Folate Intake, MTHFR Polymorphisms, and the Risk of Colorectal Cancer: A Systematic Review and Meta-Analysis

Background. The objective was to determine whether relationships exist between the methylene-tetrahydrofolate reductase (MTHFR) polymorphisms and risk of colorectal cancer (CRC) and examine whether the risk is modified by level of folate intake. Methods. MEDLINE, Embase, and SCOPUS were searched to May 2012 using the terms “folic acid,” “folate,” “colorectal cancer,” “methylenetetrahydrofolate reductase,” “MTHFR.” Observational studies were included which (1) assessed the risk of CRC for each polymorphism and/or (2) had defined levels of folate intake for each polymorphism and assessed the risk of CRC. Results. From 910 references, 67 studies met our criteria; hand searching yielded 10 studies. The summary risk estimate comparing the 677CT versus CC genotype was 1.02 (95% CI 0.95–1.10) and for 677TT versus CC was 0.88 (95% CI 0.80–0.96) both with heterogeneity. The summary risk estimates for A1298C polymorphisms suggested no reduced risk. The summary risk estimate for high versus low total folate for the 677CC genotype was 0.70 (95% CI 0.56–0.89) and the 677TT genotype 0.63 (95% CI 0.41–0.97). Conclusion. These results suggest that the 677TT genotype is associated with a reduced risk of developing CRC, under conditions of high total folate intake, and this associated risk remains reduced for both MTHFR 677 CC and TT genotypes.


Introduction
Worldwide, colorectal cancer (CRC) is the third most frequently diagnosed cancer in males and the second in females [1]. Australia and New Zealand, Europe and North America have the highest incidence rates of CRC worldwide, and Africa and South-Central Asia, the lowest [1,2]. Over 75% of CRCs occur sporadically, with only 25% of patients having a family history of CRC [3].
Folate insufficiency has been suggested as one of the possible mechanism for CRC development and progression. DNA strand breaks, impaired DNA methylation and repair have been associated with folate deficiency and CRC [4][5][6][7]. There are many enzymes involved with folates and onecarbon metabolism; however, the methylene tetrahydrofolate reductase (MTHFR) enzyme is a key enzyme responsible for determining whether reduced folates are directed towards DNA methylation pathways or pyrimidine or purine synthesis [8]. In 1995, a variant of MTHFR enzyme was identified which causes a substitution of C to T at nucleotide 677 [9]. The MTHFR C677T homozygous variant (TT genotype) is thermolabile, and its activity is reduced by 70% compared to the wild type (CC genotype) [10]. This reduced enzyme activity causes an accumulation of plasma homocysteine and higher rates of thymidylate synthesis [10,11].
The distribution of the TT genotype varies from country to country. In Europe, there would appear to be a northsouth gradient with the distribution of the TT genotype lowest in the north [12,13] while in Asia, the frequency is highest in China and lowest in India [12,[14][15][16][17][18]. In North America, African Americans have a much lower TT genotype frequency versus Caucasians [19]. Individuals with this variant are thought to be at greater risk for a number of diseases including cardiovascular disease, acute lymphocytic leukemia, and neural tube defects [10]. Some published studies have suggested that those with the TT genotype have a reduced risk of CRC versus those with the CC genotype (wild type) [20][21][22][23][24][25][26][27][28]; however, other studies have found an increased risk [29][30][31].
A second variant of the MTHFR enzyme, with a substitution of A to C at nucleotide 1298, has also been identified. Unlike the MTHFR C677T polymorphism, the enzyme activities of the variants of MTHFR A1298C polymorphism are not thermolabile, but the enzyme activity is reduced by approximately 40% of the wild type (AA genotype) in the variant genotype. Altered homocysteine levels have not been found in individuals with these variants [32]. The prevalence of the 1298CC genotype varies, with the homozygous genotype found in 7-12% of Caucasians, in Europeans, 4-12%, while in China, Japan, and Hawaiian studies the prevalence ranged between 1 and 4% [32,33].
The objective of this effort was to conduct a systematic review and meta-analysis of the published data to determine whether relationships exist between the various MTHFR polymorphisms and the incidence of CRC. A secondary objective was to examine whether there exists a relationship between the level of folate intake for each MTHFR genotype and the risk of CRC.

Inclusion Criteria.
We selected observational studies reporting on the polymorphisms of the MTHFR C677T and/or A1298C genes and the associated risk of CRC, colon, or rectal cancer in adult populations. Studies were also included if they reported on folate exposure (dietary or total) with at least two levels of folate intake and the associated rates of colorectal, colon and/or rectal cancer by genotype. Studies were excluded if they did not provide the information necessary to determine an odds ratio and 95% confidence interval for each genotype. No restrictions were placed on language of publication or country of study.

Search Strategy. The databases MEDLINE, Embase, and
Scopus on the OVID platform were searched from inception to May 2012. Both database-specific subject headings and text words were searched using the terms "folic acid" OR "folate" and "colorectal cancer" and "colorectal neoplasms" AND "methylenetetrahydrofolate reductase or MTHFR or C667T" limiting the results to humans only. The results of the search in each of the three databases were placed in a bibliography tool, and, in order to ensure blinding, an extract of author, title, and year information was made and uploaded into a spreadsheet for the purposes of title review. Title review was conducted by one reviewer (DAK) blinded to the journal of publication, place of research, and results, to determine which study articles to retrieve. The methods section of the selected journal articles were retrieved by other team members (MS and IM) not responsible for reviewing the journal articles. The method sections were reviewed by two independent reviewers (DAK, SJS) blinded to the journal of publication, place of research, and results as to their meeting the inclusion criteria. In case of disagreement between the two reviewers, a third reviewer served as a tiebreaker (GK). Previous reviews were also hand searched to identify other relevant publications to include.

Data Extraction.
Data extraction was carried out by one reviewer and independently checked for accuracy by a second reviewer. Data collected included the type of study, location, study inclusion and exclusion criteria, case and comparator group size, folate intake levels, odds ratio or risk ratio, the number, for both case and control, and percentage frequency of each genotype, relevant adjustments, and conclusions. The genotype distribution of the control group was evaluated for agreement with the Hardy-Weinberg equilibrium (HWE) using chi-squared with a significant level of 0.05 and the results incorporated into Table 1.
The Downs and Black scoring instrument was used to determine the quality of the studies included in this paper. The Down and Black scoring tool provides a means to assess the quality of a study based on 5 subscales (1) reporting of the study results, (2) external validity for the purposes of assessing generalizability of the findings, (3) bias in measurement and outcomes, (4) bias in the selection of study subjects, and (5) the power of the study [79]. The score was independently calculated for each study by two team members. Disagreements were resolved by consensus. The last question on the Downs and Black tool relates to the power of the study. If a priori power calculation was reported in the paper, this question was scored with a one, otherwise, zero was scored.

Statistical Analysis.
The meta-analysis for the genotype risk comparisons was performed using the inverse variance method under a random effects model, odds ratios (ORs) along with 95% confidence intervals (CIs) were used for the case control studies according to the DerSimonian and Laird method [80]. All identified studies with available data were included in the summary effect estimate for each genotype combination. For the meta-analysis of the risk of CRC associated with genotype, the wild type (677CC or 1298AA) was used as the reference group, and comparisons were made to either the heterozygous (677CT or 1298AC) or homozygous variant type (677TT or 1298CC). If studies grouped genotypes together for comparison purposes, or did not report ORs and 95% confidence intervals and the raw numbers were available in the paper, unadjusted ORs and associated 95% confidence intervals were calculated according to the method described by Silman and MacFarlane [81]. These are identified in Table 1 as "OR calculated, no adjustments" in the column titled Adjustments. The meta-analyses were performed using Review Manager 5.1 Software [82].
The meta-analyses for the comparison of high versus low folate intake and the associated risk of CRC were performed using the inverse variance method under a random effects  [80]. All identified studies with available data were included in the summary effect estimate for each high versus low folate intake within a genotype. For those studies that compared folate intake by "quantile" and assessed the risk of CRC by genotype, many used the 677CC or 677CC/CT lowest folate intake quantile as the reference group to determine the OR for all genotypes and folate intake levels. For the purposes of this analysis, however, the desire was to compare the risk of CRC between the highest folate intake to lowest folate intake within a genotype. The method described by Hamling et al. and the associated MS Excel spreadsheet, which recalculates the adjusted odds ratios permitting alternative comparisons, were used to derive the ORs of highest compared to the lowest folate intake within the genotype [83,84]. This analysis was performed using Microsoft Excel (Microsoft Corporation (2007), Redmond, WA, USA). An analysis of folate intake and CRC risk for the MTHFR A1298C gene was not possible due to an insufficient number of studies reporting on this data. In performing this analysis, the result from the highest "quantile" identified in the study was used to compare the lowest "quantile" in the study. Dietary folate intake for the lowest "quantile" ranged from a low of less than 115.6 to 406 mcg/day; the range for the highest was from 320 to 485 mcg/day or more. Although these ranges do overlap, they represent the highest and the lowest folate intake for the study population upon which the specific study odds ratios were derived. The meta-analyses were performed using Review Manager 5.1 Software [82]. Publication bias was assessed via the Begg and Mazumdar's rank correlation test, Egger's linear regression, and the Trim and Fill methods [85][86][87]. The assessment of publication bias was performed using the Comprehensive Metaanalysis (CMA) software (Biostat, Version 2.2, Englewood, NJ, USA) [88]. Summary effect estimates from CMA were compared with the RevMan results to ensure that they were both in agreement prior to executing the tests for publication bias.
Assessment of heterogeneity was performed using both Cochran's χ 2 and I 2 . The Cochran's χ 2 test assesses whether the differences in results are due to chance only [89]. Heterogeneity exists when the P value is low, that is, P < 0.10 [89]. The I 2 statistic is the percentage of variability in the effect estimates that is due to heterogeneity rather than chance. An I 2 statistic value over 50% indicates that substantial heterogeneity may be present [89]. The analysis was performed using Review Manager 5.1 software [82].
Kruskal-Wallis was performed on the quality of the studies to determine whether there were differences in the quality of the studies based on the directionality of the outcome. For the purposes of this analysis, directionality was assessed as positive (statistical significant OR > 1), neutral (nonsignificant OR), or negative (statistical significant OR ≤ 1). IBM's SPSS for Windows version 17 was used for the analysis (IBM SPSS, Version 17, Chicago, IL, USA).
The Forest plots of the MTHFR C677T and A1298C (Figures 2 through 5) were sorted according to the percentage of the comparator genotype (either 677CT, 677TT, 1298AC, or 1298CC) in the control group, from highest to lowest, while the remaining Forest plots ( Figure 6) were organized by increasing year of publication.

Results
The pooled search resulted in 910 records. Of these 67 met our inclusion criteria, 10 studies were found on hand searching ( Figure 1). Four identified studies were not included in the paper. In two studies, newborns comprised either all or part of the control group, which suggested that these studies were related to the determination of the prevalence of genotypes rather than risk of CRC since few newborns have had the opportunity to develop colorectal cancer [8,92]. The remaining two studies did not report the separate case control numbers for each genotype; therefore, ORs could not be calculated for all genotypes; however the folate intake results, reported on in one of these studies, are included in the high versus low folate intake analysis [31,93]. The majority of the studies included in the systematic review and meta-analysis were case control or nested case control studies, two cohort studies were identified (  [94,95]. All case control studies, with available data, were included in the meta-analysis, regardless of the quality score. Study results were reported from twenty-five countries: Asia (China, India, Japan, South Korea, Taiwan, and Thailand), Australia, Europe (EPIC Cohort (10 European Centers), Czech Republic, Croatia, France, Germany, Hungary, Italy, Norway, Poland, Romania, Spain, Sweden, and United Kingdom), Latin America (Mexico), Middle East (Egypt, Iran, and Turkey), South America (Brazil), and USA. Six papers were written in another language with an English abstract: five in Chinese: the other in Spanish [31,40,41,53,62,93]. When duplicate studies were found, for example, Nurses' Health study and Health Professionals study, only the most recently published results were used in this analysis. There were five studies whose recruitment period was during the early days of folate fortification in USA; otherwise none of the studies were conducted in an environment of food fortification [35,54,76]. A blood sample was the most often used medium to assess genotype. There were two studies that used buccal samples as the tissue source for genotyping [18,60].

Sensitivity Analysis.
In an attempt to identify the studies contributing to the heterogeneity in the genotype summary risk effect results, sensitivity analysis was performed according the sequential algorithm proposed by Patsopoulos and colleagues [96]. This method involves sequentially dropping one study from the meta-analysis to determine the impact on the I 2 statistic with the objective of identifying the study or studies that will reduce the I 2 below a set threshold.
Using this method, we were not successful in reducing the heterogeneity below the threshold value of an I 2 value of less than 25%, which would have suggested that there was minimal heterogeneity in the results. Given that the typical diets of Asian cultures can be substantially different from that of Europe and North America, separate analyses were conducted including just the studies in the Asian locations (China, India, Japan, South Korea, and Taiwan), separate from the European locations (Czech Republic, Croatia, European EPIC study, France, Germany, Hungary, Italy, Norway, Poland, Romania, Spain, Sweden, and United Kingdom), USA, and Middle East (Egypt, Iran, and Turkey) ( Table 2). The protective effect of the 677TT genotype was sustained in each geography; however, only in the USA was the risk reduction significant with no heterogeneity.
A further analysis was performed by comparing the results based on the source of controls: either hospital patients or healthy persons. The heterogeneity was sustained ( Table 2).

Publication Bias.
Publication bias was assessed using three different tests: Begg and Mazumdar's rank correlation test, Egger's linear regression, and the Trim and Fill methods. For the MTHFR 677CT genotype there may be some evidence for publication bias. The Begg and Mazumdar test returned a P value = 0.03, Egger's a P value = 0.005, and Trim and Fill found that an additional 12 studies would be necessary to form a symmetrical funnel plot. Whereas, for the MTHFR 677TT genotype, the Begg and Mazumdar test returned a P value = 0.33, Egger's a P value = 0.38, and Trim and Fill found that additional 4 studies would be necessary to form a symmetrical funnel plot, suggesting that publication bias may not be significant concern.

Correlation between Study Quality versus Results.
There was no statistically significant difference found in the quality of the studies based on outcome (positive versus neutral versus negative) (P = 0.310).

Sensitivity Analysis.
In an attempt to identify the studies contributing to the heterogeneity in the genotype summary risk effect results for 1298CC, the previously described process for sensitivity analysis was performed. The resulting summary effects estimate for 1298CC versus 1298AA was 1.04 (95% CI 0.94-1.14) χ 2 = 32.17, df = 32, P = 0.46, I 2 = 1% with no significant heterogeneity (data not shown). In this analysis, the studies contributing to the heterogeneity were conducted in Germany, India, and the USA [35,37,48,54,76].

Subgroup Analysis.
There were an insufficient number of studies that reported CRC risk by sex; however, subgroups, by geography, and source of controls were performed. Subgroup analysis by geography was performed for the MTHFR A1298C polymorphism according to the country groups previously described. There were an insufficient number of studies from the Middle East to include this location in the analysis. The subgroup analysis revealed that for European countries there was an associated, significant increased risk of CRC for those with the 1298CC genotype, while Asian and USA studies suggest a significant associated decrease in risk (Table 3). This variability in the associated risk of the 1298CC genotype by geography was also noted by Kono and Chen in their meta-analysis [95].
A further analysis was performed by comparing the results based on the source of controls; either hospital patients or healthy persons. For the CC variant, the healthy controls had a nonsignificant reduced risk associated with CRC versus hospital control, within some increase in heterogeneity (Table 3).

Publication Bias.
The results of the statistical test for publication bias for the MTHFR A1298C polymorphisms suggest that publication bias may not be a concern. For MTHFR 1298AC, the Begg and Mazumdar test returned a P value = 0.24, Egger's a P value = 0.398, and Trim and Fill found that an additional 5 studies would be necessary to form a symmetrical funnel plot whereas, for the 1298CC genotype, the Begg and Mazumdar test returned a P value = 0.88, Egger's a P value = 0.74, and Trim and Fill found that no additional studies would be necessary to form a symmetrical funnel plot.

Colorectal Cancer Risk and Combinations of the MTHFR
C677T and A1298C Genotypes. The combinations of variants of the MTHFR C677T and A1298C genotypes are in linkage disequilibrium such that rarely are there individuals with the 677TT/1298AC and 677TT/1298CC combinations [95]. The results of the summary risk estimates for the remaining combinations are presented in Table 4. The combination of 677TT/1298AA was associated with lowest risk of CRC with a summary risk estimate of 0.77 (95% CI  frequency questionnaire (FFQ) was the usual method used to collect dietary intake information. Dietary information was captured for one to two years preceding diagnosis, or for the control group, at the time of enrolment in the study. The range of dietary folate intake, defined as folate from food sources, for the lowest "quantile" ranged from a low of less than 115.6 to 406 mcg/day; the range for the highest was from 320 to 485 mcg/day or more (   Favors high folate intake Favors low folate intake

MTHFR 677TT
Heterogeneity: τ 2 = 0.06; χ 2 = 6.29, df = 5 (P = 0.28); I 2 = 21% (b) Figure 6: Forest plot of the risk of colorectal cancer comparing high versus low folate intake within each MTHFR C677T polymorphism.  Figure 6). Total folate intake information was also reported in some studies. Total folate was defined as folate from dietary and supplemental sources. The lowest "quantile" ranged from less than 264 to 450 mcg/day and the higher "quantile" ranged from 348 to 1583 mcg/day or more (  Figure 6). Only two studies had information available for the 677CT genotype; therefore, the summary risk estimate was not determined.

Discussion
The results of the analysis suggest that the homozygous variant genotype MTHFR 677TT confers a degree of protection against the development of CRC, affording an associated risk reduction of 12%. In contrast, the heterozygous genotype, MTHFR 677CT, was found to have the same risk as the genotype, MTHFR 677CC. These results are consistent with the previous meta-analysis completed in 2009 [94]. The thermolabile nature of MTHFR 677TT enzyme results in the reduced conversion of 5,10-methylenetetrahydrofolate to 5-methyl-tetrahydrofolate, which acts as cofactor in the conversion of homocysteine to methionine, permitting a larger pool of 5,10-methylene-tetrahydrofolate for thymidylate biosynthesis. This protective effect would suggest that preferential availability of folates to contribute pyrimidine synthesis, and therefore a reduction in uracil misincorporation and subsequent DNA breaks, could be important in the pathogenesis of CRC [32]. This reduced risk of CRC for the 677TT genotype was not supported by all of the included studies. In several individual studies, the 677TT genotype was associated with an increased risk of CRC [29][30][31]. The authors of these studies theorized that conditions of low folate intake, which is characteristic of the diet in these countries (Brazil, Mexico), may explain the increased risk found between the 677TT genotype and CRC. This would appear to be substantiated by the reduced risk apparent in the summary risk estimated for 677CC and 677TT genotypes when comparing high versus low total folate intake ( Figure 6) and would suggest that folate intake can alter the risk of CRC. Evidence for the alteration of disease through adequate folic acid intake has been found in other situations. For example, a maternal MTHFR 677TT genotype is associated with a higher risk of having an offspring with a neural tube defect [97]. Increased folic acid supplementation, periconceptionally and during the first trimester, has been found to reduce this risk [98].
Many of the studies incorporated both men and women into the case control groups. However, far fewer studies stratified their results based on sex. Of the eleven studies included in this subgroup analysis, representing over 7,000 case/control study participants, only one reported significant OR based on sex and genotype, which was contrary to the summary results in this meta-analysis (Table 2). Lightfoot et al. found that the men with the 677CT genotype had a reduced risk of CRC, and women with the 677TT genotype had an increased risk [58]. In the subgroup analysis on sex, the risk reduction of the 677TT genotype and significant summary risk estimate for both sexes was no longer evident. This may represent lack of statistical power; it is possible that more studies are necessary to determine whether there may be a gender bias favoring one sex over another regarding the protective nature of the 677TT genotype.
The A1298C polymorphisms would not appear to be associated with any substantial reduction in the associated risk of CRC. However, subgroup analysis did reveal some variability in the associated risk for the 1298CC genotype, with lower risks associated with Asian and USA studies. What might be contributing to these geographical differences is unclear. Perhaps, as with the subgroup analysis by sex, additional studies with larger numbers of participants with this genotype are necessary to more clearly understand the relationship.
Many of the studies included in the high versus low folate intake meta-analysis compared the risk of CRC using the 677CC or 677CC/CT genotype and low folate intake as the reference group for the calculation of the odds ratio in other genotypes and folate intake "quantiles." Generally, the findings of these studies were that high folate intake and the 677TT genotype were associated with a nonsignificant reduction in CRC risk versus low folate intake. This is the first study to perform a meta-analysis of the risk of CRC comparing high versus low folate intake within a genotype. The meta-analysis findings for the homozygous genotypes (677CC and 677TT) indicate that there is greater risk reduction with higher levels of folate intake. The upper range of high folate intake reported in the studies was, generally, over the Institute of Medicine's (IOM) recommended daily intake (RDI) of 400 mcg/day and in one case over 1 mg/day [23,99]. There were no clear boundaries in the definition of low folate intake versus high folate intake in this analysis as there was overlap in the ranges in daily folate amounts that defined the lowest folate intake versus the highest intake. This does prevent generalizing an amount of folate intake for each genotype that may be related to reducing colorectal cancer risk, which is a limitation of this analysis. Further, there is insufficient data to verify the shape (linear versus nonlinear) of the dose effect curve. More studies at this level of detail are necessary to provide further insight into the shape of the dose effect curve for folate and its associated impact on the risk of colorectal cancer.
The available studies used food frequency questionnaires (FFQs) or an adapted Coronary Artery Risk Development