Glutathione S-transferases M1, T1, and P1 and Breast Cancer: A Pooled Analysis

Inherited differences in the capacity of xenobiotic metabolizing enzymes might be an important factor of genetic susceptibility to cancer. Glutathione S-transferases (GST) are phase II enzymes involved in the detoxification of a broad range of toxic and potentially carcinogenic compounds (1). In humans, five classes of GST enzymes have been identified (GST classes α, μ, π, σ, and 𝛉). Each class is encoded by a separate gene or gene family. Allelic variants for each of these genes may result in less effective or absent enzymatic detoxification and thus increase susceptibility to cancer, although the exact biochemical processes are not yet fully understood.

The GSTM1 gene, coding for cytosolic GST class μ enzyme, is located on chromosome 1p13.3 (2) and includes a deletion polymorphism that, in the homozygous state (GSTM1 null), results in the total absence of a functional gene product (3). Several studies have shown high agreement between the GSTM1 null genotype and a lack of GST class μ function. GSTM1 is expressed in various tissues, mainly liver, stomach, and brain. The frequency of the GSTM1 null genotype varies across ethnic groups and was reported to be ∼50% in Caucasians (4-6). The GSTT1 gene (chromosome 22q11.2; ref. 7) also has an inactivating homozygous deletion polymorphism (8). Homozygosity for the deletion is present in ∼11% to 18% of Caucasians (9). In humans, the GSTT1 enzyme is primarily expressed in liver and erythrocytes. In GSTP1 (chromosome 11q13; ref. 10), an amino acid transition has been reported at codon 105 (A313G→Ile105Val), leading to expression of an active but functionally different protein (11, 12). The GSTP1 encoded enzyme GST class π is mainly found in spleen, heart, and lung tissue. Both GST classes π and μ enzymes are also expressed in breast cancer tissue (13).

Several environmental risk factors have been previously associated with increased susceptibility to breast cancer. Hormonal factors that play an important role in cell growth and several aspects of reproductive history, characterized by elevated and prolonged estrogen levels, are associated with breast cancer risk. Nulliparity, lack of or reduced breast-feeding, older age at first birth, early age at menarche, and late age at menopause increase breast cancer risk (14). High body mass index (BMI) and possibly low physical exercise are also risk factors of postmenopausal breast cancer, acting via interference with hormonal levels (15). Cigarette smoking is an established risk factor for several cancers including the respiratory and upper aerodigestive tract, kidney, bladder, stomach, and pancreas (16).

Enzymes belonging to the GST classes μ, π, and 𝛉 are involved in the detoxification of benzo(a)pyrene and other polycyclic aromatic hydrocarbons found in tobacco smoke and have received considerable attention in relation to smoking-related cancers. An association between the GST null phenotype and cancer susceptibility was described for lung cancer (17, 18), colorectal cancer, and bladder cancer (19, 20). Polycyclic aromatic hydrocarbons induced mammary tumors in animal models, and polycyclic aromatic hydrocarbon-DNA adducts have been identified in human mammary epithelial cells. A recent case-control study suggested an increased risk of breast cancer in relation to polycyclic aromatic hydrocarbon-DNA adducts (21). Although tobacco smoking does not seem to be a risk factor of breast cancer in studies of unselected populations, the possibility of an increased risk in genetically predisposed groups remains (22). GSTs are also involved in detoxifying reactive compounds generated during estrogen metabolism.

Few studies have addressed GST polymorphisms and breast cancer, although a meta-analysis of studies published before 1997 has suggested a slight risk increase in carriers of the GSTM1 null polymorphism, which was only significant in the youngest age group (23).

The Genetic Susceptibility to Environmental Carcinogen (GSEC) Study, an international collaborative project, has been initiated to investigate the relationships of polymorphisms in genes that metabolize environmental carcinogens and cancers at different sites (24). The established database offered the opportunity for investigating the association of polymorphisms in the GSTM1, GSTT1, and GSTP1 and breast cancer using individual data of several studies and taking into account the potential modifying effect of reproductive factors and tobacco consumption.

The present analysis was based on the original data of seven case-control studies on GST genotypes and breast cancer, which were provided by the investigators to the GSEC Study (24, 25). Investigators who had published case-control studies on metabolic gene polymorphisms and cancer up to June 1999 were identified through Medline and invited to provide data on all subjects (including published as well as unpublished data). The results of six of the seven studies including breast cancer had been published previously (26-32). Two groups have published their results after they had submitted data to the GSEC. The number of subjects contained in the published reports may differ slightly from the number included in this pooled analysis. The present study was restricted to women of Caucasian origin. For every subject, case-control status and genotype at the GSTM1, GSTT1, and GSTP1 loci were submitted for the pooled analysis. All studies used PCR to detect polymorphisms. Although in some studies different PCR protocols were used, the laboratory procedures in general should be comparable. If available, additional information on age, anthropometric measures (body weight and height), family history of cancer, smoking status, and reproductive factors was provided. Prior to analysis, the data were checked and coded in a standard fashion and entered into a common database.

Genotypes at the GSTM1 and GSTT1 loci were coded positive if at least one functional allele was present and null in the case of a homozygous deletion. The number of homozygote and heterozygote carriers of GSTM1 and GSTT1 wild-type alleles was not provided. GSTP1 genotype was classified in three categories: the homozygous wild-type, the heterozygous, and the homozygous mutant. BMI was calculated according to the standard formula: weight (kg)/height (m)². Information on cumulative tobacco smoking was expressed as pack-years (daily amount multiplied by duration in years and divided by 20). Available information on reproductive factors included parity (nulliparous versus parous, i.e., at least one birth), menopausal status (premenopausal versus postmenopausal), age at menopause, and age at menarche. Family history of breast cancer was considered positive if a woman had at least one first-degree relative with breast cancer.

Crude odds ratios (OR) and 95% confidence intervals (95% CI) for the risk of breast cancer associated with GSTM1 null and GSTT1 null were calculated for each study by means of meta-analysis. Fixed and random effects models were fitted. The Q test was done to assess heterogeneity between individual studies (33). Meta-analysis was based on the data submitted to the GSEC Study. Publication bias was assessed by a funnel plot and Begg's and Egger's test (34, 35).

Differences in the proportions of exposed and unexposed individuals among cases and controls were assessed using the χ² test. Differences in the distribution of continuous variables (age and BMI) were investigated with the Wilcoxon rank sum test due to the non-normal distributions of these variables. Genotype frequencies of GSTP1 were tested for Hardy-Weinberg equilibrium by comparing observed and expected frequencies using a χ² test. All significance tests were two sided at the 0.05 level.

To measure the effect of genotypes and potential risk factors (age, family history, BMI, smoking, and reproductive factors) on the risk of breast cancer, ORs and 95% CIs were estimated using unconditional logistic regression. All ORs were adjusted for study center because of possible population stratification and possible study-specific misclassification of genotype (e.g., due to differences in the laboratory protocols).

To evaluate a possible modifying effect of tobacco consumption and reproductive factors on the association between GSTs and breast cancer, separate analyses were conducted after stratifying by categories of ever versus never smokers, parous versus nulliparous women, and premenopausal versus postmenopausal women. The continuous variables pack-years in ever smokers, age at menarche, and age at menopause were each categorized into two groups based on the division at the median among controls. To assess a modifying effect of GST genotype on the effect of smoking, ORs for the risk induced by smoking were estimated stratifying by genotype. To address an effect on age at onset, the distributions of age at diagnosis among cases were compared by genotype. ORs were calculated stratifying at tertiles of age in controls (<51, 51–61, and ≥62 years).

Interaction between GSTs and variables used for stratification was formally assessed by adding a product term to a model containing the main effects of GST genotype and the categories of the stratification variable. Models with and without interaction term were compared by using the likelihood ratio test. Analyses were repeated restricting to postmenopausal women. This set of analyses does not include GSTP1 because only one study group provided information on menopausal status and GSTP1 (31).

Finally, genotypes were combined to assess a potential synergism of polymorphisms of different GST classes. ORs were computed for the effect of each possible combination of wild-type and null among GSTM1, GSTT1, and GSTP1 genotype and the risk of breast cancer. Analyses were carried out using Stata statistical software (Stata, College Station, TX; ref. 36).

The baseline characteristics of the seven case-control studies on GST and breast cancer available in the GSEC database are shown in Table 1. In total, 2,048 cases with invasive breast cancer and 1,969 controls were included in the pooled analysis. All studies investigated the association between GSTM1 genotype and breast cancer; four studies provided genotypes of GSTT1, comprising 50.9% of study subjects; and two studies provided genotypes of GSTP1, comprising 37.5% of study subjects (Table 1). Overall, the GSTM1 null genotype was found in 52.0% of controls, with similar proportion in each individual study (P for difference in proportions = 0.25). The GSTT1 null genotype was found in 21.6% of controls, ranging from 17.4 in the Norwegian study (30) to 25.8% in a study in the United States (ref. 27; P = 0.12). A homozygous mutant genotype of GSTP1 was present in 9.4% of controls (Table 1). No departure from Hardy-Weinberg equilibrium was found for GSTP1 genotypes in each study and both among cases and controls.

The meta-analytic OR (95% CI) of the effect of GSTM1 on breast cancer risk in the random effects model was 0.94 (0.77–1.15). The individual ORs in the studies ranged from 0.59 to 1.28 with considerably overlapping 95% CIs (Q test: P = 0.056). 95% CIs were not significant, except for one small study (unpublished) with OR (95% CI) of 0.59 (0.36–0.96). There was an indication of publication bias in the funnel plot (Fig. 1; Egger's test, P = 0.011; Begg's test, P = 0.035). Meta-analysis of the four studies providing GSTT1 genotypes yielded an OR (95% CI) of 1.11 (0.87–1.42). None of the single-study ORs were significant.

The availability of information on variables investigated in the pooled analysis is shown in Table 1. Cases were, on average, 2 years older than controls (median age 59 versus 57 years; P <0.001). Eighty-five (17%) cases and 40 (7%) controls had a positive family history of breast cancer (P <0.001). No significant difference between cases and controls was found for BMI (median 25 kg/m² in each cases and controls; P = 0.24), proportion of smokers (43% versus 39%; P = 0.10), number of pack-years (median 22.5 versus 27; P = 0.50), parity (82% versus 81%; P = 0.73), age at menarche (13 years in each group; P = 0.95), and menopausal status (74% versus 77% postmenopausal women; P = 0.28). The age at menopause seemed to be slightly higher in cases than in controls but the difference was not significant (median 49 versus 47 years; P = 0.07).

A positive family history of breast cancer conferred a >2-fold increased risk of breast cancer (OR, 2.45; 95% CI, 1.64–3.66). Neither BMI (OR, 0.88; 95% CI, 0.68–1.13), ever smoking (OR, 1.07; 95% CI, 0.88–1.30), and postmenopausal status (OR, 0.78; 95% CI, 0.56–1.06) nor parity (OR, 1.09; 95% CI, 0.68–1.76) was associated with breast cancer risk. ORs were recalculated controlling for age (categorized at tertiles in controls) but differed little from the previous estimates (results not shown).

The analysis of the individual patient data from seven case-control studies yielded no evidence for an increased risk of breast cancer associated with the null genotype at the GSTM1 locus (OR, 0.98; 95% CI, 0.86–1.12, adjusted for study center). This estimate is similar to the summary OR obtained in the meta-analysis (OR, 0.94; 95% CI, 0.77–1.15). No association was found between the GSTT1 null genotype and breast cancer (OR, 1.11; 95% CI, 0.87–1.41) and there was no risk increase for either the heterozygous (OR, 1.01; 95% CI, 0.79–1.28) or the homozygous (OR, 0.93; 95% CI, 0.62–1.38) mutant genotype at the GSTP1 locus (Table 2).

In stratified analyses, no risk modification was found with regard to family history of breast cancer. For GSTM1 null, the OR (95% CI) was 0.68 (0.30–1.53) among those with a positive history and 0.89 (0.69–1.16) among those with a negative family history. The respective ORs (95% CIs) for GSTT1 null were 1.70 (0.38–13.47) and 0.90 (0.57–1.42).

Smoking status, parity, age at menarche, and menopause also did not show any effect modification. There was no evidence of interaction between GST genotypes and these stratifying variables. Detailed results are given in Tables 3 and 4. In the group of individuals who had smoked <27 pack-years seemed to be a significant 2.7-fold increased cancer risk associated with GSTT1 null (95% CI, 1.13–6.37).

Higher age at menopause (stratified at 47 years) had no modifying effect on risk. In a separate analysis addressing the risk conferred by smoking and stratifying by GST genotypes, there was no evidence of a modifying effect (results not shown). All analyses were repeated restricting the sample to postmenopausal women but did not reveal any significant risk increase or modifying effect (details not shown).

The median age at diagnosis of breast cancer was 58 years for carriers of GSTM1 null compared with 60 years in cases with wild-type GSTM1 (P = 0.09). There was no significant difference in the distribution of age at diagnosis according to genotype after stratification by family history. ORs (95% CIs) for the risk of breast cancer stratified by age group were 1.01 (0.77–1.32) for the youngest group, 0.92 (0.65–1.30) for the middle group, and 1.04 (0.81–1.35) for the highest age group.

The effect of combined GST genotypes on the risk of breast cancer was assessed in a separate set of analyses permuting genotypes in GSTM1, GSTT1, and GSTP1. Results are shown in Table 5. The combination of GSTT1 null and GSTP1 mutant yielded a nonsignificant 2-fold increased risk. Compared with wild-type for both, the addition of GSTM1 null suggested no risk modification.

This analysis of pooled individual data revealed no increased risk of breast cancer associated with GST null genotypes at the GSTM1 and GSTT1 loci and the codon 105 polymorphism at the GSTP1 locus. This is consistent with the results of the published studies that contributed data to the GSEC database and in which no risk increase was found (26, 27, 30). Other studies, published recently, have also failed to detect an association between GSTs and breast cancer (37, 38). An earlier meta-analysis of three studies revealed a significant risk increase for GSTM1 null only in the youngest age group (23). We did not observe any risk difference by age group in the pooled analysis. Among cases, the age distributions of age at onset were not significantly different by GSTM1 genotype. Our results are based on the analysis of case-control data from Caucasian women. A large case-control study in Asian women from Shanghai found no association between GSTM1 or GSTT1 genotypes and breast cancer but observed a significant risk increase among homozygotes for the GSTP1 polymorphism (39), the latter not being confirmed in our analysis. These authors also did a meta-analysis of 19 published studies (predominantly Caucasian women, according to the authors) and found no association with GSTM1, GSTT1, or GSTP1 (39).

We found no evidence of interaction between GST polymorphisms and smoking. There was a suggestion of an increased risk for GSTT1 null carriers who smoked <27 pack-years. This association was not observed when controlling for age. Published studies have yielded no (40) or only weak evidence of a modifying effect of smoking in postmenopausal women (22). Our positive result might also be due to chance in a small subgroup.

Menopausal status and age at menopause did not modify the relation between GSTs and breast cancer risk. Previous studies yielded inconsistent results concerning the potential modifying effect of menopausal status. Two studies have reported a positive association between the homozygous deletion of GSTM1 and breast cancer risk among postmenopausal women (41) or women ages >50 years (42), whereas another study detected no association neither in postmenopausal nor in premenopausal women (40). Similarly, one of those studies reported a protective effect of GSTT1 null genotype in premenopausal women (40), but another study failed to show this (41). Such inconsistent results might be due to limited power in small studies and a tendency to overemphasize positive subgroup findings. Parity and age at menarche did not influence or modify breast cancer risk. The number of children, which probably is a stronger factor for breast cancer than the binary variable parity, could not be examined because this information was available of too few subjects. Family history was a strong risk factor for breast cancer in univariate analysis. Stratification by family history of breast cancer did not modify the OR estimates associated with GST genotypes.

Hypothetically, a combination of deficient GSTs could affect the outcome. Helzlsouer et al. (41), in a study on 110 cases and 113 control women, reported a gene dosage effect of GSTP1, wherein increasing number of polymorphic alleles increased breast cancer risk. The Shanghai study cited above described a risk increase for the GSTP1 polymorphism in the homozygous state, but this was not significant in postmenopausal women (39). Mitrunen et al. (38) investigated 481 cases and 483 controls and described an inverse relation between number of polymorphic alleles and risk. The present analysis, combining data of two studies, did not show any dosage effect of polymorphic GSTP1 alleles. There was a suggestion that the combination of the GSTT1 null genotype with a GSTP1 mutant might increase breast cancer risk, although the 95% CI was wide. Mitrunen et al. also observed a tendency of increased risk with GSTT1 null plus homozygous mutant GSTP1 given a null genotype at the GSTM1 locus (38).

ORs were adjusted to account for potential confounding by age and study site, but estimates differed only little from the unadjusted values. Slight differences in frequencies of the null genotype of GSTT1 were observed between studies. In the literature, the reported frequency of the GSTT1 null genotype ranged from 11% to 18%. The average frequency among controls across the pooled studies was 21.6% with maximum of 25.8% in one study. The outlying frequencies could be the result of misclassification of genotypes or inherent in the different populations investigated (population stratification).

In theory, publication bias could affect the results of the pooled analysis. Publication bias can occur if studies with significant associations are more likely to get published than studies with null results, which are subsequently not included in the pooled analysis. This would lead to biased results (43). We did two formal tests for assessing publication bias. However, the results must be interpreted with caution due to the fairly low power when applied to a few meta-analytically investigated studies. It is unlikely that publication bias does jeopardize the results of the present analysis, because none of the published studies contributing data to the GSEC database had detected a significant positive association of GSTs with breast cancer. Failure to locate and include further null studies cannot bias the presented null results (44). Additionally, data from an unpublished study were included in the pooled analysis. The crude OR for this study (Rebbeck et al.) yielded a significant negative association of GSTM1 null and breast cancer risk and seemed as outlier on the funnel plot.

Pooling and analyzing individual data from original studies has several advantages. Combining the individual efforts of different authors, a large sample size can be obtained. Allowing a type 1 error of 5%, the present study has power greater than 80% to detect an effect size of 1.2 for GSTM1, 1.4 for GSTT1, and 1.6 for GSTP1 (in the latter case, considering only homozygous mutants as exposed). This also provides the opportunity to study gene-environment interactions in stratified analysis, whereas smaller studies have not enough power to investigate this. Additionally, one has the possibility to adjust for potential confounders using the individual data. This would not be possible in meta-analysis wherein the study-specific estimates of effect are used to obtain an overall summary estimate.

Uniform coding of all data following a standard protocol is a benefit for the analysis and increases reliability of results. However, the problem of misclassification due to, for example, application of different PCR protocols, as is the case in the present pooled analysis, remains. Some information might get lost if exposure has to be categorized coarsely, as was the case with smoking exposure in this study. A drawback of the present study is that information on alcohol consumption was not available. Alcohol is involved in the etiology of breast cancers (45) and a recent report suggested an interaction of GST and alcohol consumption in breast cancer patients (46).

In conclusion, this pooled analysis provides strong evidence that GSTs do not play a major role in susceptibility to breast cancer. Furthermore, there was no evidence for a modifying effect of reproductive factors and exposure to tobacco smoke, in agreement with most previous studies. Taken together, the results of this pooled analysis and the results of the original reports and more recent studies consistently showed no significant association of GSTs with breast cancer risk. Given this strong evidence in favor of the null hypothesis, a clear compelling rationale should be provided to justify further studies into GST polymorphisms and breast cancer risk.

We thank Heide A. Stirnadel (Worldwide Epidemiology, GlaxoSmithKline) for helpful comments on an earlier version of the article.