Quantitative assessment of the influence of EPHX1 gene polymorphisms and cancer risk: a meta-analysis with 94,213 subjects

A comprehensive literature search was performed using PubMed database for relevant articles published (last search: March 14, 2014) with the following terms: ((“epoxide hydrolase 1”) OR EPHX1) AND (((polymorphism) OR (SNP)) OR variant)) AND ((((neoplasm) OR cancer) OR carcinoma) OR leukemia). All the references of retrieved articles and supplementary data were checked when key information relevant to the meta-analysis was missing.

All studies were included if they met the following criteria: (1) case-control study; (2) studies to evaluate the association between EPHX1 gene polymorphisms (Tyr113His and His139Arg) and risk of cancer; (3) sufficient data for estimating an odds ratio (OR) with 95% confidence interval (CI); (4) full-text in English available and (5) more than 100 patients. When the same population was included in several publications, only the most complete one was included in this meta-analysis.

Data were carefully evaluated and extracted from the eligible studies by two investigators independently according to the inclusion criteria listed above. The following characters were collected from eligible studies: first author’s name, year of publication, ethnicity (categorized as Asian, Caucasian, African, or mixed), source of control groups (population-based [PB], hospital-based [HB], family-based [FB] or unknown), genotype frequency of cases and controls, and the results of Hardy-Weinberg equilibrium (HWE) test. When it came to discrepancy between two investigators, another investigator was invited to discuss and check the data until a consensus was reached.

The departure from the Hardy-Weinberg equilibrium for the control group in each study was assessed with Pearson's goodness-of-fit Chi-square test with 1 degree of freedom by a web-based program (http://​ihg.​gsf.​de/​cgi-bin/​hw/​hwa1.​pl) and the violation of HWE was determined with a threshold of p < 0.05. Odds ratios (ORs) with 95% confidence intervals (CIs) were used to assess the strength of association between the EPHX1 gene polymorphisms and cancer susceptibility. Pooled ORs were performed for dominant model (aa + Aa vs. AA, a was for the minor allele and A was for the major allele), recessive model (aa vs. Aa + AA), homozygote comparison (aa vs. AA), heterozygote comparison (Aa vs. AA), and additive model (a vs. A), respectively. Heterogeneity among pooled studies were evaluated by the Chi-square-based Cochran’s Q test [15] and I² statistics [16]. To be more conservative, heterogeneity was considered to be present when the Cochran’s Q-test P-value was less than 0.1, then random-effects model (the DerSimonian and Laird method) [17] was utilized, otherwise, fixed-effects model was used (the Mantel-Haenszel method) [18]. In addition, inconsistency across studies was quantified by means of I² statistic, with I² < 25%, 25-75%, and >75% considered to represent low, moderate and high degree of heterogeneity, respectively [16]. Stratification analyses were performed to test the effects of cancer types, source of control, ethnicity and smoking status, respectively. To explore the source of heterogeneity among the studies of this meta-analysis, a multivariate meta-regression analysis subjected to 10,000 permutations was undertaken to explore the possible sources of heterogeneity. The following study characteristics were included as covariates in the meta-regression analysis: ethnicity, source of control, cancer types. Sensitivity analysis was carried out through omitting individual study in turn to check the consistency of the results. Publication bias was evaluated by visual inspection of the Begg’s funnel plots [19] and the Egger’s linear regression (P < 0.05 was considered a significant publication bias) [20]. All statistical tests were performed with metafor [21] and meta (http://​cran.​r-project.​org/​web/​packages/​meta/​) packages of R (version 3.0.1), using two-sided p-values.

This meta-analysis followed the guidelines of the preferred reporting items for systematic reviews and meta-analysis (PRISMA) statement [22] (Additional file 1: Table S1).

The initial literature search through PubMed database yielded 192 published articles. Totally, when reviewed in full-text, 4 were not concerned with Tyr113His or His139Arg polymorphisms in EPHX1 gene, 37 were not cancer risk studies, 1 was not published in English, 1 was not provided in full text, 7 were not case-control studies, 30 were no usable reported data, and 19 were meta-analysis or reviews; all these publications were excluded. Among the remaining 87 articles, studies presented separate OR by different polymorphisms, cancer types or ethnicity, and each of them was considered separately for pooling analysis. Furthermore, 32 studies not in HWE and 39 with less 100 patients were also deleted. Hence, 45 studies [8]-[10],[23]-[63] for Tyr113His polymorphism (20,091 cases and 27,396 controls) and 54 studies [8]-[10],[24]-[26],[28],[29],[32],[34]-[45],[47]-[51],[53],[55]-[61],[63]-[81] for His139Arg polymorphism (19,437 cases and 27,289 controls) were included eventually. Genotype distributions in the controls of all selected studies are in agreement with HWE. The flow of study selection was shown in Figure 1, and the main characteristics of eligible studies were summarized (Additional file 2: Table S2 and Additional file 3: Table S3).

The pooled results of meta-analysis for the association between EPHX1 polymorphisms (Tyr113His and His139Arg) and cancer susceptibility are shown in Tables 1 and 2. Heterogeneity across studies must be considered because it may affect the strengths of the meta-analysis. Significant heterogeneity was observed in some comparisons for both EPHX1 Tyr113His and His139Arg polymorphisms. Thus, random-effect model was used when heterogeneity identified.

For Tyr113His polymorphism, overall, no significantly elevated cancer risk could be observed in all genetic models (Table 1. homozygote model: OR = 1.05, 95% CI = 0.95-1.16; heterozygote model: OR = 0.94, 95% CI = 0.88-1.01; additive model: OR = 1.00, 95% CI = 0.95-1.05; dominant model: OR = 0.96, 95% CI = 0.90-1.03, Figure 2; recessive model: OR = 1.08, 95% CI = 0.99-1.18). When stratified by ethnicity, the significantly increased cancer risks were found among Asian population (homozygote model: OR = 1.46, 95% CI = 1.05-2.03; recessive model: OR = 1.39, 95% CI = 1.10-1.76) and Mixed population (homozygote model: OR = 1.17, 95% CI = 1.02-1.34; recessive model: OR = 1.17, 95% CI = 1.02-1.33). Stratified analyses by cancer types, smoking status and source of controls indicated no evidence of significant association between Tyr113His polymorphism and the cancer risk. Furthermore, individuals carrying Tyr/His or His/His genotype have a significantly reduced risk of lung cancer (heterozygote model: OR = 0.80, 95% CI = 0.65-0.98; dominant model: OR = 0.81, 95% CI = 0.68-0.98).

With respect to His139Arg polymorphism, similarly, the combined results did not show any association with the elevated risk of cancer for all genetic models (homozygote model: OR = 1.05, 95% CI = 0.93-1.18; heterozygote model: OR = 0.96; 95% CI = 0.91-1.01; additive model: OR = 0.99, 95% CI = 0.94-1.04; dominant model: OR = 0.97, 95% CI = 0.92-1.03, Figure 3; recessive model: OR = 1.02, 95% CI = 0.93-1.13). When stratified according to cancer types, no significant association with increased cancer risk was demonstrated in all subgroups for overall population. However, the result suggested a decreased risk for blood cancers (additive model: OR = 0.91, 95% CI = 0.83-0.99; dominant model: OR = 0.90, 95% CI = 0.81-0.99) and colorectal cancer (heterozygote model: OR = 0.92; 95% CI = 0.85-0.99). In the subgroup analysis by source of controls, smoking status and ethnicity, no significant association with cancer risk was observed in all subgroups (Table 2).

Heterogeneity is a potential issue that may affect the interpretation of the results. As for Tyr113His polymorphism, the heterogeneity was observed in all genetic models and the detailed data are shown in Table 1. With respect to His139Arg polymorphism (Table 2), the heterogeneity was detected in homozygote comparison, heterozygote model, dominant model and additive model. We therefore explored the source of heterogeneity by cancer type, ethnicity, and source of control by meta-regression in all comparisons with significant heterogeneity. As a result, for Tyr113His polymorphism, source of control may be the major source of heterogeneity in homozygote model (P = 0.024), heterozygote model (P = 0.000), dominant model (P = 0.002) and additive model (P = 0.005), but not recessive model. However, for His139Arg polymorphism, none of these variables showed statistically significant associations in multivariate meta-regression model (P > 0.05), suggesting factors mentioned above could not explain the heterogeneity among studies.

The influences of each individual study on the overall ORs for Tyr113His/His139Arg polymorphisms were evaluated. The results showed the pooled ORs of these two polymorphisms were not materially altered by the omission of any individual study, suggesting credibility for the conclusions (Additional file 4: Figure S1 and Additional file 5: Figure S2).

Begg’s funnel plot and Egger’s test were performed to assess the publication bias of literatures. The shape of funnel plots (Figures 4 and 5) did not reveal any evidence of asymmetry. The statistical results of Egger’s test still did not show publication bias for Tyr113His polymorphism (additive model: P = 0.146; homozygote comparison: P = 0.620; heterozygote model: P = 0.189; dominant model: P = 0.054; recessive model: P = 0.915) and His139Arg polymorphism (additive model: P = 0.125; homozygote comparison: P = 0.847; heterozygote model: P = 0.255; dominant model: P = 0.111; recessive model: P = 0.153).