CYP2D6 Genotype and Tamoxifen Activity: Understanding Interstudy Variability in Methodological Quality

Introduction

There has been great controversy over the years regarding the impact of CYP2D6 polymorphisms on the efficacy of tamoxifen in women with breast cancer. The most significant publication to date is the 2009 publication from investigators in Stuttgart and the Mayo Clinic, a 1325 patient study published in JAMA.¹ Despite this high-impact publication, there has been variable acceptance of these results due to inconsistent replication of these findings.²

The controversy

Are the studies demonstrating an association of CYP2D6 genotype with tamoxifen efficacy false-positive studies? We acknowledge that false-positive studies are endemic in the pharmacogenomic literature, generally due to failure to correct for multiple testing of associations between a large number of candidate polymorphisms and multiple phenotypes. However, this concern is not applicable to the study by Schroth and colleagues, or most of the other positive studies that have focused exclusively on testing of a single polymorphic gene, CYP2D6. Furthermore, although some of these positive studies have modest sample sizes and/or incomplete genotyping, this does not impact the probability that a positive study is a true-positive (but would increase the probability of a false-negative study).

Or, are the studies that failed to demonstrate such an association false-negative studies? If so, what would be the explanation? False-negative replication studies can occur for many reasons. One often focuses on statistical explanations for failure of replication. For example, studies may be too small to detect the hypothesized effect, or the true effect may be less than was previously observed, a phenomenon known as “the winner's curse”. Even large studies may fail to replicate due to statistical issues, as the type 2 error can never be zero. However, we believe that the multiple negative studies of CYP2D6 and tamoxifen exemplify more fundamental issues in pharmacogenomics particularly relevant to studies in cancer patients.

The pharmacology

In assessing this body of literature, it is appropriate to start with the pharmacology of tamoxifen, an antiestrogen agent with decades of experience. Tamoxifen undergoes extensive metabolism by multiple P450 enzymes, producing two major active metabolites, 4-hydroxytamoxifen and endoxifen, is the latter primarily formed by CYP2D6. In patients, endoxifen exposure is associated with CYP2D6 genotype, and has also been associated with dextromethorphan phenotype (AUC after a 30 mg dextromethorphan dose).^3,4 Thus, there is little dispute that CYP2D6 activity affects endoxifen exposure at a fixed tamoxifen dose.

However, there has been debate regarding the importance of endoxifen, since it is not the only active metabolite of tamoxifen. Many authors appear to be unaware of another large study, published in this journal in 2011 by Madlensky and colleagues.⁵ Using archived serum and leukocyte DNA samples from 1370 patients from the Women's Healthy Eating and Living (WHEL) study, a clear association between low endoxifen (lowest quintile) and recurrence was demonstrated. No such association was found between low concentrations of tamoxifen or the other two metabolites assayed (4-hydroxytamoxifen, N-desmethyltamoxifen). Furthermore, patients who were genotypically poor metabolizers had a 44-fold risk (76% probability) of being in the low endoxifen quintile, although this genotypic subset included only 23% of all patients with low endoxifen. (Although CYP2D6 genotype itself was not correlated with recurrence, this was a highly heterogeneous patient population and there was no attempt to restrict or correct for concomitant CYP2D6 inhibitors and/or adherence.)

Thus, there is strong mechanistic support for an association between CYP2D6 genotype and tamoxifen efficacy. There is a clear relationship between CYP2D6 genotype and activity, and a demonstrable relationship between the CYP2D6 metabolite (endoxifen) and tamoxifen efficacy.

Genotyping challenges

In conducting a replication study, one must carefully plan genotyping, phenotyping, and the resulting statistical analysis. We will not discuss the statistical issues here, but will focus on the former two issues. The first issue is the source of DNA. Whereas geneticists routinely collect germline DNA (from leukocytes or buccal swabs), cancer researchers have often relied on tumor tissue banks associated with clinical trial data. Although most investigators use these tissues to analyze somatic mutations and gene or protein expression, some investigators have used these samples as a putative source of germline DNA, without consideration of all of the potential pitfalls of this approach.

The use of tumor DNA to assay germline variants is problematic for several reasons. First, it is well known that a significant proportion of cancer cells have extensive chromosomal aberrations. Second, such samples are often formalin-fixed, generally precluding the use of array-based approaches to genotyping. In addition, it is difficult to detect copy number variation, particularly the CYP2D6*5 deletion, using such samples. Given that this deletion variant has an allelic frequency of up to 5%, this omission will result in misclassification, as patients will falsely be considered homozygous (usually extensive metabolizers). However, the use of breast cancer samples to genotype CYP2D6 has additional challenges, since CYP2D6 lies in a region on chromosome 22q13 that is commonly deleted in breast cancer. Thus, the assayed sample may not be representative of the germline, resulting in significant deviations from Hardy-Weinberg equilibrium (HWE).² Although modest deviation from HWE would be expected if the CYP2D6*5 deletion is not considered, massive deviation (e.g., p < 10^-91, the calculated value for one of the recent studies)² from HWE must be due to somatic deletions of CYP2D6 and/or genotyping error (e.g., the use of 60 polymerase chain reaction cycles in one recent study)². For example, if one assumes an allelic frequency for CYP2D6*5 of 4%⁶, there would be an apparent loss of 8% of the heterozygotes due to this genotyping error. In contrast, one of the recent studies demonstrates a loss of 33% of the heterozygotes, consistent with somatic deletion of CYP2D6, but not consistent with a 4% allelic frequency of CYP2D6*5.² Indeed, we calculated the probability that a CYP2D6*5 frequency less than 10% would create the magnitude of departure from HWE that is observed in these data is less than 10^-28. (If genotyping for CYP2D6*5 is performed, HWE can be tested using a triallelic test, the details of which are beyond the scope of this commentary.)

Many investigators have also approached CYP2D6 genotyping without full recognition of the complexity of variation at this locus, which may include deletions, amplifications, and hybrids with two adjacent highly homologous pseudogenes. Thus, it may even be necessary to use nested PCR to avoid nonspecific amplification prior to genotyping some variants, including the SNP most commonly genotyped to detect CYP2D6*4.⁶ It is also important to use broad allele coverage to reduce phenotype misclassification, as approximately one-third of patients will be misclassified if only CYP2D6*4 is genotyped.⁷ Investigators are encouraged to use well established methodologies, such as commercial platforms used in clinical laboratories,⁸ or to collaborate with geneticists that are familiar with the complexities of CYP2D6 genotyping.

Phenotyping challenges

Pharmacogenomic studies are often conducted by investigators with a background in molecular epidemiology, but not pharmacology. Whereas epidemiologists may be comfortable distinguishing cases and controls, pharmacogenomic studies must carefully analyze drug exposure, as well as any other drugs that also affect the phenotype. In regard to tamoxifen, some investigators have attempted to replicate the original positive studies using retrospective cohorts that have included patients receiving tamoxifen in combination with chemotherapy. Given that the effect of tamoxifen is relatively modest in this setting, it is not surprising that such studies failed to replicate the original positive findings.⁹

Although most investigators have collected and analyzed information regarding other anticancer treatment, there are many prescription and OTC drugs that can inhibit CYP2D6, potentially reducing endoxifen formation, and abrogating the efficacy of tamoxifen. This is particularly problematic in retrospective cohorts, since a complete list of medications is not always available, particularly OTC agents such as diphenydramine. Even if such information were available, the impact of such CYP2D6 inhibitors would likely depend on the time interval between the ingestion of the inhibitor and tamoxifen, data that are very unlikely to have been collected outside of a prospective study.

Quality metrics

Any genomic study needs to pay careful attention to quality control in genotyping, phenotyping, and data analysis (Table 1). Although many have expressed concern regarding false-positive pharmacogenomic studies due to multiple testing, there has been less attention to the risks of false-negative studies due to problematic genotyping and/or phenotyping. As noted above, many of the studies of tamoxifen exemplify these problems.

Table 1. Quality metrics for pharmacogenomic studies.

Element	High	Questionable	Low
Source of DNA	Blood or buccal swab	Peritumoral tissue	Tumor
Genotyping approach	Verified by at least two platforms	Using a single platform with low signal/noise ratio (poor separation of signals)	No attention to quality control
Polymorphisms tested	Well justified based on rationale, and type 1 and 2 errors	Adequate rationale, but no consideration of type 1 and 2 errors	No justification
Hardy-Weinberg Equilibrium	Present	Borderline deviation	Clear deviation or not tested
Possible population stratification	Considered and excluded	No consideration, but population homogenous	Heterogeneous population without appropriate analysis
Phenotyping approach	Prospective	Retrospective, but phenotype unambiguous	Potential ambiguities in retrospective cohort
Drug dosing	Consistent	Inconsistent, but adjusted for in analysis	No analysis of dosing
Concomitant drugs	Interacting drugs excluded prospectively	Retrospective analysis using a comprehensive data set	No analysis of interacting drugs
Data analysis	Careful correction for multiple testing	Exploratory analysis	No attention to multiple testing

Conclusions

The studies reporting an association between CYP2D6 genotype and tamoxifen efficacy cannot be explained by chance, and are most likely true-positive studies. In contrast, there are evident methodological flaws permeating the CYP2D6/tamoxifen literature, including the most recent highly visible studies.² Since the positive association studies are consistent with our understanding of tamoxifen's pharmacokinetics and pharmacodynamics, we recommend that CYP2D6 genotyping be utilized to exclude poor metabolizers from receiving tamoxifen. This recommendation is consistent with the results of a recent study demonstrating the desire of women with breast cancer to have such information, and the impact of such testing on choice of therapy.¹⁰ In addition, we recommend greater attention to methodological issues in pharmacogenomic studies in general (particularly in oncology) by investigators, reviewers, and journal editors.