The Associations between Two Vital GSTs Genetic Polymorphisms and Lung Cancer Risk in the Chinese Population: Evidence from 71 Studies

Background The genetic polymorphisms of glutathione S-transferase (GSTs) have been suspected to be related to the development of lung cancer while the current results are conflicting, especially in the Chinese population. Methods Data on genetic polymorphisms of glutathione S-transferase Mu 1 (GSTM1) from 68 studies, glutathione S-transferase theta 1 (GSTT1) from 17 studies and GSTM1-GSTT1 from 8 studies in the Chinese population were reanalyzed on their association with lung cancer risk. Odds ratios (OR) were pooled using forest plots. 9 subgroups were all or partly performed in the subgroup analyses. The Galbraith plot was used to identify the heterogeneous records. Potential publication biases were detected by Begg's and Egger's tests. Results 71 eligible studies were identified after screening of 1608 articles. The increased association between two vital GSTs genetic polymorphisms and lung cancer risk was detected by random-effects model based on a comparable heterogeneity. Subgroup analysis showed a significant relationship between squamous carcinoma (SC), adenocarcinoma (AC) or small cell lung carcinoma (SCLC) and GSTM1 null genotype, as well as SC or AC and GSTT1 null genotype. Additionally, smokers with GSTM1 null genotype had a higher lung cancer risk than non-smokers. Our cumulative meta-analysis demonstrated a stable and reliable result of the relationship between GSTM1 null genotype and lung cancer risk. After the possible heterogeneous articles were omitted, the adjusted risk of GSTs and lung cancer susceptibility increased (fixed-effects model: ORGSTM1 = 1.23, 95% CI: 1.19 to 1.27, P<0.001; ORGSTT1 = 1.18, 95% CI: 1.10 to 1.26, P<0.001; ORGSTM1-GSTT1 = 1.33, 95% CI: 1.10 to 1.61, P = 0.004). Conclusions An increased risk of lung cancer with GSTM1 and GSTT1 null genotype, especially with dual null genotype, was found in the Chinese population. In addition, special histopathological classification of lung cancers and a wide range of gene-environment and gene-gene interaction analysis should be taken into consideration in future studies.


Introduction
Lung cancer is the most common malignancy in the world and the leading cancer in males, accounting for 17% of the total new cancer cases and 23% of the total cancer deaths [1][2][3]. The burden of lung cancer mortality in females in developing countries is up to 11% of the total female cancer deaths [2]. In the United States, there were 226,160 newly diagnosed cases and 160,340 deaths due to lung cancer in 2012 [4]. In China, although females have a lower prevalence of smoking, there is still higher lung cancer rates (21.3 cases per 100,000 females) than those in European countries [5], due to indoor air pollution, cooking fumes, occupational and environmental pollutions. Besides, due to the incurable nature and less than a five-year survival rate (only 16%), lung cancer has attracted a huge attention across the whole world [6]. Pathologic diagnosis " : ALL means that all lung cancer cases were confirmed by pathologic diagnosis; PARTIAL means that partial cases were confirmed by pathologic diagnosis; NA means that relative data were not available in original studies.  Lung cancer can be divided into several types by pathological classification, such as squamous cell carcinoma (SC), adenocarcinoma (AC) and large or small cell carcinoma. It is also classified as small cell lung carcinoma (SCLC) and non-small cell lung carcinoma (NSCLC), which accounts for about 85% of all lung cancer [7]. Given the possible relapses in the local respiratory system and the metastasis in other systems after the classical treatments of radical surgery, immunotherapy has provided an innovative method for lung cancer treatment in the past 30 years to enhance the clinical outcome, alleviate the disease burden, prevent recurrences and attenuate toxicity [8][9][10][11][12][13][14].
Tobacco smoking has clearly been demonstrated to be a strong exogenous factor for lung cancer risk [15][16][17]. Polycyclic aromatic hydro-carbons (PAHs) and the tobacco-specific nitrosamine 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) are considered to be the major carcinogens, which can interact with DNA and cause the formation of DNA adducts [17]. In the meantime, free radicals from tobacco smoking can induce oxidative damage to lung tissues, and also damage DNA, which provides another clue to lung cancer development [18][19][20][21]. In this process, DNA was damaged by superoxide anions (O 2 2 ) and hydroxyl radicals (OH 2 ) and was repaired by antioxidant enzymes. This balance can be broken by both environmental and genetic factors. Available molecular epidemiology studies have shown that genetic polymorphisms play a major role in the progress of carcinoma [22,23]. Among these studies, genetic variants of carcinogenmetabolizing enzymes have received much attention, especially glutathione S-transferase (GST) genes and cytochrome P450 genes. The cytochrome P450 (CYP450) family, as the first-pass metabolism enzymes, plays an important role in many physiological and biochemical reactions in the human body, and participates in the metabolic process of endogenous and exogenous substrates (biosynthesis and degradation) [24]. Toxic materials like benzo[a]-pyrene and other PAHs could be metabolized to oxygenated intermediates and then degraded sequentially to lower toxic or non-toxic substances by the second-pass metabolic enzymes such as the glutathione S-transferases (GSTs) family [25,26]. Therefore, the polymorphisms of both gene families might affect the metabolism of tobacco toxicants in lung and finally influence the advancement of cancer.
To identify the association of two vital GST genetic polymorphisms (GSTM1 and GSTT1) with lung cancer risk, an updated systematic meta-analysis was performed in this study by selecting all eligible studies in the Chinese population.

Inclusion and exclusion criteria
Inclusion criteria: (1) individuals or samples in all eligible studies were examined and diagnosed by polymerase chain reaction (PCR), pathologic diagnosis or other methods to get a full picture of GST genetic polymorphisms and lung cancer types; (2) Chinese living in China; (3) articles providing raw data including odds ratio (OR) with 95% confidence interval (CI) and respective variance, or the relevant information could be calculated.
Exclusion criteria: (1) Chinese out of China; (2) raw data not available; (3) when there were multiple publications by the same researchers, only the latest or the largest population study was adopted; (4) meeting abstract, case reports, editorials, newsletter and review articles were excluded.

Data extraction and synthesis
To decide inclusively or exclusively, articles were identified by three independent work groups (group 1-Kui Liu and Lu Zhang; group 2-Xia Yue and Xialu Lin; group 3-Jian Chen and Guixiu Jin) using a standardized data extraction form designed by ourselves. Discrepancies among three groups were further discussed by all parties. If consenses was still not reached, another group (group 4-Huiqin Wang and Qi Zhou) would make the final decision. Firstly, the titles and abstracts of all studied articles were screened to determine their relevance. If the titles and abstracts were ambiguous, full articles would be investigated. In order to make full use of the available data, it was counted as two separated studies if two different control groups were employed in the same article, such as two different controls versus the same control. If there were more than one region to be investigated in one article, information for each region was also counted as a separated study. Information collected from each eligible study included: first author, year of publication, region, study time, pathologic diagnosis, source of control, characteristics of cases and controls, genotype frequency of null GSTM1, null GSTT1, and null of both  [47]. HWE test was usually assessed in the control group [48]. Furthermore, details of eligible studies used for detecting GSTs genotype, combined evaluation of other genes, HWE test results of CYP1A1 polymorphisms, the percent of null GSTs genotype in the control groups, smoking status, study type and quality score were also elicited ( Table 2). Study types also consisted of epidemiological design and non-epidemiological design. Epidemiological designs were comprised of case-control, cohort and nested case-control studies, all of which must satisfy three conditions for both cases and controls: explicit diagnosis of status (histology or cytology), clear description of the age period, and the same source population [49]. Those not meeting the conditions were considered non-epidemiological designs. The quality score of epidemiological studies was evaluated by Newcastle-Ottawa Scale (NOS).

Statistical analysis
(1) The pooled ORs and 95% CIs were determined by the Z test, P#0.05 was considered statistically significant. (2) Statistical heterogeneity among studies was assessed by Q and I 2 statistics [50]. In heterogeneity tests, when P#0.1, a random-effects model was used; when P.0.1, a fixed-effects model was performed [51]. Meanwhile, if I 2 $50%, 50%.I 2 $25% or I 2 ,25%, we identified the studies as high, middle or low heterogeneity, respectively. (3) Sensitivity analysis was performed by removing one study at a time to calculate the overall homogeneity and effect size; the Galbraith plot was also performed to examine the possible distinct articles. (4) The possible reasons for heterogeneity between studies were investigated by subgroup analyses. Nine subgroups were analyzed as follows: histopathological classification (SC, AC or SCLC), geographical location (North, Northeast, Northwest, East, Central, South, or Southwest of China) (See Figure S1), smoking status (smoker vs. non-smoker), CYP1A1(Msp1) polymorphisms, case number (,100 vs. $100), source of controls (population-based vs. hospital-based), research design (epidemiological design vs. nonepidemiological design), test material (white blood cells, involved tissues or other cells, or not available) and quality score (4-5, 6, 7-8). The last five items listed above were used to assess the study quality. (5) Cumulative meta-analysis was used to explore any significant changes in the variation of sample size or publication year. (6) Publication bias was investigated by the Begg's test [52], Egger's linear regression test and Trim and Fill test [53]. (7) All analyses were performed with the software Stata version 12.0 (StataCorp LP, College Station, Texas, USA), and all P values were two sided.

Synthesis results of all studies
The results showed a significant association between the GSTM1 null genotype and lung cancer risk in the Chinese population under the random-effects model (OR = 1.20, 95% CI: 1.16 to 1.25, I 2 = 45.1%, P,0.001) ( Table 3). The random-effects model showed that the GSTT1 null genotype was significantly correlated with lung cancer risk in the Chinese population (OR = 1.17, 95% CI: 1.07 to 1.28, I 2 = 55.9%, P,0.001) ( Table 4). Further analyses showed that dual-null genotype of GSTM1-GSTT1 had a significant higher association with lung cancer risk (OR = 1.29, 95% CI: 1.03 to 1.63, I 2 = 61.7%, P = 0.011) ( Table 5). Risk estimation for each study is shown in the Forest plots in Figure 3, Figure 4a and Figure 4b.

Cumulative meta-analysis
The cumulative meta-analysis was used to examine the fluctuation of the eligible studies with changes in the publication year or sample size. With the publication year development and sample size increase, the cumulative meta-analysis of GSTM1 tended to be stable. However, no significant difference in the trend was found in the GSTT1 and GSTM1-GSTT1 cumulative metaanalysis. The results for cumulative meta-analysis are shown in Figure 5 and Figure 6.

Subgroup analysis
Due to the fact that all studies were middle to high heterogeneities, analyses on nine subgroups as mentioned above were performed accordingly. No significant increase in the risk of lung cancer was detected in either null genotype of GSTM1 in the northwest, or null genotype of GSTT1 in the north, southwest or northwest of China (Table 3, Table 4). The excess lung cancer risk was found associated with null GSTM1 genotype, but not with null GSTT1 genotype, in both smokers and nonsmokers. Besides, smokers had a higher risk than non-smokers in the association between GSTM1 null genotype and lung cancer risk. The interaction of CYP1A1 (Msp1) with mt/mt genotype and GSTM1 null genotype could enhance the risk of lung cancer, and the OR of which were a little higher than the other two CYP1A1 (Msp1) genotypes with GSTM1 null.
However, high heterogeneities in the analysis of the association between GSTM1 variants and lung cancer were found in the studies from northeast and southwest China. The subgroups of AC and smokers also showed greater heterogeneities (I 2 :53.8% and 50.3%, respectively). Meanwhile, the subgroup analyses of GSTT1 genetic polymorphisms and lung cancer susceptibility demonstrat- Table 5. Subgroup analysis of the association between GSTM1-GSTT1 null genotype and lung cancer risk. When analyzing the five subgroups of case numbers $100, population-based controls, epidemiological studies, test material from white blood cells, and quality score (7)(8), all pooled results showed significant association between GSTT1 genetic polymorphisms and lung cancer risk, but high heterogeneities also appeared. However, subgroups of case numbers ,100, hospitalbased controls, non-epidemiological studies, test materials from involved tissue or cells or not available, and quality score (4)(5), all pooled results showed no significant association between GSTT1 genetic polymorphisms and lung cancer risk (Table 4).
In the analysis of the relationship of GSTM1-GSTT1 genetic polymorphisms with lung cancer risk, no significant association was found in the subgroup of case numbers ($100). Along with significant increase risks in the subgroup of population-based controls and epidemiological studies, high heterogeneity was also found ( Table 5).

Galbraith plot and sensitivity analysis
In Figure 7a, 7 articles were identified in the Galbraith plot as the outliers [60,68,86,89,93,115,122]. After omitting these records, the adjusted association of GSTM1 null genetype and lung cancer risk showed a lower heterogeneity and an increased susceptibility (fixed-effects model: OR = 1.23, 95% CI: 1.19 to 1.27, P,0.001). Besides, according to the Galbraith plot of the association of GSTT1 or GSTM1-GSTT1 interaction polymorphisms with lung cancer risk, 2 articles [98,115] were obviously spotted as the outliers, which were the possible sources for the heterogeneities. After adjustment, the association of both groups were all increased (fixed-effects model: OR GSTT1 = 1.18, 95% CI: 1.10 to 1.26, P,0.001; OR GSTM1-GSTT1 = 1.33, 95% CI: 1.10 to 1.61, P = 0.004) and the I 2 indexes were decreased to 29.5% for GSTT1 and 2.1% for GSTM1-GSTT1, respectively ( Figure 7, Table 6). Then, the sensitivity analysis was carried out in each group (data not shown).

Potential publication bias
Begg's funnel plots and Egger's linear regression test were used to evaluate the potential publication bias (Figure 8a Figure 8e and Figure 8f for GSTM1-GSTT1). No publication bias was detected by Egger's test (P E = 0.245 for GSTM1, P E = 0.510 for GSTT1 and P E = 0.320 for dual-null genotype of GSTM1-GSTT1). The Trim and Fill test further confirmed the results (data not shown).

Discussion
To our knowledge, this is the first large-scale systematic metaanalysis on the correlation of two vital GSTs genetic polymorphisms with lung cancer risk in the Chinese population over the past decade. Our pooled analysis on the original studies in the Chinese population provided efficient and effective evidences of an increased association between null GSTM1, null GSTT1 or dual null GSTM1-GSTT1 genotypes and lung cancer risk when omitting some possible heterogeneous records. This large-scale systematic review on sufficient studies helps to reduce random error and increase the statistical power. Simultaneously, by using the same inclusive criteria, it can also ensure the pooled results more precise and exact. It is well known that different populations have different genetic variations and environmental exposure factors. Previous studies paid more attention to the Asian or special environmental population [35,46]. We only focused on the Chinese ethnicity.
In subgroup analysis of GSTM1 genetic variants, the northeast and southwest of China were found to be a source of difference, and in subgroup analysis of GSTT1 genetic variants, the southwest regions of China was also suggested as the major heterogeneous source. Furthermore, no association between GSTs and lung cancer susceptibility was evident in the Chinese population living in the above regions. To our knowledge, the greatest population in the southwest and northwest areas of China is the Chinese ethnic minorities. The complex genetic backgrounds of various ethnic minorities might have an influence on lung cancer susceptibility. In the subgroup of histopathological classification, increased association between the genetic polymorphisms and SC (OR and 95% CI:1.20 [1.12,1.27]) and SCLC (OR and 95% CI:1.29[1.13,1.47]) risk were found with a low heterogeneity. These results for the first time imply a clue that SCLC could have a stronger association with GSTM1 deficiency than the other two types while no statistic difference was found among 3 pathological types from available data. Due to the limited number of studies and comparatively diversity among various studies, more well designed epidemiological studies should be performed for various pathological types of lung cancers (especially for pulmonary AC). Additionally, we found that there was increased susceptibility between GSTM1 null genotype and lung carcinoma risk in different phase I isoenzymes of CYP1A1. These results not only further confirm our conclusion, but also imply some enlightenments. For instance, under a higher OR with no heterogeneity, people with CYP1A1 (mt/mt) and GSTM1 null genotype should pay more attention to avoiding exposure to harmful environmental factors associated with lung cancer. Naturally, more studies including a genome-wide association study (GWAS) are necessary to prove this hypothesis. Due to the limited number of studies, the same analysis for the GSTT1 null genotype was not performed.
The subgroup analyses of the smoking status for GSTM1 studies further suggested that the possible risk factor of GSTM1 null genotype is different. However, eligible studies for GSTT1 failed to reach a significant association, which might be caused by a limited number of studies with high heterogeneities. Unclear smoking definition and inconsistent classification of the amount of tobacco consumed among different studies might all have an influence on the stability, reliability, as well as further in-depth analyses of the results. Therefore, clear smoking definition and consistent classification for the smoking status are necessary in any future research.
In the sensitivity analyses and Galbraith plot, 7 heterogeneous articles for GSTM1 were detected by the Galbraith plot. The potential bias of these articles might be the result of small sample size, complex population composition, distinction of testing materials [86], and/or unknown reasons [115]. After omitting these articles, no heterogeneity was detected. Additionally, the Galbraith plot for the GSTT1 and GSTM1-GSTT1 groups spotted two of the same articles [98,115] as the major source of between-heterogeneity. After removing these two articles, heterogeneity decreased substantially. Compared to the raw OR and 95% CI, the adjusted OR and 95% CI of GSTT1 and GSTM1-GSTT1 were both increased.
Cumulative meta-analysis showed a comparable change in the trend in the accumulated OR and 95% CI for GSTT1 or GSTM1-GSTT1 with the publication time development and sample size increase. Thus, to identify the real association between the GSTT1 null type, GSTM1-GSTT1 dual null type and lung cancer susceptibility, more large-scale case-control and cohort studies from multi-centers should be performed. At last, no publication biases were detected in our meta-analysis.
It's worth mentioning that Hardy-Weinberg equilibrium has been widely recommended in testing studies of genetic polymorphisms and diseases, the violations of which may have potential impacts on the results [125]. In this paper, no individual studies made any distinction between heterozygotes or homozygotes and GSTM1 and GSTT1 in the present genotype, so Hardy-Weinberg equilibrium tests could not be performed. Therefore, the Hardy-Weinberg equilibrium test results reported in some of the 71 articles might not be reliable.
It is worthy to note that several other limitations might be included in this study: (1) as common observational studies, casecontrol studies were susceptible to various biases (including recall Figure 7. Galbraith plot of association between GSTs polymorphisms and lung cancer risk. Each figure represents a unique article in this meta-analysis. The figures outside the three lines were spotted as the outliers and the possible sources of heterogeneity in the analysis pooled from the total available number. (a) Galbraith plot result of GSTM1 polymorphisms and lung cancer risk; (b) Galbraith plot result of GSTT1 polymorphisms and lung cancer risk; (c) Galbraith plot result of GSTM1-GSTT1 dual null genotype and lung cancer risk. doi:10.1371/journal.pone.0102372.g007 Table 6. Subgroup analysis of $ the adjusted association between GSTM1 null genotype, GSTT1 null genotype and GSTM1-GSTT1 dual null genotype and lung cancer risk. bias of smoking status, different diagnostic criteria and the investigation bias of NOS score). These biases could influence the final findings of this study; (2) conclusions of this study were partly based on literatures obtained from the hospital-based population, which might not represent the whole population; (3) eligible studies for this study covered nearly all regions in China, but the article number was still insufficient in some less developed or relatively sparsely regions; (4) the interaction of genes with environmental factors, especially with special external occupational exposure and environmental pollution, might all contribute to the development of lung cancer. Factors above might also contribute to a possible source of heterogeneity of our results.
Owning to the limitation of the data, this paper did not analyze the interaction effects of these factors; (5) absence of HWE test in the control group, some unbalance controls could lead to some bias in the final results. Taken together, after a decade of extensive studying on this topic, our findings suggest that GSTM1 and GSTT1 genetic polymorphisms are associated with increased lung cancer risk in the Chinese population. Because of multifactor etiology of the interaction of gene-gene and gene-environment in the development of lung cancer, large-scale and methodologically sound studies with different environmental background and other genetic polymorphisms should be carried out to explore the real association between GSTs variants and various pathological types of lung cancer. Begg's funnel plot is used to detect potential publication bias in which a symmetric funnel shape means no publication bias. Egger's linear regression test is used to quantify the potential presence of publication bias; (a) (b) GSTM1: No publication bias has been found from 68 inclusive studies about the association between GSTM1 polymorphisms and lung cancer risk by Begg's??? test and Egger's test, respectively; (c)(d) GSTT1: No publication bias has been found from 17 inclusive studies about the association between GSTT1 polymorphisms and lung cancer risk by Begg's test and Egger's test, respectively; (e)(f) GSTM1-GSTT1 dual-null genotype: No publication bias has been found from 8 inclusive studies about the association between GSTM1-GSTT1 dual-null genotype and lung cancer risk by Begg's test and Egger's test, respectively. doi:10.1371/journal.pone.0102372.g008