Inherited polymorphisms in enzymes that activate or detoxify chemotherapy drugs are thought to account for some of the variability in toxicity and efficacy of cancer treatment.1 The GST enzymes catalyze the glutathione-dependent detoxification of several chemotherapeutic drugs or their metabolites.2, 3 Polymorphisms that result in reduced (e.g., GSTP1 single nucleotide polymorphisms [SNP]) or no (e.g., GSTM1 and GSTT1 deletion polymorphisms) activity of certain GST enzymes are recognized. These polymorphisms may alter the metabolism of chemotherapeutic drugs and modify the effectiveness of therapy, as suggested by reports that GST polymorphisms predict differences in outcomes of treatment for cancers including breast cancer,4, 5, 6 leukemias7, 8, 9, 10 and colorectal cancer.11
GSTA1 and other GSTs of the α class are the predominant GSTs in human liver,12, 13 the major site of drug metabolism, and are also expressed in other tissues.14, 15In vitro studies have shown that among human GSTs, GSTA1 has the highest catalytic activity for glutathione conjugation of nitrogen mustard chemotherapy agents,3 including metabolites of cyclophosphamide (CP),16 which is used in combination chemotherapy for breast cancer. A polymorphism that influences the hepatic expression of GSTA1 has recently been described.15, 17 Liver cytosols from individuals who carried the variant GSTA1*B allele, which consists of several linked SNPs in the proximal promoter region of the GSTA1 gene, had reduced levels of GSTA1 enzyme.17 Because of the importance of the GSTA1 enzyme in metabolism of chemotherapeutic drugs, it can be hypothesized that individuals carrying the low-expression GSTA1*B allele may have altered responses to chemotherapy.18
To our knowledge, no studies have considered genotypic variants affecting α class GST enzymes in relation to outcomes of cancer treatment. We have determined genotypes at the GSTA1 proximal promoter polymorphism in normal DNA from women who received CP as part of combination chemotherapy for breast cancer and we have evaluated overall survival in relation to GSTA1 genotype.
AC, adriamycin and cyclophosphamide; AJCC, American Joint Committee on Cancer; CAF, cyclophosphamide, adriamycin and 5-fluorouracil; CI, confidence interval; CMF, cyclophosphamide, methotrexate and 5-fluorouracil; CP, cyclophosphamide; ER, estrogen receptor; GST, glutathione S-transferase; HR, hazard ratio; PR, progesterone receptor; SNP, single nucleotide polymorphism.
SUBJECTS AND METHODS
We conducted a retrospective study in a cohort of women who were treated for breast cancer at the University of Arkansas for Medical Sciences in Little Rock, Arkansas. The study protocol was approved by that institution's Institutional Review Board. Women diagnosed during the years 1985–96 who received combination chemotherapy as first course of therapy for invasive breast cancer were eligible for the study. Archived, paraffin-embedded tissue from surgery was used as a source of DNA for genotyping, so only patients who had surgery for their primary tumor at the study hospital were included in the study. We queried computerized hospital tumor registry records to identify women treated for breast cancer. Information on characteristics at diagnosis (age, ethnic origin, AJCC disease stage,19 estrogen receptor [ER] and progesterone receptor [PR] status, number of positive nodes, histology and grade), first course of treatment and follow-up (vital status, cause of death and last contact date), was obtained from the tumor registry. For some cases with missing information, the pathologic slides were reviewed to determine histology and grade. We have reported previously on survival in the same cohort of patients;5, 6 additional subjects treated by chemotherapy were added for the present analysis and patients who received radiation but no chemotherapy were excluded.
DNA isolation and genotyping
DNA was isolated from tissue preserved in paraffin blocks, as described previously.5 A histologically normal lymph node or nodes was the first choice of tissue. If no normal lymph node tissue was available, a block containing normal skin or breast tissue was selected. Tumor tissue was used if no other tissue was available for a subject. The SNP C-69T in the proximal promoter region of the GSTA1 gene was detected as described by Coles et al.17 Briefly, we used the polymerase chain reaction to amplify a 480-bp fragment of the gene, followed by digestion with the restriction enzyme Ear1 (New England Biolabs, Beverly, MA). The GSTA1*B allele is recognized by 380 and 100 bp fragments.
Associations between GSTA1 genotypes and patient or tumor characteristics were assessed using Pearson's χ2 test. Departure of the distributions of genotypes from Hardy-Weinberg equilibrium was evaluated by Pearson's χ2 test.
Patients were followed by the hospital tumor registry from the time of diagnosis until death, or until the end of follow-up for our study (October, 1999). Death from any cause was the endpoint for our primary analysis. The tumor registry classified causes of death as cancer, non-cancer or unknown. A separate analysis using deaths known to be due to cancer as the outcome was also conducted. Survival time was calculated as the time from diagnosis to death or to the last contact date for living subjects. Person-years were calculated as the sum of survival times for all subjects within a group. The association between GSTA1*B genotype and survival was evaluated using the Kaplan-Meier survival function and log rank tests and relative risks were estimated by calculating hazard ratios (HR) from Cox proportional hazards models. We evaluated several breast cancer prognostic factors as potential confounders of the GSTA1-survival association by including them as covariates or stratifying variables in the Cox model. Time-specific HRs for the association between GSTA1 genotype and survival were calculated by including a time-dependent covariate, using a dichotomous time variable, in the Cox model. Stata software (Stata Corp., College Station, TX) was used for statistical analysis.
Treatment and follow-up
The characteristics of the study population are shown in Table I. The median age at diagnosis was 49 years (range 25–78 years); 64% of the subjects had node-positive disease at diagnosis. Thus our study population, who were patients referred to a cancer research center for chemotherapy treatment, was younger and included a higher proportion with node-positive disease than would be expected in a population-based sample of breast cancer patients.20 All 245 patients had received combination chemotherapy, most frequently CP in combination with methotrexate and 5-fluorouracil (CMF) before 1990, whereas CP and adriamycin with or without 5-fluorouracil (AC or CAF) became the more common treatments in 1990 and later. Many of the patients had received at least 1 additional adjuvant therapy: 58 received tamoxifen, 50 received radiation therapy and 31 were treated with both tamoxifen and radiation.
Table I. Characteristics of 245 Women Treated For Breast Cancer, by GSTA1 Genotype1
The median follow-up time among women alive at the end of observation was 73 months. The range of follow-up times for living subjects was large, 14–159 months, because eligible patients were diagnosed over a wide span of years, 1985–96. Most patients had been followed up by the hospital tumor registry within 2 years of the date that registry records were queried for our study (October, 1999); there were only 6 living subjects with last contact dates more than 2 years before the end of the study. At the end of follow-up, 89 subjects were deceased (Table II), and cancer was recorded as the cause of death for 66 of these (74%). Only 6 deaths (7%) noted during the follow-up period were due to non-cancer causes, but cause was unknown for 17 deaths (19%). The low proportion of deaths attributed to non-cancer causes in this cohort is probably explained by the fact that study population was young and included a high proportion of patients with high-risk disease.
Table II. Vital Status of 245 Breast Cancer Patients At End of Follow-Up, By GSTA1 Genotype1
Values are n (%).
Cause of death
The distributions of GSTA1 genotypes according to subject characteristics are shown in Table I. Among Caucasian subjects, 36% were homozygous for the common allele (GSTA1*A/*A), 48% were heterozygous (GSTA1*A/*B) and 17% were homozygous for the variant allele (GSTA1*B/*B). Among African-American subjects, the proportions were similar, with 30% of subjects GSTA1*A/*A, 57% GSTA1*A/*B and 13% GSTA1*B/*B. The distributions of genotypes within each ethnic group did not depart from Hardy-Weinberg equilibrium (p = 0.99 for Caucasians and p = 0.69 for African-Americans). The distribution of genotypes in the present study population was similar across age groups and by stage at diagnosis. There were fewer ER negative and PR negative subjects in the GSTA1*B/*B genotype category, but these differences could be explained by chance.
GSTA1 genotype and survival
The Kaplan-Meier survivor function showing overall survival by GSTA1 genotype is presented in Figure 1. The shape of the curves indicated that overall survival was similar among subjects with GSTA1*A/*A andGSTA1*A/*B genotypes. We therefore combined the GSTA1*A/*A and GSTA1*A/*B groups for further analysis. The proportion of the combined GSTA1*A/*A and GSTA1*A/*B subjects surviving at 5 years was 0.66 (95% CI = 0.59–0.72), whereas for GSTA1*B/*B subjects the proportion was 0.86 (95% CI = 0.67–0.95). In a log-rank test stratified on stage and node status at diagnosis, survival for the GSTA1*B/*B subjects (11 deaths observed, 17.4 expected) was somewhat better than the combined GSTA1*A/*A and GSTA1*A/*B group (78 deaths observed, 71.6 expected), χ2(1) = 3.22, p = 0.07.
We next considered the association between GSTA1 genotype and survival in a multivariate Cox proportional hazards model. Although the analysis shown in Table I did not detect any important associations between GSTA1 genotype and patient characteristics, when covariates with strong prognostic value, i.e., stage and node status at diagnosis, were included in the model, the HR for GSTA1 genotype shifted away from the null and the confidence intervals became narrower. We therefore present results adjusted for stage at diagnosis (AJCC Stage 1–4), number of positive nodes (0, 1–3, 4+), ER and PR status, ethnic origin and age (<70 or ≥70 years). Further adjustment for histology (ductal or lobular), tumor size or tumor grade had essentially no effect on the HR for GSTA1*B/*B. Consistent with the survival curves in Figure 1, the adjusted model showed that the survival of subjects with GSTA1*B/*B genotype was longer than that of the GSTA1*A/*A and *A/*B group, with an estimated HR of 0.5 (Table III). When the analysis was repeated treating only deaths known to be due to cancer as failures, HRs were very similar to those in Table III, but confidence intervals were wider due to the smaller number of events. Table II shows that “other” and “unknown” causes of death are distributed approximately evenly across all three genotypes, so the association between GSTA1 genotype and survival is unlikely to be explained by a relationship with any cause of death other than breast cancer.
Table III. Survival By GSTA1 Genotype Among Women Treated For Breast Cancer
Hazard ratios calculated from a Cox proportional hazards model, adjusted for age, stage at diagnosis, node status, ethnic origin, and ER and PR status.
Hazard ratios from a Cox model with an interaction term between GSTA1*B/*B and a dichotomous time variable; in a test for interaction by time (likelihood ratio test), p = 0.01.
GSTA1*A/*A or *A/*B
Time-dependence of the GSTA1-survival association
In the Kaplan-Meier function (Fig. 1), a survival advantage for GSTA1*B/*B subjects was apparent during the first few years after diagnosis, but the survival curves became closer together at later time points. This suggested that the hazards were not proportional with time. We used a Cox model with a time-dependent covariate to calculate the HRs for GSTA1*B/*B during 2 time intervals: an early period within 5 years of diagnosis and a later period from 5 years to the end of followup (Table III). There was a significant interaction between GSTA1*B/*B genotype and time, with a reduced hazard of death associated with GSTA1*B/*B during the first 5 years after diagnosis, but no difference by genotype during the later time interval.
Combined GSTA1 and GSTP1 results
We have reported previously that GSTP1105Val variant was associated with longer survival in the same study population5 and based on in vitro activity of GSTs with CP intermediates,16 GSTA1 and GSTP1 are both expected to act in detoxification of CP. We therefore evaluated the effect of these genotypes in combination among 230 study subjects treated by chemotherapy and genotyped for both GSTA1 and GSTP1. When the associations between the GSTP1Val/Val and GSTA1*B/*B genotypes with overall survival were considered in a Cox model that included both genotypes, each remained independently associated with survival, with HRs essentially unchanged compared to results from separate models.
We had not explored time-dependence of the GSTP1-survival association in the earlier report. After additional chemotherapy-treated patients were added to the data set for the current analysis, we were able to re-analyze the GSTP1-survival association in a time-dependent model. There was evidence of an interaction of GSTP1 with time (p = 0.05), with reduced hazard of death more apparent during the first 5 years after treatment.
We evaluated survival in relation to a novel polymorphism that affects expression of GSTA1, an important drug-detoxifying enzyme, in a cohort of women who received CP in combination with other drugs for breast cancer. Subjects who were homozygous for the low-expression GSTA1*B allele had longer overall survival, with hazard of death reduced to 30%, during the first 5 years after diagnosis. Analysis of human tissue samples has shown that individuals who were homozygous for the GSTA1*B allele had less GSTA1 enzyme expressed in liver.17 Based on the in vitro activity of GSTA1 in glutathione conjugation of CP metabolites,16 we would predict that GSTA1*B/*B patients treated with CP would have reduced ability to detoxify CP and would receive a higher effective dose. Thus the improved survival for this group may be attributable to improved treatment efficacy.
We found significant heterogeneity in the GSTA1-survival association with time, with improved survival for the GSTA1*B/*B subjects during the first 5 years after diagnosis. Because of the heterogeneity with time, the 2 HRs from the time-dependent analysis (Table III) should be considered a better description of the GSTA1-survival association than an overall HR for the entire follow-up period. For patients who survived longer than 5 years, there was no evidence of a continued survival difference by genotype. Although combination chemotherapy for breast cancer contributes to improved cumulative survival for breast cancer patients as long as 20 years after treatment,21, 22 reports of analyses from some clinical trials have noted that, when the time-specific hazard for recurrence was considered, the risk reduction attributable to chemotherapy was strongest within 5 years of treatment.23, 24, 25 If the effect of CP-containing chemotherapy is more pronounced close to the time of treatment, then our observation that the GSTA1-survival association was limited to a time period close to treatment is in agreement with the interpretation that GSTA1 genotype modifies the effectiveness of therapy.
Although this is the first report of an association between GSTA1 genotype and cancer survival, other studies have reported differences in outcomes of therapy according to a polymorphism affecting the GSTP1 enzyme. Because GSTP1 also catalyzes glutathione conjugation of chemotherapeutic drugs including CP,16 the GSTP1 and GSTA1 polymorphisms may act via a similar mechanism. We reported an association between the GSTP1105Val variant and survival in the same study cohort.5 The results for GSTP1 and GSTA1 were similar; patients who were homozygous for the variant alleles, representing low activity and low expression, respectively, had reduced hazard of death and the associations were both stronger during the 5 years after treatment. The GSTP1105Val polymorphism has also been reported to be associated with longer survival for patients receiving combination chemotherapy for advanced colorectal cancer11 and with increased risk of therapy-related acute myelogenous leukemia after chemotherapy treatment for another primary malignancy.26 These latter observations are also consistent with a mechanism in which reduced GST activity toward drug substrates leads to increased systemic dose, which in turn can result in better therapeutic effect, but also may result in increased risk of an adverse drug effect.
Some limitations of this pilot study should be acknowledged. The survival difference that we observed was based on a relatively small group, 39 subjects with GSTA1*B/*B genotype and the time-specificity of the association was not anticipated. For these reasons, a role of chance in this association should be considered until results of additional studies are available. The analysis presented here was restricted to patients who were treated by chemotherapy. We were unable to consider interaction between chemotherapy and genotype because among breast cancer patients treated at the study hospital, the number of deaths of patients who did not receive chemotherapy, usually patients with node-negative disease, was too small to allow useful analysis. No information on drug toxicity or adverse outcomes was available for patients in the present study.
We focused our hypothesis on CP because of the in vitro data linking the GSTA1 enzyme with CP metabolites as substrates.16 However, all of the patients in the present study were treated by combination chemotherapy, either CMF, AC or CAF and some patients also received radiation therapy or tamoxifen, so the GSTA1-survival association that we observed can not be conclusively attributed to variation in response to any single drug. Despite this caveat, we regard modification of CP metabolism as the most likely explanation of the association between GSTA1*B/*B genotype and better 5-year survival. Pharmacokinetic studies should be carried out to establish whether kinetics of CP or other drugs differ according to GST genotype. Future studies of GST genotypes and cancer patient survival should include larger groups of patients and should consider clinical endpoints such as drug toxicity. Studies within clinical trial populations would be of interest to assess the role of GSTA1 by type of treatment. Such studies would be important to understand the mechanism for GST-survival associations and their clinical implications.
Research on the role of inherited enzyme polymorphisms in cancer patient prognosis may ultimately have implications for therapeutic decisions. There are now several categories of adjuvant therapy with proven benefit for breast cancer patients, including radiation, polychemotherapy, hormonal therapy and taxanes. Physicians and patients must consider the individual's potential to benefit and potential for side effects, from each of these. If larger studies show that subgroups of patients, because of inherited genotypes for metabolizing enzymes, have the most potential to benefit from a particular type of therapy, then in future the patient's pattern of inherited enzyme variants may become an additional factor to consider in deciding on the best course of therapy.
This research was supported in part by the Arkansas Breast Cancer Research Program. The authors thank L. Erkman, M.Y. Fares and C.L. Woods for technical assistance.