Estrogen Receptor Alpha Gene Amplification Is an Independent Predictor of Long-Term Outcome in Postmenopausal Patients with Endocrine-Responsive Early Breast Cancer

The use of validated gene expression assays can help oncologists to assess recurrence risks in endocrine-treated hormone-receptor positive (HR⁺), human epidermal growth factor 2–negative (HER2⁻) early-stage breast cancer, and to identify patients who are likely to benefit from the addition of chemotherapy to standard endocrine therapy. Most of the newer assays use algorithms that are based on the expression of several proliferation- and estrogen-associated genes, and mainly rely on the prognostic usage of genes arbitrarily identified by nonhierarchical clustering (1).

Disappointingly, however, despite decades of research, there is still no single ERα-associated biomarker that can be used to identify subgroups of HR⁺ early breast cancer patients with a particularly good or poor outcome.

Amplification (increased copy number) of the ESR1 gene encoding ERα has been reported in up to about 30% of early breast cancers depending on detection and scoring methods (2–10). Some studies have suggested that ESR1 amplification as detected by FISH could identify a subset of cancers that might respond particularly well to anti-ER treatment (2, 3), but others linked ESR1 amplification to therapy resistance rather (5, 7, 11).

It is likely that the compilation of patient cohorts and different detection methods, as well as variable definitions of ESR1 amplification, have contributed to these discrepant findings (7–9). In order to investigate whether ESR1 amplification is associated with long-term outcome in endocrine-responsive early breast cancer, we have therefore analyzed breast cancer samples from the prospective clinical ABCSG-6 trial for ESR1 copy-number changes using simple scoring criteria and a commercially available ESR1 probe.

This study is part of the ABCSG translational research program (abcsg.research). 537 women included in this study had been randomized into the prospectively randomized adjuvant endocrine trial ABCSG-6 between 1990 and 1995, and had received 5 years of adjuvant tamoxifen (12). Approximately 50% of patients participating in that trial were subsequently rerandomized to receive 3 years of extended anastrozole versus no further treatment (ABCSG-6a; ref. 13). For the purpose of the main analysis, these patients were censored at the time of their follow-up when they were entered into ABCSG-6a to ensure treatment homogeneity of this study (i.e., none of the patients in this analysis had been exposed to any endocrine treatment other than 5 years of tamoxifen). Notably, none of the HER2-positive patients had received HER2-directed therapy. Eventually, 394 patients who received tamoxifen monotherapy and had tumor blocks available were included in this study. In an additional exploratory analysis, we investigated the long-term outcome of 254 patients who had not experienced a recurrence by the end of 5 years of adjuvant tamoxifen, and who had subsequently been re-randomized to receive either 3 additional years of anastrozole or no further treatment as part of the ABCSG6a extension study. Trial design, inclusion criteria and the main clinical results of these trials have been previously reported (12, 13). Formalin-fixed, paraffin-embedded (FFPE) tumor blocks were collected from participating centers at the time of surgery and were stored at room temperature. Ethical Approval was obtained from Institutional Review Boards. Written informed consent forms were obtained from every patient prior to participation in the described trial.

Freshly cut 4-μm tissue sections were used for ERα IHC analysis as previously described (2). IHC detection of ERα protein was performed using the antibody NCL-L-ER-6F11 (Novocastra). In brief, slides were deparaffinized and subjected to antigen retrieval in a pressure cooker at 120°C for 12 minutes in citrate buffer at pH 6. The primary antibody NCL-L-ER-6F11 (Novocastra) was prediluted 1:1,000 and incubated overnight at 4°C. The Vectastain ABC Elite system was used for detection of antibody binding. IHC scoring was performed according to the Allred score (14, 15). In brief, ERα staining intensity was recorded on a 4-tiered scale (0–3) and the percentage of ERα-positive tumor cells on a 5-tiered (1–5) scale. Addition of both parameters resulted in an 8-tiered score, and a score of >2 was considered positive.

Large FFPE sections were treated using the ZytoLightSPEC ESR1/CEN 6 Dual Color Probe Kit (Zytovision) according to the manufacturer's instructions with minor modifications: Slides were heated overnight at 58°C before deparaffinization. Probe hybridization time at 37°C was extended to 48–72 hours.

A pathologist marked areas for FISH scoring on a consecutive hematoxylin and eosin–stained reference slide. All slides were centrally analyzed by an experienced scientist. Tumor regions with clearly detectable FISH signals for both ESR1 and CEP6 were selected for analysis. In case of tumor heterogeneity of elevated ESR1/ CEP6 signal ratios, the areas with the highest ratios were chosen. In 10 to 20 representative tumor cell nuclei showing distinguishable FISH signals, the number of ESR1 and CEP6 signals was determined. The average number of ESR1 and centromere 6 (CEP6) signals were used to calculate the ESR1/CEP6 copy-number ratio indicating focal copy number increase. High-level ESR1 amplification was defined as an ESR1/CEP6 ratio ≥2.0, and elevated ESR1/CEP6 copy-number ratios from ≥1.3 to <2 were considered as low-level amplification (gains; refs. 6, 16, 17). Tumors with an ESR1/CEP6 ratio of <1.3 were classified as “not amplified.” For statistical analysis, tumors with ESR1 low- and high-level amplification (see representative examples in Fig. 1) were combined into one group of ESR1 amplification (“increased ESR1 copy number”) in the predefined analysis plan.

The primary endpoints of the statistical analyses were distant recurrence-free survival (DRFS) and breast cancer–specific survival (BCSS). DRFS was defined as the interval between the date of surgery and the first evidence of relapse at any distant site. Because of the median age of 65 at trial initiation and the long-term follow-up, patients were censored if they had, in the absence of breast cancer recurrence, died from confirmed reasons unrelated to their malignancy. Baseline data, dichotomized according to ESR1 amplification status (increased versus normal copy numbers), were compared in univariate analyses using the χ² test and in a multiple logistic model. Survival rates were estimated using the Kaplan–Meier method. The prognostic value of ESR1 amplification status (including low- and high-level amplifications in separate or combined analysis) was studied using univariate and multivariable Cox models. Interaction terms between ESR1 copy number and HER2 were assessed in Cox models. All P values are shown as the results of two-sided tests. A P value of ≤0.05 was considered statistically significant. All statistical analyses were performed using SPSS software version 15.0 (SPSS, Inc.)

Data were generated by the authors. The dataset used is not publicly available as it may contain information that would compromise patient consent. Please contact the corresponding author for more information and to request access to these data.

Of the 996 female patients with early breast cancer who were randomized into the tamoxifen-only arm in ABCSG 6, FFPE tumor samples were available in 537. Of these, ESR1 gene and centromere 6 copy numbers were assessable in 394 cases (Fig. 2). Focal low-level (9%) and high-level (38%) amplifications were detected in 187 of 394 interpretable cancers (47%). Eight of 394 (2%) samples exhibited a ratio of <1, while the vast majority (272/394; 69%) ESR1/CEP6 ratios ranged from 1.0 to 2.0. (Fig. 3). Because there was no statistically significant difference in survival between tumors with ESR1 low- and high-level amplifications (HR for relapse, 4.93; 95% CI, 0.66–36.64, P = 0.12; HR for death, 2.49; 95% CI, 0.58–10.62, P = 0.22), we combined these tumors into one group of tumors with ESR1 amplification (increased copy number) for further analysis. ESR1 amplification was significantly linked to patient age >60 (P = 0.006) but were unrelated to breast cancer tumor size, tumor grade, or nodal status. These data are summarized in Table 1.

Increased ESR1 copy numbers (amplification) were significantly correlated with ERα protein expression both, when measured locally in participating trial centers using the Remmele Score (P = 0.001, χ² test), and when measured centrally using the Allred score (P < 0.0001, χ² test). With one exception (4%) out of 24 samples, we did not observe low- or high-level ESR1 amplifications in confirmed ERα-negative tumors (i.e., Allred score of 0–2), while ESR1 amplification was found in 127 of 206 (62%) tumors with an Allred score of 7–8 (Table 2).

Median follow-up of the patients was 10 years. Tumor size, nodal status, and ESR1 amplification were significantly associated with DRFS in univariate analyses (Table 3).

Likewise, tumor size, nodal status, tumor grade, and ESR1 amplification were significantly associated with BCSS in univariate analysis (Table 3). The independent effect of ESR1 amplification on DRFS and BCSS was assessed by multivariable Cox proportional Hazard models adjusted for age, tumor size, nodal status, tumor grade, and HER2. In these multivariable analyses, ESR1 amplification remained significantly associated with prolonged DRFS (adjusted HR for relapse, 0.48; 95% CI, 0.26–0.91; P = 0.02) and improved breast cancer–specific survival (adjusted HR for death, 0.47; 95% CI, 0.27–0.80; P = 0.01) when compared with women with tumors exhibiting normal ESR1 copy numbers. Within the group of ESR1-amplified tumors, however, we did not observe an association between the level of amplification and outcome for both, DDFS (adjusted HR, 0.88; 95% CI, 0.70–1.13; P = 0.317) and BCSS (adjusted HR, 0.89; 95% CI, 0.72–1.46; P = 0.89)

In contrast, assessment of ERα protein by Allred score did not allow for discrimination between DRFS and BCSS in tamoxifen-treated women (adjusted HR for relapse, 0.86; 95% CI, 0.59–1.25; P = 0.43 and adjusted HR for death, 0.85; 95% CI, 0.58–1.23; P = 0.38; Fig. 4A–D).

No significant interaction was observed between ESR1 amplification and HER2 with respect to DRFS (P_interaction = 0.63) and BCSS (P_interaction = 0.82). We then investigated possible associations between ESR1 amplification and DRFS and BCSS in relation to the patient´s nodal status, and found that ESR1 amplification was associated with an improved DRFS and BCSS in nodal-positive tumors, while this correlation was absent in nodal-negative tumors (Fig. 5A–D).

We also performed an additional outcome analysis in patients who had not experienced a recurrence by the end of 5 years of adjuvant tamoxifen had been rerandomized to receive either 3 additional years of anastrozole (n = 125; 51%) or no further (n = 120; 49%) treatment. We did not find a significant interaction between AI intake and ESR1 amplification. Furthermore, both DRFS and BCSS were not associated with ESR1 amplification in either the anastrozole-treated, or the untreated patient population (data not shown).

Qualitative ERα expression identifies endocrine-responsive tumors, and is both an established prognostic parameter and a predictor for response to endocrine therapy in early breast cancer. A recent meta-analysis of 19 studies including 30,754 patients, however, found no clear evidence for a correlation between higher quantitative ERα and better disease outcome in patients with stage 1–3 breast cancer (18). These results are in line with our observation which also demonstrated that semiquantitative ERα protein expression using the Allred score was unable to predict the long-term outcome.

Furthermore, in vitro studies show that mutations in the hotspot ligand-binding domain of the ERα gene ESR1 confer ligand-independent activity and relative resistance to tamoxifen and fulvestrant, but these mutations develop under the selective pressure of endocrine treatments, and are infrequent in untreated early ER⁺ breast cancers (19, 20).

Consequently, whereas ESR1 mutations in circulating DNA are commonly detected in metastatic disease, they are rarely detectable at the initial diagnosis or during adjuvant treatment, thus limiting their utility to predict long-term outcome in endocrine-treated early breast cancer (21).

In this study, we retrospectively analyzed breast cancer tissues from the prospective adjuvant ABCSG-6 trial using a commercial ESR1 FISH probe and simple scoring criteria to detect cancers with elevated ESR1 gene copy numbers.

Using FISH, we detected focal ESR1 amplifications (copy-number elevations) in 47% of 394 hormone receptor–positive breast cancers. This is somewhat higher than the combined fraction of ESR1 amplifications and gains in previous studies performed by us (36%) or others (34%; refs. 2, 3). However, using a different FISH assay and an ESR1/CEP6 ratio of 1.3 as cutoff for ESR1 gain, a recent study by Laenkholm and colleagues has reported comparable rates of ESR1 copy-number elevations in 42% of breast cancer cases (16). Of note, we selected ERα-positive tumors in our current study (94% ER positive by IHC), while there were only 77% ER-positive tumors in our previous study and 82% in the study by Tomita and colleagues (2, 3). The higher fraction of cancers with increased ESR1 copy numbers in our current study therefore also reflects the known close association between ERα expression and ESR1 copy-number alterations (2).

Because the initial description of ESR1 amplification in breast cancer, there has been considerable discrepancy regarding the reported prevalence and the clinical relevance of ESR1 amplification in early breast cancer (8). Several groups have independently detected ESR1 copy-number elevations in breast cancer samples in more than 20% by using FISH technology (2–4, 8, 16, 22). These results were, however, challenged by studies in which predominantly alternative techniques were employed: Quantitative PCR (qPCR), multiplex-ligation dependent probe amplification (MLPA), comparative genomic hybridization (CGH), and single nucleotide polymorphism (SNP) microarrays yielded considerably lower amplification rates of ESR1 amplification of <10% and did not demonstrate a prognostic role for ESR1 gene copy changes (6, 8, 23–27).

Although these discrepancies could be the consequence of patient selection in individual studies, it is obvious that significant percentages of these differences could in part be caused by the applied technology: Nonmorphological methods such as qPCR, MLPA, SNP microarrays, or Southern blotting are particularly sensitive to contamination with nonmalignant breast tissue and to tumor heterogeneity in the copy-number status, which could conceal amplifications of low copy-number level and consequently result in a considerable underestimation of ESR1 amplification rates (8, 9, 28).

Another factor that might have led to an underestimation of ESR1 amplification rates in some studies is the choice of improper cut-off levels for calling amplification. By setting the cut-off levels too high, copy-number elevations of low level would be considered unamplified, which has already been shown to have impact for several other genes, although it is still unclear how low gain rates translate into clinically relevant clinical outcomes (6, 8, 9).

Our findings are supported by an earlier report describing a prolonged survival of women whose tumors exhibited ESR1 amplification among the 175 patients who had undergone tamoxifen treatment (2). They are also in line with findings by Babyshkina and colleagues who recently found that in tamoxifen-treated luminal tumors, a significant decrease in ESR1 mRNA expression levels and a heterogeneous ERα protein expression pattern were both associated with a significantly shorter PFS (29).

Our results are, however, contradictory to a retrospective case–control study of 91 patients in whom ESR1 amplifications were significantly more common in primary breast carcinomas which recurred within the first 4 years after diagnosis (11). Furthermore, in a subset of the BIG 1–98 trial, ESR1 amplification status alone did not predict for DFS after 5 years of endocrine therapy, but was prognostic in combination with the HER2 amplification status (5). In both studies, however, only tumors with an ESR1/CEP-6 ratio of ≥2 (i.e., high-level amplification) were considered positive and low-level ESR1 amplification (copy number gains) were not included. It is well possible that the prognostic value of the ESR1/CEP-6 ratio assessment in these trials was simply compromised by selection of an inadequate cut-off value.

The mechanism by which ESR1 amplification renders an endocrine-responsive tamoxifen-treated tumor less aggressive remains unclear. One hypothesis, however, suggests that the amplification a particular gene might be driven by the tumors’ addiction to a pathway in which the protein product of the respective gene is involved (30). Such a mechanism has been suggested in two independent studies that observed focal ESR1 amplifications of low-level copy-number change in long-term estrogen-deprived (LTED) MCF7 breast cancer cell lines. Yet another experimental study showed that breast-cancer-derived xenografts respond to estrogen treatment of tumor cells that harbor ESR1 amplification (31, 32). Further evidence comes from a phase II study which evaluated antiestrogen treatment, and found focal ESR1 amplification in response to endocrine deprivation by (33). Taken together, these functional studies provide strong evidence for the potential clinical relevance of ESR1 amplification as a mechanism of ERα pathway regulation.

In summary, we have detected focal ESR1 amplification (copy-number increase) in 47% of ERα-positive and endocrine-responsive early stage breast cancers from postmenopausal women who had been enrolled in the ABCSG-6 trial. In this well-defined and meticulously documented prospective clinical trial patient population receiving 5 years of tamoxifen therapy, we demonstrate that ESR1 amplification status is a predictor for long-term DRFS. In contrast to most multigenomic assays, whose reported prognostic value is limited to DFS and DDFS, we here demonstrate that the ESR1 copy-number ratio is also predictive of survival (34, 35). An analysis of ESR1/CEP17 ratios in patients included in prospectively designed studies with aromatase inhibitors will help to further establish the prognostic role of this easily available biomarker in endocrine-responsive early breast cancer.

C.F. Singer reports grants from Deutsche Forschungsgemeinschaft (DGF SI 1347/3-1) and Roche Inc. during the conduct of the study; grants and non-financial support from Roche, Novartis, AstraZeneca, and Amgen outside the submitted work. F. Holst reports grants from Deutsche Forschungsgemeinschaft during the conduct of the study; personal fees from ZytoVision GmbH outside the submitted work; in addition, F. Holst has a patent for US8101352B2 issued, licensed, and with royalties paid from ZytoVision GmbH and a patent for EP2046984B1 issued, licensed, and with royalties paid from ZytoVision GmbH. E. Burandt reports personal fees from Novartis and EISAI outside the submitted work. S. Lax reports personal fees from Biomedica, Glaxo-Smith-Kline (GSK), Merck Sharp Dome (MSD), Pharma Mar, AstraZeneca, and Novartis outside the submitted work. R. Greil reports personal fees from Celgene, Roche, Merck, Takeda, AstraZeneca, Novartis, Amgen, BMS, MSD, Sandoz, AbbVie, Gilead, Daiichi Sankyo, and Sanofi during the conduct of the study and personal fees from Celgene, Roche, Merck, Takeda, AstraZeneca, Novartis, Amgen, BMS, MSD, Sandoz, AbbVie, Gilead, Daichii Sankyo, and Sanofi outside the submitted work. M. Filipits reports personal fees from AstraZeneca, Biomedica, Bio-Rad, Boehringer Ingelheim, Eli Lilly, Merck, Novartis, and Pfizer outside the submitted work. M. Gnant reports personal fees from Amgen, Daiichi Sankyo, AstraZeneca, Eli Lilly, Lifebrain, NanoString, Novartis, Merck Sharp & Dohme, and Pierre Fabre outside the submitted work; and an immediate family member is employed by Sandoz. No disclosures were reported by the other authors.

C.F. Singer: Data curation, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. F. Holst: Investigation, methodology, writing–original draft, writing–review and editing. S. Steurer: Data curation, investigation, writing–review and editing. E. Burandt: Data curation, investigation, writing–review and editing. S. Lax: Investigation, writing–review and editing. R. Jakesz: Data curation, supervision, investigation, writing–review and editing. M. Rudas: Supervision, investigation, writing–review and editing. H. Stöger: Data curation, supervision, investigation, writing–review and editing. R. Greil: Data curation, supervision, investigation, writing–review and editing. G. Sauter: Data curation, investigation, writing–review and editing. M. Filipits: Formal analysis, investigation, methodology, writing–original draft, writing–review and editing. R. Simon: Supervision, investigation, writing–review and editing. M. Gnant: Data curation, supervision, investigation, methodology, writing–original draft, writing–review and editing.

This study was supported by the Deutsche Forschungsgemeinschaft (DGF SI 1347/3-1) and by a research grant from Roche Inc. We thank Sylvia Schnöger and Sascha Eghdessadi from the Department of Pathology, University Medical Center Hamburg-Eppendorf, Hamburg, for excellent technical support. No compensation was received for the persons named in this section.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.