Gene Polymorphisms in Cyclophosphamide Metabolism Pathway,Treatment-Related Toxicity, and Disease-Free Survival in SWOG 8897 Clinical Trial for Breast Cancer

Adjuvant chemotherapy for early stage breast cancer patients has greatly improved both disease-free and overall survival by eliminating residual tumor cells after surgery (1, 2). However, because of lack of target specificity, systemically administered cytotoxic agents also cause damage to normal tissues, leading to severe and sometimes life-threatening toxicities (3). Treatment outcomes, including drug toxicities and disease-free survival (DFS), may vary greatly even among women receiving the same dose of the same drug. This is likely due, in part, to metabolic differences resulting in variability in effective drug dose delivery, as well as cell sensitivity to and defense against drug cytotoxicity.

Cyclophosphamide (CP) is a commonly used alkylating drug for breast cancer patients. There are large variations in pharmacokinetics of CP, with interindividual variability estimated at 52% in the central volume of distribution, 36% in peripheral volume of distribution, and 22% in clearance of the active metabolites (4). The variation may be partially due to patient characteristics such as weight and age but may also be due to genetic background (5). A number of genes including those encoding for phase I activation enzymes and phase II detoxification enzymes form complex pharmacokinetic networks that determine the effective dose delivered to target cells (6, 7). These key genes are highly polymorphic among individuals, and functions of some common polymorphisms have been characterized, making them excellent candidates for pharmacogenetic studies.

CYP2B6 and CYP3A4 are 2 major hepatic enzymes that convert CP prodrug to its active 4-hydroxylated metabolites. A neutral amino acid replacement in exon 4 of CYP2B6 resulting from a single nucleotide polymorphism (SNP) rs3745274 (Gln172His) is associated with increased catalytic activity in both cell cultures and human liver specimens (8, 9). For CYP3A4,SNP rs2740574 in the 5' promoter region is most commonly studied. Although it has little impact on constitutive expression (10, 11), it may result in lower enzymatic activity toward CP as shown recently in breast cancer patients (12). GSTA1 and GSTP1 are the 2 major detoxification enzymes for CP, conjugating reduced glutathione to CP metabolites. SNP rs3957356 in the 5' promoter of GSTA1gene was reported to alter binding affinity of transcription factors and thus result in lower expression (13). A nonsynonymous coding SNP rs977894 (Ile105Val) in exon 5 of GSTP1 decreases its glutathione-conjugating affinity, resulting in compromised catalytic detoxification functions (14–16).

In the Southwest Oncology Group (SWOG) S8897 trial, a clinical trial with CP-containing regimens, we conducted an ancillary pharmacogenetic study to examine 4 common SNPs of CYP2B6CYP3A4GSTA1, and GSTP1 in relation to acute hematologic toxicity and DFS after adjuvant chemotherapy.

This correlative study was nested in a completed clinical trial led by the SWOG for The Breast Cancer Intergroup of North America (S8897, INT0102). Details regarding trial design, patient characteristics, treatment plan, and outcomes were published previously (17–20). Briefly, women with stage T1 to T3a node-negative invasive breast cancer were eligible to participate. On the basis of tumor prognostic characteristics, including tumor size and hormone (estrogen and progesterone) receptor status, they were assigned to groups depending upon risk of recurrence (high risk, low risk, or indeterminate risk). Those in the indeterminate-risk group were reassigned to either the high- or low-risk group, based on S-phase fraction of their primary cancers as determined by flow cytometry. Women in the high-risk group (either initially or after flow cytometry) were randomly assigned to an anthracycline-based regimen of 6 cycles of oral CP, intravenous doxorubicin, 5-fluorouracil (CAF; ref. 21), or a nonanthracycline-containing regimen of 6 cycles of oral CP, intravenous methotrexate, and 5-fluorouracil (CMF; ref. 22). After chemotherapy, patients in the treated group were subsequently randomized to tamoxifen for 5 years or not. Women in the low-risk group served as an observational cohort and did not receive adjuvant chemotherapy or endocrine therapy. Prophylactic hematopoietic colony-stimulating factors (CSF) were not administered at the time of the study.

Between 1989 and 1993, a total of 3,965 eligible patients were registered onto the study. Of these, 2,690 patients were assigned to the high-risk group (either initially or after flow cytometry) and received adjuvant therapy and 1,206 were assigned to the low-risk group and followed without adjuvant therapy, as shown in Figure 1. An additional 69 patients were not classifiable as at either low or high risk and were excluded. Patient characteristics of age, race, and type of primary surgery were similar among all treatment groups. After a median follow-up time of 10.8 years, the study showed slightly better overall survival but not DFS in the CAF arm than in the CMF arm, at an expense of higher toxicity (17).

For the present ancillary pharmacogenetic study, only women with archived, uninvolved lymph node tissue available for DNA extraction in the SWOG tissue bank were included. This group consisted of women assigned to the low-risk group (initially and after flow cytometry, n = 1,206) and women initially in the indeterminate-risk group who were subsequently reassigned to the high-risk group (n = 527). Because all women with hormone receptor–negative breast cancer were classified as at high risk initially, nodal tissue was not collected. Therefore, this analysis is restricted to women who were estrogen receptor positive and/or progesterone receptor positive. The study also predates routine testing of HER2/neu status, so HER2/neu status is also unknown. Written consent was obtained from patients for use of their tissues for ancillary research studies, and this study was approved by the Institutional Review Board at Roswell Park Cancer Institute. The number of specimens collected and the number of patients for whom DNA was available in each of the cohorts are included in Figure 1, according to REMARK criteria (23). These specimens have been used for previously reported studies of the association of germline variations of other genes and outcomes (18–20).

Two 5-μm paraffin-embedded, formalin-fixed slides of normal lymph node tissue procured from the SWOG tissue bank in San Antonio were deparaffinized and removed from slides. DNA was extracted with Qiagen DNeasy Tissue Kit (Qiagen Inc.) as previously described (24), using a modified method of Goelz et al. (25). Adequate DNA for genotyping was obtained from 458 patients in the high-risk group who received adjuvant therapy and 874 patients in the low-risk group who were followed without adjuvant therapy (Fig. 1).

Potentially functional SNPs in the 4 selected key genes involved in activation and detoxification of CP, including rs3745274 in CYP2B6, rs2740574 in CYP3A4, rs3957356 in GSTA1, and rs1695 in GSTP1, were genotyped by Sequenom matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry. Successful genotyping call rates varied between 88.4% and 99.6%, related to adequacy of DNA, with 100% concordant rates between 5% repetitive samples blindly included in the analyses.

Statistical analyses were prospectively planned in a formal, written protocol and conducted by the SWOG statistical center. For the analysis, we separated treated and untreated groups. Primary endpoints assessed in the study were high-grade hematologic toxicity and DFS. Per the protocol, DFS was defined as the time from the date of registration to the date of a second breast primary tumor, first recurrence, or death due to any cause, but not a new non–breast primary cancer (26). S8897 was conducted prior to the development of the NCI (National Cancer Institute) Common Toxicity scales and therefore standard SWOG toxicity criteria applied per protocol at the time were used (27). In brief, grade 2 to 4 neutropenia was defined by a neutrophil count of 1,000–1,500, 500–1,000, and <500/mm³, respectively; and grade 2 to 4 leucopenia was defined by a leukocyte count of 2,000–3,000, 1,000–2,000, and <1,000/mm³, respectively. The highest grade toxicity that occurred during the treatment was recorded and used in this analysis. Prevalence of acute toxicities within 1 year of registration by treatment group in the trial has been previously reported (17) and was similar among the subset of patients included in this ancillary study. Hematologic toxicities, mainly neutropenia and leucopenia, were the most common acute toxicities in all treatment groups; grade 4 nonhematologic toxicity, including severe cardiac toxicity, was rare (n = 2/458). Thus, primary endpoints in toxicity analysis were grade 3 and 4 neutropenia, grade 3 and 4 leucopenia, and combined grade 4 hematologic toxicity.

For toxicity analysis in women who received adjuvant therapy, chi-square test or Fisher's exact test was first done to examine genotype distribution by toxicity grade. Considering the incomplete knowledge of exact functions of the genes and the genotypes, if homozygous and heterozygous genotypes had similar frequency distributions by toxicity grade, the categories were combined under an assumption of dominant or recessive genetic models. Similarly, rare homozygous genotypes were collapsed with heterozygous genotypes for power considerations. Unconditional logistic regression was used to estimate associations between each SNP and risk of high-grade toxicity. For neutropenia and leucopenia, grade 3 and 4 toxicities were first combined and tested together and then tested in an ordinal model under an assumption that there is a linear relationship between toxicity grades. In addition to a binary outcome of grade 4 hematologic toxicity (yes vs. no), discrete count of grade 4 toxicity based on neutropenia and leucopenia (0, 1, and 2) was also used as an ordinal outcome in logistic regression models. Unadjusted odds ratios (OR) and 95% CIs were first derived and then were adjusted for age, menopausal status, chemotherapy arms (CAF vs. CMF), tamoxifen treatment, and time from surgery to randomization. Additional control for race or excluding nonwhite race from the analyses did not substantially change the results. According to the preplanned analytic protocol, for a SNP with a minor allele frequency in the range of 0.05 to 0.50, we would have at least 80% power to detect a minimum OR in the range of 0.56 to 0.64 for grade 4 toxicity. Four SNPs were tested in the analyses; for each SNP, 2 genetic models including dominant and ordinal models and 3 toxicity outcomes including neutropenia, leucopenia, and grade 4 hematologic toxicity were tested, yielding a total of 24 tests. To address the concern of false positivity due to multiple testing, we calculated the false-positive report possibility (FPRP), using the method by Wacholder et al. (28), assuming a prior probability for associations with toxicities of 0.1, 80% statistical power, and FPRPs below 0.5 and below 0.2 for suggestive associations and more definitive associations, respectively.

Potential modifying effects of genotypes on DFS were examined by contrasting event-free time between those carrying variant alleles and those with homozygous common alleles. Kaplan–Meier survival curves were generated and the differences were tested by log-rank test. To control for potential confounders, Cox proportion hazard models were used to adjust for age, menopausal status, and for the treated group only, also time from surgery to randomization, chemotherapy arms (CAF vs. CMF), and tamoxifen treatment.

To test interacting effects from chemotherapy regimens and tamoxifen treatment, we fitted in the regression models a productive term between treatment regimen and genotypes. The significance of the interactions was assessed by the likelihood ratio test. To test synergistic and/or additive effects among multiple SNPs, the number of “at-risk” alleles that increases CP activation or decreases detoxification was tallied (range = 0–8) and categorized into ranks. Toxicities and DFS were then compared across these ranks.

Characteristics of patients included in this ancillary study are summarized in Table 1; there were no differences between this subset of patients and those in the entire trial (17). As expected, women in the treated group tended to be younger, premenopausal, and have larger tumors than those in the untreated group who were at lower risk of recurrence. During a median 10.8 years of follow-up, there were 129 events in the treated group and 257 in the untreated group. S8897 was conducted prior to the era of CSF support. Of the patients who received adjuvant therapy, there were few reports of thrombocytopenia (2%), but 67% had grade 3 and 4 neutropenia and 62% had grade 3 and 4 leucopenia. Thirty-five percent of patients had 1 grade 4 hematologic toxicity, and 12% experienced 2 grade 4 events. Genotype frequencies are summarized in Table 2 and were similar to those in the SNP500Cancer database.

As shown in Table 3, there were no significant associations between the polymorphisms in CYP2B6, CYP3A4, or GSTA1 and the risk of grade 4 hematologic toxicity, or with the risk of high-grade neutropenia or leucopenia (data not shown). However, there were notable associations between the GSTP1 polymorphism and reduced risk of high-grade hematologic toxicity, with similar ORs in unadjusted and adjusted models (Table 4). Compared with women homozygous for the common allele, those carrying the less active variant G allele had significantly lower risk of grade 3 and 4 neutropenia (adjusted OR = 0.63, 95% CI = 0.41–0.97, P = 0.04), and grade 3 and 4 leucopenia (adjusted OR = 0.59, 95% CI = 0.39–0.89, P = 0.01). Associations with risk of any grade 4 hematologic toxicity were of borderline significance (adjusted OR = 0.72, 95% CI = 0.48–1.07, P = 0.11). Results from the models with the ordinal outcomes (grade ≤2, 3, and 4; or count of grade 4 of hematologic toxicity 0, 1, and 2) were consistent with these models with binary outcomes. FPRP calculation indicates that the strength of the observed associations between the GSTP1 polymorphism and hematologic toxicity was between suggestive (FPRP < 0.5) to more definitive (FPRP < 0.2).

Associations between each individual SNP and DFS are shown in Table 5. There was no statistically significant association of any SNP in any of the evaluated genes with DFS in women receiving or not receiving adjuvant therapy. No significant interactions between genotypes and treatment randomization (CMF vs. CAF, tamoxifen yes vs. no). In analyses to examine synergistic and/or additive effects of the 4 selected SNPs, there were also no associations of hematologic toxicities or DFS with total number of at-risk alleles (data not shown).

In this pharmacogenetic study, there were significant associations between a common genetic polymorphism in GSTP1 and acute hematologic toxicity but no associations with DFS. Selected polymorphisms in 3 other key genes involved in CP metabolism (CYP2B6, CYP3A4, and GSTA1) were not associated with hematologic toxicity or DFS following adjuvant therapy. Because CSFs for neutropenia prophylaxis are now commonly used in patients receiving breast cancer adjuvant chemotherapy, our findings of significant associations between the GSTP1 polymorphism and hematologic toxicity suggest that a subgroup of women may be at low risk. Thus, CSF support may not be necessary when considering the cost and potential side effects from growth factors, including bone pain and decreased bone mineral density. However, further confirmation in other data sets using archived specimens, or in a prospective trial, from patients who received CP without CSF support is required before such a strategy would be recommended

The variant G allele of GSTP1 polymorphism has been associated with increased risk of toxicity in colorectal cancer patients (29) and increased risk of neutropenia in patients with lupus erythematosus treated with CP (30). However, in women with breast cancer, this SNP was not associated with adverse drug reactions after CP-containing chemotherapy in 2 previous studies (31, 32). In a small group of 94 women receiving 6 cycles of anthracycline-based cyclophosphamide, epirubicin, and 5-fluorouracil (CEF) regimen for breast cancer, GG genotypes of GSTP1 were associated with increased risk of grade 3 and 4 hematologic toxicity (OR = 6.4, 95% CI = 1.05–39.0; ref. 33). Surprisingly, we found in our study an inverse association, which cannot be explained by direct drug detoxification function of GSTP1. However, there are other reports in the literature showing similar trends. Compared with GG genotypes, the more active AA genotype has been associated with increased risk of oxaliplatin-induced neuropathy (34), docetaxel-induced neuropathy (35), and cisplatin-induced hearing impairment (36). A possible alternative mechanism to explain these unexpected findings in drug-related toxicity is through the novel roles of GSTP1 in cellular stress response signaling as an inhibitor of c-Jun N-terminal kinase (JNK; ref. 37), a key cellular stress mediator that can activate downstream transcription factors to upregulate a number of antioxidant cellular response genes. It was speculated that less active G alleles result in increased JNK activity and elevated expression of downstream cellular stress defense genes, which may protect cells from drug cytotoxicity (34). Slower consumption of reduced glutathione by the less active GSTP1 variant may also facilitate the defense against cellular stress and may prevent cell apoptosis (38). In addition, it is possible that the functional alteration of the G allele may be drug-specific, as supported by findings in cultured cells that the “less active” GG genotype showed enhanced cisplatin detoxification compared with the AA genotype (39). Our study, along with others, calls for further drug- and tissue-specific functional characterization of this SNP.

Despite significant associations with reduced risk of toxicity, GG genotypes of GSTP1 were not associated with poorer DFS [hazard ratio (HR) = 1.19, 95% CI = 0.66–2.13] among women receiving chemotherapy (Table 5). The lack of concordance between survival and toxicity may be due to differential capacities of normal and malignant cells in dealing with drug cytotoxicity, which may be further attributed to somatic changes incurred during tumorigenesis in cancer cells. This null result, however, is unexpected on the basis of our previous findings in a smaller study, in which the HR for overall breast cancer survival was 0.3 (95% CI = 0.1–1.0) for GG genotypes compared with AA genotypes (40). Similar results were also reported in a study with Chinese breast cancer patients (41). Heterogeneity in patient characteristics and treatment regimens among those studies may partially account for the inconsistent findings.

There were no significant associations between treatment outcomes and polymorphisms in CYP2B6, CYP3A4, or GSTA1 genes. The null findings in our study could be due to redundancy in drug metabolism genes. Although we selected 4 key genes in CP metabolism pathway, a number of other genes in the same pathway may neutralize any functional changes due to the selected genetic variations.

In conclusion, in a cooperative group clinical trial for breast cancer, we found that genetic polymorphisms in key genes in CP metabolism were not associated with DFS following adjuvant chemotherapy. Of the genes evaluated, only GSTP1 polymorphism was significantly associated with risk of acute hematologic toxicity. Our results indicate that genetic variations in CP pharmacokinetic pathways may not be candidate markers for predicting treatment outcomes. Women carrying GSTP1 variants may be at lower risk of neutropenia, suggesting that genetic markers in combination with other clinical and molecular factors may be useful in predicting risk of neutropenia, and for those at low risk, prophylactic CSF may not be necessary. These results need to be further validated in independent studies before clinical recommendations can be made.

No potential conflicts of interest were disclosed.

This investigation was supported in part by the following R01 and PHS Cooperative Agreement grant numbers awarded by the National Cancer Institute, DHHS: CA095222, CA32102, CA38926, CA02599, CA13612, CA22433, CA27057, CA37981, CA46282, CA20319, CA35431, CA76447, CA45560, CA12644, CA14028, CA58416, CA04919, CA35090, CA35176, CA58686, CA58861, CA46113, CA58882, CA35128, CA74647, CA46136, CA45450, CA35261, CA35192, CA12213, CA16385, CA58658, CA46441, CA58723, CA45377, CA35119, CA42777, CA73590, CA114558–02, CA35178, and CA35262. C.B. Ambrosone, G.N. Hortobagyi, and J.M. Rae are recipients of funding from the Breast Cancer Research Foundation. S. Yao was partially supported by a Department of Defense award DAMD W81XWH-08-1-0223.