Association Between CYP2D6 Polymorphisms and Outcomes Among Women With Early Stage Breast Cancer Treated With Tamoxifen FREE

Tamoxifen has been the gold standard for the last 25 years for endocrine treatment of breast cancer. It is estimated that the lives of half a million women have been saved with adjuvant tamoxifen therapy.¹ Although long-term experience showed that tamoxifen is safe and effective,² de novo and acquired drug resistance is a major issue. Several lines of evidence indicate that most of the tamoxifen antiproliferative effects in breast cancer models are mediated by the metabolites 4-hydroxytamoxifen and endoxifen,^{3 - 10} with a different mode of action for endoxifen being suggested.¹¹ Endoxifen is predominantly formed by the polymorphic enzyme cytochrome P450 (CYP) 2D6.¹² Approximately 100 CYP2D6 genetic variants have been identified, which manifest in the population in 4 distinct phenotypes, extensive (normal activity), intermediate (reduced activity), poor (no activity), and ultrarapid (high activity) metabolism,^{13 - 17} and a gene-dose effect with respect to endoxifen plasma concentrations has been demonstrated.¹⁸ Thus, it can be speculated that genotype-related differences in the formation of active metabolites influence therapeutic response to tamoxifen. Recent clinical studies in Europeans and Asians showed that nonfavorable outcome with tamoxifen was associated with poor or intermediate metabolizer genotypes^{19 - 24} ; however, 3 studies showed discrepant results.^{25 - 27}

Data from clinical trials comparing the efficacy of aromatase inhibitors (AIs) and tamoxifen showed an advantage of AIs in the adjuvant endocrine treatment of early stage breast cancer.^{28 - 30} However, because the absolute difference in recurrence rates comparing tamoxifen and AIs is less than 5%,²⁹ it has been hypothesized that treatment outcomes following tamoxifen in patients with normal CYP2D6 enzyme function would be similar to treatment outcomes following an AI. ³¹

While CYP2D6 tamoxifen pharmacogenetics have generated considerable interest and have been subject to extensive review,^{17 ,32 - 35} the clinical relevance has been questioned because of discrepant results and limited sample sizes. Therefore, we conducted an adequately powered multicenter study including retrospectively and prospectively collected patient data based on strict inclusion criteria.

The study included patients recommended to receive 5 years of adjuvant tamoxifen. Inclusion criteria were a histologically proven diagnosis of primary breast cancer (stage I, II, or III), no previous chemotherapy or endocrine treatment other than tamoxifen, no metastatic disease at diagnosis, and hormone receptor positivity (estrogen receptor–positive and/or progesterone receptor–positive) assessed by locally performed immunohistochemistry.

A total of 1580 patients, a consecutively collected retrospective German breast cancer cohort (Stuttgart, Karlsruhe, Erlangen, and Mainz) and prospectively collected patients from the US North Central Cancer Treatment Group (NCCTG) 89-30-52 trial (Mayo Clinic, Rochester, Minnesota), were included. Patients in the German cohort were treated according to standard hospital practice, and inclusion followed the defined inclusion criteria. The NCCTG 89-30-52 trial is a randomized phase 3 clinical trial in postmenopausal women comparing tamoxifen alone vs tamoxifen in combination with the androgen fluoxymesterone, of which patients from the tamoxifen-only group were included.³⁶

Last follow-up was between August and December 2008. The present study included 350 patients from previous reports^{19 - 21} with extended follow-up and extensive genotyping; 219 patients were noneligible (123 had chemotherapy, 60 had metastatic disease, and/or 47 had negative or unknown hormone receptor status or inconsistent follow-up data). Thus, 1361 patients underwent pharmacogenetic analysis. Ethical approval was obtained for all participating institutions. The need for additional informed consent beyond the NCCTG trial was waived by institutional review boards. For the majority of German patients, written informed consent was obtained. Among cases for whom informed consent could not be obtained, inclusion in the study is in agreement with German ethical standards and approved by ethical committees.

The primary objective was to determine the association between CYP2D6 genetic variants and tamoxifen outcome. End points were as follows: (1) Time to recurrence was defined as time from diagnosis or randomization to documentation of a breast event, any local, locoregional, or distant recurrence of breast cancer or a contralateral breast cancer. (2) Event-free survival was defined as time to the first occurrence of a breast event or death from any cause. (3) Disease-free survival was defined as time to first occurrence of a breast event, a second nonbreast primary cancer, or death from any cause. (4) Overall survival was estimated as the time from registration to death from any cause. The secondary objective was to perform an exploratory analysis of clinical outcomes of tamoxifen-treated patients within CYP2D6 genotype subgroups vs the outcome stratification derived from the Arimidex, Tamoxifen, Alone or in Combination (ATAC) 100-month analysis of patients randomized to tamoxifen and anastrozole.²⁹

Genomic DNA was extracted from whole blood (n = 601), fresh frozen tumor (n = 101), and formalin-fixed, paraffin-embedded tumor tissues (n = 659) according to standard procedures. Genotyping of German samples was at Dr Margarete Fischer-Bosch-Institute of Clinical Pharmacology (IKP), Stuttgart, and of NCCTG samples was at Mayo Clinic (eAppendix). The CYP2D6 polymorphisms refer to the CYP Allele Nomenclature Committee (http://www.cypalleles.ki.se) (eTable).

The CYP2D6 pharmacogenetic analysis uses defined genotype-phenotype relationships based on known biochemical and pharmacological effects and included major CYP2D6 alleles within a population of European descent.^{15 - 16} We refer to the following phenotypes: those with poor metabolism (PM), lacking active enzyme function, are homozygous or compound heterozygous for CYP2D6*3, *4, or *5 alleles. Intermediate metabolizers (IM) have reduced enzyme activity and carry *10 and *41 alleles either homozygous or in combination with a PM allele. Since heterozygous carriers of PM or IM alleles (hetEM) have no distinct phenotype, this group was combined with IM to define a patient group associated with intermediate impairment of CYP2D6 activity (hetEM/IM). Patients with extensive metabolism (EM) have normal enzyme function and were characterized by the absence of PM and IM alleles. Ultrarapid metabolizer (UM) patients, with high enzyme activity, have duplicated gene copies without a PM or IM allele.

Clinical and genotyping data were analyzed at the IKP central database. Because outcome effects were previously described to be strongest between PM and EM phenotypes, the sample size was estimated with respect to those 2 patient groups, assuming a ratio of 1 in 10. To detect a hazard ratio (HR) of 1.85 between CYP2D6 phenotypes PM and EM (originally reported by Goetz et al¹⁹ ), with a statistical power of 90% and a 1-sided α level of .05, the number of required events is 166, resulting in a minimum sample size of 1279.

Log-rank tests were used to compare 3 genotype-phenotype states: 2 functional alleles (EM), heterozygous EM or homozygous IM alleles (hetEM/IM), and homozygous PM alleles (PM), and Kaplan-Meier estimates were calculated (patients with the UM phenotype with duplicated gene copies were included in the EM group). Cox proportional hazards modeling with adjustment for tumor size, node status, and histological grade was used to adjust for clinical prognostic factors. In all Cox models, menopause status and subcohort assignment (ie, mode of patient recruitment—retrospective vs prospective) were used as stratum variables to account for population heterogeneity. To decide whether the consideration of CYP2D6 phenotypes improves the prognostic accuracy, Cox models with and without CYP2D6 genotype status were compared by means of concordance probability estimates and analysis of deviance including a likelihood ratio test.³⁷

To model the differences in outcome between AIs and tamoxifen, HRs between CYP2D6 phenotypes in the present study were compared with HRs assigned to tamoxifen in the ATAC trial²⁹ using a 1-sided Mantel-Haenszel log-rank test. Supposing the Cox proportionality hazard assumption, a hypothetical anastrozole survival curve was calculated (eAppendix) based on the HR from the ATAC trial and compared with survival curves of the present study. We used R statistical software, version 2.6.2, including libraries survival and Hmisc (http://www.r-project.org), as well as SPSS, version 16 (SPSS Inc, Chicago, Illinois). Unless stated otherwise, all statistical tests were 2-sided and statistical significance was defined as P < .05.

Pharmacogenetic analyses were performed for 1325 of 1361 eligible patients for whom genotypes could be assigned. Baseline and tumor characteristics were similar among CYP2D6 phenotype subgroups (Table 1). Overall median age at diagnosis was 66 years and median follow-up was 76.1 months.

Numbers of CYP2D6 genotype frequencies (given as homozygous EM/heterozygous/homozygous variant with allele frequencies in parentheses) were as follows: CYP2D6*3: 917/31/1 (1.7%); CYP2D6*4: 854/400/71 (20.0%); CYP2D6*5: 646/33/0 (2.4%); CYP2D6*10: 1175/40/4 (2.0%); and CYP2D6*41: 1080/206/15 (9.0%). For technical reasons, gene duplication (UM) and deletion of allele *5 could be assessed only in blood- or fresh tumor–derived samples (n = 679) but not in paraffin-derived DNA samples. The prevalence of the UM phenotype was 2.3% (20/679), and prevalence of deletion of allele *5 was 4.9% (33/679). Among the 1325 patients with genotyping were 609 EM (45.9%), 637 hetEM/IM (48.1%), and 79 PM (5.9%).

There was a variant-allele dose-dependent increase of the incidence of recurrence and death events among CYP2D6 phenotypes. Patients with EM had the lowest event probabilities, followed by hetEM/IM patients, and PM patients had the highest rates (Table 2). Kaplan-Meier estimates showed significantly higher breast cancer event rates in patients with reduced or absent CYP2D6 function vs EM patients (time to recurrence: P < .001; event-free survival: P = .003; and disease-free survival: P = .005) (Figure 1). The recurrence rates at 9 years of follow-up were 14.9% for EM, 20.9% for hetEM/IM, and 29.0% for PM. The unadjusted HR for PM was highest at 2.12 (95% confidence interval [CI], 1.28-3.50; P = .003) and for hetEM/IM was somewhat less at 1.49 (95% CI, 1.12-2.00; P = .006) (Figure 2). High enzyme activity in UM patients was associated with nonsignificantly fewer recurrences vs patients with 1 to 2 functional CYP2D6 gene copies. (eFigure).

The effect size for impaired CYP2D6 metabolism was greater in the prospectively collected group (HR, 3.47 [n = 212]) than in the retrospectively collected cohort (HR, 1.88 [n = 1113]); however, there was no statistical difference between the HRs (P = .15). We further evaluated the effect of decreased CYP2D6 metabolism within each prognostic subgroup and demonstrated that the HR was consistently greater than 1 within each subgroup (Figure 2), with no significant interactions being detectable.

Because tumor size, node status, and histological grade were independently associated with outcome, we estimated gene variant–associated risks by Cox proportional hazards models adjusted for prognostic factors, conditional on menopausal and subcohort status. The adjusted HR for time to recurrence in PM relative to EM was 1.90 (95% CI, 1.10-3.28; P = .02) and was approximately half of this effect size in hetEM/IM patients (HR, 1.40; 95% CI, 1.04-1.90; P = .03). Decreased CYP2D6 activity was associated with worse event-free survival (HR, 1.33; 95% CI, 1.06-1.68; P = .01) and disease-free survival (HR, 1.29; 95% CI, 1.03-1.61; P = .02). Genotypes did not show a significant effect on overall survival (Table 3 and Figure 2). For all end points, we compared the Cox models that incorporated CYP2D6 with corresponding models without CYP2D6 genotypic information. Analysis of deviance and likelihood ratio tests showed that the predictive accuracy of the Cox models for time to recurrence (deviance, 7.51; P = .02), event-free survival (deviance, 5.1; P = .01), and disease-free survival (deviance, 6.19; P = .02) improved significantly with the incorporation of CYP2D6 genotypic information. Concordance probability estimates were 0.658 (model with genotypic status) vs 0.648 (model without genotypic status) for time to recurrence, 0.646 vs 0.639 for event-free survival, and 0.642 vs 0.636 for disease-free survival. For overall survival, no significant increase of discriminatory power by genotypic information was found (P = .31).

Patients with moderately impaired enzyme activity (hetEM/IM) had similar time-to-recurrence Kaplan-Meier survival curves vs unselected patients (Figure 1A). In contrast, recurrence risk for EM was lower vs unselected patients (P = .03) and was markedly increased for PM (P = .02). In an analogous way, Kaplan-Meier curves for CYP2D6-associated event-free and disease-free survival were compared with the entire cohort (Figure 1, B and C). Compared with the entire cohort unselected by genotype, after 9 years the absolute risk of CYP2D6 PM was increased by 10.7% (time to recurrence) and 6.4% (event-free survival), and was reduced for EM patients by 3.4% (time to recurrence) and 3.3% (event-free survival).

We evaluated whether CYP2D6 genotype stratification of tamoxifen outcome in the present study was more pronounced than the difference between tamoxifen vs anastrozole in the ATAC trial. The absolute difference in the proportion of patients with relapse between EM and PM was 6.7% at 5 years and 13.4% at 9 years. This effect size corresponds to an unadjusted HR of 2.12, which was significantly larger than the HR assigned to tamoxifen relative to anastrozole of 1.31 (P = .01) (Table 4).²⁹ Supposing the proportional hazard assumption, a hypothetical survival function for anastrozole was calculated on the basis of a recurrence HR of 0.76 (1/1.31) for anastrozole vs tamoxifen²⁹ and the Kaplan-Meier estimate for time to recurrence of the entire cohort from the present study. This hypothetical AI curve was plotted onto the survival curves of the CYP2D6 phenotypes (Figure 3). The hypothetical anastrozole survival function is located within the 95% CI of the Kaplan-Meier estimate for CYP2D6 EM patients, suggesting similar outcomes for patients treated with an AI vs patients with 2 functional CYP2D6 alleles receiving tamoxifen.

This study, for the first time to our knowledge, provides sufficiently powered evidence for an association between CYP2D6 genetics and clinical outcome of tamoxifen. Since tamoxifen is standard of care for premenopausal women with estrogen receptor–positive breast cancer, and tamoxifen as well as AIs are valid treatment options for postmenopausal patients, our findings provide a powerful argument for refined endocrine treatment. Unequivocal conclusions have not been possible to date because there have been contradictory reports with respect to this association^{25 - 27} and data from prospective clinical trials are lacking. To address this shortcoming, we took advantage of accessible data and biological materials from patients with long-term follow-up. We defined strict inclusion criteria to exclude confounding effects from unclear hormone receptor status and combinatorial chemotherapy. We recruited a large number of patients to reach sufficient statistical power and compensated for sampling heterogeneity known to cause bias in retrospective studies.

Patients lacking CYP2D6 enzyme function (PM) had an almost 2-fold increased risk of developing breast cancer recurrence compared with patients with 2 functional CYP2D6 alleles (EM). The lower effect in patients with intermediate impairment of enzyme function (hetEM/IM) is indicative of a variant-allele dose-dependent associated risk and, thus, underscores the primary role of CYP2D6-mediated tamoxifen activation to its active metabolite; ie, endoxifen.^{6 ,12 ,18} A relationship between the number of functional alleles and clinical outcome is further evident from CYP2D6 UM phenotypes, accounting for 2.3% of all patients, who may have the best tamoxifen outcome.

CYP2D6 is not known to be associated with breast cancer biology but is thought to be a predictive factor specific to tamoxifen treatment outcomes. While a direct comparison between tamoxifen-treated and non- or AI-treated patients was not possible within the context of this study, several lines of evidence suggest that the observed association between the constitutional CYP2D6 genotype and tamoxifen outcome is a valid hypothesis. In our recent study, a control group of 280 patients not treated with tamoxifen showed that CYP2D6 genetic variants had no effect on the outcome.²¹ Together with the strong pharmacological evidence of CYP2D6 being a key enzyme in tamoxifen bioactivation,^{6 ,12} the observed association in the present study is best explained in relation to tamoxifen rather than treatment outcome per se. This is supported by subgroup analyses and analysis of deviance in Cox models indicating a role of CYP2D6 to predict tamoxifen outcome independent of prognostic factors.

Although the clinical end points of event-free and disease-free survival showed lower effect sizes than in previous smaller studies,^{20 - 21 ,23} the gain of statistical power stresses the significance of defective CYP2D6 activity for nonfavorable tamoxifen outcome. Overall survival was not significantly associated, but a median follow-up of 6.3 years prevents this study from being conclusive because of the limited number of deaths or more common non–breast cancer deaths.³⁸

Our findings were not shown in 3 other studies.^{25 - 27} This discrepancy can be explained by a number of reasons. As demonstrated by our sample size calculation, the number of patients/events required for CYP2D6 analysis is much higher than in any of the previously published studies. The reason why Goetz et al^{19 - 20} and Schroth et al²¹ observed significant associations between PM status and poor outcome may be explained by the prospective patient recruitment^{19 - 20} and the comprehensive coverage of PM and IM predisposing alleles.²¹ In contrast, the patient cohort characteristics of other studies^{25 - 27} appear to be different with respect to treatment modes, estrogen receptor status, and CYP2D6 genotyping. Of note, an incomplete coverage of PM genotypes likely results in missing up to 30% of PMs in a given patient cohort.³⁹

Recently, a modeling analysis questioned the superiority of AI treatment to tamoxifen³¹ based on assumptions made from the CYP2D6*4 variant and assumed effect sizes from a small data set.¹⁹ Based on a more comprehensive CYP2D6 phenotype definition, our data fit well with the model predictions by Punglia et al.³¹ Accordingly, EM patients have the best benefit from tamoxifen compared with the unstratified tamoxifen cohort, with an absolute difference in recurrence risk of 3.4% at 9 years. While Punglia et al used 5-year follow-up comparisons from the Breast International Group Trial 1-98 data,³¹ we performed a 9-year follow-up comparison between our data and results from the ATAC study.²⁹ We superimposed a hypothetical survival curve for AIs based on the reported HR of 0.76 for time to recurrence between anastrozole and tamoxifen²⁹ onto the time-to-recurrence curves for tamoxifen of the present study. Of note, within this comparison, the outcome of the hypothetical AI curve is within the 95% CI of the CYP2D6 EM survival curve. For CYP2D6 EM, which accounts for 46% of patients, this suggests that CYP2D6 stratification attains the therapeutic benefit of AIs. Notably, we regard this approach as hypothesis-generating, yet this has been the best option, considering that a direct comparison of recurrence rates between both studies has not been possible. In support of the strong signal inherent to CYP2D6 stratification and by use of a statistical test independent of modeling, we demonstrated that the effect size of tamoxifen outcome between PM and EM patients in the present study was significantly larger than the HR assigned to tamoxifen relative to anastrozole in the ATAC study (Table 4). This amounts to an absolute risk increase of greater than 10% for PM patients compared with unselected tamoxifen patients, a finding that has implications for many thousands of women each year. Final conclusions must await confirmation from pharmacogenetic analysis in the studies comparing tamoxifen with AIs.

There are potential limitations to our study. The retrospective patient recruitment may have been prone to suboptimal documentation of events, insufficient control of adherence, lack of data regarding the coprescription of CYP2D6 inhibitors with tamoxifen, and differences in the length of follow-up. Also, because of difficulties inherent in the analysis of paraffin-derived DNA, we may have missed 15% to 20% of expected PM phenotypes, as it was apparent from comparison with PM frequency in blood-derived DNA. Although these limitations must be appreciated, our final conclusions are likely to have underestimated the effect size rather than overestimated the CYP2D6-related outcome effect. Importantly, we demonstrated by subgroup cross-comparison between prospectively and retrospectively collected patients that a CYP2D6 PM associated risk of breast cancer recurrence was detectable irrespective of patient sampling schemes. However, different degrees of effect size as well as large CIs indicate that both follow-up time and sample size are critical factors in the assessment of robust CYP2D6-associated risk calculations, once more underscoring the need for high-powered data sets and possibly explaining spurious results in underpowered previous studies.

Our findings indicate that CYP2D6 PM patients have a substantially higher risk of tamoxifen treatment failure. Our analysis indicating that EM patients treated with tamoxifen had an outcome similar to patients treated with AIs in the ATAC trial should provide new impetus to the medical and scientific community to revisit the issue of relative efficacy of these 2 approaches in women with early breast cancer. Genotyping has the potential for identification of women who have the CYP2D6 PM phenotype and for whom the use of tamoxifen is associated with poor outcomes, thus indicating consideration of alternative forms of adjuvant endocrine therapy.