Tamoxifen (TAM) is a widely prescribed antiestrogen for the control of estrogen receptor (ER)-positive breast cancer, yet its efficacy is reduced due to the development of endocrine resistance and intrinsic patient characteristics that prevent drug response. The latter has been partially attributed to a lack of TAM bioactivation toward its active metabolite, endoxifen. Pharmacological and pharmacogenetic evidence strongly support that in vivo endoxifen formation is mainly mediated from the primary metabolite N-desmethyl-TAM by the cytochrome P450 2D6 (CYP2D6) enzyme (Stearns et al., 2003; Desta et al., 2004). As distinct genetically determined functional variants are present in the general population, inter-patient variability of plasma endoxifen is expected to be predictable, at least in part, by CYP2D6 (de Vries Schultink et al., 2015).

The CYP2D6 polymorphism with more than 100 known alleles contributes to inter-individual differences in enzyme activities and plasma exposure of metabolized drugs and are commonly grouped into four CYP2D6 metabolizer phenotypes: ultra-rapid (UM), extensive (EM), intermediate (IM), and poor (PM) metabolizers. Traditionally, these have been defined using probe substrates, however, due to probe drug differences to derive phenotypes, CYP2D6 genotyping has emerged as the method of choice to predict enzyme activity (Hicks et al., 2014). Activity scores (ASs) of 0, 0.5, and 1 for null (PM), reduced-function (IM), and fully-functional (EM) alleles, respectively, have been used to infer metabolizer phenotypes from diplotypes (Gaedigk et al., 2008). Of note, there is no universally accepted method to assign allele ASs that reflect metabolic activity across all enzyme substrates. In particular, this refers to the activities of reduced-function alleles ^∗9, ^∗10, ^∗17, ^∗29, and ^∗41, of which ^∗10 has been suggested to have a more deleterious effect on enzyme function compared to the remaining IM alleles (Steimer et al., 2004; Shen et al., 2007). A recent literature review across various CYP2D6 substrates suggested that a ^∗10 AS downgrade from 0.5 to 0.25 may better reflect the functional impairment in ^∗10 defined IM individuals (Hicks et al., 2014).

Prediction of an impaired TAM metabolizer phenotype (IM, PM) with low endoxifen formation capacity is potentially important for personalized treatment decisions in breast cancer such as increasing the therapeutic TAM dose or replacing TAM with an aromatase inhibitor (AI). Although prospective data that demonstrate a clinical benefit of dose adjustment are lacking, a retrospectively defined clinical threshold of 5.9 ng/mL (15.8 nM) plasma endoxifen separated patients below this cutoff into those with reduced clinical benefit from the remainders (Madlensky et al., 2011; Saladores et al., 2015), suggesting a clinical relevance of predicting endoxifen formation capacity. Since routine therapeutic endoxifen monitoring is not standard clinical practice, genotyping has been put forward and tested in several studies as a prospective tool to select patients for TAM dose escalation or to establish its predictive value (Irvin et al., 2011; Kiyotani et al., 2012; Dezentjé et al., 2015; Hertz et al., 2015; Fox et al., 2016). In the absence of standardized guidelines (Hicks et al., 2014), studies used different phenotype assignments including that based on CYP2D6-dependent codeine metabolism (Crews et al., 2014). As a consequence, low endoxifen predictability from codeine-specific CYP2D6 phenotype assessment in a recent study (Fox et al., 2016) led to recommendations against the use of CYP2D6 genotype to guide clinical decisions (Hertz and Rae, 2016a).

While the combination of multi-locus genotypes into diplotypes based on the AS system (Gaedigk et al., 2008) appears straightforward, their attribution to a specific metabolizer phenotype has been inconsistent preventing meaningful clinical conclusions. Here, we used different metabolizer phenotype definitions to test the power of CYP2D6 diplotype and phenotype-based prediction of impaired endoxifen metabolism with the goal to provide a robust algorithm toward the standardization of CYP2D6 in personalized endocrine treatment.

We re-evaluated a comprehensive data set of CYP2D6 genotypes and TAM metabolite concentrations of breast cancer patients treated with adjuvant TAM to assess the prediction of impaired TAM metabolism by CYP2D6. We applied the power of diplotype-based assignments (Gaedigk et al., 2008) to further refine the discriminatory value of metabolizer phenotype as the most intuitive concept to interpret CYP2D6 polymorphism. To shed light on current controversies on the utility of CYP2D6 for TAM efficacy prediction (Ratain et al., 2013; Brauch and Schwab, 2014; Hertz and Rae, 2016b) standardized genotype–phenotype relationships for the validation of an association between CYP2D6 and impaired TAM metabolism are mandatory.

Currently, the extent to which CYP2D6 determines the up to 20–30 fold (Mürdter et al., 2011) inter-patient variability of plasma endoxifen under standard TAM treatment is poorly characterized. On the assumption that variable TAM metabolism is prognostic for a patient’s response to treatment, it can be argued that drug level monitoring of endoxifen in the first months after treatment start would be straightforward. Yet, since CYP2D6 genotyping requires only a standard molecular biology laboratory and can be used to guide upfront treatment decisions, this approach has been put forward and tested for its predictive value to select patients for TAM dose interventions. Of note, the lack of standardized guidelines to deduce phenotypes from genotype led to the use of a CYP2D6 phenotype classification scheme for impaired Tam metabolism (Fox et al., 2016) that was previously recommended for codeine metabolism (Crews et al., 2014). However, it has become increasingly clear that CYP2D6 variants may exert substrate-dependent effects (Bogni et al., 2005; Gaedigk et al., 2008; Zhou, 2009), and therefore, diplotype specific phenotype data obtained with codeine cannot be extrapolated to other CYP2D6 substrates such as TAM. This functional discrepancy may have important clinical implications as others, based on the inappropriately used codeine scoring concluded that CYP2D6 has no value for the prediction of TAM metabolism (Hertz and Rae, 2016a). Our re-evaluation of existing pharmacogenetic data challenges these findings.

We showed that plasma endoxifen prediction highly depends on the phenotypical grouping of CYP2D6 variant alleles and on the choice of metabolite readout, i.e., absolute metabolite concentrations versus metabolic ratio E/DMT. CYP2D6 diplotypes were superior in predicting endoxifen variability, independent of ethnicity. This was less pronounced when diplotypes were collapsed into fewer class levels (phenotypes), indicating that the effects of functional variants are maximally exploited by a score reflecting the number of null- or reduced-activity haplotypes such as diplotype grouping or AS. Importantly, the codeine-specific phenotype grouping (Crews et al., 2014) poorly predicted CYP2D6-based endoxifen formation. In particular, it was inferior in Asians (median R² < 20%) most likely due to a misclassification of abundant IM/IM (^∗10) diplotypes as EM. From this it follows that CYP2D6 variants act differentially on TAM and codeine substrates, underscoring the need for substrate-specific CYP2D6 genotype–phenotype assessments (Hicks et al., 2014). Within this context, the suggested extra deleterious effect on enzyme function of ^∗10 compared to other IM alleles (Shen et al., 2007; Hicks et al., 2014) was addressed by downgrading its phenotypic activity and by placing ^∗10 homozygous patients together with IM/PM diplotypes into a new phenotype category of slow metabolizers. A moderate increase of explanatory power (TAM4) by approximately 5% compared to the best explaining phenotype that did not incorporate a ^∗10 downgrade (TAM2) supports the notion of an increased deleterious effect of ^∗10 compared to other IM alleles on reduced TAM metabolism. However, given the absence of such an effect in non-Caucasians with a relatively small effect size in Caucasians, the significance of downgrading ^∗10 for an improved prediction of TAM metabolism must be replicated in larger cohorts and/or meta-analyses.

Our study showed that the metabolite endpoint closely linked to CYP2D6 activity is active metabolite-to-precursor ratio, as more than two thirds of the variability (median R²: 68–82%) of E/DMT was explained by CYP2D6 diplotype. Notably, a portion of unexplained variability in this study may be related to CYP2D6 alleles that were either not accounted for, e.g., hybrid alleles, or were only partially captured (^∗2A and other ^∗2 alleles contributing to EM were genotyped only in Asians), or depend on cis/trans-regulatory regions with an influence on CYP2D6 expression that are not yet integrated in biomarker panels. However, the effect of undetected hybrid genes on phenotype is more related to gene duplications and therefore minor (Black et al., 2012), and outliers that strongly differed to expected CYP2D6 function were dispersed over several phenotypic categories (AS ≥ 1) rather than being limited to EM. Therefore, it is plausible to postulate the existence of additional genetic loci, that, similar to a previously described enhancer (Wang et al., 2013, 2015) influence CYP2D6 expression in a modest number of patients, a hypothesis which needs to be further investigated. Moreover, plasma endoxifen variability was predictable to a lesser extent (39–58%) compared to the highly CYP2D6-dependent E/DMT. Therefore, factors other than CYP2D6 genetics may account for the unexplained portion of endoxifen variability such as non-compliance, CYP2D6 inhibitor use, environmental factors, and other cytochrome P450 isoenzymes including CYP3A phenotype (Teft et al., 2013; ter Heine et al., 2014). Although our genetic model did not incorporate CYP3A4 activity predicted by the ^∗22 variant (Wang et al., 2011; Teft et al., 2013; Antunes et al., 2015), fluctuations in plasma levels of endoxifen precursors (4OH-TAM, desmethyl-TAM) were accounted for by adjusting for CYP2C9 and CYP3A5 variants, while strong CYP2D6 inhibitor use was low to absent in the majority of patients. Thus, our E/DMT-based translations of diplotypes into metabolizer phenotypes TAM4 and TAM2 with or without downgrading of ^∗10 activity, respectively, capture most of the variability attributable to CYP2D6 and are superior to previous CYP2D6 metabolizer assignments such as TAM1 (Irvin et al., 2011; Hertz et al., 2015) and the codeine score (Fox et al., 2016).

Although our study does not provide direct data for clinical outcome prediction, an endoxifen threshold concentration of 5.9 ng/mL useful to predict breast cancer recurrence risk during TAM therapy (Madlensky et al., 2011) is predictable by CYP2D6. Overall, a genotyping test discriminating patients based on an AS cutoff of 0.5 revealed a greater than 95% sensitivity to predict whether patients will have beneficial endoxifen concentrations. This simplified test interpretation may help clinicians to reassure their upfront treatment decisions, i.e., standard TAM above threshold, versus TAM dose adjustment or AI choice around and below threshold. Given our observation of a moderate specificity, e.g., in Caucasians, an approach of maximizing specificity at the cost of decreased sensitivity to avoid the risk of false-positive CYP2D6 testing could be a strategy which needs further investigation based on independent cohorts. Moreover, the increased FDR in Caucasians (11%) compared to Asians and Middle-Eastern Arabs (2%; Table 2) point to a modest proportion of TAM treated patients tested positive but having sub-therapeutic endoxifen concentrations, which would bear a risk of undertreatment. Together with the fact that genotyping does not provide information on treatment adherence, a combination of upfront genotype-informed treatment allocation followed by therapeutic blood monitoring for active metabolite concentrations therefore appears promising to personalize TAM treatment use.

In summary, we provided an improved algorithm to predict CYP2D6-dependent impaired TAM metabolism from genotype underscoring its essential role in drug bioactivation toward endoxifen independent of age and ethnicity. Clinical evaluation of standardized TAM-specific CYP2D6 activity assignments may shed new light on linking impaired endoxifen formation with TAM outcome prediction aiding the selection of patients for TAM dose increase or AI treatment.

WS, SW, TM, MS, and HB designed and performed the study. DE, BE, BC, CK, AT, and NZ recruited study participants and provided specimens for analyses. WS, SW, TM, ES, ME, MS, and HB are responsible for data interpretation, critical review, and manuscript preparation. All authors approved the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.