The potential role of MGMT rs12917 polymorphism in cancer risk: an updated pooling analysis with 21010 cases and 34018 controls

In the present study, we aimed at determining the potential role of rs12917 polymorphism of the O-6-methylguanine-DNA methyltransferase (MGMT) gene in the occurrence of cancer. Based on the available data from the online database, we performed an updated meta-analysis. We retrieved 537 articles from our database research and finally selected a total of 54 case–control studies (21010 cases and 34018 controls) for a series of pooling analyses. We observed an enhanced risk in cancer cases compared with controls, using the genetic models T/T compared with C/C (P-value of association test <0.001; odds ratio (OR) = 1.29) and T/T compared with C/C+C/T (P<0.001; OR = 1.32). We detected similar positive results in the subgroups ‘Caucasian’, and ‘glioma’ (all P<0.05; OR > 1). However, we detected negative results in our analyses of most of the other subgroups (P>0.05). Begg’s and Egger’s tests indicated that the results were free of potential publication bias, and sensitivity analysis suggested the stability of the pooling results. In summary, the T/T genotype of MGMT rs12917 is likely to be linked to an enhanced susceptibility to cancer overall, especially glioma, in the Caucasian population.


Introduction
In humans, the O-6-methylguanine-DNA methyltransferase (MGMT) protein, encoded by the MGMT gene located on chromosome 10 (10q26) [1], is involved in the DNA repair process [2,3]. By means of methyl transfer, MGMT removes alkylating agents from the DNA direct reversal repair pathway and thus repairs the DNA [2,3]. Two potential functional polymorphisms have been identified in the MGMT gene, namely rs12917 (Leu84Phe) and rs2308321 (Ile143Val) [4,5]. In addition, the promoter methylation status of the gene is reportedly correlated with several clinical diseases, such as glioblastoma [6,7], gastric cancer [8], and oral carcinoma [9].
Both genetic and environmental factors contribute to the occurrence and progression of clinical cancers [10,11]. A number of studies have been conducted on the potential genetic effect of MGMT rs12917 polymorphism on its susceptibility to cancer, but the results were inconclusive. Before 2013, only three relative meta-analyses investigated the potential role of this polymorphism in the overall risk for cancer [12][13][14]. Based on the currently available data, we performed an updated meta-analysis to reassess the genetic relationship between MGMT rs12917 polymorphism and cancer risk. We enrolled a total of 54 case-control studies for the study.

Article screening strategy
We designed our inclusion and exclusion criteria according to Patient, Intervention, Comparison and Outcome and Study design (PICOS) principles. We ruled out duplicates and screened improper articles. Exclusion criteria were as follows: (P), non-cancer patients; (I), other variants, gene expression or methylation; (C), lack of study controls or P-value of Hardy-Weinberg equilibrium (HWE) <0.05; (O), lack of full genotype frequency data; (S), review, meta, poster, or meeting abstract. Eligible articles had to be designed as case-control studies, targetting the genetic relationship between MGMT rs12917 and cancer risk and containing the full genotype (C/C, C/T, T/T) frequencies in both cancer cases and negative controls.

Data extraction and quality assessment
After extracting usable data, we listed the basic information in tables. We assessed methodological quality via the Newcastle-Ottawa Scale (NOS) [15]. High-quality articles with NOS score > 5 were regarded as eligible and included in our statistical analysis.

Statistical analysis
We used STATA software version 12.0-SE (StataCorp, College Station, TX) to perform our analyses. We first assessed the inter-study heterogeneity using Cochran's Q statistic and the I 2 test. A P-value of Cochran's Q statistic < 0.1 or I 2 value > 50% was considered to show a high level of heterogeneity. We thus used the DerSimonian-Laird association test with a random-effects model. Otherwise, we used the Mantel-Haenszel association test with a fixed-effects model. The P-value of association test, summary odds ratio (OR), along with the corresponding 95% confidence interval (CI) could be obtained for the allele (T compared with C), homozygous (T/T compared with C/C), recessive (T/T compared with C/C+C/T), heterozygous (C/T compared with C/C), dominant (C/T+T/T compared with C/C), and carrier (T compared with C) models.
We performed subgroup analyses by race, cancer type, and control source. Additionally, we assessed possible publication bias by means of Begg's and Egger's tests and evaluated the robustness of the results through sensitivity analysis.

Results
Eligible case-control studies Figure 1 depicts the flowchart for the identification of eligible case-control studies. We initially obtained a total of 537 articles by searching four databases, including PubMed (245 articles), Cochrane Library (1 article), Embase (241 articles), and WANFANG (50 articles). We then excluded 233 duplicates plus another 258 articles based strictly on our screening strategy. Finally, we identified 46 full-text articles for inclusion [4,5,. After data extraction and quality evaluation, we enrolled a total of 54 case-control studies free of poor quality (all NOS score > 5) in our pooling analyses. The basic information and genotype frequency distribution are presented in Supplementary Table S1 and Table 1, respectively.

Meta-analysis data
First, we studied the association between the MGMT rs12917 polymorphism and cancer risk via an overall meta-analysis. As shown in Table 2, we included a total of 54 case-control studies with 21010 cases and 34018 controls under the genetic models of allele T compared with C, C/T compared with C/C, C/T+T/T compared with C/C, and carrier T compared with C; meanwhile, we included 50 studies with 20716 cases and 33608 controls under the models of T/T compared with C/C and T/T compared with C/C+C/T. For the homozygous, recessive and carrier genetic models, we performed a Mantel-Haenszel association test with a fixed-effects model, and we observed no high degree of heterogeneity (Table 2; all P-values of heterogeneity > 0.1; I 2 < 50%). For other models (all P-values of heterogeneity <0.001), we performed a DerSimonian-Laird association test with a random-effects model. Pooling data (Table 2) indicated an increased risk of cancer in cases compared with controls for the T/T compared with C/C (P-value of association test <0.001; OR = 1.29) and T/T compared with C/C+C/T (P<0.001; OR = 1.32) genetic models. Nevertheless, we failed to detect any statistical difference between cancer cases and negative controls under other genetic models (Table 2; all P>0.05). Forest plot data are shown in Figure 2 and Supplementary Figures S1-S5; they revealed that the T/T genotype of the MGMT rs12917 polymorphism was likely to be associated with an increased susceptibility to cancer.

Subgroup analysis data
Next, we carried out four subgroup analyses by race, cancer type, and control source. For the T/T compared with C/C model (Table 3) Table S5; P=0.020; OR = 0.92) under the carrier T compared with C model. Thus, the T/T genotype of MGMT rs12917 may have been associated with an increased risk of cancer in cases, especially the glioma cases, in the Caucasian population.

Publication bias and sensitivity analysis
Begg's and Egger's tests indicated that results were free of possible publication bias (Supplementary Table S6; P>0.05 for Begg's test, >0.05 for Egger's test). A Begg's funnel plot with pseudo-95% confidence limits under the T/T compared with C/C model is shown in Figure 4. In addition, we observed the same stable results in our subsequent sensitivity analysis; data from this analysis under the homozygous model ( Figure 5) are presented as an example.

Discussion
We observed conflicting conclusions about the genetic role of MGMT rs12917 polymorphism in its susceptibility to different cancers. For instance, the polymorphism seems to be associated with the risk of esophageal cancer in the Chinese population [41], but not in the Kashmiri population [50]. This merits a quantitative synthesis via the meta-analytic approach. Although there were already three meta-analyses of the MGMT rs12917 polymorphism and its role in the overall risk for cancer [12][13][14], expanding the sample size and employing a distinct analysis strategy led to better results in our updated pooling analysis.  -, OR (95% CI) data were not provided, when P-value of association > 0.05.
We did our best to gather candidate articles from four online databases. After screening them based on strict inclusion and exclusion criteria, we enrolled only the case-control studies that were of high quality and those that followed HWE. We ultimately included a total of 46 articles in our updated meta-analysis. After data extraction, we enrolled 54 case-control studies with 21010 cases and 34018 controls in the meta-analysis. We used the carrier, allele,  -, OR (95% CI) data was not provided, when P-value of association > 0.05.  homozygous, recessive, heterozygous, and dominant genetic models, and also confirmed the stability of the statistical results via sensitivity analysis.
Our updated pooling analysis data demonstrated that cases had an overall enhanced risk for cancer when compared with negative controls under the T/T compared with C/C and T/T compared with C/C+C/T genetic models, especially in the European-descended population, which is partly consistent with the data of previous analyses [12][13][14].
Moreover, we observed that the MGMT rs12917 polymorphism is likely to be associated with the susceptibility to glioma, which is partly in-line with the two studies on the association between DNA repair gene polymorphisms and glioma risk [67,68]. Nevertheless, owing to the limitation of sample size, the previous three meta-analyses of the overall risk for cancer did not conduct subgroup analyses of 'glioma' [12][13][14].
Some of the limitations to our meta-analysis are as follows: (1) Although the sample sizes enrolled were quite large (21010 cases and 34018 controls), genotype data were very limited in many subgroup analyses. For instance, we used only three case-control studies in our analyses of the subgroups for gastric [33,44,47], blood [52,53,56], and colorectal [54,55] cancers. Even for the subgroup analysis of 'glioma' , with positive correlations under the T/T compared with C/C and T/T compared with C/C+C/T models, only five case-control studies [23,29,30,42,48] were included.
(2) We did not investigate the genetic effects of the MGMT rs12917 polymorphism in combination with other variants, such as rs2308321 of MGMT, rs25487 of X-ray cross-complementing group 1 (XRCC1), and rs13181 of xeroderma pigmentosum complementation group D (XPD), in certain specific cancers.
(3) We extracted certain demographic information such as the mean age at diagnosis and the sex of subject, but not other confounding factors such as lifestyle and clinical features. Moreover, we did not perform the relevant stratified meta-analyses due to lack of sufficient usable data.
(4) We detected significant heterogeneity amongst studies under the allele T compared with C, C/T compared with C/C, C/T+T/T compared with C/C, and carrier T compared with C genetic models. Complicating factors such as race and cancer type may be sources of inter-study heterogeneity. For instance, we detected decreased levels of heterogeneity in the 'Caucasian' and 'esophageal cancer' subgroups. Although we observed a positive conclusion in the 'glioma' subgroup, we failed to detect reduced inter-study heterogeneity. Only five case-control studies [23,29,30,42,48] were enrolled.
(5) There may be other undetected or unpublished articles containing potential eligible case-controls in other geographical locations or languages; in other words, our study may suffer from selection bias.
(6) Last but most important, our meta-analysis found a positive conclusion between MGMT rs12917 and the risk of cancer in general for the T/T compared with C/C and T/T compared with C/C+C/T models. Considering the distinct etiopathogenesis or pathogenesis of different kinds of cancers, more studies of large-scale populations of different ethnicities are required for a more scientific elucidation of MGMT rs12917's functional role in each particular cancer type.
To sum up, our updated pooling analysis offered additional evidence that MGMT rs12917 polymorphism is likely to be associated with an enhanced susceptibility to cancer overall, especially glioma, in the Caucasian population.

Author contribution
Z.S. and H.W. conceived and designed the study. Z.S. and M.K. were responsible for the data extraction and statistical analysis. Z.S. wrote the manuscript and H.W. revised the manuscript.