Abstract
Background: Folate is essential for one-carbon metabolism, a pathway required by DNA synthesis, methylation, and repair. Low dietary and circulating folate and polymorphic variation in this pathway are associated with increased risk of colorectal adenoma and cancer.
Methods: We genotyped 882 single nucleotide polymorphisms (SNP) in 82 one-carbon metabolism genes for 1,331 cases of advanced colorectal adenoma, identified by sigmoidoscopy at baseline, and 1,501 controls from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO). We evaluated associations between one-carbon genes and adenoma risk in all subjects and stratified by folate intake. We applied the Adaptive Rank Truncated Product (ARTP) method to assess statistical significance at the gene and pathway levels.
Results: Folate intake was inversely associated with advanced colorectal adenoma risk [odds ratio (OR) by quartile = 0.85, P = 1.9 × 10−5]. We found no statistically significant associations between one-carbon genes and adenoma risk in all subjects. As hypothesized, we observed a statistically significant pathway-level association (P = 0.038) in the lowest quartile of folate; no significant associations were found in higher quartiles. Several genes including adenosine deaminase (ADA) and cysteine dioxygenase (CDO1) contributed to this signal (gene-level P = 0.001 and 0.0073, respectively). The most statistically significant SNP was rs244072 in ADA (P = 2.37 × 10−5).
Conclusions and Impact: Stratification by dietary folate and application of the ARTP method revealed statistically significant pathway- and gene-level associations between one-carbon metabolism genes and risk of advanced colorectal adenoma, which were not apparent in analysis of the entire population. Folate intake may interact with associations between common variants in one-carbon metabolism genes and colorectal adenoma risk. Cancer Epidemiol Biomarkers Prev; 21(3); 417–27. ©2012 AACR.
This article is featured in Highlights of This Issue, p. 389
Introduction
Rates of colorectal cancer incidence and mortality have decreased in recent years; however, colorectal cancer remains the third most common cancer and third leading cause of cancer-related death in both men and women in the United States (1). Although most colorectal adenomatous polyps will not develop into cancer, these are commonly accepted as precursor lesions of colorectal cancer (2, 3), and screening by colonoscopy or sigmoidoscopy and polyp excision is standard of care for adults ages 50 and older (4, 5).
Folate is essential for one-carbon metabolism, a pathway required by DNA synthesis, methylation, and repair (6). Folate is a water-soluble B-vitamin found in leafy green vegetables, beans and peas, many fruits, and fortified grains and cereals. In epidemiologic studies, low folate intake and low circulating folate levels have consistently been associated with increased risk of colorectal adenoma and colorectal cancer (7–11).
Genes in the folate-mediated one-carbon metabolism pathway have been of interest as candidate genes for influencing colorectal adenoma risk (10, 12–15). Studies of these folate-related genes and folate intake may reveal gene–environment interactions and shed light on complex patterns of risk. There is expanding discussion of the possible importance of gene–environment interaction in chronic disease etiology, and how to assess efficiently whether interaction exists (12, 13, 15–18). Notably, several studies have found associations between colorectal adenoma risk and common polymorphisms in methylenetetrahydrofolate reductase (MTHFR; 12, 15) and thymidylate synthase (TYMS; 13, 14), along with evidence of gene–environment interaction; specifically some of these studies reported elevated risk for polymorphisms in these genes among individuals with low folate intake and colorectal adenoma (14, 15) as well as for colorectal cancer (19, 20). Recently, there have been several genome-wide association studies (GWAS) for colorectal cancer (21–23), but so far no GWAS have been reported for colorectal adenoma.
In this study, we investigated associations between advanced colorectal adenomas and 82 one-carbon metabolism genes, selected on the basis of the known biology of this pathway and an extensive literature search. We studied 1,331 cases and 1,501 controls from the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO), a large, multicenter randomized screening trial coordinated by the National Cancer Institute (NCI). We systematically looked for associations with each of 882 tag single nucleotide polymorphisms (SNP) in these genes in the entire population, and then examined associations by level of folate intake. To further investigate gene- and pathway-level associations both in the entire population and by the folate levels, we applied the Adaptive Rank Truncated Product (ARTP) method that combines association signals over a set of SNPs and genes while accounting for multiple testing.
Methods
Study population
The PLCO is a randomized screening trial in which men and women, ages 55 to 74 years, were enrolled from 10 different centers in the United States between 1993 and 2001 (24). All subjects included in this study had been assigned to the screening arm of the trial, completed a baseline risk factor questionnaire, provided a blood specimen at baseline, and consented to participate in etiologic studies of cancer and related diseases. In addition, subjects had completed at baseline a food frequency questionnaire, called the PLCO Dietary Questionnaire (DQx).
Cases were all those participants with a histologically confirmed advanced distal colorectal adenoma diagnosed at baseline sigmoidoscopy. Advanced colorectal adenomas were defined as adenomas more than 1 cm in size, with villous characteristics, and/or with high-grade dysplasia, and carcinomas in situ. Controls were randomly selected from those participants without distal colorectal polyps at baseline sigmoidoscopy, and frequency matched to cases on age at baseline, sex, and race. A total of 1,331 cases (850 men and 481 women) and 1,501 controls (950 men and 551 women) were included in this analysis.
A subset (n = 500 cases and 500 controls) was chosen for folate measurements in baseline plasma. During the PLCO randomization period (1993–2001), the United States had required that enriched breads, cereals, flours, corn meals, pastas, rice, and other grain products be fortified with folic acid, specifically by January 1, 1998 (25). Therefore, this subset of cases was randomly selected from those advanced adenoma cases with baseline blood draws either before December, 1997 or after January, 1998 so that plasma folate values would represent circulating levels either prior or subsequent to the mandatory folate fortification in 1998. The subset of controls was matched to the cases on date of baseline blood draw.
Dietary assessment
Folate intake and alcohol consumption were derived from the DQx, which asked about the typical frequency of intake during the past year of 137 food items, including beer, wine, and hard liquor, and typical portion sizes for 77 of the food items (24).
We calculated 7 measures of folate intake, in micrograms per day, all of which were based on the single DQx administered to each participant at the time of study entry. Two of the dietary folate estimates were based on USDA (U.S. Department of Agriculture) food composition databases: the first for the U.S. food supply prior to folate fortification in 1998 and the second for the U.S. food supply after folate fortification. The USDA databases relied on microbiologic folate assays to estimate the bioavailability of the folate naturally occurring in foods and the synthetic folic acid added in fortification. A third measure, of total folate, combined vitamin supplement information from the DQx with dietary folate estimates based on the postfortification USDA database.
Four University of Minnesota food composition databases from the Nutrition Data Systems for Research software (NDS-R) were also used; all were for the U.S. food supply after folate fortification in 1998. The first estimated natural folate in foods and the second estimated synthetic folate added to foods. The third and fourth combined natural and synthetic folate occurring in foods by bioavailability based on microbiologic assays and by weighting synthetic folate as 1.67 times more bioavailable than natural folate, respectively.
Unless otherwise specified, for all statistical analyses, we used dietary folate estimated with the USDA prefortification food composition database. Quartile cut points were based on the frequency distribution in the study controls. The quartiles of dietary folate intake were 0 to 262 μg/d (Q1), 263 to 359 μg/d (Q2), 360 to 466 μg/d (Q3), and more than 466 μg/d (Q4).
Laboratory assays
For the subset of 500 cases and 500 controls, heparin plasma was assayed for folate, using a Lactobacillus casei microbial assay, at the USDA Human Nutrition Research Center on Aging at Tufts University (26). Plasma collections from 4 subjects were used for quality control. Four anonymized quality control samples were randomly inserted into each of the 28 assay batches, with 2 of the quality control samples in each batch from the same individual. The total laboratory coefficient of variation (CV) for plasma folate, based on all 28 batches, averaged 14.5%, with individual CVs of 4.4%, 10.4%, 15.0%, and 18.0% for the 4 subjects. Quartiles of plasma folate were calculated based on the plasma folate distribution within the controls.
Gene selection and genotyping
A total of 82 genes related to one-carbon metabolism and regulation were genotyped at the NCI Core Genotyping Facility as part of a custom iSelect bead chip (Illumina Custom Infinium) that included over 29,000 SNPs with coverage of more than 1,300 genes postulated to be related to cancer (27). The one-carbon metabolism pathway genes included on the platform were selected through an extensive literature search and based on the known biology of this pathway.
Tag SNPs were selected for the 82 genes inclusive of the region 20 kb upstream and 10 kb downstream of each gene. We used the HapMap CEU, JPT, CHB, and YRI populations and the Carlson method, as implemented in Tagzilla with a r2 threshold of 0.8 and minor allele frequency (MAF) of 5% or more. Coverage of tag SNPs for each gene was calculated with Tagzilla; the median percentage coverage of loci needed to be tagged at a threshold of r2 = 0.8 was 79.62%. SNPs with known or putative functional significance (i.e., nonsynonymous, promoter, intron–exon splice sites) were also included whenever possible (27), for an initial total of 1,082 SNPs selected in one-carbon metabolism genes.
Of the tag SNPs originally selected, we excluded SNPs with a call rate less than 90%, in violation of Hardy–Weinberg equilibrium (P < 0.05/1,082), and/or with an MAF less than 5%. We also excluded SNPs on the X chromosome with a large number of haploid heterozygotes (>10% among men). After these exclusions, 882 SNPs remained for the 82 one-carbon metabolism genes and were used for this analysis.
Statistical analyses
Association between dietary folate and colorectal adenoma risk.
The association between dietary folate and advanced colorectal adenoma was assessed with multivariate logistic regression adjusting for age, gender, race, and study center. We used quartiles of folate as a continuous variable and a trend test with one degree of freedom.
SNP-level genetic association tests.
Genetic association tests for individual SNPs were conducted with multivariate logistic regression, adjusting for the same set of covariates described above and additionally including dietary folate using all subjects in the data. In addition, analyses stratified by folate intake were conducted by applying the association test to each SNP among subjects within each quartile of dietary folate. An additive genetic model and the corresponding one degree of freedom likelihood ratio test were used.
To correct for multiple comparisons involved in these genetic association tests, we used a permutation method that takes into account both the multiple SNPs and multiple strata used in the analyses. We simulated 5,000 null data sets by permuting case/control status; for each data set, we conducted 4 sets of association tests across quartiles of dietary folate. The minimum P value from each data set across the 882 SNPs and across the 4 strata was recorded, and a distribution of minimum P values was used as a reference distribution for assessing the statistical significance of our findings for individual SNPs.
Gene- and pathway-level genetic association tests.
To test our prior hypothesis that gene- and pathway-level associations existed within the lowest quartile of dietary folate, we applied the ARTP method that combines association signals over a set of SNPs while accounting for multiple testing (28). An underlying assumption of this method is that for each gene (or pathway) to which the test is applied, there may be more than one SNP (or gene) that contributes to disease risk. Thus, instead of using only the smallest P value, a summary statistic based on the top K smallest P values is used (K = 1, 2, 3, 4, and 5). Specifically, the product of the top K P values is used as the test statistic and its significance is assessed with permutations. The ARTP method is sensitive and powerful when there are a number of small association signals. Additional information is provided elsewhere (29, 30). We applied the ARTP method to each gene and obtained 82 gene-level P values; we then applied the ARTP method again to the 82 gene-level P values and obtained a single summary pathway-level P value. Gene- and pathway-level P values were calculated independently for each quartile of dietary folate, and do not take into account testing across the multiple strata.
Testing for gene–environment interaction.
To test formally for the statistical significance of interactions between a SNP and folate intake, we used a likelihood ratio test with 6 degrees of freedom based on the model using 4 categories of dietary folate and 3 categories for a given SNP (0, 1, or 2 copies of a minor allele). In addition, we conducted a one degree of freedom test treating both the quartiles of dietary folate and the number of copies of the minor allele as continuous variables.
Results
Characteristics of cases and controls are presented in Table 1. A moderate inverse association between dietary folate and advanced colorectal adenoma was observed [odds ratio (OR) = 0.85 per quartile, 95% confidence interval (95% CI) = 0.79–0.92, Ptrend = 1.92 × 10−05]. Compared with persons in the highest quartile of dietary folate, persons in the lowest quartile were at 50% increased risk of advanced colorectal adenoma (OR = 1.46, 95% CI = 1.17–1.82).
Characteristics of advanced colorectal adenoma cases and controls in PLCO genotyped for this study
Characteristic . | Cases . | % . | Controls . | % . | p-value . |
---|---|---|---|---|---|
No. of individuals | 1,331 | 1,501 | |||
Gender | 0.75 | ||||
Female | 481 | 36.1 | 551 | 36.7 | |
Male | 850 | 63.9 | 950 | 63.3 | |
Age at randomization (y) | 0.11 | ||||
55–59 | 395 | 29.7 | 509 | 33.9 | |
60–64 | 423 | 31.8 | 458 | 30.5 | |
65–69 | 330 | 24.8 | 340 | 22.7 | |
70–74 | 183 | 13.8 | 194 | 12.9 | |
Race | 0.13 | ||||
White non-Hispanic | 1,234 | 92.7 | 1,368 | 91.1 | |
Other | 97 | 7.3 | 133 | 8.9 | |
Family history of colorectal cancer | 0.01 | ||||
Yes | 169 | 13.2 | 147 | 10.2 | |
No | 1,331 | 86.8 | 1,298 | 89.8 | |
Alcohol consumption (g/d) | 0.06 | ||||
<1 | 486 | 40.1 | 580 | 41.5 | |
1–15 | 392 | 32.3 | 479 | 34.3 | |
15–30 | 131 | 10.8 | 158 | 11.3 | |
≥30 | 203 | 16.8 | 181 | 13.0 | |
Prefortification dietary folate (mcg/d) | 0.0002 | ||||
<400 | 822 | 67.8 | 850 | 60.8 | |
≥400 | 390 | 32.2 | 548 | 39.2 | |
Postfortification dietary folate (mcg/d) | <0.0001 | ||||
<400 | 604 | 49.8 | 587 | 42.0 | |
≥400 | 608 | 50.2 | 811 | 58.0 | |
Multivitamin use | |||||
Ever | 141 | 16.5 | 156 | 15.1 | 0.42 |
Never | 714 | 83.5 | 875 | 84.9 | |
Body mass index (kg/m2) | 0.08 | ||||
<18.5 | 8 | 0.6 | 8 | 0.5 | |
18.5–25 | 348 | 26.3 | 440 | 29.5 | |
25–30 | 597 | 45.2 | 683 | 45.8 | |
≥30 | 369 | 27.9 | 359 | 24.1 | |
Strenuous physical activity (h/wk) | |||||
0 | 225 | 18.6 | 182 | 13.1 | 0.0002 |
<1 | 217 | 18.0 | 231 | 16.6 | |
1 | 155 | 12.8 | 179 | 12.9 | |
2 | 184 | 15.2 | 218 | 15.7 | |
3 | 184 | 15.2 | 215 | 15.4 | |
≥4 | 243 | 20.1 | 368 | 26.4 | |
Smoking status | <0.0001 | ||||
Never | 510 | 38.4 | 718 | 47.8 | |
Former | 196 | 14.7 | 105 | 7.0 | |
Current | 624 | 46.9 | 678 | 45.2 |
Characteristic . | Cases . | % . | Controls . | % . | p-value . |
---|---|---|---|---|---|
No. of individuals | 1,331 | 1,501 | |||
Gender | 0.75 | ||||
Female | 481 | 36.1 | 551 | 36.7 | |
Male | 850 | 63.9 | 950 | 63.3 | |
Age at randomization (y) | 0.11 | ||||
55–59 | 395 | 29.7 | 509 | 33.9 | |
60–64 | 423 | 31.8 | 458 | 30.5 | |
65–69 | 330 | 24.8 | 340 | 22.7 | |
70–74 | 183 | 13.8 | 194 | 12.9 | |
Race | 0.13 | ||||
White non-Hispanic | 1,234 | 92.7 | 1,368 | 91.1 | |
Other | 97 | 7.3 | 133 | 8.9 | |
Family history of colorectal cancer | 0.01 | ||||
Yes | 169 | 13.2 | 147 | 10.2 | |
No | 1,331 | 86.8 | 1,298 | 89.8 | |
Alcohol consumption (g/d) | 0.06 | ||||
<1 | 486 | 40.1 | 580 | 41.5 | |
1–15 | 392 | 32.3 | 479 | 34.3 | |
15–30 | 131 | 10.8 | 158 | 11.3 | |
≥30 | 203 | 16.8 | 181 | 13.0 | |
Prefortification dietary folate (mcg/d) | 0.0002 | ||||
<400 | 822 | 67.8 | 850 | 60.8 | |
≥400 | 390 | 32.2 | 548 | 39.2 | |
Postfortification dietary folate (mcg/d) | <0.0001 | ||||
<400 | 604 | 49.8 | 587 | 42.0 | |
≥400 | 608 | 50.2 | 811 | 58.0 | |
Multivitamin use | |||||
Ever | 141 | 16.5 | 156 | 15.1 | 0.42 |
Never | 714 | 83.5 | 875 | 84.9 | |
Body mass index (kg/m2) | 0.08 | ||||
<18.5 | 8 | 0.6 | 8 | 0.5 | |
18.5–25 | 348 | 26.3 | 440 | 29.5 | |
25–30 | 597 | 45.2 | 683 | 45.8 | |
≥30 | 369 | 27.9 | 359 | 24.1 | |
Strenuous physical activity (h/wk) | |||||
0 | 225 | 18.6 | 182 | 13.1 | 0.0002 |
<1 | 217 | 18.0 | 231 | 16.6 | |
1 | 155 | 12.8 | 179 | 12.9 | |
2 | 184 | 15.2 | 218 | 15.7 | |
3 | 184 | 15.2 | 215 | 15.4 | |
≥4 | 243 | 20.1 | 368 | 26.4 | |
Smoking status | <0.0001 | ||||
Never | 510 | 38.4 | 718 | 47.8 | |
Former | 196 | 14.7 | 105 | 7.0 | |
Current | 624 | 46.9 | 678 | 45.2 |
NOTE: May not sum to total number of participants due to missing covariate values.
The top findings from association tests conducted for individual SNPs in the entire population are summarized in Table 2. The strongest association was for rs11627387 in methylenetetrahydrofolate dehydrogenase (MTHFD1; OR = 0.84 per allele, 95% CI = 0.75–0.94, nominal P = 0.0027, nonrisk/risk allele: A/G). However, this result, along with the other top findings in Table 2, did not remain statistically significant after adjusting for multiple testing using a permutation method (multiple-testing adjusted P = 0.34 for rs11627387). We then applied the ARTP method to all genotyped SNPs in each of the 82 one-carbon metabolism genes. The pathway-level P value for the one-carbon metabolism pathway, based on the 82 gene-level P values using ARTP methodology, was not statistically significant (P = 0.89). All of these analyses included all cases and controls, irrespective of folate intake.
Top findings from association tests in the entire population
Gene . | SNP . | Location . | Risk allele . | Nominal P . | OR (95% CI) . |
---|---|---|---|---|---|
MTHFD1 | rs11627387 | IVS26-955G>A | G | 0.0027 | 0.84 (0.75–0.94) |
GLS | rs12185688 | IVS14-3683T>C | T | 0.0033 | 0.79 (0.67–0.92) |
MTHFD1 | rs2230491 | Ex1-1015C>T | C | 0.0045 | 0.80 (0.69–0.93) |
UNG2 | rs231625 | -4520T>C | C | 0.0088 | 1.26 (1.06–1.50) |
TYMS | rs2298582 | IVS13+122A>C | A | 0.0090 | 0.80 (0.67–0.95) |
MTHFS | rs622506 | IVS2-1411T>G | G | 0.0121 | 1.15 (1.03–1.28) |
MTHFD1 | rs2281603 | IVS27-380T>C | T | 0.0138 | 0.85 (0.75–0.97) |
GART | rs9984077 | Ex11+195G>A | A | 0.0138 | 1.17 (1.03–1.34) |
UNG2 | rs231622 | IVS1+5C>T | T | 0.0141 | 1.16 (1.03–1.30) |
MTHFD2 | rs7587117 | IVS29-27A>G | G | 0.0146 | 1.15 (1.03–1.29) |
Gene . | SNP . | Location . | Risk allele . | Nominal P . | OR (95% CI) . |
---|---|---|---|---|---|
MTHFD1 | rs11627387 | IVS26-955G>A | G | 0.0027 | 0.84 (0.75–0.94) |
GLS | rs12185688 | IVS14-3683T>C | T | 0.0033 | 0.79 (0.67–0.92) |
MTHFD1 | rs2230491 | Ex1-1015C>T | C | 0.0045 | 0.80 (0.69–0.93) |
UNG2 | rs231625 | -4520T>C | C | 0.0088 | 1.26 (1.06–1.50) |
TYMS | rs2298582 | IVS13+122A>C | A | 0.0090 | 0.80 (0.67–0.95) |
MTHFS | rs622506 | IVS2-1411T>G | G | 0.0121 | 1.15 (1.03–1.28) |
MTHFD1 | rs2281603 | IVS27-380T>C | T | 0.0138 | 0.85 (0.75–0.97) |
GART | rs9984077 | Ex11+195G>A | A | 0.0138 | 1.17 (1.03–1.34) |
UNG2 | rs231622 | IVS1+5C>T | T | 0.0141 | 1.16 (1.03–1.30) |
MTHFD2 | rs7587117 | IVS29-27A>G | G | 0.0146 | 1.15 (1.03–1.29) |
We then conducted a stratified analysis of individual SNP associations within quartiles of dietary folate (Fig. 1). Consistent with our hypothesis that one-carbon metabolism gene signals might be substantially stronger in persons with low folate intake, the smallest P value was observed with rs244072 (MAF = 10%) in the adenosine deaminase gene (ADA; nominal P = 2.37 × 10−5, OR = 2.33 per allele, 95% CI = 1.60–3.50; nonrisk/risk allele: T/C). For this SNP, the OR per allele decreased to 1.1 to 1.2 in the intermediate quartiles and to 1.0 in the highest quartile of folate intake (Table 3). The tests for interaction between rs244072 and quartiles of dietary folate had P of 0.029 (6 df test) and 0.01 (1 df test). The next top 4 SNPs from the stratified analyses were in the folate receptor 1 gene (FOLR1), TYMS, cysteine dioxygenase gene (CDO1), and methionine adenosyltransferase II, β gene (MAT2B; Table 3). Interestingly, as with rs244072 in ADA, the SNPs in FOLR1 and TYMS also showed notable associations with risk only in the lowest quartile of dietary folate, and the SNP in CDO1 showed notable associations with risk only in the lowest 2 quartiles.
A, association tests between SNPs and advanced colorectal adenoma risk stratified by dietary folate. Each quartile is a different color. For example, the dots in red represent nominal P values for each individual one-carbon metabolism SNP for subjects in the lowest quartile of folate intake (Q1), and the dots in blue represent nominal P values for each individual SNP for subjects in the highest quartile of folate intake (Q4). B, gene-level P values using the ARTP method in lowest folate quartile. C, gene-level P values using the ARTP method in quartile 2. D, gene-level P values using the ARTP method in quartile 3. E, gene-level P values using the ARTP method in quartile 4. Each dot represents a P value (y-axis) for each gene, as listed along the x-axis.
A, association tests between SNPs and advanced colorectal adenoma risk stratified by dietary folate. Each quartile is a different color. For example, the dots in red represent nominal P values for each individual one-carbon metabolism SNP for subjects in the lowest quartile of folate intake (Q1), and the dots in blue represent nominal P values for each individual SNP for subjects in the highest quartile of folate intake (Q4). B, gene-level P values using the ARTP method in lowest folate quartile. C, gene-level P values using the ARTP method in quartile 2. D, gene-level P values using the ARTP method in quartile 3. E, gene-level P values using the ARTP method in quartile 4. Each dot represents a P value (y-axis) for each gene, as listed along the x-axis.
Top findings from association tests in each quartile of prefortification dietary folate
Gene . | SNP . | Location . | Risk allele . | Dietary folate quartile . | Nominal Pa . | OR per allele . | 95% CI . |
---|---|---|---|---|---|---|---|
ADA | rs244072 | IVS2-1496T>C | C | 1 | 2.37E-05b | 2.33 | 1.60–3.50 |
2 | 0.7194 | 1.07 | 0.75–1.50 | ||||
3 | 0.4952 | 1.15 | 0.77–1.70 | ||||
4 | 0.9398 | 0.99 | 0.68–1.40 | ||||
FOLR1 | rs10501409 | 7502T>G | T | 1 | 0.0010 | 1.77 | 1.25–2.50 |
2 | 0.8014 | 1.05 | 0.74–1.48 | ||||
3 | 0.8105 | 0.96 | 0.72–1.52 | ||||
4 | 0.3134 | 1.22 | 0.83–1.81 | ||||
TYMS | rs11664283 | −6774A>G | A | 1 | 0.0016 | 1.45 | 1.15–1.85 |
2 | 0.0598 | 0.79 | 0.63–1.01 | ||||
3 | 0.3690 | 1.12 | 0.91–1.45 | ||||
4 | 0.9589 | 1.01 | 0.77–1.30 | ||||
CDO1 | rs34869 | −599C>G | C | 1 | 0.0018 | 1.41 | 1.14–1.75 |
2 | 0.0177 | 1.31 | 1.05–1.64 | ||||
3 | 0.1311 | 1.20 | 0.66–1.06 | ||||
4 | 0.4069 | 0.90 | 0.71–1.15 | ||||
MAT2B | rs4869087 | IVS2-1783C>A | C | 1 | 0.7900 | 1.03 | 0.81–1.30 |
2 | 0.1579 | 1.18 | 0.94–1.50 | ||||
3 | 0.0020 | 0.65 | 0.49–0.86 | ||||
4 | 0.2827 | 1.16 | 0.89–1.50 |
Gene . | SNP . | Location . | Risk allele . | Dietary folate quartile . | Nominal Pa . | OR per allele . | 95% CI . |
---|---|---|---|---|---|---|---|
ADA | rs244072 | IVS2-1496T>C | C | 1 | 2.37E-05b | 2.33 | 1.60–3.50 |
2 | 0.7194 | 1.07 | 0.75–1.50 | ||||
3 | 0.4952 | 1.15 | 0.77–1.70 | ||||
4 | 0.9398 | 0.99 | 0.68–1.40 | ||||
FOLR1 | rs10501409 | 7502T>G | T | 1 | 0.0010 | 1.77 | 1.25–2.50 |
2 | 0.8014 | 1.05 | 0.74–1.48 | ||||
3 | 0.8105 | 0.96 | 0.72–1.52 | ||||
4 | 0.3134 | 1.22 | 0.83–1.81 | ||||
TYMS | rs11664283 | −6774A>G | A | 1 | 0.0016 | 1.45 | 1.15–1.85 |
2 | 0.0598 | 0.79 | 0.63–1.01 | ||||
3 | 0.3690 | 1.12 | 0.91–1.45 | ||||
4 | 0.9589 | 1.01 | 0.77–1.30 | ||||
CDO1 | rs34869 | −599C>G | C | 1 | 0.0018 | 1.41 | 1.14–1.75 |
2 | 0.0177 | 1.31 | 1.05–1.64 | ||||
3 | 0.1311 | 1.20 | 0.66–1.06 | ||||
4 | 0.4069 | 0.90 | 0.71–1.15 | ||||
MAT2B | rs4869087 | IVS2-1783C>A | C | 1 | 0.7900 | 1.03 | 0.81–1.30 |
2 | 0.1579 | 1.18 | 0.94–1.50 | ||||
3 | 0.0020 | 0.65 | 0.49–0.86 | ||||
4 | 0.2827 | 1.16 | 0.89–1.50 |
NOTE: P value that is the most significant in each SNP is marked bold.
aNominal P before adjustment for multiple comparisons.
bP value that is statistically significant (P < 0.05) after adjustment for multiple testing.
To correct for multiple comparisons involved in the above stratified association tests for individual SNPs, we used a permutation method that takes into account both the multiple SNPs and multiple strata used in the analyses. The simulated 5% significance threshold for our analysis was at P = 6.7 × 10−5. Only rs244072 in ADA was found to remain statistically significant in the lowest quartile of folate intake, with an adjusted P of 0.019 (Fig. 1B). After correction, none of the other 881 SNPs were statistically significantly associated with risk in any of the quartiles of folate intake.
The ARTP method was then used to assess gene- and pathway-level significance in the lowest dietary folate quartile by combining association signals from SNPs within each of the 82 one-carbon metabolism genes. The pathway-level P value was statistically significant among subjects in the lowest quartile of folate intake (pathway-level P = 0.038). The genes contributing to this significance included ADA (gene-level P = 0.001), CDO1 (gene-level P = 0.0073), and FOLR1 (gene-level P = 0.0076) with gene-level P values smaller than 0.01. Other notable genes with gene-level P values smaller than 0.05 were: serine hydroxymethyltransferase (SHMT1; gene-level P = 0.031), cysteine sulfinic acid decarboxylase (CSAD; gene-level P = 0.0327), and TYMS (gene-level P = 0.042). In applying the ARTP method in each of the other quartiles of dietary folate, results did not reveal any statistically significant gene- or pathway-level P values (Figs. 1C–E).
To investigate the sensitivity of the association of rs244072 in ADA—which was the only statistically significant SNP after multiple testing correction, we repeated the analysis using 6 additional measures of folate intake. The results were consistent with USDA prefortification folate (Table 4); for each measure of folate intake, statistically significant associations for rs244072 (ADA) were seen only in the lowest quartile of intake; the nominal P values ranged from 6.45 × 10−5 to 0.015 and per-allele ORs ranged from 1.55 to 2.12. When using plasma folate levels, measured with a microbiologic assay for a subset of 500 cases and 500 controls, the OR for rs244072 was elevated among persons in the lowest 3 quartiles (per-allele OR = 1.41–1.87; Table 4). However, none of the associations attained statistical significance, possibly because of the reduced sample size. In addition, we also considered different cut points to define low folate intake for the USDA prefortification folate variable. Stratifying at the 20th, 15th, 10th, and 5th percentiles of the prefortification estimate of folate intake, we found ORs were consistently increased in the low folate subpopulation. Interestingly, the per-allele OR modestly increased as the definition of low folate became more stringent; the OR per allele increased from 2.4 for individuals in the lowest 20% of folate intake to 3.0 for individuals in the lowest 5%.
Associations between rs244072 (ADA) and advanced colorectal adenoma risk in subjects in the lowest folate quartile, using different folate measures
. | . | Per-allele OR (Nominal P) . | |||
---|---|---|---|---|---|
Food composition database . | Folate measure . | Quartile 1a 356/350b . | Quartile 2 301/349 . | Quartile 3 281/349 . | Quartile 4 274/350 . |
USDA | Prefortification dietary folate | 2.33 (2.37E-05) | 1.07 (0.72) | 1.15 (0.50) | 0.99 (0.94) |
Postfortification dietary folate | 2.12 (9.28E-05) | 1.07 (0.71) | 1.01 (0.92) | 1.06 (0.74) | |
Total folate from diet (post-fortification) and supplements | 1.61 (8.90E-03) | 1.17 (0.41) | 0.93 (0.73) | 1.28 (0.21) | |
University of Minnesota Nutrition Data Systems for Research | Postfortification dietary folatec | 1.88 (6.80E-04) | 1.1 (0.61) | 1.02 (0.90) | 1.11 (0.58) |
Postfortification dietary folated | 1.89 (5.70E-04) | 1.09 (0.65) | 0.88 (0.57) | 1.18 (0.36) | |
Natural folatee | 2.16 (6.45E-05) | 0.85 (0.40) | 1.26 (0.20) | 1.04 (0.81) | |
Synthetic folatef | 1.55 (1.53E-02) | 1.2 (0.33) | 0.93 (0.74) | 1.23 (0.25) | |
N/A | Plasma folate | 1.68 (0.14) | 1.87 (0.13) | 1.41 (0.29) | 0.93 (0.84) |
. | . | Per-allele OR (Nominal P) . | |||
---|---|---|---|---|---|
Food composition database . | Folate measure . | Quartile 1a 356/350b . | Quartile 2 301/349 . | Quartile 3 281/349 . | Quartile 4 274/350 . |
USDA | Prefortification dietary folate | 2.33 (2.37E-05) | 1.07 (0.72) | 1.15 (0.50) | 0.99 (0.94) |
Postfortification dietary folate | 2.12 (9.28E-05) | 1.07 (0.71) | 1.01 (0.92) | 1.06 (0.74) | |
Total folate from diet (post-fortification) and supplements | 1.61 (8.90E-03) | 1.17 (0.41) | 0.93 (0.73) | 1.28 (0.21) | |
University of Minnesota Nutrition Data Systems for Research | Postfortification dietary folatec | 1.88 (6.80E-04) | 1.1 (0.61) | 1.02 (0.90) | 1.11 (0.58) |
Postfortification dietary folated | 1.89 (5.70E-04) | 1.09 (0.65) | 0.88 (0.57) | 1.18 (0.36) | |
Natural folatee | 2.16 (6.45E-05) | 0.85 (0.40) | 1.26 (0.20) | 1.04 (0.81) | |
Synthetic folatef | 1.55 (1.53E-02) | 1.2 (0.33) | 0.93 (0.74) | 1.23 (0.25) | |
N/A | Plasma folate | 1.68 (0.14) | 1.87 (0.13) | 1.41 (0.29) | 0.93 (0.84) |
aQuartiles were calculated based on the distribution of the folate variable among the controls.
bThe number of cases and controls.
cNatural + synthetic, based on microbiologic folate assays.
dNatural + synthetic, based on estimated bioavailability.
eFood folate from diet.
fFolic acid from diet (postfortification).
To assess the biological consistency of the association of rs244072 with advanced colorectal adenoma risk, we considered its interaction with alcohol intake. High alcohol consumption is thought to modify folate homeostasis by interfering with the intestinal absorption and hepatic uptake of folate and thereby influencing folate bioavailability (31). We conducted an exploratory analysis of rs244072 (ADA) within quartiles of alcohol intake (g/d). Interestingly, similar to dietary folate, a statistically significant effect was observed only in the highest risk group, specifically those subjects in the highest quartile of alcohol consumption (nominal P = 0.0009, OR = 1.87 per allele, 95% CI = 1.28–2.73). Two-way stratified analyses by tertiles of folate intake and tertiles of alcohol intake showed that those in the highest risk group (i.e., with the lowest folate intake and highest alcohol intake) were at significantly increased risk of adenoma if they carried the C allele of rs244072 (nominal P = 0.0014, OR = 2.63 per allele, 95% CI = 1.41–4.92), while those in the lowest risk group (i.e., with the highest folate intake and lowest alcohol intake) were not at increased risk (nominal P = 0.82, OR = 0.93 per allele, 95% CI = 0.52–1.67).
Discussion
Our study found a statistically significant pathway-level association between common variants in one-carbon metabolism genes and advanced colorectal adenoma risk in persons with low folate intake (pathway-level P = 0.04). Of the 82 one-carbon metabolism genes explored, the ones contributing to this pathway-level association included ADA, CDO1, and FOLR1 (gene-level P values of 0.001, 0.007, and 0.008, respectively). We found no statistically significant associations at the SNP-, gene-, or pathway-level in quartiles 2 through 4 of folate intake.
Several studies previously reported interactions between polymorphisms in one-carbon metabolism genes and folate intake. For example, evidence for an interaction was reported between the 677C>T polymorphism in the MTHFR gene and folate by Ulrich and colleagues (15), in which they found persons with the TT genotype in the lowest tertile of folate, vitamin B12, or vitamin B6 intake were at increased risk for colorectal adenoma. However, in another study by Chen and colleagues (32), no statistically significant interaction between this polymorphism and folate intake was detected with respect to colorectal adenoma risk. Interactions between 3 TYMS polymorphisms and folate intake have also been reported in the literature, with reduced colorectal adenoma risk observed among carriers of variant alleles who had high folate levels (33).
Our data included both of these genes. Analysis of the 12 SNPs in MTHFR, including the 677C>T polymorphism (rs1801133), in our study did not reveal any statistically significant SNP- or gene-level associations in the entire population or in any folate quartile under an additive genetic model. Previous reports suggested that MTHFR 677C>T has a recessive effect in subjects with low folate intake, that is, the TT genotype is associated with elevated risk compared with the CT and CC genotypes (15). Hence, we analyzed this SNP under a recessive model. The association test was not statistically significant in the entire population (nominal P > 0.2); however, stratified analysis revealed increased risk for the TT genotype in each of the lowest 2 quartiles of folate (OR = 1.45, P = 0.13 and OR = 1.72, P = 0.02, for the lowest and the second lowest quartiles of folate, respectively). In contrast, the risk was not elevated for the TT genotype in the second highest and highest quartiles of folate intake. This finding is consistent with the previous report by Ulrich and colleagues (15).
In contrast, our results do not support the previous report indicating that TYMS polymorphisms modify risk of colorectal adenoma at high folate intake. In our study, TYMS showed a gene-level P value of 0.042, but only in the lowest quartile of dietary folate intake; no associations were found in other quartiles (gene-level P > 0.4).
Therefore, although our results implicate the one-carbon metabolism pathway, further studies are needed for replication and elucidation of the magnitude and direction of genetic effects for specific SNPs and their interactions with folate intake. Our statistical approach focused on an additive genetic model for testing associations between SNPs and adenoma risk, which is known to be robust under several genetic models. However, it is possible that any association signal from genetic variants under a recessive model could have been missing from our top findings. Supplementary Table S1 shows P values of all the SNPs analyzed in our data, both in the entire population and by levels of folate.
We reported that the only statistically significant SNP-level association was for rs244072 in ADA in the lowest folate quartile (nominal P = 2.37 × 10−5, adjusted P = 0.019), while no associations for this SNP were observed in the other quartiles. The functional relevance of SNP rs244072 is unclear, and the association could be due to other SNPs in linkage disequilibrium that have functional significance. ADA is an enzyme involved in DNA metabolism, specifically purine metabolism, and in the salvage pathway activity of cancerous tissues and cells (34, 35). The role of ADA in the one-carbon metabolism pathway is the deamination of adenosine produced through the hydrolysis of S-adenosylhomocysteine into adenosine and homocysteine (36). A fully functional ADA, facilitating the irreversible degradation and removal of adenosine, may be necessary in the setting of increased purine and pyrimidine metabolism in carcinogenesis. Previous studies have observed higher levels of ADA activity in cancerous colorectal tissue, as compared with normal adjacent tissue from the same subjects, and in cancerous bladder tissue, as compared with normal adjacent tissue from the same subjects and from cancer-free subjects (34, 37). Deficiencies in ADA have also been associated with a type of severe combined immunodeficiency disease (SCID; ref. 38).
The other top gene-level associations were observed with the folate receptor gene (FOLR1; gene-level P = 0.007) and the CDO1 (gene-level P = 0.008) in the lowest quartile of folate intake. FOLR1 may be involved in cellular proliferation (39), and its gene product is a useful prognostic marker for certain types of cancer (40, 41). CDO1 catalyzes the production of cysteine sulfinate from cysteine, which is metabolized from homocysteine and serine. Overexpression of CDO1 is associated with Sézary syndrome, a type of T-cell lymphoma (39).
Our data included 7 measures of folate intake derived from a single food frequency questionnaire administered to each participant at the time of study entry based on different food composition databases. Our findings for the ADA gene were consistent using each measure, including an integrated measure of dietary and supplemental folate. Furthermore, in our control group, the dietary folate intake estimates were highly correlated (Pearson r ranging from 0.78 to 0.95). Therefore, our results do not seem to be sensitive to the estimates of folate intake.
Our study has several strengths. First, our study included a substantially larger number of cases and controls than previous studies of one-carbon metabolism and colorectal adenoma. Second, our analyses considered advanced colorectal adenomas because advanced adenomas are more likely to progress to colorectal cancer than other adenomas, and hence these lesions are more relevant to colorectal cancer etiology than adenomas in general. Third, although previous studies of one-carbon metabolism genes and colorectal adenomas or cancers focused mostly on individual genes in the pathway, to our knowledge, our study is the first report to systematically investigate the pathway-level association between one-carbon metabolism genes and advanced colorectal adenoma risk. Fourth, we used permutation methods to control the false positive rate accounting for multiple testing of SNPs and the ARTP method to access significance at the gene- and pathway-level by combining small association signals into a single test.
Despite the strengths described above, this study has several limitations. First, only a subset of participants had plasma folate measured at baseline, which is a more accurate measure of folate bioavailability than dietary intake. However, the biologic interaction of one-carbon metabolism genes could be primarily with dietary folate and help determine the levels of circulating folate. Second, our results are for distal advanced adenomas only. In PLCO, only sigmoidoscopy was used for screening at baseline, and we did not consider the presence or absence of proximal adenomas in the cases or controls. Distal and proximal adenomas have been shown to vary in their tumor initiation and growth rates, which may represent somewhat different pathogenic processes (40, 41). However, we know of no evidence that the role of common genetic variation in the etiology of colorectal cancer is restricted to either distal or proximal tumors. Third, this study may not be powered to detect some important gene–environment interactions because the required sample size for detecting interactions is much larger than for detecting main effects. We therefore focused on stratified analyses by folate intake and systematically applied interaction tests to our top findings.
To our knowledge, no associations between one-carbon metabolism genes and colorectal cancer have been reported in GWAS. Given our results, this is not surprising as the primary hypothesis of GWAS is that susceptible SNPs can be detected on the basis of main effects of SNPs. Hence, SNPs that interact with environmental exposures may be overlooked. In contrast, we detected associations using stratified analysis by folate intake. Our results suggest that the development and application of more powerful tests to detect interactions in GWAS studies may be very useful. These tests could be based on versions of the methods developed here.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
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.