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

Aim To evaluate the association between the EPHX1 Tyr113His and His139Arg polymorphisms in the EPHX1 gene and the risk of head and neck cancer. Materials and Methods Studies on the association of EPHX1 Tyr113His and His139Arg polymorphisms with HNC performed up until June 1st, 2014, were identified using a predefined search strategy. Summary odds ratios (ORs) and 95% confidence intervals (CIs) were used to evaluate the strength of these associations. Results In this meta-analysis, 10 case-control studies, which included 9 studies of Tyr113His (1890 cases and 1894 controls) and 10 studies of His139Arg polymorphisms (1982 cases and 2024 controls), were considered eligible for inclusion. Overall, the pooled results indicated that the EPHX1 Tyr113His polymorphism was significantly associated with increased HNC risk (Tyr/His vs. Tyr/Tyr, OR = 1.26, 95%1.02–1.57;His/His+ Tyr/His vs. Tyr/Tyr, OR = 1.29, 95% I = 1.03–1.61). However, no significant association was found between the His139Arg polymorphism and HNC risk. In the subgroup analysis, a statistically significant association between the EPHX1 Tyr113His polymorphism and HNC was observed in population-based case-control studies (PCC), which involved less than 500 participants and genotype frequencies in HWE. This association showed minimal heterogeneity after excluding studies that were determined to contribute to heterogeneity. After categorizing the studies by publication time, a sensitivity analysis and cumulative meta-analysis of the two associations were conducted, and the results of the two analyses were consistent. Conclusion Our meta-analysis suggests that EPHX1 Tyr113His polymorphism may be a risk factor for HNC, while the EPHX1 His139Arg polymorphism has no association with HNC risk.


Results
In this meta-analysis, 10 case-control studies, which included 9 studies of Tyr113His (1890 cases and 1894 controls) and 10 studies of His139Arg polymorphisms (1982 cases and 2024 controls), were considered eligible for inclusion. Overall, the pooled results indicated that the EPHX1 Tyr113His polymorphism was significantly associated with increased HNC risk (Tyr/His vs. Tyr/Tyr, OR = 1.26, 95%1.02-1.57;His/His+ Tyr/His vs. Tyr/Tyr, OR = 1.29, 95% I = 1.03-1.61). However, no significant association was found between the His139Arg polymorphism and HNC risk. In the subgroup analysis, a statistically significant association between the EPHX1 Tyr113His polymorphism and HNC was observed in population-based case-control studies (PCC), which involved less than 500 participants and genotype frequencies in HWE. This association showed minimal heterogeneity after excluding studies that were determined to contribute to heterogeneity. After categorizing the studies by publication time, a sensitivity analysis and cumulative meta-analysis of the two associations were conducted, and the results of the two analyses were consistent.

Introduction
Head and neck cancer (HNC) is the fifth most common type of human cancer [1], with approximately 434,000 new cases diagnosed annually worldwide [2]. HNC occurs more frequently in developing countries, such as India, Brazil, and Thailand [3][4]. Although the mechanism underlying HNC is still not fully understood, accumulating evidence suggests that tobacco smoking, drinking alcohol, and chewing betel quid are three main risk factors for HNC [5][6]. Several previous studies reported that tobacco smoke and alcohol metabolites may induce defects in DNA and genomic instability, which can lead to mutations and malignant transformation [7][8][9].
Microsomal epoxide hydrolase (mEH) (EPHX1), which is one of the phase I detoxification enzymes found on the endoplasmic reticulum in many tissues, plays a role in the metabolism of potential carcinogens in tobacco smoke [10]. Epoxides, with their highly reactive oxidative metabolites, are often the most toxicologically active drugs and environmental chemical [11]. EPHX1 catalyzes the hydrolysis of reactive epoxides to trans-dihydrodiols, and some dihydrodiols can be subsequently metabolized to highly mutagenic polycyclic hydrocarbon diol epoxides [12][13]. Therefore, EPHX1 plays a dual role in the activation and detoxification of procarcinogens. Moreover, some studies have reported that the role of EPHX1 in carcinogenesis may depend on exposure to different environmental substrates [14].
The EPHX1 gene is 35.48 kb in length with nine exons and eight introns [11], and it is located on chromosome 1q42.1. More than 110 single nucleotide polymorphisms (SNPs) have been identified in the EPHX1 gene, and these SNPs can be found in the dbSNP database (http:// www.ncbi.nlm.nih.gov/SNP). Two alleles of EPHX1 are common and have been associated with altered enzyme activity [15]. With the differential effect of EPHX1 alleles on the detoxification of procarcinogens, we proposed that the two functional polymorphisms may affect HNC risk.
Several previous studies were conducted to evaluate the association between EPHX1 polymorphisms (Tyr113His and His139Arg) and the risk of HNC in different populations; however, the results of these studies were inconclusive. Up to now, no remarkable evidence has been presented to suggest the precise role of EPHX1 (Tyr113His and His139Arg) in HNC. Accordingly, a comprehensive evaluation of the associations between EPHX1 polymorphisms (Tyr113His and His139Arg) and HNC risk is urgently needed. In this study, we reviewed the existing literature and performed a meta-analysis to evaluate the association between Tyr113His and His139Arg polymorphisms in the EPHX1 gene and HNC susceptibility.

Search strategy
We conducted a computerized literature search of Medline, PubMed, EMBASE and China National Knowledge Infrastructure Whole Article Database (CNKI) prior to the June 1st, 2014 using the following key words: ('microsomal epoxide hydrolase' OR 'EPHX1' OR 'mEH') AND ('head and neck' OR 'oral' or 'oropharynx' OR 'larynx' OR 'pharynx') AND ('cancer' OR 'tumor' OR 'carcinoma') AND ('polymorphism' OR 'variant' OR 'allele' OR 'genotype'). The search was not restricted by language but was restricted to human studies. The references cited by the identified original studies and relevant reviews were searched at the same time. Furthermore, we contacted some experts in the area to obtain information about unpublished results or ongoing trials nearing completion.

Inclusion and exclusion criteria
Relevant articles and abstracts were selected and reviewed independently by two of the authors (Hong Chen and Lin Ge). The following inclusion criteria were used for published studies: (i) case control studies that were performed to assess the association between at least one of the two polymorphisms (Tyr113His and/or His139Arg) and HNC risk; (ii) papers that clearly describe HNC diagnoses and the sources of cases and controls; (iii) papers that include sufficient genotype data such that the odds ratios (ORs) and 95% confidence intervals (CIs) can be calculated. If the samples described in two studies overlapped, we selected the study in which the larger sample was identified. The major exclusion criteria were as follows: (i) papers classified as reviews, editorials, letters, comments, or case reports; (ii) cell line studies; (iii) duplicate studies; (iv) studies in which the participants had distant metastasis or synchronous malignancy in other organs.

Data extraction
Working independently, two authors (Hong Chen and Lin Ge) extracted the data from all eligible publications according to the inclusion criteria listed above. Any disagreements in data quality scores and abstraction were assessed further and dealt with by discussion between the authors. The following characteristics were extracted: the first author, year of publication, country of origin of participants, ethnicity, cancer type, source of control group (populationor hospital-based controls), number of cases and controls, genotypes, minor allele frequency (MAF) in controls, and P for Hardy-Weinberg equilibrium (HWE) ( Table 1). In this metaanalysis, we defined a population-based case-control study (PCC) as a study that enrolled controls from healthy populations, and we defined a hospital-based case-control study (HCC) as a study that enrolled controls from hospitalized patients, as reported in previous studies [11,16].

Data synthesis
All data from the eligible studies were abstracted. We first evaluated HWE in the controls for each eligible study using the chi-square goodness of fit test, and P>0.05 was considered statistically significant. Crude odds ratios (ORs) and 95% confidence intervals (95% CIs) were used to estimate the strength of the association between HNC and the EPHX1 Tyr113His and His139Arg polymorphisms. A Z-test was also used, and P values <0.05 indicated statistically significant associations. Pooled ORs were estimated for genetic models including the dominant model, recessive model, homozygote comparison and heterozygote comparison. Using EPHX1 Tyr113His as an example, the codominant model would be His/His vs. Tyr/Tyr and Tyr/His vs. Tyr/Tyr, the dominant model would be His/His+ Tyr/His vs. Tyr/Tyr, and the recessive model would be His/His vs. Tyr/His+ Tyr/Tyr [17].
We tested statistical heterogeneity by using Cochran's Q statistic [18] and the I 2 statistic [19], where P<0.1 was considered significant heterogeneity, and I 2 >50% indicated large heterogeneity. If heterogeneity existed, a random-effects model (the DerSimonian and Laird method) was adopted [20]; otherwise, a fixed-effects model (the Mantel-Haenszel method) was adopted [21] as appropriate. In addition, heterogeneity was also examined in subgroup analyses by ethnicity, source of controls (HCC/PCC), study sample size (500/<500 subjects), matched control (Yes/No), and HWE in controls (Yes/No). The reliability of the results was evaluated by performing sensitivity analysis. A funnel plot, Begg's rank correlation method [22] and the Egger's weighted regression method [23] were adopted to statistically assess publication bias (P < 0.05 was considered statistically significant). All analyses were performed using STATA software, version 12.0 (STATA Corp., College Station, TX, USA).

Description of the included trials
267 publications relevant to the search words were identified, and all of them were written in English. The two reviewers (Hong Chen and Lin Ge) independently screened the title and abstract with the focus question, and12 articles were identified [24][25][26][27][28][29][30][31][32][33][34][35]. Of these articles, three were excluded after full-text assessment for eligibility for the following reasons: one study was irrelevant, one study did not examine the association between these two polymorphisms (Tyr113His and His139Arg), and one study (an HNC risk study) did not yield an accurate result. A flow chart illustrating the study selection and specific reasons for exclusion is presented in Fig 1. Thus, 10 studies reported in 9 articles [24][25][26][27][28][29][30][31][32], which included 9 studies of Tyr113His (1890cases and 1894 controls) and 10 studies of His139Arg polymorphisms (1982 cases and 2024 controls), were found to match our inclusion criteria. One article [31] mentioned two independent case-control studies (Caucasians and African Americans), and the article was thus treated as two separate estimates.
The characteristics of the studies included in the meta-analysis are presented in Table 1. The ethnicities studied included Asians, African Americans, Caucasians, and mixed ethnicities. The studies were carried out in India, the Netherlands, Spain, Germany, Italy, Chile, France, and the USA. Among these studies, 5 studies focused on oral/pharyngeal/laryngeal cancer, 3 on oral /laryngeal cancer, 1 on oral cancer and 1 on laryngeal cancer. Most of the studies involved extraction of DNA from peripheral blood and employed the classic PCR-RFLP assay and PCR for genotyping. The genotype distributions among the controls of all studies followed HWE except for two studies [28,31] that examined the Tyr113His polymorphism.

Quantitative synthesis
The EPHX1 Tyr113His polymorphism and HNC susceptibility. Eight articles [24][25][26][27][28][29][31][32] included in this meta-analysis described 9 case-control studies, with 1890 cases and 1894 controls, revealing an association between the EPHX1 Tyr113His polymorphism and HNC susceptibility. The main results of this pooled analysis are presented in Table 2  To determine the reason underlying the potential underestimation of the true effect of these polymorphisms on HNC risk, we performed subgroup analysis according to ethnicity, source of controls, study sample size, matched controls, and HWE in controls. Different ethnicities  Table 2).
The EPHX1 His139Arg polymorphism and HNC susceptibility. Nine articles [24][25][26][27][28][29][30][31][32] were included in this meta-analysis that described 10 case-control studies, with 1982 cases and 2024 controls, reporting the association between the EPHX1 His139Arg polymorphism and HNC susceptibility. The main results of this pooled analysis are presented in Table 3 and Fig 3 shows forest plots illustrating the effect of the EPHX1 Tyr113His polymorphism on HNC risk.

Heterogeneity analysis
For the Tyr113His polymorphism, significant heterogeneity was found in the overall comparisons in four genetic models: dominant model P = 0.01, recessive model P = 0.01, homozygote comparison P = 0. 00, and heterozygote comparison P = 0.02.
Significant heterogeneity was also detected for the His139Arg polymorphism. No significant heterogeneity was found in the homozygote comparison or the recessive model comparison; however, significant heterogeneity was detected in the heterozygote comparison and dominant model (dominant model P = 0.02, recessive model P = 0.60, homozygote comparison P = 0.51, and heterozygote comparison P = 0.00.). Galbraith plot analyses were used to evaluate the potential sources of heterogeneity in this article. In this analysis, three studies [24,28,32] were found to be contributors of heterogeneity for the Tyr113His polymorphism (S1 Fig). After excluding these three outlier studies, we re-evaluated the association with reduced heterogeneity (His/His vs. Tyr/Tyr: P = 0.55; Tyr/His vs. Tyr/Tyr: P = 0.66; His/His+ Tyr/His vs. Tyr/Tyr: P = 0.86; His/His vs. Tyr/His+ Tyr/Tyr: P = 0.32). Regarding the His139Arg polymorphism, two studies [24,31] were found to be contributors of heterogeneity (S1 Fig), and this heterogeneity was significantly reduced after the two studies were excluded(Arg/Arg vs. His/His: P = 0.58; Arg/Arg vs. Arg/His+ His/His: P = 0.89).

Heterogeneity analysis
For the Tyr113His polymorphism, significant heterogeneity was found in the overall comparisons in four genetic models: dominant model P = 0.01, recessive model P = 0.01, homozygote comparison P = 0. 00, and heterozygote comparison P = 0.02.
Significant heterogeneity was also detected for the His139Arg polymorphism. No significant heterogeneity was found in the homozygote comparison or the recessive model comparison; however, significant heterogeneity was detected in the heterozygote comparison and dominant model (dominant model P = 0.02, recessive model P = 0.60, homozygote comparison P = 0.51, and heterozygote comparison P = 0.00.). Galbraith plot analyses were used to evaluate the potential sources of heterogeneity in this article. In this analysis, three studies [24,28,32] were found to be contributors of heterogeneity for the Tyr113His polymorphism (S1 Fig). After excluding these three outlier studies, we re-evaluated the association with reduced heterogeneity (His/His vs. Tyr/Tyr: P = 0.55; Tyr/His vs. Tyr/Tyr: P = 0.66; His/His+ Tyr/His vs. Tyr/Tyr: P = 0.86; His/His vs. Tyr/His+ Tyr/Tyr: P = 0.32). Regarding the His139Arg polymorphism, two studies [24,31] were found to be contributors of heterogeneity (S1 Fig), and this heterogeneity was significantly reduced after the two studies were excluded (Arg/Arg vs. His/His: P = 0.58; Arg/Arg vs. Arg /His+ His/His: P = 0.89).

Sensitivity analysis
Sensitivity analysis was conducted to detect the influence of a single study on the overall metaanalysis by omitting one study at a time. Regarding the association of the EPHX1 Tyr113His polymorphism with HNC risk, the study [32] seemed to have the greatest influence on the overall pooled estimates. Without that study, a re-evaluation of the meta-analysis showed that the OR became 1.13 (95% CI: 0.91-1.41). Compared with the previous result (OR = 1.35, 95% CI: 0.93-1.96), this result was not obviously different, which indicated the stability of the results. Regarding the association of the EPHX1 His139Arg polymorphism with CRC risk, the study [31] seemed to have the most influence on the overall pooled estimates. After the removal of this study, the meta-analysis showed that the OR changed from 1.11 (95% CI: 0.74-1.65) to 1.98 (95% CI: 0.84-1.14), which indicated the stability of the results. When the studies that were not in HWE were excluded, the estimated pooled OR was only altered in the homozygote comparison for the Tyr113His polymorphism ( Table 2), demonstrating that our results were reliable.

Cumulative meta-analysis
Cumulative meta-analyses were used to examine how the evidence has changed over time. In this article, we conducted cumulative meta-analyses of the 2 associations by categorizing the studies by publication time. S2 Fig displays the results from the cumulative meta-analysis of the association between the EPHX1 Tyr113His polymorphism (His/His+ Tyr/His vs. Tyr/Tyr) and overall HNC in chronological order. The results showed a minimally significant association between the EPHX1 Tyr113His polymorphism and HNC risk when the data were categorized by publication year. A lack of significant associations between the EPHX1 His139Arg polymorphism and HNC risk was observed when the data were categorized by publication year (S2 Fig).

Publication bias
Publication bias appears if no significant findings remain unpublished, which will artificially expand the apparent magnitude of an effect. In this meta-analysis, funnel plot, Begg's and Egger's tests were used to evaluate publication bias of the literature on HNC. S3 Fig and S4 Fig  show funnel

Discussion
Some previous articles reported a significant association between EPHX1polymorphism and several human diseases, such as type 2 diabetes mellitus [36], alcohol dependence [37], and chronic obstructive pulmonary disease [38], while others found no such association [39][40][41]. In this meta-analysis, the pooled results indicated that subjects carrying the EPHX1 Tyr113His genotype had an increased risk of developing HNC. Our approach also allowed us to identify some potential differences, such as ethnicity, source of controls, study sample size and others.
In different ethnicities, the EPHX1 polymorphism did not have a significant association with HNC risk. Our result differed from that of Zhong J H [42] and Liu F [11], which may be due to the limited number of studies in these analyses. Thus, well-designed studies with a large sample size that evaluate multiple ethnicities are required. The results of meta-analyses often depend on control selection procedures [43], such as the source of participants. In this metaanalysis, a statistically significant association between the EPHX1 Tyr113His polymorphism and HNC was observed in PCC studies, but not HCC studies, as the latter may contribute to some selection biases. Controls from a hospital population may not be a representative of the general population, especially when the investigated genotypes are associated with disease conditions [11]. In a subgroup analysis of controls in HWE, the 2 associations were only statistically significant in studies in which the controls followed HWE. Potential errors may occur in studies examining populations that deviate from HWE, such as laboratory or genotyping errors, population stratification, selection bias in the choice of controls and other confounding factors [44][45]. Simultaneously, the limited sample size may be an obstacle to obtaining accurate results. In this meta-analysis, a statistically significant association between the EPHX1 Tyr113His polymorphism and HNC was observed in studies enrolling less than 500 participants.
The degree of heterogeneity in a meta-analysis partially determines the difficulty in drawing overall conclusions [46]. Q-test and I 2 statistics were calculated to test the impact of heterogeneity on the overall comparisons and some subgroup analyses where P<0.1 was considered significant heterogeneity and I 2 >50% indicated large heterogeneity. Galbraith plot analyses were used to evaluate the potential sources of heterogeneity in this article. In this analysis, three studies [24,28,32] were found to be contributors of heterogeneity for the Tyr113His polymorphism, while two studies [24,31] were found to be contributors of heterogeneity for the His139Arg polymorphism. The heterogeneity was significant reduced after excluding these outlier studies; however, the conclusion was still consistent in the overall comparisons. Furthermore, although we conducted a sensitivity analysis by omitting one study at a time, we found no significant difference, indicating that our results were statistically reliable. Cumulative meta-analyses were also conducted to examine how the evidence has changed over time, and the results were borderline significant. However, the meta-analysis described herein had some limitations. First, unadjusted OR estimates were adopted in this meta-analysis because we could not obtain sufficient information to calculate adjusted ORs with potential confounders, such as age and sex. Second, the study size and the sample size for some subgroup analyses were limited, which contributes to the possibility of type I and type II errors. Thus, the reliability of our results may need to be further tested. Third, similar to previous studies [11], this meta-analysis only focused on analyzing two individual SNPs-a combination (two-SNP) analysis was not performed. However, the activity of EPHX1 can be affected by single Tyr113His and His139Arg polymorphisms or a combination of these polymorphisms [46][47]. In this analysis, only combination (two-SNP) analyses were considered, which prevented us from performing a pooled analysis. Finally, significant heterogeneity was found in studies examining EPHX1 polymorphisms, and in some studies, the genotype distribution showed a deviation from HWE. As a retrospective study, the quality of a meta-analysis is dependent on the methodological limitations of the original studies. To minimize bias, we first designed a detailed protocol and performed a meticulous search as a predefined search strategy.
In conclusion, this meta-analysis evaluates the relationship between EPHX1 polymorphisms and HNC risk. Our meta-analysis suggests that the EPHX1 Tyr113His polymorphism may be a risk factor for HNC, while the EPHX1 His139Arg polymorphism has no association with HNC risk. In subgroup analysis, a statistically significant association between the EPHX1 Tyr113His polymorphism and HNC was observed in population-based case-control studies, studies that enrolled more than 500 participants and studies in which the genotype frequencies were in HWE. When stratified by ethnicity, no significant associations were found in this study; thus, well-designed studies with a large sample size that examine multiple ethnicities are required. Moreover, further meta-analyses must be performed to estimate the effects of both single SNP analysis and combination (two-SNP) analysis. Gene-environment interactions should also be examined to determine the association between the EPHX1 polymorphisms and HNC risk.