Predicting steady‐state endoxifen plasma concentrations in breast cancer patients by CYP2D6 genotyping or phenotyping. Which approach is more reliable?

Abstract In previous studies, steady‐state Z‐endoxifen plasma concentrations (ENDOss) correlated with relapse‐free survival in women on tamoxifen (TAM) treatment for breast cancer. ENDOss also correlated significantly with CYP2D6 genotype (activity score) and CYP2D6 phenotype (dextromethorphan test). Our aim was to ascertain which method for assessing CYP2D6 activity is more reliable in predicting ENDOss. The study concerned 203 Caucasian women on tamoxifen‐adjuvant therapy (20 mg q.d.). Before starting treatment, CYP2D6 was genotyped (and activity scores computed), and the urinary log(dextromethorphan/dextrorphan) ratio [log(DM/DX)] was calculated after 15 mg of oral dextromethorphan. Plasma concentrations of TAM, N‐desmethyl‐tamoxifen (ND‐TAM), Z‐4OH‐tamoxifen (4OH‐TAM) and ENDO were assayed 1, 4, and 8 months after first administering TAM. Multivariable regression analysis was used to identify the clinical and laboratory variables predicting log‐transformed ENDOss (log‐ENDOss). Genotype‐derived CYP2D6 phenotypes (PM, IM, NM, EM) and log(DM/DX) correlated independently with log‐ENDOss. Genotype‐phenotype concordance was almost complete only for poor metabolizers, whereas it emerged that 34% of intermediate, normal, and ultrarapid metabolizers were classified differently based on log(DM/DX). Multivariable regression analysis selected log(DM/DX) as the best predictor, with patients’ age, weak inhibitor use, and CYP2D6 phenotype decreasingly important: log‐ENDOss = 0.162 ‐ log(DM/DX) × 0.170 + age × 0.0063 ‐ weak inhibitor use × 0.250 + IM × 0.105 + (NM + UM) × 0.210; (R 2 = 0.51). In conclusion, log(DM/DX) seems superior to genotype‐derived CYP2D6 phenotype in predicting ENDOss.


| INTRODUC TI ON
Tamoxifen (TAM) is a selective estrogen receptor antagonist used as adjuvant therapy to prevent estrogen-receptor-positive breast cancer recurrence. TAM is de facto a prodrug because its anti-estrogen activity is 30-100 times less than that of its metabolites Z-4OH-tamoxifen (4OH-TAM) and Z-endoxifen (ENDO). 1,2 ENDO is considered the most effective metabolite in vivo, with plasma concentrations 5-10 times higher than 4OH-TAM. 3 Two well-powered trials found ENDO plasma concentrations correlated with recurrence risk. 4,5 Madlensky et al 4 reported that women with ENDO concentrations >5.97 ng ml-1 (>16 nM) had a 30% lower relative risk of breast cancer recurrence.
Saladores et al 5 found ENDO levels <5.2ng ml-1 (<14 nM) associated with a shorter relapse-free survival (RFS) compared with >13.0 ng ml-1 (>35 nM). Hence the suggestion that monitoring ENDO concentrations can be used to individualize adjuvant TAM therapy. 6,7 An alternative strategy involves measuring predictors of steadystate ENDO levels (ENDOss) before starting TAM therapy. While several enzymes contribute to ENDO formation (CYP3A4/5, CYP2C9, CYP2C19, CYP1A2) and elimination (UGTs, SULTs), the main metabolic pathway is ENDO formation from N-desmethyl-tamoxifen (ND-TAM) by the cytochrome CYP2D6 8 (Figure 1). CYP2D6 activity has been estimated indirectly by combining the several CYP2D6 allelic variants with a different gene expression, 9 or calculated directly from the dextromethorphan (DM)/dextrorphan (DX) urinary metabolic ratio [log(DM/DX)]. 10,11 Both methods can predict ENDOss. CYP2D6 phenotyping is considered superior to genotyping because non-genetic factors like age, drug-drug interactions, or co-morbidities can affect phenotype (phenoconversion phenomenon), 12 but the two methods' performance had yet to be compared directly.
Our primary aim was to ascertain which method -CYP2D6 genotyping or phenotyping -can predict ENDOss more accurately.
The findings presented here are part of an ongoing prospective trial (TAM study) to correlate ENDOss with breast cancer recurrence.

| Patients and study design
This study concerned 203 Caucasian women with estrogen-receptor-positive breast cancer (stage IA 67.2%, IIA 23.2%, IIB 6.8%, and One, 4, and 8 months after starting TAM, blood was sampled before a drug dose to assay plasma concentrations of TAM, ND-TAM, 4OH-TAM, and ENDO. All other routine procedures were completed according to local clinical practice. The study protocol was approved by the Ethics Committee of Rovigo Hospital (Italy) and all participants gave their written informed consent.

What this study adds
• An algorithm including log(DM/DX), patient's age and weak inhibitor use predicts 48% of log-ENDOss variability.
• The model based on log(DM/DX) may identify patients who will have ENDOss <5.97 ng ml-1, and consequently require a higher starting dose of tamoxifen. Tamoxifen and its metabolites were analyzed in patients' plasma using a validated high-performance liquid chromatography (HPLC) method, 13 with partial adaptations. Briefly, blood was centrifuged within an hour of sampling and plasma was stored at −20°C until analysis. One mL of plasma was alkalinized with 1 ml-glycine/ NaOH buffer (1M, pH: 11.3) and extracted with 7 ml of hexane/2-

| CYP2D6 genotyping procedure
Germline DNA was isolated from blood using Wizard Genomic DNA Purification Kit (Promega) according to the manufacturer's recommendations.

| Urinary DM and DX assay
Urinary DM and DX were tested using HPLC according to Flores-Péres et al, 22

| CYP2D6 phenotyping procedure
The logarithm of the ratio of urinary DM to DX molar concentra-

| Statistical analysis
In the tables, continuous variables are presented as means ± standard deviations (unless otherwise stated), and categorical variables as absolute numbers and percentages.
Continuous variables with a normal distribution were compared with Student's t test. One-way ANOVA was used for comparing more than two independent groups, followed by Bonferroni posthoc tests for pairwise comparisons, and the test for linear trend, as needed. Two-way ANOVA was used to compare repeated measures from the same patient. The homoscedasticity assumption was verified with the Bartlett and Levene test. Equivalent non-parametric tests (the Mann-Whitney U, Kruskal-Wallis and Friedman tests) were used whenever applicability conditions were not met. Categorical data frequencies were examined using Pearson's chi-square and Fisher's exact test, as appropriate.
ENDOss was calculated as the mean of ENDO concentrations at 4 and 8 months, if both measurements were available, or at 4 months otherwise. A log-transformation was applied (log-ENDOss) to achieve a normal distribution of ENDOss.
The following independent variables were used in the regres- First, univariable linear regression analyses were run, taking one independent variable at a time. Then, a stepwise multivariable forward regression was conducted (P < .05 for variable inclusion, and P < .15 for variable removal) to select the best log-ENDOss prediction model. Multicollinearity was checked using the tolerance and the variance inflation factor (VIF); variables with a tolerance <0.4 (VIF > 2.5) were discarded from the analysis.
Possible confounding effects were investigated with all variables excluded by the stepwise selection. Residuals analysis was performed to examine the models' goodness of fit and adherence to the regression assumptions. The validity of the final model was assessed by measuring the R 2 coefficient and the mean absolute and percentage errors (MAE and MAPE) of the ENDOss predicted.
Pearson's correlation coefficients (r) were used to test the relationship between log(DM/DX) and log-ENDOss, and between log(DM/DX) and log(NDT/ENDO).

| Patients' characteristics
Of the population of 203 women, only 164 were suitable for multivariable regression analyses as all the independent variables were available. Table 1 summarizes patients' characteristics for the whole group and for the regression group. No patients were taking strong CYP2D6 inhibitors, while 12 in the whole group and 9 in the regression group were taking weak inhibitors (citalopram, sertraline, duloxetine, venlafaxine). There were no statistically significant differences between the variables considered in the two groups, except for age (P = .0083), and menopausal status (P = .023).

| Log(DM/DX) and CYP2D6 phenotype
One-way ANOVA followed by testing for linear trends showed a significant difference in the mean log(DM/DX) values across groups
The stepwise multivariable regression analysis identified log(DM/ DX), patient's age, weak inhibitor use, and CYP2D6 phenotype as significant independent predictors of log-ENDOss, ruling out BMI:

| D ISCUSS I ON
This study showed that the co-variables log(DM/DX), age and weak inhibitor use, and CYP2D6 phenotype correlated significantly with log-ENDOss on multiple regression analysis, explaining 51.0% of log-EN-DOss variability. The best predictor was log(DM/DX) (partial R 2 = 0.39), with the contributions of age (partial R 2 = 0.051), weak inhibitor use (partial R 2 = 0.035), and CYP2D6 phenotype (partial R 2 = 0.032) decreasingly important (Table 3). When CYP2D6 phenotype was removed from the regression, the overall R 2 was marginally lower (0.48), whereas removing log(DM/DX) resulted in a greater decrease in R 2 (0.41).
These results were not unexpected because the expression of showed that downgrading CYP2D6*10 activity score from 0.5 to 0.25 improved ENDOss prediction, so new guidelines have recently been updated. 21 In short, phenoconversion and activity score misclassification can both weaken the predictive power of CYP2D6 genotype. On the other hand, genotype is stable for life, whereas DM levels may change due to intervening, non-genetic factors (Table 7).

Several clinical studies investigated the correlation between
CYP2D6 genotype-derived activity and ENDOss, using various methods and phenotyping criteria, and with mixed results. In general, individuals labelled as PMs have significantly lower ENDOss  Incidentally, our phenotyping method based on the urinary log(DM/DX) ratio was validated by comparison with the reference debrisoquine test 10 and found to correlate with the partial metabolic clearance of DM to DX. 11 A good correlation was also demonstrated in our population between urinary log(DM/DX) ratio and plasma log(ND-TAM/ENDO) ratio, which is another measure of CYP2D6 activity ( Figure 5). Notably, Saladores et al 5  Whatever the method used, patients whose predicted ENDOss concentration is lower than the threshold of 5.97 ng mL-1 should start therapy with doses >20 mg. Assuming a linear dose-concentration relationship, the dose should be increased by the threshold-to-predicted-concentration ratio and rounded up to 30 or 40 mg. Should higher (off-label) doses be required, aromatase inhibitors may be an alternative option, since little is known about the long-term safety of higher doses of TAM. 36 Adequately-powered prospective trials are obviously needed to test this strategy.
Our study has some limitations. First, we assumed that all patients adhered to their TAM treatment. Though we could not prove as much, it is reasonable to assume a good compliance at the start of the therapy. The long half-lives of TAM and its metabolites should also guarantee stable ENDO concentrations even if a TAM dose is missed occasionally. Second, we did not genotype all the known CYP2D6 variants, but only those most common in Caucasians, so our results cannot be extended to other ethnicities. Third, ENDOss are reportedly 20% lower in winter than the mean year-round levels. 40 Our study covered a period of 8 months, so ENDOss may have been influenced by seasonal changes. That said, a post-hoc analysis of our data (not shown) found no differences in ENDOss measured in January-March versus July-September. Fourth, the results of urinary DM testing may be affected by changes in urinary pH and renal function, thus leading to misphenotyping CYP2D6. Although we cannot exclude this possibility, such a bias can be minimized by collecting urine over a long period (10 hours), as we did. The log(DM/DX) ratio also correlated strongly with the log(ND-TAM/ENDO), which more closely reflects ENDO production by CYP2D6.

| CON CLUS IONS
Our study found that phenotyping CYP2D6 activity by means of a urinary DM test is the single best predictor of ENDOss. A model including log(DM/DX), patient's age, and use of CYP2D6 inhibitors has an acceptable predictive performance, and could be used as an alternative to genotyping tests. Despite some uncertainty regarding the optimal ENDOss, a therapeutic approach that aims at personalizing TAM dosage early on is worth testing in a prospective trial.

CO N FLI C T O F I NTE R E S T S
None of the authors have any competing interests to disclose.

AUTH O R S' CO NTR I B UTI O N S
MG, FP, NM, and RP contributed to study conception and design.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data analyzed in this study are available from the corresponding author on reasonable request.