Association between Folate Levels and CpG Island Hypermethylation in Normal Colorectal Mucosa Free

DNA methylation is increasingly recognized as an alternative mechanism to mutation in the silencing of gatekeeper genes during colorectal carcinogenesis (1). Hypermethylation in CpG islands (CGI), cytosine-guanine dinucleotide clusters located within the promoter region of many genes, is commonly observed in colorectal tumors, yet these sites are largely unmethylated in normal tissues (1, 2). During the course of normal aging, however, low-level methylation of CGIs in several regulatory or mismatch repair genes has been observed in the normal colorectal mucosa (3–6). The age-related increases in both CGI methylation and risk of colorectal neoplasia have led some to propose methylation as a predisposing phenomenon for colorectal neoplasia (7, 8). Increased DNA methylation is believed to lead to the silencing of the expression of regulatory or mismatch repair genes, which, in turn, may provide a growth advantage to the affected cells (7, 8).

Although the etiology of these age-related epigenetic changes is not well understood, a correlation between methylation levels and personal characteristics of an individual (e.g., diet, lifestyle factors) may provide a biologic link between established risk and protective factors for colorectal neoplasia and risk of disease. The ERα and SFRP1 genes are expressed in normal colorectal tissue, exhibit intense CGI methylation in colon cancer, and are believed to regulate growth and differentiation, two factors associated with tumorigenesis (3–6). ERα was selected because of strong prior data connecting it with aging (a significant risk factor of colorectal neoplasia) and SFRP1, a known antagonist of the Wnt signaling pathway (9), was studied because of suggestions that it is a gatekeeper gene in the earliest phases of colorectal carcinogenesis (4). Frequent CGI methylation of the ERα and SFRP1 genes has been observed in several different types of tumors, including breast, prostate, lung, and most colorectal adenocarcinomas (4, 10–13). Methylation of these genes has also been frequently observed in the mucosa adjacent to carcinomas (4, 5, 11) and in normal mucosa in patients with and without colorectal neoplasia (14–18).

In the present study, we examined the association among demographic, lifestyle, dietary, and genetic factors and the degree of gene-specific DNA methylation (ERα, SFRP1) in normal colorectal mucosa among subjects enrolled in a multicenter chemoprevention trial testing the efficacy of aspirin and folate on the risk of large bowel adenomas. Specifically, we examine the association between these factors and CGI methylation in the ERα and SFRP1 genes in both the right colon and the rectum.

Our data were collected as part of a randomized, double-blind, placebo-controlled trial testing the efficacy of aspirin, folic acid, or both, in the prevention of colorectal adenomas (19, 20). Written informed consent was obtained from each participant, and the institutional review board (IRB) of each participating institution approved the studies.

Study participants were recruited between July 1994 and March 1998 at 9 clinical centers in North America. Each participant had a history of at least 1 colorectal adenoma excised before study entry and had no remaining polyps in the entire large bowel after complete colonoscopic examination. The study had a 3-by-2 factorial design, comparing 81 and 325 mg of aspirin per day with placebo and comparing 1 mg of folic acid per day with placebo. The trial was initially designed to investigate only aspirin, but soon after recruitment began, the study was extended to examine folic acid also. One hundred patients who were randomly assigned to receive aspirin or placebo could not be included in the factorial design for folic acid, but they were in the analyses of aspirin.

Follow-up colonoscopy was done approximately 3 years after the qualifying examination. Polyps were classified as neoplastic (adenomatous) or nonneoplastic by the central study pathologist, who also assessed the degree of dysplasia and the extent of villous component in each adenoma. We defined advanced lesions as invasive carcinoma or adenomas with at least 25% villous component, high-grade dysplasia, or an estimated size of 1 cm or greater. In addition to collection of polyps during the year 3 colonoscopy, 84.6% of subjects (n = 914) were approached for permission to obtain normal mucosal biopsy specimens; 781 (85.4%) consented. Of the 167 individuals who were not approached, 92 (55%) were from 1 center that could not participate in the biopsy study and the remaining individuals had been randomized only to aspirin.

At enrollment, participants completed a questionnaire addressing basic demographic characteristics, lifestyle factors, medical history, and usual diet (assessed using a validated food frequency questionnaire; ref. 21). We analyzed several demographic and lifestyle factors, including age (quartiles), sex, race (Caucasian, African American, Hispanic, other), smoking status (categorized as “never,” “former,” and “current”), alcohol use (none, ≤1 drink per day, >1 drinks per day), and body mass index (BMI), which was divided into 3 categories by using the standard established by the World Health Organization: normal (<25 kg/m²), overweight (25–29.9 kg/m²), and obese (BMI ≥ 30 kg/m²).

At baseline, we also assessed several dietary measures by using a validated semiquantitative food frequency questionnaire (21): daily intakes of folate, protein, carbohydrates, fiber, and fat. The estimated intakes of these nutrients were adjusted for caloric intake, using residuals computed from the linear regression of the log of the nutrient intake on the log of caloric intake. During the course of the trial, we also asked subjects twice a year about their use of nutritional supplements, including a multivitamin provided to subjects on request that did not contain folic acid. We coded multivitamin use as positive if subject reported usage on 1 or more of the semiannual questionnaires.

Blood specimen collection and analysis and genotyping assays were previously described (22). Briefly, nonfasting blood samples were obtained from subjects both at baseline and at the follow-up approximately 3 years after study entry by using 7-mL EDTA Vacutainer brand tubes. After collection, specimens were immediately put on ice and then centrifuged at 1,100 × g for 10 minutes. Whole blood, plasma, and buffy coat fractions were stored at −20°C for 6 to 12 months and then transferred to Dartmouth Medical School where they were stored at −80°C until analysis. Plasma levels of folate were determined by a microbiological assay by using a colistin sulfate–resistant strain of Lactobacillus leichmannii (23). EDTA samples with low (<2 nmol/L) or no folate, attributable to inhibition of bacterial growth by antibiotics, were reanalyzed with a method based on measurement of folate as p-aminobenzoylglutamate equivalents (24). Circulating levels of plasma B₂, B₆, B₁₂, and homocysteine were also measured, as were genotypes for key enzymes in the 1-carbon metabolism pathway related to DNA synthesis and methylation: CBS-1080 C>T, CBS-699 C>T, MTHFR-677 C>T and 1298 C>T, MTR-2756 A>G, and D919G and MTRR-66 A>G, as described previously (25, 26). RBC folate was determined by the ACS:180 folate assay, a competitive immunoassay using direct chemiluminescent technology (Bayer Corporation). The year 3 RBC concentrations were calculated using baseline hematocrit values because these were not obtained at the year 3 blood draw. Biochemical analyses were conducted blinded to methylation features, randomized treatment assignment, and other subject characteristics.

At the beginning of the year 3 follow-up colonoscopy, 2 biopsy specimens were taken from normal appearing mucosa in the rectum (10 cm above the anus). After the endoscope was advanced to the cecum, 2 biopsy specimens were obtained from the mid-ascending colon (5 cm above the ileocecal valve). Immediately after each biopsy was taken, the specimens were removed from the forceps, placed into freezer tubes, and immersed in liquid nitrogen or dry ice/ethanol slurry in which they remained until stored in a −70°C freezer at the central facility of the clinical center before shipment on dry ice for analysis.

We obtained the biopsy specimens (as described earlier) of normal mucosa from 768 (98.3%) of the 781 individuals who were originally consented. Of the 13 individuals who consented but did not provide biopsy specimens, the reasons were as follows: schedule conflicts (n = 9), no IRB approval at hospital (n = 3), and unknown (n = 1). Of the 768 participating subjects, we collected a total of 3,072 biopsy samples. Our analysis includes a convenience sample of 1,000 (499 from the rectum and 501 from the right colon) taken from 388 individuals. Of these, 1 subject had only 1 biopsy from the rectum, 274 subjects had 2 biopsies (273 from both sides of the colon and 1 from the right colon only), 1 subject had 3 biopsies (1 from the right colon and 2 from the rectum), and 112 subjects had 4 biopsies (111 had 2 biopsies from each side of the colorectum and 1 subject had 3 from the right colon and 1 from the rectum).

DNA extraction followed standard protocols in the laboratory. Briefly, the biopsy specimen was minced using a razor blade and transferred to a tube containing lysis buffer (1% SDS, 1 mg/mL of proteinase K, LTE pH 8.0). The solution was incubated at 50°C for 1 hour, followed by phenol/chloroform extraction and ethanol precipitation. The quantity and quality of DNA were then determined by running a small aliquot on a 1% agarose gel and comparing it to a set of DNA standards.

Quantitative bisulfite pyrosequencing method was used for the ERα and SFRP1 genes (27). Briefly, 2 μg of genomic DNA was denatured with 2 mol/L of NaOH for 10 minutes, followed by incubation with 3 mol/L of sodium bisulfite (pH 5.0) for 16 hours at 50°C. After treatment, DNA was purified by using a Wizard Miniprep Column (Promega), precipitated with ethanol, and resuspended in 30 μL of distilled water. A 2-μL of aliquot was used as the template for PCR. For ERα, the primers: TGTGTTTTTTTTTTAGGTGG (sense) and 5′-biotin AACCATCCCAAATACTTTAATA (antisense) were used to amplify a 125-bp fragment of the promoter CGI. For SFRP1, the primers: TGTTTTTTAAGGGGTGTTGA (sense) and 5′-biotin CTCCRAAAACTACAAAACTAAAAT (antisense) were used to amplify a 218-bp fragment of the promoter CGI. PCR cycling condition for ERα was denaturation at 95°C for 5 minutes, followed by 50 cycles at 95°C for 30 seconds, 56°C for 1 minute, 72°C for 45 seconds, and a final extension at 72°C for 7 minutes. For SFRP1, touchdown PCR was done at 61°C for 5 cycles, 59°C for 5 cycles, 57°C for 5 cycles, and 55°C for 35 cycles. For pyrosequencing, the sequencing primers were GGATAAGGTTTGTATTTTG for ERα and GTTTGGTTTTAGTAAAT for SFRP1. The methylation level was averaged over the following CpG sites (relative to transcription start site): ERα 218, 220, 226 bp; SFRP1 −140, −134, −130, −128, −119 bp. For each assay, setup included positive controls (samples after SssI treatment) and negative controls (samples after whole genomic amplification), mixing experiments to rule out bias, and repeat experiments to assess reproducibility. Optimizing annealing temperature of PCR was used to overcome PCR bias as reported (28).

To assess the reproducibility of measurements in duplicate biopsies, we computed intraclass correlation coefficients and to assess the reliability of the assays, we calculated the coefficient of variation (CV) using replicate measures from the same biopsy on separate PCR gels. We had 500 replicate measures for ERα and 1,000 for SFRP1. Agreement between measures on biopsies on the right colon and rectum was examined by Spearman's rank correlation coefficient.

We used generalized estimating equation (GEE) regression analysis to assess the association between ERα and SFRP1 gene methylation levels and selected variables. We also analyzed the average of ERα and SFRP1 methylation levels and selected variables using standardized z-scores (29) as described previously (30). Briefly, for each gene we calculated the z-score as (X – m)/s, where X represents the methylation data of each gene in each sample; m represents the mean of methylation of each gene among all samples; and s stands for standard deviation. We then determined the average methylation score by summing the z-scores for ERα and SFRP1 and dividing by 2. We used GEE modeling because most individuals had more than 1 sample to account for the within-subject correlation of measurements from the same site; it is a minor component compared with the demographic differences. For each analysis, we produced means and standard errors (SE) while adjusting for age as continuous variable, sex, folate treatment assignment, smoking status (never, former, and current smoker), and daily alcohol intake (none, ≤1/d, ≥2 drinks/d) during the 3-year randomized period. Means for dietary intakes of nutrients were also adjusted for log-transformed daily calories. To measure the association between methylation and adenoma risk, we estimated risk ratios (and 95% confidence intervals) for 1 or more adenomas after randomization, calculated with generalized linear regression analyses using a logarithmic link and a Poisson distribution. We obtained the relative risks and P for trend by using orthogonal linear contrasts. We used Wald tests to assess statistical interactions. All effect estimates were adjusted for age, sex, folic acid and aspirin treatment assignment, smoking status, daily alcohol intake, and time since randomization. All tests of statistical significance were 2-sided.

We used analysis of variance methodology to determine the proportion of variance in RBC folate levels explained by folate treatment and other covariates including age, sex, smoking, alcoholic drinks per day, dietary folate, and log of calories. We could not obtain an R² using the GEE methodology, because no R² extension has been developed that incorporates the correlations estimated in a GEE model. We also computed a coefficient of partial determination to assess the unique contribution of folate treatment to the proportion of variation in RBC folate levels in the multivariate context.

The mean age of the study participants was 57.8 years (SD = 9.1) and 63.4% were men. The participants in this methylation study were broadly similar to those in the parent study (Table 1). Mean baseline dietary folate intake, baseline plasma folate, and year 3 RBC folate levels were significantly lower (P < 0.05) in the subjects included in the methylation analysis than in the entire parent population.

We first examined the reliability and reproducibility of the measurements. The CV of the ERα assay was 8.6% and for SFRP1 was 13.6%. For ERα methylation, the intraclass correlation coefficient between duplicate biopsies taken from the rectum was ρ = 0.66 (P < 0.0001) and ρ = 0.63 (P < 0.0001) for the right colon. For SRFP1, the intraclass coefficient between biopsies of the rectum was ρ = 0.45 (P< 0.0001) and of the right colon, ρ = 0.52 (P < 0.0001). Next, we assessed the patterns of methylation in the right colon and rectum. Overall, for ERα, the correlation between biopsies of the right colon and rectum was r = 0.42 (P < 0.0001); for SFRP1, the correlation was r = 0.37, (P < 0.0001). For ERα, mean percent methylation was 12.1% in the rectum and 9.6% in right colon (P = 0.001). A similar pattern of lower values in the right colon was observed for SFRP1: 23.3% methylation in the rectum and 20.8% in the right colon (P = 0.001).

Increasing age was significantly associated with higher mean methylation levels for both ERα and SFRP1 (Tables 2–4). The percentage increase in methylation for age increments of 10 years was 1.7% for ERα (P < 0.001) and 2.9% for SFRP1 (P < 0.001). In general, methylation levels were similar for men and women for both ERα and SFRP1. However, methylation levels among Caucasians, African Americans, and Hispanics exhibited marked differences for both ERα and SFRP1 genes (Tables 2–4). ERα methylation levels were higher among Caucasians and lower among African Americans and Hispanics (P for heterogeneity = 0.008). We observed a different pattern for SFRP1, with lower mean levels among Caucasians and African Americans than among Hispanics (P for heterogeneity = 0.0005). For both genes, we also observed significant interactions between race and side of colorectum (P for interaction for Erα = 0.05; P for SFRP1 = 0.009; results are summarized in Tables 2–4).

We observed no significant associations between BMI level (data not shown), smoking status, or amount of alcohol consumed and percentage methylation for either ERα or SFRP1 (Tables 2–4). However, for Erα, there was a suggestion that increasing alcohol intake was associated with lower percentage methylation (P for trend = 0.07). No pattern was observed for SFRP1.

Neither aspirin nor folic acid treatment was significantly associated with ERα or SFRP1 percentage methylation (Tables 2–4). However, subjects assigned to aspirin treatment exhibited slightly lower ERα methylation in the right colon than placebo (P for trend = 0.10). We observed no association between ERα methylation and any dietary variable, including intakes of total fat, carbohydrate, total dietary fiber (data not shown), protein intake, or dietary folate (Tables 2 and 4). Use of study multivitamins (which did not contain folic acid) during the trial was associated with a lower percentage methylation for the ERα genes (Table 2) and for the combined ERα/SFRP z-score measure (Table 4). We also observed a significant interaction between multivitamin use and age for ERα (P < 0.01) and a borderline interaction for SFRP1 (P <0.06). Subjects above the median age (>58 years) who reported taking a multivitamin had lower mean methylation levels (11.4%) than those not taking a multivitamin (12.8%). For subjects below the median age who reported multivitamin use, mean methylation levels were higher (9.8%) than those who did not (9.5%). A similar pattern was observed for SFRP1: older/multivitamin use (23.5%), older/none (25.3%), younger/multivitamin use (20.0%), and younger/none (19.6%).

For SFRP1, we observed a strong inverse association between higher levels of protein intake and lower methylation (Tables 3 and 4). Among those in the highest quartile of dietary protein intake, the average percentage methylation was 20.9% versus 22.7% for the lowest quartile (P for trend = 0.006). The association was observed in both the right colon and rectum. None of the other dietary variables exhibited an association with percentage SFRP1 methylation (data not shown).

There was no significant association between baseline plasma levels of B2, B6, B12, or homocysteine and percentage methylation of the ERα or SFRP1 gene (data not shown). We observed no significant association between MTHFR polymorphisms C677T or A1298C and percentage CGI methylation for either ERα or SFRP1 (data not shown).

Year 3 RBC folate levels were associated with higher levels of methylation in ERα, SFRP1, and the combined ERα-SFRP1 measurements (Table 5 and Fig. 1). We did not observe any statistically significant interactions between folic acid treatment and RBC folate levels on ERα methylation (P for interaction = 0.54), SFRP1 methylation (P for interaction = 0.15), or combined ERα-SFRP1 methylation (P for interaction = 0.23). There were no significant associations between year 3 plasma folate levels and SFRP1 or ERαmethylation (Table 5), nor were there associations for baseline dietary folate levels.

Folic acid treatment was significantly associated with RBC folate levels. The means (SE) in the folic acid and placebo groups were 931 ng/mL (SE) and 682 ng/mL (SE), respectively (P < 0.0001). Age and current smoking status were significantly related to RBC folate levels but not to sex, number of alcoholic drinks per day, dietary folate, or log of calories. All of these variables (including folic acid treatment) explained 19% of the variance in RBC folate levels. The coefficient of partial determination for folic acid treatment explained 13% of the variation in RBC folate levels.

We observed no significant association between percentage ERα or SFRP1 methylation level and risk of adenoma or hyperplastic polyps (Table 6). For ERα, those in the highest compared with the lowest tertile of methylation had an R² for any adenoma of 1.00 (0.83–1.21) and for any hyperplastic polyp, 1.33 (0.94–1.84). For SFRP1 in these same tertile comparisons, the RR for any adenoma was 0.93 (0.76–1.13) and for any hyperplastic polyp, 0.98 (0.69–1.38). For both ERα and SFRP1, there were nonsignificant trends for associations between higher levels of methylation and hyperplasic polyp occurrence in the right colon, but the number of hyperplasic polyps in this study was small.

In this large, multicenter clinical polyp prevention study, we found that methylation levels of the ERα and SFRP1 CGIs in the normal colorectal mucosa differed by large bowel region, age, race, and RBC folate levels. Methylation levels were significantly higher in the rectum than in the right colon, and increasing age was strongly associated with higher levels of methylation in both the right colon and the rectum. For both ERα and SFRP1, we observed that Caucasians had higher levels of methylation than African Americans whereas Hispanics had relatively low levels of ERα and relatively high levels of SFRP1. We found no significant relationship of the level of methylation for CGI site with aspirin or folic acid treatment, dietary folate, or any 1-carbon folate metabolism polymorphism. However, we did observe a relationship between higher levels of year 3 RBC folate and increasing levels of ERα and SFRP1 methylation.

Our findings of an age-related increase in CGI methylation in the normal mucosa parallel those of several other investigations (3–6). Our study is 1 of 2 to document age-related increases in SFRP1 methylation within the normal mucosa (14). Another investigation (31) reported no evidence of SFRP1 methylation in the normal mucosa in a younger study population (ages 20–54). Other studies have reported low-level SFRP1 methylation in normal tissue adjacent to tumors but no effort was made to examine an association with age (11). For ERα, the evidence of an age-related increase in methylation in the normal colorectal tissue has been documented in most other investigations (4–6) yet not all (14). Although the differences detected are small in some instances (2%–4% methylation), they can be significant if they represent differences in the numbers of colorectal crypts with dense methylation. This has previously been shown to be the case (32).

Investigators have become increasingly aware that there are clear physiologic, morphologic, and biochemical differences between the right and left colorectums (33, 34), and these differences may shed light on why we observed differences in methylation between the right colon and the rectum (33, 34). Our results parallel an earlier study that also showed higher ERα methylation levels in the left colorectum than in the right (4), although in that study the samples were selected from a mix of patients with colorectal cancer, adenomas, other disease, and sigmoidoscopy-only “normal controls,” which represents a subject population that may limit comparability with ours. Of the 2 other investigations (6, 35) that examined ERα methylation in normal colorectal mucosa, one (35) reported no difference by anatomic location and in the other biopsies were selected from the rectum only (6). In the first study to show no difference by region of the large bowel (35), no description is provided about the number of right versus left biopsies.

Why methylation patterns may differ by large bowel region is not well understood. Theoretically, the differences in CGI methylation between the right and left colorectums may be driven by physiologic/biological differences in the anatomic regions. For example, there are many known differences in the biology of the right and left colorectums such as different embryologic origin, crypt length, and apoptotic index (33, 34). Alternatively, the higher methylation levels that we observed in the rectum may be specific to the ERα and SFRP1 genes, and other CGIs might show a different pattern.

In our study population, African Americans had lower levels of methylation than Caucasians and Hispanics. We believe that our study is the first to consider CGI methylation levels in normal colorectal mucosa of different racial groups. One other study reported no significant difference in ERα methylation levels in colorectal tumor tissue when comparing Caucasian and non-Caucasian persons, yet the average ERα CGI methylation level was higher in Caucasians (36). There are several possible explanations for our findings, including lifestyle factors, genetic protection or predisposition to methylation, and polymorphisms in the genes studied themselves. Still, the number of African Americans studied is small and our findings could also be a result of chance.

Very few of the lifestyle or dietary variables we examined were related to CGI methylation levels. Previous research has suggested a relationship between dietary folate and genomic methylation in normal mucosa (37, 38), but our study is the first to examine the association between folic acid supplementation and CGI methylation in normal tissue. In findings consistent with ours, another study reported no association between serum folate levels and ERα methylation in normal mucosa (6). However, we did observe that RBC folate levels were positively associated with ERα and SFRP1 methylation levels. It is interesting and biologically relevant that RBC folate levels showed a stronger association than plasma folate. The former more closely reflect long-term intake and also reflect intracellular folate directly. In light of these data, it might seem surprising that randomization to the folate supplementation arm was not associated with a difference in DNA methylation. This is likely due in part to the fact that the difference in RBC folate between the placebo and folate treatment arms (249 ng/mL) would correspond to a less than 0.5% difference in ERα and SFRP1 methylation, something that would be difficult to detect with this technology and sample size. These data have important implications regarding the safety of supplementary folate administration in healthy adults, given the hypothesis that methylation in normal mucosa may be a predisposing phenomenon for colorectal neoplasia (discussed later).

All patients had detectable levels of ERα and SFRP1 methylation, and no significant differences were observed among patients with and without recurrence of colorectal adenomas. This implies that the methylation of ERα and SFRP1 is not rate limiting in the development of adenomas in this population and is consistent with the idea that a mutation in APC (39) is a critical first step in this regard. However, there may be other age-related methylation genes such as MYOD1, DKK1, and HPP1 (14, 16–18) that could be associated with adenoma occurrence. It has also been proposed that mutations of these genes may be more likely to occur in a “predisposed” mucosa, and that aberrant DNA methylation is one of the factors that results in this predisposition (40).

To test whether DNA methylation does in fact lead to this predisposition, one would need to study a mixed population with patients at high and low risk of adenoma formation, something that was not done here, given that all patients had adenomas at study entry. Comparing results across studies is difficult for many reasons, including differences in (1) the genes examined (2), the control for confounding variables such as age and sex (3), or the colorectal location(s) selected for mucosal sampling (18). Despite these challenges, several studies have examined CGI methylation levels in the normal mucosa of persons with and without neoplastic lesions (14–18, 41, 42). Of these, 5 have specifically examined ERα and SFRP1 (14–18). Four of these studies have reported lower ERα methylation (14, 16–18) among those with advanced neoplasia than among those with adenomas or disease-free controls, whereas 1 reported higher ERα levels in patients with adenoma (15). SFRP1 was associated with lower levels of methylation in mucosa of patients with neoplasia than those without in one study (18) and was unrelated to pathology in another (14).

Although not statistically significant, we observed that a higher level of ERα methylation was associated with an increased risk of hyperplastic polyps, especially right-sided hyperplastic polyps. Interestingly, 2 other investigations (6, 18) also observed increased levels of ERα CGI methylation in the normal mucosa in persons with hyperplastic polyps or proximal serrated polyps compared with controls. Historically, hyperplastic polyps have been considered indolent, nonneoplastic lesions (43, 44). However, currently it is understood that a subset of “hyperplastic” polyps are in fact sessile, serrated adenomas, the precursor lesion to cancers evolving along the serrated adenoma pathway (45, 46). The lesions of the serrated pathway have been closely associated with the CpG island methylator phenotype, which involves the methylation of multiple cancer-specific CGIs (1, 47–49). Future studies will be needed to determine whether CGI methylation in the normal mucosa predisposes toward to serrated lesions, as these lesions seem to have a somewhat distinct epidemiologic profile (50).

There are several strengths to our study. First, we examined methylation levels in a well-characterized population, with a systematic protocol, using a quantitative technique capable of detecting low-level methylation. ERα was selected because of strong prior data connecting its methylation state with aging, and SFRP1 was studied because of suggestions that it is a gatekeeper for neoplasia in the colon. The fact that ERα and SFRP1 methylation was consistently correlated with age (and with each other) confirms the validity of the measurements. In addition, subjects were part of a large, multicenter polyp prevention study with excellent follow-up and compliance with treatment. Ours is the first study with the magnitude and rigor to examine the effect of folic acid supplementation on methylation levels. Limitations to the study include the entry criteria: all subjects had at least 1 adenoma and so may be different from patients with no history of polyps. Thus, it remains to be determined whether mucosal methylation levels are lower in individuals who never develop adenomas. Furthermore, we used a convenience sample of specimens, which may not reflect the entire population of 768 persons with biopsies. There was a relatively low intraclass correlation between biopsies in a given individual. Given the relatively low CV of the assay, the primary reason for the low correlation is likely due to the natural variation of the biopsies. It will be important in future studies to obtain duplicate biopsies from many segments of the colorectum, given the known biological and anatomic differences that occur by region.

Overall, our study results point to differences in methylation levels by age, race, different regions of the large bowel, and RBC folate levels. This is one of the first large-scale studies to document associations between epigenetic measures and dietary factors in adults.

No potential conflicts of interest were disclosed.

J.P. Issa is an American Cancer Society Clinical Research professor supported by a generous gift from the F. M. Kirby Foundation.

Grant Support

This work was supported by the National Cancer Institute at the National Institute of Health (R01-CA105346 and R01-CA89245 to J.P. Issa and N01-CO-12400, R01-CA-059005, and U54-CA100971 to J.A. Baron).

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.