Mismatch Repair Proteins Initiate Epigenetic Alterations during Inflammation-Driven Tumorigenesis

Approximately 25% of all cancers are associated with chronic inflammation (1). Colorectal cancer is a substantial contributor to morbidity and mortality worldwide (2). Although the vast majority of colorectal cancer cases are sporadic, 5%–10% of cases have been attributed to hereditary conditions (3). In addition to family history, several risk factors associated with colorectal cancer have been identified including age, inflammatory bowel disease, bacterial infection, obesity, excessive alcohol consumption, and smoking (4). Most of these risk factors are characterized by chronic inflammation (5). Moreover, nonsteroidal anti-inflammatory drug use was reported to reduce the risk of colorectal cancer, reinforcing the notion that inflammation is a central contributor to colorectal cancer development (6).

In addition to inflammation, genetic mutations and epigenetic alterations play a key role in driving colorectal cancer (7, 8). Several genetic mutations that contribute to the initiation of colorectal cancer have been identified and well characterized (9). Epigenetic alterations also contribute to colorectal cancer initiation and progression. DNA methylation is the most extensively studied epigenetic alteration in cancer and cancer cells exhibit a global loss of DNA methylation that is thought to lead to genomic instability (8). In addition, there is concurrent hypermethylation at distinct regions, often within promoter CpG islands of tumor suppressor genes (TSG; ref. 8). Such hypermethylation can lead to TSG silencing and contribute to cancer. While the fundamental importance of cancer-specific DNA methylation alterations is clear, how they are initiated is not well understood.

Our group previously demonstrated that treatment of cancer cells with the reactive oxygen species (ROS) hydrogen peroxide (H₂O₂) causes localization of an epigenetic silencing complex containing DNA methyltransferases DNMT1 and DNMT3B, and the chromatin silencers Sirtuin-1 (SIRT1) and Enhancer of Zeste-2 (EZH2) to sites of damaged chromatin. Notably, there was enrichment of these silencing proteins at the promoter CpG islands of TSGs, resulting in reduced gene expression (10). On the basis of these findings, we hypothesized that inflammation would induce similar events in vivo.

In the current study, we uncover a novel mechanism responsible for initiating cancer-relevant epigenetic alterations in a murine model of inflammation-induced tumorigenesis. Our results demonstrate that epigenetic alterations that occur during inflammation-driven tumorigenesis are distinct from those that occur during inflammation-independent tumorigenesis. To our knowledge, we are the first to demonstrate the involvement of the mismatch repair (MMR) pathway in initiating epigenetic alterations during inflammation-induced tumorigenesis. As inflammation and epigenetic alterations are critical in the development of many cancer types and diseases, a deeper understanding of the interplay between these two factors, as gained from this study, is broadly relevant.

C57BL/6J (Jackson Laboratories) and Min^{ApcΔ716+/−} mice were handled and inoculated with Enterotoxigenic Bacteroides fragilis (ETBF) as in Wu and colleagues (2009; ref. 11). Msh2^l/lVC are a result of crossing B6.Cg-MSH2^tm2.1Rak/J and B6.Cg-Tg(Vil1-cre)997Gum/J mice (Jackson Laboratories) to create mice homozygous for MSH2^tm2.1Rak and expressing the Vil1-cre transgene. Littermates not expressing Vil1-cre were used as WT controls. Msh2^l/lVC/Min mice are the result of crossing Msh2^l/lVC and Min^{ApcΔ716+/−} mice. For all experiments, both males and females were used, mice were randomized between mock and ETBF groups, and mice of different genotypes were cohoused. Individual tumors were removed from dissected colons with the aid of a dissecting microscope and stored at −80°C until further analysis. Distal (0–2 cm measured from the rectum) and proximal (feathered portion adjacent to cecum) epithelium was collected by scraping the mucosal surface of the dissected colon (after removal of any tumors), washed three times in PBS, and then subjected to the indicated protocol. Such scraping has been shown by others to be an effective method to obtain samples of intestinal epithelial cells (12). All mouse experiments were covered under protocol number 16-027, which was approved by the Indiana University Bloomington Animal Care and Use Committee in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care International.

DNA was isolated from tumors or epithelium using the QIAamp DNA Mini Kit (Qiagen) following the manufacturer's protocol. To identify differentially methylated regions (DMR), methyl CpG binding domain (MBD) enrichment was performed from DNA from epithelium collected from different mice (n = 5 WT/Min mock, n = 3 WT/Min ETBF, n = 3 Msh2^l/lVC/Min mock) or individual tumors (n = 3 WT/Min mock, n = 7 WT/Min ETBF, n = 3 Msh2^l/lVC/Min mock, n = 3 Msh2^l/lVC/Min ETBF) using Diagenode's MethylCap kit. Libraries were prepared following Bioo Scientific's Methyl Sequencing kit. Single-end 75-bp sequencing was performed using an Illumina Nextseq. Z-scores were calculated using a 500-bp fixed sized bin spanning CpG islands based on the distribution of coverage from uniquely mapped reads. Z-ratios were derived from the comparison of z-scores for the different sample types for the 500-bp regions. See Supplementary Methods for more detail.

DNA was bisulfite treated (EZ DNA methylation-Gold kit, Zymo Research) and used for pyrosequencing. Pyrosequencing and quantitative methylation specific PCR (qMSP) assays were first tested using a standard curve of bisulfite-treated mixtures of unmethylated and methylated DNA (data not shown). Only methylated DNA assays with little to no amplification of unmethylated samples, close to 100% efficiency and R² close to 1 were used further. See Supplementary Table S1 for assays and primers used.

RNA was prepared from epithelium or tumors using TRIzol followed by cleanup with a RNeasy Kit (Qiagen). cDNA was prepared and qPCR was done using TaqMan assays (see Supplementary Table S1 for assays used). Expression of candidate genes was normalized to expression of a housekeeping gene (PPIA or 18S).

Tight chromatin fractionation of washed epithelium was performed as described in Ding and colleagues (13). Blots presented are representative of three independent experiments.

Chromatin immunoprecipitation (ChIP) was performed using anti-H3K27me3 (Diagenode) or anti-EZH2 (D2C9, Cell Signaling Technology) and the iDeal CHIP-Seq kit for histones and transcription factors (Diagenode) according to the manufacturer's instructions.

Coimmunoprecipitations (co-IP) were performed from nuclear extracts that were treated with oligoamines to release chromatin-bound proteins as described in Ding and colleagues (13). Antibodies used were rabbit IgG (Millipore) and anti-EZH2 (5246, Cell Signaling Technology). Blots presented are representative of two independent experiments.

Microsatellite instability (MSI) analysis was performed as in Woerner and colleagues (14). MSI was determined by comparing DNA fragment analysis of mononucleotide markers in tumors relative to tail DNA from the same mouse.

DNA was isolated from stool using the ZR Fecal DNA micro kit (Zymo Research). Libraries were made from 16S V1-V3 PCR amplicons from the stool DNA using the NEXTflex 16S V1-V3 kit (Bio Scientific). Paired-end sequencing (300 bp) of pooled libraries was performed on a MiSeq. Read quality filtering was performed using mothur. AbundantOTU+ was used for clustering of sequences to operational taxonomic units (OTU) and further classification. MetagenomeSeq was used to determine differential abundance across samples using normalized OTU read counts. OTUs with significant (5% FDR) differential abundance between any of the sample groups were used to create the heatmap (65 OTUs).

Total CpG methylation in DNA from epithelium and tumor samples was determined using an ELISA-based assay [MethylFlash Global DNA Methylation (5-mC) – Epigenetek].

Expression data, densitometry, qMSP, and local ChIP are presented as the mean ± SEM. These data are evaluated by one-tailed t test and considered statistically significant with a P < 0.05. Sample sizes are indicated in associated figure legends. MBD-seq statistical analysis is detailed in the Supplementary Methods.

MBD and 16S sequencing data have been deposited at the Sequence Read Archive (SRA) data repository (project accession number SRP105286).

ETBF is a clinically relevant strain of B. fragilis (15). When Multiple intestinal neoplasia (Min) mice, which are heterozygous for mutant adenomatous polyposis coli (Apc^Δ716), are infected with ETBF, tumors rapidly form at the site of inflammation, in the distal colon (11). Studies have demonstrated that inflammation-induced tumors have unique DNA methylation alterations compared with uninflamed tissue; therefore, we evaluated DNA methylation in our model (16, 17). Min mice spontaneously develop rare colon tumors (18) allowing us to uniquely compare DNA methylation alterations of inflammation-induced ETBF tumors to methylation alterations of background Min (mock) tumors. Interestingly, relative to mock epithelium, the ETBF tumors compared with mock tumors had more hypermethylated DMRs (203 compared with 6) but fewer hypomethylated DMRs (194 compared with 700; Table 1; Fig. 1A). Of the methylation gains that occurred in mock tumors, 5 overlap with the ETBF tumors (Supplementary Table S2).

As Naumov and colleagues (19) demonstrated in human tumors, regions that gained methylation in ETBF tumors had low DNA methylation levels in mock epithelium (mean z-score of 0.4), whereas regions that lost methylation in ETBF or mock tumors had relatively high DNA methylation levels in mock epithelium (mean z-score 3.4 or 3.6, respectively; Fig. 1B). Both gains and losses occurred more often in CpG islands in exons or promoters of genes than introns, 3′ UTRs, or intragenic regions (Supplementary Fig. S1A). As expected, regions targeted for DNA hypermethylation were enriched for regions that are bivalent (containing both H3K4me3 and H3K27me3) in mouse embryonic stem cells (Supplementary Fig. S1B; refs. 20–24).

As chronic inflammation is known to induce DNA methylation changes in inflamed epithelium (16), we also assayed methylation changes in the distal colon epithelium of ETBF-infected mice (ETBF epithelium). ETBF epithelium had fewer hypermethylated DMRs in comparison with the ETBF tumors, but more than the mock epithelium and mock tumors (Table 1). Eleven of the 40 hypermethylated DMRs observed in the ETBF epithelium overlapped with those in the ETBF tumors (Supplementary Table S2). When samples were clustered on the basis of regions with DNA hypermethylation in tumors relative to mock epithelium, the ETBF epithelium clustered with the mock epithelium (Fig. 1A). Interestingly, ETBF tumors clustered separately from the mock tumors and epithelium, suggesting that they have distinct methylation gains from the other samples.

Gene ontology (GO) analysis revealed that the genes associated with DMRs in the mock and ETBF tumors are involved in multiple biological processes, many of which are associated with aspects of development, differentiation, and regulation of transcription (Supplementary Table S3). Interestingly, several of the methylation gains observed in the ETBF tumors occurred within CpG islands of genes with known tumor-suppressive function, such as Hoxa5, Polg, Runx1, Runx3, CD37, Stx11, Tceb2, Lgr6, Cdx1, Fut4 (Fig. 1C; Supplementary Table S2; refs. 25–34). CpG island hypermethylation at 18 of 21 candidate genes in ETBF tumors versus mock epithelium were validated by pyrosequencing or qMSP (Fig. 1D and E; Supplementary Fig. S1C). No DNA methylation changes were detected in the promoter CpG island of Gapdh, a negative control (Supplementary Fig. S1C). Eighteen candidate genes are associated with human digestive organ tumors by Ingenuity Pathway Analysis, demonstrating their relevance to human disease (Supplementary Fig. S1D). Runx3, Gata2, and Hoxa5 are also known to undergo aberrant DNA hypermethylation in human colorectal cancer (35, 36). qMSP analysis at candidate genes demonstrated an intermediate level of methylation in the ETBF epithelium, between that observed in the mock epithelium and ETBF tumor (Fig. 1E) consistent with the data in Fig. 1A.

As hypothesized, many genes containing hypermethylated promoter CpG islands in ETBF tumors also had reduced mRNA expression (Fig. 1F). In addition, several candidate genes had lower expression in ETBF epithelium compared with ETBF tumors, but higher expression compared with mock epithelium (Fig. 1F), consistent with the methylation results in Fig. 1E. Genes that contained hypermethylated CpG islands in nonpromoter exons had increased gene expression as has been demonstrated previously (Supplementary Fig. S1E – Cldn4, Spi1, Stx11; ref. 37). Overall, these findings indicate that ETBF tumors contain distinct cancer-specific DNA hypermethylation alterations that distinguish them from mock and ETBF epithelium and mock tumors.

We hypothesized that ETBF-induced oxidative DNA damage may initiate epigenetic changes in the distal colon that ultimately result in the DNA hypermethylation observed in the inflammation-induced tumors. ETBF inoculation caused oxidative DNA damage (increased 8-oxoguanine) in the distal colon epithelium 2 days post-ETBF, which returned to background levels 7 days post-ETBF (Fig. 2A). We confirmed that the oxidative DNA damage was occurring in epithelial cells by assaying 8-oxoguanine in cells positive for the epithelial marker, EPCAM (Fig. 2B).

The increased oxidative DNA damage observed in the distal colon epithelium 2 days post-ETBF corresponded to a decrease in the mRNA expression of several candidate genes that were found to be hypermethylated in the ETBF tumors with expression levels returning back to normal 7 days post-ETBF (Fig. 2C). To address whether these changes occurred specifically in epithelial cells, we assessed the purity of our colon epithelial samples. Our samples contained approximately 84% EPCAM-positive cells and fewer than 2% CD45-positive cells (Supplementary Fig. S2A), irrespective of ETBF treatment status. To probe this further, we also cultured organoids from the distal colon epithelium following an established protocol (38), treated the epithelial organoids with H₂O₂ to mimic inflammation/oxidative damage in vitro, and examined expression of several candidate genes. Indeed, the candidate genes examined (Cdx1, Fut4, Hoxa5, Polg, Runx1) had reduced expression in response to oxidative stress (but not housekeeping genes Rpl0, Tbp), analogous to our in vivo results (Supplementary Fig. S2B). This result suggests that the changes observed in the colon epithelium in response to ETBF are likely occurring in epithelial cells and not an alternative cell type present in the mucosal surface.

As EZH2 has been implicated in transcriptional repression at sites of oxidative DNA damage, we hypothesized that this protein would be enriched in the promoters of candidate genes that have reduced expression in the inflamed epithelium. Indeed, EZH2 was significantly enriched at several candidate genes in ETBF epithelium compared with mock epithelium 2 days post-ETBF (Fig. 2D). These regions also had enrichment of the repressive chromatin mark catalyzed by EZH2, trimethyl H3K27 (H3K27me3; Fig. 2E). These findings indicate that ETBF initiates early epigenetic and transcriptional alterations in several candidate genes from the MBD-seq dataset, and these changes are temporally and spatially associated with the oxidative DNA damage induced by ETBF.

The MSH2-MSH6 heterodimer facilitates repair of clustered oxidative DNA lesions in a PCNA-dependent and S-phase–independent manner (39). In accordance with this finding, our group observed in vitro that MSH2 and MSH6 become more tightly bound to chromatin in response to oxidative damage and are indispensable for the recruitment of epigenetic silencing proteins to damaged chromatin (13). Therefore, we hypothesized that MSH2 and MSH6 contribute to early epigenetic alterations that occur in response to ETBF.

First, we demonstrated that EZH2 coimmunoprecipitated with DNMT1, MSH2, MSH6, and PCNA in the distal colon epithelium 2 days after ETBF, suggesting an interaction between epigenetic proteins and MMR proteins in response to ETBF (Fig. 3A). This interaction did not occur in mock epithelium. SUZ12 and EED are known EZH2-interacting partners and are positive controls for the EZH2 co-IP.

To evaluate whether the MMR pathway is involved in ETBF-mediated epigenetic alterations, we examined mice that lack expression of Msh2 in intestinal epithelial cells. In Msh2^l/lVC mice, the Msh2 gene is flanked by LoxP sequences and Cre recombinase is driven by a Villin promoter, leading to deletion of Msh2 in intestinal epithelial cells (40). We verified that MSH2 protein is present in colon crypt cells in mock and ETBF inoculated wild-type (WT) mice, but not in crypts from Msh2^l/lVC mice (Fig. 3B). In distal colon epithelium from WT ETBF-infected mice, epigenetic silencing proteins (EZH2, DNMT1, and SIRT1) and MMR proteins (MSH2, MSH6 and PCNA) have a higher affinity for chromatin than in mock mice (Fig. 3C). This enhanced affinity occurred in the distal colon and not in the proximal colon that lacks ETBF-mediated inflammation. Loss of Msh2 reduced the ETBF-mediated increase in binding of EZH2, DNMT1, and SIRT1 to chromatin (Fig. 3C), highlighting the necessity of the MMR pathway in the response of epigenetic proteins to ETBF-induced damage. LAMB (Lamin-B), a loading control, was consistent across the samples. Levels of phosphorylated H2AX (γH2AX), a marker of DNA damage, were elevated in ETBF samples relative to mock samples. Interestingly, γH2AX levels were consistently higher in mock epithelium from Msh2^l/lVC than WT mice, corresponding to a relative increase in binding of EZH2, SIRT1, and PCNA to chromatin. Moreover, we examined the effect of loss of Msh2 on mRNA expression of several candidate genes and their expression was unaffected by ETBF in Msh2^l/lVC mice (Fig. 3D; Supplementary Fig. S3A).

To rule out the possibility that loss of Msh2 alters the initial immune response to ETBF we examined the expression of several cytokines in response to ETBF 2 and 7 days postinfection and found that their expression was unaltered by loss of Msh2 (Supplementary Fig. S3B and S3C). Cell proliferation measured by Ki-67 staining and phosphorylation of STAT-3 was also unaffected by loss of Msh2 (Supplementary Fig. S3D and S3E). ETBF-infected Msh2^l/lVC mice also have similar numbers of γH2AX foci per crypt compared with ETBF-infected WT mice (Fig. 3E). Collectively, these findings implicate a role for MSH2 in initiating early epigenetic and transcriptional alterations observed in the inflamed colon epithelium of mice 2 days post-ETBF.

It is possible that an alteration of the microbiota composition caused by loss of Msh2 could explain reduced ETBF-induced epigenetic alterations in Msh2^l/lVC mice. Therefore, we examined whether loss of Msh2 alters the initial or ETBF-induced intestinal microbiota composition. 16S ribosomal RNA gene sequencing was performed on stool DNA and sequences were clustered into OTUs. Principal component analysis (PCA) and hierarchical clustering analysis based on the OTUs that had significantly differential abundance among the sample types revealed a high degree of similarity between the bacterial populations present in stool from mock-infected mice, regardless of genotype (Fig. 4A and B). None of the OTUs were significantly different between the mock groups, suggesting that loss of Msh2 does not alter the background microbiota composition (Supplementary Table S4). There was separation between samples from ETBF-infected and mock-infected mice, irrespective of genotype (Fig. 4A and B; Supplementary Table S4). PCA revealed a small degree of separation between ETBF-infected WT/Min versus ETBF-infected Msh2^l/lVC/Min samples, suggesting that loss of Msh2 might impact the ETBF-induced microbiota (Fig. 4A and B; Supplementary Table S4). However, loss of Msh2 correlates with a significant increased abundance of ETBF (Fig. 4C) and colony formation units of ETBF per gram of stool (Supplementary Fig. S4), ruling out the possibility that loss of Msh2 reduces ETBF-mediated epigenetic alterations by reducing the abundance of ETBF. Altogether, loss of Msh2 did not alter the background microbiota of these mice, although it does alter the relative abundance of some bacterial populations upon ETBF infection.

Although loss of Msh2 reduced early ETBF-induced epigenetic alterations (Fig. 3), sporadic and germline mutations in the MMR pathway are commonly implicated in colorectal cancer pathogenesis (4). Therefore, we examined the effect of Msh2 deletion from intestinal epithelial cells on tumorigenesis in our model. Mock Msh2^l/lVC/Min mice were similar to uninfected WT/Min mice in terms of tumor burden (Fig. 4D). Interestingly, ETBF-infected Msh2^l/lVC/Min mice developed significantly more tumors in the distal colon than the WT/Min mice infected with ETBF, but the same regions had the highest number of tumors in both sets of mice (Fig. 4D).

The Th17/STAT-3 immune response drives ETBF-mediated tumorigenesis in Min mice (11). Msh2 deletion did not alter ETBF-induced phosphorylation of STAT-3 in ETBF-induced tumors (Fig. 4E). Furthermore, inactivation of APC leads to constitutive activation of the WNT/β-Catenin signaling pathway in tumors in Min mice. Loss of Msh2 did not alter the magnitude of increase in levels of β-Catenin or PCNA, a marker of proliferation, in ETBF-induced tumors (Fig. 4E).

Mutations in MMR genes are known to cause genomic instability by altering the length of repetitive DNA sequences known as microsatellites (4). Interestingly, the mock and ETBF Msh2^l/lVC/Min tumors tested had microsatellite instability (MSI), whereas most tumors from the WT/Min mice were microsatellite stable (MSS; 9/11; Table 2). These results are consistent with the observation in humans that mutated MSH2 can lead to colon tumorigenesis (4) and suggest that ETBF-induced tumors from Msh2-deficient mice are more genetically unstable than ETBF-induced tumors from Msh2-sufficient mice.

We initially demonstrated that Min ETBF tumors have a unique hypermethylation signature (Fig. 1; Table 1). As the Msh2^l/lVC/Min mice still developed tumors in response to ETBF, this model provided us with a unique tool to answer the question: is MSH2 required for ETBF-induced tumor-specific DNA hypermethylation alterations during tumorigenesis? Therefore, we examined the effect of Msh2 deletion on methylation in ETBF tumors using MBD-seq. Importantly, loss of Msh2 dramatically reduced ETBF-mediated hypermethylation alterations with the Msh2^l/lVC/Min ETBF tumors having only 11 DNA hypermethylated regions relative to Msh2^l/lVC/Min mock epithelium (Fig. 5A and B; Supplementary Table S2). When samples are clustered using regions with DNA hypermethylation in tumors relative to the respective mock epithelium, the two mock epitheliums cluster together and the Msh2^l/lVC/Min ETBF tumors cluster with the Min mock tumors. The WT/Min ETBF tumors fall on a distinct arm from the other samples, suggesting that their DNA hypermethylation alterations are distinct from all other tumors types, including the ETBF-induced tumors from Msh2^l/lVC/Min mice. Plots of the DNA methylation data for Hoxa5 and Polg demonstrate the reduction of promoter CpG island hypermethylation in Msh2^l/lVC/Min ETBF tumors compared with WT/Min ETBF tumors (Fig. 5C). This observation was validated at several candidate genes with tumor-suppressive function including Hoxa5, Polg, Runx3, and Stx11 (Supplementary Fig. S5A and S5B). Both the mock and ETBF Msh2^l/lVC/Min tumors had many hypomethylated DMRs relative to Msh2^l/lVC/Min mock epithelium (Fig. 5B; Supplementary Table S2), suggesting that the losses of methylation are general to the tumorigenesis process in these mice.

Furthermore, we examined the effect of loss of Msh2 on mRNA expression of several candidate genes in tumors that formed at sites of ETBF-mediated inflammation. Interestingly, loss of Msh2 partially restored expression of candidate genes in tumors that formed at sites of ETBF-mediated inflammation, including Cdx1, Fut4, Hoxa5, Polg, Runx1, and Runx3 (Fig. 5D).

As mentioned, it has been widely reported that there is a global loss of DNA methylation in cancer. Therefore, we examined the effect of Msh2 deficiency on global CpG DNA methylation. Consistent with what is observed in humans, the WT/Min ETBF tumors exhibited a global loss of DNA CpG methylation. Interestingly, loss of Msh2 restored total CpG methylation back to mock epithelium levels (Fig. 5D). Thus, not only did loss of Msh2 reduce regional hypermethylation alterations genome-wide, it also reduced ETBF-mediated global DNA hypomethylation. These results reinforce the notion that MSH2 is an essential player in both the recruitment of epigenetic proteins to chromatin during the early stages of inflammation and in the permanent silencing of several TSGs in tumors that form at sites of ETBF-mediated inflammation and oxidative DNA damage.

Many studies have demonstrated a strong relationship between inflammation and epigenetic alterations (1). Although the mechanism connecting the two is not well understood, oxidative damage is a prominent feature associated with both. On the basis of previous work by our group, we hypothesized that sustained and/or repeated oxidative damage to chromatin, during inflammation, may result in epigenetic silencing (O'Hagan, 2011). To test this hypothesis, we used an established mouse model of inflammation-induced colon tumorigenesis. The bacterium ETBF has been shown to rapidly induce acute colitis and tumorigenesis in Min mice that resembles the pathology observed in humans with colorectal cancer (11, 41). Importantly, several studies revealed a strong association between ETBF colonization in the gut and colorectal cancer incidence in humans, suggesting that this mouse model is highly relevant to human colon carcinogenesis and an appropriate model to use to investigate the mechanism underlying how inflammation initiates DNA methylation alterations (42, 43).

Whether ETBF-mediated inflammation can induce DNA methylation alterations in humans remains to be determined; however, colitis, which can be induced by ETBF, is strongly associated with aberrant DNA methylation alterations in humans (44). Furthermore, others have demonstrated an association between bacterial infection and DNA methylation alterations in both animals and humans. A well-characterized example is induction of aberrant DNA methylation alterations and gastric cancer in Mongolian gerbils due to exposure to the bacterium Helicobacter pylori (45). A human study reported that the gastric mucosa of H. pylori infected individuals had higher methylation levels at 8 marker CpG islands compared with uninfected individuals (46). The results presented in our study are highly consistent with these findings in animals and humans in that compared wih mock colon epithelium, ETBF-induced tumors have altered DNA methylation, including hypermethylation of promoter CpG islands of several TSGs. Furthermore, the epithelium surrounding the tumors in ETBF-infected mice had more DNA hypermethylation alterations than the mock epithelium, but still considerably fewer hypermethylation changes compared with ETBF tumors, confirming an inflammation-produced field effect that has previously been demonstrated (1). Interestingly, the DNA methylation pattern observed in ETBF-induced tumors was remarkably similar to what is observed in human cancers, including colorectal cancer, namely global hypomethylation concomitant with focal hypermethylation. As has been shown in human colorectal cancer, the DNA hypermethylation changes observed in our study occurred at regions that are bivalent in embryonic stem cells and genes associated with DMRs were enriched in processes associated with development and differentiation.

Min mice spontaneously develop occasional tumors in the colon. Conveniently, this allowed us to compare DNA methylation changes in inflammation-induced tumors to changes in noninflammation-induced tumors. There were very few hypermethylation changes in the mock tumors compared with the ETBF tumors. A critical point here is that the vast majority of DNA hypermethylation alterations observed in our study in the ETBF-induced tumors are specifically driven by ETBF or ETBF-mediated inflammation and this is highly consistent with what has been reported in similar animal models and in humans.

Niwa and colleagues ruled out the possibility that H. Pylori itself is responsible for the induction of DNA methylation alterations in the gastric mucosa of Mongolian gerbils, and confirmed that innate immune-mediated inflammation resulting from H. pylori is the culprit (45). However, the molecular mechanism that underlies how inflammation initiates epigenetic alterations in cancer is not well understood. Our hypothesis in this study is that innate immune-mediated inflammation, caused by ETBF, induces oxidative DNA damage and thereby triggers early transient silencing of genes, rendering them susceptible to stable silencing through DNA methylation. Importantly, at the time point when ETBF-infection induces robust oxidative DNA damage, we see increased enrichment of EZH2 at promoters of genes that become methylated in tumors that form at sites of inflammation, coinciding with reduced expression. We demonstrate that Msh2 deletion in intestinal epithelial cells abrogates the ETBF-mediated recruitment of epigenetic proteins to chromatin and restores expression of candidate genes. Knowing that we now had a system where we could block the recruitment of epigenetic proteins to sites of oxidative DNA damage without affecting the inflammatory response or DNA damage levels, we could determine whether DNA hypermethylation in the inflammation-induced tumors depends on this early epigenetic response to oxidative damage. Importantly, Msh2 deletion reduced ETBF-mediated genome-wide hypermethylation alterations and reversed ETBF-induced global hypomethylation in tumors that formed at sites of inflammation.

Intriguingly, loss of Msh2 led to increased tumorigenesis in response to ETBF, despite a lack of DNA hypermethylation and reexpression of TSGs. This result suggests that another mechanism, independent of aberrant epigenetic alterations, can drive tumorigenesis in the context of ETBF infection. We speculate that increased genomic defects, due to MSI, underlie tumorigenesis in Msh2^l/lVC/Min mice exposed to ETBF. To our advantage, the fact that Msh2^l/lVC/Min mice develop tumors in response to ETBF allowed us to elucidate a mechanism to explain how epigenetic alterations occur during tumorigenesis, a task that could not have been achieved if tumors did not form in these mice. Whether or not DNA hypermethylation changes are necessary to drive inflammation-induced tumorigenesis in the context of a normal mutational burden needs to be studied further.

It remains to be determined precisely how the early ETBF-induced epigenetic alterations observed in our model are maintained and converted into permanent epigenetic alterations in tumors that form at sites of inflammation. We speculate that, in addition to the epithelial cells, these alterations occur and persist in the intestinal stem cells, which are known to eventually go on to transform into tumor cells (47). This hypothesis is supported by the observation that some epigenetic alterations that occur early in stem cells lead to permanent silencing of TSGs, thereby contributing to predisposing the cell to malignant transformation (48).

The work described here provides a mechanism to explain how inflammation, in this case inflammation mediated by ETBF, a pathogen associated with colorectal cancer in humans, causes early epigenetic alterations, some of which persist in tumors that form at sites of inflammation. A challenge in elucidating the mechanism by which inflammation initiates epigenetic alterations, is that it is difficult to selectively modulate responses caused by inflammation without also modulating the initial inflammatory response altogether. We were able to prevent inflammation-induced hypermethylation in tumors without modulating the initial inflammatory response by using mice lacking Msh2. Our findings suggest a role for MSH2 in initiating early ETBF-mediated epigenetic alterations and maintaining DNA methylation alterations that occur in the ETBF-induced tumors. A recent study demonstrated that MSI tumors from patients with Lynch syndrome, a disease in which patients harbor genetic mutations in MMR genes and eventually develop colorectal cancer, have less DNA hypomethylation and fewer regional hypermethylation alterations (49), analogous to what we observed in our study.

While in sporadic colorectal cancer, loss of expression of the MMR protein MLH1 is associated with CpG Island Methylator Phenotype (CIMP), we demonstrate here that loss of MSH2 or MSH6 plays an opposite role. This finding is supported by colorectal cancer TCGA data where, of the hypermutated tumors studied, those with MLH1 methylation were CIMP, but those with mutations in MSH2/6/3 and WT MLH1 expression were not CIMP (9).

The results presented herein fill in critical gaps in our understanding of the relationship between inflammation and epigenetic alterations by elucidating a role for inflammation-induced DNA damage and consequent MMR in initiating cancer-specific epigenetic alterations during tumorigenesis. Knowledge gained from this study has notably increased our understanding concerning how epigenetic alterations are initiated during tumorigenesis in the colon and this information can likely be extrapolated to other types of cancer. Interestingly, regions with DNA hypermethylation in the inflammation-induced tumors were surprisingly consistent between tumors from different mice and different experimental cohorts, suggesting a strong selective pressure for these silencing events. We hypothesize that the mechanism of DNA hypermethylation is similar in other inflammatory-driven epithelial cancers, but the specific genes silenced may differ depending on the selective pressures of tumorigenesis in other tissues. Understanding the mechanism of initiation of DNA methylation changes will allow the scientific community to therapeutically target inflammation-induced epigenetic changes and potentially reduce the cancer risk of individuals with chronic inflammatory diseases.

No potential conflicts of interest were disclosed.

Conception and design: A.R. Maiuri, C.E. DeStefano Shields, H.M. O'Hagan

Development of methodology: A.R. Maiuri, C.E. DeStefano Shields, C.L. Sears, H.M. O'Hagan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Peng, S. Sriramkumar, C.M. Kamplain, C.E. DeStefano Shields, H.M. O'Hagan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.R. Maiuri, R. Podicheti, S. Sriramkumar, D.B. Rusch, C.E. DeStefano Shields, H.M. O'Hagan

Writing, review, and/or revision of the manuscript: A.R. Maiuri, C.E. DeStefano Shields, C.L. Sears, H.M. O'Hagan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.M. Kamplain, H.M. O'Hagan

Study supervision: H.M. O'Hagan

We thank the Indiana University Center for Genomics and Bioinformatics and the Indiana Molecular Biology Institute for their assistance. We also thank Sue Childress for her assistance with tissue processing.

This work was supported by grants from the NIEHS (R01ES023183 to H.M. O'Hagan), NCI (R01CA151325 to C.L. Sears), the Department of Medicine, and the Bloomberg Kimmel Immunotherapy Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.