CYP2C19*2 predicts substantial tamoxifen benefit in postmenopausal breast cancer patients randomized between adjuvant tamoxifen and no systemic treatment

From 1982 to 1994, a randomized clinical trial was conducted in the Netherlands, studying the benefit from adjuvant tamoxifen (IKA-trial) versus no adjuvant therapy [12, 13]. Study data were part of the Oxford meta-analysis [14]. Patients were randomized in a 2:1 ratio between 1-year tamoxifen (30 mg per day) and no adjuvant therapy. Eligible patients were postmenopausal, <76 years of age, and had a T_1–4, N_0–3, M₀ breast tumor [15] with no mastitis or palpable supra- or infraclavicular lymph nodes. After 1 year, for patients in the tamoxifen arm who were on study, a second randomization was performed to receive another 2 years of tamoxifen or to stop further treatment. From 1989, based on two interim analyses showing a significant improvement in recurrence-free survival among lymph node-positive patients, these node-positive patients were all allocated to the tamoxifen treatment arm (i.e., skipped the first randomization). In total, 1,662 patients were included. The patient characteristics and clinical outcome of tamoxifen treatment have been presented elsewhere [13].

We have traced tissue blocks of participating patients and recollected sufficient tumor material of 739 patients who did not differ in prognostic factors from the total group (Table S1). After revision of estrogen receptor α (ERα) status as assessed with immunohistochemistry (IHC), a total of 563 ERα-positive tumors were used for subsequent analysis. The number of patients in each treatment arm of randomization 1 and randomization 2, pre- and post-interim analysis, is shown in Figure S1.

Tissue microarrays (TMAs) were constructed using formalin-fixed paraffin-embedded (FFPE) tumor blocks. The TMAs were stained for ERα, progesterone receptor (PgR), and HER2. ERα and PgR were considered positive when ≥10 % of invasive cells showed nuclear reactivity. This cutoff was chosen because it is common practice in the Netherlands and in addition this would avoid the potential inclusion of basal-like tumors [16] in our analysis. HER2 was considered positive when membranous staining was score 3. In case of a membranous score of 2, chromogenic in situ hybridization (CISH) was performed on whole-tissue slides. For tumors that did not have sufficient cores in the TMA, whole slides were cut and could adequately be assessed for ERα (N = 60), PgR (N = 55), and HER2 (N = 36). Tumor grade was scored on a hematoxylin–eosin (HE)-stained slide using the modified Bloom–Richardson score [17].

From paraffin-embedded tissue blocks, 10-μM-thick sections were cut and attached to microscope slides. A total of 10 slides per tumor were used for DNA isolation. Slides were deparaffinized in xylene, rehydrated, and stained with hematoxylin. The slides were incubated with sodiumthiocyanate overnight. Exact tumor location was circled by the pathologist on a HE-stained slide, which was used as a template. After adding a drop of tissue lysis buffer, tumor tissue was scraped from the slides, added to a 1.5-μl micro-centrifuge tube containing a 200-μl mix of tissue lysis buffer/proteinase K. This tube was incubated in a Thermomixer at 55 °C for 48 h. An additional 27-μl proteinase K (2 mg/μl) was added after 24 and 36 h. After 48 h, the tube was incubated at 80 °C for 10 min to inactivate proteinase K. After centrifuging, the supernatant was pipetted into a new tube. A total of 150 μl was purified using a QIAquick PCR purification kit. DNA isolation was successful for 535 tumor samples.

Genotyping for CYP2C19*2 (681G>A, rs4244285) and CYP2C19*17 (−806C>T, rs12248560) was performed on tumor DNA. The concordance between selected genotyping from FFPE-derived tumor DNA and DNA from serum has previously been established [18]. Taqman allelic discrimination assays were used (Applied Biosystems, Nieuwerkerk ad IJssel, The Netherlands) on an ABI Prism 7500 Sequence Detection system (Applied Biosystems). Each assay consisted of two allele-specific minor groove-binding (MGB) probes, labeled with fluorescent dyes VIC and FAM. The assay IDs are C_25986767_70 (CYP2C19*2) and C_469857_10 (CYP2C19*17). Thermal profile for genotyping consisted of 95 °C for 10 min, followed by 50 cycles of 15 s at 92 °C and 90 s at 60 °C. Genotypes were scored by allele-specific fluorescence using 7500 fast system SDS software (Applied Biosystems).

Recurrence-free interval (RFI) was taken as the time from the date of (first) randomization until the occurrence of a local, regional, or distant recurrence or breast cancer-specific death [19]. Since these CYP2C19 variants are associated with breast cancer risk and the duration of treatment in this trial was relatively short to prevent the occurrence of a secondary breast cancer, patients with a secondary contra-lateral breast cancer were censored at the date of this occurrence (Table S2). In the subset of 563 ERα-positive patients, median follow-up of patients without a recurrence event is 7.8 years. Genotypes were tested for Hardy–Weinberg equilibrium using a Chi square test. The distribution of clinico-pathological characteristics by the CYP2C19*2 and CYP2C19*17 variants was evaluated using Chi square tests. Survival curves were constructed using the Kaplan–Meier method and compared using log-rank tests. To determine whether the benefit from tamoxifen was different in CYP2C19 (one or two *2 alleles vs. no *2 allele and one or two CYP2C19*17 alleles vs. no CYP2C19*17 allele) genotypes, covariate adjusted Cox proportional hazard regression models were constructed with an interaction between the treatment and the genotype. Treatment groups were defined according to the results of the first randomization (1–3 years of tamoxifen versus no adjuvant systemic treatment). The change in randomization that occurred after the interim analysis resulted in an enrichment of lymph node-positive patients in the group of tamoxifen-treated patients. Therefore, Cox proportional hazard regression models were stratified for nodal status. The following factors were included as covariates: age (≥65 vs. <65), grade (grade 3 vs. grade 1–2), tumor size (T3–T4 vs. T1–T2), HER2 status (positive versus negative), estrogen receptor expression (10–99 vs. 100 %), and progesterone status (positive versus negative). No adjustments for multiple testing were performed. To assess the prognostic value of the CYP2C19 genotypes, we performed covariate adjusted Cox proportional hazard regression (including lymph node status) in the subgroup of patients who were randomized to the control arm. This study complied with reporting recommendations for the tumor marker prognostic studies (REMARK) criteria [20] outlined in Table S3.

Adequate genotype data for CYP2C19*2 and CYP2C19*17 were available for 494 and 504 patients, respectively (Fig. S2). For CYP2C19*2, a total of 12 (2.4 %) patients were homozygous carrier and 127 (25.7 %) heterozygous carriers were found. The homozygote CYP2C19*17/*17 genotype was seen in 28 (5.6 %) of the patients, while 151 (30.0 %) were heterozygous. Genotype frequencies of CYP2C19*2 and CYP2C19*17 were in Hardy–Weinberg equilibrium (p = 0.86 and p = 0.06, respectively). A weak association between CYP2C19*2 and low ERα expression was observed (Table 1). A higher frequency of grade III was observed in patients carrying a CYP2C19*17 allele. (Table S4). In general, known prognostic factors were equally divided over the treatment arms for all genotypes (Tables S5, 6), with the exception of lymph node status which can be explained by the change in randomization. In patients without a CYP2C19*17 allele, the tamoxifen arm included more HER2-positive patients than the control arm, while in patients with a CYP2C19*17 allele, the patients in the control arm were younger than the patients in the tamoxifen arm.

When stratified by nodal status, the tamoxifen effect in all 563 ERα-positive patients is 0.54 (95 % CI 0.36–0.83, p = 0.004). In univariate analysis, positive lymph node status was associated with an unfavorable outcome (HR 2.61, p < 0.001). No significant interaction between lymph node status and tamoxifen was observed.

We did not find a significant interaction between the CYP2C19*17 genotype and treatment (adjusted p for interaction = 0.62) (Table S7). However, a significant interaction between CYP2C19*2 genotype and treatment was found (both unadjusted and adjusted p for interaction = 0.04). Patients carrying one or two CYP2C19*2 alleles derived more benefit from tamoxifen than patients with no CYP2C19*2 allele (CYP2C19*2 carriers: adjusted HR = 0.26, p = 0.001; non-CYP2C19*2 carriers: adjusted HR = 0.64, p = 0.18; Figs. 1 and S3; Table 2). Performing a sensitivity analysis, using a 3-level factor for T-stage (T1, T2, and T3–4), did not substantially change these results (Table S8).

Tamoxifen-untreated patients with a CYP2C19*2 variant allele (N = 33) had an unfavorable RFI when compared with tamoxifen-untreated patients without a CYP2C19*2 allele (N = 86). After correcting for the prognostic factors as described in the methods section, CYP2C19*2 remained an independent adverse prognostic factor (HR = 2.77, p = 0.01) (Table S9). We did not find a difference in RFI in control patients with a CYP2C19*17 allele (p = 0.91) (Table S10).