Methylation of Polycomb Target Genes in Intestinal Cancer Is Mediated by Inflammation

Analogous to mutations, epigenetic changes are strongly associated with cancer development (1). Aberrant DNA hypermethylation is associated with gene silencing and is often observed in CpG islands. Changes in DNA methylation can also occur in premalignant cells or even in normal tissue, for example, as a function of aging (2–6). Such epigenetic events are regarded as early steps in carcinogenesis.

Recent data suggest that Polycomb (PcG) proteins may play a critical role in tumorigenesis (7–9). PcG proteins are repressors involved in maintaining gene expression patterns during development and differentiation (10–13). Binding of PcG complexes is highly correlated with the repressive chromatin mark, H3K27 trimethylation (H3K27me3), catalyzed by PcG protein complexes (14–16). Recently, several groups reported that aberrant DNA hypermethylation in cancer often is associated with PcG target genes (17–21). However, the mechanisms responsible for PcG target gene methylation in tumorigenesis are not clear.

A strong link between cancer and chronic inflammation has been established (22–24). Inflammatory bowel disease (IBD) correlates with an increased risk for development of colorectal cancer (25). In most IBD animal models, such as mice deficient for transforming growth factor-β1, for T-cell receptor α, and for interleukin-10, carcinogenesis in the gastrointestinal tract follows the chronic inflammation phase, which is induced by aberrant microflora (26–28). The inflammation process is always associated with the production of reactive oxygen species (ROS). Phagocytic WBCs produce ROS for killing invading pathogens. However, ROS can harm an inflamed tissue by damaging proteins, lipids, and DNA. Some of these damages have mutagenic effects and are associated with cancer (29). DNA damage caused by oxidative stress can result in different types of modifications, including cross-link lesions, base and sugar damage, deletions, DNA strand breaks, and halogenation of deoxycytosine (30–32). It has been proposed that 5-halogenated cytosine can be a cause for inappropriate de novo DNA methylation because DNMT1 cannot distinguish methylated from halogenated cytosines in vitro (33, 34). This proposed mechanism provides a possible link between inflammation and cancer through aberrant DNA methylation.

To understand how inflammation may modulate DNA methylation patterns, we have analyzed DNA methylation during chronic inflammation in glutathione peroxide 1 and 2 (Gpx1/2) double knockout (Gpx1/2-KO) mice, which are a mouse IBD model (35–37). These mice lack two antioxidant proteins: Gpx1 and Gpx2. Gpx proteins are responsible for neutralization of ROS and for reduction of hydroperoxides, including H₂O₂. Hydrogen peroxide is the product of reduction of superoxide radicals (O₂·⁻) and is the source for potential cytotoxins such as HOCl and HOBr, which are used in cytosine halogenation reactions. In gastrointestinal epithelium, the ubiquitous Gpx1 and the epithelium-specific Gpx2 are the major H₂O₂-reducing Gpx activities. Mice with homozygous disruption of Gpx1 or Gpx2 are disease-free under normal housing conditions, whereas inactivation of both genes (Gpx1/2-KO) leads to high susceptibility to ileocolitis, which begins around weaning (35, 36). Depending on the genetic background, the Gpx1/2-KO genotype causes different susceptibility to cancer development. B6 Gpx1/2-KO mice have milder ileocolitis and a lower mortality, and only 2.5% of B6 mice develop tumors in the lower gastrointestinal tract (37). B6;129 double knockout (DKO) mice have higher levels of inflammatory markers compared with B6 DKO mice, and tumors are observed in 20% to 25% of the mice housed under non–germ-free conditions. This animal model offers the opportunity to follow epigenetic changes from birth through chronic inflammation to tumor formation.

Animals. The establishment and maintenance of the Gpx1/2-KO mouse colonies has been described previously (36, 37). For healthy controls, we used non-DKO mice, which carry at least one wild-type (WT) Gpx1 or Gpx2 allele and do not develop intestinal inflammation. Gpx1/2-KO and control mice were housed under normal, non–germ-free conditions. Pathology of the tumors was performed as described previously (36, 37).

DNA isolation. Epithelial cells were isolated from the distal 10 to 12 cm of the small intestine (distal ileum, corresponding to the diseased segment in Gpx1/2-KO mice at the height of pathology and distribution of tumors) by the everted sac method as described previously (38) to recover the villus and crypt compartments. DNA was isolated using a phenol-chloroform extraction method.

Methylated CpG island recovery assay–assisted CpG island microarrays. Methylated CpG island recovery assay (MIRA) and protein purification for MIRA were done as described previously (39). DNA was digested with MseI and HhaI and ligated to MseI linkers, and the methylated fragments were enriched by MIRA for hybridization to microarrays as described previously (17, 40). We used mouse CpG island microarrays (Agilent Technologies) with two-dye Cy3/Cy5 labeling. This microarray covers 16,030 CpG islands and contains 97,652 oligonucleotide probes. All microarray hybridizations (except for the age-dependent DNA methylation study) were done by comparing the experimental sample (DKO mice) with the control sample of the same age group. For inflammation-dependent and age-dependent DNA methylation analysis, we analyzed the MIRA-enriched fractions from pooled DNA from three to five mice (on quadruplicate microarrays for each genetic background). In tumor cases, methylation patterns of individual tumors were compared with a mixed DNA sample from five control mice at the age of 8 mo. Labeling and array hybridization were as described (19). The data were analyzed with Axon GenePix v.5.1 software. Statistical analysis was performed with ChIP Analytics v.1.3 (Agilent Technologies) to determine genes affected by DNA methylation. This software includes user-configurable heuristics for binding event identification based on P values and adjacent probes, as well as intensity normalization and error modeling. The default settings were used with Lowess normalization and the requirement of three bound oligonucleotides for a gene to be considered as a methylation-positive candidate. For indication of methylated regions in evaluated genes, we used oligonucleotides with P[Xbar] <10⁻⁴. Because in our experiments only a few genes (17 of 552 genes) were affected by methylation of several CpG islands, we considered these 35 CpG islands as separate genes in our statistical analysis. Because we used mice of both genders, we excluded data for the X chromosome. Microarray data were deposited into the Gene Expression Omnibus database (accession number GSE12315).

Tiling microarrays. We used NimbleGen tiling microarray, MM8 Tiling Set 16. This microarray contains a large part of chromosome 7, from 47,370,227 to 115,300,979. MIRA-enriched DNA fractions from 8-mo-old DKO mouse ileum and 8-mo-old ileum from control mice were compared with input DNA. The labeling of dsDNA, microarray hybridization, and scanning were performed by the NimbleGen Service Group (Reykjavik, Iceland). Data were extracted from scanned images by using NimbleScan 2.3 extraction software. The obtained methylation patterns were compared between DKO and control mice.

Combined bisulfite restriction analysis and bisulfite sequencing. To verify the microarray data, we performed combined bisulfite restriction analysis (COBRA) and bisulfite sequencing. Mouse genomic DNA was bisulfite converted using the Zymo Gold kit (Zymo Research). The bisulfite-treated DNA was amplified with DNA fragment-specific primers available on request. PCR products were digested with BstUI, HpyCH4IV, or TaqIα restriction enzymes (New England Biolabs) according to the COBRA method described by Xiong and Laird (41). The PCR products obtained after bisulfite conversion were cloned into the pGEMTeasy vector (Promega) and 10 individual clones were sequenced.

ChIP and microarrays (ChIP on chip). ChIP from ileum epithelial cells (prepared as described above) was performed as described (42). We used chromatin containing 8 μg DNA with a size range on agarose gels of 300 bp to 1 kb. Chromatin was incubated overnight with 4 μg antibodies against H3K27me3 or H3K9me3 (Upstate). Normal rabbit immunoglobulin G (Santa Cruz Biotechnology) was used as a control. For generation of blunt ends, ChIP and input DNA were incubated with T4 DNA polymerase (New England Biolabs) in the presence of deoxynucleotide triphosphates at 12°C for 20 min. After purification with QIAquick kits (Qiagen), DNA was ligated to a blunt linker (5′-AGCAACTGTGCTATCCGAGGGAT and 5′-ATCCCTCGGA) with T4 ligase (New England Biolabs) at 16°C overnight. Amplification was done by using Taq polymerase (Qiagen) according to the MIRA protocol (17). We performed two different types of ChIP on chip experiments: (a) ChIP versus input for detection of localization of chromatin modifications along chromosomes and (b) ChIP versus ChIP for detection of changes in chromatin modifications during inflammation in the ileum of Gpx1/2-KO mice. ChIP on chip experiments were performed on Agilent Technologies CpG island microarrays with Cy3/Cy5 labeling. For ChIP versus ChIP experiments, statistical analysis was performed with ChIP Analytics v.1.3. The default settings were used with Lowess normalization and the requirement of three bound oligonucleotides for a gene to be considered as an H3K27me3-positive candidate. For indication of bound regions in evaluated genes, we used oligonucleotides with P[Xbar] <10⁻³. Binding of H3K27me3 antibodies to chromatin in ChIP versus input arrays was performed after feature extraction and normalization by Axon GenePix v.5.1. A CpG island was considered as positive for H3K27me3 binding if it contained three or more probes with a binding ratio of H3K27me3/input of >1.9 and a distance between each probe of <1,000 bp. For real-time PCR verification, 2 μL of ChIP DNA and input DNA from three independent mice were amplified with gene-specific primers (sequences available on request). A standard curve method was applied and data were presented relative to input (100%). Hoxa10 promoter primers were used as positive control for binding of H3K27me3; H19 primers were used as a positive control for H3K27me3 and H3K9me3. The Gapdh promoter was used as a negative control. Binding to IgG antibodies was used as background control.

Analysis of gene expression patterns using the UniGene database. To compare gene expression with DNA methylation data, we used the National Center for Biotechnology Information UniGene database, which provides information on tissue-specific expressed sequence tag (EST) counts. Up to May 2008, 85,872 ESTs were available for mouse intestine. A gene was considered as unexpressed if zero EST counts of this gene were detected per 85,872 ESTs. A gene was considered as “low expression” if one or two ESTs were detected. A gene was considered as “expressed” if three or more ESTs were found per 85,872 ESTs.

Real-time reverse transcription-PCR. Tissues from mouse brain cerebrum and ileum were used. RNA isolation was performed using the RNeasy Mini kit according to the manufacturer's protocol (Qiagen). Synthesis of cDNA targets was performed with the iScript cDNA Synthesis kit (Bio-Rad). Using real-time PCR, cDNA was amplified with transcript-specific primers (sequences available on request). For verification of the amount of a particular cDNA in the samples, a standard curve method was applied. Gene transcription was normalized to Gapdh expression.

DNA methylation changes in Gpx1/2-KO B6 mice. For genome-wide analysis of DNA methylation patterns during chronic inflammation, aging, and tumorigenesis, we used the MIRA coupled with a microarray approach (17). This method is based on the enrichment of the genomic DNA fraction containing CpG-methylated DNA using the MBD2b and MBD3L1 protein complex and then hybridization using microarrays. We hybridized 5-methylcytosine–enriched DNA from Gpx1/2-KO and control mice onto Agilent Technologies CpG island microarrays. To examine DNA methylation changes during ileum inflammation, we analyzed tumor-free ileum of Gpx1/2-KO and WT control mice at the ages of 28 days and 8 months. For mice on the B6 background, the microarray statistical analysis revealed that Gpx1/2-KO mice at age 28 days had only seven genes that have an increased level of CpG island methylation relative to control mice (Fig. 1; Supplementary Tables S1 and S2). However, at the age of 8 months, Gpx1/2-KO B6 mice had 249 genes with increased DNA methylation in comparison with 8-month-old B6 control mice (Fig. 1; Supplementary Table S1). The DNA methylation status of microarray gene candidates Robo1, Gpc6, and Gabrg3 was verified by COBRA and by bisulfite sequencing (Fig. 2). Confirming the microarray results, these data show that during inflammation in the DKO mice, DNA methylation is substantially increasing (Fig. 2). COBRA analysis of other genes, Lepr, Parc, Lhx8, Barhl2, Cdo1, Sez6l, Fbn1, and Jam3, also indicated a strong increase in DNA methylation in the inflamed ileum of DKO mice (Fig. 3; Supplementary Fig. S1). As expected, we also saw an increase in DNA methylation of several CpG islands as a function of age (8 months versus 28 days) in control mice. Aberrant DNA methylation during inflammation can occur independently of age-associated DNA methylation (e.g., Barhl2, Cdo1, and Sez6l) or can take place at an increased level in genes that are affected by age-dependent DNA methylation (e.g., Robo1, Gpc6, Gabrg3, Lhx8, Fbn1, and Jam3; Fig. 2; Supplementary Fig. S1). To determine if inactivation of Gpx1 and Gpx2 in DKO mice may affect DNA methylation independently of tissue inflammation, we performed MIRA-assisted microarray analysis with liver DNA obtained from DKO and control mice at age 28 days and 8 months because this organ is not affected by inflammation. The microarrays indicated that DNA methylation in liver is increased in only one gene (Vamp4) at the age of 28 days and in 20 genes (11700023E05Rik, 4933433P14Rik, AI854703, Cdkn1b, Ece1, Flt4, Gpx1 exon2, Hapln4, Hnrpk, Mab21l2, Pcdha4, Ppp1r11, Ptch1, Ran, Sema3b, Shroom1, Sox15, Spag1, Yrdc, and Zfp710) at the age of 8 months compared with control liver (Fig. 1). Therefore, in the inflamed ileum of DKO mice, aberrant DNA methylation was over 10 times more frequent than in liver.

DNA methylation changes in the ileum during inflammation in Gpx1/2-KO B6;129 mice. In addition to the B6 background, inflammation-dependent DNA methylation was analyzed in B6;129 mice (Fig. 1; Supplementary Tables S1 and S2), which have higher levels of inflammation and are more prone to tumorigenesis (37). According to the microarray data, these mice had an increase in DNA methylation in 26 genes at age 28 days and in 273 genes at age 8 months when comparing with control mice (Fig. 1; Supplementary Table S2). As for B6 mice, microarray data for B6;129 mice were confirmed by COBRA and bisulfite sequencing of CpG islands of gene candidates, Zfp329, Pdx1, and Fat4 (Supplementary Fig. S2). We verified several gene candidates (Vax1, Tnfaip8, and Nefm) with the highest P values to prove that these evaluated genes were highly methylated (Supplementary Table S1; Supplementary Fig. S3). All three genes had strongly increased DNA methylation in DKO mice compared with control mice. Similar to B6 mice, aberrant DNA methylation during inflammation in B6;129 mice was inflammation specific in one type of CpG islands, that is, occurring only in DKO mice (e.g., Zfp329, Msx1, Barhl2, Parc, Tnfaip8, Gabrg3, and Itgb4) and also occurred during normal aging—albeit at a reduced level—in other CpG islands (e.g., Fat4, Pdx1, Dbc1, Cdh7, Cyp7b1, Cdh20, Tbx2, Vax1, Nefm, Robo1, and Lepr; Fig. 3; Supplementary Figs. S2 and S3).

To further confirm our CpG island microarray data, we performed NimbleGen tiling array analysis of chromosome 7 from position 47,370,227 to 115,300,979 with DKO and control mouse ileum at age 8 months versus input DNA. According to the CpG island microarrays, chromosome 7, from 47,370,227 to 115,300,979, contains eight genes (Dbx1, Tmem16e, Gabrg3, Nr2f2, A830059I20Rik, Phox2a, Ric3, and Cyp2r1) affected by aberrant DNA methylation during inflammation in DKO mice (Supplementary Table S1). Tiling array data confirmed CpG island microarray results because analysis of peaks of MIRA-enriched DNA showed that all eight genes have intense peaks in DKO mice compared with control mice (Supplementary Fig. S4). We searched for hypomethylation of flanking DNA regions around repetitive elements in inflamed tissues. However, we did not detect any substantial hypomethylation in ileum DNA from DKO mice in comparison with healthy control mice.

Because different backgrounds of DKO mice have different susceptibility to tumorigenesis, we compared inflammation-dependent aberrant DNA methylation in B6 and B6;129 mice (Fig. 3; Supplementary Table S1). We found that both backgrounds had ∼80% (213) similarities of inflammation-dependent DNA methylation patterns (Fig. 3A–C). This observation suggests that genetic background may affect DNA methylation of only a small proportion of all CpG islands. We compared methylation patterns between mice of the same genetic background by using COBRA. We observed that individual mice with the same background are characterized by similar methylation patterns during inflammation (Figs. 2 and 3; Supplementary Figs. S1-S3). These data indicate that inflammation-dependent aberrant methylation is not just a stochastic event but is predisposed to occur in specific genes.

Inflammation-dependent DNA methylation and PcG marking. Because DNA methylation may be directed by PcG binding (18–21), we compared genes affected by aberrant DNA methylation in DKO mice with genes having promoters marked by H3K27me3, Suz12, Eed1, Phc1, or Rnf2 in mouse embryonic stem (ES) cells (11). We considered a gene as a PcG target if this gene is marked by H3K27me3 and/or is bound by at least one of the analyzed PcGs (Suz12, Eed1, Phc1, or Rnf2) in ES cells. We found that ∼70% of genes affected by aberrant DNA methylation in the ileum of DKO mice were PcG targets in ES cells (Table 1, row A). From these PcG targets, 97% were marked by H3K27me3 and 35% were associated with all five PcG marks (Table 1, rows B and C). We observed that PcG targeting is associated with aberrant DNA methylation independently of CpG island localization because ∼73% of genes affected by inflammation-dependent DNA methylation in the promoter region and ∼67% of genes affected by aberrant DNA methylation outside of the promoter were PcG targets (Table 1, rows D and E). These data were very similar for both genetic backgrounds. Because homeobox genes are regulated by PcG, we analyzed inflammation-dependent DNA methylation in homeobox genes. We found that ∼7% of inflammation-dependent DNA methylation takes place at CpG islands associated with homeobox genes (Table 1, row F). Age-dependent DNA methylation affects PcG targets less frequently than inflammation-dependent DNA methylation. Only approximately 49% to 53% of age-dependent DNA methylation events occurred at PcG targets (Table 1, row A). Further analysis revealed that 58% of tumor-associated aberrant DNA methylation occurred at PcG targets. These data suggest that PcG targeting established in early development is critical for aberrant DNA methylation occurring during inflammation of the intestinal tissue of mice.

To determine if the PcG mark H3K27me3 is found in intestinal epithelial tissue of mice, we carried out ChIP on chip experiments with ileum epithelial cells isolated from 8-month-old control and Gpx1/2-KO mice. In total, 146 of the 249 CpG-methylated genes (59%) were positive for the PcG mark H3K27me3 in ileum of control mice (Fig. 4A; Supplementary Table S1). Genome-wide, 171 CpG islands showed a decrease of PcG marking (H3K27me3) and 68 CpG islands showed an increase of this modification in the Gpx1/2-KO mice versus control mice. Only 2 of the 68 CpG islands showing an increase of H3K27me3 were DNA CpG methylated. However, strikingly, 46 of the 171 CpG islands showing a decrease of H3K27me3 in Gpx1/2-KO mice were on the list of 249 genes having increased DNA methylation in the DKO mice (Fig. 4A). To verify the array data with conventional ChIP assays, we conducted real-time PCR assays of several candidate genes that showed decreased H3K27 trimethylation in the DKO mice (Fig. 4). These experiments confirmed reduction of the PcG mark for CpG islands that became CpG methylated in the DKO mice (Gabrg3, Gpc6, Sez6l, Cdh20, Pcdh17, Pcdh7, and Cbln4). Simultaneously, there was a small increase in H3K9 trimethylation at the same gene targets. These results indicate that inflammation leads to a frequent rearrangement of PcG marks. Our results indicate that the repressive chromatin mark H3K27me3 can be removed and replaced by the presumably more permanent repressive mark, DNA CpG methylation, during the intestinal inflammation process in Gpx1/2-KO mice.

Age-dependent DNA methylation in the ileum of Gpx1/2-KO mice. Analogous to inflammation, aging plays a critical role in tumorigenesis and is characterized by CpG island DNA hypermethylation (3). To establish a differential role of age-dependent and inflammation-dependent DNA methylation in tumorigenesis, we analyzed DNA methylation in the ileum during aging. Age-dependent DNA methylation was determined by hybridization of MIRA-enriched DNA from the ileum of control, Gpx-WT mice at age 28 days versus MIRA-enriched DNA from ileum of control mice at the age of 8 months. Experiments were done for both B6 and B6;129 backgrounds (Supplementary Tables S1 and S2). Microarray analysis revealed that the ileum from control mice on the B6 background was more affected by age-dependent DNA methylation compared with ileum from B6;129 mice. In ileum of B6 control mice, we detected 213 genes affected by age-dependent DNA methylation in contrast to B6;129 mice, which were characterized by 124 genes methylated with aging. Analysis of the similarities in age-dependent DNA methylation between B6 and B6;129 mice revealed that both genetic backgrounds have 100 genes in common that were affected by age-dependent DNA methylation (Fig. 3D). During aging, 24 genes were B6;129 background specifically methylated in contrast to 113 genes specifically affected by age-dependent DNA methylation on the B6 background (Fig. 3D). These data indicate that age-dependent DNA methylation is affected by genetic background.

Comparison of age-dependent and inflammation-dependent DNA methylation patterns of DKO B6;129 mouse ileum revealed that 34 genes have accelerated age-dependent DNA methylation in inflamed DKO B6;129 ileum (Fig. 3D). These genes become more methylated from 28 days to 8 months in control tissues and are even more highly methylated in inflamed tissues of DKO mice in comparison with control mice at age 8 months. For DKO B6 mice, we detected 65 genes affected by accelerated age-dependent DNA methylation during inflammation (Fig. 3D). However, age-dependent and inflammation-dependent DNA methylation showed only approximately 20% to 30% overlap. Eighty-eight percent of genes for B6;129 mouse ileum and 74% of genes for B6 mouse ileum affected by inflamed ileum-specific methylation were not affected by aging (Fig. 3D). Seventy-three percent of genes for B6;129 mouse ileum and 70% of genes for B6 ileum affected by age-dependent methylation did not undergo inflammation-dependent DNA methylation. This observation indicates that inflammation-dependent and age-dependent DNA methylations preferentially affect different CpG islands.

DNA methylation changes during tumorigenesis in Gpx1/2-KO mice. To elucidate the effect of inflammation-dependent DNA methylation on tumorigenesis, we performed MIRA-assisted microarray analysis on tumors arising in B6;129 DKO mice versus control tissue from WT mice at the age of 8 months. Microarray data from six independent tumors revealed that tumorigenesis, as expected, is associated with aberrant DNA methylation. Surprisingly, however, the tumor-associated aberrant DNA methylation was less pronounced than inflammation-dependent DNA methylation because inflamed ileum from DKO B6;129 8-month-old mice was characterized by 273 affected genes in contrast to 76, 50, 107, 92, 76, and 52 affected genes in the six individual tumors (Supplementary Tables S1 and S2). Comparing the DNA methylation patterns of the two events, tumorigenesis and inflammation, we observed three types of genes: genes affected by inflammation-specific DNA methylation in DKO mice, genes affected by tumorigenesis-specific DNA methylation, and genes methylated in both inflamed and tumor tissue samples (Fig. 5A). The last group contained ∼60% of all the genes affected by aberrant DNA methylation in DKO tumors. Comparison of methylation patterns of tumors reveled similarities: 9 genes from a total of 209 tumor-methylated genes were affected by DNA methylation in all six tumors (Fig. 5B). However, six of these nine genes were also methylated in inflamed tissues (Supplementary Table S1). Twelve genes were methylated in 5 of 6 tumors and 8 of these 12 were affected by aberrant DNA methylation during inflammation. These observations suggest that inflammation plays a critical role in tumor-associated DNA methylation.

To determine the effect of aging and inflammation on tumorigenesis-associated DNA methylation separately, we compared genes affected by aberrant DNA methylation in aging, inflammation, and tumorigenesis. This study was performed for each individual tumor (Fig. 5D). We observed that tumor-associated aberrant DNA methylation consists of tumor-specific, inflammation-dependent, and age-dependent DNA methylation. However, we found that only ∼2% of genes affected by aberrant DNA methylation in tumors specifically showed age-dependent DNA methylation (these genes were not affected by inflammation-dependent DNA methylation in DKO mice). In contrast, ∼60% of tumor-associated DNA methylation occurred also during inflammation and was not affected by age-dependent DNA methylation. These data were also confirmed when genes affected by tumor-associated DNA methylation in all 6 tumors (209) were plotted together (Fig. 5C). This observation indicates that, in this mouse model, inflammation has a dominant effect relative to aging on aberrant DNA methylation in tumors.

Analysis of expression of genes affected by aberrant DNA methylation at the 5′ end. We analyzed how gene expression status may affect the acquisition of DNA methylation at the 5′ ends of genes. For this approach, we used real-time reverse transcription-PCR (RT-PCR) and the UniGene database, which gives a semiquantitative estimate of tissue-specific gene expression, based on EST counts. Only 1% of methylation-susceptible genes were expressed at substantial levels (>2 of >85,000 ESTs) according to the UniGene database. Inflammation-associated DNA methylation in the ileum of B6 and B6;129 DKO mice at 5′ gene ends was correlated with 86% and 84% of silent or poorly expressed genes (no EST counts), respectively (Table 1, row G). This 5′ methylation was also associated with 13% and 15% of genes with low expression (one or two EST counts) in B6 and B6;129 DKO mice, respectively. These data indicate that inflammation-dependent DNA methylation occurs preferentially in already tissue specifically silenced or poorly expressed genes. We performed real-time RT-PCR with control and DKO B6 mouse tissues (Supplementary Fig. S5). Analysis of the UniGene database revealed that the brain frequently expresses genes affected by inflammation-dependent DNA methylation at the 5′ gene ends at high levels. Therefore, we used gene expression in brain as a positive control. We analyzed gene expression of eight genes, among which four genes (Dbc1, Robo1, Barhl2, and Parc) were unexpressed, three genes (Sez6l, Fbn1, and Jam3) had low expression, and one gene (Cdo1) was expressed according to the UniGene database. For most genes tested, transcription in ileum was strongly reduced in comparison with brain (Supplementary Fig. S5).

Our data reveal that inflammation of the ileum leads to aberrant DNA methylation of several hundred CpG islands in Gpx1/2-KO mice on two genetic backgrounds. Both backgrounds showed ∼80% similarities in aberrant DNA methylation caused by inflammation and the numbers of genes affected by inflammation-dependent DNA methylation were comparable (249 and 271, respectively). However, B6 Gpx1/2-KO mice have milder ileocolitis and an 8- to 10-fold lower tumor incidence than B6;129 Gpx1/2-KO mice. These data suggest that although inflammation is clearly associated with methylation of a large number of CpG islands, an increase in such methylation events per se rarely leads to neoplastic transformation in B6 Gpx1/2-KO mice having a relatively tumor-resistant genetic background. We have analyzed the methylated genes for known roles in human cancer. Eighteen of the promoter-methylated genes in our mouse model of ileum inflammation are methylated in human cancers (Pcdh10, Grm7, Robo1, Sall1, Dbc1, Zik1, Cadm1, Dapk1, Hspa2, Fbn1, Prima1, Htr1b, Sez6l, Thbd, Hhip, Sox17, Nell1, and Ptgis), and four of these genes (Hhip, Sox17, Nell1, and Ptgis) are methylated in human colorectal cancer (43–45).

Our study revealed that ∼60% of cancer-associated DNA methylation events were present in the inflammation-prone tissue before tumor formation. This finding indicates an important role of the inflammation process in aberrant DNA methylation in cancer as previously suggested for human cancers of the gastrointestinal tract (46–48). By comparing the genes affected by inflammation-dependent DNA methylation in the ileum of Gpx1/2-KO mice with genes marked by PcG proteins and H3K27me3 in mouse ES cells (11), we found that 70% of genes affected by inflammation-dependent DNA methylation are PcG targets. In addition, 59% of the CpG-methylated genes showed occupancy with the PcG mark H3K27me3 in the ileum of WT mice. This observation indicates a role of H3K27me3 and the PcG complexes in establishment of aberrant DNA methylation. Our data indicate that during tissue inflammation, the PcG mark may be rearranged. At CpG islands undergoing de novo methylation, H3K27me3 can be lost. As a consequence, DNA methylation is a more permanent silencing mark, which may be difficult to remove during subsequent stages of cell proliferation.

Inflammation-dependent DNA methylation has a much stronger correlation with tumor-associated DNA methylation patterns than age-dependent methylation. Age-dependent and inflammation-dependent DNA methylation occurred mostly in different gene targets, although some targets also showed accelerated age-dependent DNA methylation during inflammation in DKO mice. This argues against a scenario in which inflammation can be simply depicted as accelerated aging at the cellular level manifested, for example, by enhanced cell proliferation. Aberrant DNA methylation caused by aging, inflammation, and cancer may have different epigenetic mechanisms.

We found that aberrant DNA hypermethylation during inflammation at 5′ ends of genes is most frequently associated with tissue specifically silenced or very weakly expressed genes. Promoter DNA methylation is likely a consequence of gene silencing directed by PcG binding and low expression levels of the target gene. The biochemical mechanism of inflammation-associated methylation of PcG gene targets remains to be determined. In one hypothetical model, DNA methyltransferases are associated with PcG complexes in somatic stem cells. The effect of inflammation would be to induce proliferation of the stem cell pool, thus leading to methylation errors at PcG target loci where DNMT proteins are preferentially localized. Alternatively, but not mutually exclusive, chlorination of cytosines at CpG sites as a consequence of oxidative stress and production of HOCl during inflammation may lead to aberrant methylation of chlorocytosine-containing CpG sites by DNMT1 as previously proposed (33). If DNMT proteins were preferentially associated with PcG targets, CpG sites within these loci could become methylated by this mechanism even in nondividing cells. This model may also explain the gradual DNA methylation found around tumors because tumors often contain inflammatory infiltrates, which may cause oxidative stress.

In summary, inflammation leads to aberrant DNA methylation, which is predisposed by binding of PcG proteins and by lack of gene expression. Future investigations should be aimed at further elucidating the biochemical mechanism of PcG-associated aberrant DNA methylation. In addition, it will be important to determine the functional importance of PcG target gene methylation in tumorigenesis with the goal of identifying those specific PcG targets important in tumor suppression.

No potential conflicts of interest were disclosed.

Grant support: NIH grants CA084469 and CA128495 (G.P. Pfeifer) and CA114569 (F-F. Chu).

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.

We thank Tibor Rauch and Zunde Wang for help with the MIRA-assisted microarray method.