epithelial organoids derived from 2 gut segments (ie, terminal ileum [TI] and SC) of CD and control patients. (C) Quantile-quantile plot generated from organoid-derived genome-wide DNAm P values. Plotted are P values (observed vs expected) comparing specific CD-associated DMPs (from Figure 3) for each gut segment with randomly selected CpGs. (D) Examples of CD-associated DMPs being retained in patient-derived organoids. Plotted are beta values derived from genome-wide array data generated from purified colonic epithelium and respective organoids. GREB1, Growth Regulation By Estrogen In Breast Cancer 1; TMEM173, Transmembrane Protein 173; PDE1B, Phosphodiesterase 1B; CtrlP, Control purified IEC (n = 14); CDP, CD purified IEC (n = 13); CtrlO, Control organoids (n = 7); CDO, CD organoids (n = 5). (E) Validation of CpGs shown in D. Validation of genome-wide DNAm data using pyrosequencing. n = 5–7 per group; *P < .05; ***P < .001; unpaired, 2-tailed t-test between Ctrl and CD.

Figure 6

Correlation of intestinal epithelial cells (IEC)-specific molecular signatures with diagnosis and clinical outcome measures: IEC-derived epigenetic, transcriptomic, and microbial signatures were tested for their potential to predict diagnostic status (A and B) and correlation with disease outcome parameters (C–F). (Ai) Bar chart indicating area under the curve (AUC) of the best model to accurately differentiate samples based on diagnosis (ie, IBD vs controls). (Aii) ROC curve for the best diagnostic model (inflammatory bowel disease [IBD] vs control) using colonic DNAm data. (Bi) Bar chart of the AUC of the best models to differentiate between Crohn’s disease (CD) and ulcerative colitis (UC). (Bii) Receiver operator characteristic (ROC) curve for the best model separating CD from UC using ileal IEC DNAm data. (C) Weighted Gene Co-Expression Network Analysis (WGCNA) of CD terminal ileum (TI)-derived RNA-Seq data showing correlations between key gene-expression modules and clinical parameters. Each cell on the heatmap displays Pearson correlation coefficient and corresponding P value. Outlined cells indicate significant correlations. (D) Heatmap and hierarchical clustering of patients based on gene expression (ie, RNAseq counts) for strongest module. (Dii and Diii) Kaplan-Meier curves based on patient grouping derived from 7Di, ie, top gene expression module for use of biologics and time to third treatment escalation during 75 weeks of follow-up (n = 10 patients, P = .049 and P = .032, log-rank test). (Ei) Heat-map and hierarchical clustering of CpGs within strongest module identified by applying WCGNA to CD TI DNA methylation profiles. (Eii and Eiii) Kaplan Meier curves based on patient grouping derived from 7Ei for use of biologics (Eii) and time to third treatment escalation (Eiii) during 75 weeks follow-up (n = 29 patients, P = .025 and P = .043, log-rank test). (F) Venn-diagram showing the overlap between annotated genes that were present in the top modules for both gene expression and DNA methylation.