The Glutathione S-Transferase-μ and -θ Genotypes in the Etiology of Prostate Cancer: Genotype-Environment Interactions with Smoking¹

The μ and θ classes of the GSTs³have been implicated in the etiology of cancer at several sites (1, 2, 3, 4, 5). To date, most studies have focused on homozygous deletions in these genes as risk factors for chemical carcinogenesis;homozygous deletions exist at GSTM1 in ∼40–60% of the Caucasian population in the United States and at GSTT1 in∼20–30% of the Caucasian population in the United States (1). Homozygous deletions at GSTM1 have been associated with cancers of the lung (6), colorectum (7), and stomach (8). Less is known about cancer risk attributable to homozygous deletions at the GSTT1 locus, although increased breast and larynx cancer risk have been associated with GSTT1 deletion (9, 10). A recent case-control study showed that individuals carrying the non-deleted GSTT1 allele, either as heterozygotes or homozygotes, were at increased risk of developing prostate cancer (OR, 1.83; 95% CI, 1.19–2.80; Ref. 2). That sample contained many of the same individuals who are studied in the present paper. This finding was consistent with the hypothesis that GST-θ catalyzes the bioactivation of certain xenobiotics to genotoxic metabolites (11, 12, 13, 14) such as dichloromethane and other halogenated alkanes.

Results from previous studies evaluating smoking as a risk factor for prostate cancer have been equivocal (15, 16). However, few studies have evaluated interactions between exposures (e.g.,smoking or occupational exposures) and genes encoding carcinogen metabolism enzymes in prostate cancer etiology. We hypothesized that interactions between smoking and the non-deleted GSTM1 or GSTT1 genotype may be associated with increased risk of prostate cancer. We investigated this hypothesis using a case-control study of 276 men (ages 41–80 years) that developed prostate cancer between 1994 and 1998, and 499 controls identified during the same time period from clinics of HUP.

A sample of incident cases was identified through Urological Oncology Clinics at HUP between 1994 and 1998. All study subjects provided informed consent for participation in this research under a protocol approved by the Committee for Studies Involving Human Subjects at the University of Pennsylvania. Men were excluded from the study if they reported having exposure to finasteride (Proscar), were diagnosed with prostate cancer more than 12 months before joining the study, or had a prior diagnosis of cancer at any site.

The controls studied here were men attending HUP general medicine clinics. These clinics see a patient population that is demographically similar to those seen in the Urological Oncology Clinics at HUP. These men were ascertained concurrently with the CaP cases (i.e.,between September 1994 and April 1998). Controls were excluded from this study if they ever had an abnormal prostate-specific antigen test(i.e., ≥4 ng/dl), if they had ever had an abnormal digital rectal examination, if they had a previous cancer diagnosis, or if they reported having had exposure to finasteride (Proscar) at the time of study ascertainment. Prostate-specific antigen values were available from medical records and/or self-report from all controls. Information was not available about non-cancer diagnoses or exposures to medications used to treat these conditions. Controls were frequency-matched to cases on the basis of age and race. However, because the frequency matching was approximate, analyses adjusted for age and race were also undertaken to account for residual variation attributable to these factors. The mean age in the cases was 61.2 years versus 60.2 years in the controls. The vast majority of cases and controls were Caucasian (84.0%), 13.3% were African-American, and about 3% of the study populations were of Asian or other racial category.

Medical history and prostate cancer diagnostic information were obtained by using a standardized questionnaire and a review of medical records. With regard to smoking history, subjects were asked if they had ever smoked cigarettes for a period of 1 year or longer. Those who replied in the affirmative were considered smokers in the present analyses.

Samples of genomic DNA were obtained by self-collection by each study subject using sterile cheek swabs. The samples were processed and analyzed using the protocol described previously by Rebbeck et al. (2). A small number of genotypes were not determined for either GSTM1 or GSTT1 because of genotype assay failures. The following genotype categories were used in this study: (a) homozygous deletion genotypes at GSTM1, denoted “GSTM1–0,” were compared with genotypes for which at least one non-deleted allele was present,denoted “GSTM1–1”; (b) the same definition was used to distinguish between the deleted and non-deleted GSTT1 genotypes (“GSTT1–0 ” and“ GSTT1–1,” respectively).

Genotype-disease associations, smoking-disease associations, and genotype-smoking interactions were analyzed using logistic regression models. A likelihood ratio test was used to evaluate whether genotype-smoking interactions significantly improved the prediction of CaP risk by comparing a reduced model, which included the main effects of smoking and genotype as predictors, with a complete model, which included those terms plus an interaction term. The likelihood ratio test statistic is approximately distributed as aχ ² with degrees of freedom equal to the difference in the number of model parameters between the reduced and complete models. We also investigated whether the interaction between genotype and smoking was age-dependent using the Cochran-Armitage test for trend (17) and the Breslow-Day test of homogeneity (18).

To estimate the potential public health implications of an interaction,we calculated the AR of the interaction using the following equation (19):

where AF_ge is the attributable fraction of cases because of the interaction, P_ge is the proportion of cases with both the risk-conferring genotype and exposure to the environment of interest (i.e., smoking), R_ge is the relative risk of CaP comparing those who were exposed to both smoking and genotype compared with those who were never smokers and who did not have the risk-conferring genotype, R_e is the relative risk attributable to the exposure (i.e., smoking)compared with those who were never smokers, and R_g is the relative risk attributable to the risk-conferring genotype compared with those who did not carry this genotype.

Table 1 presents a description of the sample population. Because of incomplete genotype data on some study subjects, the total number of subjects with GSTM1 genotype data were 675 (257 cases, 418 controls), and the total number of subjects with GSTT1 genotype data were 725 (256 cases, 469 controls). Case and controls ranged in age from 41–80 years. There were no significant differences in the age and smoking distribution between the groups with only GSTM1 genotype data and the total sample set. Approximately 53% of both cases and controls were GSTM1–1. GSTT1–1 made up 76.6% of cases,whereas 70.0% of controls were GSTT1–1.

Among 257 cases and 418 controls for whom GSTM1 genotype data were available, 146 (56.8%) and 244 (58.4%) smoked for 1 year or more, respectively. One hundred and forty-seven (57.4%) of the 256 cases with GSTT1 genotype smoked for 1 year or longer, as compared with 275 of the 469 controls (58.6%).

The results of the analyses evaluating the effect of smoking and genotype are presented in Table 2, A and B. Ever having smoked for 1 year or more was not associated with the probability of having CaP (OR, 0.95; 95% CI, 0.69–1.29), nor was the GSTM1 genotype (OR, 1.00; 95% CI, 0.73–1.36). Overall, the model containing no interaction was not statistically significant(P = 0.923). Mutually adjusting for smoking and for GSTM1 genotype did not substantially alter the observed ORs associated with smoking or with the GSTM1 genotype (Table 2A). A model containing a smoking×GSTM1 interaction was also not statistically significant (P = 0.590). Using nonsmokers with the GSTM1–0 genotype as the reference category, the ORs for CaP were: (a) OR, 0.94 (95% CI, 0.69–1.29) in GSTM1–0 smokers; (b) OR, 1.00 (95% CI,0.73–1.37) in GSTM1–1 nonsmokers; and (c) OR,0.94 (95% CI, 0.49–1.61) in smokers with GSTM1–1.

The GSTT1 genotype was a significant predictor of CaP with an OR of 1.61 (95% CI, 1.14–2.28; Wald P = 0.007). More importantly, the addition of the smoking×GSTT1interaction term to form the complete model was also statistically significant (Wald P = 0.049), as was the likelihood ratio test comparing the reduced and complete models(Δχ² = 3.86; P < 0.05). Using nonsmokers with the GSTT1–0 genotype as the reference category, the ORs for CaP were OR, 0.98 (95% CI, 0.72–1.33) in GSTT1–0 smokers; OR, 1.61 (95% CI, 1.14–2.28) in GSTT1–1 nonsmokers; and OR, 1.57 (95% CI, 1.09–2.22) in smokers with GSTT1–1.

Because race is known to be strongly associated with CaP risk, we undertook a subset analysis of Caucasians to evaluate our initial results in a more racially homogeneous population (n =607; 222 cases; 385 controls). The same logistic regression models were built, and results were obtained that were very similar to those described above. In fact, the magnitude of the smoking×GSTT1 interaction term was greater (OR, 2.21; 95%CI, 1.03–4.76) and slightly more significant (Wald P =0.043) in the Caucasian-only group (results not shown).

To further examine the interaction between GSTT1 and smoking across age groups, we stratified our sample by age decades and evaluated the differences in relative frequency of the GSTT1–1 genotype among smokers and nonsmokers in cases and controls. ORs and 95% CIs were also calculated and are presented concurrently with the genotype frequencies in Figs. 1 and 2. In nonsmokers (Fig. 1), the relative frequency of GSTT1–1genotype among cases and controls was fairly similar across the age groups. An exception was in the first age group in which 100% of the cases had the GSTT1–1 genotype, although the sample size for cases in this age group was small (n = 11). In comparison, among smokers, the relative frequency of the GSTT1–1 genotype among cases was consistently greater than among the controls across the four age groups of smokers, indicating the increased likelihood of cases who smoked to have the GSTT1–1 genotype. These same patterns of GSTT1–1 genotype frequency and ORs were observed when the sample was limited to Caucasians only.

Because there appeared to be a slight increase in GSTT1–1genotype frequency across age groups in cases who smoked, we used the Cochran-Armitage test for trend to evaluate the pattern of GSTT1–1 genotype frequencies. In smokers, the increase in GSTT1–1 genotype frequency in controls across the four age groups was significant (P = 0.040), but the increase in GSTT1–1 genotype frequency in cases was not(P = 0.820). The test of the trend of GSTT1–1 genotype frequency in nonsmokers was not significant for cases or controls (P = 0.200 and P = 0.987, respectively). Tests of homogeneity (Breslow and Day; Ref. 18) of the ORs across the age groups in smokers and nonsmokers were not statistically significant(P = 0.306 for smokers; P = 0.225 for nonsmokers). We conclude that there was no effect of age on the interaction.

Using the AR equation in the “Materials and Methods”section (Khoury et al., Ref. 19) to estimate the AR of disease because of the smoking×GSTT1interaction, we estimate P_ge to be 44.92%, R_ge to be 2.03, R_e to be 0.58, and R_g to be 1.06 (taken from Table 2B). Substituting these estimates into the AR equation yields a value of 30.9%.

We report a significant interaction between GSTT1 and smoking that elevates the risk of CaP among carriers of the non-deleted GSTT1 genotype. No interaction exists between GSTM1 and smoking. Additionally, the GSTT1×smoking interaction could explain ∼27.2% of the CaP cases in our study. Age did not have an appreciable effect on the interaction between smoking and GSTT1, although the sample size available for the test of trend in some age groups was small.

In several studies, genetic risk of environmentally induced disease has been observed to be higher in individuals who lack genes coding for carcinogen metabolism enzymes or express less active forms of these enzymes (20, 21, 22). This has been observed with the GSTs (3, 4, 5, 6, 7, 8, 10). For example, persons lacking GSTM1are at higher risk of developing lung cancer (5). In our case, we have observed the opposite; that is, individuals possessing the non-deleted GSTT1 genotype (GSTT1–1) were found to be at higher risk of prostate cancer, and this risk is heightened by an interaction with smoking.

The mechanisms by which such an interaction between GSTT1and smoking elevate the risk of prostate cancer are unknown at this point. However, two plausible mechanisms can be proposed to explain the observed relationship. First, some metabolic intermediates in the GST-θ pathway of glutathione conjugation are mutagenic (11, 12, 13, 14). At least three studies have shown that GST-θ-dependent glutathione conjugation of dichloromethane yields formaldehyde, which is capable of producing DNA-protein crosslinks or formaldehyde-RNA adducts in the human and mouse liver (12, 13, 14). Byproducts of tobacco smoke include methyl chloride (23), which is a substrate of GST-θ detected at appreciable levels in cigarette smoke. Methyl chloride undergoes a biotransformation similar to that of dichloromethane (24). Conjugation of methyl chloride by GST-θ produces S-methylglutathione and subsequently S-methylcysteine after cleavage by transpeptidases. Methylthiopyruvic acid is then formed by a transamination reaction, and then is decarboxylated to methylthioacetic acid, with methanethiol being the next metabolite. Experimental evidence from Garnier et al. (25) and Chellman et al.(26) suggests that this may be a toxifying rather than a detoxifying pathway, because the oxidation of methanethiol produces formaldehyde and hydrogen sulfide (24), both of which are toxic metabolites (27). Formaldehyde can form DNA-protein crosslinks or formaldehyde-RNA adducts, as seen during the metabolism of dichloromethane, and hydrogen sulfide is an inhibitor of cytochrome oxidase (28). Given that GST-θ is highly expressed in the prostate (29) and that systemic circulation of methyl chloride could perfuse the prostate, the production of mutagenic intermediates in the prostate itself may be a mechanism promoting tumorigenesis. The pharmacokinetic issues of whether or not methyl chloride reaches the prostate and in what concentration are important,unstudied issues that would need to be addressed to substantiate this etiological model.

Alternatively, it is plausible that smoking directly or indirectly induces inflammation of the prostate. Chronic inflammation has been linked to carcinogenesis in several organs (30, 31, 32). Systemic depletion of antioxidants by chronic exposure to tobacco combustion products as well as lung tissue injury by constituents of tobacco smoke can result in inflammation and lead to a state of oxidative stress (33, 34, 35, 36). The subsequent release of inflammatory cytokines can then promote inflammation. Endogenous substrates of GST-θ have not yet been identified, but it is known that the GSTs, in general, participate in leukotriene synthesis and are mediators of the inflammatory response. The conversion of Leukotriene A4 to Leukotriene C4 is catalyzed by the GSTs (37). Because GST-θ is highly expressed in the prostate,it is possible that upon receiving the signal of inflammatory cytokines, a more vigorous or more chronic inflammation response is mounted in the prostate gland of individuals with the GSTT1–1 genotype versus the GSTT1–0genotype individual. Inflammatory cells release reactive oxygen and nitrogen species, which can bind and alter DNA (38, 39). The prostate may already be vulnerable because of a preexisting condition of oxidative stress as well. There is evidence which suggests that inflammation of the prostate may be linked to the development of neoplasia (40, 41, 42, 43), lending credence to this model. Additionally, inflammatory cells such as neutrophils have been shown to activate procarcinogens, e.g., constituents of tobacco smoke, aromatic amines, and polycyclic aromatic hydrocarbons, to DNA-damaging species by oxidant-dependent mechanisms, providing another mechanism by which inflammation of the prostate could promote carcinogenesis (44). Finally, other lines of evidence from different types of studies also suggest that oxidative stress may be etiologically important and may be responsible for inducing carcinogenesis or disease progression in the prostate. One study has shown that dietary antioxidants inhibit prostate tumor cell proliferation in mice (45), and similar studies have led to a large-scale clinical trial that is testing the hypothesis that dietary antioxidants can prevent CaP disease progression (46).

There are a number of limitations of our study. One limitation is the way in which smoking status was assessed. Study participants were asked in a yes/no format whether they had ever smoked cigarettes for a period of 1 year or more. This question does not discern between individuals who may have vastly different smoking histories in terms of type of cigarette smoked (i.e., filtered versusnonfiltered cigarettes), amount smoked, or type of exposure(e.g., cigarette versus pipe and cigar smoking). Hence, we were unable to stratify our analyses by smoking exposure type or level. Had such stratification been feasible, it would have been possible to assess a potential dose-response relationship, perhaps revealing greater magnitudes of interaction between smoking and GSTT1. Additionally, diet and occupational exposures are confounders that may have influenced our results. Both have been implicated in CaP development by several studies (47, 48, 49, 50)and were not simultaneously addressed in this study. Finally, we detected an effect of age on genotype frequencies at GSTT1in smokers but not in nonsmokers. The observation of a trend toward decreasing GSTT1–0 frequency with age in smokers could reflect the fact that smokers with this genotype were selectively removed from the population because of other illnesses. Therefore, the observation that prostate cancer may be associated with the presence of GSTT1 genotype could be attributable to the fact that GSTT1–0 controls were in deficit in the population because of the effects of this genotype on the risk of other smoking-related diseases. However, our sample of smokers in each age group was relatively small, and may have not been adequate to evaluate trends of genotype by age. Therefore, any potential inferences about age-specific effects by genotype are limited in the present sample.

Additional research is needed to confirm these findings, preferably using more detailed assessments of smoking exposures. Future directions for this area of research should also aim to further assess interactions between GSTT1 and other environmental exposures. Occupational exposure to other substrates of GST-θ might also prove to be important in prostate cancer etiology. The mechanism(s) of these interactions also needs to be addressed to help support or diminish the strength of the hypotheses raised by the present results. Biomarkers of exposure could be used in molecular epidemiological studies to help elucidate the mechanism. The use of a prostate cell line or an animal model of prostate cancer in which human GSTT1 is expressed might also provide pathophysiological data to help differentiate between the potential inflammation-induced and bioactivation pathways of prostatic carcinogenesis.

We thank Dr. Joe McCord for his insights and helpful discussion and Dr. Leonard Sweet for his critical review of the manuscript.