Main

Aromatase inhibitors (AIs) confer a disease-free survival (DFS) benefit over and above that achieved with adjuvant tamoxifen in postmenopausal women with early oestrogen-or progestrone-receptor (ER/PgR)-positive breast cancer (Thurlimann et al, 2005; Forbes et al, 2008; van de Velde et al, 2011). Recent data from the Tamoxifen and Exemestane Adjuvant Multinational (TEAM) and Breast International Group (BIG) 1–98 trials suggest that DFS is similar in patients treated with either an AI for 5 years or ‘switched’ to an AI following 2–3 years of tamoxifen (Thurlimann et al, 2005; van de Velde et al, 2011).

This observation has generated debate regarding the optimal treatment strategy (upfront AIs vs switch) for postmenopausal ER/PgR-positive breast cancer. As, for all strategies, the benefit of AIs vs tamoxifen is modest when compared with tamoxifen vs no endocrine treatment (Abe et al, 2005; Viale et al, 2009), there is considerable impetus for translational studies aimed at identification of those patients most likely to benefit from upfront AIs.

The differences in benefit between hormonal regimens may be explained, in part, by the diverse biology of breast cancer, particularly the differences between luminal A and luminal B cancers (Perou et al, 2000). Clearly, multiple factors may influence differential response to hormonal treatments. Including HER2, Ki-67, and RAS/RAF or PI3K/Akt signalling (Beeram et al, 2007; Viale et al, 2008). Future selection of optimal endocrine therapy for early breast cancer will need to be personalised based on studies identifying an increasing number of patient subsets with unique molecular profiles, for example (Curtis et al, 2012).

Selecting optimal adjuvant endocrine therapy and/or chemotherapy is currently influenced by measures of residual risk (Van Belle et al, 2010). However, selection between endocrine agents (AIs vs tamoxifen) requires specific markers indicating differential benefit from these agents. We have previously shown, within TEAM, that quantitative analysis of ER and PgR expression combined with clinicopathologic factors (age, tumour size and grade, and nodal status) can identify patients at higher risk for early recurrence (Bartlett et al, 2011). We confirmed previous data indicating that PgR, although prognostic, is not a predictive marker of benefit from AI vs tamoxifen (Dowsett et al, 2008; Simon et al, 2009). Our study was, unlike previous studies, based on an adequately powered and prospectively planned treatment-by-marker analysis, satisfying criteria for high-level evidence (Simon et al, 2009).

Type 1 receptor kinase expression (HER1, HER2, and HER3 (HER1–3)) is associated with a higher probability of early relapse in tamoxifen-treated patients (Tovey et al, 2004, 2005), consistent with both preclinical and clinical data suggesting overexpression of HER2, and HER1/EGFR confer resistance to tamoxifen (Benz et al, 1992; Carlomagno et al, 1996; Houston et al, 1999). Conversely, neoadjuvant studies suggest that AIs are effective regardless of HER1 or HER2 overexpression (Ellis et al, 2001; Dixon et al, 2004). On the basis of these observations, we hypothesised that overexpression of HER1, HER2, and/or HER3 is associated with a differential benefit of an AI compared with tamoxifen in the adjuvant setting, and that outcome in patients with HER1–3-positive tumours would be improved by initiating treatment with an AI rather than tamoxifen. The analysis presented here was prospectively planned and powered to test the hypothesis, within the TEAM study, that HER1–3 status acts as a predictive biomarker for benefit of exemestane vs tamoxifen during the 2.75 years prior to the switch point.

Materials and Methods

Study design

The TEAM trial, an international, open-label, phase III trial in postmenopausal women with ER/PgR-positive early breast cancer (van de Velde et al, 2011), included two prospectively planned and powered pathology studies. This, the second TEAM pathology study, tests the hypothesis that upfront exemestane improves DFS compared with tamoxifen in patients with HER1–3-positive tumours, defined as tumours expressing high levels of at least one of HER1, HER2, or HER3 proteins. Outcomes in patients with high levels of HER1–3 expression were compared with those in patients without high HER1–3 expression. This intent-to-treat analysis was planned at 2.75 years follow-up and was not event-driven (van de Velde et al, 2011). Using a two-sided α=0.05 assuming a hazard ratio (HR) of 1.93 and HER1–3-positive prevalence of 25%, a sample size of 4000 patients would give >90% power to detect a treatment–biomarker interaction. Secondary exploratory analyses including DFS at 2.75 years, censoring patients at the actual time of switching, and DFS at a median follow-up of 6.5 years were performed.

Patients

Overall samples from 4781 patients were received from the United Kingdom (1097), The Netherlands (2722), Belgium (122), Germany (745) and Greece (95). Patient demographics and tumour characteristics were similar between the analysed subset and all patients in the pathology substudy; patients in the pathology subset were at slightly higher risk than the entire TEAM population (Bartlett et al, 2010) (Supplementary Table 1: CONSORT table).

In general, patients had histologically or cytologically confirmed T1-3 N0-2 M0 breast adenocarcinoma and were treated with surgical resection followed by radiotherapy and/or adjuvant chemotherapy. Among countries participating in the TEAM trial, five (United Kingdom/Ireland, The Netherlands, Belgium, Germany, and Greece) provided tumour samples for this substudy after appropriate ethical review.

Staining methodology

Tissue microarrays (TMAs) were constructed as reported previously (Bartlett et al, 2011), in line with current guidelines (Leyland-Jones et al, 2008). Standard immunohistochemical techniques were used to stain TMAs for HER1–3 (HER1 clone 31G7: Invitrogen, Paisley, UK; HER2, HercepTest, and HER3 clone DAK-H3-IC: Dako, Cambridgeshire, UK) and Ki67 (Clone MIB1, 1 : 50 dilution, Dako). Assays were performed to good laboratory practice (GLP) in a GLP-monitored laboratory using single batches of each antibody and reagent; incubations were rigorously controlled for temperature. In each assay, quality controls with varying HER expression were included as described previously for ER/PgR (Bartlett et al, 2011). Cores were scored manually for HER1–3 using highly trained observers (Kirkegaard et al, 2006). Histoscores for HER1–3 (membrane staining only) were recorded. For Ki67, scoring was performed using the Ariol SL-50 Image analysis system (Genetix, New Milton, UK) as previously described for ER/PgR (Bartlett et al, 2011) using an algorithm developed specifically for Ki67. Results for Ki67 were recorded as the percentage of Ki67 positive cells. HER2 status was confirmed by fluorescence in situ hybridisation (FISH) with 95% concordance between immunohistochemical and FISH results (Wolff et al, 2007).

Statistical analysis

Disease-free survival, was defined as time from randomisation to earliest documentation of disease relapse (primary tumour recurrence (locoregional or distant) and ipsilateral/contralateral breast cancer) or death from any cause. The primary aim of this study was evaluation of the interaction between HER1–3 expression and treatment at 2.75 years on an intent-to-treat basis. Exploratory intent-to-treat analyses of interactions between HER1–3 expression and treatment were performed with 6.5 years median follow-up (van de Velde et al, 2011). A sensitivity analysis of the impact of censoring patients at the recorded time of treatment switch (Tam-AI) was performed. Exploratory analysis of DFS from 2.75 years postrandomization (excluding patients without follow-up after 2.75 years) until last recorded follow-up was performed. Data included in these analyses were locked on 1 February 2013.

The predictive value of HER expression was assessed using Cox proportional hazards regression models. Interactions between treatment arms and HER expression levels were evaluated using the Wald chi-square (χ2) statistic. The predictive value of HER expression was further investigated in multivariate analyses, consistent with REMARK guidelines (McShane et al, 2006), adjusting for known prognostic factors: patient age (continuous variable); tumour size (continuous variable); number of positive nodes (continuous variable); treatment with chemotherapy (yes/no); treatment arm (tamoxifen/exemestane); and expression of HER1–3 (negative/positive), ER, PgR, and Ki67 (each a continuous variable). Continuous variables were evaluated for nonlinearity by applying simple log transformations followed by more complex fractional polynomials, and the best-fitting transformation was applied as assessed by the change in Akaike’s information criterion between univariate Cox proportional hazard models of transformed and untransformed data (Collett, 1994). Treatment allocation was included as a time-dependent covariate to investigate the impact of switching on the tamoxifen randomised arm. The proportional hazards assumption was investigated and time and covariate interactions were analysed to evaluate changing effects with time. All data were analysed using SAS/STAT statistical software (SAS Institute, Cary, NC, USA).

Results

Study population

Of 4541 eligible samples (Bartlett et al, 2011), 4225 (93%) had complete HER1–3 data: 199 (5%) were HER1-positive; 547 (13%) HER2-positive, and 875 (21%) HER3-positive. Altogether, 1351 (32%) tumours were positive for at least one HER1–3 biomarker. Patient demographics and tumour characteristics were similar between the analysed subset and all patients in the pathology substudy; patients in the pathology subset were at slightly higher risk vs the entire TEAM population (Bartlett et al, 2010) (Supplementary Table 1: CONSORT table).

Efficacy

At 2.75 years, prospectively powered intent-to-treat analysis of treatment by HER1–3 marker expression

Among patients analysed in this substudy (n=4225), 408 DFS events were recorded over 2.75 years of follow-up. A trend towards DFS benefit of exemestane vs tamoxifen (HR=0.84; 95% confidence interval (CI), 0.69–1.02) similar to the entire TEAM trial population was observed at 2.75 years (van de Velde et al, 2011). Among 2874 patients with HER1–3-negative tumours, 237 (8%), and among the 1351 patients with HER1–3-positive tumours, 171 (13%) DFS events occurred. HER1–3-positive patients had a 57% increased risk of a DFS event vs HER1–3-negative patients (HR=1.57; 95% CI: 1.29–1.91; P<0.0001). In the pre-planned treatment-by-marker analysis, there was a significant interaction between HER1–3 expression and treatment (HR=0.58; 95% CI, 0.39–0.87; P=0.008) in favour of increased benefit from exemestane in HER1–3-negative patients (Figure 1A). Among HER1–3-negative patients, there was a DFS benefit associated with exemestane vs tamoxifen (HR=0.67; 95% CI, 0.52–0.87; Figure 1B). Conversely, there was no marked treatment effect between exemestane and tamoxifen in HER1–3-positive patients (HR=1.15; 95% CI, 0.85–1.56; Figure 1C). In multivariate regression analysis, this treatment-by-marker interaction remained significant (P=0.005; HR=0.55; 95% CI, 0.36–0.85), along with nodal status, tumour grade/size, patient age, ER, PgR, Ki67 (all as continuous variables); HER1–3 expression; and treatment (Table 1). Prior chemotherapy did not contribute significantly to risk at 2.75 years. Supplementary analysis at 2.75 years stratified by country shows similar interaction HRs across all countries (Table 2).

Figure 1
figure 1

Disease-free survival in the intent-to-treat population at 2.75 years: (A) hazard ratio plot of treatment/biomarker interaction; (B) tumours negative for HER1, HER2, or HER3 (n=2874; 68%); and (C) tumours positive for HER1, HER2, or HER3 (n=1351; 32%). Abbreviations: CI=confidence interval; Exe=exemestane; HER=human epidermal growth factor receptor; HR=hazard ratio; O–E=observed minus expected; Tam=tamoxifen; Var=variance.

Full size image
Table 1 Multivariate analysis of disease-free survival at 2.75 years (N=3779; 360 DFS events)
Full size table
Table 2 Hazard ratios for the individual countries for HER1–3 expression and interaction with treatment arm
Full size table

Exploratory analyses: HER1, HER2, and HER3 expression

Further exploratory analyses were performed for the individual HER receptors; benefit from upfront exemestane vs tamoxifen treatment was apparent in HER1-negative (HR=0.80; 95% CI, 0.65–0.98) vs HER1-positive tumours (HR=1.60; 95% CI, 0.79–3.25; interaction test HR=0.50; 95% CI, 0.24–1.03; P=0.06). Similarly, HER2-negative tumours indicated benefit from upfront exemestane vs tamoxifen (HR=0.71; 95% CI, 0.57–0.89), vs HER2-positive tumours (HR=1.67; 95% CI, 1.09–2.55; interaction test HR=0.43; 95% CI, 0.26–0.70; P<0.001). However, there was no apparent differential benefit among HER3-negative or HER3-positive patients (HR=0.80; 95% CI, 0.64–0.99 vs HR=1.00; 95% CI, 0.65–1.53; interaction test HR=0.80; 95% CI, 0.50–1.29; P=0.36; Figure 2A). In a second exploratory analysis, tumours expressing either HER1 or HER2 were assumed (by the formation of active homo- or heterodimers) to exhibit ‘active HER signalling’, whereas tumours lacking HER1, HER2, and HER3 expression or expressing HER3 only were assumed to exhibit limited ‘HER signalling’ (Bartlett et al, 2010). A significant (exemestane vs tamoxifen) treatment-by-marker (‘active HER signalling’ vs cases without ‘active HER signalling’) interaction (HR=0.42; 95% CI, 0.27–0.65; P<0.001; Figure 2B) suggests that patients with active HER signalling do not derive benefit from early exemestane treatment.

Figure 2
figure 2

Contribution of each HER to disease-free survival benefit by treatment group: (A) interaction for individual HER receptors; and (B) activity of HER ‘active’ vs ‘inactive’ signalling. Abbreviations: CI=confidence interval; Exe=exemestane; HER=human epidermal growth factor receptor; HR=hazard ratio; O–E=observed minus expected; Tam=tamoxifen; Var=variance.

Full size image

Exploratory 2.75-year censored analysis

The primary intent-to-treat analysis (DFS at 2.75 years) evaluated benefit of exemestane and tamoxifen with relation to HER1–3 expression at the expected ‘switch point’ (tamoxifen patients switching to exemestane) of 2.75 years. However, 45% (949/2113) of tamoxifen patients switched treatment before the 2.75-year follow-up, whereas 21% (439/2113) discontinued tamoxifen early and did not switch. Among exemestane-treated patients, 12% (257/2112) stopped treatment early. A sensitivity analysis censored all patients at the actual time of switch, at treatment cessation, or at 2.75 years, whichever occurred first. The time until treatment cessation (excluding those who switched) for patients who stopped treatment early was different in the two treatment arms (median treatment duration, 0.94 vs 0.67 years for tamoxifen-treated and exemestane-treated patients, respectively), leading to potential bias in this analysis. Among patients included in the sensitivity analysis (n=4225), 278 events were recorded. Comparing DFS at 2.75 years between HER1–3-negative and HER1–3-positive patients, as with the 2.75-year analysis, HER1–3-positive patients had significantly increased risk of a DFS event (HR=1.61; 95% CI, 1.27–2.04; P<0.001). Analysis revealed that significant treatment-by-marker interaction between HER1–3 expression and treatment with exemestane/tamoxifen remained significant in both univariate (HR=0.50; 95% CI, 0.31–0.81; P=0.0049; Figure 3), and multivariate analyses (HR=0.44; 95% CI, 0.26–0.73; P=0.002).

Figure 3
figure 3

Hazard ratio plot of treatment-by-marker analysis, censoring patients at the time of treatment switch. This analysis presents disease-free survival at 2.75 years; data in this figure are censored=whereas Figure 1A data are not. Abbreviations: CI=confidence interval; Exe=exemestane; HER=human epidermal growth factor receptor; HR=hazard ratio; O–E=observed minus expected; Tam=tamoxifen; Var=variance.

Full size image

Extended follow-up

In an unplanned analysis, data were also evaluated at a median follow-up of 6.5 years, including the period when all patients were treated with exemestane (unless they discontinued treatment). Among patients analysed (n=4225), 1021 DFS events were recorded. No significant interaction between HER1–3 and treatment with tamoxifen vs exemestane was observed (HR=1.05; 95% CI, 0.82–1.36; P=0.68). Among HER1–3-negative patients (n=2874), 632 (22%) DFS events occurred, and among HER1–3-positive patients (n=1351), 389 (29%) DFS events occurred (Figure 4A). There was no significant treatment effect (exemestane/tamoxifen) in HER1–3-positive patients (HR=0.93; 95% CI, 0.76–1.14; Figure 4B). In HER1–3-negative patients, there was no longer a significant treatment effect between exemestane and tamoxifen (HR=0.98; 95% CI, 0.84–1.15; Figure 4C). Relapses in the exemestane monotherapy group increased after the 2.75-year assessment and did not follow an extrapolation of the 2.75-year curve.

Figure 4
figure 4

Disease-free survival in the intent-to-treat population at 5 years: (A) hazard ratio plot of treatment/biomarker interaction; (B) tumours positive for HER1, HER2, or HER3; and (C) tumours negative for HER1, HER2, or HER3. Abbreviations: CI=confidence interval; Exe=exemestane; HER= human epidermal growth factor receptor; HR= hazard ratio; O–E= observed to expected; Tam= tamoxifen; Var= variance.

Full size image

Analysis of treatment-by-marker interaction as a time-dependent covariate

By including each of the covariates as an interaction term with time in a Cox proportional hazard model, it is possible to investigate how the effect of each covariate on outcome varies with time. There is no significant interaction with time for either treatment or HER1–3 positivity. However, the hazard associated with the interaction between treatment and HER1–3 positivity significantly increases with time suggesting that HER1–3-negative patients treated with exemestane have an increased relapse risk as time progresses (HR=1.07, 95% CI 1.01–1.14, P=0.017). Relapse risk increases by 7% per annum for HER1–3-negative exemestane-treated patients when compared with all other patients.

As expected, given the time dependence of the HER1–3/treatment interaction term, assessment of the proportional hazard assumptions of the Cox model shows that the hazard of disease is not proportional between the two groups (Figure 5). Relapse risks clearly diverge between 0–3 years indicating that the hazard of disease during this time period for the two groups are not proportional. After 3 years, the proportionality assumptions are met. Therefore, the inclusion of the interaction with time of the treatment-by-marker interaction term is justified and explains the lack of evidence for treatment-by-marker interaction with extended follow-up without the use of a time-dependent model.

Figure 5
figure 5

Log–log survivor plot for HER1–3-negative cases treated with exemestane (dashed line) vs all other cases (solid line). Divergence of lines prior to 3–4 years postrandomisation is evidence of non-proportional hazard rates between groups at this time. After this time, the hazard of relapse appears to be proportional between groups.

Full size image

Discussion

The results of this prospectively planned translational study show that expression of HER1, HER2, or HER3 predicts a differential benefit from initial adjuvant therapy with an AI compared with tamoxifen, which is shown to be both real and time dependent. In a prospectively planned and powered analysis, a significant DFS benefit in favour of initiating treatment with exemestane was observed among patients with HER1–3-negative tumours, in both univariate and multivariate analyses including the treatment-by-marker interaction (Figure 1, Table 1). Strikingly, this study did not show any benefit associated with initial exemestane treatment vs tamoxifen in patients with HER1-, HER2-, or HER3-positive tumours suggesting these tumours are partially resistant to endocrine therapy (Shou et al, 2004; Folgiero et al, 2008; Massarweh et al, 2008; Osborne et al, 2011). However, lack of overexpression of HER1, HER2, or HER3 is confirmed as an independent predictive biomarker for early AI benefit in patients with ER/PgR-positive early breast cancer. Exploratory analyses suggested that this effect was largely driven by HER1/HER2 expression, consistent with predicted HER signalling activity (HER3 lacks significant signalling potential) (Yarden and Pines, 2012). Therefore, assessing HER1/HER2 could provide valuable information in clinical practice in ER-positive disease (Hudelist et al, 2003; Sassen et al, 2008). Finally, in an exploratory time-dependent analysis (Figure 5), we identified a difference in the risk of relapse, relative to other tumours, associated with HER1–3-negative tumours treated with exemestane as time progressed. This time-dependent effect ultimately negates the treatment benefit observed in the pre-planned analysis (performed at 2.75 years) such that by 6.5 years median follow-up, no significant interaction between initial endocrine treatment and outcome is observed. Strikingly, in an exploratory analysis of DFS from treatment switch time point (2.75 years), the interaction term was inverted owing to the time-dependent effect in the HER1–3-negative exemestane group (data not shown). Nonetheless the time-dependent analysis confirms the statistical robustness of the interaction during the initial treatment period prior to switching from tamoxifen to exemestane.

The lack of DFS benefit associated with exemestane in the HER1–3-positive subset observed in the current study is entirely consistent with data from transATAC (an ATAC substudy) indicating no difference between benefit from tamoxifen and anastrozole in HER2-positive cancers (Dowsett et al, 2008). In HER2-positive ATAC patients, recurrence rates were 18.8% among tamoxifen-treated patients vs 19.8% among anastrozole-treated patients (Dowsett et al, 2008). Conversely, in HER2-negative patients, 5-year recurrence rates were 9.0% for tamoxifen-treated vs 5.9% for anastrozole-treated patients (Dowsett et al, 2008). These data are consistent with a HER2 treatment-by-marker interaction HR of ∼0.6, similar to that observed in the present study when patients received either tamoxifen or exemestane (prior to 2.75 years). In BIG-1–98, the possibility of a similar interaction between treatment and HER2 status is suggested by the fact that 48 fewer events were observed in the AI treatment arm for HER2-negative patients vs 13 more events in the AI-treated vs tamoxifen-treated HER2-positive group (Viale et al, 2009). A future meta-analysis of this effect across multiple trials (in collaboration with the AI overview group) will be important in evaluating whether effects of HER2 signalling are consistent across multiple trials.

Another striking observation in the TEAM study is the time dependency of the interaction between HER1–3 and treatment (Figures 1 and 4). The subgroup of patients with HER1–3-negative tumours treated with exemestane experience a time-dependent increase in risk of disease relapse when compared with all other patients (Figure 5). This increase progressively erodes the benefit of early treatment with exemestane relative to tamoxifen in the HER1–3-negative group. This apparently paradoxical effect does not appear to occur in either the ATAC or BIG-1–98 studies. How then, could it be explained? The key difference between ATAC/BIG-1–98 and TEAM is that the TEAM addresses specifically a switch from tamoxifen to AIs, whereas ATAC/BIG-1–98 predominantly address AIs vs 5 years of tamoxifen. Further analysis of the effect observed in the TEAM study could be performed in the relatively small switching arms within BIG-1–98. Exploration of a time-dependent effect of these different strategies is warranted; however, if such a time-dependent effect is not observed, the challenge of explaining our observations remains. We speculate that a proportion of HER1–3-negative early breast cancers are primed to develop endocrine resistance, as distinct from those with primary endocrine resistance, and that for a proportion of these cases AIs prevent or delay early recurrence. If our admittedly speculative hypothesis is correct, those cases where AIs delay recurrence may explain the increase in risk for HER1–3-negative patients observed in TEAM, while cases where switching from tamoxifen to AIs provides benefit may explain the convergence of the event rates for HER1–3-negative patients treated with tamoxifen followed by exemestane to those treated with AIs alone. Although we cannot speculate as to the molecular mechanisms relating to these trends, they reflect clinical experience with delayed recurrence following endocrine therapy.

Biomarker analyses raise questions relating to which biomarkers should be included in a risk assessment panel to achieve an optimal result, and how should data be interpreted? Overexpression of HER2 is associated with poor prognosis (Slamon et al, 1989), and the current analysis suggests that patients with HER1–3-positive tumours are at increased risk of early relapse regardless of treatment with exemestane vs tamoxifen consistent with previous data that signalling through multiple members of the HER family is associated with endocrine resistance (Tovey et al, 2004, 2005; Naresh et al, 2006). In patients with HER1–3-negative tumours, the question becomes whether these patients benefit from upfront AI treatment rather than tamoxifen. Assessing signalling potential suggests that exemestane provides a significant differential DFS benefit vs tamoxifen in tumours with ‘inactive’ HER1–3 signalling (Figure 3). These results suggest that the receptor activation may better indicate tumour response to adjuvant therapy than simple expression. The TEAM trial was conducted before introduction of adjuvant HER2-directed therapies, and for women eligible for HER2-directed therapy, an additional dimension exists in understanding the impact of such therapy.

In conclusion, upfront exemestane provided a superior DFS benefit compared with tamoxifen in tumours that were HER1–3-negative or had inactive HER signalling. The time dependency of this effect was explained by a progressive increase in relapse risk, over time, in HER1–3-negative patients treated with exemestane. These results warrant further confirmation in meta-analyses. Tumours that are HER1–3-positive appear to be relatively resistant to endocrine therapy. Pragmatically in clinical practice, HER2 results should provide adequate information for selection of early endocrine therapy although the addition of HER1/EGFR results will be of benefit to a small proportion of patients.