An understanding of early genetic/epigenetic changes in colorectal cancer would aid in diagnosis and prognosis. To identify these changes in human preneoplastic tissue, we first studied our mouse model in which Mthfr+/− BALB/c mice fed folate-deficient diets develop intestinal tumors in contrast to Mthfr+/+ BALB/c mice fed control diets. Transcriptome profiling was performed in normal intestine from mice with low or high tumor susceptibility. We identified 12 upregulated and 51 downregulated genes in tumor-prone mice. Affected pathways included retinoid acid synthesis, lipid and glucose metabolism, apoptosis and inflammation. We compared murine candidates from this microarray analysis, and murine candidates from an earlier strain-based comparison, with a set of human genes that we had identified in previous methylome profiling of normal human colonic mucosa, from colorectal cancer patients and controls. From the extensive list of human methylome candidates, our approach uncovered five orthologous genes that had shown changes in murine expression profiles (PDK4, SPRR1A, SPRR2A, NR1H4, and PYCARD). The human orthologs were assayed by bisulfite-pyrosequencing for methylation at 14 CpGs. All CpGs exhibited significant methylation differences in normal mucosa between colorectal cancer patients and controls; expression differences for these genes were also observed. PYCARD and NR1H4 methylation differences showed promise as markers for presence of polyps in controls. We conclude that common pathways are disturbed in preneoplastic intestine in our animal model and morphologically normal mucosa of patients with colorectal cancer, and present an initial version of a DNA methylation-based signature for human preneoplastic colon. Cancer Prev Res; 6(11); 1171–81. ©2013 AACR.

Nearly one million people worldwide develop colorectal cancer every year (1). Colorectal cancer results from a combination of environmental and genetic factors that convert normal epithelium into a malignant tumor through multiple stages. An understanding of early events in tumorigenesis will lead to timely diagnoses and improved outcomes.

Epigenetic changes are early events in colorectal cancer and other neoplasias. There are numerous genes with methylation differences between colorectal tumors and adjacent tissues (2–4). However, there are limited data on differential methylation in normal colonic mucosa between controls and patients with colorectal cancer (5, 6 and our own recent report (7)). To identify protumorigenic changes in normal human colonic mucosa, we began with the identification of candidate genes in our mouse model that had previously been shown to develop intestinal tumors after administration of low folate diets (8). Since folates generate methyl groups for DNA methylation, we predicted that there would be genetic/epigenetic changes in preneoplastic intestine in our mouse model and that some of these changes might be similar to those seen in human colonic mucosa. We therefore compared murine candidates to the unrestricted list of preliminary human genes that had shown changes following methylation profiling of normal colonic mucosa in colorectal cancer patients and controls (7 and unpublished data).

Our mouse model reflects some of the genetic and nutritional factors that affect risk for human colorectal cancer. Individuals with low folate intake are more susceptible to colorectal cancer than individuals with adequate intake (9). A polymorphism in methylenetetrahydrofolate reductase (MTHFR), which generates methyl groups for S-adenosylmethionine–dependent methylation reactions, also modulates colorectal cancer risk. The association between the MTHFR 677C→T polymorphism and risk for colorectal cancer has been found in several epidemiologic studies and the proposed mechanisms involve epigenetic remodeling through DNA methylation changes (9–11).

The impact of folate and MTHFR deficiency on tumorigenesis is apparent in our mouse model for spontaneous intestinal neoplasia in BALB/c mice. BALB/c and C57BL/6 mice were fed folate-deficient (FD) or control diets (CD) for one year. Tumors were only observed in BALB/c mice fed FD (8, 12); a single functional copy of the Mthfr gene increased the number of FD mice with tumors (8). Several tumor-predisposing candidate genes involved in cell-cycle control, cell survival, and DNA repair were identified by comparing expression profiles of tumor tissue with normal tissue (13). These observations, in combination with increased DNA damage (14, 15), decreased expression of tumor suppressors and increased retinoid/PPARA pathway activity in BALB/c normal preneoplastic intestine (15), could explain the susceptibility of these mice to intestinal tumorigenesis. We observed differential expression of Bcmo1, Aldh1a1, and Sprr2a when comparing CD- and FD-fed BALB/c mice. Expression differences were also seen for Bcmo1 between Mthfr+/+ and Mthfr+/− mice. These changes were consistent with the hypothesis that enhanced RXR/PPAR activity would increase oxidative stress/damage and lead to neoplasia (15).

Our unique mouse model, without germline mutation or carcinogen induction, provides an opportunity to study early events in intestinal neoplasia. In this report, to identify specific candidate genes in tumorigenesis, we used microarrays to compare BALB/c Mthfr+/+ CD mice and BALB/c Mthfr+/− FD mice, which have relatively lower and higher intestinal tumor susceptibility, respectively (15). We found significant differences in retinoid/PPARA pathway genes between BALB/c mice fed FD and CD. The activation of this pathway is consistent with the findings from our previous report on gene expression profiling between tumor-susceptible BALB/c and tumor-resistant C57BL/6 mice (15).

We compared our murine candidates, obtained from both of the aforementioned expression microarrays, to the extensive list of candidate human genes identified through methylation profiling (7 and unpublished data). Our inter-species comparison led to the identification of 5 human genes that showed significant pyrosequencing-based methylation differences, in 14 CpG dinucleotides, as well as significant expression differences, in normal human colonic mucosa between patients with colorectal cancer and controls. Our results suggest that common tumorigenic mechanisms, reflecting an altered metabolic state, are shared by our mouse model and human colorectal cancer. Furthermore, these methylation differences contribute to an epigenetics signature for diagnosis of colonic neoplasia.

Mice and diets

Animal experimentation was approved by the Montreal Children's Hospital Animal Care Committee, in accordance with Canadian Council on Animal Care guidelines. After weaning, BALB/c Mthfr+/+ and Mthfr+/− mice were fed CD (2 mg folate/kg diet) or FD diets (0.3 mg folate/kg diet) for one year. Incidence of neoplasia was consistent with our previous reports (8, 15).

Control subjects

Research was approved by the Temple University Office for Human Subjects Protections Institutional Review Board, protocol 11910. We collected biologic specimens from subjects undergoing routine screening colonoscopy at Temple University Medical Center to serve as the control arm of the study (Supplementary Table S1). We excluded subjects with a personal or first-degree family history of any cancer and subjects with a previous colonoscopic finding of polyps. Subjects who were not excluded underwent complete colonoscopic evaluation by a board certified gastroenterologist. If the colonoscope could not be passed to the appendiceal orifice, the subject was excluded. If the complete colon was visualized, two cold forceps biopsies of normal colonic mucosa from the ascending colon (proximal to the hepatic flexure) were pooled.

Cancer patients

We collected biologic specimens from patients undergoing colon resection for presumed or biopsy-proven colon cancers. Patients were considered eligible if they had no personal or family history of colon cancer before this encounter. Patients with known or clinical features of hereditary cancer syndromes (specifically, hereditary nonpolyposis colorectal cancer, or familial adenomatous polyposis syndrome) were excluded. Patients with any personal history of chemotherapy or radiotherapy were also excluded. Patients who remained eligible (Supplementary Table S2) underwent colon resection at a single National Cancer Institute designated Comprehensive Cancer Center (Fox Chase Cancer Center/Temple University, Philadelphia, PA). Specimens, determined by a board certified pathologist to be normal appearing colon mucosa, were obtained well away from the lesion (∼10 cm).

DNA and RNA isolation from human normal tissue

Morphologically normal colon mucosa specimens were obtained from colorectal cancer patients or from controls undergoing screening colonoscopy (7). Samples were treated as described (7).

RNA extraction from murine normal preneoplastic intestine

RNA was extracted as described (15). Eight samples for microarrays were prepared from 4 BALB/c Mthfr+/− mice fed FD and 4 BALB/c Mthfr+/+ mice fed CD. High quality of RNA was verified (Supplementary Fig. S1). In addition, 16 RNA samples were extracted from BALB/c Mthfr+/− and BALB/c Mthfr+/+ mice fed CD and FD (4 mice per group). These biologic replicates were used to confirm microarray results by quantitative real-time RT-PCR (qRT-PCR) and verify effects of genotype and diet on expression.

Microarray analysis and quantitative real-time RT-PCR

Microarray experiments were performed using Affymetrix Mouse Gene 1.0 ST Array Chips, as previously described (15). We considered BALB/c Mthfr+/+ mice fed CD as the group with higher tumor resistance and BALB/c Mthfr+/− FD group as the tumor-susceptible group. Genes with expression fold changes more than 1.4 and a P value less than 0.01 after false discovery rate correction were considered significant. Ingenuity Pathways Analysis (IPA) was used to assess biologic processes with the most significant changes.

qRT-PCR was performed as described to confirm microarray data (15). Primers were designed (Supplementary Table S3) and amplified fragments of expected sizes (data not shown). We used 16 mice in 4 groups (4 mice per group); the groups were Mthfr+/+ CD, Mthfr+/+ FD, Mthfr+/− CD, and Mthfr+/− FD.

RNA extraction, cDNA synthesis, and gene-specific TaqMan probes (Applied Biosystems) were done as described (7), to measure steady-state levels of PDK4, SPRR1A, SPRR2A, NR1H4, and PYCARD in normal colon mucosa from patients with cancer and controls. Primers and probes are described in Supplementary Table S4.

Quantitative CpG methylation analysis by pyrosequencing

We used bisulfite pyrosequencing to measure methylation of specific CpGs in the 5′ region of human PDK4, SPRR2A, NR1H4, SPRR1A, and PYCARD. CpGs in mouse orthologous regions were also assessed, as described (15). Genes and relevant oligonucleotides are presented in Supplementary Table S5. Representative pyrograms are shown in Supplementary Fig. S2A–S2F.

Statistical analysis

Quantitative data are presented as average value of replicates ± SEM. Levene test was performed to assess equality of variance. Two-factor ANOVA was used to evaluate effects of diet and genotype on gene expression; strain and diet were also compared in some cases. Student t test for independent samples was performed for specific comparisons where indicated. Analyses were done using SPSS for WINDOWS software, version 11.0. P values < 0.05 were considered significant.

Differences between BALB/c Mthfr+/− FD and Mthfr+/+ CD gene expression profiles

Our microarray results have been deposited in the Gene Expression Omnibus database (GEO, ref. 16, GEO accession no. GSE34011). There were 63 genes with significant expression changes (51 increased and 12 decreased; Supplementary Fig. S3A) in FD Mthfr+/− BALB/c mice compared with CD Mthfr+/+ BALB/c mice (Supplementary Table S6).

These 63 genes were grouped based on functions using IPA (Supplementary Fig. S3B). The top 3 categories were lipid metabolism, small-molecule biochemistry, and nucleic acid metabolism. Fatty acid metabolism, LPS/IL-1–mediated inhibition of RXR function, and PXR/RXR activation were identified as pathways with the most significant changes (data not shown).

Evidence for involvement of PPARA in tumorigenesis

Our previous study compared gene expression between the C57BL/6 and BALB/c mouse strains with different sensitivity to intestinal tumorigenesis. In that report, we showed that the PPARA-oxidation pathway plays a critical role (15). The present work, based on diet and genotype comparisons in BALB/c, confirms the involvement of PPARA. Our gene expression profiling identified several genes related to PPARA activation and oxidative stress that are affected by diet and/or Mthfr genotype (Supplementary Table S7), as well as PPARA-responsive genes (Supplementary Table S8). Responsiveness to PPARG was also reported for some of these genes (references listed in Supplementary Table S8).

Confirmed expression changes for eight genes that may promote tumorigenesis in Mthfr+/−FD mice

Expression of eight genes that may influence tumorigenesis was confirmed by qRT-PCR (Supplementary Fig. S4). These genes are involved in regulation of proliferation (Atf3; ref. 17), apoptosis (Plscr1 and Plscr2; ref. 18), cell survival (Ppme; ref. 19–21), recognition of aberrant unmethylated DNA (Pdctrem; ref. 22), overexpression or chromosomal rearrangement related to cancer (Lhfpl2; refs. 23, 24), or reduction of retinaldehyde levels (Rdh18 and Akr1c13; ref. 25).

Five additional murine genes and their human orthologs show changes in expression or methylation in mice and in normal intestinal mucosa of colorectal cancer patients

We hypothesized that some protumorigenic mechanisms in our mice might be shared by human preneoplastic colon. To pinpoint the involved genes, we first looked at genes identified in the aforementioned murine microarray (affected by diet or Mthfr genotype, Supplementary Table S6), that would match human orthologs with demonstrated methylation changes in our recent genome-wide profiling of DNA methylation of normal colonic mucosa in patients with colorectal cancer and controls (ref. 7 and C. Sapienza, 2012, unpublished data). We limited our selection to human genes with methylation changes more than 2% and for which increased/decreased methylation could correspond to decreased/increased expression in murine mucosa. We also focused on genes that were related to the PPAR/oxidative stress pathway. Only three genes passed this stringent screen: PDK4, SPRR2A, and NR1H4. We applied the same filters to the list of mouse genes deduced from our previously published strain comparison (15); this scan generated only two additional candidates: SPRR1A and PYCARD. Microarray-based methylation changes in PYCARD had been published (7); the other four genes had not been reported.

Confirmation of microarray data for these five murine genes was undertaken. Pdk4, a target of PPARA, is a positive regulator of glycolysis (26). It is upregulated by FD in mice with both Mthfr genotypes (Fig. 1A). Expression of Pdk4 is also higher for Mthfr+/− mice than Mthfr+/+ mice, for both diets. We previously reported that Sprr2a is downregulated in BALB/c compared with C57BL/6 (15). This ROS scavenger plays a role in tumorigenic events related to oxidative stress. FD lowered Sprr2a in both genotypes (Fig. 1B). Nr1h4, also known as Fxr, upregulates cell growth and induces PPARA (27). FD Mthfr+/− mice demonstrated higher expression of Fxr than wild-type mice. Folate deficiency also stimulated its expression in Mthfr+/− mice (Fig. 1C). Sprr1a has a similar role to Sprr2a and showed decreased levels in BALB/c mice compared with C57BL/6 (Fig. 1D). Surprisingly, a marked elevation by FD was seen in C57BL/6 (Fig. 1D), whereas this was not observed for Sprr2a (Fig. 1B). PYCARD downregulation is well documented in colorectal cancer (28). Pycard showed lower expression in BALB/c compared with C57BL/6, in both diets (Fig. 1E).

The human orthologs of the five mouse genes presented in Fig. 1 showed altered methylation levels in our genome-wide methylation array (7). We confirmed these changes by bisulfite pyrosequencing-based assays for 6 CpGs in the 5 genes in 6 controls and 6 patients with colorectal cancer. There was excellent correlation between the two methods when β-values from microarrays were compared with %methylation from pyrosequencing data for the 12 individuals: Spearman r = 0.94, a total of 72 points; linear regression r2 = 0.93; P < 0.001 (Supplementary Fig. S5). We then examined a new cohort of 29 controls and 29 patients with colorectal cancer, who had not been tested in the original arrays, by pyrosequencing (Supplementary Fig. S6, left panel for each marker). They were compared with 12 controls and 24 patients with colorectal cancer that had been tested for these same 6 CpGs by arrays only (Supplementary Fig. S6, right panel for each marker). There are significant differences between controls and patients, for these independent cohorts of 58 and 36 individuals tested by pyrosequencing and microarrays, respectively.

We then expanded our pyrosequencing assessment to a total of 14 CpGs for these 5 genes (for 35 controls and 35 patients) as seen in Fig. 2. The additional dinucleotides were in the vicinity of the originally interrogated CpGs. All 14 tested CpGs showed significant differences in methylation in normal mucosa between patients with colorectal cancer and controls (Fig. 2A–E). CpGs showed significantly decreased methylation in normal colon of patients with colorectal cancer for PDK4 (Fig. 2A) and NR1H4 (Fig. 2C). Significantly higher methylation was observed for SPRR2A (Fig. 2B), SPRR1A (Fig. 2D), and PYCARD (Fig. 2E). We then questioned whether these differences could be used as methylation-based biomarkers for presence or absence of polyps in controls. Approximately half of the controls had polyps. The average methylation of NR1H4 was lower for subjects with polyps, although the difference did not reach statistical significance (Fig. 3A). However, we observed an increased average methylation for the five PYCARD CpGs, with significance for two CpGs (16:31121937 and 16:31121918), and borderline significance for three CpGs (16:31121929, 16:31121927 and 16:31121902; Fig. 3B).

We compared transcript levels of these five genes in normal mucosa of colorectal cancer patients and controls, and observed significantly increased expression for the five markers in patients with colorectal cancer (Fig. 4A–E). No expression differences were seen between controls with polyps and those without polyps.

Although average methylation of our five candidate genes was significantly different between patients with colorectal cancer and controls (Fig. 2A–E), there is some overlap between the two groups, for each of the five markers. We hypothesized that the methylation pattern for a collection of markers could have significant diagnostic power to distinguish normal mucosa between colorectal cancer patients and controls. Indeed, hierarchical clustering for classification of 70 individuals led to the identification of 2 major clusters: one cluster comprised predominantly of colorectal cancer patients and a second cluster composed exclusively of controls (Fig. 5A).

We applied the same principle to evaluate the discriminatory ability for classification of 35 controls with (n = 17) or without (n = 18) polyps. Empiric combinatory comparison using different sets of these five biomarkers (data not shown) revealed that NR1H4 and PYCARD aggregated data provide the best combination for a potential methylation signature (Fig. 5B). Hierarchical clustering yielded one group almost exclusively composed of controls with polyps (8/9 subjects) and a second group enriched in controls without polyps (17/26 subjects).

DNA-based biomarkers in normal colonic mucosa would be extremely useful because they have the potential to be diagnostic of colon cancer in the near term or, upon further development, may become prognostic indicators of colon cancer risk. Such biomarkers would provide discriminatory and quantitative biochemical measures to supplement the current endoscopic screening test that is both invasive and subjective.

Environmental factors, such as diet, may be the most important influences on colorectal cancer risk. Low dietary folate is one such risk factor. The polymorphism in MTHFR (677C→T) can also increase cancer risk when folate status is inadequate. Our mouse model, which develops intestinal neoplasia after low dietary folate, is a relevant model for human sporadic colorectal cancer because the mice do not have germline mutations and develop tumors over an extended period of time, without carcinogen induction. The use of Mthfr-deficient mice allows us to examine gene-nutrient interactions that have also been observed in human colorectal cancer. DNA methylation is altered by MTHFR 677C→T genotype and folate levels, with folate-deficient TT individuals showing the lowest global DNA methylation and the highest prevalence of cancer history (29).

Another dietary risk factor for colorectal cancer is high fat (30). Expression profiling in mice revealed significant differences for genes downstream of PPARA, a major regulator of lipid and glucose metabolism. We hypothesize that a disturbance in folate metabolism can result in activation of the RXR/PPARA pathway that increases fatty acid oxidation, generates oxidative stress/damage, and enhances glycolysis, setting the stage for tumorigenesis. Tumors have altered energy metabolism, with a preference for aerobic glycolysis (Warburg effect), instead of the tricarboxylic acid cycle (31). Because of the strong link between high fat diets and development of colon cancer, we had previously argued that the epigenetic reprogramming of lipid and carbohydrate metabolism probably preceded tumor development and was not programmed by the tumor at distant sites (7). Interestingly, only 25% of BALB/c Mthfr+/− mice developed tumors (15), and the BALB/c Mthfr+/− mice used for confirming mouse expression microarray data by qRT-PCR did not have tumors. These findings directly support the hypothesis that gene expression changes and reprogramming occur before tumor formation.

The retinoid pathway and metabolism of lipids and carbohydrates were predominant in murine expression profiling (Supplementary Fig. S3B) and human genome-wide methylation assessment (7). Correlation between vitamin A deficiency and tumor initiation has been confirmed in several studies (32 and references therein). Retinoic acid regulates gene expression through the retinoic acid receptor and retinoid X receptor (RAR/RXR) heterodimer. In addition, RXR interacts with other nuclear receptors such as PPARs. Retinoids cannot be synthesized in humans; they are converted from dietary carotenoids (33). Beta-carotene is the major provitamin A carotenoid. Retinaldehyde, the product of BCDO1, prevents formation of the RXR/PPAR heterodimer. Downregulation of Bcmo1, as observed in Mthfr+/− and FD mice, would result in lower retinaldehyde levels and increased PPARA activity. Retinaldehyde can be converted to retinoic acid or retinol by aldehyde dehydrogenases or aldo-keto reductases, respectively. We observed higher Aldh1a1 expression in FD-fed BALB/c mice (Supplementary Table S7 and Fig. 4B of ref. 15) and higher expression of Akr1c13 in BALB/c Mthfr+/− FD mice (Supplementary Fig. S4); these changes would also contribute to lowering retinaldehyde. There was increased expression of Rdh18 in BALB/c Mthfr+/− FD mice; retinol dehydrogenases can also metabolize retinaldehyde, although it is unclear whether the Rdh18 transcript encodes a functional protein (GenBank accession# AY053573). We have recently shown that inhibiting BCMO1 expression increases invasion and migration in human colorectal cancer cells, and that β-carotene, the BCMO1 substrate, upregulates the gene and reverses these effects (34).

To identify early human colorectal cancer events, we compared our list of murine candidate genes (Supplementary Table S6 in the present report and Supplementary Table S6 in ref. 15) with genes identified in methylation profiling of normal human colon (7 and C. Sapienza, 2012, Unpublished data) because disturbances in folate metabolism perturb methylation in both species (12, 35). This approach identified five mouse and human orthologous genes that showed, respectively, changes in expression in murine microarrays and changes in human methylation profiling. Bisulfite-pyrosequencing of human DNA in independent samples of normal mucosa of patients with cancer and controls confirmed significant methylation differences in a total of 14 CpGs (Fig. 2). These genes were: NR1H4 which can activate PPARA (27); PDK4, a target of PPARA that enhances glycolysis (26); PYCARD, a proapoptotic gene (28); and two different members of the SPRR family, involved in protection against oxidative damage. Although different methods and sites of biopsy could result in samples with different cell types, our analysis of methylation at more than 27,000 CpGs revealed differences for only 909 (3.3%) between cancer and control specimens at our least stringent statistical threshold (7). To address the issue of tumor sidedness, we examined normal colon from colorectal cancer patients with left- or right-sided tumors (n = 15 in each group). When we compared the 14 CpGs between groups, there was only one marker (one of the three CpGs in SPRR2A) that showed a significant difference in methylation (data not shown). Aging can also influence DNA methylation and modify colorectal cancer risk (36). However, since the age ranges in controls and patients were similar and there was no difference between the mean ages of these groups in our original study (7), from which the five candidates were selected for validation in this study, it is unlikely that age was a confounder.

Some of the controls had polyps, and one of the above genes, PYCARD, showed significant methylation differences between controls without polyps and controls with polyps (Fig. 3). Interestingly, controls included two cases of HPP (hyperplastic polyps), thought not to give rise to colon tumors, and they are at the very low end of the normal PYCARD methylation distribution while other controls with polyps are all in the upper part of the normal PYCARD distribution (Supplementary Table S1). Furthermore, in the heatmap in Fig. 5B, these 2 subjects cluster with the controls without polyps.

A major outcome of this study is the potential for using the methylation differences per se as molecular biomarkers for diagnosis. Although some methylation differences were relatively small, they were observed in two different cohorts, using two different methodologies (Supplementary Figs. S5 and S6), and are therefore reliable. These five genes also exhibited significant expression differences in normal human mucosa between controls and patients with colorectal cancer, with increased expression in patients (Fig. 4). Increased expression was associated with both decreased and increased methylation. While it is true that there is a general (but not absolute) inverse correlation between DNA methylation in the promoter region and transcript levels, our finding is not unusual since methylation within the body of genes and distal to genes is often positively correlated with transcript levels (37, 38). Specific mechanisms involving binding efficiency of transcriptional repressor(s) to methylated regions for example, may also result in changes in expression (39). This mechanism may be relevant here, since the assessed CpGs are all in the 5′ region of genes.

The increased expression of PDK4 and NR1H4 in patients with colorectal cancer (Fig. 4A and C) is consistent with the higher expression of their orthologs in BALB/c FD+/− mice (Fig. 1A and C). These results suggest that similar molecular changes exist in human and murine preneoplastic intestine. Higher expression of SPRR1A, SPRR2A, and PYCARD in normal tissue of patients with colorectal cancer may represent compensatory effects in response to the tumorigenic environment. Only 20% of mice show tumors at one year and they are quite small (1–2 mm); patients with colorectal cancer have well-developed tumors and their normal mucosa may have had sufficient time for compensatory expression changes. Mice lacking Pycard demonstrate polyp formation (40); PYCARD is an inflammasome-associated molecule with influences on diabetes, obesity, and cancer (41). SPRR genes are often induced in response to stress and extensively upregulated in various types of cancer (42). SPRR2A is overexpressed in early stages of prostate tumorigenesis (43), and is modulated by 5-aza-2′-deoxycytidine despite the absence of CpG islands (44).

Increased expression of PDK4 and NR1H4 in normal mucosa of colorectal cancer patients and in Mthfr+/− mice fed FD may be highly tumorigenic. As mentioned, the shift away from mitochondrial respiration is a hallmark of tumor metabolism (31, 45). Inhibition of pyruvate dehydrogenase (PDH) through phosphorylation, by pyruvate dehydrogenase kinases (PDK), results in decreased respiration in tumors (45). PDKs are a family of four kinases in humans (26); siRNA-based knockdown of PDK1 reverses PDH inhibition and the Warburg effect, and can inhibit tumor growth (45). PDK4 expression is increased by PPARA, by consumption of high fat diets and in diabetic states (26); its role in transformation has not been well studied. Decreased methylation of PDK4 in colorectal cancer mucosa is consistent with increased expression. We also observed significantly decreased methylation in the 5′ region of Pdk4 in Mthfr-deficient mice, for both diets (data not shown).

NR1H4 encodes the farnesoid-X-receptor (FXR). Bile acids, natural ligands for this receptor, can induce PPARA through a FXR response element in the human PPARA promoter (27). NR1H4 activation increases PDK4 expression (46). Decreased methylation of NR1H4 in human colorectal cancer mucosa is consistent with increased expression. In mice, we observed increased Nr1h4 methylation for Mthfr+/− FD mice compared with Mthfr+/+ FD mice, and a trend for increased methylation at two CpGs in the 5′ region, in Mthfr+/− FD mice compared with Mthfr+/− CD mice (data not shown).

Because tumorigenesis is a complex process, there are certainly other genes identified through microarrays in mice or humans that could contribute. For example, in Supplementary Fig. S4, we show 6 genes, in addition to Rdh18 and Akr1c13 (already mentioned), with confirmed expression changes due to folate or Mthfr deficiency.

Our cluster analysis of bisulfite pyrosequencing-based DNA methylation data provides an initial epigenetic signature for distinguishing normal mucosa from controls versus normal mucosa in patients with colorectal cancer (Fig. 5A). Additional methylation markers could improve the power of this diagnostic assay, and some of the genes discussed above, as well as other genes in our microarrays, could potentially improve the discriminatory power. However, we cannot exclude the possibility that some misclassification in Fig. 5 may have been due to a false negative finding during colonoscopy. Clustering of controls without polyps versus those with polyps (Fig. 5B) is also of interest, but requires additional markers. We used a systematic human genome-wide methylation marker discovery study with patients that were not screened for a specific cause of colorectal cancer, although none of the subjects had a history of familial cancer, colon polyps, or inflammatory bowel disease (7). Candidate genes were intersected with expression profiling data from our mouse model. This original, two-filter approach, resulting in gene identification within common pathways, provides a solid basis for an epigenetics signature for normal intestine in colon cancer. As colorectal cancer development proceeds through multiple stages, it would be useful to develop diagnostic tests for early intervention. Although many studies have reported methylation differences between normal colon and tumors, there are very few genes that have been confirmed by quantitative methods to exhibit methylation differences between controls and patients with colorectal cancer in normal mucosa (5, 7). Our approach has identified 5 genes, at 14 CpG sites, using the highly quantitative pyrosequencing method which could be adapted into a clinical setting (47). Methylation changes accumulate over years and could serve as sentinel markers before the appearance of polyps. Shedding of colon-derived DNA into stool could allow noninvasive testing (48). If differences are systemic, measurements of methylation changes in peripheral blood or saliva would also be extremely useful. Additional studies using our two-pronged approach may lead to identification of other biomarkers for the establishment of a biochemical measure of cancer risk that may be more objective than routine endoscopy.

No potential conflicts of interest were disclosed.

Conception and design: D. Leclerc, C. Sapienza, R. Rozen

Development of methodology: D. Leclerc, N. Lévesque, Y. Cao, L. Deng, C. Sapienza

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Leclerc, N. Lévesque, Y. Cao, L. Deng, Q. Wu, J. Powell, C. Sapienza, R. Rozen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Leclerc, N. Lévesque, Y. Cao, J. Powell, C. Sapienza, R. Rozen

Writing, review and/or revision of the manuscript: D. Leclerc, N. Lévesque, Y. Cao, L. Deng, Q. Wu, J. Powell, C. Sapienza, R. Rozen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Sapienza, R. Rozen

Study supervision: C. Sapienza, R. Rozen

The authors thank Drs. Elin Sigurdson and Andrew Godwin (Chase Cancer Center) for assistance with colon biopsies and tissue acquisition, and Nuala O'Leary (NCBI) for discussions on the Rdh18 gene.

This research was supported by the Canadian Institutes of Health Research (grant no. MOP 77596 to R. Rozen) and NIH (T32 CA 103652-05) and the Fels Institute for Cancer Research (to C. Sapienza). This work was also supported by a McGill University Health Centre Research Institute Fellowship (to N. Lévesque). The Research Institute is supported by a Centres grant from the Fonds de Recherche du Québec - Santé.

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.

1.
Moghaddam
AA
,
Woodward
M
,
Huxley
R
. 
Obesity and risk of colorectal cancer: a meta-analysis of 31 studies with 70,000 events
.
Cancer Epidemiol Biomarkers Prev
2007
;
16
:
2533
47
.
2.
Goel
A
,
Boland
CR
. 
Epigenetics of colorectal cancer
.
Gastroenterology
2012
;
143
:
1442
60
.
3.
Al-Sohaily
S
,
Biankin
A
,
Leong
R
,
Kohonen-Corish
M
,
Warusavitarne
J
. 
Molecular pathways in colorectal cancer
.
J Gastroenterol Hepatol
2012
;
27
:
1423
31
.
4.
Li
X
,
Yao
X
,
Wang
Y
,
Hu
F
,
Wang
F
,
Jiang
L
, et al
MLH1 Promoter methylation frequency in colorectal cancer patients and related clinicopathological and molecular features
.
PLoS ONE
2013
;
8
:
e59064
.
5.
Shen
L
,
Kondo
Y
,
Rosner
GL
,
Xiao
L
,
Hernandez
NS
,
Vilaythong
J
, et al
MGMT promoter methylation and field defect in sporadic colorectal cancer
.
J Natl Cancer Inst
2005
;
97
:
1330
8
.
6.
Milicic
A
,
Harrison
LA
,
Goodlad
RA
,
Hardy
RG
,
Nicholson
AM
,
Presz
M
, et al
Ectopic expression of P-cadherin correlates with promoter hypomethylation early in colorectal carcinogenesis and enhanced intestinal crypt fission in vivo
.
Cancer Res
2008
;
68
:
7760
8
.
7.
Silviera
ML
,
Smith
BP
,
Powell
J
,
Sapienza
C
. 
Epigenetic differences in normal colon mucosa of cancer patients suggest altered dietary metabolic pathways
.
Cancer Prev Res
2012
;
5
:
374
84
.
8.
Knock
E
,
Deng
L
,
Wu
Q
,
Leclerc
D
,
Wang
XL
,
Rozen
R
. 
Low dietary folate initiates Intestinal Tumors in Mice, with altered expression of G2-M Checkpoint Regulators Polo-Like Kinase 1 and cell division cycle 25c
.
Cancer Res
2006
;
66
:
10349
56
.
9.
Ma
J
,
Stampfer
MJ
,
Giovannucci
E
,
Artigas
C
,
Hunter
DJ
,
Fuchs
C
, et al
Methylenetetrahydrofolate reductase polymorphism, dietary interactions, and risk of colorectal cancer
.
Cancer Res
1997
;
57
:
1098
102
.
10.
Ulvik
A
,
Vollset
SE
,
Hansen
S
,
Gislefoss
R
,
Jellum
E
,
Ueland
PM
. 
Colorectal cancer and the methylenetetrahydrofolate reductase 677C → T and methionine synthase 2756A → G polymorphisms: a study of 2,168 case-control pairs from the JANUS cohort
.
Cancer Epidemiol Biomarkers Prev
2004
;
13
:
2175
80
.
11.
Teng
Z
,
Wang
L
,
Cai
S
,
Yu
P
,
Wang
J
,
Gong
J
. 
The 677C>T (rs1801133) polymorphism in the MTHFR gene contributes to colorectal cancer risk: a meta-analysis based on 71 research studies
.
PLoS ONE
2013
;
8
:
e55332
.
12.
Knock
E
,
Deng
L
,
Wu
Q
,
Lawrance
AK
,
Wang
XL
,
Rozen
R
. 
Strain differences in mice highlight the role of DNA damage in Neoplasia induced by low dietary folate
.
J Nutr
2008
;
138
:
653
8
.
13.
Garcia-Crespo
D
,
Knock
E
,
Jabado
N
,
Rozen
R
. 
Intestinal neoplasia induced by low dietary folate is associated with altered tumor expression profiles and decreased apoptosis in mouse normal intestine
.
J Nutr
2009
;
139
:
488
94
.
14.
Knock
E
,
Deng
L
,
Krupenko
N
,
Mohan
RD
,
Wu
Q
,
Leclerc
D
, et al
Susceptibility to intestinal tumorigenesis in folate-deficient mice may be influenced by variation in one-carbon metabolism and DNA repair
.
J Nutr Biochem
2011
;
11
:
1022
9
.
15.
Leclerc
D
,
Cao
Y
,
Deng
L
,
Mikael
LG
,
Wu
Q
,
Rozen
R
. 
Differential gene expression and methylation in the retinoid/PPARA pathway and of tumor suppressors may modify intestinal tumorigenesis induced by low folate in mice
.
Mol Nutr Food Res
2013
;
57
:
686
97
.
16.
Barrett
T
,
Wilhite
SE
,
Ledoux
P
,
Evangelista
C
,
Kim
IF
,
Tomashevsky
M
, et al
NCBI GEO: archive for functional genomics data sets–update
.
Nucleic Acids Res
2013
;
41
:
D991
5
.
17.
Bottone
FG
 Jr
,
Moon
Y
,
Kim
JS
,
Alston-Mills
B
,
Ishibashi
M
,
Eling
TE
. 
The anti-invasive activity of cyclooxygenase inhibitors is regulated by the transcription factor ATF3 (activating transcription factor 3)
.
Mol Cancer Ther
2005
;
5
:
693
703
.
18.
Huang
Y
,
Zhao
Q
,
Chen
GQ
. 
Phospholipid scramblase 1
.
Sheng Li Xue Bao
2006
;
58
:
501
10
.
19.
Chen
W
,
Possemato
R
,
Campbell
KT
,
Plattner
CA
,
Pallas
DC
,
Hahn
WC
. 
Identification of specific PP2A complexes involved in human cell transformation
.
Cancer Cell
2004
;
5
:
127
36
.
20.
Puustinen
P
,
Junttila
MR
,
Vanhatupa
S
,
Sablina
AA
,
Hector
ME
,
Teittinen
K
, et al
. 
PME-1 protects extracellular signal-regulated kinase pathway activity from protein phosphatase 2A-mediated inactivation in human malignant glioma
.
Cancer Res
2009
;
69
:
2870
7
.
21.
Westermarck
J
,
Hahn
WC
. 
Multiple pathways regulated by the tumor suppressor PP2A in transformation
.
Trends Mol Med
2008
;
14
:
152
60
.
22.
Takeshita
F
,
Ishii
KJ
. 
Intracellular DNA sensors in immunity
.
Curr Opin Immunol
2008
;
20
:
383
8
.
23.
Rizzatti
EG
,
Falcão
RP
,
Panepucci
RA
,
Proto-Siqueira
R
,
Anselmo-Lima
WT
,
Okamoto
OK
, et al
Gene expression profiling of mantle cell lymphoma cells reveals aberrant expression of genes from the PI3K-AKT, WNT and TGFbeta signalling pathways
.
Br J Haematol
2005
;
130
:
516
26
.
24.
García-Escudero
R
,
Paramio
JM
. 
Gene expression profiling as a tool for basic analysis and clinical application of human cancer
.
Mol Carcinog
2008
;
47
:
573
9
.
25.
Porté
S
,
Xavier Ruiz
F
,
Giménez
J
,
Molist
I
,
Alvarez
S
,
Domínguez
M
, et al
Aldo-keto reductases in retinoid metabolism: search for substrate specificity and inhibitor selectivity
.
Chem Biol Interact
2013
;
202
:
186
94
.
26.
Jeong
JY
,
Jeoung
NH
,
Park
KG
,
Lee
IK
. 
Transcriptional regulation of pyruvate dehydrogenase kinase
.
Diabetes Metab J
2012
;
36
:
328
35
.
27.
Goto
T
,
Kim
YI
,
Funakoshi
K
,
Teraminami
A
,
Uemura
T
,
Hirai
S
, et al
Farnesol, an isoprenoid, improves metabolic abnormalities in mice via both PPARα-dependent and -independent pathways
.
Am J Physiol Endocrinol Metab
2011
;
301
:
E1022
32
.
28.
Riojas
MA
,
Guo
M
,
Glöckner
SC
,
Machida
EO
,
Baylin
SB
,
Ahuja
N
. 
Methylation-induced silencing of ASC/TMS1, a pro-apoptotic gene, is a late-stage event in colorectal cancer
.
Cancer Biol Ther
2007
;
6
:
1710
6
.
29.
Friso
S
,
Udali
S
,
Guarini
P
,
Pellegrini
C
,
Pattini
P
,
Moruzzi
S
, et al
Global DNA hypomethylation in peripheral blood mononuclear cells as a biomarker of cancer risk
.
Cancer Epidemiol Biomarkers Prev
2013
;
22
:
348
55
.
30.
Sung
MK
,
Yeon
JY
,
Park
SY
,
Park
JH
,
Choi
MS
. 
Obesity-induced metabolic stresses in breast and colon cancer
.
Ann N Y Acad Sci
2011
;
1229
:
61
8
.
31.
Menendez
JA
,
Joven
J
,
Cufí
S
,
Corominas-Faja
B
,
Oliveras-Ferraros
C
,
Cuyàs
E
, et al
The Warburg effect 2.0: metabolic reprogramming of cancer stem cells
.
Cell Cycle
2013
;
12
:
1166
79
.
32.
Altucci
L
,
Gronemeyer
H
. 
The promise of retinoids to fight against cancer
.
Nat Rev Cancer
2001
;
3
:
181
93
.
33.
D'Ambrosio
DN
,
Clugston
RD
,
Blaner
WS
. 
Vitamin A metabolism: an update
.
Nutrients
2011
;
3
:
63
103
.
34.
Pham
DNT
,
Leclerc
D
,
Lévesque
N
,
Deng
L
,
Rozen
R
. 
β-Carotene 15,15′-monooxygenase 1 and its substrate β-carotene modulate migration and invasion in colorectal carcinoma cells
.
Am J Clin Nutr
2013
;
98
:
413
22
.
35.
Bottiglieri
T
,
Arning
E
,
Wasek
B
,
Nunbhakdi-Craig
V
,
Sontag
JM
,
Sontag
E
. 
Acute administration of L-DOPA induces changes in methylation metabolites, reduced protein phosphatase 2A methylation, and hyperphosphorylation of Tau protein in mouse brain
.
J Neurosci
2012
;
32
:
9173
81
.
36.
Wallace
K
,
Grau
MV
,
Levine
AJ
,
Shen
L
,
Hamdan
R
,
Chen
X
, et al
Association between folate levels and CpG Island hypermethylation in normal colorectal mucosa
.
Cancer Prev Res
2010
;
3
:
1552
64
.
37.
Hahn
MA
,
Wu
X
,
Li
AX
,
Hahn
T
,
Pfeifer
GP
. 
Relationship between gene body DNA methylation and intragenic H3K9me3 and H3K36me3 chromatin marks
.
PLoS ONE
2011
;
19
:
e18844
.
38.
Jjingo
D
,
Conley
AB
,
Yi
SV
,
Lunyak
VV
,
Jordan
IK
. 
On the presence and role of human gene body DNA methylation
.
Oncotarget
2012
;
3
:
462
74
.
39.
Pipaon
C
,
Real
PJ
,
Fernandez-Luna
JL
. 
Defective binding of transcriptional repressor ZEB via DNA methylation contributes to increased constitutive levels of p73 in Fanconi anemia cells
.
FEBS Lett
2005
;
579
:
4610
4
.
40.
Allen
IC
,
TeKippe
EM
,
Woodford
RM
,
Uronis
JM
,
Holl
EK
,
Rogers
AB
, et al
The NLRP3 inflammasome functions as a negative regulator of tumorigenesis during colitis-associated cancer
.
J Exp Med
2010
;
207
:
1045
56
.
41.
Wen
H
,
Ting
JP
,
O'Neill
LA
. 
A role for the NLRP3 inflammasome in metabolic diseases-did Warburg miss inflammation?
Nat Immunol
2012
;
13
:
352
7
.
42.
Hong
SH
,
Lee
JE
,
Jeong
JJ
,
Hwang
SJ
,
Bae
SN
,
Choi
JY
, et al
Small proline-rich protein 2 family is a cluster of genes induced by estrogenic compounds through nuclear estrogen receptors in the mouse uterus
.
Reprod Toxicol
2010
;
30
:
469
76
.
43.
Song
H
,
Zhang
B
,
Watson
MA
,
Humphrey
PA
,
Lim
H
,
Milbrandt
J
. 
Loss of Nkx3.1 leads to the activation of discrete downstream target genes during prostate tumorigenesis
.
Oncogene
2009
;
28
:
3307
19
.
44.
Yamashita
S
,
Tsujino
Y
,
Moriguchi
K
,
Tatematsu
M
,
Ushijima
T
. 
Chemical genomic screening for methylation-silenced genes in gastric cancer cell lines using 5-aza-2′-deoxycytidine treatment and oligonucleotide microarray
.
Cancer Sci
2006
;
97
:
64
71
.
45.
Fujiwara
S
,
Kawano
Y
,
Yuki
H
,
Okuno
Y
,
Nosaka
K
,
Mitsuya
H
, et al
PDK1 inhibition is a novel therapeutic target in multiple myeloma
.
Br J Cancer
2013
;
108
:
170
8
.
46.
Mencarelli
A
,
Cipriani
S
,
Renga
B
,
D'Amore
C
,
Palladino
G
,
Distrutti
E
, et al
FXR activation improves myocardial fatty acid metabolism in a rodent model of obesity-driven cardiotoxicity
.
Nutr Metab Cardiovasc Dis
2013
;
23
:
94
101
.
47.
Mikeska
T
,
Bock
C
,
El-Maarri
O
,
Hübner
A
,
Ehrentraut
D
,
Schramm
J
, et al
Optimization of quantitative MGMT promoter methylation analysis using pyrosequencing and combined bisulfite restriction analysis
.
J Mol Diagn
2007
;
9
:
368
80
.
48.
Jain
S
,
Wojdacz
TK
,
Su
YH
. 
Challenges for the application of DNA methylation biomarkers in molecular diagnostic testing for cancer
.
Expert Rev Mol Diagn
2013
;
13
:
283
94
.

Supplementary data