Glutathione S-Transferase M1 (GSTM1) Polymorphisms and Lung Cancer: A Literature-based Systematic HuGE Review and Meta-Analysis

¹ Department of Medicine, Faculty of Medicine, University of British Columbia, Vancouver, British Columbia, Canada
² School of Environmental Health, University of British Columbia, Vancouver, British Columbia, Canada
³ United Kingdom HuGENet Coordinating Centre, Strangeways Research Laboratory, University of Cambridge, Cambridge, United Kingdom
⁴ MRC Biostatistics Unit, Institute of Public Health, University of Cambridge, Cambridge, United Kingdom
⁵ Department of Medicine, School of Medicine, University of Washington, Seattle, WA
⁶ Center for Ecogenetics and Environmental Health, University of Washington, Seattle, WA

Correspondence to Dr. Chris Carlsten, Vancouver General Hospital, 2775 Laurel Street, 7th Floor (The Lung Center), Vancouver, British Columbia V5Z 1M9, Canada (e-mail: chris.carlsten@vch.ca ).

Received for publication July 5, 2007. Accepted for publication December 7, 2007.

Multiple genes have been studied for potential associations with lung cancer. The gene most frequently associated with increased risk has been glutathione S-transferase M1 (GSTM1). The glutathione S-transferase enzyme family is known to catalyze detoxification of electrophilic compounds, including carcinogens, therapeutic drugs, environmental toxins, and products of oxidative stress. In this review, the authors summarize the available evidence associating lung cancer with the GSTM1 gene. They describe results from an updated meta-analysis of 98 published genetic association studies investigating the relation between the GSTM1 null variant and lung cancer risk including 19,638 lung cancer cases and 25,266 controls (counting cases and controls in each study only once). All studies considered, the GSTM1 null variant was associated with an increased risk of lung cancer (odds ratio (OR) = 1.22, 95% confidence interval (CI): 1.14, 1.30), but no increase in risk was seen (OR = 1.01, 95% CI: 0.91, 1.12) when only the five largest studies (>500 cases each) were considered. Furthermore, while GSTM1 null status conferred a significantly increased risk of lung cancer to East Asians (OR = 1.38, 95% CI: 1.24, 1.55), such a genotype did not confer increased risk to Caucasians. More data regarding the predictive value of GSTM1 genetic testing are needed before population-based testing may be reasonably considered.

Keywords: epidemiology; genetics; genome; human; glutathione S-transferase M1; glutathione transferase; GSTM1; lung neoplasms; meta-analysis
Abbreviations: CI, confidence interval; CYP, cytochrome P-450; CYP1A1, cytochrome P-450 1A1; GST, glutathione S-transferase; GSTM1, glutathione S-transferase M1; GSTT1, glutathione S-transferase T1; HuGE, Human Genome Epidemiology; HuGENet, Human Genome Epidemiology Network; OR, odds ratio

The GSTs [EC 2.5.1.18 (4)] are a family of cytosolic enzymes known to catalyze the detoxification of electrophilic compounds, including carcinogens, therapeutic drugs, environmental toxins, and products of oxidative stress, by conjugation with glutathione (5). This conjugation reaction also facilitates excretion and thus constitutes a detoxification step. In addition to this role in phase II detoxification, GSTs are able to modulate the induction of other enzymes and proteins important in cellular functions, such as DNA repair, and are therefore important in maintaining genomic integrity (5). The GST enzymes could potentially play an important role in susceptibility to cancer. Five distinct loci (alpha, mu (M), theta, pi, and gamma) are known to encode the GST enzymes. Here we consider the relation between the GSTM1 gene and lung cancer.

GSTM1 (OMIM number 138350 (6)) has been mapped to the GST mu gene cluster on chromosome 1p13.3. Two variants in GSTM1 have been identified: a deletion and a substitution. The alleles of the substitution variant differ by a C-to-G transition at base position 534, resulting in a lysine-to-asparagine substitution at amino acid 172 (7, 8). The deletion (GSTM1 null variant) has been examined extensively in epidemiologic studies. Persons with a homozygous deletion of the GSTM1 locus have no enzymatic functional activity. Phenotype assays have confirmed this lack of function by demonstrating a strong concordance (≥94 percent) between phenotype and genotype (9, 10). The GSTM1 gene and the null variant have been the focus of previous HuGE reviews of colorectal cancer (9) and squamous-cell carcinoma of the head and neck (11) and previous pooled and meta-analyses (table 1).

click to view

Table 1  Characteristics and findings of previously conducted meta- and pooled analyses of glutathione S-transferase M1 (GSTM1) polymorphisms and lung cancer

Lung cancer has been the most common cancer in the world since 1985 (12). In 2002, 1.35 million new cases of lung cancer were diagnosed, representing more than 12 percent of all new cancer cases. Lung cancer is also the most common cause of death from cancer, with 1.18 million deaths, accounting for 17.6 percent of the world total (12). Cancer rates have peaked among men in many parts of the world, but rates are continuing to rise among women, with almost half of all cases occurring in the developing world (12, 13).

Lung cancer is generally divided into two types, small-cell and non-small-cell, although there are other, rarer types, such as carcinoid tumors. Small-cell lung cancer accounts for approximately 20 percent of all lung cancer cases and is almost exclusively caused by smoking. Non-small-cell lung cancer accounts for approximately 80 percent of all lung cancers and includes three subtypes: squamous-cell carcinoma (almost always caused by smoking), adenocarcinoma, and large-cell undifferentiated carcinoma. Recent decades have seen an increase in the frequency of adenocarcinoma and a decline in squamous-cell carcinoma in developed countries. This could be partly explained by an increase in the use of filtered cigarettes (14).

Lung cancer is frequently diagnosed at an incurable stage. Treatment for non-small-cell lung cancer (stages I, II, and occasionally IIIa) is based on surgery with adjuvant irradiation and/or chemotherapy. Patients with advanced non-small-cell lung cancer usually receive only chemotherapy. Surgery plays only a limited role in the management of small-cell lung cancer. Depending on the stage of disease and its complications, patients typically receive some combination of radiation and chemotherapy. Prognosis is poor in general, but it is considerably better in cases of non-small-cell lung cancer than in cases of small-cell lung cancer. For the approximately 70 percent of non-small-cell lung cancers that are unresectable, survival time varies greatly, from a few weeks to a few years, depending on the functional status of the patient at the time of diagnosis. In contrast, given its very aggressive nature, the median survival of patients with small-cell lung cancer is approximately 1 year (15).

Tobacco smoking is clearly the strongest risk factor for lung cancer, and despite its original description by Rottman (16) in 1898, smoking-induced lung cancer continues to be a major public health problem. In 2001, 856,000 of the annual trachea, bronchus, and lung cancer deaths (70 percent of the total number of such deaths) were attributable to smoking (17). The risk among smokers as compared with never smokers was increased 8–15 times in men and 2–10 times in women (18). Cessation of smoking is known to significantly reduce lung cancer risk, with the most marked effect being observed in heavy smokers, particularly among women (19). However, many persons who smoke continue to do so. Other risk factors for lung cancer include environmental tobacco smoke exposure, diet, and occupational exposures such as soot and asbestos (20).

In the last two decades, and especially in recent years, a large body of medical and epidemiologic literature has described genetic variants that appear to affect susceptibility to lung cancer. Multiple genes—including several in the GST group, cytochrome P-450 1A1 (CYP1A1) and several other genes in the cytochrome P-450 (CYP) group, microsomal epoxide hydrolase (mEH), aryl hydrocarbon receptor (AhR), NAD(P)H quinone oxidoreductase 1 (NQO1), myeloperoxidase (MPO), and N-acetyltransferase (NAT)—have been variably associated with the disease (14, 21–25). These effects are at least partly independent of the effects of tobacco; an excess risk of lung cancer has been observed in relatives of lung cancer patients regardless of smoking status (26, 27). While some of the familial risk could be due to environmental tobacco smoke exposure, a shared genetic risk is strongly suggested. Regardless, the independent effect on lung cancer risk is strongly amplified by cigarette smoking (17, 23, 28). Variants in several genes have now been shown to be associated with increased lung cancer risk specifically in smokers; smokers with the "at risk" genotype are at a significantly higher risk of lung cancer than smokers without the "at risk" genotype. Several general reviews of the topic are available (25, 29–33).

Seven meta-analyses and pooled analyses published to date have been consistent in finding a modest but statistically significant increase in risk for persons carrying the null variant (table 1); summary odds ratios from these meta-analyses range from 1.17 to 1.54. For the present review, we sought all population-based cohort, case-control, or cross-sectional studies reporting associations between the GSTM1 null variant and lung cancer. Cases had to be diagnosed with lung cancer, and controls had to be healthy or hospital-based controls without cancer. Full details on the methods used for collating and synthesizing data from these association studies are provided in the Appendix. Our literature search retrieved 2,597 papers published up to March 2006. We identified 98 studies for inclusion in the meta-analysis, and these are individually characterized in Web table 1.

The 98 studies were undertaken in a wide range of ethnogeographic settings (Web table 1), with 46 percent (9,071 of 19,638) of cases being reported as Caucasian (data from 36 studies), 31 percent (6,088 of 19,638) of cases being reported as East Asian (data from 42 studies), and 23 percent (4,479 of 19,638) of cases being reported as of nonspecific ethnicity (included African-American, mixed ethnicity, and ethnicity not stated; data from 20 studies). Five studies accounted for just over one quarter of all cases (26 percent; 5,112 of 19,638). Forty-four studies used general population controls, 33 used hospital-based controls, and 21 used controls from other sources (included healthy workers, friends and spouses of cases, and source not stated). In several studies, investigators also reported results broken down by lung cancer clinical subtype, such as adenocarcinoma (40 studies), squamous-cell carcinoma (37 studies), small-cell carcinoma (22 studies), or large-cell carcinoma (7 studies).

Using a random-effects meta-analysis with a dominant genetic model, the combined odds ratio for lung cancer among persons with the GSTM1 null genotype was 1.22 (95 percent confidence interval (CI): 1.14, 1.30) (see Web figure 1). The fixed-effect meta-analysis odds ratio for lung cancer was 1.16 (95 percent CI: 1.12, 1.21). There was some evidence of heterogeneity among these studies (I² = 58 percent, 95 percent CI: 46, 66; p < 0.0001) and also of funnel plot asymmetry (Begg's test, p = 0.003). Ethnicity accounted for some of this heterogeneity (21 percent of the between-study variance, p < 0.001). Subgroup analyses were also undertaken (figure 1). When studies were subgrouped by ethnicity, the odds ratio for Caucasians was 1.04 (95 percent CI: 0.97, 1.11), with I² equal to 22 percent (95 percent CI: 0, 48; p = 0.117), and the odds ratio for East Asians was 1.38 (95 percent CI: 1.24, 1.55), with I² equal to 56 percent (95 percent CI: 34, 68; p < 0.0001). The odds ratios for general population and hospital-based control groups were 1.21 (95 percent CI: 1.10, 1.33) with I² equal to 54 percent (95 percent CI: 31, 66; p < 0.0001) and 1.32 (95 percent CI: 1.14, 1.52) with I² equal to 69 percent (95 percent CI: 54, 77; p < 0.0001), respectively. When only the large (>500 cases) studies were considered, the odds ratio for persons with the GSTM1 null genotype was 1.01 (95 percent CI: 0.91, 1.12), with I² equal to 31 percent (95 percent CI: 0, 74; p = 0.216). Phenotyping rather than genotyping was conducted in five studies which, combined, gave an odds ratio of 1.63 (95 percent CI: 0.96, 2.74) with I² equal to 75 percent (95 percent CI: 8.6, 88; p = 0.0028).

click to view

Figure 1 : Results from a random-effects meta-analysis of studies of glutathione S-transferase M1 (GSTM1) polymorphisms and lung cancer, according to various characteristics. OR, odds ratio; CI, confidence interval.

The combined odds ratio for adenocarcinoma cases (n = 4,005; 40 studies) was 1.18 (95 percent CI: 1.05, 1.32), with I² equal to 48 percent (95 percent CI: 19, 63.3; p = 0.0005), for the GSTM1 null genotype. Small-cell carcinoma cases (n = 807; 22 studies) had an odds ratio of 1.35 (95 percent CI: 1.12, 1.64), with I² equal to 31 percent (95 percent CI: 0, 58; p = 0.08). The combined odds ratio for squamous-cell carcinoma cases (n = 3,700; 37 studies) with the GSTM1 null genotype was 1.24 (95 percent CI: 1.10, 1.40), with I² equal to 55 percent (95 percent CI: 30, 68; p < 0.0001). The large-cell carcinoma cases (n = 112; 7 studies) had an odds ratio of 1.06 (95 percent CI: 0.58, 1.93), with I² equal to 50 percent (95 percent CI: 0, 77; p = 0.06).

Gene-gene interactions
An association between enzymes in either the CYP or GST families and a smoking-related cancer such as lung cancer is biologically plausible. Most toxic compounds are detoxified in two phases. In phase 1, atomic oxygen is introduced in a reaction catalyzed by the CYP gene family. This generates an oxygenated intermediate, which is a substrate for phase 2, in which several families of enzymes (including GST) add moieties that detoxify the substrate (34). With cigarette smoking, benzo[a]pyrene is considered a primary toxic byproduct, and it is metabolized by CYP1A1 to benzo[a]pyrene epoxide, which is the reactive intermediate. GSTM1 then converts this intermediate to benzo[a]pyrene-S-glutathione. As a result, either high CYP1A1 activity (conferred by the MspI "m2" variant of the CYP1A1 gene) or low GSTM1 activity (conferred by the null variant of the GSTM1 gene), or particularly their combination, will increase benzo[a]pyrene levels and therefore toxicity (35). Further mechanistic support is provided by research that correlates the GSTM1 null genotype with the DNA adducts (polycyclic aromatic hydrocarbon–deoxyguanosine monophosphate) that are known markers for carcinogenesis (36).

Few studies have investigated the role of gene-gene interactions in lung cancer, mainly because of the large numbers of participants that would be required to provide adequate statistical power. Nakachi et al. (37) found that persons with the CYP1A1 MspI or Ile/Val variant and persons with the GSTM1 null variant with low levels of cigarette smoking were at high risk of lung cancer, with odds ratios of 16.0 (95 percent CI: 3.76, 68.02) and 41.0 (95 percent CI: 8.68, 193.61), respectively. Although the evidence suggests that the risk of lung cancer is increased in carriers of both the GSTM1 null variant and the CYP1A1 variant, the wide confidence intervals obtained leave the results difficult to interpret (38–45). Studies investigating the interaction between the GSTM1 null variant and the glutathione S-transferase T1 (GSTT1) null variant have observed conflicting results, showing both reduced risk (42, 46) and increased risk (42, 44) of lung cancer for double null carriers.

Recently, Vineis et al. (47) conducted a pooled analysis through the GSEC (International Collaborative Study on Genetic Susceptibility to Environmental Carcinogens) initiative (48), including six case-control studies with 611 lung cancer cases and 870 controls genotyped for GSTM1 null, GSTT1 null, and CYP1A1 MspI, Ile/Val, and Thr/Asn. Associations with lung cancer were observed in carriers of either CYP1A1 MspI, Ile/Val, or Thr/Asn and the double deletion of both GSTM1 and GSTT1. For the CYP1A1 Thr/Asn and double GSTM1 and GSTT1 deletion carriers, an odds ratio of 8.25 (95 percent CI: 2.29, 29.77) was observed. The gene-gene interaction between GSTM1 and CYP1A1, simplistically summarized here, is the topic of another registered HuGE review (49).

Gene-environment interactions
An increase in lung cancer risk is favored when the effect of increased smoking is assessed along with that of GSTM1 variation (50–53), but at least one study (54) has demonstrated an opposite effect, that is, an increased odds ratio at a lower level of smoking. The discrepancy may be based on the lack of consistent controls for concomitant polymorphisms (e.g., multiple variants of CYP and GST) other than the primary one (e.g., GSTM1) being tested for in an individual study. In theory, induction of some polymorphisms (e.g., CYP1A1 MspI) by cigarette smoke leads to increased carcinogen exposure, while induction of others leads to decreased carcinogen exposure. Studies controlling for all relevant polymorphisms have been lacking, making it difficult to fully assess interactions between smoking and genotype. Therefore, it is currently unknown whether any potentiating interaction is occurring. Stucker et al. (53) have argued that the GSTM1 null genotype and cigarette smoking are independent risk factors for lung cancer but are not synergistic. The interaction between GSTM1, smoking, and lung cancer has been registered as the topic of a separate HuGE review (49).

There are several dietary compounds and toxic exposures that will also need to be controlled for in order to fully elucidate gene-environment interactions related to lung cancer risk. The most notable of these are isothiocyanates, found in high concentrations in cruciferous vegetables. London et al. (55, 56) found a decreased risk (odds ratio (OR) = 0.36, 95 percent CI: 0.20, 0.63) associated with the GSTM1 null genotype when patients were stratified by urinary isothiocyanate level, and Spitz et al. (57) found increased risk in persons reporting lower isothiocyanate intake. Lewis et al. (58, 59) found decreased risk with higher consumption of cruciferous vegetables (OR = 0.27, 95 percent CI: 0.06, 1.33), but the wide confidence interval makes this finding inconclusive.

Other potentially significant interactions include use of smoky coal, which Lan et al. (60) found to confer increased risk in GSTM1-null subjects, and rural living, which conferred increased risk in one study (61). There have thus far been mixed data for an effect of vitamin C intake. Garcia-Closas et al. (62) found a protective effect, but London et al. (54) found no significant association between vitamin C intake and GSTM1 status for lung cancer risk. Both Woodson et al. (63) and London et al. (54) failed to find significantly altered odds ratios for lung cancer when a GSTM1-null population was stratified by β-carotene intake.

Molecular methods for determining GSTM1 genotype have been reviewed by Cotton et al. (9).

Given the uncertain positive predictive value of GSTM1 genetic testing as a predictor for lung cancer risk, the clinical value of such testing is questionable. From a public health perspective, an optimistic goal would be to use genetic testing to supplement current efforts to motivate people to stop smoking, but there are considerable obstacles to achieving this goal (64). The theory that knowledge of polymorphism-related lung cancer may somehow guide behavioral change (given the "voluntary" nature of smoking) has been tested by Audrain et al. (65) and Lerman et al. (66). They measured motivation to quit, ultimate quitting rates, and depressive symptoms in patients randomized to receive quit-smoking counseling, patients randomized to receive counseling plus biofeedback, and a third group in which genotype testing was added to these two methods. While persons who were told of their genetic predisposition to cancer experienced short-term positive gains in perceived risk, perceived quitting benefit, and fear arousal, cessation rates were not affected by genetic risk knowledge. Initially, the biomarker group experienced increased levels of depressive symptoms, but these were not maintained over 12 months. The authors suggested (65, 66) that genetic susceptibility information might prove more compelling to persons who received more intensive counseling and/or newer pharmacologic support (nicotine patches, etc.) than was provided in these studies.

In a subsequent study by McBride et al. (67), in which smokers were randomized to "usual care" or biofeedback (consisting of GST genetic testing and counseling), there was a greater prevalence of smoking abstinence in the biofeedback group at 6 months but not at 12 months (although a trend persisted at 12 months). Interestingly, the difference at 6 months was based generically on the biofeedback/counseling process; no difference was noted between persons told that they had the GSTM1-null genotype ("susceptible") and persons told that they were GST-normal ("not susceptible"). On the basis of these limited trials, there is no current justification for any population-based testing. However, this question will need to be revisited as gaps in our understanding of this issue are addressed through further research.

In this paper, we have reviewed available evidence for the role of GSTM1 in predisposition to lung cancer. We conducted an updated meta-analysis of association studies involving a total of 19,638 cases and 25,266 controls from 98 studies, carefully avoiding the double-counting of participants in the analysis. The GSTM1 null variant was observed to be associated with a small increase in lung cancer risk (OR = 1.22, 95 percent CI: 1.14, 1.30), although no increased risk was apparent when only the five largest studies (>500 cases each) were considered (OR = 1.01, 95 percent CI: 0.91, 1.12). There was a suggestion that the GSTM1 null variant may confer increased risk in persons with an East Asian ethnic background (OR = 1.38, 95 percent CI: 1.24, 1.55), with a lack of convincing evidence for persons of Caucasian ethnicity (OR = 1.04, 95 percent CI: 0.97, 1.11). Although the studies that examined the relation of GSTM1 phenotype with lung cancer found a larger association (OR = 1.63, 95 percent CI: 0.96, 2.74), the confidence intervals were wide.

Several methodological issues should be considered in interpreting these findings. First, the key threat to literature-based reviews and meta-analyses is the possibility of reporting bias (the possibility that only the most exciting findings are available in the literature). We cannot rule out this possibility, not least because we observed a lack of association in the largest studies, which may be less prone to selective reporting. Second, higher levels of smoking may accentuate or minimize the effect of adverse genotypes on lung cancer risk. Tobacco smoking is the most firmly established risk factor for lung cancer (28, 68, 69). However, reporting of smoking exposure is not standardized, varies considerably across studies, and is difficult to address adequately in a review like this. Vineis et al. (70) have shown that the relation between lung cancer and smoking may level off at approximately 20 cigarettes per day. Third, polymorphism frequencies are known to vary by ethnicity (71), but the effect of this on risk has not yet been adequately studied. In the studies we identified, the frequencies of the GSTM1 null genotype among controls were similar in Caucasian and East Asian populations. The observed difference in the magnitude of the association between these populations does not appear to be explained by differences in genotype frequencies, suggesting more complex factors that warrant further investigation. Fourth, some studies have suggested that females may accumulate more adducts than males, even when smoking level and other confounding factors are controlled for (72). The clinical significance of this finding remains to be studied. When studies that reported results for females only (n = 6) and males only (n = 7) were subgrouped, odds ratios of 1.50 (95 percent CI: 1.06, 2.12) and 1.08 (95 percent CI: 0.91, 1.28), respectively, were observed for the GSTM1 null genotype and lung cancer (using a random-effects meta-analysis; data not otherwise shown).

It also appears that the effect of the GSTM1 genotype may vary according to histologic subtype. In our analyses, we evaluated the risk for each of the three major lung cancer subtypes. In spite of the variation in subtypes between studies, the odds ratios were elevated for squamous-cell carcinoma (OR = 1.23, 95 percent CI: 1.09, 1.39), small-cell carcinoma (OR = 1.33, 95 percent CI: 1.10, 1.60), and adenocarcinoma (OR = 1.13, 95 percent CI: 1.02, 1.25) when each type was considered independently. Our analyses indicated that previous meta-analyses have overestimated the effect of the GSTM1 null variant on each of the three main histologic subtypes (table 1). This is an area in which more research is warranted.

In addition to these questions, contributions from other gene variants may also be responsible for differences between studies. For example, genetic polymorphisms in the CYP family may modulate nicotine metabolism (73) or its effects on dopamine receptors (66) and therefore addiction. Possible interaction between GSTM1 and these CYP genotypes, and other polymorphisms theorized to modulate lung cancer risk, were infrequently investigated and rarely accounted for in the studies outlined in Web table 1. Bartsch et al. (74) have suggested that the interactions result in a greater-than-additive risk. These effect-modifying interactions were not taken into account in our analyses of the association between lung cancer and GSTM1 genotype. Realistically, however, comprehensive studies of genetic and environmental factors contributing to lung cancer may not be feasible until chip array technology allows for ready characterization of multiple relevant genes. Furthermore, making use of such technology when it becomes available will require large study samples in order to generate sufficient power to evaluate multiple potential contributors to risk. Researchers will need to consider the ability of the latest technology to address these concerns.

Because of the complex pathways of carcinogen metabolism and the various enzymes involved, any single gene might play a smaller, more limited role in the risk of lung cancer. In this review, we observed a modest effect of the GSTM1 null variant on lung cancer risk, and we would therefore encourage much larger studies than have traditionally been conducted in this area. Larger, more comprehensive studies would allow for meaningful stratification and allow stronger conclusions to be drawn regarding the effects of study characteristics such as ethnicity or histologic subtype. Larger studies would also permit evaluation of gene-gene and gene-environment interactions, factors that are clearly important in complex diseases such as lung cancer. In the process of exploring such research, it is imperative to use foresight in targeting it towards clear applicability to public health (64).

Editor's note: References 89–228 are cited in Web table 1.

Selection criteria and identification of studies
We sought all population-based cohort, case-control, and cross-sectional studies reporting associations between the GSTM1 null variant and lung cancer. Cases had to be diagnosed with lung cancer, and controls had to be healthy or hospital-based controls without cancer. Electronic searches, not limited to the English language, were performed using MEDLINE, EMBASE, BIOSIS, and the Science Citation Index, and we also perused the reference lists of retrieved articles and previous meta-analyses. The latest searches were undertaken on March 13 and 14, 2006. The MEDLINE search strategy, using PubMed Medical Subject Headings (MeSH), for assessing the association between the GSTM1 null variant and lung cancer was the following: (glutathione S-transferase* or glutathione S transferase* or glutathione transferase[MeSH] or GSTM1 or aryl hydrocarbon hydroxylases[MeSH]) and (lung or respiratory tract or lung[MeSH] or cancer* or neoplasm* or neoplasms[MeSH] or carcino* or carcinoma[MeSH] or tumour* or tumor* or tumour[MeSH] or DNA adduct* or DNA adducts[MeSH] or squamous cell carcinoma* or large-cell carcinoma* or small cell carcinoma* or adenocarcinoma* or non-small cell carcinoma* or lung neoplasms[MeSH] or respiratory tract neoplasms[MeSH]). Two reviewers (C. C. and G. S. S.) scanned relevant articles identified by the search independently on the basis of title, keywords, and abstract (where available) and rejected on an initial screen any article that clearly did not meet the inclusion criteria. The full text of all remaining articles was obtained for further evaluation by the same two reviewers. In the case of uncertainty about eligibility, a third reviewer (A. J. F.) was consulted before a decision was made.

Data collection and analysis
Data were extracted independently by two reviewers (A. J. F. and G. S. S.), using a prepiloted data extraction form (with any discrepancies being resolved by discussion). Variables on which information was collected were study design; geographic location; genotype frequencies, by categorical disease outcome (including clinical subtypes if presented); mean ages of cases and controls; proportions of males and persons in ethnic subgroups (defined as European continental ancestry, East Asian ancestry, or other, including African-American); genotyping method used; and blinding of laboratory workers to participant case-control status. Where multiple publications on the same study were identified, data were extracted from each article and the most complete and up-to-date information was identified. Studies that presented results for different ethnic groups or different control sources were considered as a single study for the overall analysis but were considered as individual studies for the ethnicity and control-source subanalyses, in order to avoid double-counting of individuals.

Primary analyses were conducted using a dominant inheritance model. Meta-analyses used a standard approach, weighting by precision and incorporating random effects to allow for the variation in true associations across studies. Funnel plots were used to assess assumptions involved in meta-analysis and to explore the relation between precision and magnitude of association. Consistency of the gene effect sizes across studies was assessed using the I² statistic, which describes the percentage of total variation in point estimates attributable to genuine variation rather than sampling error (75). Variation was further explored by prespecified subgrouping of studies according to sample size (<100, 100–499, or ≥500 cases), ethnicity (Caucasian, East Asian, other), source of controls (general population, hospital-based, other), study design (retrospective, prospective), and blinding of genotype to disease outcome (yes, no, unknown). Sensitivity analyses were also conducted by performing fixed-effect meta-analyses. All ranges presented are 95 percent confidence intervals unless otherwise specified.

This review was supported by the University of Washington's Center for Ecogenetics and Environmental Health, via US National Institute of Environmental Health Sciences grant P30ES07033. Aspects of this work were supported by funding from the United Kingdom Department of Health and the United Kingdom Department of Trade and Industry held under the program of the Cambridge Genetics Knowledge Park. G. S. S., A. J. F., and J. P. T. H. were supported by a research grant from the PHG Foundation (Cambridge, United Kingdom) held by the United Kingdom Medical Research Council.

The authors thank Dr. Zheng Ye for allowing access to previously extracted data from the Chinese literature and Dr. Karen Edwards for a review of the manuscript.

Drs. C. Carlsten and G. S. Sagoo contributed equally to this article.

Conflict of interest: none declared.

• Table 1
• Figure 1
• Web Table 1
• Web Figure 1

List of References