Genetic variants of glutathione S-transferase as possible risk factors for hepatocellular carcinoma: a HuGE systematic review and meta-analysis.

Correspondence to Dr. Hashem El-Serag, Michael E. DeBakey Veterans Affairs Medical Center, Room 152, 2002 Holcombe Boulevard, Houston, TX 77030 (e-mail: hasheme@bcm.tmc.edu).

Received for publication March 8, 2007. Accepted for publication September 24, 2007.

The authors performed a systematic review and meta-analysis to determine the effect of polymorphisms in genes encoding glutathione S-transferases (GSTs), phase II isoenzymes involved in cellular detoxification, on risk of hepatocellular carcinoma (HCC). Fifteen eligible studies were identified: 14 evaluated GSTM1; 13, GSTT1; three, GSTP1; and one each evaluated GSTM2, GSTM3, GSTA1, GSTA4, GSTO1, and GSTO2, respectively. All were case-control studies performed in populations with high (Asian, African) and medium (European) HCC incidence rates. Random-effects meta-analyses suggested a small excess risk of HCC with GSTT1 null (odds ratio (OR) = 1.19, 95% confidence interval (CI): 0.99, 1.44) and possibly GSTM1 null (OR = 1.16, 95% CI: 0.89, 1.53) genotypes. Cumulative meta-analyses demonstrated that both pooled estimators generally trended toward a small excess risk with publication of more recent studies. Results for GSTP1 A313G suggested no excess risk (OR = 0.75, 95% CI: 0.50, 1.15). A number of potentially interesting gene-gene and gene-environment interactions were reported, but these were too few and inconsistent to allow meta-analysis. The overall results suggest that there may be a small excess risk of HCC in individuals with GSTT1 null and possibly also with GSTM1 null genotypes. However, given the relatively limited total number of subjects examined and observed between-study heterogeneity, chance could not be excluded.

Keywords: carcinoma, hepatocellular; epidemiology; genetics; glutathione S-transferase pi; glutathione transferase; humans; liver neoplasms; meta-analysis
Abbreviations: CI, confidence interval; GST, glutathione S-transferase; HCC, hepatocellular carcinoma; OR, odds ratio

Primary liver cancer is the fifth most common cancer worldwide and the third most common cause of cancer mortality (1). Hepatocellular carcinoma (HCC) accounts for between 85 percent and 90 percent of all primary liver cancers. There are marked geographic variations in the distribution of HCC, with most cases occurring in either sub-Saharan Africa or in eastern Asia. Racial and ethnic variations in incidence have been observed even within the same geographic region. Further, in virtually all geographic regions and racial/ethnic groups, men are at a two- to fourfold greater risk of developing HCC (2). Environmental risk factors such as viral hepatitis seem to explain most but not all of these variations.

A very small minority of HCC cases are associated with familial disorders of mendelian inheritance, such as hereditary hemochromatosis or alpha-1-antitrypsin deficiency, whereas the great majority of adult-onset HCC cases are sporadic, with many having at least one established nongenetic risk factor, such as habitual alcohol abuse, chronic infection with hepatitis B or C virus, or exposure to aflatoxin B1, a powerful dietary hepatocarcinogen. However, most individuals with these known environmental risk factors never develop cirrhosis or HCC, while a sizable minority of HCC cases develop among individuals without known risk factors.

Genetic variation has been postulated to influence the variable risk for HCC observed both within and across populations. It has recently become possible to perform large-scale epidemiologic studies to evaluate genetic risk factors given rapid advances within the field of genomics, including the completion of the Human Genome Project and a dramatic reduction in the cost of genetic testing.

Currently, the most extensively studied inherited genetic risk factors for HCC are variants of glutathione S-transferases (GSTs) (Enzyme Commission (EC) number 2.5.1.18). GSTs are a broadly expressed family of phase II isoenzymes that protect against endogenous oxidative stress, as well as exogenous potential toxins. They detoxify a variety of electrophilic compounds, including oxidized lipid and DNA products generated by reactive oxygen species damage to intracellular molecules (3). Several GST variants are principally expressed in the liver, an organ whose primary functions include detoxification and metabolism. In the liver, inflammation is related to a variety of insults, including hepatitis B virus, alcohol, and hepatitis C virus, which are sources of reactive oxygen species. The liver is also exposed to other carcinogenic by-products as the result of both normal and drug metabolism, as well as exposure to exogenous environmental toxins such as aflatoxin (3–5).

Cytosolic and membrane-bound forms of GST are encoded by two distinct supergene families. At present, eight distinct classes of the soluble cytoplasmic mammalian GST have been identified: alpha, kappa, mu, omega, pi, sigma, theta, and zeta (6). GSTM1 is one of the genes encoding the mu class of enzymes located on chromosome 1p13.3 and contains 10 exons (7). The theta class of GST enzymes is encoded by the GSTT1 gene, which is mapped to chromosome 22q11.23 and contains six exons (8). The GSTP1 gene encodes the pi class of enzymes, the gene is located on chromosome 11q13, and it has nine exons (7).

The GST genes are highly polymorphic and frequently inducible. Among the numerous GST genes, GSTM1, GSTT1, and GSTP1 genes have been extensively examined in association with risk of cancer (6, 9) and clinical outcomes of cancer patients (7, 10–13). The most common variant of the GSTM1 and GSTT1 genes is homozygous deletion (null genotype), which has been associated with the loss of enzyme activity and increased vulnerability to cytogenetic damage (14, 15). According to a report from pooled data of 12,525 Caucasians, 2,136 Asians, and 996 African Americans (16), the frequency of the GSTM1 null genotype is 53 percent (range: 42–60 percent) in Caucasians, 27 percent (range: 16–36 percent) in African-American subjects, and 53 percent (range: 42–54 percent) in Asians. The frequency of the GSTT1 gene deletion is 20 percent (range: 13–26 percent) and 47 percent (range: 35–53 percent) for Caucasians and Asians, respectively.

In the liver, GST has been proposed to protect against hepatitis B virus-related injury, which is partly manifested as extensive oxidative DNA damage (17). GSTs, in particular GSTM1, are also aflatoxin-metabolizing enzymes and therefore may be involved in regulating individual ability to metabolize the important aflatoxin-related hepatocarcinogen, exoepoxide. Therefore, it has been proposed that the decreased production of GSTM1 and GSTT1 in the null types is expected to be associated with an increased HCC risk in the presence of hepatitis B virus infection and/or aflatoxin exposure.

The most extensively studied GSTP1 variants are exon 5 A1404G encoding an Ile→Val exchange at codon 105 (Ile105Val) (reference single nucleotide polymorphism (rs) accession identification number (rs#947894)) and exon 6 C2294T encoding an Ala→Val exchange at codon 114 (Ala114Val) (rs#1799811); both have been shown to confer lower levels of metabolic activity (18–20). The minor allele frequency of the Ile105Val variant is 31 percent, 54 percent, and 17 percent for Caucasians, African Americans, and Asians, respectively. The Ala114Val minor allele is present in about 10 percent of Caucasians but absent in African Americans and Asians (21).

Individual genetic association studies, such as those that have evaluated selected GST polymorphisms and risk of HCC, are frequently highly underpowered and often report small or variable effects. Meta-analysis has been recognized as an important tool to more precisely define the effect of selected polymorphisms on risk of disease and to identify potentially important sources of between-study heterogeneity (22, 23). We therefore conducted a systematic review and meta-analysis to more precisely define the effect of GST polymorphisms on risk for HCC.

Eligibility criteria
We followed recognized guidelines for the conduct and reporting of meta-analyses (24). All epidemiologic studies examining the effect of GST polymorphisms on HCC risk within a geographically defined human population were eligible for inclusion if they had a case group with medically documented HCC, a nonrelated and comparable control group without HCC, and GST genotype confirmed by polymerase chain reaction or related DNA identification methods.

To minimize potential publication bias, no restrictions were placed on time period, sample size, population, language, or type of report (e.g., abstract, dissertation, or manuscript). However, given that a critical requirement in meta-analysis is the statistical independence of observations (25), when multiple reports were available for a single unique study population, we included only the most recent or largest report.

Studies were excluded if the case group consisted of unspecified liver cancer or if the control group consisted of nondiseased tissue from the cases or first-degree relatives of the HCC cases or had liver disease or cancer without adequate exclusion of possible HCC.

Search strategy
To identify all potentially eligible studies, two investigators independently conducted keyword searches in selected databases (PubMed, OMIM, EMBASE, ISI Web of Science, and Dissertation Abstracts) and in the electronic conference proceedings for the International Genetic Epidemiology Society (1997–2005) and the American Society for Human Genetics (1999–2006). Searches included all possible combinations of selected keywords including "hepatocellular carcinoma," "liver cancer," or "HCC" with "genes," "inherited factors," "genetic polymorphisms," "GST"; "glutathione S-transferase"; "GSTA"; "GSTA1"; "GSTA2"; "GSTA4"; "GSTK'"; "GSTK1"; "GSTM"; "GSTM1"; "GSTM2"; "GSTM3"; "GSTS"; "GSTS1"; "GSTT"; "GSTT1"; "GSTT2"; "GSTP"; "GSTP1"; "GSTO"; "GSTO1"; "GSTO2"; "GSTZ"; and "GSTZ1." Searches were updated as of November 1, 2006.

We also used ancestry methods, including review of the bibliographies of all eligible studies and from relevant review articles, to identify additional studies not captured by our keyword searches.

Data abstraction
Two investigators independently reviewed titles, abstracts, and manuscripts identified by keyword searches to determine if an individual study was eligible for inclusion in the meta-analysis. All disagreements about eligibility were resolved during a consensus meeting with a third reviewer.

Data on study methods and findings were entered directly into a structured database. When the most recent or largest report for an eligible study population did not include sufficiently detailed information on methods, data from any earlier or smaller reports were used. When specific results were not reported, we used available tabular data to calculate them. When data were otherwise unavailable, we contacted the corresponding author by e-mail for additional information.

Analysis
The effect measure of choice was the odds ratio and associated 95 percent confidence interval. Because GSTM1 and GSTT1 have common deletion polymorphisms leading to absent or reduced enzymatic activity, studies have typically compared null with nonnull genotypes for HCC risk. For all other polymorphisms, planned comparisons included wild-type versus rare homozygotes and heterozygotes. We recalculated all reported odds ratios comparing differences in genotype distribution among cases and controls, as well as tests for Hardy-Weinberg equilibrium in controls, using reported tabular data. When there was a discrepancy between reported and calculated results, the tabular calculations were used for all subsidiary analyses.

We evaluated heterogeneity across studies using the I² of Higgins and Thompson (26), which quantifies the proportion of total between-study variation attributed to actual between-study differences or heterogeneity as opposed to random error or chance.

Our decision to perform a fixed or a random-effects meta-analysis was based on our heterogeneity assessment. When substantive between-study heterogeneity is observed, a random-effects meta-analysis (27) is the preferred method for obtaining the pooled estimator. All meta-analyses are presented as forest plots that include odds ratios and 95 percent confidence intervals for all individual studies, as well as the pooled estimator. Shaded figures provided for all odds ratios have dimension proportional to study weight.

We decided a priori to evaluate the following variables as potential sources of heterogeneity with random-effects meta-regression: ancestry, source of the controls, sample size, date of publication, and whether the predominant nongenetic risk factor was infection with hepatitis B. We performed an analysis of influence to determine the effect of omission of individual studies on overall pooled estimators, as well as a cumulative meta-analysis by year of publication, to determine how the pooled estimator changed as evidence accumulated over time. Finally, we used Egger's regression test, which assesses whether the relation between effect size and variance differs between large and small studies, to assess whether there was potential small-study or publication bias (28).

All analyses were conducted using STATA, version 9.0, software (StataCorp LP, College Station, Texas).

Searches
We identified 30 potentially eligible reports (29–58). All were case-control studies, and most were reported in English (80 percent). The majority (87 percent) were identified by structured keyword searches in PubMed . However, four reports were identified solely by ancestry methods (38, 47, 48, 54), of which three were in Chinese (38, 48, 54).

Review of abstracts and manuscripts resulted in exclusion of 15 reports because they were earlier or smaller reports from an eligible study population (n = 11) (44–50, 53–55, 58), included the same results presented in the eligible report (n = 2) (43, 52), or had an ineligible control group (n = 2) (56, 57).

Study characteristics
Baseline characteristics for the 15 eligible studies are provided in table 1. Fourteen evaluated GSTM1 (29–40, 42, 51), 13 evaluated GSTT1 (29–38, 40, 42, 51), three evaluated GSTP1 A313G (36, 38, 51), one evaluated both GSTO1 A140D and GSTO2 N142D (41), and one evaluated GSTA1, GSTA4, GSTM2, and GSTM3 (51).

click to view

Table 1  Characteristics of studies eligible for review and meta-analyses evaluating the effect of the glutathione S-transferase genotype on risk of hepatocellular carcinoma

 
Most of the eligible studies were performed in populations of Asian ancestry (n = 11). Sample size was variable though generally small, with six studies having less than 100 HCC cases and five studies with a total sample size of less than 250. More than half of all studies used hospital-based controls (n = 8), and many used prespecified criteria to match cases to controls (n = 12). A minority of studies (n = 3) were specifically restricted to hepatitis B carriers (31, 37, 39).

Association of GST polymorphisms and HCC
GSTM1.
The 14 studies evaluating the association of the GSTM1 polymorphism and HCC included a total of 2,514 cases and 4,416 controls (29–40, 42, 51), with most DNA samples successfully genotyped for each polymorphism (96.7 percent). Thirteen studies solely determined the presence versus the absence of GSTM1 (null vs. nonnull). Therefore, data were unavailable to assess Hardy-Weinberg equilibrium among controls.

The relative frequency of the GSTM1 null genotype among control groups varied widely from 25.9 percent in ancestral Africans from The Gambia (29) to 60.2 percent in ancestral Chinese from Taiwan (37). Eight of 14 studies showed small to modest excess risk with GSTM1 ranging from 11 percent (35) to 147 percent (42), although the excess was statistically significant in only five studies (32, 38–40, 42).

Because substantial between-study heterogeneity was observed (I² = 84 percent), we used a random-effects model. Overall, the pooled estimator for GSTM1 null suggested that it might convey a small excess risk for HCC, although this effect was not statistically significant (odds ratio (OR) = 1.16, 95 percent confidence interval (CI): 0.89, 1.53) (figure 1).

click to view

Figure 1 : Study-specific and random-effects pooled odds ratios with 95% confidence intervals for risk of hepatocellular carcinoma with GSTM1 null genotype. The box size is proportional to the meta-analysis study weight. Ref, reference citation.

 
We used random-effects meta-regression to evaluate five potential sources of between-study heterogeneity specified a priori, including the following: source of controls (hospital based vs. not), year of publication, sample size, ancestry of controls, and whether the study was restricted solely to hepatitis B carriers. Univariate analysis suggested that the source of the controls (τ² = 0.14, p < 0.01) and year of publication (τ² = 0.06, p < 0.055) were potentially important and explained 38 percent and 21 percent of between-study heterogeneity, respectively. However, the small number of studies prohibited further exploratory evaluation with multivariate analysis.

We found no evidence of substantive small-study or publication bias among these 14 studies (p_Egger's = 0.71). Further, our analysis of influence demonstrated that omission of any single study did not markedly change overall findings, with pooled estimators for the GSTM1 null genotype ranging from an odds ratio of 1.25 (95 percent CI: 0.97, 1.62) to an odds ratio of 1.10 (95 percent CI: 0.85, 1.41) with the removal of the smallest (37) and largest (42) study, respectively (data not shown). Finally, our cumulative meta-analysis by year of publication showed that the pooled estimator generally trended toward a small though nonsignificant excess risk with publication of the final eight studies between 2005 and 2006 (Appendix figure 1) (29–32, 35, 39, 40, 42).

GSTT1.
Thirteen studies (29–38, 40, 42, 51) with a total of 2,423 cases and 4,327 controls also evaluated GSTT1. As with GSTM1, only the presence versus the absence of the GSTT1 polymorphism was determined, so data were unavailable to assess Hardy-Weinberg equilibrium.

The relative frequency of the GSTT1 null genotype among control groups varied widely from 18.0 percent in Caucasians from Italy (30) to 60.2 percent in Chinese from Taiwan (31). Although a greater proportion of studies reported excess risk of HCC with GSTT1 null (n = 9) (29, 31–33, 35, 36, 38, 40, 42), the excess risk was generally very small to modest, and fewer studies reached statistical significance (38, 40, 42).

As substantial between-study heterogeneity was observed with GSTT1 (I² = 63 percent), we again used a random-effects model. Overall, the GSTT1 null genotype conveyed a similarly small increased risk of HCC that approached significance (OR = 1.19, 95 percent CI: 0.99, 1.44) (figure 2).

click to view

Figure 2 : Study-specific and random-effects pooled odds ratios with 95% confidence intervals for risk of hepatocellular carcinoma with GSTT1 null genotype. The box size is proportional to the meta-analysis study weight. Ref, reference citation.

 
Meta-regression identified the source of the controls as the only significant source of between-study variability (τ² = 0.035, p < 0.01), explaining 55 percent of total observed variability. Most of the additional analyses performed for GSTT1 had results similar to those for GSTM1, including the following: absence of small-study or publication bias (p_Egger's = 0.49), robustness of the pooled estimator to removal of single studies (data not shown), and a trend toward a small excess risk that approached significance with GSTT1 null with the publication of the final seven studies between 2005 and 2006 (29–32, 35, 40, 42) (Appendix figure 2).

GSTP1.
Three studies evaluated GSTP1 A313G, and all were conducted in populations of Asian ancestry (36, 38, 51). There were a combined total of 393 cases and 538 controls, all with successfully genotyped DNA data. One study solely reported a lack of association between GSTP1 polymorphisms and HCC (p < 0.17) and did not provide by-genotype data (51). Among the other two studies, one solely reported tabular data that combined heterozygotes and rare homozygotes into a singular "high-risk" genotype ("any G") (36). Therefore, only a similarly combined pooled estimator could be calculated, and Hardy-Weinberg equilibrium could be confirmed only in the other study. A small and nonsignificant decreased risk of HCC was observed with the high-risk genotype (OR = 0.75, 95 percent CI: 0.50, 1.15) (figure 3).

click to view

Figure 3 : Study-specific and random-effects pooled odds ratios with 95% confidence intervals for risk of hepatocellular carcinoma with GSTP1 A313G high-risk genotype by year of publication. The box size is proportional to the meta-analysis study weight. Ref, reference citation.

GSTO1 and GSTO2.
Only one study evaluated GSTO1 A140D and GSTO2 N142D (41) (table 2). All 28 cases and 30 controls had DNA samples that were successfully genotyped. Both polymorphisms were reported and confirmed to be in Hardy-Weinberg equilibrium. Odds ratios were reported only for comparison of the high-risk (any "A" and any "G," respectively) with wild genotype. Although no excess risk was observed with the high-risk genotype for GSTO1 (OR = 0.96, 95 percent CI: 0.33, 2.80), a very large and significant excess was observed for the high-risk genotype for GSTO2 (OR = 23.83, 95 percent CI: 5.11, 123.8).

click to view

Table 2  Reported effects of GSTO1 and GSTO2 polymorphisms on risk of hepatocellular carcinoma

GSTA1 and GSTA4.
Only one study evaluated GSTA1 and GSTA4 (51). There were a total of 231 cases and 256 controls. Neither GSTA1 nor GSTA4 was associated with HCC in a by-polymorphism analysis (p < 0.09 and p < 0.56, respectively). A by-genotype comparison was reported only for GSTA4. Although there was no association between the GSTA4 genotype and HCC risk overall (OR = 0.84, 95 percent CI: 0.58, 1.20), there was significant increased risk of HCC when the analysis was restricted to males (OR = 1.55, 95 percent CI: 1.03, 2.35).

GSTM2 and GSTM3.
Finally, only one study evaluated GSTM2 and GSTM3 (51). There were a total of 231 cases and 256 controls. The study population was monomorphic for GSTM3, and therefore no test of association was performed. There was no significant association between GSTM2 polymorphisms and HCC (p < 0.12).

Interaction among GST polymorphisms.
Eight studies reported assessment of interaction between GSTM1 and GSTT1 (32, 34–38, 40, 44). This includes seven of 13 studies that evaluated both polymorphisms in the current meta-analysis, as well as in another study (44), which (although ineligible for the meta-analysis because it was an earlier report from an eligible study population) provided unique information on the GSTM1-GSTT1 interaction.

Excess risk of HCC with dual deletion of GSTM1 and GSTT1 was reported in four studies (ORs ranging from 1.46 (95 percent CI: 0.74, 2.87) (35) to 4.13 (95 percent CI: 1.70, 10.14) (38)). Further, there was evidence of significant interaction in three studies (32, 38, 40), with reported odds ratios with the dual null genotype from 1.3 to 2.0 times greater than those observed for either main effect. However, other studies reported only absence of interaction without providing specific measures of effect (34, 36, 37).

There was no evidence of interaction between GSTP1 and GSTM1 or GSTT1 (36).

Interaction among GST and other genes.
Eleven studies evaluated whether risk of HCC was modified by the presence of other genes (29–31, 34, 36, 42–44, 46, 49, 58). This includes several studies (43, 44, 46, 49, 58) that, although ineligible for the current meta-analysis, nonetheless provided unique data on possible interaction.

For GSTM1, there was no evidence of interaction with IL-IRN (31), NAT2 (43), CYP2E1 (36, 49), L-myc (46), ALDH2 (36), or ADH3 (30). Results with HYL1*2 were equivocal, with one study reporting significant interaction between GSTM1 null and HYL1*2 (42) and the other finding no evidence of interaction (29). Results with EPHX1 were contradictory: Two studies found evidence of significant interaction with GSTM1 null (29, 58), while another found significant interaction only with GSTM1 nonnull (34). The most suggestive finding of possible interaction was with XRCC1, a gene involved in DNA repair. Two of three studies found evidence of strong, significant, and dose-dependent interaction between GSTM1 null and XRCC1 (29, 42). However, a third study found a suggestive although much weaker and nonsignificant interaction only with GSTM1 nonnull (44).

Fewer studies evaluated potential interactions between GSTT1 and other genes. Evidence of significant interaction was found between GSTT1 null and XRCC1 (44) and between GSTT1 nonnull and EPHX1 (34). However, no evidence of interaction was reported for IL-IRN (31), NAT2 (43), ALDH2 (36), or CYP2E1 (36).

There was no reported interaction between GSTP1 and either CYP2E1 (36) or ALDH2 (36).

Interaction between GST and environmental exposure.
Evidence suggestive of possible interaction between GSTM1 null genotype and alcohol consumption was reported in two of three (30, 33, 36) studies, with elevated risk among heavy drinkers reported in one study (Synergism Index: 2.28) (30) and among drinkers with low plasma carotenoid levels reported in the other (33). Evidence of possible interaction between GST polymorphisms and smoking was also reported in two of three studies (33, 36, 43), with nonsignificant excess risk reported among light smokers with a GSTT1 null genotype in one study (43) and a significant excess risk among smokers with a GSTM1 or GSTT1 null genotype and low levels of plasma beta-carotene reported in the other (33). Overall, the most frequently evaluated environmental exposure was aflatoxin B1 (29, 34, 37, 40, 42), with evidence of significant interaction with GSTM1 null reported in four studies (29, 34, 40, 42) and with GSTT1 null reported in two studies (37, 40).

Laboratory testing
All studies except one (51) evaluating GSTM1 and GSTT1 used multiplex polymerase chain reaction to visually determine the presence versus the absence of the polymorphism. Studies evaluating GSTP1, GSTO1, GSTO2, GSTM3, GSTA1, and GSTA2 used polymerase chain reaction/restriction fragment length polymorphism to identify specific polymorphic variants.

Although most reports contained detailed information on the methods used for DNA isolation and on the use of specific primers and internal controls, few reported explicitly on information related to other aspects of quality control including the following: performance of duplicate testing (two studies only (34, 37)), use of a single laboratory to perform DNA testing (two studies only (29, 37)), blinding of laboratory personnel to disease status (two studies only (33, 37)), and timing of testing, including whether matched case and control DNA samples were processed at the same time (one study only (37)).

This report is the first published meta-analysis specifically examining the effect of GST polymorphisms on risk of HCC. We found that null variants of GSTT1 and, to a lesser extent, GSTTM1 might be associated with a small increase in HCC risk. Although considerably less examined, there was no evidence for an association between GSTP1 and HCC with insufficient evidence to make a statement for the other GST variants evaluated. There were also a number of potentially interesting gene-gene and gene-environment interactions reported in individual studies. However, these were too few and too inconsistent to allow a meta-analysis. All of the studies identified for possible inclusion in this meta-analysis were conducted in populations with high (Asian, African) or medium (European) incidence rates for HCC, with most conducted in Asian populations (11 of 15 studies) and in countries where hepatitis B is endemic (12 of 15 studies).

There is currently a remarkable absence of studies on GST and HCC from low-incidence regions including North America, where HCC incidence has more than doubled over the past two decades (59), and neither hepatitis B nor aflatoxin exposure is a major etiologic risk factor for HCC. Oxidative stress is proposed to be an important pathogenic factor in liver damage related to alcohol and hepatitis C, the major risk factors for HCC in these regions, either alone or in combination with other factors including obesity-related nonalcoholic fatty liver disease. Further, there is preliminary evidence indicating that GST polymorphisms are associated with increased risk of advanced liver disease that serves as a precursor for HCC in the setting of alcoholic liver disease. One study from the United Kingdom compared the frequency of GST polymorphisms in patients with alcoholic liver disease from heavy drinking and normal local controls. A significantly increased prevalence of the GSTT1 null genotype was observed in alcoholic liver disease patients in comparison with nondrinking controls (OR = 2.1, 95 percent CI: 1.1, 4.7) (60). Another study from Finland reported that homozygous deletion of the GSTM1 gene may indicate increased susceptibility to irreversible liver damage in response to the toxic effects of ethanol. In that study, the GSTM1 null genotype indicating absent enzymatic activity was assessed in 33 abstainers, 43 moderate alcohol consumers, and 313 heavy alcohol consumers; the study found that the null genotype was nearly significantly more frequent among heavy consumers with at least slight liver fibrosis (p = 0.05; OR = 1.8) and statistically significant more frequently among alcoholics with advanced liver fibrosis (p < 0.025; OR = 2.3) (61). Finally, nonalcoholic fatty liver disease has become the most common liver disorder in several countries in North America and Europe (62). A study from Italy analyzed the catalytic activities of GST in the blood of 21 children with nonalcoholic steatohepatitis and 28 controls. GST, which provides a second defense line against oxidative stress, was 17.8 percent increased in cases compared with controls (63). Similar arguments can be made for hepatitis C, where GSTs are involved in the metabolism of endogenously generated, cancer-causing reactive oxygen species continuously produced through hepatitis C-induced inflammatory disease. In summary, the evidence of a potential role of GST polymorphisms in liver disease related to common risk factors in low-risk populations in North America and Europe justifies further examination of these GST polymorphisms as risk factors for HCC in these currently unstudied and understudied regions.

Among the medium- and high-incidence populations where studies evaluating GST polymorphisms have been performed, we identified a total of 14 eligible studies evaluating GSTM1 (13 of which also evaluated GSTT1) (29–40, 42, 51). GSTM1 and GSTT1 protect cells from the natural by-products of lipid peroxidation and oxidative stress; deletion of the GSTM1 and GSTT1 genes is associated with enhanced endogenous mutagenic processes that are implicated in susceptibility to other inflammation-related cancers of the gastrointestinal tract, such as pancreatic cancer (64, 65). The overall results of our random-effects meta-analyses suggest a small excess risk of HCC with GSTT1 (OR = 1.19, 95 percent CI: 0.99, 1.44) and possibly GSTM1 (OR = 1.16, 95 percent CI: 0.89, 1.53) null genotypes. The pooled estimators generally trended toward a small excess risk with publication of the most recent studies.

We also performed meta-regression to evaluate the effect of several preselected factors on the observed variability among studies evaluating GSTM1 or GSTT1 null genotype. Univariate analysis suggested that the source of controls and possibly the year of publication and hepatitis B predominance may be important sources of between-study heterogeneity. However, these analyses are limited by multiple comparisons, the assumption that the association found at the study level applied at the individual level (66), and our inability to perform multivariate analyses given the small number of studies.

GSTP1 was evaluated in only three studies (36, 38, 51). All reported no association with HCC, although only two provided by-genotype data that could be combined in a pooled estimator (OR = 0.75, 95 percent CI: 0.50, 1.15). These results suggest that GSTP1, combined with the fact that its greatest expression is in the brain and not the liver, is unlikely to be a strong independent risk factor for HCC.

Only one study evaluated each of GSTO1 (41), GSTO2 (41), GSTA1 (51), GSTA4 (51), GSTM2 (51), and GSTM3 (51). Significantly increased risk of HCC was observed with GSTO2 (41) and with GSTA4 in men only (51). However, these results should be considered as provisional, because results of the earliest genetic association study tend disproportionately to report the strongest or most significant findings and are frequently not replicated by subsequent research (67, 68).

The current meta-analysis has several strengths, most notably a rigorous design that included structured searches in multiple databases for published reports, unpublished abstracts, and dissertations along with the use of ancestry or bibliography searches to identify nondatabase-indexed reports. These efforts, as well as our inclusion of non-English language reports, were important in minimizing a major potential threat to the validity of any meta-analysis—publication bias and the related threat of a language bias. Another important strength is our performance of auxiliary analyses, including meta-regression and cumulative meta-analyses, that allowed a more thorough examination and appropriate qualification of our results.

Our meta-analysis is, however, subject to the same potential limitations that affect all meta-analyses, including the direct comparability and quantity and quality of available reports. The issue of comparability is of lesser concern here, given the nature of our exposure (inherited genotype) and outcome (medically confirmed HCC) and the uniform use of a case-control design.

Limited sample size is of concern in the current meta-analysis. Only 14 eligible studies evaluated GSTM1 null, the most frequently studied GST variant, with a total sample size of 966 in the largest study and with five studies having a total sample size of less than 250. It is well accepted that the strength of an association is not an inherent biologic property with small associations potentially reflecting important causal relations (69). However, the following three points have only recently become understood: 1) It may take relatively few common genetic variants, each conveying only small to modest excess risk, to account for a sizable portion of the population attributable fraction for common diseases (e.g., 10–18 genes, each with a 20–30 percent prevalence and conveying an odds ratio of only 1.2–1.5 to explain between 30 and 50 percent of the population attributable fraction) (70); 2) there is meta-analytical evidence suggesting that some common genetic variants do convey small to modest excess risk for several common diseases (e.g., DRD3 and schizophrenia and PPARG and type II diabetes) (71); and 3) the minimum sample size needed to reliably detect the likely small individual effects of common genetic variants contributing to common chronic diseases is much larger than previously thought (i.e., often multiple thousands) (72). Given our limited total sample size and small observed effects, performance of additional larger genetic association studies with subsequent update of the current meta-analysis will be needed to confirm our preliminary suggestive findings for the common deletion polymorphisms for GSTT1 and GSTM1.

It is difficult to assess how variable individual study quality may have influenced the overall findings of our review. Although the reporting was generally good among individual studies, for more technical aspects of study design and conduct, the reporting was generally more limited and variable regarding the use of quality control measures such as blinding of laboratory personnel to disease status or evaluations of possible biases, particularly regarding the selection and participation of controls. Although some meta-analysts have provided quality scores (73), these scores have also been criticized because of their potential to introduce bias of unknown dimensions (74). Because we could not determine reliable or valid methods to distinguish between "absence of explicit reporting" and "absence of actual performance" of these quality control procedures, we elected not to assign quality scores and instead to provide the reader with collective data to evaluate.

In conclusion, taken together, our findings suggest a possible small excess risk of HCC with the GSTT1 null genotype and possibly also with the GSTM1 null genotype in populations with high or medium incidence of HCC. Larger studies will be needed to explore potential interactions between GST polymorphisms and other genetic and environmental risk factors for HCC in these high- and medium-risk populations. We suggest that future studies might also be conducted in previously unevaluated populations, including low-risk populations from North America and Europe, in which HCC incidence has more than doubled over the last two decades (59) and where important risk factors, such as alcohol drinking and obesity-related nonalcoholic fatty liver disease, acting either alone or in combination with hepatitis C, are major sources of oxidative stress.

• Human Genome Epidemiology Network prepared by the National Office of Public Health Genomics, Centers for Disease Control and Prevention (http://www.cdc.gov/genomics/hugenet/default.htm)
  
• Online Mendelian Inheritance in Man by The Johns Hopkins University (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=OMIM)
  
• Information Hyperlinked over Proteins (http://www.ihop-net.org/UniPub/iHOP/ ) (75)
  
• American Society of Human Genetics (http://www.ashg.org/genetics/ashg/menu-about.shtml )
  
• International Genetic Epidemiology Society (http://iges.biostat.wustl.edu/ )

We thank George Kitsios for comments on the manuscript.

Dr. El-Serag is a Veterans Affairs Health Services Research and Development Service Career Development awardee (RCD 00-013-2).

Conflict of interest: none declared.

1. Parkin DM. Global cancer statistics in the year 2000. Lancet Oncol (2001) 2:533–43.
2. El-Serag HB. Hepatocellular carcinoma: recent trends in the United States. Gastroenterology (2004) 127(suppl 1):S27–34.
3. Hayes JD, Flanagan JU, Jowsey IR. Glutathione transferases. Annu Rev Pharmacol Toxicol (2005) 45:51–88.
4. Landi S. Mammalian class theta GST and differential susceptibility to carcinogens: a review. Mutat Res (2000) 463:247–83.
5. Salama SA, Abdel-Rahman SZ, Sierra-Torres CH, et al. Role of polymorphic GSTM1 and GSTT1 genotypes on NNK-induced genotoxicity. Pharmacogenetics (1999) 9:735–43.
6. Strange RC, Spiteri MA, Ramachandran S, et al. Glutathione-S-transferase family of enzymes. Mutat Res (2001) 482:21–6.
7. Okcu MF, Selvan M, Wang LE, et al. Glutathione S-transferase polymorphisms and survival in primary malignant glioma. Clin Cancer Res (2004) 10:2618–25.
8. Daniel V. Glutathione S-transferases: gene structure and regulation of expression. Crit Rev Biochem Mol Biol (1993) 28:173–207.
9. Rebbeck TR. Molecular epidemiology of the human glutathione S-transferase genotypes GSTM1 and GSTT1 in cancer susceptibility. Cancer Epidemiol Biomarkers Prev (1997) 6:733–43.
10. Yang G, Shu XO, Ruan ZX, et al. Genetic polymorphisms in glutathione-S-transferase genes (GSTM1, GSTT1, GSTP1) and survival after chemotherapy for invasive breast carcinoma. Cancer (2005) 103:52–8.
11. Stoehlmacher J, Park DJ, Zhang W, et al. Association between glutathione S-transferase P1, T1, and M1 genetic polymorphism and survival of patients with metastatic colorectal cancer. J Natl Cancer Inst (2002) 94:936–42.
12. Hohaus S, Di RA, Di FA, et al. Glutathione S-transferase P1 genotype and prognosis in Hodgkin's lymphoma. Clin Cancer Res (2005) 11:2175–9.
13. Ambrosone CB, Sweeney C, Coles BF, et al. Polymorphisms in glutathione S-transferases (GSTM1 and GSTT1) and survival after treatment for breast cancer. Cancer Res (2001) 61:7130–5.
14. Hayes JD, Strange RC. Glutathione S-transferase polymorphisms and their biological consequences. Pharmacology (2000) 61:154–66.
15. Norppa H. Cytogenetic biomarkers and genetic polymorphisms. Toxicol Lett (2004) 149:309–34.
16. Garte S, Gaspari L, Alexandrie AK, et al. Metabolic gene polymorphism frequencies in control populations. Cancer Epidemiol Biomarkers Prev (2001) 10:1239–48.
17. Hagen TM, Huang S, Curnutte J, et al. Extensive oxidative DNA damage in hepatocytes of transgenic mice with chronic active hepatitis destined to develop hepatocellular carcinoma. Proc Natl Acad Sci U S A (1994) 91:12808–12.
18. Harris MJ, Coggan M, Langton L, et al. Polymorphism of the Pi class glutathione S-transferase in normal populations and cancer patients. Pharmacogenetics (1998) 8:27–31.
19. Ali-Osman F, Akande O, Antoun G, et al. Molecular cloning, characterization, and expression in Escherichia coli of full-length cDNAs of three human glutathione S-transferase Pi gene variants. Evidence for differential catalytic activity of the encoded proteins. J Biol Chem (1997) 272:10004–12.
20. Watson MA, Stewart RK, Smith GB, et al. Human glutathione S-transferase P1 polymorphisms: relationship to lung tissue enzyme activity and population frequency distribution. Carcinogenesis (1998) 19:275–80.
21. Packer BR, Yeager M, Burdett L, et al. SNP500Cancer: a public resource for sequence validation, assay development, and frequency analysis for genetic variation in candidate genes. Nucleic Acids Res 2006;34(database issue):D617–21. (http://snp500cancer.nci.nih.gov/home.cfm). (Accessed December 1, 2006).
22. Khoury MJ, Little J. Human genome epidemiologic reviews: the beginning of something HuGE. Am J Epidemiol (2000) 151:2–3.
23. Little J, Khoury MJ. Mendelian randomisation: a new spin or real progress? Lancet (2003) 362:930–1.
24. Stroup DF, Berlin JA, Morton SC, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-Analysis of Observational Studies in Epidemiology (MOOSE) group. JAMA (2000) 283:2008–12.
25. Petitti DB. Meta-analysis, decision analysis, and cost-effectiveness analysis: methods for quantitative synthesis in medicine. 2nd ed. (2000) New York, NY: Oxford University Press.
26. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med (2002) 21:1539–58.
27. Dersimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials (1986) 7:177–88.
28. Egger M, Davey SG, Schneider M, et al. Bias in meta-analysis detected by a simple, graphical test. BMJ (1997) 315:629–34.
29. Kirk GD, Turner PC, Gong Y, et al. Hepatocellular carcinoma and polymorphisms in carcinogen-metabolizing and DNA repair enzymes in a population with aflatoxin exposure and hepatitis B virus endemicity. Cancer Epidemiol Biomarkers Prev (2005) 14:373–9.
30. Covolo L, Gelatti U, Talamini R, et al. Alcohol dehydrogenase 3, glutathione S-transferase M1 and T1 polymorphisms, alcohol consumption and hepatocellular carcinoma (Italy). Cancer Causes Control (2005) 16:831–8.
31. Chen CC, Yang SY, Liu CJ, et al. Association of cytokine and DNA repair gene polymorphisms with hepatitis B-related hepatocellular carcinoma. Int J Epidemiol (2005) 34:1310–18.
32. Deng ZL, Wei YP, Ma Y. Polymorphism of glutathione S-transferase mu 1 and theta 1 genes and hepatocellular carcinoma in southern Guangxi, China. World J Gastroenterol (2005) 11:272–4.
33. Yu MW, Chiu YH, Chiang YC, et al. Plasma carotenoids, glutathione S-transferase M1 and T1 genetic polymorphisms, and risk of hepatocellular carcinoma: independent and interactive effects. Am J Epidemiol (1999) 149:621–9.
34. Tiemersma EW, Omer RE, Bunschoten A, et al. Role of genetic polymorphism of glutathione-S-transferase T1 and microsomal epoxide hydrolase in aflatoxin-associated hepatocellular carcinoma. Cancer Epidemiol Biomarkers Prev (2001) 10:785–91.
35. Ladero JM, Martinez C, Garcia-Martin E, et al. Glutathione S-transferase M1 and T1 genetic polymorphisms are not related to the risk of hepatocellular carcinoma: a study in the Spanish population. Eur J Cancer (2006) 42:73–7.
36. Munaka M, Kohshi K, Kawamoto T, et al. Genetic polymorphisms of tobacco- and alcohol-related metabolizing enzymes and the risk of hepatocellular carcinoma. J Cancer Res Clin Oncol (2003) 129:355–60.
37. Sun CA, Wang LY, Chen CJ, et al. Genetic polymorphisms of glutathione S-transferases M1 and T1 associated with susceptibility to aflatoxin-related hepatocarcinogenesis among chronic hepatitis B carriers: a nested case-control study in Taiwan. Carcinogenesis (2001) 22:1289–94.
38. Liu CZ, Bian JC, Jiang F, et al. Genetic polymorphisms of glutathione S-transferases M1, T1, P1 on susceptibility hepatocellular carcinoma. (In Chinese). China Public Health (2002) 18:935–6.
39. Zhu MH, Chen XH, Zhou LF. Association of genetic polymorphisms in glutathione S-transferases M1 with hepatitis beta-related hepatocellular carcinoma. (In Chinese). Zhejiang Da Xue Xue Bao Yi Xue Ban (2005) 34:126–30.
40. Long XD, Ma Y, Wei YP, et al. Study on the detoxication gene gstM1-gstT1-null and susceptibility to aflatoxin B1 related hepatocellular carcinoma in Guangxi. (In Chinese). Zhonghua Liu Xing Bing Xue Za Zhi (2005) 26:777–81.
41. Marahatta SB, Punyarit P, Bhudisawasdi V, et al. Polymorphism of glutathione S-transferase omega gene and risk of cancer. Cancer Lett (2006) 236:276–81.
42. Long XD, Ma Y, Wei YP, et al. The polymorphisms of GSTM1, GSTT1, HYL1*2, and XRCC1, and aflatoxin B1-related hepatocellular carcinoma in Guangxi population, China. Hepatol Res (2006) 36:48–55.
43. Gelatti U, Covolo L, Talamini R, et al. N-Acetyltransferase-2, glutathione S-transferase M1 and T1 genetic polymorphisms, cigarette smoking and hepatocellular carcinoma: a case-control study. Int J Cancer (2005) 115:301–6.
44. Yu MW, Yang SY, Pan IJ, et al. Polymorphisms in XRCC1 and glutathione S-transferase genes and hepatitis B-related hepatocellular carcinoma. J Natl Cancer Inst (2003) 95:1485–8.
45. Chen CJ, Yu MW, Liaw YF, et al. Chronic hepatitis B carriers with null genotypes of glutathione S-transferase M1 and T1 polymorphisms who are exposed to aflatoxin are at increased risk of hepatocellular carcinoma. Am J Hum Genet (1996) 59:128–34.
46. Hsieh LL, Huang RC, Yu MW, et al. L-myc, GST M1 genetic polymorphism and hepatocellular carcinoma risk among chronic hepatitis B carriers. Cancer Lett (1996) 103:171–6.
47. Ma Y, Deng Z, Wei YP. Study of the deletion mutation of glutathione S transferase M1 gene and its role in susceptibility to hepatocellular carcinoma. Chin J Cancer Res (2001) 13:176–8.
48. Deng Z, Wei Y, Ma Y. Glutathione-S-transferase M1 genotype in patients with hepatocellular carcinoma. (In Chinese). Zhonghua Zhong Liu Za Zhi (2001) 23:477–9.
49. Yu MW, Gladek-Yarborough A, Chiamprasert S, et al. Cytochrome P450 2E1 and glutathione S-transferase M1 polymorphisms and susceptibility to hepatocellular carcinoma. Gastroenterology (1995) 109:1266–73.
50. Omer RE, Verhoef L, Van't Veer P, et al. Peanut butter intake, GSTM1 genotype and hepatocellular carcinoma: a case-control study in Sudan. Cancer Causes Control (2001) 12:23–32.
51. McGlynn KA, Hunter K, LeVoyer T, et al. Susceptibility to aflatoxin B1-related primary hepatocellular carcinoma in mice and humans. Cancer Res (2003) 63:4594–601.
52. London WT, Evans AA, Buetow K, et al. Molecular and genetic epidemiology of hepatocellular carcinoma: studies in China and Senegal. Princess Takamatsu Symp (1995) 25:51–60.
53. Bian JC, Shen FM, Shen L, et al. Susceptibility to hepatocellular carcinoma associated with null genotypes of GSTM1 and GSTT1. World J Gastroenterol (2000) 6:228–30.
54. Bian JC, Shen F, Shen L, et al. Susceptibility to hepatocellular carcinoma is associated with the null genotypes of GSTM1 and GSTT1. (In Chinese). Chin J Dig (1999) 19:87–90.
55. Long XD, Ma Y, Wei YP, et al. A study about the association of detoxication gene GSTM1 polymorphism and the susceptibility to aflatoxin B1-related hepatocellular carcinoma. (In Chinese). Zhonghua Gan Zang Bing Za Zhi (2005) 13:668–70.
56. Zhou T, Evans AA, London WT, et al. Glutathione S-transferase expression in hepatitis B virus-associated human hepatocellular carcinogenesis. Cancer Res (1997) 57:2749–53.
57. Chen SY, Wang LY, Lunn RM, et al. Polycyclic aromatic hydrocarbon-DNA adducts in liver tissues of hepatocellular carcinoma patients and controls. Int J Cancer (2002) 99:14–21.
58. McGlynn KA, Rosvold EA, Lustbader ED, et al. Susceptibility to hepatocellular carcinoma is associated with genetic variation in the enzymatic detoxification of aflatoxin B1. Proc Natl Acad Sci U S A (1995) 92:2384–7.
59. El-Serag HB, Mason AC. Rising incidence of hepatocellular carcinoma in the United States. N Engl J Med (1999) 340:745–50.
60. Brind AM, Hurlstone A, Edrisinghe D, et al. The role of polymorphisms of glutathione S-transferases GSTM1, M3, P1, T1 and A1 in susceptibility to alcoholic liver disease. Alcohol Alcohol (2004) 39:478–83.
61. Savolainen VT, Pjarinen J, Perola M, et al. Glutathione-S-transferase GST M1 "null" genotype and the risk of alcoholic liver disease. Alcohol Clin Exp Res (1996) 20:1340–5.
62. Angulo P. GI epidemiology: nonalcoholic fatty liver disease. Aliment Pharmacol Ther (2007) 25:883–9.
63. Nobili V, Pastore A, Gaeta LM, et al. Glutathione metabolism and antioxidant enzymes in patients affected by nonalcoholic steatohepatitis. Clin Chim Acta (2005) 355:105–11.
64. Wiencke JK, Pemble S, Ketterer B, et al. Gene deletion of glutathione S-transferase theta: correlation with induced genetic damage and potential role in endogenous mutagenesis. Cancer Epidemiol Biomarkers Prev (1995) 4:253–9.
65. Duell EJ, Holly EA, Bracci PM, et al. A population-based, case-control study of polymorphisms in carcinogen-metabolizing genes, smoking, and pancreatic adenocarcinoma risk. J Natl Cancer Inst (2002) 94:297–306.
66. Thompson SG, Higgins JP. How should meta-regression analyses be undertaken and interpreted? Stat Med (2002) 21:1559–73.
67. Trikalinos TA, Ntzani EE, Contopoulos-Ioannidis DG, et al. Establishment of genetic associations for complex diseases is independent of early study findings. Eur J Hum Genet (2004) 12:762–9.
68. Ioannidis JP, Trikalinos TA. Early extreme contradictory estimates may appear in published research: the Proteus phenomenon in molecular genetics research and randomized trials. J Clin Epidemiol (2005) 58:543–9.
69. Rothman K, Greenland S. Causation and causal inference. In: Modern epidemiology—Rothman K, Greenland S, eds. (1998) Philadelphia, PA: Lippincott-Raven. 7–28.
70. Yang Q, Khoury MJ, Friedman J, et al. How many genes underlie the occurrence of common complex diseases in the population? Int J Epidemiol (2005) 34:1129–37.
71. Lohmueller KE, Pearce CL, Pike M, et al. Meta-analysis of genetic association studies supports a contribution of common variants to susceptibility to common disease. Nat Genet (2003) 33:177–82.
72. Hattersley AT, McCarthy MI. What makes a good genetic association study? Lancet (2005) 366:1315–23.
73. Thakkinstian A, Bowe S, McEvoy M. Association between apolipoprotein E polymorphisms and age-related macular degeneration: a HuGE review and meta-analysis. Am J Epidemiol (2006) 164:813–22.
74. Greenland S, O'Rourke K. On the bias produced by quality scores in meta-analysis, and a hierarchical view of proposed solutions. Biostatistics (2001) 2:463–71.
75. Hoffmann R, Valencia A. A gene network for navigating the literature. (Letter). Nat Genet (2004) 36:664.