Individual and combined effects of GSTM1, GSTT1, and GSTP1 polymorphisms on breast cancer risk: A meta-analysis and re-analysis of systematic meta-analyses

Background Fourteen previous meta-analyses have been published to analyze the polymorphisms of individual GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val on breast cancer (BC) risk. However, their meta-analyses did not explore the combined effects of the three genetic polymorphisms on BC risk. In addition, they did not evaluate the credibility of statistically significant associations. Furthermore, a multitude of new articles have been published on these themes, and therefore a meta-analysis and re-analysis of systematic previous meta-analyses were performed to further explore these issues. Objectives To determine the association between the individual and combined effects of GSTM1, GSTT1, and GSTP1 polymorphisms on breast cancer risk. Methods Crude odds ratios (ORs) and their 95% confidence intervals (CIs) were applied to estimate the association between individual and combined effects of GSTM1, GSTT1, and GSTP1 polymorphisms on BC risk. To evaluate the credibility of statistically significant associations in the current and previous meta-analyses, we applied the the false-positive report probabilities (FPRP) test and the Venice criteria. Results 101 publications were selected to evaluate the individual and combined effects of GSTM1, GSTT1 and GSTP1 polymorphisms on BC risk. Overall, statistically significant elevated BC risk was found in any individual and combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms. However, when we restricted studies only involving with high-quality, matching, HWE, and genotyping examination performed blindly or with quality control, significantly increased BC risk was only found in overall population for GSTM1 null genotype, among all populations, Caucasians, and postmenopausal women for the combined effects of GSTM1 and GSTT1 polymorphisms, and in overall analysis for the combined effects of GSTM1, GSTT1, and GSTP1 IIe105Val polymorphisms. Further, less-credible positive results were identified when we evaluated the credibility of positive results of the current and previous meta-analyses. Conclusions This meta-analysis indicates that the individual and combined effects of GSTM1, GSTT1 and GSTP1 polymorphisms may be not associated with increased BC risk.


Introduction
Breast cancer (BC) is one of the most common diseases and an important public health challenge among women worldwide, although the incidences of BC are not the same in different countries and ethnic groups [1,2]. Risk factors that have been confirmed including age, family history and several reproductive factors only explain one-third of BC cases [3]. Studies on the pathologic mechanism of BC remain enigmatic, and it is a multifactorial and polygenic disease which may be influenced by both environmental and genetic factors [4,5]. Therefore, studies on gene polymorphisms have become much more important in the progression of BC worldwide [6,7].
In recent years, some genes have been confirmed as potential cancer susceptible genes. Glutathione S-transferases (GSTs) are overwhelmingly important genes, which play key role in the detoxification of toxic, potentially carcinogenic compounds, and a host of basic physiological processes of the human body [8][9][10][11]. In human, five classes of GST enzymes have been found (α,μ,π,σ, andθ) [12] and each class is encoded by an independent gene or family genes (such as GSTA, GSTM, GSTP, GSTO, and GSTT genes). Among these genes, both GSTM1 and GSTT1 genes show deletion polymorphisms (null genotype) [13,14], which cause the absence of expression and enzyme activity loss [15]. They are located on chromosome 1 (1p13.3) and chromosome 22 (22q11.2), respectively [16]. An codon 105 A to G mutation at exon 5 in GSTP1 polymorphism results in a change isoleucine (IIe) to valine (Val), which also decreases enzymatic activity [17,18]. Therefore, the three gene mutations may increase BC risk on the basis of their biological effects.
In 1993, the first publication was reported on the association between GSTM1 null genotype and BC cancer susceptibility [Reference 1 in S1 Appendix]. The first study investigated the association between individual GSTT1 null genotype and the combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms on BC cancer risk was published in 1998 [Reference 5 in S1 Appendix], and it is the first article that was published to explore the association between GSTP1 IIe105Val polymorphism and BC cancer risk [Reference 110 in S1 Appendix]. So far, 116 publications [References 1-116 in S1 Appendix] have been reported on these themes. Nevertheless, the results of these studies were contradictory. Fourteen previous meta-analyses [19][20][21][22][23][24][25][26][27][28][29][30][31][32] have been published to analyze the individual GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms on BC risk. However, their meta-analyses did not conduct the combined effects of the three genes on BC risk, in addition, they did not evaluate the credibility of statistically significant associations, furthermore, a lot of new studies have been published, and therefore a meta-analysis and reanalysis of previous meta-analyses were carried out to further explore the individual and combined effects of these genes on BC risk.

Search strategy
Literature search was performed using PubMed and CNKI databases in this meta-analysis (update to 18 May, 2018). The following search strategy was applied: (glutathione S-transferase T1 OR GSTT1 OR glutathione S-transferase P1 OR GSTP1 OR glutathione S-transferase M1 OR GSTM1) AND breast AND (polymorphism OR genotype OR allele OR variant OR mutation). Language was not restricted in the present meta-analysis. It was implemented to identify additional studies manually (references of the original and review studies). Finally, the corresponding authors were contacted via e-mail if necessary.

Inclusion and exclusion criteria
The eligible publications were selected applying the following criteria: (1) case-control study; (2) detailed genotype frequencies were afforded between case and control groups; (3) studies must assess the association between the individual and combined effects of GSTM1 present/ null, GSTT1 present/null and GSTP1 IIe105Val polymorphisms on BC risk. Studies were removed if they were case reports, duplicate data or incomplete data, meta-analysis, and so on.

Data extraction
Information was carefully collected independently by two investigators from all selected studies. Potential disagreements were judged through the corresponding authors if necessary. The following information was collected: first author's surname, year of publication, country, ethnicity, source of cases, source of controls, type of controls, matching, single nucleotide polymorphism (SNP), sample size, and genotype frequencies of the individual and combined effects of GSTM1, GSTT1 and GSTP1 polymorphisms on BC risk.

Quality score assessment
The quality of the studies were appraised independently by two of all authors. We designed quality assessment criteria on the basis of two previous meta-analyses [33,34]. S1 Table lists the scale for quality assessment of molecular association studies of BC. S3 and S4 Tables list the quality assessment by included studies of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms with BC risk. They were considered as low quality studies if quality scores were �10, while scores of > 10 were regarded as high quality in this meta-analysis.

Statistical analysis
We applied crude odds ratios (ORs) and their 95% confidence intervals (CIs) to estimate the association between individual and combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms on BC risk. We used null vs. present model to calculate the pooled ORs with their 95% CIs for the GSTM1 present/null and GSTT1 present/null polymorphisms. Analysis was conducted employing the following genetic models for GSTP1 IIe105Val polymorphism: Val/Val vs. IIe/IIe, IIe/Val vs. IIe/IIe, Val/Val vs. IIe/IIe + IIe/Val, Val/Val + IIe/Val vs. IIe/IIe, and Val vs. IIe. For the combined effects of GSTM1 present/null and GSTT1 present/null polymorphisms, we applied the following genetic models: + − vs. + +, − + vs. + +, − − vs. + +, (+ −) + (− +) vs. + +, (− −) + (+ −) + (− +) vs. + +, and − − vs. (+ +) + (+ −) + (− +). − − was GSTM1 null/GSTT1 null, + + was GSTM1 present/GSTT1 present, + − was GSTM1 present/GSTT1 null, and − + was GSTM1 null/GSTT1 present. The following genetic models were used for the combined effects of GSTM1 present/null and GSTP1 IIe105Val polymorphisms: GSTM1 null/GSTP1 31,32,33,35,45,54,76,82,91,107,115 in S1 Appendix] were removed because their sample had been overlapped with another eleven studies [8,9,17,23,41,47,55,64,71, 88, 105 in S1 Appendix]. In the end, 101 publications were selected to evaluate the individual and combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms on BC risk. S2 Table shows the general characteristics of studies included in this meta-analysis. There were 88 case-control studies from 82 publications on GSTM1 present/null polymorphism (involving 28,676 BC cases and 32,539 controls, S5 Table), 67 case-control studies from 62 publications on GSTT1 present/null polymorphism (involving 23,092 BC cases and 26,381 controls, S5 Table), 56 case-control studies from 53 articles on GSTP1 IIe105Val polymorphism (involving 25,331 BC cases and 27,424 controls, S5 Table), 31 case-control studies from 30 articles on the combined effects of both GSTM1 and GSTT1 null genotypes (involving 10,497 BC cases and 10,242 controls, S8 Table), 15 case-control studies on the combined effects of GSTM1 present/null and GSTP1 IIe105Val polymorphisms (involving 6,272 BC cases and 6,739 controls, S10 Table), 13 case-control studies on the combined effects of GSTT1 present/null and GSTP1 IIe105Val polymorphisms (involving 5,413 BC cases and 5,567 controls, S11 Table), and 13 case-control studies on the combined effects of three GSTM1 present/null, GSTT1 present/ null, and GSTP1 IIe105Val polymorphisms (involving 5,395 BC cases and 5,544 controls, S12 Table). In addition, twenty, fifteen, ten, and seven case-control studies were conducted to analyze GSTM1 null genotype (including 7,934 BC cases and 11,059 controls), GSTT1 null genotype (including 6,786 BC cases and 9,477 controls), GSTP1 IIe105Val (including 3,448 BC cases and 3,676 controls), and the combined effects of GSTM1 and GSTT1 polymorphisms (including 1,916 BC cases and 2,268 controls) among postmenopausal women, and seventeen, twelve, fifteen, and six case-control studies were conducted to analyze GSTM1 null genotype (including 2,840 BC cases and 3,393 controls), GSTT1 null genotype (including 1,605 BC cases and 1,830 controls), GSTP1 IIe105Val (including 8,493 BC cases and 11,040 controls), and the combined effects of GSTM1 and GSTT1 polymorphisms (including 981 BC cases and 1,185 controls) among premenopausal women, respectively, as shown in S6-S9 Tables. Furthermore, there were five, three, and zero current smoking studies, seven, six, and one past smoking studies, and eleven, nine, and three no-smoking studies on GSTM1, GSTT1, and GSTP1 polymorphisms, respectively, as shown in S7 Table. Finally, there were 31 high-quality studies  and 57 low-quality studies on GSTM1 present/null, 23 high-quality studies and 44 low-quality  studies on GSTT1 present/null, 30 high-quality studies and 26 low-quality studies on GSTP1  IIe105Val, 13 high-quality studies and 18 low-quality studies on the combined effects of GSTM1 and GSTT1, nine high-quality studies and six low-quality studies on the combined effects of GSTM1 and GSTP1, eight high-quality studies and five low-quality studies on the combined effects of GSTT1 and GSTP1, and eight high-quality studies and five low-quality studies on the combined effects of GSTM1, GSTT1, and GSTP1 polymorphisms as determined by quality assessment of molecular association studies (S1 Table).

Quantitative synthesis
At the overall analysis, the GSTM1 null genotype was associated with elevated BC risk (OR = 1.12, 95% CI = 1.06-1.09). In addition, significantly elevated BC risk was also observed in a slice of subgroups, such as Asians, population-based studies, healthy women, cancer-free women, cancer-free patients, matching studies, no matching studies, large-sized studies, small-sized studies, high-quality studies, low-quality studies, postmenopausal and premenopausal women, as shown in Table 1.
The characters who carried GSTT1 null genotype had a significantly elevated BC risk (OR = 1.15, 95% CI = 1.06-1.25) in overall analysis. Significant association was also shown in quite a few subgroups, for instance, Caucasians, hospital-based studies, healthy women, cancer-free patients, matching studies, no matching studies, large-sized studies, small-sized studies, low-quality studies, and premenopausal women, as shown in Table 2.
No significantly raised BC risk was observed for GSTP1 IIe105Val polymorphism in pooling all studies. However, significantly increased BC risk was yielded in some subgroup analyses, such as Asians, Indians, hospital-based studies, no matching, and low-quality studies, as shown in Table 3.
Finally, we also performed a pooled analysis to investigate the combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val polymorphisms on BC risk. The results indicated that a significantly increased BC risk was found in overall populations (  Table 7.

Heterogeneity and sensitivity analyses
Heterogeneity was observed in the current meta-analysis, as shown in Tables 1-7. Then, we evaluated heterogeneity source by applying a meta-regression analysis method. The results suggested that source of controls (P = 0.027 for + − vs. + +), type of controls (P = 0.005 for − − vs. + +), and quality score of articles (− + vs. + +: P = 0.045 for − − vs. + +) were source of heterogeneity between the combined effects of GSTM1 present/null and GSTT1 present/null polymorphisms with BC risk. For the combined effects of GSTM1 present/null and GSTP1 IIe105Val polymorphisms, matching (GSTM1 present/GSTP1 Val � vs. GSTM1 present/GSTP1 IIe/IIe: P = 0.041; all one risk genotypes vs. GSTM1 present/GSTP1 IIe/IIe: P = 0.018) was source of heterogeneity. Sensitivity analysis was estimated by applying two methods in this meta-analysis. First, results did not change when removing a single study each time to appraise the robustness in the current meta-analysis. However, when we restrained only high-quality studies, HWE, matching, and genotyping examination performed blindly or with quality control, significantly increased BC risk was found in the overall analysis for the combined effects of GSTM1, GSTT1, and GSTP1 IIe105Val polymorphisms (GSTM1 null/GSTT1 null/GSTP1 null vs.  Tables 7-9, respectively; no significant association was observed for the combined of GSTT1 and GSTP1 IIe105Val polymorphisms, GSTT1,GSTP1, and the combined effects of GSTM1 and GSTP1 IIe105Val polymorphisms, and, as shown in Tables 6, 10, 11 and 12, respectively.

Credibility of the current and previous meta-analyses
Statistically significant associations were considered as "positive results" when they met the following criteria [44]: (1) P value < 0.05 was observed in at least one of the two genetic model (individual GSTM1 and GSTT1 polymorphisms with BC risk (there was no need to meet this condition between GSTM1 and GSTT1 polymorphisms and BC risk because they only used null vs. present); (2) FPRP < 0.2; (3) statistical power > 0.8; and (4) I 2 < 50%. Associations were considered to be "less-credible positive results" if they did not meet the above criteria. Tables 13 and 14 list the statistically significant association, I 2 value, statistical power and FPRP value for the current and previous meta-analyses, respectively. We identified "less-credible positive results" for the current and previous meta-analyses on the basis of above criteria.

Discussion
A meta-analysis involving 101 publications was done to evaluate the relationship between individual and combined effects of GSTM1, GSTT1, and GSTP1 polymorphisms on BC risk. We also used FPRP test and Venice criteria to re-analyze the previously published systematic meta-analyses. As far as we know, this is the first meta-analysis to investigate whether there was an increased BC risk for the combined effects of GSTM1 present/null, GSTT1 present/ null, and GSTP1 IIe105Val polymorphisms.  Among these genes, both GSTM1 and GSTT1 genes show deletion polymorphisms (null genotype), which cause the absence of expression and enzyme activity loss. GSTP1 IIe105Val polymorphism also decreases enzymatic activity. Given the involvement of GSTs in deactivating and detoxifying carcinogens, deletions in GSTM1 and GSTT1 and IIe105Val polymorphism in GSTP1 resulting in no enzyme activity may compromise an individual's ability to deactivate carcinogens, thus increasing risk of cancer. The exact mechanism involved are still mysterious, these combined effects might be due to the involvement of GSTM1, GSTT1, and GSTP1 in metabolism. Moreover, each gene expresses an increased risk genotype (GSTM1 null, GSTT1 null and GSTP1 Val/Val), which may be involved in breast cancer susceptibility when more than one are expressed in each individual. Overall, statistically significant increased BC risk was found in any individual and combined effects of the GSTM1, GSTT1 and GSTP1 polymorphisms. In addition, significant association was also observed in some subgroups for these genes on BC risk. However, when we restrained only high-quality studies, HWE, matching, and genotyping examination performed blindly or with quality control, significantly increased BC risk was found in the overall analysis for GSTM1 null genotype, all populations, Caucasians, and postmenopausal women for the combined effects of GSTM1 and GSTT1 polymorphisms, and overall analysis for the combined effects of GSTM1, GSTT1, and GSTP1 IIe105Val polymorphisms. This was an attempt to avoid random errors and confounding bias that sometimes distorted the results of molecular epidemiological studies [45][46][47]. Furthermore, the current meta-analysis were analyzed by applying several subgroups and different genetic models at the expense of multiple comparisons, under these circumstances, the pooled P-value must be adjusted [48]. With regard to the Venice criteria, statistical power and I 2 were important indicator by Ioannidis et al. [49]. Hence, we used FPRP test and Venice criteria to assess positive results. Finally, we identified "less-credible positive results" for the current and previous meta-analyses when we evaluated the credibility of significant associations in the current and previous meta-analyses. Heterogeneity was also observed in the current meta-analysis. The results of meta-regression analysis suggested that source of controls, type of controls and quality score of articles were source of heterogeneity between the combined effects of GSTM1 and GSTT1 polymorphisms and BC risk. For the combined effects of GSTM1 and GSTP1 IIe105Val polymorphisms, matching was source of heterogeneity in this meta-analysis. Therefore, we should perform subgroup analyses to reduce heterogeneity, because HB, patients, low quality studies and no-matching studies were important confounding bias. In addition, random error and bias were common in the studies with small sample  sizes, and the results were unreliable, especially in molecular epidemiological studies [48]. Furthermore, small sample studies were easier to accept if there was a positive report as they tend to yield false-positive results because they may be not rigorous and are often of low-quality. S8-S19 Figs indicates that the asymmetry of the funnel plot was caused by a study with lowquality small samples. A total of fourteen previous meta-analyses [19][20][21][22][23][24][25][26][27][28][29][30][31][32] between 2004 and 2016 have been published to analyze the individual GSTM1 present/null, GSTT1 present/null, and/or GSTP1 IIe105Val polymorphisms on breast cancer (BC) risk. Table 14 lists the statistically significant  [21] indicated that the GSTM1 and GSTT1 null genotypes were associated with an increased BC risk in Asians and the GSTP1 IIe105Val polymorphism was associated with an increased BC risk in Caucasians. The examination of 27 studies of GSTM1, 23 studies of GSTT1, and 20 studies of GSTP1 by Tang et al. [22] indicated that the GSTM1 and GSTP1 polymorphisms were associated with an increased BC risk in Asian population, especially in East Asian, while the GSTT1 polymorphism may be not associated with BC risk. Xiao et al. [23] conducted an association of 13 studies involving 3,387 cases and 5,085 controls in Chinese population, and suggested that the  [32] suggested that the GSTM1 and GSTT1 null genotypes were associated with an increased BC risk in postmenopausal and all women, respectively. However, quality assessment of the eligible studies was not assessed in 12 previous meta-analyses [19-21, 23, 25-32], source of heterogeneity was not explored in 13 previous meta-analyses [19][20][21][22][23][24][25][26][27][28][29][30][31][32] on the basis of meta-regression     analysis, the false-positive report probabilities of statistically significant association and statistical power was not evaluated in all previous meta-analyses [19][20][21][22][23][24][25][26][27][28][29][30][31][32], and I 2 value was not showed in 11 previous meta-analyses [19,20,23,[25][26][27][28][29][30][31][32]. Therefore, results of their meta-analyses may be not credible. This meta-analysis has several advantages over previous meta-analyses [19][20][21][22][23][24][25][26][27][28][29][30][31][32]. First, the sample size was much larger, with 88 studies involving 28,676 BC cases and 32,539 controls assessed for the GSTM1 null genotype, 67 studies involving 23,092 BC cases and 26,381 controls for the GSTT1 null genotype, and 56 studies involving 25,331 BC cases and 27,424 controls in all populations. Second, this is the first meta-analysis to investigate the combined effects of these genes in overall population. Third, we evaluated quality assessment of the eligible studies. Forth, we used meta-regression analysis method to explore the source of heterogeneity. Fifth, we collected more detailed data. Sixth, an important sensitivity analysis was conducted on studies that were high-quality, matching, HWE, and or in which genotyping was performed blindly or with quality control. Seventh, we applied FPRP and Venice criteria to investigate the significant association with BC risk. The current meta-analysis also has several limitations. First, only published articles were included in the current meta-analysis, therefore, publication bias may be exist as shown in S8-S19 Figs. Positive results are known to be published more readily than negative ones. If negative results were included, an underestimation of the GSTM1 null effect may be observed. Second, we did not consider whether the genotype distribution in the controls was in HWE for GSTM1 and GSTT1 polymorphism because we cannot calculate the HWE on the both genes. Third, no data were extracted on other risk factors, such as hormonal readiness, obesity, smoking, and so on.

Conclusions
In summary, this meta-analysis indicates that individual and combined effects of GSTM1, GSTT1 and GSTP1 polymorphisms may be not associated with increased BC risk.   Table. Genotype frequencies of the GSTM1, GSTT1, and GSTP1 IIe105Val polymorphisms between postmenopausal and premenopausal breast cancer and control groups. (PDF) S7 Table. Genotype frequencies of the GSTM1, GSTT1, and GSTP1 IIe105Val polymorphisms between and breast cancer and control groups by smoking status. (PDF) S8 Table. Genotype frequencies of the combined effects of GSTM1 present/null and GSTT1 present/null between breast cancer and control groups. (PDF) S9 Table. Genotype frequencies of the combined effects of GSTM1 and GSTT1 between postmenopausal and premenopausal breast cancer and control groups. (PDF) S10 Table. Genotype frequencies of the combined effects of GSTM1 present/null and GSTP1 IIe105Val between breast cancer and control groups. (PDF) S11 Table. Genotype frequencies of the combined effects of GSTT1 present/null and GSTP1 IIe105Val between breast cancer and control groups. (PDF) S12 Table. Genotype frequencies of the combined effects of GSTM1 present/null, GSTT1 present/null, and GSTP1 IIe105Val between breast cancer and control groups. (PDF)