Early Enrichment of ESR1 Mutations and the Impact on Gene Expression in Presurgical Primary Breast Cancer Treated with Aromatase Inhibitors

Over 80% of patients diagnosed with breast cancer present with tumors that are estrogen-receptor positive (ER⁺) and proliferate in response to the female hormone estrogen (E; ref. 1). Aromatase inhibitors (AI) block the conversion of androgens to estrogens and are first-line treatment for postmenopausal women with ER⁺ breast cancer. Despite their efficacy, over 20% of patients with early-stage disease will eventually relapse, and those with metastatic disease will inevitably recur despite initial response to AI therapy (2).

Currently, few mechanisms of resistance to AI therapy have been identified with most being attributed to cross-talk between ER and growth factor signaling pathways, allowing tumor cells to circumvent the need for steroid hormone (3). Furthermore, although studies have shown that AIs lead to a marked remodeling of the clonal mutational landscape (4–6), few mutations have been shown to be enriched in the metastatic setting with the exception of ESR1. Mutations in ESR1 have been observed in 30% to 40% of patients who progress on AI therapy, but only sporadically in patients who have not received AI for metastatic disease (6–11). The identification of new biomarkers and therapeutic strategies that can target early resistance is therefore of paramount importance.

Neoadjuvant (presurgical) AI (NAI) therapy, which is used to downstage primary tumors to enhance the likelihood of breast conserving surgery being a treatment option, provides an ideal opportunity to observe biological changes as a result of AI treatment. This can yield both prognostic and predictive information and facilitate the design of novel clinical trials targeting endocrine-resistant disease (12). Most of the clinical trials exploiting this concept have been restricted to short-term presurgical exposure to AI, such as the PeriOperative Endocrine Therapy for Individualising Care (POETIC, CRUK/07/015) and Alliance for Clinical Trial in Oncology (ACOSOG) Z1031B studies, where patients were treated for 2 to 4 weeks before surgery (13–15). Although informative, these studies do not address the long-term effect of NAI therapy that may be necessary to evaluate the full impact of AI-induced phenotypic/genotypic alterations (16) or the effects that might limit response and lead to clinical resistance.

Here, we report a detailed study of the molecular alterations associated with extended (>4 weeks) NAI treatment in the index primary ER⁺ breast cancer and show that ESR1 mutations are enriched with longer duration of therapy and become a key mitogenic driver. Using the validated proliferation marker Ki67 to identify endocrine-resistant tumors (17), we show that ESR1 wild-type (ESR1^Wt) tumors with high residual proliferation after NAI therapy appear partially independent of “classical ER signaling,” highlighting the high degree of heterogeneity in adaptive mechanisms circumventing E-deprivation.

We retrospectively identified 109 postmenopausal women diagnosed with primary ER⁺ breast cancer and treated with NAI for at least 1 month at the Royal Marsden Hospital (RMH) between 2003 and 2016. Inclusion criteria included the requirement for generic consent to conduct tissue-based research and the availability of both the diagnostic core-biopsy and paired surgical excision post-NAI. Exclusion criteria were (i) multifocal disease; (ii) previous breast cancer in a period of 10 years; (iii) involvement in a neoadjuvant clinical trial; (iv) concomitant anticancer treatments including chemotherapy, biological response modifiers, endocrine therapy (including steroids) and radiotherapy. Paired biopsies with >40% invasive cell areas were available from 87 patients (Supplementary Fig. S1).

Clinical and histologic details are shown in Supplementary Table S1. Given the focus on the molecular characteristics in the excision, clinical response by RECIST (17) was characterized based on ultrasound changes between start of AI and surgery.

All the research was carried out in accordance with the provisions of the Declaration of Helsinki of 1975. Ethical approval for the study was received from an NHS research ethics committee (reference 17/EM/0145) and patients had to have given consent for their tissues to be used for ethically approved research.

IHC for ER, PgR, and Ki67 was performed as previously described (18, 19). H-Score was used for the assessment of ER and percentage positivity for PgR. Ki67 percentage positivity was scored as a continuous variable, according to the method described by the International Ki67 Working Group (20). Ki67 proliferation was used as an end-of-neoadjuvant-treatment endpoint to identify index tumors that are endocrine resistant (17).

Formalin-fixed paraffin-embedded (FFPE) tissue sections were microdissected before coextraction of RNA and DNA using the AllPrep DNA/RNA FFPE Kit (Qiagen), according to the manufacturer's instructions with the exception of an extended overnight digestion for the DNA extraction. Nucleic acid quantification was done using high-sensitivity RNA and DNA Qubit assays (Thermo Fisher Scientific).

Gene expression was evaluated using nCounter FLEX Analysis System (NanoString Technologies) with two panels (744 and 106 genes, including 30 in common; Supplementary Table S2). The panels included reference genes, PAM50 gene set (panel #1), and genes involved in the most important aspects of breast cancer or with evidence of an association with AI resistance, including ERGs, proliferation, invasion, growth factor receptors, PI3K–AKT–mTOR pathway, MAPK signaling, cholesterol metabolism, inflammation, and epithelial mesenchymal transition (EMT) genes. For three patients, gene-expression analysis was conducted using only the smallest panel due to the low availability of RNA.

Raw counts were normalized by NanoStringNorm package in R (21). Briefly, the geometric mean of the counts from the six External RNA Controls Consortium (ERCC) positive controls were used to take into account the efficiency of the hybridization. Background correction was done by subtracting the geometric mean of the nine ERCC-negative control probes. Data were scaled and normalized by nine reference genes (Supplementary Table S2), which were confirmed as representative of the lowest combined variation across the studied samples. Expression values were log2 transformed for statistical analysis.

Tumors were classified into one of the intrinsic subtypes (luminal A, luminal B, Basal-like, and HER2-enriched) based on the PAM50 classifier algorithm (22, 23). ERGs expression was defined as the mean of TFF1, GREB1, PDZK1, and PGR (24) and proliferation-associated gene (PAGs) expression as the mean of 11 proliferation genes in the PAM50 gene set (BIRC5, CCNB1, CDC20, CDCA1, CEP55, KNTC2, MKI67, PTTG1, RRM2, TYMS, and UBE2C). An E2F activation metagene was developed based on the 24-gene E2F signature devoid of cell-cycle–associated genes described by Miller and colleagues (ref. 25; Supplementary Methods).

Seven hotspot ESR1 mutations within the ligand-binding domain were evaluated by droplet-digital PCR (ddPCR). Initially, we screened all post-AI samples using two multiplexed reactions for the following mutations: (i) E380Q, L536R, Y537C, and D538G; (ii) S463P, Y537N, and Y537S. ddPCR was performed with 5 ng of DNA on an automated droplet generator and QX100 system (Bio-Rad). The results were validated using singleplex ddPCR. ESR1 mutations were also assessed in the pre-AI samples from those patients with a mutation in their residual tumor. ESR1 mutation was considered positive with at least two mutant droplets detected. Mutation allele fraction was calculated as previously described (26).

Tumors with variant allele frequency (VAF) <1% were validated by ddPCR after fluorescent activated cell sorting (FACS) to enrich the number of cytokeratin-positive neoplastic cells evaluated (Supplementary Methods). This approach was also used to confirm the lack of detectable mutations in pre-NAI samples.

ESR1 copy number was evaluated by fluorescence in situ hybridization (FISH) in residual tumors harboring ESR1 mutations and in one pretreatment sample. Dual color FISH probes hybridizing at 6q25 (ESR1) and chromosome-6 (CEN6) were applied (ZytoLight). Briefly, 4-μm FFPE sections were deparaffinized and incubated for 20 minutes in Target Retrieval Solution Citrate pH 6.1 (Agilent) at 98°C, followed by pepsin digestion for 15 minutes at 37°C and RNase A treatment for 30 minutes at 37°C. Codenaturation was performed for 10 minutes at 75°C followed by hybridization for 24 hours at 37°C. Sections were mounted in DAPI-containing Vectashield (VectorLabs) and analyzed using fluorescence microscopy (Leica Biosystems).

FISH scoring was performed by counting 40 representative nonoverlapping nuclei. Average copy-number ratio ESR1/CEN6 was assessed. A ratio ≥2 was rated amplification and ≥1.3 as copy-number gain (27).

All analyses were performed using R v3.4.4. P value < 0.05 was considered statistically significant. For gene-expression analysis involving multiple comparisons, false discovery rate (FDR) was applied as indicated. GSEA was run using the GSEA v.3.0 software (http://software.broadinstitute.org/gsea) with 1,000 permutations.

Paired pre- and post-NAI therapy tumors were available from 87 ER⁺ breast cancer patients (Supplementary Fig. S1; Supplementary Table S1), in which key biomarkers ER, PgR, HER2, and Ki67 were assessed by IHC, together with the expression of 820 genes (Supplementary Table S2) using NanoString technology (Fig. 1A).

The mean age was 72.1 years (ranging between 50 and 93); 58 (66.7%) of patients had grade 2 and 23% had grade 3 ER⁺ tumors; 63 (72.4%) were ductal subtype (Supplementary Table S1). PgR expression was detected in 69 patients (79.3%) and 5 (5.7%) were classified as HER2⁺ by IHC and FISH (Fig. 1B; Supplementary Table S1). Based on intrinsic subtypes (n = 84), four (4.6%) tumors were HER2 enriched and two (2.3%) were basal-like (Fig. 1B; Supplementary Table S1).

The mean ± standard deviation duration of treatment was 26.3 ± 16.2 weeks (Fig. 1A and B). Three patients (3.4%) received second- or third-line endocrine therapy after AI. One (1.1%) patient showed a complete response (CR) as measured by ultrasound, 55 (63.2%) partial response (PR), 13 (14.9%) stable disease (SD), and 5 (5.7%) progressive disease (PD; Fig. 1B). Among the PRs, six (12.7%) showed >20% increased tumor volume before surgery compared with previous ultrasound, with five being treated for less than 32 weeks and one for 70 weeks. Among those with SD, six (46.2%) showed an initial objective response to AI therapy, which was subsequently followed by an increase of tumor volume.

Clinical response was not associated with clinical, pathologic or protein biomarkers tested at diagnosis (pre-NAI) or surgery (post-NAI; P > 0.05, χ² test or t test) with the exception of pre-NAI PgR levels (P = 0.007; t test) and expression of E-regulated genes (ERGs; P = 0.019; ref. 24) that were lower in SD/PD in comparison with CR/PR (Supplementary Fig. S2). In post-NAI tumors, proliferation-associated genes (PAG) were higher in SD/PD compared with CR/PR (P = 0.013, t test; Supplementary Fig. S2). As expected, Ki67 abundance correlated with its transcript level and also with PAGs (r = 0.59–0.77; P < 0.001; Pearson correlation; Supplementary Fig. S3A and S3B). Based on this observation and the wealth of the data supporting the use of residual Ki67 (Ki67r) as a biomarker of benefit from AI in the adjuvant setting (13, 15), Ki67r by IHC was used as a measure of response in this study.

Assessment of the on-treatment change in Ki67 (mean reduction: 21.7%) and PAGs (3.8-fold) showed that overall the majority of tumors responded to NAI at surgery (P < 0.001; paired t test; Fig. 2A; Supplementary Fig. S3A and S3B), with 55 (63.2%) tumors showing complete cell-cycle arrest (28) based on Ki67r (low-Ki67r; ≤2.7%) and 15 (17.2%) high residual proliferation (high-Ki67r; Ki67r ≥ 10%).

In further confirmation of the response to NAI, PgR/PGR and ER/ESR1 were significantly suppressed on therapy (protein mean reduction: 41% and 21%, gene-expression reduction: 4.4- and 1.8-fold, respectively; P < 0.001; paired t test; Fig. 2A; Supplementary Fig. S3A), with two (2.3%) tumors classified as ER⁻/PgR⁻ and 64 (73.6%) ER⁺/PgR⁻ post-NAI. In keeping with this observation, ERGs post-NAI were similarly suppressed (6.8-fold; P < 0.001; Fig. 2A; Supplementary Fig. S3C). There was no clear difference between HER2⁻ and the small number of HER2⁺ (5.7%) tumors with regard to these biomarkers (Figs. 1B, 2A, 3A; Supplementary Figs. S3 and S5A).

In addition, analysis of intrinsic subtypes showed that most of tumors were phenotypically luminal-A–like post-NAI (Figs. 1A and 2B), and gene set enrichment analysis (GSEA) confirmed the inhibition of genes involved in E-response and proliferation, including E2F targets (Fig. 2C). Most of tumors also showed a significant reduction of an E2F activation signature (ref. 25; P < 0.001, paired t test; Supplementary Fig. S4), which was also associated with SD/PD in post-NAI tumors (P = 0.022; t test; Supplementary Fig. S4).

Comparison of gene expression between pre- and post-NAI revealed that 554 genes were differentially expressed (FDR 5%; Supplementary Table S3). Hierarchical clustering based on the changes in expression of these genes with >25% change (410 genes) separated tumors into four main branches labeled A–D (Fig. 2B). Branches A and B showed less inhibition of cell-cycle genes and of genes involved in E-response and contained 5 of 9 patients with recurrence. Branch B also showed less inhibition of genes involved in immune response, focal adhesion, MAPK, and cytokine signaling and in this aspect was distinct from the other branches. Notably, this branch was enriched of tumors with poor response based on clinical response and with Ki67r ≥10% and was also enriched for tumors with post-NAI ESR1 mutations described in more detail below. Branch C was characterized mainly by downregulation of E-related and PAGs, but also by upregulation of genes involved in immune response (expanded immune pathways are shown in Supplementary Table S4), and making this branch was distinct from D. Overall, branch C tumors changed more with treatment based on the intrapatient correlation score. Branch C contained three of the four patients with late distant recurrence (≥5 years). Both branches C and D showed greater upregulation of genes associated with focal adhesion, MAPK, and cytokine signaling compared with branches A and B.

Twelve (13.8%) tumors showed increased expression of ERGs after treatment (Fig. 3A), which led us to investigate the presence of ESR1 mutation in post-NAI tumor samples. Seven ESR1 hotspot mutations were identified in six tumors (Figs. 1B and 3A; Supplementary Table S5): five with D538G mutation (VAF: 0.2%–27.6%), one with Y537N/D538G (VAF_Y537N: 12.3%; VAF_D538G: 27.6%) and one with Y537S (VAF: 17.3%). Tumors with VAF <1% were validated by ddPCR after enrichment of neoplastic cells (cytokeratin positive) using FACS; Supplementary Methods; Fig. 3B).

Noteworthy, one case harbored a D538G ESR1 mutation pre-NAI (VAF: 2%), which was further enriched in the post-NAI sample (VAF: 19.3%; Supplementary Table S5). To further determine if ESR1 mutations could be detected in the pre-NAI biopsies, we used FACS to enrich tumor cells in the other five pre-NAI tumors from patients harboring post-NAI mutation and were able to extract enough DNA to test for the presence of ESR1 mutations by ddPCR in four specimens. No ESR1 mutation was detected by doing this.

None of the patients with ESR1-mutant (ESR1^Mut) tumors were among the ones treated with a second- or third-line therapy presurgery. Only one (1.1%) patient showed SD and local recurrence, which had a VAF in the recurrence similar to that detected in the primary tumor (VAF_post-NAI: 24.3%; VAF_recurrence: 22.1%; Supplementary Table S5). The remaining five patients achieved PR prior to surgery.

In pre-NAI samples, ERGs (Supplementary Fig. S5A), Ki67, PAGs, and ESR1 (Supplementary Table S6) expression did not differ between ESR1^Mut and ESR1^Wt tumors (P > 0.05). However, ESR1^Mut tumors showed less suppression of ERGs (P = 0.002, Mann–Whitney test) and PAGs (P = 0.039) and greater ESR1 (P = 0.016; Supplementary Tables S7 and S8) expression post-NAI compared with ESR1^Wt tumors (Fig. 3A and C).

We further accessed the number of ESR1 copies by FISH in the residual ESR1^Mut tumors (Supplementary Table S5; Fig. 3D) and found one case (VAF: 0.23%) presenting copy-number gain (>1.3 ESR1/CEP6 ratio). Despite the copy-number gain, this patient showed a reduction in the expression of both ERGs and proliferation after NAI probably reflecting the large majority of ESR1 being wild-type; however, increased on-treatment ESR1 expression was detected (fold change: 1.96). Additionally, copy-number analysis of the pre-NAI sample from this patient confirmed that the gain of ESR1 copies was acquired with treatment; however, three copies of chromosome 6 were evident in both pre- and post-NAI samples. Based on the frequencies, our data suggest that copy-number gain preceded the mutation.

GSEA showed a lack of inhibition of E-response and less inhibition of proliferation-related pathways (including E2F targets) in ESR1^Mut tumors in comparison with ESR1^Wt (Fig. 3E). The E2F metagene was similarly less suppressed in ESR1^Mut (P = 0.016, Mann–Whitney test; Fig. 3F). Furthermore, dependence on ESR1 as a mitogenic driver was confirmed by the observation that several genes linked with ER signaling, including CCND1, RET, and FOXM1 (P = 0.023–0.047, Mann–Whitney test; Supplementary Fig. S5B), showed smaller change in response to NAI (Supplementary Tables S7 and S8).

Of particular note, post-NAI ESR1^Mut tumors were treated for longer with NAI in comparison with ESR1^Wt (P = 0.011; Mann–Whitney test; Fig. 3G). Furthermore, all five acquired mutations occurred in the third of patients with the longest duration of NAI third tertile: > 191 days; >6 months), with a prevalence of 5/29 (17%) in this tertile. Taken together, these data support that ESR1 mutations are enriched with extended NAI treatment in primary ER⁺ breast cancer.

In order to identify putative resistance mechanisms independent of ESR1 mutation, we analyzed the molecular changes associated with high Ki67r in tumors harboring ESR1^Wt (Supplementary Tables S9–S11). Overall, the expression profile between pre-NAI and post-NAI samples from tumors with high-Ki67r changed less than those from tumors with low-Ki67r (P = 0.023, t test; Figs. 4A and Fig. 2B). Moreover, tumors with high-Ki67r tended to maintain their baseline intrinsic subtype (Fig. 2B). Both these results are consistent with the molecular phenotype of these responsive tumors being refractory to the NAI therapy.

As expected, higher baseline expression of ERGs was correlated with reduced proliferation after treatment (P < 0.001, r = −0.38; Supplementary Fig. S6), highlighting their dependence on ER signaling as the main mitogenic driver. Conversely, high Ki67r was associated with less inhibition of ERGs (P = 0.012; t test; Fig. 4B; Supplementary Fig. S6), which was paralleled by less reduction in PgR abundance (P = 0.023; Fig. 4B). Furthermore, genes involved either directly or indirectly in cell-cycle control were less inhibited in tumors with high-Ki67r in comparison with low-Ki67r (Supplementary Fig. S7), including genes regulated by E, such as CCND1 and RET (P = 0.01 and P = 0.011, respectively; Fig. 4C; Supplementary Table S11). In addition, an ER-dependent E2F activation signature (25) was less inhibited in high-Ki67r tumors (P = 0.002–0.031; t test, Fig. 4D). Moreover, tumors with high-Ki67r did not show significant inhibition of pathways involved in E-early and -late response (Fig. 4E).

Further interrogation of the data showed that tumors with high-Ki67r had lower ESR1/ER expression/abundance at diagnosis (pre-NAI) compared with those with low-Ki67r (ESR1: P = 0.044; ER: P = 0.013 t test; Fig. 4F; Supplementary Table S9). This observation was paralleled by the lower ER abundance in high-Ki67r tumors compared with low-Ki67r and medium-Ki67r (2.7% > Ki67r < 10%) tumors post-NAI (P = 0.021, P = 0.025, respectively; Fig. 4F). These findings endorse the hypothesis that a subset of ER⁺ tumors are less dependent on classic ER signaling at diagnosis.

Although there was a high degree of similarity in the gene-expression profiles at diagnosis irrespective of Ki67r (FDR >10% for all genes; Supplementary Table S9), two key genes involved in the regulation of cell proliferation and inflammation (Supplementary Table S12), CDK2 (P_pre-NAI = 0.028, 1.3-fold; P_post-NAI = 0.001, 1.4-fold, t test in relation with low-Ki67r) and FGFR4 (P_pre-NAI = 0.007, 4.34-fold, P_post-NAI = 0.013, 3.93-fold), showed higher expression in those tumors with high-Ki67r at both time points investigated (Fig. 5A). In addition, both CDK2 and FGFR4 showed higher expression in tumors of patients with SD/PD in comparison with CR/PR in both pre-NAI (P = 0.017; P = 0.012, respectively; t test) and post-NAI (P = 0.017, P = 0.007, respectively; Fig. 5B).

In this study, we focused on understanding mechanisms of resistance that emerge in primary ER⁺ breast cancer treated with extended NAI therapy and the importance to evaluate paired pre- and post-treatment biopsy. Our study is the first to show the early enrichment of ESR1 mutation in the neoadjuvant setting. Here, we provide further insights into early mechanisms of endocrine resistance, which may inform on combination treatment either before or after surgery or in recurrent disease.

Overall, we observed that cell proliferation was suppressed in most tumors, an observation paralleled by the reduced expression of E-responsive genes. At surgery, tumors tended to be more phenotypically luminal-A–like, but the degree to which this occurred varied across the patient cohort, emphasizing the high degree of heterogeneity.

ESR1 mutations were identified in 7% of tumors within the study and were enriched among patients treated with NAI for more than 6 months. This is consistent with previous studies, which have shown that ESR1 mutations appear enriched almost exclusively in response to AI therapy (8–11, 29, 30). However, in this study, the reported frequency was lower than that noted in metastatic ER⁺ breast cancer patients who had relapsed on AI therapy (11%–54%; refs. 8–11, 29, 30). Thus, our results indicate that the selection of ESR1 mutations can occur frequently in primary as well as metastatic disease.

Nonetheless, ESR1 mutations were detected at a higher frequency in our cohort compared with a previous study of NAI therapy, which reported 1.5% (16). One explanation for this discrepancy is the difference in technologies used to call the mutation status. In the latter study, mutations were identified by exome sequencing, while we used targeted ddPCR and microdissected tumors, allowing identification of VAFs as low as 0.2%, which we also supported using FACS. We are the first to describe a temporal association of duration of AI as first-line treatment in a neoadjuvant setting with enrichment of ESR1 mutation. Our findings support our previous studies in ER⁺ breast cancer cell lines that demonstrated enrichment of ESR1 mutations with time post E-deprivation (31). Further study of the emergence of ESR1 mutations during NAI therapy may improve our understanding of the tissue dynamics that underpin clinical relevance of treatment-dependent clonal selection during extended E-deprivation.

Our findings provide evidence that ESR1 mutations are the mitogenic driver of AI resistance. Thus, tumors harboring a ESR1 mutation in their residual disease showed activation of genes involved in E-response and of pathways associated with proliferation, highlighted by smaller change of the E-regulated CCND1, the ESR1 coactivator FOXM1 (32) together with downstream E2F targets. Moreover, ESR1^Mut tumors showed increased expression of the oncogene RET, which has previously been associated with ligand-independent ER activity (33). Our findings provide functional evidence for the gain of ESR1 mutations being a bona fide resistance mechanism to AI. These data provide further support for recent in vitro characterizations of ESR1 mutations, which show that these mutations govern an altered cistrome leading to the engagement of E2-independent—ER-driven transcriptional programs (31, 34). They also support for the concept that selective ER downregulators (SERD) or combination of AI with CDK4/6 inhibitors may provide greater benefit than AI alone in the adjuvant setting for patients with primary ESR1^Mut ER⁺ breast cancer.

The present study was focused on molecular changes that underpin response in the index tumor and to gain a better understanding of the de novo and acquired resistance mechanisms as opposed to deriving a direct tool that predicts clinical response. There is strong evidence to support Ki67 as the primary endpoint of neoadjuvant endocrine therapy study from multiple previous clinical trials [Preoperative Anastrozole, Tamoxifen, or Combined with Tamoxifen (IMPACT); P024 study; American College of Surgeons Oncology Group (ACOSOG) Z1031] and the level of Ki67 after treatment had been associated with recurrence-free and overall survival (17, 31, 35). Noteworthy, clinical response per se is poorly related to recurrence risk on adjuvant endocrine therapy, in contrast to pCR with chemotherapy in ER⁻ and HER2⁺ disease. Moreover, Ki67 can be used as a marker for endocrine-resistant tumor to discriminate patients requiring more aggressive treatment (15).

We identified a subgroup of ESR1^Wt tumors in which proliferation remained high after NAI therapy with less inhibition of classic and nonclassic ERGs. Although ER⁺, tumors with higher Ki67r showed lower ESR1/ER expression at diagnosis, confirming that tumors with decreased dependence of ER gain less benefit from AI therapy. Interestingly, previous clinical studies have shown patients with higher levels of ER abundance measured by ligand-binding assays gained greater benefit from tamoxifen in the adjuvant setting (2). Indeed, the measure of ESR1/ER expression may also help in the prediction of patients who would gain greater response with extended NAI (36). Moreover, the decreased dependence on ER signaling associated with high expression of several cyclins and E2F targets support, as noted above, that patients with this phenotype may benefit from the combined use of an endocrine agent with a CDK4/6 inhibitor targeting the RB/E2F regulon.

Notably, RET expression appeared to increase in tumors with high residual proliferation and decreased in tumors gaining greater benefit from AI therapy. As noted above, RET has been linked with resistance to therapy and its potential as a therapeutic target has been suggested (37, 38).

Tumors showing less dependence on ESR1/ER signaling at baseline and higher residual proliferation also showed increased expression of cell-cycle control and immune response genes pretreatment. In this context, high expression of CDK2 was evident in tumors with high Ki67r in both pre- and post-NAI therapy. CDK2 may act (i) as a direct mitogenic driver or (ii) to phosphorylate ER, leading to ligand-independent ER signaling (35, 39). Although CDK2 may be a contributing factor for AI resistance in primary ER⁺ breast cancer, it is important to note that only minimal differences were observed at the transcriptional level between groups based on Ki67r. Although CDK2 is an obvious therapeutic target, no specific inhibitors have thus far been developed for clinical use.

Similarly, the growth factor receptor FGFR4 showed higher expression in pre-NAI tumors with high Ki67r in our cohort. FGFR4 can stimulate the proliferation of breast cancer cells via an ER-independent mechanism (40, 41). Furthermore, increased expression of FGFR4 has previously been associated with poor response to tamoxifen (42, 43). FGFR4 is a possible targetable alteration (DGIdb: http://www.dgidb.org/) and FGFR4 inhibitors (FGF401, H3B-6527, and BLU554) have been tested in phase I and II trials targeting other cancer types, such as in the trials NCT02508467, NCT02834780, and NCT02325739.

Tumors with high Ki67r are a very heterogeneous group in our study; however, our findings show that tumors with acquired resistance exhibit smaller changes in gene expression compared with sensitive tumors and that these tumor therefore more closely resemble their diagnostic samples, an observation in keeping with the study from Selli and colleagues (44). Taken together, this supports the notion that the presurgical exposure of ER⁺ tumors to AI markedly enhances the ability to reveal their dependence on classic ER signaling and therefore identify mechanisms of resistance.

It is important to underscore that four cases with recurrences were observed in the branch C of our hierarchical cluster analysis in which tumors with the greatest inhibition of proliferation and estrogen signaling were grouped. We and others previously reported that, although patients whose tumors were rated as more highly estrogen responsive at diagnosis had a lower risk of recurrence up to 5 years, their risk was greater with further follow-up after such treatment discontinuation (40, 41). This is consistent with such patients only showing lower recurrence rates when the disease is controlled by endocrine therapy.

Although our study had biological strength in tumors phenotypic characterization, some limitations should be noted. First, only about half of the patients had a follow-up of more than 5 years, impairing our ability to directly link phenotypic/genotypic alterations with risk of recurrence. Historically, NAI treatment has been selected for postmenopausal women with large ER⁺ tumors or for those who may be too frail to undergo surgery. This patient population is often older, with limited long-term follow-up (12). Second, although we have a representative cohort of ER⁺ breast cancer treated with NAI, the largest to date with extended NAI (4, 16), subgroup analysis was restricted due to lack of statistical power. Nonetheless, a significant strength of the study was our access to both pre- and post-NAI samples, which enabled us to conduct comparative gene-expression profiling and mutation analysis to define the acquisition of ESR1 mutations.

In summary, overall, most tumors showed little evidence for the emergence of resistant disease after NAI therapy, highlighted mainly by the continued reduced expression of proliferation genes/proteins and several genes involved in E-response. Two main groups of tumors showing possible resistance to long-term NAI therapy were observed: (i) tumors with ESR1 mutations that were enriched with longer exposure to AI and (ii) ESR1^Wt tumors with relatively low ESR1 expression at diagnosis and high Ki67r. In both groups, ligand-independent ER signaling was detected, and it can be used to inform on subsequent adjuvant treatment in early ER⁺ breast cancer.

L.-A. Martin reports receiving speakers bureau honoraria from Pfizer and reports receiving commercial research grants from Pfizer, Radius, and PUMA. M.C.U. Cheang is a co-inventor for the US Patent No. 9,631,239 (Method of classifying a breast cancer intrinsic subtype) which is licensed to NanoString for the commercial assay named Prosigna. M. Dowsett is a consultant/advisory board member for and reports receiving commercial research support from Radius. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.F. Leal, B.P. Haynes, M. Dowsett

Development of methodology: M.F. Leal, B.P. Haynes, E. Schuster, A. Dodson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.F. Leal, B.P. Haynes, B. Yeo, L. Zabaglo, V. Martins, I.E. Smith

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.F. Leal, B.P. Haynes, E. Schuster, V. Martins, R. Buus, M.C.U. Cheang

Writing, review, and/or revision of the manuscript: M.F. Leal, B.P. Haynes, E. Schuster, L. Zabaglo, R. Buus, M.C.U. Cheang, L.-A. Martin, M. Dowsett

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.F. Leal, M. Afentakis, A. Dodson

Study supervision: M. Dowsett

Other (discussion of findings): I.E. Smith

The authors are thankful for Ricardo Ribas' critical reading. This study was supported by fellowship from Le Cure to M.F. Leal and by a grant from The Breast Cancer Research Foundation. We thank Breast Cancer Now for funding this work as part of Programme Funding to the Breast Cancer Now Toby Robins Research Centre. We acknowledge support from the National Institute for Health Research through the National Cancer Research Network and the Royal Marsden/The Institute of Cancer Research Biomedical Research Centre.

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.