|Home | About | Journals | Submit | Contact Us | Français|
There is an overwhelming abundance of genetic association studies available in the literature, which often can be collectively difficult to interpret. To address this issue, the Venice interim guidelines were established for determining the credibility of the cumulative evidence. The objective of this report is to evaluate the literature on the association of common GST variants (GSTM1 null, GSTT1 null and GSTP1 Ile105Val polymorphism) and lung cancer, and to assess the credibility of the associations using the newly proposed cumulative evidence guidelines. Information from the literature was enriched with an updated meta-analysis and a pooled analysis using data from the Genetic Susceptibility to Environmental Carcinogens (GSEC) database. There was a significant association between GSTM1 null and lung cancer for the meta- (meta OR = 1.17, 95% CI: 1.10–1.25) and pooled analysis (adjusted OR = 1.10, 95% CI: 1.04–1.16), although substantial heterogeneity was present. No overall association between lung cancer and GSTT1 null or GSTP1 Ile105Val was found. When the Venice criteria was applied, cumulative evidence for all associations were considered “weak”, with the exception of East Asian carriers of the G allele of GSTP1 Ile105Val, which was graded as “moderate” evidence. In spite of large amounts of studies, and several statistically significant summary estimates produced by meta-analyses, the application of the Venice criteria suggests extensive heterogeneity and susceptibility to bias for the studies on association of common genetic polymorphisms, such as with GST variants and lung cancer.
The literature is awash with a plethora of genetic association studies on low penetrance genes, often with conflicting results, making it difficult to discern the true nature of the risk associated with variants in these genes. There have been many attempts at summarizing overall effects through systematic literature reviews and meta-analyses. According to the Human Genome Epidemiology Literature Finder, 1,167 meta-analyses on associations of various genes and diseases had been published as of October 30, 2009 . Unpublished data and potential for bias among the published literature threatens the validity and credibility of summary estimates produced by meta-analyses; as bias increases, the chances of findings being true decreases . Thus particular caution must be exercised when assessing the true effect of genes with borderline significant results , which is often the case with low penetrance variants. To address these issues, the Human Genome Epidemiology Network Working Group on the Assessment of Cumulative Evidence was established. In November 2006, the group convened and developed interim guidelines for assessment of the credibility of the cumulative epidemiologic evidence surrounding genetic association studies that were published one year later .
Lung cancer is the leader in cancer incidence and mortality in the United States, accounting for an estimated 215,020 cases and 161,840 deaths in 2008 . Although smoking is still the main risk factor responsible for lung cancer, individual susceptibility to the harmful effects of tobacco exists . This has elicited a strong interest in and a multitude of studies on polymorphisms involved in tobacco carcinogen metabolism and DNA repair to determine how genetic factors might modulate tobacco-related lung cancer risk.
Among the metabolic gene products, glutathione S-transferases (GSTs; EC 220.127.116.11) are a large family of cytosolic phase II xenobiotic metabolizing enzymes, whose function is to catalyze the detoxification of electrophilic metabolites, including benzo[α]pyrene and other polycyclic aromatic hydrocarbons (PAHs) found in tobacco smoke, forming soluble, nontoxic peptide derivatives to be excreted or compartmentalized by phase III enzymes . The most commonly studied GST variants include GSTM1 deletion, GSTT1 deletion, and GSTP1 Ile105Val (rs1695) polymorphism. The GSTM1 and GSTT1 deletions (null) are recessive variants for which homozygous deletions result in null activity of their respective enzymes. GSTP1 Ile105Val is a non-synonymous A>G SNP that leads to a substitution of isoleucine by valine resulting in alterations in the substrate binding site and enzyme activity .
The glutathione S-transferase gene family is the most extensively studied genetic susceptibility factor in relation to lung cancer, thus offering the opportunity to conduct the exercise of assessing and quantifying the credibility of the association, using the aforementioned Venice guidelines. The objective of this report is to summarize the evidence of the association of GSTM1 deletion, GSTT1 deletion and GSTP1 Ile105Val with lung cancer as it is currently reported in the literature and to apply the published interim cumulative evidence guidelines (4) to assess the credibility of the associations. The published information was enriched with a newly performed updated meta-analysis and a pooled (individual-level data) analysis.
All genotype-based case-control studies reporting on GSTM1 deletion, GSTT1 deletion or GSTP1 Ile105Val and susceptibility to lung cancer, published in English language from January 1991 (the first study examining the association of lung cancer and GST genotype ) through May 2009 were identified, along with all genome-wide association studies of lung cancer. The results of the search were further validated by reviewing the references included in recent published meta-analyses to ensure that all appropriate studies were captured. When a publication reported data in an ancestry-specific manner, each ancestral subset was considered as a discreet study. Overlap of study subjects between publications was evaluated by comparing sources of data described in the methods and through cross-referencing using dataset identifiers available through the Genetic Susceptibility to Environmental Carcinogens (GSEC) database (www.gsec.net) ; in the case of overlap, the more inclusive publication was selected for inclusion in the meta-analysis.
Data were combined across studies utilizing inverse-variance weighting to calculate fixed effects and random effects estimates (DerSimonian and Laird method) with corresponding 95% confidence intervals . In the absence of between-study heterogeneity (Q-statistic P > 0.05), fixed and random effects estimates are identical and therefore fixed effects were reported to conserve statistical power; otherwise random effects were reported. Heterogeneity was assessed using the Q-statistic and I2 metric, and susceptibility to bias was assessed using the Harbord  and significance chasing bias  tests. Details on these tests are provided in the Application of Venice Criteria section below.
Subset analyses were also conducted by ancestry (Whites, East Asian, African descent) and type of controls (hospital and community based) as potential sources of heterogeneity.
None of the six genome-wide association studies on lung cancer [14–19] reported results on GSTM1, GSTT1 or GSTP1 polymorphisms. Two of the GWAS studies employed platforms containing probes for copy-number variants [15, 17], whereas the other platforms require special protocols for GSTM1 or GSTT1 deletion . GSTP1 Ile105Val can be discerned by any of the platforms used in these studies. The authors were contacted, requesting results on the association between GSTM1, GSTT1, GSTP1 and lung cancer but they were unable to provide this information.
The Genetic Susceptibility to Environmental Carcinogens (GSEC) database , which compiles published and unpublished case-control data from participating investigators on genetic metabolic polymorphisms and various types of cancer was used to obtain data sets on lung cancer and GSTM1 deletion, GSTT1 deletion and GSTP1 Ile105Val for performing analyses on individual participant data. Analyses of pooled data were conducted for each GST variant using unconditional logistic regression models to produce estimates for the association with lung cancer, adjusted for age, gender, smoking status (ever/never), ancestry (White, East Asian, African descent and Other) and study. Heterogeneity and susceptibility to bias were tested using the same techniques as described for the meta-analysis; study-specific adjusted odds ratios were used for these analyses.
The cumulative evidence for the association of GSTM1 deletion, GSTT1 deletion and GSTP1 Ile105Val polymorphism with lung cancer was assessed according to the Venice interim guidelines  that use 3 major criteria: 1) amount of evidence; 2) replication of results; and 3) protection from bias. In the present paper for each gene-disease association, the cumulative evidence was derived for the current updated meta-analysis and from re-analyzing individual-level data from the GSEC database. The criteria were applied as previously described [4, 21]. In brief:
For each genetic polymorphism, the size of the smallest genetic group (nminor) was calculated by adding the total number of cases and controls with the least frequent genotype. The frequency of the minor genetic group (fminor) was estimated using the frequency of the minor genotype among the control subjects. When considering dominant models, the frequency of the less common genetic group was used. Grade A requires nminor > 1000, grade B is given for 100 ≤ nminor ≤ 1000 and grade C for nminor < 100.
When a non-significant association (P > 0.05) was observed, the association was considered as unreplicated (grade C). In addition, heterogeneity was assessed with the Q-statistic  and the I2 metric, given by (Q-df)/Q)  and 95% confidence intervals. I2 is a measure of how much heterogeneity is beyond chance, with values ranging from 0–100%. Values of I2 < 25% were classified as low heterogeneity, 25–50% as moderate heterogeneity, and >50% as large heterogeneity (grades A, B, and C, respectively) . However, qualitative aspects should also be considered when assessing consistency of replication.
Lack of protection from bias was claimed when any of the following criteria applied:
A Medline search was conducted to identify all previous meta-analyses published to date on GSTM1 deletion, GSTT1 deletion, or GSTP1 Ile105Val polymorphisms and lung cancer. The I2 metric and corresponding 95% confidence intervals were calculated for the meta-analysis whenever possible based on available information if not specifically reported. Results were contrasted with the results of the updated meta-analyses and each previous meta-analysis was graded whenever possible according to the same Venice criteria in an attempt to assess temporal changes in cumulative evidence.
All statistical analyses were performed using the Stata 10 software package (Stata Corporation, College Station, TX).
There were 82 publications on distinct datasets reporting on GSTM1 deletion and lung cancer [9, 24–104], 3 of which reported ancestry-specific subset analyses [36, 55, 68], for a total of 85 studies, consisting of 18,777 cases and 24,050 controls (Table 1).
Individual-level data for 48 of the 85 studies (56.5%) was available through the GSEC study, plus 8 unpublished datasets, for a total of 11,392 cases and 16,165 controls. The GSEC datasets corresponding to the published studies that were used in the meta-analysis contained 10,320 cases and 14,394 controls. The 8 unpublished datasets accounted for 1,072 cases (9.4%) and 1,771 controls (11.0%).
For published studies included in the meta-analysis, nminor was 21,042 and fminor in the control subjects was 0.493 (Table 1).
There were 43 publications on distinct datasets reporting on GSTT1 deletion and lung cancer [25, 27, 28, 33, 34, 36, 38, 40, 50, 53, 55, 59–63, 65–67, 69, 71, 74, 82–86, 88–92, 94, 97, 100, 101, 103, 105–110], 2 of which reported ancestry-specific subset analyses [36, 55], for a total of 45 studies with 12,569 cases and 15,545 controls (Table 1).
Individual-level data for 34 of 45 studies (75.6%) included in the meta-analysis was available through the GSEC study, plus 5 unpublished data sets, totaling 8,556 cases and 12,287 controls. The GSEC datasets corresponding to the published studies included in the meta-analysis contained 8,308 cases and 11,729 controls. The 5 unpublished datasets accounted for 248 cases (2.9%) and 558 controls (4.5%).
For published studies used in the meta-analysis, nminor = 6,585 and fminor = 0.232 (Table 1).
There were 34 publications on distinct datasets reporting on GSTP1 Ile105Val and lung cancer that analyzed the combined AG and GG genotypes (*G) [30, 32–34, 36, 47, 50, 57, 59, 60, 63, 66, 74, 77, 82, 83, 85, 86, 88, 89, 92, 96, 100, 101, 110–119], consisting of 10,125 cases and 10,867 controls, 30 of which (9,427 cases and 9,887 controls) also reported the association for the GG genotype separately from the AG genotype [30, 32, 36, 47, 50, 57, 59, 60, 63, 66, 74, 83, 85, 86, 88, 89, 92, 96, 100, 101, 110–113, 115–120] (Table 1). One of 34 publications reported subset data by ancestry , resulting in 35 and 31 datasets, respectively.
Individual-level data from 19 of 35 studies (52.9%) was available through the GSEC study, plus an additional 2 unpublished data sets, for a total of 5,506 cases and 6,206 controls. The GSEC datasets corresponding to published studies that were used in the meta-analysis contained 5,149 cases and 5,810 controls. The 2 unpublished datasets accounted for 357 cases (6.5%) and 396 controls (6.4%).
The nminor for GSTP1 Ile105Val publications used in the meta-analysis was 1,870 for the GG genotype, with fminor = 0.161 (Table 1). For the meta-analysis combining AG and GG genotypes, the AA genotype was the minor genetic group, where nminor = 10,458 and fminor = 0.497.
The current updated meta-analysis (Table 1) showed an overall significant association between GSTM1 null and lung cancer (meta OR = 1.17, 95% CI: 1.10–1.25). These findings were consistent with the adjusted estimates from the pooled analysis (OR = 1.10, 95% CI: 1.04–1.16). After conducting subset analyses by ancestry, a significant association among East Asian populations was observed in both the meta-analysis (meta OR = 1.21, 95% CI: 1.05–1.39) and pooled analysis (OR = 1.25, 95% CI: 1.07–1.45); there was a significant association among White populations in the meta-analysis (meta OR = 1.15, 95% CI: 1.06–1.26), which remained robust after excluding the first published study , while the adjusted estimate from the pooled analysis was not significant (OR = 1.05, 95% CI: 0.98–1.12); and there was no significant association in the meta-analysis for populations of African descent (meta OR = 1.25, 95% CI: 0.95–1.64), while the adjusted estimate from the pooled analysis was significant (OR = 1.32, 95% CI: 1.00–1.74).
The updated meta-analysis shows significant heterogeneity among studies (P < 0.001; I2 = 55%, 95% CI: 44%–63%). When stratified by ancestry (Table 1), heterogeneity was still present between studies conducted on East Asian and White populations, but not between the 4 studies on populations of African descent. Studies using community-based controls still showed a higher degree of heterogeneity (P < 0.001; I2 = 60%, 95% CI: 49%–69%), than those using hospital-based controls (P = 0.034; I2 = 34%, 95% CI: 5%–55%).
There are 6 previous meta-analyses for GSTM1 deletion and lung cancer [121–126], with summary point estimates ranging from 1.14 to 1.34, all statistically significant (Table 1), and similar to the results of the current updated meta- and pooled analyses.
The ORs in the meta-analyses restricted to East Asian populations [121–125, 127, 128] ranged from 1.38 to 1.60, all statistically significant and similar, although slightly higher in magnitude, to those of the current meta- and pooled analyses.
The summary estimates for the White population meta-analyses [121–125, 127] ranged from 1.04–1.32, with three meta-analyses showing statistically significant results. The current updated meta- and pooled analyses produced results that are in the middle of the continuum of the previously published meta-analyses.
The only previous meta-analysis on populations of African descent  reported a summary OR = 1.19 (95% CI: 0.88–1.62), similar to the findings of the current updated meta-analysis, although differing from the pooled analysis for which the adjusted estimate was significant.
Results on heterogeneity were not reported by the oldest overall meta-analysis ; all but one  of the most recent five meta-analyses reported significant heterogeneity (Table 1). When tested, heterogeneity was present in meta-analyses on East Asian studies [121, 122, 128], while no significant heterogeneity was reported by 2 of the 3 meta-analyses on Whites that tested for heterogeneity [121–123].
The current updated meta-analysis (Table 1) found no significant association of GSTT1 null and lung cancer (meta OR = 1.07, 95% CI: 0.97–1.17), with similar results from the pooled analysis (OR = 1.05, 95% CI: 0.97–1.14). A borderline association was observed among East Asians in the meta-analysis (meta OR = 1.19, 95% CI: 0.99–1.45), with a stronger association observed in the adjusted estimate from the pooled analysis (OR = 1.25, 95% CI: 1.07–1.45); no association was observed in White populations in the meta- (meta OR = 1.01, 95% CI: 0.91–1.14) or pooled analysis (OR = 1.04, 95% CI: 0.94–1.14).
Significant heterogeneity was observed in the current meta-analysis (Table 1), and was still observed in the subset analysis according to type of controls (community-based: P = 0.026 and I2 = 38%, 95% CI: 43%–76%; hospital-based: P < 0.001 and I2 = 63%, 95% CI: 8%–56%). The amount of heterogeneity was reduced in the subset analysis by ancestry.
There have been 2 published meta-analyses on GSTT1 deletion and lung cancer (Table 1) yielding comparable results (Raimondi : meta OR = 1.07, 95% CI: 0.96–1.19; Ye : meta OR = 1.09, 95% 1.02–1.16) to those of the current updated meta- and pooled analyses.
There was 1 published meta-analysis that conducted subset analysis by ancestry for GSTT1 polymorphism  (Table 1), reporting a significant association among East Asians (meta OR = 1.28, 95% CI: 1.10–1.49), with borderline heterogeneity (P = 0.09); null findings were observed in White populations (meta OR = 0.99, 95% CI: 0.87–1.12) with significant heterogeneity (P = 0.02), consistent with the updated estimates.
The current updated meta-analysis (Table 1) shows no association of lung cancer with combined AG and GG genotypes (meta OR = 1.03, 95% CI: 0.97–1.09) or for GG genotype alone (meta OR = 1.05, 95% CI: 0.95–1.16). The results from the pooled analysis are consistent with these findings (OR = 0.98, 95% CI: 0.86–1.13 and 1.06, 95% CI: 0.97–1.15 respectively).
There was a significant association between G allele carriers and lung cancer in East Asian populations in the current meta-analysis (OR = 1.16, 95% CI: 1.01–1.32); and for the AG genotype in the pooled analysis (OR = 1.31, 95% CI: 1.03–1.67). There was no significant association with the GSTP1 Ile105Val polymorphism and lung cancer among White populations.
There was no significant between-study heterogeneity overall or in any of the subset analyses other than in East Asian populations (Table 1).
Two recent meta-analyses on GSTP1 Ile105Val and lung cancer reported borderline associations of the G allele and lung cancer (Ye : meta OR = 1.04, 95% CI: 0.99–1.09; Cote : meta OR = 1.04, 95% CI: 0.97–1.10), with no heterogeneity, consistent with the current updated meta-analysis (Table 1).
The recently run meta-analysis of GSTM1 deletion and lung cancer suggests the presence of small-study effect (Table 2) overall (Harbord test: P = 0.011) and in the community-based control subset (P = 0.010) but not in the ancestry-specific analyses.
In the overall analysis, the significant findings among the published studies were not statistically significantly more than expected (Table 2). However, the test was statistically significant in the East Asian subset (P = 0.002) and in studies involving hospital-based controls (P = 0.010).
There was no evidence of small-study effect in the overall meta-analysis of GSTT1 deletion and lung cancer (Table 2). In the subset analyses, there was evidence of small-study effect only in studies involving community-based controls (P = 0.079)
The significant findings were more than expected among the published studies (Table 2) overall (significance chasing bias: P < 0.001), in the subset analyses according to control source (P = 0.001 for both hospital- and community-based) and in the subset of studies conducted in Whites (P < 0.001).
There was evidence of small-study effect (Table 2) among publications on the association between GSTP1 Ile105Val G allele carriers and lung cancer (Harbord test: P = 0.014), but not for studies evaluating the association of GG genotype and lung cancer. In the subset analyses, only the community-based control group had significant evidence for small study effect (P = 0.024).
There were not more than expected significant findings for the overall associations of GSTP1 Ile105Val G allele carriers or GG genotype (Table 2). In the subset analyses, there were more than expected significant findings among studies on the GG genotype in White populations (P = 0.002) and in hospital control studies (P = 0.067).
Five publications did not report the genotype distribution of the controls [33, 34, 77, 82, 114] and therefore Hardy-Weinberg equilibrium (HWE) could not be calculated. Controls for 5 (16.1%) of the remaining 31 studies were not in HWE (Supplemental Table S1). When the meta-analysis was run excluding those 5 studies not in HWE, the results did not significantly change (data not shown).
Based on the previously proposed guidelines , the amount of evidence was categorized as A since its nminor is above 1,000 (nminor = 20,941); replication was assigned to category B because, despite the amount of between-study heterogeneity (I2 = 55%), the vast majority of the point estimates from the individual studies are positive (OR > 1.0); even after performing subset analyses by ancestry, the meta and pooled analyses consistently showed a significant association with lung cancer, with the exception of the meta-analysis for people of African descent; and protection from bias was graded as category C due to considerable potential for bias, as evidenced by the Harbord test for small study effect. The HWE could not be evaluated due to lack of publication of necessary genotyping data (homozygous present genotype was always reported combined with heterozygotes). The overall assessment of the association between GSTM1 deletion and lung cancer would be: weak cumulative evidence (Table 3). A similar overall assessment is reached for the ancestry-specific subset analyses.
The amount of evidence was scored as A since nminor exceeds 1,000 (nminor = 6,585); replication was classified as C as a result of large-scale between-study heterogeneity (I2 = 51%) and lack of statistically significant findings; and protection from bias was placed into category C since there were more than expected significant results (significance chasing bias). The overall classification of the association between GSTT1 deletion and lung cancer is: weak evidence (Table 3). A similar assessment is reached for the ancestry-specific subset analyses.
For the association between GSTP1 Ile105Val GG genotype and lung cancer, the amount of evidence was classified as A since it exceeds the nminor threshold of 1,000 (nminor = 1,870); replication was considered in category C because there was no overall significant association; and protection from bias was considered as category B because the Harbord and excess-significance tests were non-significant, but the ORs consistently deviated less than 1.15-fold from the null, thus potential for bias cannot be entirely ruled out. The overall classification of the association between GSTP1 Ile105Val GG genotype and lung cancer is: weak evidence (Table 3). A similar assessment is reached for the ancestry-specific subset analyses.
For the association of the G allele (AG and GG genotypes combined) and lung cancer, the amount of evidence was categorized as A since nminor greatly exceeds 1,000 (nminor = 10,458); replication was classified as C due to lack of overall significant findings; and protection from bias was categorized as C due to significant potential for bias as evidenced by significance of the Harbord test. The overall classification of the association between GSTP1 Ile105Val AG and GG genotype and lung cancer is: weak evidence (Table 3). A similar assessment is reached for the White subset, although for East Asian populations there was moderate evidence.
Among the meta-analyses providing sufficient information for grading, all would be assessed as having weak cumulative evidence based on the Venice criteria (Table 4). The criteria used to score the existing meta-analyses are presented in the Supplemental Table S2. For most of the meta-analyses, incomplete information was given for fully assessing heterogeneity and publication bias.
Glutathione S-transferase polymorphisms have been widely studied in the literature, and thus there have been many attempts over time to synthesize the effect of genetic variants through meta-analyses. Compared to previous attempts of meta-analyzing data, we have used more rigorous and standardized criteria, as defined through the Venice initiative . These have been specifically developed for genetic data, and include an evaluation of the amount of evidence available, the degree of consistency/replication of findings, and of bias. With such strict criteria we have found that overall there is not strong evidence of an involvement of commonly studied GST polymorphisms in lung cancer.
Application of the Venice criteria indicates weak evidence in support of an association with GSTM1 null, GSTT1 null, and GSTP1 Ile105Val polymorphisms and lung cancer, both overall and after analysis of ancestral subsets. The only exception is for the GSTP1 Ile105Val G allele in East Asian populations, where the evidence for association with lung cancer is moderate. This result may be attributable to differences in genotype frequencies in East Asian populations, more carefully designed studies, or local differences in exposure to relevant carcinogens, although bias cannot be fully excluded .
The available a priori evidence on a role of GSTs in lung cancer is rather strong, i.e. a criterion of biological plausibility can be easily met. This is why, in fact, they have been studied extensively. Glutathione S-transferases catalyze the detoxification of electrophilic metabolites, including benzo[α]pyrene and other polycyclic aromatic hydrocarbons (PAHs) found in tobacco smoke, which are likely to play a key role in lung carcinogenesis. The GSTM1 and GSTT1 deletions result in null activity of their respective enzymes, and so a strong functional impact would be expected. However, the loss of function associated with these gene variants is likely compensated by the existence of alternative metabolic pathways, which may help to explain why the effect, if any, is weak.
Despite the fact that GST variants are considered to be among the most extensively studied genetic markers of susceptibility, and that GSTM1 deletion in particular exhibited statistically significant associations with lung cancer in previous meta-analyses, the application of the Venice criteria shows that when this standardized method for evaluating the cumulative evidence of association is applied, the previous meta-analyses would fall into the category of “weak evidence”. While the estimates of effect size changed over time for some associations, inconsistencies dominate and the evaluation of “weak” is mainly related to heterogeneity and/or lack of protection from bias.
In spite of the systematic approach outlined by the Venice interim criteria, the present exercise shows that there may be room for improvement in the standardization and performance of the Venice criteria . One issue that emerged from this analysis is the possibility that different weights should be assigned to the different criteria included in the Venice method. For example, the assessment of protection from bias should probably have less weight than the consistent replication across studies and meta-analyses of a very strong association, with a large effect estimate (e.g. an odds ratio of 2.0). However, when small effects are observed, as in the case of the associations of GST variants and lung cancer, the presence of a modest bias could easily contribute to producing a spurious, small association of the magnitude reported in this re-analysis. In the area of genetic susceptibility, where small associations are expected, hints for bias cannot be discarded even though tests for bias are not perfect [132, 133].
One aspect emerging from our exercise is the good correspondence of the results obtained with the meta- and respective pooled (individual-level) analyses. These two approaches have different principles, strengths, and limitations and may not always agree in their results . For some analyses, such as those restricted to populations of African descent, the pooled analysis offered the appropriate data that were missing in the meta-analysis, thus allowing the drawing of a more complete picture. Another potential strength is that in several instances the replication of the meta-analysis was classified with a grade of C due to large amounts of heterogeneity, while the pooled analysis constantly produces less heterogeneity. This could be attributable to the ability of pooled analyses to deliver adjusted estimates but also may reflect the smaller sample size relative to the meta-analyses. It should also be acknowledged that heterogeneity estimates have substantial uncertainty, as evidenced by the large 95% confidence intervals of heterogeneity metrics, even when a large number of studies are available .
As expected from the role that metabolic genes have in carcinogenesis, the associations derived from individual studies and from summary estimates in meta- and pooled analysis confirm the definition of “low penetrance” genes, with increases in lung cancer risk ranging from 15%–21% at the most, but with many studies showing no significant associations. The original Venice criteria applied here weight the credibility that an association is present, but do not score the certainty that the association is absent. Very small effects (e.g. odds ratios of 1.05) cannot be excluded with certainty.
The effort to date regarding elucidation of the true effects of GSTM1 deletion, GSTT1 deletion, and GSTP1 Ile105Val polymorphism has been a resource-consuming, laborious task with a lot of uncertainties surrounding the evidence collected so far. In spite of the biologic plausibility due to involvement in xenobiotic detoxification, GST polymorphisms have not fulfilled their early promise as biomarkers of lung cancer susceptibility. Rather, the large body of literature either displays wide variability between studies, yields no overall significant summary effect, or both. This seems to be a common feature of studies on metabolic gene variants and cancer. Metabolic gene variants are known to produce very small odds ratios in both directions, either for a positive or negative association, and the significant heterogeneity and susceptibility to bias shown in most of the analyses conducted so far may have contributed to dilute the real effect, if any exist. It is certainly possible that poor replication – leading to a high degree of heterogeneity - is related to local differences in exposures affecting the populations from which samples are drawn, (for example differences in environmental pollution, smoking habits or dietary factors) which are difficult to incorporate in the score proposed by the Venice guidelines. It is possible that restricting the assessment to studies including a formal test for gene-environment interaction would yield more positive evaluations of the overall associations. One should also consider the degree of misclassification that is much higher for environmental or lifestyle exposures than for gene variants .
A further aspect highlighted by this exercise is that certain ethnic groups, such as populations of African descent, are still highly understudied. No conclusion about a possible association between susceptibility genes and cancer can be drawn at present.
In light of the limitations and determination of weak cumulative evidence shown by this exercise that evaluated the credibility of the association between commonly studied GST polymorphisms and lung cancer, future studies may try to evaluate interaction of GST variants with other genes in adequately powered datasets, as well as include the assessment of gene-environment interaction. This may allow for identification of gene–disease associations and interactions where the cumulative evidence for association is strong, while evidence for main effects is conclusively uniformly weak. Application of criteria for assessment of the strength of cumulative evidence, such as the Venice criteria, to meta-analyses will help to clarify the validity of summary estimates for gene-disease associations, making it easier for researchers and clinicians to comprehend the true nature of the risk involved, if any.
We would like to thank Barbara Stadterman for her assistance in preparation of the GSEC data.
Funding sources: the Environmental Cancer Risk, Nutrition and Individual Susceptibility European Union Network of Excellence (ECNIS, Contract No. FOOD-CT-2005-513943); Clinical and Translational Science Institute and the Institute for Clinical Research Education at the University of Pittsburgh (grant 5TL1RR024155-03 to SM Langevin)
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
*Genetic Susceptibility to Environmental Carcinogens group; collaborators to the present analysis: José A.G. Agundez, Anna-Karin Alexandrie, Ravindran Ankathil, Herman Autrup, Judith L.C. Autrup, Juan M. Barros-Dios, Simone Benhamou, Paolo Boffetta, Katja Breskvar, Jurgen Brockmoller, Dorota Butkiewicz, Ingolf Cascorbi, Margie L. Clapper, Michele L. Cote, Ioanna A. Dialyna,Vita Dolzan, Tommaso Dragani, K.M. Fong, Martha P. Gallegos-Arreola, Seymour Garte, Andrea Gsur, Curtis C. Harris, Aage Haugen, Evgeny N. Imyanitov, Magnus Ingelman-Sundberg, Ivan Kalina, Daehee Kang, Masahiro Kihara, Chikako Kiyohara, Pierre Kremers, Qing Lan, H. Lee, Loïc Le Marchand, Stephanie J. London, D. Lucas, Maria Li Lung, Valle Nazar-Stewart, Kazumasa Noda, Hatice Pinarbasi, Paola Pisani, Andrew C. Povey, Yuepu Pu, Agneta Rannug, Angela Risch, Liliane Roelandt, Marjorie Romkes, David Ryberg, Joachim Schneider, Bernadette Schoket, Janeric Seidegard, Adeline Seow, Peter G. Shields, Ranbir C. Sobti, Margaret R. Spitz, Richard C. Strange, Isabel Stücker, Haruhiko Sugimura, Jordi To-Figueras, Jingwen Wang, John Wiencke, Ping Yang, Jun Yokota, Lair Zambon