Systematic Review and Meta-Analysis of the Relationship between EPHX1 Polymorphisms and Colorectal Cancer Risk

Background Microsomal epoxide hydrolase (EPHX1) plays an important role in both the activation and detoxification of PAHs, which are carcinogens found in cooked meat and tobacco smoking. Polymorphisms at exons 3 and 4 of the EPHX1 gene have been reported to be associated with variations in EPHX1 activity. The aim of this study is to quantitatively summarize the relationship between EPHX1 polymorphisms and colorectal cancer (CRC) risk. Methods Two investigators independently searched the Medline, Embase, CNKI, and Chinese Biomedicine Databases for studies published before June 2012. Summary odds ratios (ORs) and 95% confidence intervals (CIs) for EPHX1 Tyr113His (rs1051740) and His139Arg (rs2234922) polymorphisms and CRC were calculated in a fixed-effects model and a random-effects model when appropriate. Results This meta-analysis yielded 14 case-control studies, which included 13 studies for Tyr113His (6395 cases and 7893 controls) and 13 studies for His139Arg polymorphisms (5375 cases and 6962 controls). Overall, the pooled results indicated that EPHX1 Tyr113His polymorphism was not associated with CRC risk; while the His139Arg polymorphism was significantly associated with decreased CRC risk (Arg/His vs. His/His, OR = 0.90, 95%CI = 0.83–0.98; dominant model, OR = 0.92, 95%CI = 0.85–0.99). The statistically significant association between EPHX1 His139Arg polymorphism and CRC was observed among Caucasians and population-based case-control studies. This association showed little heterogeneity and remained consistently strong when analyses were limited to studies in which genotype frequencies were in Hardy–Weinberg equilibrium, or limited to studies with matched controls. When cumulative meta-analyses of the two associations were conducted by studies’ publication time, the results were persistent and robust. Conclusion This meta-analysis suggests that EPHX1 Tyr113His polymorphism may be not associated with CRC development; while the EPHX1 His139Arg polymorphism may have a potential protective effect on CRC.


Introduction
Colorectal cancer (CRC) is the third most commonly diagnosed cancer in males and the second in females worldwide, with over 1.2 million new cancer cases and 608,700 deaths estimated to have occurred in 2008 [1]. In the United States, CRC is the third most common cancer and third leading cause of cancer death for both men and women [2]. In Europe, CRC represents one of the primary causes of cancer deaths [3] and in Asia, CRC is the fourth leading cause of mortality by cancer, and its incidence is increasing [4]. In recent years, the incidence of CRC is increasing in China, which accounts for about 6.5% of total cancers in urban areas and 4.6% in rural areas [5]. However, the mechanism of colorectal carcinogenesis is still not fully understood. As with other complex diseases, CRC is caused by both genetic and environmental factors [6]. Because well-recognized genetic predisposition syndromes account for less than 3% of CRC, low-penetrance genetic factors alone or in combination with environmental factors probably contribute to CRC development [7].
Red meat consumption has frequently shown an association with an increased risk of CRC. It has been proposed that this risk may be due to carcinogenic polycyclic aromatic hydrocarbons (PAHs) and heterocyclic amines produced when meat is cooked at high temperatures [8]. Microsomal epoxide hydrolase (mEH) (EPHX1) is an enzyme found on the endoplasmic reticulum of many tissues and is responsible for the hydrolysis of various epoxides, including PAHs [9]. Epoxides are often the most toxicologically active form of a drug or environmental chemical, because they are highly reactive oxidative metabolites. EPHX1 breaks the three-membered epoxide ring structure by the transaddition of water to form a less-reactive diol that can be conjugated and more readily excreted. Nevertheless, EPHX1 plays a dual role in the detoxification and activation of procarcinogens, and its role in carcinogenesis may depend on exposures to different environmental substrates [10].
The human EPHX1 gene is 35.48 kb with nine exons and eight introns on chromosome 1q42.1. There are more than 110 validated single nucleotide polymorphisms (SNPs) in EPHX1 gene reported in the dbSNP database (http://www.ncbi.nlm.nih.gov/ SNP), two of which are common and the two alleles of EPHX1 in codons 113 (site T337C, amino acid change Tyr113His, dbSNP: rs1051740) and 139 (A415G, His139Arg, rs2234922) affect enzyme activity [11]. The tyrosine to histidine substitution in exon 3 (Tyr113His) of the EPHX1 gene decreases in vitro enzyme activity by 40%, whereas the histidine to arginine substitution in exon 4 (His139Arg) increases in vitro enzyme activity by 25% [11]. Given the known differential effect of EPHX1 alleles in the detoxification of procarcinogens, it has been proposed that the two functional polymorphisms may affect cancer risk.
Over the last two decades, a number of studies were conducted to investigate the association between EPHX1 polymorphisms and CRC risk in different populations. However, the results of these studies are conflicting rather than conclusive. Until recently, few studies had been conducted to examine association between EPHX1 Tyr113His and His139Arg polymorphism and CRC risk by the systematic review or meta-analysis. In order to derive a comprehensive estimation of the associations between EPHX1 polymorphisms and CRC risk, we conducted a meta-analysis to assess the association between Tyr113His and His139Arg polymorphisms of the EPHX1 gene and CRC susceptibility.

Literature Search Strategy
We searched the PubMed, Embase, CNKI (China National Knowledge Infrastructure) and Chinese Biomedicine databases for all articles on the association between EPHX1 polymorphisms and CRC risk (last search update 5th June 2012). The following key words were used: ''microsomal epoxide hydrolase'' or ''EPHX1'' or ''mEH'', ''colorectal'' or ''colo*'', ''cancer'' or ''tumor'' or ''carcinoma'', and ''polymorphism'' or ''variant'' or ''allele'' or ''genotype''. The search was without restriction to the language and on studies conducted on human subjects. The reference lists of reviews and retrieved articles were hand searched at the same time. We did not consider abstracts or unpublished reports. If more than one article was published by the same author using the same case series, we selected the study where the most individuals were investigated.

Inclusion and Exclusion Criteria
We reviewed abstracts of all citations and retrieved studies. The following criteria were used to include published studies: (i) casecontrol studies were conducted to evaluate the association between at least one of these two polymorphisms (Tyr113His and His139Arg) and CRC risk; (ii) sufficient genotype data were presented to calculate the odds ratios (ORs) and 95% confidence intervals (CIs); (iii) The paper should clearly describe CRC diagnoses and the sources of cases and controls. Major reasons for exclusion of studies were (i) review, or editorial, or comment; (ii) duplicated studies; (iii) cell line studies.

Data Extraction
Two investigators (Fei Liu and Ding Yuan) extracted information from all eligible publications independently according to the inclusion criteria listed above. Disagreements were resolved by discussion between the two investigators. The following characteristics were collected from each study: the first author's name, year of publication, the country of participants, ethnicity, source of control group (population-or hospital-based controls), number of cases and controls, genotypes, genotyping methods, minor allele frequency (MAF) in controls, and evidence of Hardy-Weinberg equilibrium (HWE) ( Table 1). According to definitions in previous study [12], population-based case-control study (PCC) was defined as controls from healthy people, and hospital-based case-control study (HCC) were from hospitalized patients.

Statistical Analysis
We first assessed HWE in the controls for each study using goodness-of-fit test (chi-square or Fisher's exact test) and a P,0.05 was considered as statistically significant. The strength of the association between CRC and the EPHX1 Tyr113His and His139Arg polymorphisms were estimated using ORs, with the corresponding 95% CIs. In addition, Z-test was also used, and the P value ,0.05 indicated statistical significance for the association. The crude ORs and 95%CIs were calculated by several comparisons. Taking EPHX1 Tyr113His as an example: codominant model (His/His vs. Tyr/Tyr and Tyr/His vs. Tyr/Tyr), dominant model (His/His+Tyr/His vs. Tyr/Tyr) and recessive model (His/His vs. Tyr/His+Tyr/Tyr) respectively [13].
Both the Cochran's Q statistic [14] to test for heterogeneity and the I 2 statistic to quantify the proportion of the total variation due to heterogeneity [15] were calculated. A P value of more than the nominal level of 0.10 for the Q statistic indicated a lack of heterogeneity across studies, allowing for the use of a fixed-effects model (the Mantel-Haenszel method) [16];otherwise, the randomeffects model(the DerSimonian and Laird method) was used [17]. All meta-analyses are presented as forest plots that include ORs and 95% CIs for all individual studies, as well as the pooled estimator. Shaded figures provided for all ORs have dimension proportional to study weight. The Galbraith plot was used to detect the potential sources of heterogeneity [18]. Heterogeneity was also explored using subgroup analysis with ethnicity, study sample size ($1000/,1000 subjects), matched control (Yes/No), HWE in controls (Yes/No) and source of controls (HCC/PCC).
Sensitivity analyses were performed to assess the stability of the results, namely, a single case-control study in this meta-analysis was omitted each time to reflect the influence of the individual data set to the pooled OR. Several methods were used to assess the potential publication bias. Visual inspection of funnel plot asymmetry was conducted. The Begg's rank correlation method [19] and the Egger's weighted regression method [20] were used to statistically assess publication bias (P,0.05 was considered statistically significant). All analyses were done using STATA software, version 11.0 (STATA Corp., College Station, TX, USA). All the P values were two-sided.
The literature search and study selection procedures are shown in Figure 1.
The characteristics of selected studies are summarized in Table 1. There were two studies of subjects of Asian descent, 11 studies of subjects of Caucasian descent and one of subjects Mixed descent. Studies had been carried out in China, UK, USA, Spain, Canada, Czech, Japan, Turkey, Norway, Hungary, and Netherlands. The cases definition used in the individual studies were pathologically or histologically diagnosed with CRC. Controls were mainly from healthy populations and matched for age and/ or sex, of which 10 were population-based and four were hospitalbased. Most of studies extracted DNA from peripheral blood and the classic PCR-RFLP assay and Taqman PCR were mainly used for genotyping. The genotype distributions among the controls of all studies followed HWE except for four studies [21,22,28,33] for the Tyr113His polymorphism and one study [33] for the His139Arg polymorphism.

Quantitative Synthesis
Association of the EPHX1 Tyr113His polymorphism with CRC susceptibility. 13 case-control studies [7,21,22,[24][25][26][27][28][29][31][32][33] with 6395 cases and 7893 controls for EPHX1 Tyr113His were included eventually. On the basis of the potential underestimation of the true effect of the polymorphism on the CRC risk, we stratified these studies according to ethnicity, source of controls, study sample size, matched control, and HWE in controls. Different ethnicities were categorized as Caucasians and others; while different source of controls were defined as HCC and PCC. In stratified analyses, the variant genotypes (His/His and Tyr/His) had no significant relationship with CRC in all of the subgroups except that a significantly increased CRC risk was observed among the HCC populations in the homozygote comparison. Also, no significant associations were found in the dominant and recessive models in any subgroup ( Table 2).
When stratifying by ethnicity and source of controls, the significantly decreased CRC risk was observed among Caucasians

Heterogeneity Analysis
For Tyr113His polymorphism, there was substantial heterogeneity among these studies for homozygote comparison (His/His vs. Tyr/Tyr: P heterogeneity = 0.004), and recessive model comparison  (His/His vs.Tyr/His+Tyr/Tyr: P heterogeneity ,0.001). For Hi-s139Arg polymorphism, mild between-study heterogeneity was also detected the homozygote comparison, and recessive model comparison. Galbraith plot analyses of all included studies were used to assess the potential sources of heterogeneity. Three studies [21,22,28] were found to be contributors of heterogeneity for Tyr113His polymorphism ( Figure S1A). We re-evaluated the association after excluding these three outlier studies with reduced heterogeneity (His/His vs. Tyr/Tyr: P heterogeneity = 0.614; His/His vs.Tyr/His+Tyr/Tyr: P heterogeneity = 0.495). Only one study was found to be contributor of heterogeneity for His139Arg polymorphism ( Figure S1B) and the heterogeneity was significant reduced when excluding the outlier study (Arg/Arg vs. His/His: P heterogeneity = 0.521; Arg/Arg vs.Arg/His+His/His: P heterogeneity = 0.212).

Sensitivity Analysis
In the sensitivity analysis, the influence of each study on the pooled OR was examined by repeating the meta-analysis while omitting each study, one at a time. As for the association of the EPHX1 Tyr113His with CRC risk, the study that had the most influence on the overall pooled estimates ( Figure S2A) seemed to be the one conducted by Kiss et al. [28]; however, the sensitivity analysis showed that the ORs were 1.02 (95% CI: 0.96, 1.09) and 1.01 (95% CI: 0.94, 1.08) before and after the removal of that study, respectively, indicating high stability of the results. Because there is known methodological issue with PCR-RFLP analysis of Tyr113His SNP [27], we performed analysis without studies using the biased method. When excluding the studies using PCR-RFLP analysis of Tyr113His SNP, the estimated pooled OR still did not change at all (Table S1). As for the association of the EPHX1 His139Arg with CRC risk, the study that had the most influence on the overall pooled estimates ( Figure S2B) seemed to be the one conducted by Robien et al. [25]; however, the sensitivity analysis showed that the ORs were 0.92 (95% CI: 0.85, 0.99) and 0.91 (95% CI: 0.83, 0.99) before and after the removal of that study, respectively, indicating high stability of the results. When excluding the studies that were not in HWE, the estimated pooled OR still did not change at all ( Table 2 and Table 3). This procedure proved that our results were reliable and robust.

Cumulative Meta-analysis
Cumulative meta-analyses of the 2 associations were also conducted via the assortment of studies by publication time. Figure S3A shows results from the cumulative meta-analysis of the association of the EPHX1 Tyr113His with overall CRC in chronologic order. Inclinations toward null significant associations were evident with each accumulation of more data over time. Figure S3B shows results from the cumulative meta-analysis of the association of the EPHX1 His139Arg with overall CRC in chronologic order. Inclinations toward decreased significant associations were evident with each accumulation of more data over time, although associations were initially null.

Publication Bias
Funnel plot, Begg's and Egger's tests were performed to evaluate publication bias of the literature on CRC. Figure S4 displayed funnel plots that examined the EPHX1 polymorphisms and overall CRC risk included in the meta-analysis in dominant comparison model. The shape of funnel plots did not reveal any evidence of funnel plot asymmetry.

Discussion
The present meta-analysis, including 14 case-control studies, explored the association between the Tyr113His and His139Arg polymorphisms of the EPHX1 gene and CRC risk. We found that EPHX1 Tyr113His polymorphism was not associated with CRC risk (6395 cases and 7893 controls). When subgroup analyses were performed by ethnicity, source of controls, study sample size, matched control, and HWE in controls; significant association was still not observed in any subgroup except for among hospital-based studies. Nevertheless, we found that EPHX1 His139Arg polymorphism was associated with decreased CRC risk. When stratifying by ethnicity and source of controls, the significant association was observed among Caucasians and among PCC studies. Moreover, this association showed little heterogeneity (I 2 = 0) and remained consistently strong when analyses were limited to studies in which genotype frequencies were in HWE, or limited to studies with matched controls. When cumulative meta-analyses of the two associations were conducted by studies' publication date, the results were persistent and robust.
EPHX1 is a critical enzyme in xenobiotic metabolism [35], which plays an important role in both the activation and detoxification of PAHs and aromatic amines. EPHX1 catalyzes the hydrolysis of arene, alkene, and aliphatic epoxides from PAHs and aromatic amines. This hydrolysis is generally a detoxification reaction because less reactive and more water-soluble transdihydrodiols are produced [36]. In a sense, EPHX1 is a protective enzyme involved in general oxidative defenses against a number of environmental substances, and its genetic polymorphisms, EPHX1 Tyr113His and His139Arg, may affect enzyme activity [11]. Previous in vitro study found that EPHX1 Tyr113His was associated with 40% of decreased enzyme activity, while His139Arg was associated with 25% of increased enzyme activity [11]. Based on the assumption that the Tyr allele at exon 3 and the His allele at exon 4 confer normal activity, whereas the His allele at exon 3 confers low activity and the Arg allele at exon 4 confers high activity, Benhamou et al [37] classified predicted EPHX1 activity as low (113HisHis/139HisHis, 113TyrHis/139HisHis and 113HisHis/139HisArg), intermediate (113 TyrTyr/139HisHis, 113 HisHis/139ArgArg and 113TyrHis/139HisArg) or high (113TyrTyr/139ArgArg, 113TyrTyr/139 HisArg and 113Tyr-His/139ArgArg) on the presence or absence of the 2 polymorphisms. Similarly, Smith and Harrison [38]classified predicted EPHX1 activity as rapid (113 TyrTyr/139 HisArg or 113 TyrTyr/139 ArgArg); normal (113 TyrTyr/139 HisHis or 113 TyrHis/139 HisArg); slow (113 TyrHis/139 HisHis or 113 TyrHis/139 ArgArg); and very slow (113 HisHis/139 HisHis). Given the different enzyme (the EPHX1 protein) activity which depends on the polymorphic form, it is biologically plausible that the EPHX1 His139Arg polymorphism may decrease the risk of CRC.
Interestingly, we found that the EPHX1 His139Arg heterozygotes, but not the homozygotes, had a significantly decreased risk of CRC. The observed effect is due mostly to the presence of heterozygous genotype and homozygous variant genotype rather dilutes this effect (Table 3 -heterozygous vs. dominant model). From the functional view there is lack of dose-relationship where the highest activity should exert the most significant effect.
Although the reason for a significantly decreased risk associated with the His139Arg variant heterozygote remains unknown, it is possible that these heterozygotes may have impaired function because of the potential imbalance of the protein structure. Another possible explanation is that the heterozygous genotype may be in linkage disequilibrium with other susceptibility loci. Similar phenomenon was observed by Ma et al. [39], who studied the variant genotypes of CDKN1A and CDKN1B and breast cancer risk. They found that the CDKN1B C -79T heterozygotes, but not the homozygotes, had a significantly increased risk of breast cancer.
Our results were in part consistent with previous studies. For example, Li et al. [40] performed a comprehensive meta-analysis of published epidemiological studies aims to systematically evaluate putative EPHX1 enzyme activity and risk of cancers and found that putative EPHX1 enzyme activity is related with risk of lung and upper aerodigestive tract cancers. However, they did not find any association between EPHX1 Tyr113His and His139Arg polymorphism and CRC risk. In recent, Zhao et al [41] published a meta-analysis for the relationships between five metabolic gene (including EPHX1) polymorphisms and colorectal adenoma risk and found that EPHX1 Tyr113His and His139Arg did not have any associations with colorectal adenoma risk. Although the reasons for this difference are as yet unknown, some possibilities should be considered. First, those gene-variant associations vary in different kinds of diseases and may result from the different mechanisms of carcinogenesis among different kinds of tumor. Second, different ethnic composition may contribute to the discrepancy. Different meta-analyses included different original studies which were performed in different races and the ethnic composition in different meta-analyses may be diversity. Third, some methodological diversity, such as inclusion criteria, the quality of original studies, selection bias, Type I error and study sample size, also can contribute to the discrepancy.
Because the allele frequencies of polymorphisms and their effects on the cancer risk were diverse in the different ethnicities, we carried out subgroup analysis by ethnicity. The results demonstrated that EPHX1 His139Arg polymorphism was associated with a decreased CRC risk among Caucasians, while there was no association between EPHX1 His139Arg polymorphism and CRC risk among Asians. The null result in Asians may be due to the limited number of studies with only two studies from Asian available in this meta-analysis. It is critical that larger and welldesigned multicentric studies based on Asian patients should be performed to re-evaluate the association. Moreover, results of meta-analyses often depend on control selection procedures [42]. Different controls source may be a confounding factor which may impact on the conclusion of our study because of case-control studies. For instance, some studies used a healthy population as the reference group (PCC), whereas others selected hospitalized patients as the reference group (HCC). In order to eliminate interference from the confounding factor, we performed subgroup analysis by source of controls. Our results showed that the significant association between EPHX1 His139Arg polymorphism and CRC was observed among PCC, but not among HCC. This may be due to that the HCC studies have some selection biases because such controls might be ill-related population, and may not be a representative of the general population, especially when the investigated genotypes were associated with the disease conditions hospital-based controls might have. Although hospital controls are relatively easier, more convenient and economical to be recruited, a proper population-based control subject may be better to reduce biases in such genetic association studies.
One of the major concerns in a sound meta-analysis is the degree of heterogeneity that exists between the component studies because non-homogeneous data are liable to results in misleading results. In the present study, the Q-test and I 2 statistics were carried out to test the significance of heterogeneity. Obvious heterogeneity between studies was observed in overall comparisons and also some subgroup analyses. In an attempt to find the sources of heterogeneity, a Galbraith plot was drawn, and three studies were thought to serve as the main contributors for the Tyr113His polymorphism and only one study for the His139Arg polymorphism. The heterogeneity was significantly reduced when excluding the outlier studies. Moreover, we re-analyzed the association after excluding the outlier studies; the conclusion was still consistent in overall comparisons. Another important issue for any meta-analysis is publication bias due to selective publication of reports. In the current study, Begg's funnel plot and Egger's test were performed to evaluate this problem. Both the shape of funnel plots and statistical results did not show publication bias. It is worth mentioning that the results held when the sensitivity analysis was performed, which implied that the results were reliable.
Some limitations of this meta-analysis should be addressed. First, our meta-analysis was based on unadjusted OR estimates because not all published studies presented adjusted ORs or when they did, the ORs were not adjusted by the same potential confounders, such as age, sex, ethnicity and exposures. Lacking of the information for the data analysis may cause serious confounding bias. Second, this paper was limited by analyzing two single-SNPs respectively and lack of combination of two-SNP analysis. EPHX1 enzyme activity is affected by single or combination of polymorphisms Tyr113His and His139Arg [11,43]. Based on the genotype combination of these two functional polymorphisms, Benhamou and colleagues [37] classified EPHX1 activity as putative low, intermediate and high. Thus, a meta-analysis that performed both single-SNP analysis and combined two-SNPs analysis may provide insights into the relationship between EPHX1 enzyme activity and CRC risk. However, only limited studies in this meta-analysis reported combination of two-SNP analyses (Table S2), which prevented us to perform pooled analysis. Third, there was significant between-study heterogeneity from studies of the EPHX1 polymorphism, and the genotype distribution also showed deviation from HWE in some studies. In spite of these, our meta-analysis also had some advantages. First, we did not detect any publication bias indicating that the whole pooled result should be unbiased. Second, the quality of casecontrol studies included in current meta-analysis was satisfactory and met our inclusion criterion.
In conclusion, this meta-analysis evaluates the relationship between genetic polymorphisms and CRC risk and reveals that EPHX1 Tyr113His polymorphism may be not associated with CRC development; while the EPHX1 His139Arg polymorphism may have a potential protective effect on CRC. Since limited studies were from Asian populations, it is critical that larger and well-designed multicentric studies based on Asians should be performed to re-evaluate the association. Moreover, further studies estimating the effect of both single-SNP analysis and combination of two-SNP analysis and gene-environment interactions may eventually provide a better, comprehensive understanding of the association between the EPHX1 polymorphisms and CRC risk. Figure S1 Galbraith plots for heterogeneity test of Tyr113His and His139Arg polymorphisms. (A) Galbraith plot of the association between Tyr113His polymorphism and CRC risk (The studies outside the range between -2 and 2 were seen as the outliers and the major source of heterogeneity); (B) Galbraith plot of the correlation between His139Arg polymorphism and CRC risk. (TIF) Figure S2 Influence analysis of the summary odds ratio coefficients on the association between EPHX1 polymorphisms and colorectal cancer risk. Results were computed by omitting each study (left column) in turn. Bars, 95% confidence interval. (A), For EPHX1 Tyr113His His/His -plus-Tyr/His genotypes vs. Tyr/Tyr genotype; (B), For EPHX1 His139Arg Arg/Arg-plus-Arg/His genotypes vs. His/His genotype. (TIF) Figure S3 Results from cumulative meta-analysis of associations between EPHX1 polymorphisms and colorectal cancer risk. The circles and horizontal lines show the accumulation of estimates as results from each study were added, rather than the estimate for each individual study. Studies sorted by publication time; Bars, 95% confidence interval.