MTHFR C677T and colorectal cancer risk: A meta-analysis of 25 populations

Abstract

The common functional methylenetetrahydrofolate reductase (MTHFR) C677T polymorphism may influence the risk of colorectal cancer (CRC), but data from published studies with individually low statistical power are conflicting. To clarify the role of MTHFR C677T genotype in CRC, we considered all available studies in a meta-analysis. Studies reporting on MTHFR C677T genotype and CRC were searched in PubMed up to April 2006. The principle prior hypothesis was that homozygosity for MTHFR 677TT would be associated with reduced risk of CRC. Data were available for 29,931 subjects, including 12,243 with CRC, from 25 independent populations. Compared to the homozygous CC genotype, the MTHFR 677TT genotype was associated with a reduced risk of CRC (odds ratio (OR): 0.83; 95% confidence interval (CI): 0.75–0.93; p = 0.001). There was some heterogeneity among the results of individual studies, but this was not statistically significant (heterogeneity p = 0.12; I² = 25.8%). Heterozygosity for MTHFR 677 did not influence CRC risk (OR: 0.99; 95% CI: 0.94–1.04). These findings indicate that individuals homozygous for the MTHFR 677TT genotype are at moderately reduced risk of CRC, and support the proposal that common genetic variation in the MTHFR gene contributes to CRC susceptibility, probably accounting for at least 9% of the total incidence. © 2006 Wiley-Liss, Inc.

Colorectal cancer (CRC) is the third most common cause of cancer-related mortality in the Western world, and in the United States, it represents the second most common cause of cancer death.1 A recent twin study indicates that around 35% of all CRCs can be ascribed to inherited susceptibility.2 Mendelian predisposition syndromes associated with mutations in known genes (such as APC, DNA mismatch repair genes, MYH, SMAD4, BMPR1A/ALK3 and STK11/LKB1), however, together account for only ∼5% of the overall incidence.3 The nature of the remaining heritability is at present undefined, but a model in which high-risk alleles account for all of the excess inherited risk seems improbable. An alternative hypothesis concerning the allelic architecture of CRC susceptibility proposes that part of the inherited risk is conferred by common, low-penetrance alleles, the “common disease-common variant” hypothesis.4 This hypothesis also implies that association studies should be a powerful strategy for identifying such alleles.5

The common functional polymorphism, C677T, in the gene encoding the methylenetetrahydrofolate reductase enzyme (MTHFR), has been widely studied as a possible low-penetrance susceptibility allele for a variety of cancers, and in particular, CRC. Chen et al. first reported that homozygosity for the MTHFR 677T allele was associated with a reduced risk of CRC.6 Findings from a number of subsequent studies, however, have been inconsistent with many reporting no significant association and some reporting an opposite relationship (reviewed in Ref.7). Many studies have shown a nonsignificant trend towards reduced CRC risk in MTHFR TT genotype individuals, but have lacked sufficient power individually to detect the modest genetic effect that would be expected under the common disease-common variant hypothesis. Interactions between MTHFR C677T genotype and dietary factors in determining risk of CRC have also been reported, suggesting that the effects of MTHFR genotype may be modulated by folate status.6

To clarify the effect of variation within MTHFR defined by the C677T polymorphism on risk of CRC, we have undertaken a systematic review and meta-analysis of published studies. To further identify issues that may aid in the design of future genetic association studies, we also explored potential sources of heterogeneity between published studies. Our analysis was limited to the main effect of MTHFR C677T genotype on CRC risk, as methodological differences between studies do not allow pooling of data from gene-nutrient subanalyses to be undertaken.

Methods

Study identification

A literature search for all studies reporting on the association between MTHFR C677T genotype and CRC risk was conducted using the electronic database PubMed (www.ncbi.nml.nih.gov/pubmed) up to April 21st 2006. The search strategy included the keywords “CRC”, “colon cancer”, “rectal cancer”, “polymorphism” and “MTHFR”. We searched for any additional studies in the bibliographies of all identified publications, including previous review articles and meta-analyses.

Selection criteria

Studies were eligible if they were based on unrelated individuals, and examined the association between CRC and presence of the MTHFR C677T polymorphism. Only studies published as full-length articles or letters in peer-reviewed journals in English were included in the analysis. For duplicate publications, the smaller dataset was excluded.

Data extraction

Data for analyses, including study design, sample size, ethnicity and allele and genotype frequencies were abstracted from published articles and summarized in a consistent manner to aid comparison. When a study reported results on different subpopulations according to ethnicity, we considered each subpopulation as a separate study in the meta-analyses.

Statistical analysis

Population stratification.

In any large, randomly mating population, in which there is a constant mutation rate, and no migration or selection against a particular genotype, the proportions of the various genotypes will remain unchanged from one generation to another. To test for population stratification, the distribution of genotypes in control subjects of each individual population was tested for departure from Hardy-Weinberg equilibrium by means of the χ² test.

Meta-analysis.

For the purpose of this analysis, all estimates of risk were considered to represent odds ratios (ORs), and raw data for genotype frequencies, without adjustment, were used for calculation of the study-specific estimates of the OR wherever possible. DerSimonian and Laird's method was used to calculate random-effects summary ORs and their associated 95% confidence intervals (CIs).8 Summary ORs were calculated based on comparison of individuals with TT homozygous genotype when compared to those with CC homozygous genotype, and also comparison of CT heterozygotes and CC homozygotes. Cochran's Q statistic was used to formally test for heterogeneity, and the percentage variability of the pooled OR attributable to heterogeneity between studies was quantified using the I² statistic.9 Meta-regression analysis was used to identify characteristics contributing to heterogeneity. A random-effects weighted linear regression model was used, whereby the study-specific logOR was regressed on the study characteristic variable of interest.10 The influence of individual studies on the summary OR was also evaluated by plotting the summary OR in the absence of each study in turn.

All statistical analyses were undertaken using the program STATA version 8.0 (Stata Corporation, TX).

Assessment of publication bias.

The presence of publication bias was examined by visual inspection of funnel plots, and formally evaluated with Begg's adjusted rank correlation test and Egger's regression asymmetry test.11 Begg's test examines the correlation between the effect estimates and their variances. Egger's test is based on inverse-variance weighted regression of the effect sizes on their precision (the inverse of standard error) to test whether the intercept deviates significantly from zero.

Population-attributable risk.

The population-attributable risk (PAR) associated with the MTHFR C677T polymorphism is a measure of the proportion of CRC that can be ascribed to the effect of genotype; PAR was estimated by (x − 1)/x, where x = (1 − p)² + 2p(1 – p)OR₁ + p²OR₂, given population allele frequency p, and OR₁ and OR₂ are the summary ORs associated with hetero- and homozygosity respectively.

Results

Characteristics of published studies

Twenty-three published articles reporting MTHFR C677T genotype and risk of CRC were identified.6, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33 Five were prospective studies, including 1 cohort study22 and 4 case-control studies nested within cohort studies,6, 12, 28, 29 while the remainder were all reterospective case-control studies. One study was excluded,14 since it reported data from a population expanded and updated in a subsequent study.26 Two studies subgrouped cases and controls according to ethnicity, and in the meta-analyses data on the different ethnic groups were treated as separate studies.17, 18 A third study also grouped subjects according to ethnicity but did not present data on genotype and CRC risk for all ethnic subgroups, thus this study was treated as a single study in the meta-analysis using the summary data for all ethnic subgroups combined.29 The final data set for the meta-analyses thus included 20 case-control studies, 4 nested case-control studies and 1 cohort study (Table I).

Table I. Details of Studies Included in the Analyses

Study	Year	Country	Sex	Age range (yrs)	Cancer	Study	Case source	Control source	Ethnicity	n1
NS, not stated; HPFS, Health Professionals Follow-up Study; PHS, Physicians Health Study; NCCCS, North Carolina Colon Cancer Study; HTR, Hawaii Tumour Registry; ZES, Zutphen Elderly Study; KPMCP, Kaiser Permanente Medical Care Program of Northern California; JANUS, biobank of Norwegian health screening program participants and blood donors. 1 Cases/controls. 2 Different ethnic groups within the same study. 3 CRC cases within cohort/total cohort. 4 Predominantly Caucasian.
Chen et al.6	1986–1994	US	Male	40–75	CRC	Nested case–control	HPFS	HPFS	Caucasian	144/627
Ma et al.12	1982–1995	US	Male	40–84	CRC	Nested case–control	PHS	PHS	Caucasian	202/326
Park et al.13	NS	Korea	Both	23–81	CRC	Case–control	Hospital series	Hospital	Korean	200/460
Ryan et al.15	1996–1997	Ireland	Both	NS	CRC	Case–control	Hospital series	Hospital	NS	136/848
Shannon et al.16	1985–1998	Australia	Both	32–100	CRC	Case–control	Hospital series	Population	NS	501/1,207
Keku et al.17
A2	1996–2000	US	Both	40–79	Colon	Case–control	NCCCS	Population	Caucasian	308/539
B2									African American	244/329
Le Marchand et al.18
A2	1994–1998	US	Both	<85	CRC	Case–Control	HTR	Population	Japanese	322/397
B2									Caucasian	149/171
C2									Hawaiian	77/88
Matsuo et al.19	NS-2001	Japan	Both	NS	CRC	Case–control	Hospital series	Hospital	Japanese	142/241
Sachse et al.20	1997–2001	UK	Both	45–80	CRC	Case–control	Hospital series	Hospital	Caucasian	490/592
Toffoli et al.21	1999–2000	Italy	Both	23–91	Colon	Case–control	Hospital series	Population	NS	276/279
Heijmans et al.22	1985–1995	Netherlands	Male	65–84	CRC	Cohort	ZES	ZES	NS	18/7933
Plaschke et al.23	NS	Germany	Both	NS	CRC	Case–control	Hospital series	Hospital	NS	287/346
Pufulete et al.24	2000–2001	UK	Both	38–90	CRC	Case–control	Hospital series	Hospital	NS	28/76
Kim et al.25	1998–2000	Korea	Both	30–79	CRC	Case–control	Hospital series	Hospital	Korean	243/225
Curtin et al.26	1991–1994	US	Both	30–79	Colon	Case–control	KPMCP	Population	Caucasian4	1,608/1,972
Yin et al.27	2000–2003	Japan	Both	20–74	CRC	Case–control	Hospital series	Population	Japanese	685/778
Ulvik et al.28	1973–2001	Norway	Both	NS	CRC	Nested case–control	JANUS	JANUS	NS	2,159/2,190
Le Marchand29	1993–1996	US	Both	45–75	CRC	Nested case–control	Multiethnic Cohort	Population	Mixed	822/2,021
Otani et al.30	1998–2002	Japan	Both	20–74	CRC	Case–control	Hospital series	Hospital	Japanese	106/222
Matsuo et al.31	2001–2004	Japan	Both	NS	CRC	Case–control	Hospital series	Hospital	Japanese	256/771
Wang et al.32	1999–2001	India	Both	17–88	CRC	Case–control	Hospital series	Population	Indian	302/291
Rudd et al.33	1994–2005	UK	Both	NS	CRC	Case–control	Hospital series	Population	Caucasian	2,556/2,692

The 24 case-control studies provided data on MTHFR C677T genotype for a total of 12,243 individuals with CRC and 17,688 controls. The cohort study reported MTHFR genotype on a cohort of 793 individuals who were followed-up for a 10-year period, during which 18 individuals developed CRC.

Nine studies were performed in the US,6, 12, 17^A,B,18^A–C,26, 29 8 in Europe,15, 20, 21, 22, 23, 24, 28, 33 4 in Japan,19, 27, 30, 31 2 in Korea,13, 25 1 in Australia16 and 1 in India.32 Three studies were restricted to male subjects,6, 12, 22 4 studies were based on only colon cancer cases17^A,B,21, 26 and 6 studies provided data on colon cancer and rectal cancer cases separately.12, 19, 25, 27, 28, 29 Seventeen of the 24 studies provided ethnicity data on study participants.6, 12, 13, 17^A,B,18^A–C,19, 20, 25, 26, 27, 30, 31, 32, 33 Of the individuals with CRC included in the meta-analyses approximately a third were European Caucasians, a fifth American Caucasians, a fifth Asians and the remainder were of other or unknown ethnicity. Nine studies recruited hospital-based rather than population-based controls.13, 15, 19, 20, 23, 24, 25, 30, 31

Frequencies of MTHFR C677T genotypes in the controls did not follow the expected distribution (p < 0.05). Seven of the 24 studies showed evidence of departure from Hardy-Weinberg equilibrium (2 at the 5% level,12, 29 and a further 5 at the 10% level6, 13, 15, 16, 20), indicating possible population stratification.

Allele frequency of MTHFR 677T

There was a wide variation in the MTHFR 677T allele frequency in the control groups across different populations, ranging from 0.06 in an Indian population32 to 0.45 in an Italian population21 (Table II). The mean T allele frequency was 0.32 in European Caucasian populations, 0.32 in American Caucasians and 0.40 in Asian populations.

Table II. MTHFR 677T Allele and Genotype Frequencies

Study	n1	Ethnicity	T allele frequency		Genotype frequencies						PHWE2
			Cases	Controls	Cases			Controls
					CC	CT	TT	CC	CT	TT
NS, not stated. 1 Cases/controls. 2 p value for Hardy Weinberg equilibrium in control group. 3 Different ethnic groups within the same study. 4 CRC cases within cohort/total cohort. 5 Predominantly Caucasian. 6 Depending on ethnic subgroup.
Chen et al.6	144/627	Caucasian	0.31	0.34	67 (46.5)	64 (44.4)	13 (9.0)	280 (44.6)	263 (41.9)	84 (13.4)	0.08
Ma et al.12	202/326	Caucasian	0.32	0.35	92 (45.5)	92 (45.5)	18 (8.9)	145 (44.5)	132 (40.5)	49 (15.0)	0.04
Park et al.13	200/460	Korean	0.41	0.43	65 (32.5)	107 (53.5)	28 (14.0)	140 (30.4)	246 (53.3)	74 (16.1)	0.05
Ryan et al.15	136/848	NS	0.37	0.29	49 (36.0)	73 (53.7)	14 (10.3)	439 (51.8)	326 (38.4)	83 (9.8)	0.05
Shannon et al.16	501/1,207	NS	0.31	0.33	249 (49.7)	197 (39.3)	55 (11.0)	533 (44.2)	560 (46.4)	114 (9.4)	0.06
Keku et al.17
A3	308/539	Caucasian	0.31	0.30	144 (46.7)	140 (45.5)	24 (7.8)	265 (49.2)	223 (41.4)	51 (9.4)	0.68
B3	244/329	African American	0.10	0.11	198 (81.2)	43 (17.6)	3 (1.2)	264 (80.2)	59 (17.9)	6 (1.8)	0.21
Le Marchand et al.18
A3	322/397	Japanese	0.37	0.42	126 (39.1)	153 (47.5)	43 (13.4)	138 (34.8)	182 (45.8)	77 (19.4)	0.22
B3	149/171	Caucasian	0.34	0.38	66 (44.3)	64 (43.0)	19 (12.8)	66 (38.6)	81 (47.4)	24 (14.0)	0.57
C3	77/88	Hawaiian	0.11	0.22	44 (57.1)	27 (35.1)	6 (7.8)	53 (60.2)	32 (36.4)	3 (3.4)	0.49
Matsuo et al.19	142/241	Japanese	0.44	0.41	39 (27.5)	81 (57.0)	22 (15.5)	81 (33.6)	124 (51.5)	36 (14.9)	0.30
Sachse et al.20	490/592	Caucasian	0.31	0.31	238 (48.6)	199 (40.6)	53 (10.8)	271 (45.8)	272 (45.9)	49 (8.3)	0.09
Toffoli et al.21	276/279	NS	0.40	0.45	93 (33.7)	145 (52.5)	38 (13.8)	83 (29.8)	140 (50.2)	56 (20.1)	0.83
Heijmans et al.22	18/7754	NS	0.42	0.29	7 (38.9)	7 (38.9)	4 (22.2)	392 (50.6)	322 (41.5)	61 (7.9)	0.65
Plaschke et al.23	287/346	NS	0.33	0.34	133 (46.3)	120 (41.8)	34 (11.8)	149 (43.1)	159 (45.9)	38 (11.0)	0.65
Pufulete et al.24	28/76	NS	0.32	0.27	16 (57.2)	6 (21.4)	6 (21.4)	41 (53.9)	29 (38.2)	6 (7.9)	0.78
Kim et al.25	243/225	Korean	0.40	0.39	86 (35.4)	122 (50.2)	35 (14.4)	83 (36.9)	109 (48.4)	33 (14.7)	0.77
Curtin et al.26	1,608/1,972	Caucasian5	0.32	0.33	734 (45.6)	724 (45.0)	150 (9.3)	887 (45.0)	858 (43.5)	227 (11.5)	0.37
Yin et al.27	685/778	Japanese	0.36	0.41	270 (39.4)	330 (48.2)	85 (12.4)	278 (35.7)	367 (47.2)	133 (17.1)	0.53
Ulvik et al.28	2,159/2,190	NS	0.28	0.30	1,103 (51.1)	899 (41.6)	157 (7.3)	1,092 (49.9)	886 (40.5)	212 (9.7)	0.10
Le Marchand29	822/2,021	Mixed	Variable6	Variable6	NS	NS	NS	NS	NS	NS	Variable6
Otani et al.30	106/222	Japanese	0.47	0.51	32 (30.2)	49 (46.2)	25 (23.6)	51 (23.0)	114 (51.4)	57 (25.7)	0.68
Matsuo et al.31	256/771	Japanese	0.36	0.40	106 (41.4)	114 (44.5)	36 (14.1)	289 (37.5)	348 (45.1)	134 (17.4)	0.10
Wang et al.32	302/291	Indian	0.08	0.06	257 (85.1)	43 (14.2)	2 (0.7)	255 (87.6)	36 (12.4)	0 (0.0)	0.26
Rudd et al.33	2,556/2,692	Caucasian	0.33	0.34	1,135 (44.4)	1,161 (45.4)	260 (10.2)	1,173 (43.6)	1,193 (44.3)	326 (12.1)	0.40

Risk of CRC

Figure 1 shows a forest plot of the OR of CRC associated with the MTHFR 677TT genotype when compared to CC genotype for each of the individual studies. Only 1 study did not present raw genotype frequency data, thus for this study, the OR is adjusted for age, sex and ethnicity.29 In 15 studies, subjects with TT genotype had a reduced risk of CRC,6, 12, 13, 17^A,B,18^A,B,21, 26, 27, 28, 29, 30, 31, 33 while in 9 studies, TT genotype was associated with an increased CRC risk.15, 16, 18^C,19, 20, 22, 24, 25, 32 The ORs and 95% CIs for risk of CRC associated with the CT heterozygote genotype when compared to CC homozygotes are plotted in Figure 2. The heterozygote genotype was associated with a reduced CRC risk in 12 studies,13, 16, 17^B,18^A,B,20, 21, 23, 24, 27, 30, 31 and an increased risk in 12 studies.6, 12, 15, 17^A,18^C,19, 22, 25, 26, 29, 32, 33

Figure 1. Forest plot of ORs of CRC risk for MTHFR 677TT genotype when compared to CC genotype.

Figure 2. Forest plot of ORs of CRC risk for MTHFR 677CT genotype when compared to CC genotype.

The summary OR under a random-effects model indicated that individuals with MTHFR 677TT genotype had a significantly reduced risk of CRC when compared to individuals homozygous for the C allele (OR: 0.83; 95% CI: 0.75–0.93; p = 0.001) (Fig. 1). There was heterogeneity among the results of individual studies, although this was not statistically significant (Q = 32.3, p = 0.12; I² = 25.8%). For individuals with heterozygote genotype, however, the summary OR did not indicate a significantly altered CRC risk when compared to those with CC genotype (OR: 0.99; 95% CI: 0.94–1.04). Again there was no significant heterogeneity (Q = 30.6, p = 0.17; I² = 21.5%).

Meta-regression analysis exploring heterogeneity in the individual study ORs for the comparison of CRC risk in TT homozygotes with CC homozygotes indicated that studies using hospital-based controls contributed significantly to heterogeneity (p = 0.02). When the meta-analysis was confined to only studies with population-based controls, heterogeneity was reduced (Q = 17.6, p = 0.28; I² = 14.9%), while the summary OR remained significant (OR: 0.78; 95% CI: 0.70–0.87). Other variables, including study-design, ethnicity and departure from Hardy-Weinberg equilibrium in control subjects, did not contribute significantly to heterogeneity.

Sensitivity analyses were performed to assess the robustness of the summary estimate of reduction in CRC risk conferred by the MTHFR 677TT genotype. First, ranking studies according to sample size of CRC cases, and repeating the meta-analysis including only the larger studies in the upper half of sample size, had no effect on the summary estimate of risk (OR: 0.80; 95% CI: 0.73–0.88). Second, the influence of each individual study on the summary estimate was examined by repeating the meta-analysis omitting each study in turn. In all cases, the estimate of overall effect did not change appreciably and remained statistically significant. Third, there was no significant difference in the reduction in risk of CRC with the MTHFR 677TT genotype when the analysis was confined to subjects of only Caucasian or only Asian ethnicity (OR: 0.83; 95% CI: 0.73–0.93, and OR: 0.75; 95% CI: 0.63–0.89 respectively).

Further analyses were performed to assess the effects of MTHFR 677TT genotype on the risk of colon cancer and rectal cancer separately. Restricting the analysis to the 8 studies that provided data on 5,049 colon cancer cases, or the 6 studies providing data on 1,558 rectal cancer cases, showed similar risk reductions for individuals with TT genotype when compared to those with CC genotype (OR: 0.78; 95% CI: 0.66–0.91, and OR: 0.82; 95% CI: 0.60–1.12 respectively). The heterogeneity between the results of individual studies reporting colon cancer risk was less than that between studies reporting rectal cancer risk (Q = 12.1, p = 0.21, I² = 25.6% and Q = 9.6, p = 0.09, I² = 47.6% respectively).

The effects of geographical location of study subjects on the relationship between MTHFR 677TT genotype and CRC risk was also explored. Limiting the analysis to the 9 studies that provided data on subjects in the US, or to the 7 studies on subjects from countries in Asia, resulted in similar summary estimates (OR: 0.76; 95% CI: 0.66–0.88, and OR: 0.75; 95% CI: 0.63–0.89 respectively), with little heterogeneity (Q = 4.9, p = 0.77, I² = 0.0% and Q = 5.3, p = 0.51, I² = 0.0% respectively). However, when the analysis was restricted to the 8 studies reporting on subjects from Europe, individuals with MTHFR 677TT genotype were not at reduced CRC risk when compared to those with CC genotype (OR: 0.98; 95% CI: 0.76–1.25), although there was a significant heterogeneity between individual studies (Q = 18.2, p = 0.01, I² = 61.6%).

Publication bias

A funnel plot for visual assessment of publication bias is shown in Figure 3. Formal evaluation of publication bias using Begg's and Egger's tests paradoxically indicated the existence of publication bias towards studies reporting increased risk of CRC in MTHFR 677TT genotype individuals (p = 0.06 and 0.03 respectively).

Figure 3. Begg's funnel plot (using OR of CRC risk in TT genotype individuals when compared to CC genotype individuals). The horizontal line represents the meta-analysis summary estimate, and the diagonal lines pseudo-95% CI limits about the effect estimate. In the absence of publication bias, studies will be distributed symmetrically above and below the horizontal line. Asymmetry on the right of the graph indicates evidence of publication bias towards studies reporting a positive logOR (increased CRC risk with TT genotype). logOR, natural logarithm of the OR; selogOR, standard error of the logOR.

Discussion

On the basis of 22 studies providing data on MTHFR C677T genotype and CRC risk in 25 populations, including over 29,000 subjects, our meta-analysis provides good evidence that homozygosity for the T allele is associated with a modest, but significantly, reduced risk of CRC. Heterozygosity, however, was not shown to impact on CRC risk.

Our sensitivity analyses indicated that the estimate of the summary effect was robust, and did not alter materially when factors contributing to the heterogeneity between individual studies were addressed. Our meta-analysis was, however, based on raw data points and although the process of Mendelian randomization makes confounding unlikely in analyses of association of genotype and disease, an imbalance in the distribution of CRC risk factors by MTHFR genotype cannot be entirely excluded.34

Publication bias was considered as a possible explanation for the observed association between genotype and CRC risk. However, formal evaluation in fact indicated the existence of publication bias towards studies reporting an increased CRC risk in MTHFR 677TT genotype individuals. If such bias genuinely existed, then it would reduce the likelihood of observing the effect seen in the meta-analysis. Population stratification represents another potential source of bias in association studies, particularly when data from diverse populations are analyzed. Empirical evidence, however, indicates that population stratification is likely to be minimal in well-designed population-based studies that match or control for ethnicity.35 The control groups in 7 of the studies included in the meta-analysis showed some evidence of departure from Hardy-Weinberg equilibrium. However, excluding these studies from the analysis had no real effect on the summary OR. Selection bias in recruitment of control subjects may also occur, and the use of population-based, rather than hospital-based control subjects, provides a means of minimizing such bias. Meta-regression analyses indicated that the studies recruiting hospital-based controls were a source of bias; however, this did not impact on the overall summary effect.

Failure to replicate initial findings is a major concern in association studies, and false-positive and false-negative reports from underpowered studies may explain a large part of this inconsistency.36 It is therefore noteworthy that none of the published studies individually had adequate statistical power to robustly detect the association between MTHFR C677T genotype and CRC risk estimated by our meta-analysis, although the 4 largest individual studies all reported associations similar to the meta-analysis summary estimate. Although the OR of 0.83 for CRC in carriers of the TT genotype when compared to the CC genotype is modest, because of the high C allele frequency, this translates into a relatively high population attributable risk. On the basis of this pooled data we estimate that MTHFR C677T genotype contributed to ∼15% (95% CI: 9–22%) of the total incidence of CRC.

Causal arguments for the effects of genetic variants on disease susceptibility detected in association studies can be strengthened if the observed effect can be supported by biological plausibility and functional data.36 While it is not possible to fully exclude linkage disequilibrium (LD) as an explanation for the observed association between MTHFR C677T genotype and CRC risk, the functional effects of this polymorphism on the expressed protein thermolability and activity are well documented. Homozygosity for the T allele results in an enzyme with 30%, and heterozygosity 65%, of the wild-type enzyme activity.37 Our finding of a recessive inheritance model for the effect of MTHFR C677T genotype on CRC risk is compatible with this functional data, assuming a threshold of reduced enzyme activity somewhere between 30 and 65% at which the phenotype emerges.

The MTHFR enzyme plays a pivotal role in one-carbon metabolism, catalyzing the irreversible conversion of 5,10-methyleneTHF to 5-methylTHF (Fig. 4). The MTHFR substrate, 5,10-methyleneTHF, is also used in purine synthesis and as a substrate by the thymidylate synthetase (TS) enzyme in the methylation of dUMP to dTMP, which is the sole de novo source of thymidine required for DNA synthesis and repair.38 Reduced availability of dTMP results in misincorporation of uracil into DNA, repair of which may lead to double strand DNA breaks, which have carcinogenic effects.39 The MTHFR product, 5-methylTHF, is the methyl group donor for the remethylation of homocysteine to methionine, which is subsequently used for DNA methylation, abnormalities of which are also known to play a role in carcinogenesis.40 Reduced MTHFR activity would be expected to result in increased availability of 5,10-methyleneTHF for DNA synthesis, which may underlie the reduced CRC risk in TT genotype individuals. The carcinogenic effects of abnormalities in DNA synthesis and repair would not necessarily be expected to be limited to the colorectum, and in this regard it is interesting to note that the MTHFR C677T variant has been associated with reduced risk of other malignancies including lymphoma, acute lymphoblastic leukemia, bladder and prostate cancer.41, 42, 43, 44, 45

Figure 4. Schematic representation of one-carbon metabolism. MTHFR, methylenetetrahydrofolate reductase; MTR, methionine synthase; MTRR, methionine synthase reductase; TS, thymidylate synthetase; THF, tetrahydrofolate; DHF, dihydrofolate; dUMP, deoxyuridine monophosphate; dTMP, deoxythymidine monophosphate; SAM, S-adenosylmethionine; SAH, S-adenosylhomocysteine.

One-carbon metabolism may also be influenced by dietary factors, such as folate, alcohol and B vitamin intake, and variants in genes encoding other enzymes such as TS, methionine synthase (MTR) and methionine synthase reductase (MTRR). Furthermore, there is a second functional variant (A1298C) within the MTHFR gene itself, although this variant is in LD with the C677T variant and its functional effects are weaker.46, 47 An interaction between MTHFR C677T genotype and folate status in determining CRC risk is supported by in vitro studies, which indicate that the reduced activity of the MTHFR TT enzyme is due to enhanced propensity to dissociate into monomers and lose its FAD (flavin adenine dinucleotide) cofactor, and that this effect is enhanced under conditions of low folate.48 In our analysis, the MTHFR 677TT genotype was associated with a reduced risk of CRC when studies conducted in the US or Asia were considered, but not when studies conducted in Europe were considered. It is conceivable that this difference may reflect different levels of dietary folate intake between these geographical regions. Dietary folate intake is higher in the US when compared to Europe, where supplements are not regularly used and food is not enriched with folate, which may enhance the protective effect of MTHFR 677TT genotype.

Potential gene–environment and gene–gene interactions between these factors and the MTHFR C677T variant in determining CRC risk were investigated in most of the studies included in our analyses.6, 12, 16, 17, 18, 19, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33 In some, but not all, studies, significant interactions were observed, and this inconsistency may be due to a number of factors, apart from the well-documented difficulties in accurately recording dietary intakes.49 The sample sizes required to adequately investigate interactions are frequently significantly greater than those required to demonstrate main effects.50 Our analysis indicated that no study had sufficient power for robust detection of the main effect of MTHFR C677T genotype, and power for detection of interactions would be even lower. Furthermore, at present, it is not clear exactly how fluctuations in dietary factors impact upon one-carbon metabolism, and the functional effects of other genetic variants in genes involved in one-carbon metabolism are less well defined, making interpretation of any observed interactions speculative. Nevertheless, our analysis indicates that the MTHFR C677T polymorphism exerts an independent influence on CRC risk, without consideration of these other potential modifiers.

In conclusion, pooled analysis of data from 25 separate populations including individuals with different ethnicities indicates that the MTHFR 677TT genotype is associated with a modest, but significant, reduced risk of CRC. This association was complicated by between study heterogeneity; however, the association remained significant after removal of factors contributing to the heterogeneity. Future large-scale, population-based association studies are now required to investigate potential gene–gene and gene–environment interactions involving the MTHFR C677T polymorphism in determining CRC risk.

Acknowledgements

RAH was in receipt of a Cancer Research UK Clinical Research Training Fellowship.