Exploiting androgen receptor agonism as a treatment strategy in estrogen receptor-positive metastatic breast cancer

Abstract

The androgen receptor (AR) is expressed in 75% of estrogen receptor-positive (ER+) breast cancers (BC). Selective AR modulators (SARMs), like EP0062, present a promising therapeutic strategy for ER + BC, particularly in patients who cannot tolerate endocrine therapy (ET) or whose tumors have developed resistance. We aimed to study the antitumor activity of EP0062 in ER+ patient-derived xenograft (PDX) models. EP0062 displayed comparable antitumor efficacy to selective ER degraders (SERDs), including in PDXs with ESR1, PIK3CA, or PTEN mutations. Tumors sensitive to SARMs were enriched in GATA3 mutations. EP0062 treatment induced AR-target genes across all models tested. A transcriptional signature associated with SARM sensitivity was identified, primarily driven by proliferation-related processes, consistent with a significant decrease in S-phase cell cycle proteins upon treatment in EP0062-sensitive models. In some EP0062-resistant tumors, the combination with palbociclib enhanced the antitumor effect of EP0062, suggesting a potential strategy for metastatic patients with acquired ET resistance.

Introduction

Endocrine therapy (ET) targeting the estrogen receptor (ER) signaling pathway is the main treatment strategy for early and advanced ER-positive (ER+) and/or progesterone receptor-positive (PgR+) breast cancer (BC). Despite the significant impact of ET on improved survival in early BC, up to 20% of these patients eventually relapse with metastatic disease¹. In addition, most patients with ER+ metastatic BC (mBC) initially respond to ET, but the acquisition of resistance is almost inevitable. ET-resistance can be driven by ER-independent mechanisms, such as the crosstalk with growth factor receptors or oncogenic signaling pathways like PI3K or MAPK pathways and cell cycle dysregulation². However, it can also occur by ER-dependent mechanisms, e.g., by acquired alterations in ESR1 which arise in 20 to 40% of patients treated with aromatase inhibitors (AIs) in the metastatic setting^3–5. ESR1 mutations lead to constitutive ER signaling in the absence of an activating ligand, conferring partial or total resistance to anti-estrogens^3,4,6,7. In recent years, several oral selective estrogen receptor degraders (SERDs) have shown promising preclinical data regarding ER-degradation and antitumor activity in ER + BC models, including tumors harboring ESR1 mutation^8–13. In the clinic, elacestrant and imlunestrant are currently approved for ER+/HER2− ESR1-mutant mBC patients who have progressed to at least one line of ET^14,15. Camizestrant has recently shown superiority in terms of progression-free survival (PFS) compared with continued AI therapy in combination with cyclin-dependent kinases 4 and 6 inhibitors (CDK4/6i) in patients with ER+/HER2− BC who acquired an ESR1-mutation in the first-line setting, while the phase III trial in the first-line is still ongoing^16,17. Giredestrant has recently shown significant improvement in invasive disease-free survival (iDFS) compared with the current standard-of-care ET in patients with early-stage, ER + /HER2− BC¹⁸, while the activity in combination with CDK4/6i is being evaluated in a phase III clinical trial¹⁹. CDK4/6i combined with ET are the standard of care treatment for patients with ER+/HER2− mBC, based on results of phase III clinical trials, where CDK4/6i in combination with ET has shown improvement in PFS and even overall survival (OS) in comparison with ET plus placebo²⁰. Still, some patients have intrinsic resistance to CDK4/6i plus ET, and eventually, most patients develop acquired resistance.

The androgen receptor (AR) is co-expressed with ER in 75% of ER + BC tumors and has emerged as a potential therapeutic target given its tumor suppressor role in ER + /HER2- BC^21,22. Even before the discovery of selective ER modulators (SERMs) and AIs, steroidal androgens such as fluoxymesterone were used to treat BC²³. Importantly, selective androgen receptor modulators (SARMs) that activate AR in a tissue-specific manner do not exhibit virilizing effects and have been proposed as a therapeutic strategy for ER + BC^24,25. In addition to their antitumor activity, SARMs exhibit beneficial effects in bone and muscle, e.g., preventing osteoporosis and cachexia²⁶. Mechanistically, it has been proposed that agonist activation of AR alters the genomic distribution of ER and essential co-activators, resulting in the upregulation of AR target genes, including known tumor suppressors and further repression of ER-regulated cell cycle genes²¹. However, the antitumor activity of SARMs in ET-resistant tumors remains to be explored. Additionally, SARMs' efficacy can be increased with a CDK4/6i^21,22, but the comparison with the standard of care in ET + CDK4/6i-resistant tumors has not been reported so far. Two SARMs, EP0062 and enobosarm, are currently being developed and have shown clinical benefit in patients with ET-resistant BC with AR expression^27,28. Indeed, results from the phase II clinical trial with enobosarm demonstrated that patients with 40% AR-positivity had the best overall response rate²⁷, but no additional biomarkers of response to SARMs have been proposed so far. We selected the SARM EP0062 because it is currently in phase 1/2 clinical development (NCT05573126) and actively recruiting patients at our institution, making it a clinically relevant compound for translational studies. In this work, we used a panel of 45 ER + BC patient-derived xenografts (PDXs) to study if EP0062 exhibits an antitumor activity as potent as ET alone or combined with CDK4/6i. We further performed genomic, transcriptomic, and protein analyses in 12 ER + PDX models with ER/AR expression, to unveil the mechanism underlying the sensitivity to the SARM.

Results

ER + PDX models exhibit different sensitivity to hormone deprivation in vivo

To study the activity of the SARM EP0062 in ER-positive (ER+) PDXs, we developed and characterized a panel of 45 ER + / human epidermal growth factor receptor 2 (HER2−) PDXs derived from 37 patients. 76% of the patients had received ET, and 27% had been exposed to cyclin-dependent kinases 4 and 6-inhibitors (CDK4/6i) (Supplementary Table 1). The median of prior lines in the metastatic setting was 2 (95% confidence interval 2–3). According to ABC7 guidelines²⁹, twenty-six out of 37 (70%) patients were endocrine-sensitive, whereas 7 patients (19%) showed a primary resistance, and 4 patients were ET-naïve (11%). All PDX models recapitulated the patients’ tumor clinicopathological subtype classification, preserving hormone receptor expression by immunohistochemistry (IHC) (ER and/or progesterone receptor; PgR, Supplementary Table 2). In addition, PAM50 subtyping identified 3 HER2-enriched (7%) and 3 basal-like models (7%) within the ER + PDX panel, a similar proportion to the expected for this population³⁰ (Fig. 1A). We employed a capture-based sequencing platform, MSK-IMPACT^TM, to detect genetic alterations in >505 cancer-related genes and we identified TP53 (49%), CCND1 (42%), FGFR1 (33%), PIK3CA (36%), GATA3 (36%) and ESR1 (20%) as the most frequently altered genes in the PDX cohort (Supplementary Fig S1A). PDXs harboring ESR1 missense mutations were derived exclusively from patients having progressed to treatment with aromatase inhibitors in the metastatic setting (Fig. 1B). The most frequent ESR1 missense mutations were p.Y537S and p.D538G, with a minor representation of p.Y537N, p.G415E, and p.E380Q variants (Supplementary Fig 1B). Analysis of RNA-sequencing data from untreated PDX tumors revealed 16 ESR1 fusion events, involving 10 different fusion partners, in 11 PDX models (Supplementary Table 3 and Supplementary Fig S1C). However, only three out of the 16 ESR1 fusion events preserved exon 1–6 of ESR1 (ESR1-e6 > IMPG1, ESR1-e6 > HNF1A, and ESR1- e6 > CMSS1), encoding for the AF-1 domain, the DNA-binding domain, and part of the ligand-binding domain, potentially resulting in a functional ER. The expression of the ESR1 fusion proteins was further confirmed in two PDX models, PDX433.3 and PDX329 (Supplementary Fig S1D, E).

Fig. 1. ER + PDXs exhibit different sensitivity to hormone deprivation in vivo.

A Sankey plot of concordance of patients’ tumor clinicopathological surrogate subtype, PDX clinicopathological surrogate subtype, and PDX molecular subtype (based on the PAM50 assay), n = 45 PDX models. In cases where Ki67 measurement was not available, the patient subtype is labeled as “Luminal-like”. Schematic illustrations were created using BioRender.com. B Bar plot showing the percentage of PDX models with ESR1-mutations derived from patients that i) did not receive AI (n = 15), ii) received AI in the adjuvant setting (n = 4), ii) received AI in both the adjuvant and metastatic setting (n = 2), and iv) received AI exclusively in the metastatic setting (n = 24). C Waterfall plot representing the best response (mean change from initial tumor volume) for n = 41 PDXs in ovariectomized (OVX, circles) mice compared to non-OVX control mice (squares). Dashed lines indicate the range of PD (>20%), SD (20% to −30%), PR (−30% to −95%), and CR (<−95%) according to the best response of each model. For resistant models (PD) and tumors from control mice, day 21 is shown. Bars indicate mean ± StDev for OVX-tumors, black lines show mean ± StDev for control tumors, symbols indicate individual tumors. Heatmap summarizes: i) ER, ii) PgR and iii) AR protein levels (IHC), iv) PDX molecular subtype (PAM50 assay), v) PI3K-pathway alterations (PIK3CA, AKT1 and PTEN), and vi) ESR1 alterations identified by MSK-IMPACT^TM panel or RNA-sequencing (for ESR1 fusions). Mann–Whitney two-tailed test compared ER, PgR, and AR levels in resistant vs sensitive models, p values are shown. Fisher’s exact test analyzed the association of mutations with hormone response. D, E. Quantification of the percentage of Ki67- (D) and ER-positive cells (E) in tumors of control vs OVX-mice from hormone-resistant (n = 35) and -sensitive PDXs (n = 4) by IHC at the end of the experiment. Representative images of a hormone-resistant PDX (PDX244) and a hormone-sensitive PDX (BB3RC31) are shown. Pictures at 200× magnification, scale bar: 100 µm. Each symbol represents the mean of at least 2 biological replicates per model. For models with CR, samples were collected at day 20 in a separate experiment. The proportion of positive cells in each model was analyzed with a generalized linear model. AI aromatase inhibitor, AR androgen receptor, ER estrogen receptor, CR complete response, IHC immunohistochemistry, PD progressive disease, PgR progesterone receptor, PR partial response, Pt patient, SD stable disease, StDev standard deviation, OVX ovariectomized.

To identify hormone-resistant PDX models for SARM testing, we compared tumor growth in ovariectomized (OVX) mice vs non-OVX controls in 41 available ER + PDX models. This approach allowed us to characterize hormone dependency and identify models resistant to hormone deprivation, where SARMs might exhibit antitumor activity despite resistance to ER-targeting therapy. Six models exhibited sensitivity to hormone deprivation: complete response (CR, n = 3), partial response (PR, n = 2), and stable disease (SD, n = 1). The remaining 35 PDX models were resistant to hormone deprivation (PD, progressive disease) (Fig. 1C and Supplementary Fig S1F). Regarding hormone receptor expression, a substantial fraction of hormone-resistant models had low levels of ER and/or PgR, as expected (Supplementary Fig S1G, H). As anticipated, the 6 PDX models with a non-luminal molecular subtype were hormone-resistant. PIK3CA/PTEN/AKT1-alterations (summarized as PI3K-pathway) were significantly associated with a lack of response to hormone deprivation (p = 0.006, Fig. 1C). Similarly, ESR1 alterations were numerically more frequent in the resistant group. We used Ki67 modulation as a biological surrogate of antitumor activity. A significant decrease in the proportion of Ki67-positive tumor cells was observed in tumors from OVX-mice compared to controls in both groups but the difference was more pronounced in sensitive compared to resistant models (interaction test, p < 0.001) (Fig. 1D). As expected, no effect on the percentage of ER-positive cells was observed in either hormone-resistant or sensitive PDXs (Fig. 1E). Overall, the PDX panel reflected the heterogeneity of ER+ mBC at the molecular and hormone sensitivity levels.

EP0062 has a similar antitumor activity to SERDs in ER+/AR + PDX models

The androgen receptor has emerged as an attractive therapeutic target in ER + BC. Here, we aimed to investigate the potential of SARMs to exhibit antitumor activity, with a focus on models with resistance to ER-targeting therapy. To do so, we first conducted short-term in vivo treatment experiments with EP0062 or dihydrotestosterone (DHT), a natural AR ligand which is not a substrate for the aromatase enzyme, thus cannot be aromatized to estrogen and is poorly anabolic in the musculoskeletal system, in 35 hormone-resistant PDX models (Fig. 2A). As expected, EP0062 increased nuclear AR-protein levels similarly to DHT in ER + /AR + BC PDX models (Supplementary Fig S2A–D). Eleven out of 35 (31%) PDX models exhibited a significant reduction in the percentage of Ki67-positive cells in response to either EP0062 or DHT, with a correlation between treatment groups (p < 0.001) (Fig. 2B, Supplementary Fig. S2E). To further study the antitumor activity of EP0062 in comparison with selective ER degraders (SERDs) fulvestrant and camizestrant, we selected PDXs with i) AR-expression, ii) clinically relevant alterations (ESR1, PIK3CA, AKT1, or PTEN), and iii) that exhibited a decrease in the percentage of Ki67-positive cells upon EP0062/DHT, along with matched controls (Supplementary Table 4). Treatment with EP0062 elicited antitumor activity as a single agent in seven models (1 with CR, 2 PR, and 4 SD) while the remaining five PDX models were resistant to the SARM (Fig. 2C and Supplementary Fig S2F). In addition, we observed a positive correlation between the antitumor response to the SARM and the percentage of Ki67-positive cells after EP0062 short-term treatment (p = 0.0013, Fig. 2D). Noteworthily, two models derived from the same patient before and after progression to AI + CDK4/6i, PDX476.1 and PDX476.2, respectively, displayed different responses to EP0062, being the pretreatment sample resistant to the AR modulator, likely due to the low levels of AR expression in this model (Supplementary Fig S2G). Among the seven EP0062-sensitive models, four were also sensitive to the SERD camizestrant (PDX376, PDX512, PDX476.2, PDX527, Fig. 2E). EP0062 had superior activity than the SERDs in three models (PDX423, PDX490, and PDX491), while two EP0062-resistant models (PDX244 and PDX476.1) were sensitive to camizestrant (Fig. 2C, E). Altogether, these results show that SARMs have not only overlapped, but also superior antitumor activity compared to SERDs in some PDX models.

Fig. 2. Treatment with EP0062 induces a potent antitumor activity in a subset of ER + PDXs.

A Workflow of the experiment. Tumor-bearing mice were OVX, and after 10 days, untreated control or treated with EP0062 6 mg/kg or DHT 10 mg/pellet for 5 days. Snap-frozen and formalin-fixed embedded tissues were collected. Schematic illustrations were created using BioRender.com. B Correlation plot of delta Ki67-levels by EP0062 and DHT vs control. Delta was calculated per the PDX model with at least 2 biological replicates per arm. Each dot represents a different PDX model, n = 35. Pearson’s correlation coefficient is shown. PDX models with a statistically significant variation in the percentage of Ki67-positive cells are shown in green dots. C Antitumor response of EP0062 (x-axis) vs camizestrant (y-axis) in 12 PDXs represented as the percentage of tumor volume change compared to the initial volume. D Heatmap of biomarkers in relation to EP0062 response in 12 ER+ PDXs: i) patients’ sensitivity to ET, ii) PDX molecular subtype by PAM50 subtyping, iii) hormone dependence, iv) alterations in ESR1, v) PI3K-pathway (PIK3CA, PTEN, AKT1), and vi) GATA3, vii) the percentage of nuclear ER-, viii) nuclear AR-protein levels and ix) Ki67-positive tumor cells upon EP0062-treatment. For ER, AR, and Ki67 measurements, n = 3 biological replicates were analysed per model (E) Venn diagram of sensitivity to EP0062 and camizestrant in 12 PDX models. F, G Percentage of nuclear ER or nuclear AR-positive cells by IHC (F) and its ratio (AR/ER) (G) in control tumors from EP0062-resistant (n = 5, grey circles) and EP0062-sensitive PDX (n = 7, pink). Each dot represents the mean of at least 3 tumors for each PDX. Bars represent the mean $\pm$SEM of resistant (grey) or sensitive (pink) PDX models. An unpaired two-tailed t-test was applied for (F), and a two-tailed Mann–Whitney test was applied for (G). H. Western blot of GATA3, frameshift GATA3 proteins, AR, and ER-protein levels in the 12 ER+ PDXs, ranked according to EP0062-sensitivity. Human GAPDH was used as a control. MCF-7 and T47D cell lines were used as a control of GATA3-mutant and GATA3-wild type, respectively. The GATA3 antibody also detects a non-specific band at ~45 kDa across all models. I. Viability assay of parental T47D cells and CRISPR-engineered GATA3 p.R330Efs mutant T47D cells to treatment with 0.01 and 0.1 µM of EP0062 for 7 days. Data is normalized to controls. Bars represent mean ± SEM, n = 4 independent experiments, one-way ANOVA with Tukey’s multiple comparisons test is shown. ANOVA analysis of variance, CR complete response, DHT dihydrotestosterone, IHC immunohistochemistry, PD progressive disease, PR partial response, SD stable disease, SEM standard error of the mean, µM micromolar.

GATA3 mutations are enriched in EP0062-sensitive PDX models

Preclinical and clinical evidence suggest that AR, ER, or a high AR/ER ratio at the protein level is associated with sensitivity to SARMs^22,27. Hence, we examined whether drug sensitivity was associated with protein level expression of these receptors in the PDX cohort. No significant differences were observed among resistant and sensitive models, neither in the percentage of ER-, AR-positive cells nor the AR/ER ratio at baseline (Fig. 2F, G). We also observed that PDX models harboring alterations in PIK3CA, PTEN, and AKT1 (summarized as PI3K-pathway) or in ESR1 were equally distributed in resistant and sensitive groups, suggesting that these mutations are not associated with EP0062 sensitivity (Fig. 2D). Interestingly, we identified heterozygous GATA3-frameshift mutations caused by small indels, in four out of seven EP0062-sensitive models. Notably, these four models were also camizestrant sensitive. These mutations were localized in exons 4 and 6 of GATA3 involving the ZnFn2 domain (p.D335Gfs*17) and the C-terminal part of the protein (p.S405Qfs*97, p.G431Wfs*76 and S437SP*39), in line with previous reports³¹ (Supplementary Fig. 2H). We confirmed that GATA3-variants were expressed at the expected molecular weights in all PDX models tested, compared to T47D (GATA3-wild type) and MCF-7 (GATA3-ZnFn2 mutant) (Fig. 2H).

To further explore the impact of GATA3 mutations on therapeutic response, we leveraged a GATA3-mutant model in T47D cells³¹. The GATA3 p.R330Efs-mutant T47D cells showed increased sensitivity to EP0062 and DHT compared to parental cells (p = 0.0373, p = 0.0074, respectively) (Fig. 2I, Supplementary Fig. S2I). These results suggest that endogenous GATA3 frameshift mutations, may alter SARM therapeutic response. Overall, our findings suggest that the proposed biomarkers associated with EP0062-sensitivity require further refinement and exploring additional biomarkers in clinical trials, beyond AR and ER protein expression, e.g., GATA3 status.

ESR1 downmodulation by EP0062 is partly associated with antitumor response

Preclinical studies suggested that the antiproliferative effect of SARMs is partially mediated by an upregulation of AR-target genes, which act as tumor suppressors²¹. Hence, we analyzed the expression of two AR-target genes with tumor suppression function (SEC14L2 and ZBTB16), in sensitive and resistant models by quantitative polymerase chain reaction (qPCR) and observed an induction of both of them regardless of EP0062 sensitivity (Fig. 3A, B). To study if SARMs also induce an ESR1 downmodulation that leads to a suppression of the ER program²⁴, we evaluated ESR1 expression levels after EP0062 or DHT treatment in the 12 PDX cohort. EP0062 and, more evidently, DHT reduced ESR1 expression in PDX models exhibiting a PR or CR (Fig. 3C and Supplementary Fig. S2J). Consistent with these results, we confirmed a concomitant decrease in ESR1 expression levels and ER-protein levels by DHT in the most sensitive models (PR + CR) (Fig. 3D and Supplementary Fig. S2K). Still, the decrease in ER protein levels by EP0062 in sensitive models (PR + CR) occurred to a lesser extent than the one observed for camizestrant (Fig. 3E, F). Overall, these results suggest that AR agonists can modulate ER levels, probably due to transcriptional regulation, but this effect might not be the exclusive mechanism of action in vivo.

Fig. 3. Induction of AR target genes and modulation of ER expression in PDX models treated with AR agonists.

A, B Induction of AR-target genes SEC14L2 (A) and ZBTB16 (B) upon treatment with EP0062 or DHT for 5 days in 12 PDX models, analyzed by qPCR. Expression data was normalized to ACTB and GAPDH housekeeping genes and to control samples. PDX models are ranked according to EP0062-sensitivity. Each symbol represents an individual tumor; bars represent the mean $\pm$StDev. C, D Modulation of ESR1 expression by qPCR (C) and of ER protein levels by Western blot (D) in samples of 12 PDX models treated for 5 days with EP0062 (red), DHT (orange), or control (grey). Expression data was normalized to ACTB and GAPDH housekeeping genes for expression and to control samples. Protein levels were normalized to human GAPDH (hGAPDH) and to control samples. Each symbol represents a different PDX model, which are classified according to sensitivity (PD, SD, and PR + CR). The mean $\pm$SEM of biological replicates is shown. E, F Quantification (E) and representative images (F) of the modulation of nuclear ER protein levels upon treatment with EP0062 (red) or camizestrant (dark blue) compared to control (grey) in 5 PDX models. Sensitivity to EP0062 and to camizestrant is shown. Pictures taken at 200× magnification, scale bar: 100 μm. One-way ANOVA with Dunnett’s multiple comparisons test is shown for all panels. ANOVA analysis of variance, CR complete response, DHT dihydrotestosterone, hGAPDH human GAPDH, PD progressive disease, PR partial response, qPCR quantitative polymerase chain reaction, SD stable disease, SEM standard error.

Transcriptional changes associated with SARM sensitivity in ER + BC

Next, we explored whether EP0062 sensitivity was associated with differential transcriptional changes, including AR-associated genes, upon treatment in three sensitive (PR + CR) and four resistant models (PD). Differential expression analysis of EP0062-treated vs control samples revealed a significant enrichment of the hallmarks E2F-targets, of AR activity, and of angiogenesis in sensitive but not in resistant PDXs (Supplementary Fig S3A–D). Integration of genes modulated by EP0062 in sensitive models identified a transcriptional signature associated with response, largely reflecting changes in cell cycle and proliferation-related processes (Supplementary Fig S3E, F and Supplementary Table 5). Notably, a similar pattern of gene downregulation was observed upon treatment with camizestrant in sensitive models, suggesting partially shared transcriptional effects (Supplementary Fig. S3F, G). The potential clinical relevance of this signature was explored in the METABRIC cohort, where it showed prognostic stratification comparable to established proliferation-based signatures (Supplementary Fig S3H and Supplementary Table 6), despite limited overlap with these signatures (Supplementary Fig S3I). Overall, these findings suggest that the transcriptional changes associated with SARM sensitivity are primarily driven by modulation of proliferation-related processes.

EP0062 can induce E2F1-pathway downmodulation

Since E2F-targets were the top enriched hallmark in the gene set enrichment analysis (GSEA) of sensitive models and E2F1 is an ER-target gene^32,33, we hypothesized that the decrease in proliferation by EP0062 could be mediated by E2F1. AR agonists induced E2F1 transcriptional downmodulation in both resistant and sensitive PDX models, but the decrease at the protein level was only observed in sensitive ones (Fig. 4A–C). Given that E2F1 is a transcription factor that modulates cyclin E and cyclin A expression³⁴, we analyzed the effect of EP0062 on cell cycle proteins. Sensitive models exhibiting tumor regression (PR or CR) upon treatment with EP0062 exhibited a significant decrease in cyclin E2 and cyclin A2 protein levels upon treatment with this AR agonist (Fig. 4D, E and Supplementary Fig. S4A–G). In addition, we observed a decrease in Rb phosphorylation at S807/S811 in sensitive models, while cyclin D1 levels were unaltered (Fig. 4F and Supplementary Fig. S4A–G). These results suggest that an AR agonist can have a potent cell cycle effect on the G1-S checkpoint by a downmodulation of E2F1-pathway.

Fig. 4. AR agonists downmodulate cell cycle genes in sensitive PDX models.

A Modulation of E2F1 expression by qPCR in biological replicates of 12 PDX models treated for 5 days with EP0062 (red), DHT (orange), or control (grey). Expression data was normalized to ACTB and GAPDH housekeeping genes and to control samples. B Representative immunoblot of the modulation of cell cycle proteins after 5 days of treatment with EP0062, DHT, or untreated from biological replicates of an EP0062-sensitive model, PDX512, and a DHT-resistant model, PDX244. C–F Quantification of E2F1- (C), cyclin E2- (D), cyclin A2-protein (E), and phospho-Rb S807/S811 (F) levels measured by Western blot and normalized to human GAPDH (hGAPDH) in EP0062 (red) or DHT (orange) vs control (grey)-treated samples in 12 PDX, ranked according to EP0062-sensitivity. Each symbol represents a different PDX model, bars represent the mean ± SEM of biological replicates, and one-way ANOVA with Dunnett’s multiple comparisons test is shown for (A, C–F). ANOVA analysis of variance, DHT dihydrotestosterone, hGAPDH human GAPDH, SEM standard error of the mean, qPCR quantitative polymerase chain reaction.

Inhibition of CDK4/6 can overcome SARM resistance

Our previous results suggest that resistance to SARMs in PDX models is likely due to an insufficient E2F1-downmodulation, which is a key mediator of the antiproliferative effect of SARMs. Prior studies have already described that the addition of a CDK4/6i enhances the antitumor activity of SARMs^21,24, but its comparison with the activity of the standard ET plus CDK4/6i combinations for patients with ER+/HER2− mBC has not been reported. Hence, we interrogated whether treatment with EP0062 has at least an equal activity to the SERD fulvestrant when combined with the CDK4/6i palbociclib in 8 PDX models.

We leveraged the 4 PDX models resistant to EP0062 to test if palbociclib could overcome resistance (PDX244, PDX479, PDX476.1, and PDX433.3), and our data confirmed this in PDX244, PDX476.1, and PDX433.3 (Fig. 5A and Supplementary Fig. S5). In addition, we tested whether the addition of palbociclib to EP0062 in 4 sensitive PDX models (PDX512, PDX376, PDX490, and PDX527) could ensure faster or more durable responses, as confirmed in PDX512 and PDX376, respectively (Supplementary Fig. S5K, L). Most of the PDX models had a similar response to EP0062 plus palbociclib compared to the combination of fulvestrant plus palbociclib (Fig. 5A). Notably, we observed a superior activity of EP0062 plus palbociclib compared to the standard of care in the ESR1-mutants PDX512 and PDX244 (Fig. 5A). Overall, all models harboring ESR1 alterations had significant and durable response to EP0062 plus palbociclib contrariwise to PIK3CA-AKT1-PTEN altered or HER2-enriched PDX models in which treatment with the SARM plus palbociclib could not achieve better response than SD (Fig. 5A). These results suggest that SARMs might be as potent as fulvestrant when combined with CDK4/6i for ER + BC, including patients with ESR1-altered tumors.

Fig. 5. Palbociclib enhances the antitumor activity and transcriptomic changes induced by EP0062 in ER + PDX models.

A Waterfall plot representing the best response (mean change from initial tumor volume) of n = 8 PDX models untreated (control, black bars) or treated with EP0062 100 mg/kg plus palbociclib 50 mg/kg (pink bars), fulvestrant 5 mg/mice plus palbociclib 50 mg/kg (green bars). Dashed lines indicate the range of PD (>20%), SD (20% to −30%), PR (−30% to −95%), and CR (<−95%) according to the best response of each model. For resistant models (PD) and tumors of control mice, day 21 is shown. Each symbol represents an individual tumor, and bars represent the mean of individual tumors) ± StDev. Stripped bars indicate that the experiment was performed in OVX mice. Heatmap of biomarkers associated with EP0062 plus palbociclib response in 8 ER + PDXs: PDX response to EP0062 or fulvestrant monotherapy; PDX intrinsic subtype by PAM50 assay and mutations in ESR1 or the PI3K-pathway (PIK3CA, PTEN, AKT1). B, F Waterfall plot showing individual tumor changes of PDX433.3 (B) or PDX244 (F) treated with EP0062 100 mg/kg (red), palbociclib 50 mg/kg (yellow), and palbociclib 50 mg/kg plus EP0062 100 mg/kg (pink) at its best response or control (untreated, grey), at day 21. C, G Venn diagram of genes significantly modulated (log2FC > 1, p adj < 0.05) in at least 2 biological replicates treated for 5 days with EP0062, palbociclib, or the combination compared to control in PDX433.3 (C) and PDX244 (G). D, H Dot plot of the GSEA in Hallmark gene sets MsigDB database from tumors treated with EP0062, palbociclib, or the combination in PDX433.3 (D) and PDX244 (H). Normalized enriched score (NES) is indicated, FDR < 0.05. E, I Quantification of the percentage of Ki67-positive cells in tumors treated with EP0062, palbociclib, or the combination in PDX433.3 (E) and PDX244 (I). Each symbol represents an individual tumor, bars represent the mean ± StDev, one-way ANOVA with Dunnett’s multiple comparisons test, and p value is shown. ANOVA analysis of variance, CR complete response, FDR false discovery rate, GSEA gene set enrichment analysis, MsigDB molecular signatures database, NES normalized enrichment score, PD progressive disease, PR partial response, p adj adjusted p value, SD stable disease, StDev standard deviation, OVX ovariectomized.

To elucidate the effect of AR activation and CDK4/6 inhibition on gene expression, we performed a transcriptomic analysis in two PDX models with intrinsic resistance to the single agents but sensitivity to the combination. Notably, in PDX433.3, a PDX model with modest resistance to EP0062 and to palbociclib monotherapy (Fig. 5B), EP0062 did not induce a potent transcriptomic effect on cell cycle hallmarks, but both palbociclib and its combination with EP0062 did so under these experimental conditions (Fig. 5C, D). Consistent with an early pharmacodynamic effect, we observed a potent decrease in the percentage of Ki67-positive cells with both palbociclib and the combination (Fig. 5E), despite the lack of significant tumor volume reduction at later time points. In contrast, PDX244 displayed a substantial modulation of gene expression with the treatment combination, with no significantly modulated pathways in the GSEA of single agents, consistent with its intrinsic resistance to EP0062 or to palbociclib (Fig. 5F–H). As expected, the top pathways significantly impacted by the combination treatment were cell cycle-related, like E2F-targets, G2M checkpoint, and mitotic spindle. Accordingly, a modest decrease in the percentage of Ki67-positive cells was observed upon treatment with EP0062, with a trend towards significantly higher reduction upon addition of palbociclib (Fig. 5I). Overall, these results suggest that the efficacy of combining SARMs with a CDK4/6i is partially explained by cell cycle deregulation.

Discussion

ET combined with cyclin-dependent kinases 4 and 6 inhibitors (CDK4/6i) has significantly improved PFS in the metastatic setting, which has placed the combination as the preferred first-line treatment for patients with ER+/human epidermal growth factor receptor 2 (HER2−) mBC³⁵. Still, the emergence of resistance and the selection of the subsequent therapy upon progression to ET + CDK4/6i remains a clinical challenge. In parallel, the combinations of CDK4/6i with oral selective ER degraders (SERDs), new generation selective ER modulators (SERMs), proteolysis targeting chimeras (PROTACs), selective ER covalent antagonists (SERCAs), complete estrogen receptor antagonists (CERANs), and selective AR modulators (SARMs) are being evaluated in clinical trials to overcome ET-resistance, focusing on ESR1 alterations^15,36–45. In this study, we investigated the activity of a SARM in a cohort of ER + /HER2- BC PDXs with molecular features that parallel those of patients’ tumors in the clinical setting. In contrast with prior reports, we demonstrated that EP0062 had equal or superior activity to the next-generation oral SERD camizestrant in 7 PDX models, and it was inferior in only two. These data also provide preclinical evidence that supports the development of SARMs in the clinic in earlier lines of treatment for mBC.

Preclinical and clinical studies have yielded context-specific evidence for the role of AR in ER + BC. On one hand, AR expression has been associated with tamoxifen resistance, and mechanistic studies suggest that AR can cooperate with ER to drive an ER-transcriptional program in AI-resistant tumors^46,47. However, the addition of enzalutamide to exemestane improved PFS exclusively in ET-naïve ER + BC patients with high AR mRNA levels in a randomized phase II clinical trial, compared to previously ET-treated patients for whom no benefit was observed⁴⁸. In addition, enzalutamide in combination with fulvestrant has shown modest activity in heavily pretreated ER+/HER2− mBC⁴⁹. On the other hand, preclinical studies have underscored a tumor suppressor role for AR^21,22,24,25, suggesting that AR alters the genomic distribution of ER and essential co-activators, thereby repressing ER-transcriptional activity²¹. Also, EP0062 and enobosarm have demonstrated promising clinical activity in ER + BC patients^27,28. Importantly, exploratory analysis from the phase II clinical trial with enobosarm revealed that patients with more than 40% of nuclear staining had a clinical benefit rate of 80%²⁷. Other studies suggest an association of the AR/ER ratio with clinical response to SARMs²². Here, we show that EP0062 has antitumor activity across baseline AR and ER levels. Interestingly, we identified the enrichment of GATA3 frameshift mutations that result in either elongation or truncation of the protein among sensitive models. Prior reports have shown that mutations in its second zinc-finger domain have loss-of-function and gain-of-function effects at different loci and are associated with poor prognosis^31,50. Importantly, Velimirovic and colleagues have shown that patients with tumors harboring GATA3 mutations had shorter PFS and OS to ET and CDK4/6i than those with wild-type GATA3⁵¹.

Growing evidence suggests that GATA3 facilitates the tumor suppressor function of AR in ER + BC, working as a molecular bridge between AR and ER at growth regulatory genes⁵². Since GATA3 variants have distinct chromatin binding sites from GATA3-wild type⁵⁰, we speculate that sensitivity to AR agonists could be determined by a differential AR-recruitment to specific genomic loci driven by GATA3 variants. Mechanistically, AR agonist activation has been reported to alter the genomic distribution of ER and essential co-activators such as SRC-3 and p300, thereby repressing ER-transcriptional activity and cell cycle genes²¹. GATA3 mutations reprogram the ER/FOXA1/GATA3 network and redistribute ER-associated chromatin binding sites⁵⁰, whereas AR interacts with wild-type GATA3 to co-occupy luminal differentiation loci⁵². However, direct genome-wide comparisons of ER and AR chromatin occupancy in GATA3 wild-type versus GATA3-mutant models are currently lacking. Thus, whether GATA3 variants enhance AR-mediated competition for chromatin sites and coactivators in ER+ BC remains an attractive hypothesis for future investigation. Importantly, previous preclinical studies have reported limited antitumor activity of SARMs in tumors with ESR1-mutations (p.Y538S, p.D538G, and p.L536P)^21,25. Here, we show that two ESR1 p.D538G-, but not two ESR1 p.Y537S-mutant models, exhibited profound sensitivity to EP0062. Since ESR1 p.D538G and p.Y537S have different chromatin binding sites⁵³, future studies are needed to evaluate if SARMs have activity against specific ESR1-mutants due to a differential recruitment of AR to ER-mutant binding sites. In addition, we report that PIK3CA, AKT1, or PTEN mutations are not associated with resistance to EP0062 in our study. The combination of SARMs plus AKTi or PI3Ki has not been explored so far and might be a therapeutic strategy for patients progressing to ET and CDK4/6i, given the improvement in PFS of the AKTi capivasertib with fulvestrant vs fulvestrant in the CAPItello-291 trial⁵⁴.

Herein, we generated a 122-gene signature based on genes modulated in EP0062-sensitive models. Although there was minimal overlap with other gene signatures, the SARM-response signature demonstrated comparable performance to existing prognostic signatures⁵⁵, indicating its potential as a complementary tool for predicting response to ET, such as other signatures (e.g., PAM50-based chemoendocrine score)⁵⁶. Still, we acknowledge that the gene signature was potently governed by proliferation-associated genes, which might reflect treatment response rather than intrinsic sensitivity, and therefore should be interpreted cautiously.

Contrary to studies suggesting that SARMs’ antitumor activity in ER + BC is mediated by an upregulation of AR-target genes²¹, our data suggest a potent induction of AR-program in PDXs independent of their sensitivity. In agreement with prior reports by Yu and collaborators²⁴, we observed an ESR1-downmodulation upon EP0062 treatment in the most sensitive models. Importantly, we report that EP0062 induces a potent E2F1 downmodulation, which leads to a decrease in S-phase cyclins, in sensitive models. Notably, in PDX433.3, early decreases in proliferation markers did not translate into sustained tumor volume reduction, suggesting that CDK4/6 inhibition may induce a transient cytostatic effect that can be bypassed over time. Our data provide molecular evidence for combining SARMs with a CDK4/6i to enhance G1-S blockade. This is exemplified by PDX244, in which we observed a potent downmodulation of the cell cycle induced only by the combination treatment. Notably, while prior studies have demonstrated an enhanced antitumor activity of the SARM when combined with a CDK4/6i^21,24, our study is the first to directly compare this combination with one standard of care combination (fulvestrant plus CDK4/6i), highlighting its superior efficacy. Importantly, all ESR1-mutant PDX reached PR or CR upon treatment, but we did not observe any additional activity against HER2-enriched nor PIK3CA-, PTEN-mutant models. Nevertheless, we acknowledge that treatment with a next-generation SERD in monotherapy (elacestrant) might already overcome ET + CDK4/6i-resistance in patients with tumors harboring ESR1-mutations. The side effects of ET, e.g., reduced bone mineral density, arthralgia, back pain, hot flushes, and dryness of the vaginal mucosa, deteriorate quality of life and affect treatment adherence. In contrast, treatment with SARMs is not expected to induce these side effects but rather suggests attenuation of osteoporosis and cachexia due to AR activation in muscle and bone^26,57. Of note, SARMs cannot be metabolized to DHT, reducing the risk of androgenic effects, nor to estradiol, as they are not substrates of the aromatase enzyme.

A limitation of our study is the use of PDX models lacking a fully competent immune microenvironment. Consequently, we cannot rule out that the immune compartment may also influence EP0062 response, and future studies incorporating humanized or syngeneic models will be important to address this point. A remaining mechanistic question is whether GATA3 variants directly modulate ER and AR chromatin occupancy. Future studies using isogenic models will be required to test if GATA3 alterations enhance AR agonist efficacy through differential nuclear receptor competition at chromatin sites and coactivator sequestration. Another limitation is that we have not explored whether AR exerts a pro-tumorigenic role and if enzalutamide shows an antitumor effect in EP0062-resistant PDXs. In the clinical setting, existing biomarkers—such as nuclear AR staining—currently provide a practical means to select patients who may benefit from SARMs, helping guide therapy decisions until more definitive predictive markers are identified.

In conclusion, we highlight that SARMs are as potent as next-generation SERDs in models with ESR1 or PI3K-pathway alterations. Furthermore, we showed that the percentage of AR and ER-positive cells does not predict EP0062 response and that GATA3 frameshift mutations could be associated with response to SARMs. In addition, EP0062 induces a potent E2F1-downmodulation, which explains the antitumor response and provides a rationale for its combination with CDK4/6i. Finally, the addition of palbociclib can overcome SARMs resistance, and the combination showed a superior antitumor activity than the common standard of care, fulvestrant plus palbociclib combination for ER+/HER2− models. This study, along with others, supports the clinical investigation of SARMs alone or in combination with CDK4/6i for patients with ET-resistant tumors.

Methods

Cell lines and reagents

MCF-7 (HTB-22) and T47D (HTB-133) cell lines were obtained from the American Type Culture Collection (ATCC) and were authenticated utilizing short tandem repeat (STR) profiling (FTA Sample Collection Kit for Human Cell Authentication Service; ATCC services). Mycoplasma test was routinely performed (MycoAlert™ Mycoplasma Detection Kit; LONZA), and cell lines were used for a maximum of 20 passages. MCF-7 and T47D were grown in RPMI 1640 with GlutaMAX medium (Gibco, #61870010) supplemented with 10% of fetal bovine serum (FBS) (Gibco, #10270-106) and, in the case of MCF-7, supplemented with 1 nM of β-estradiol (Sigma-Aldrich, #E8875). T47D CRISPR GATA3 p.R330Efs were generated by Dr. Takaku (previously published in ref. ³²) and were grown in DMEM high-glucose medium with 10% FBS (Thermo Fisher Scientific). All cells have been cultured at 37 °C with a 5% CO₂ atmosphere. EP0062 and fulvestrant were purchased from Selleckchem. DHT pellets of 10 mg/pellet 60-day release were purchased from Innovative Research of America.

Cell viability assays

To assess cell viability, T47D GATA3 p.R330Efs mutant cells (3 × 10³ cells/well) and parental T47D cells (2 × 10³ cells/well) were seeded in DMEM high-glucose medium supplemented with FBS. Because T47D GATA3 mutant cells exhibit slower proliferation than parental T47D cells, different seeding densities were used to achieve comparable cell densities at the time of treatment. Twenty-four hours after seeding, cells were washed once with PBS (−), and the medium was replaced with DMEM supplemented with 10% charcoal-stripped FBS. After an additional 48 h, cells were treated for 7 days with 1 nM β-estradiol in combination with EP0062 (final concentrations: 10 and 100 nM) or DHT (final concentrations: 10 and 100 nM), or with the corresponding vehicle controls (dimethyl sulfoxide [DMSO] or ethanol, respectively). Media and compounds were refreshed every 2 days. Cell viability was measured on day 7 using the CellTiter-Glo® 2.0 Cell Viability Assay (Promega), according to the manufacturer’s instructions. Viability values were normalized to vehicle-treated controls and to a positive control treated with cisplatin (100 µM final concentration), which was used as a cytotoxic control to define minimal viability. Each experiment was independently repeated three times, with two to three technical replicates per condition.

Clinical samples

Fresh tumor samples from female breast cancer patients were collected under a protocol approved by the Clinical Research Ethics Committee of Vall d’Hebron University Hospital (PR(AG)170/2019) for the use of these patient samples. All procedures involving human participants were conducted in accordance with the Declaration of Helsinki. Written informed consent to participate in the study was obtained from all patients. For PDX generation, tumor samples were implanted into immunocompromised mice in accordance with institutional and national guidelines. All animal procedures were conducted following the European Union’s animal care directive (2010/63/EU) and were approved by the Ethical Committee of Animal Experimentation of the Vall d’Hebron Research Institute (FUE-2020-01541918). All animal experiments were conducted and reported in accordance with the ARRIVE guidelines. Clinical information regarding tumor subtype, metastatic site, and treatment response was retrospectively collected. According to ESMO guidelines²⁹, patients were classified as ET-naïve, primary ET resistance, or secondary ET resistance in the adjuvant and/or metastatic setting.

PDXs

Fresh human breast tumors were obtained from patients at the time of surgical resection or when a tumor biopsy was performed, and were immediately implanted subcutaneously into the lower flank of 6-week-old female Crl:NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ mice under inhalation anesthesia (isoflurane at 4% for induction and 1.5–3% for maintenance). Analgesia was administered for 3 days post-surgery (meloxicam, 5 mg/kg) to minimize animal pain. Animals were housed in air-filtered laminar flow cabinets under a 12-h light/dark cycle with food and water ad libitum, and provided with environmental enrichment. Mice were continuously supplemented with 1.5 μM 17β-estradiol (Sigma-Aldrich, #E8875) in drinking water, except for OVX mice. Upon tumor engraftment, PDX models were maintained by serial transplantation into recipient mice. Depending on tumor growth characteristics, models were propagated in either NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ or Rj:NMRI-Foxn1^nu/nu mice. BB3RC31 was generated in Manchester Breast Centre as previously described in ref. ⁵⁸. Flash-frozen and formalin-fixed paraffin-embedded (FFPE) samples were taken for genotyping and histological studies in each passage. The molecular characterization (ER, PR, HER2) of the PDXs was performed as described in ref. ⁵⁹. HER2 was evaluated according to ASCO/CAP guidelines, which define positive status as a ratio of >2.2, normal as <1.8, and equivocal as between 1.8 and 2.2. Genomic characterization of ER+/HER2− PDX models was conducted by targeted exome sequencing using the MSK-IMPACT^TM (Integrated Mutation Profiling of Actionable Cancer Targets) panel as described in ref. ⁵⁹. Gene alterations of unknown significance were excluded from the analyzes hence, only missense mutations (putative drivers), splice mutations, truncating mutations (putative drivers), gene amplifications, or deletions were considered. Transcriptomic characterization was performed by a custom nCounter gene expression panel and by bulk RNA-sequencing. RNA was extracted from snap-frozen or FFPE tumor samples using the High Pure FFPET RNA isolation kit (Roche, Indianapolis, IN, USA) for RNA-sequencing and PAM50 molecular subtype classification, respectively. 250 ng of total RNA was used to measure the expression of 50 genes of the PAM50 intrinsic subtype predictor assay and 5 housekeeping genes using the nCounter platform (NanoString Technologies). Data were log base2-transformed and normalized to the housekeeping genes, and all PDX tumors were assigned to an intrinsic molecular subtype of breast cancer (luminal A, luminal B, HER2-enriched, basal-like, or normal-like) using the PAM50 subtype predictor⁶⁰ using R (3.4.3) software. Total RNA was extracted from flash-frozen tumor sample pieces after tissue homogenization with a POLYTRON® system PT 1200 E (Kinematica) using the RNeasy Mini Kit (#74106, Qiagen) according to the manufacturer’s instructions. RNA integrity was assessed using the RNA 6000 Nano Bioanalyzer 2100 Assay (Agilent). One microgram of total RNA with RIN > 8 was used to generate sequencing libraries to library preparation using KAPA Stranded mRNA-sequencing Illumina Platforms Kit (Roche) and sequenced on Illumina NovaSeq 6000 platform in paired-end mode (2 × 150bp), following manufacturer’s protocol. Raw reads were mapped to the human reference genome (Gencode, release 38) and counted at the gene level using STAR version 2.55.3a. STAR-Fusion (v1.12.0) with default options was used to analyze RNA-sequencing data and to identify ESR1 fusion transcripts. ESR1 fusions with >0.1 fusion fragments per million (FFPM) and >10 spanning fragment counts, and junction reads were selected.

In vivo experiments

OVX mice were used as a preclinical model of menopause, replicating the hormonal changes associated with ovarian hormone deficiency. 8 to 12-week-old mice with palpable tumors ranging from 100–500 mm³ were OVX as previously described in ref. ⁶¹. In hormone dependence experiments, mice were OVX once tumors reached 100–300 mm³ and tumor growth was monitored in OVX compared to non-OVX mice supplemented with 1.5 μM 17β-estradiol. Animals were allocated to treatment groups once tumors reached the specified volume range, ensuring comparable mean tumor volume and variance between groups at baseline. Group allocation was performed using GraphPad Prism 10 (version 10.5.0) to minimize inter-group differences, and no statistically significant differences in tumor volume were observed prior to treatment initiation. For pharmacodynamic studies, tumor tissue was implanted into at least two mice per treatment arm, and when tumors reached 500 mm³, animals were OVX. Upon 10 days of surgery, mice were allocated into treatment arms detailed below and treated for 5 days.

To evaluate the sensitivity to targeted therapies, tumor tissue from a PDX model was reimplanted into 4–5 mice per treatment arm. Efficacy experiments were performed in non-OVX mice except where otherwise stated. Once tumors reached 100–300 mm³, animals were allocated into treatment groups: EP0062 was administered at 6 mg/kg/day p.o BID 6 times per week for pharmacodynamic studies and 100 mg/kg/day p.o QD 6 times per week for efficacy experiments and was diluted in 0.5% carboxymethyl cellulose. DHT was administered at 10 mg/pellet 60-day-release subcutaneous (sc). Fulvestrant was administered at 5 mg/mouse sc SID twice a week and diluted in 95% peanut oil, and 5% DMSO. Camizestrant was administered at 10 mg/kg p.o SID 6 times per week and diluted in 40% diethylene glycol and 60% captisol. Palbociclib was administered at 50 mg/kg/day p.o SID 6 times a week in 0.5% methylcellulose, and 0.2% polysorbate 80 in drinking water. Tumor growth was monitored bi-weekly by an electronic calliper from the first day of treatment until the end of the experiment. Tumor volume was calculated as V = (length × width²) × (π/6). Efficacy experiments contained an untreated control arm with a percentage of change in tumor growth superior to 20% from the initial volume. Investigators were blinded during group allocation and tumor measurement.

Antitumor activity was determined by comparing tumor volume on the last day of treatment to its baseline (day 1): % tumor volume change = (V_{X days} – V_baseline)/V_baseline × 100. The antitumor response was classified according to the relative change in tumor volume upon treatment, similar to the Response Evaluation Criteria in Solid Tumors (RECIST) and denominated mRECIST. Complete response (CR) was set as best response ≤−95%; partial response (PR) as −95%<best response ≤ −30%; stable disease (SD) as −30%<best response ≤ +20%; and progressive disease (PD) as best response > +20%. The best response was based on at least three repeated measures from day 21. Models that displayed a preclinical benefit from treatment (SD, PR, and CR) were categorized as sensitive, whereas all PD models were categorized as resistant. For resistant tumors or for control tumors, tumor volume at day 21 was considered. Animal welfare was monitored every day based on the following parameters: weight loss, physical appearance, behavior and posture, and tumor status. Mice were euthanized using a gradual-fill CO₂ method (30–70% chamber volume per minute), followed by cervical dislocation as a secondary method to confirm death. Euthanasia was performed at the end of the experiment or upon reaching predefined humane endpoints, including weight loss (>20%), poor body condition (emaciation), abnormal physical appearance (e.g., hunched posture), altered behavior (e.g., immobility or hyperactivity), and excessive tumor burden (tumor volume > 1500 mm³ or presence of ulceration), in accordance with institutional guidelines. All efforts were made to minimize animal suffering. At the end of the experiment, flash-frozen and FFPE tumors were taken for posterior proteomic and genomic analyses. Samples were collected 2 h after the administration of EP0062, camizestrant, and palbociclib or 24 h after the administration of fulvestrant.

Quantitative RT-PCR

mRNA from snap-frozen tumors was extracted using the RNeasy Mini Kit (Qiagen, #74104) according to the manufacturer’s protocols. 0.5 µg of RNA was retrotranscribed using PrimeScript RT reagent Kit from Takara, following the manufacturer’s instructions. 1 ng of cDNA was amplified by quantitative PCR, using the following TaqMan® MGB probes: SEC14L2 (Thermo Fisher Scientific, #Hs01052192_m1), ZBTB16 (Thermo Fisher Scientific, #Hs00232313_m1), ESR1 (Thermo Fisher Scientific, #Hs00174860_m1), E2F1 (Thermo Fisher Scientific, #Hs00153451_m1), ACTB (Thermo Fisher Scientific, #Hs99999903_m1) and GAPDH (Thermo Fisher Scientific, #Hs99999905_m1), mixed with TaqMan Fast Advanced Master Mix, in a QuantStudio 6 Flex Real-Time PCR System. Relative gene expression of SEC14L2, ZBTB16, ESR1, and E2F1 was determined by the comparative Ct method⁶². The geometric means of the expression values for both ACTB and GAPDH housekeeping genes were used to normalize the expression and to calculate the normalized SD of all transcripts analyzed.

IHC assays

IHC was performed in 3 μM-thick FFPE tumor samples. Staining of ER, AR, PgR, and Ki67 was undertaken following the protocol provided by Ventana Medical System, Inc. Briefly, slides were heated in the instrument at 75 °C for 28 min and deparaffinized with EZ prep solution (Ventana Medical System). Then, antigen retrieval was performed at a slightly basic pH at 95 °C using the Cell Conditioning 1 buffer (CC1; Ventana Medical System) for all antibodies. Next, samples were stained with the following primary HQ antibodies: rabbit anti-Ki67 (Roche, #790-4286, ready-to-use), rabbit anti-ER (Roche, #790-4325, ready-to-use), rabbit anti-AR (Roche, #760-4605, ready-to-use), mouse anti-PgR (Roche, #790-4296, ready-to-use), and finally with anti-HQ-horseradish peroxidase (HRP) and 3,3’-diaminobenzidine (DAB) staining (Roche). Slides were counterstained with Hematoxylin II and bluing reagent (Ventana Medical System) and mounted with a xylol-based mounting medium. For cyclin D1 staining, sections were dewaxed, rehydrated, and antigen retrieval was performed using a microwave at maximum power in citrate pH 6 (DAKO S2367) for 20 min. After peroxide blocking, slides were stained for 2 h with rabbit anti-cyclin D1 (Thermo Fisher Scientific, #RM-9104-S1, 1:100) diluted in DAKO Antibody Diluent (DAKO S0809). Subsequently, EnVision+ System HRP-labeled polymer anti-rabbit (DAKO K4002) was exposed to the sections for 30 min and then washed 3 times. Finally, slides were treated with Liquid DAB + Substrate Chromogen System (DAKO, K3468) for 3 min, counterstained with Hematoxylin II, dehydrated, and mounted with xylol-based mounting medium. Positive and negative controls were run along with the tested slides for each marker. Allred score of cyclin D1 staining was calculated from 0 to 8, considering the percentage of positive cells with nuclear staining (0–5) plus the staining intensity (0–3). AR staining was quantified using an H-Score, based on the percentage of cells with weak staining (1+), intermediate staining (2+), and strong staining (3+). For PgR and Ki67 staining, the percentage of cells with nuclear staining was quantified. ER staining was quantified by the percentage of positive cells with nuclear staining, or H-score, as stated in each experiment.

Protein isolation and Western blot

Flash-frozen pieces of tumor and cell plates were lysed in ice-cold 1× radioimmunoprecipitation assay buffer (RIPA) containing Tris–HCl pH 8.0, 10 mM, EDTA 1 mM, Triton-X-100 0.1%, SDS 0.1%, SDC 0.1%, and NaCl 140 mM, supplemented with 1× protease inhibitor cocktail (cOmplete, Roche), NaF 25 mM, Na₃VO₄ 1.35 mM, and PMSF 5 mM. Homogenization was performed on ice with a POLYTRON® system PT 1200 E (Kinematica). Lysates were centrifuged at 15,000×g at 4 °C for 15 min and supernatants were collected. Protein concentration was calculated using DCTM Protein Assay (Bio-Rad). A total of 20 μg of protein was separated on 12% SDS-PAGE at 100 V and transferred to a nitrocellulose membrane for 1.5 h at 100 V. Membranes were blocked for 1 h in 5% non-fat dry milk (NFDM) in Tris-buffered saline (TBS) with 0.1% Tween20 (T-TBS) and then hybridized using the primary antibodies in 5% bovine serum albumin (BSA, Sigma-Aldrich, #A9647) with T-TBS overnight. The primary antibodies used were the following: rabbit anti-ERα N-term (Millipore, #04-820, 1:1000 in 5% BSA T-TBS), rabbit anti-ERα C-term (Cell Signalling Technology, #8644, 1:1000 in 5% BSA T-TBS), rabbit anti-AR (Cell Signalling Technology, #5153, 1:1000 in 5% BSA T-TBS), rabbit anti-GATA3 (abcam, #ab199428, 1:2000 in 5% BSA T-TBS), rabbit anti-phospho-Rb S780 (Cell Signalling Technology, #9307, 1:1000 in 5% BSA T-TBS), rabbit anti-phospho-Rb S807/S811 (Cell Signalling Technology, #9308, 1:1000 in 5% BSA T-TBS), mouse anti-Rb (BD Pharmigen, #554136, 1:1000 in 5% NFDM), rabbit anti-E2F1 (Cell Signalling Technology, #3742, 1:1000 in 5% BSA T-TBS), rabbit anti-cyclin D1 (abcam, #ab40754, 1:1000 in 5% BSA T-TBS), rabbit anti-cyclin E2 (abcam, #ab40890, 1:1000 in 5% BSA T-TBS), rabbit anti-cyclin A2 (abcam, #ab32386, 1:1000 in 5% BSA T-TBS) and rabbit anti-GAPDH (abcam, #ab128915, 1:5000 in 5% BSA T-TBS). Membranes were incubated for 1 h with mouse and rabbit HRP-conjugated secondary antibodies (GE Healthcare, #GENA934 and #GENA931) in 5% milk in T-TBS, diluted 1:2000 for all primary antibodies except 1:4000 for anti-GATA3 and 1:10000 for anti-GAPDH. Proteins were detected with Immobilon Western Chemiluminescent HRP substrate (Millipore). Immunoblots were captured at a chemiluminescence imager: Amersham Imager 600, GE Healthcare, or IBright CL 1500, Invitrogen. Images were processed in Adobe Photoshop or IBright Analysis Software. ImageJ 1.54 (https://fiji.sc/) was used for immunoblotting band quantification.

RNA-sequencing analysis

Raw reads were mapped to the human reference genome (Gencode, release 38), excluding potential contributions from the murine stromal compartment, and counted at the gene level using STAR version 2.55.3a⁶³ and parameters ‘--outSAMunmapped None --outSAMtype BAM SortedByCoordinate --quantMode GeneCounts’. Qualimap⁶⁴ was used to check the quality of the alignments. As the library preparation protocol was reverse-stranded, the 4th column of the STAR gene counts output of each sample was taken as input for DESeq2. Genes with less than 10 read counts across all samples were filtered out before processing the data. Differential expression analysis was performed on the human genes with DESeq2 with the following contrasts for each PDX model: (i) EP0062-treated samples vs control samples, (ii) camizestrant-treated samples vs control samples, (iii) palbociclib-treated samples vs control samples, and (iv) EP0062 plus palbociclib-treated samples vs control samples. Genes were selected as significantly differentially expressed at FDR < 0.05, and the absolute log2 fold change (log2FC) > 1. GSEA was performed using the fgsea (v.1.22.0) package and the Hallmark gene set collection from Molecular Signatures Database (MSigDB). The gene rank was established using the Wald test statistic of all genes, obtained from DESeq2 differential expression analysis. The FDR Benjamini–Hochberg (BH FDR) method was applied to adjust p values, and gene sets with FDR < 0.05 were considered significant.

SARM-response signature

The 122-gene signature was derived from genes induced and repressed by EP0062 in sensitive PDX models (PDX423, PDX512, and PDX376), which were not modulated in any of the resistant models (PDX244, PDX479, PDX476.1, and PDX433.3). Genes were selected as significantly differentially expressed at FDR < 0.05, and the absolute log2 fold change (log2FC) > 1. Raw counts of each PDX model were joined, and low-expressed genes were filtered using a threshold of a minimum 1 count per million (cpm) in at least 2 samples (the smallest group). EP0062 and Camizestrant were dealt in separate matrices, including untreated controls in both. Reads were normalized by gene-length using transcript per million (TPM) transformation to calculate signature scores. Singscore⁶⁵ version 1.22.0 was used to compute the scores of the signatures. Scores reflect the relative mean percentile rank of the target gene sets within each sample and were compared with a Wilcoxon test. To analyze the predictive power of the signature, the METABRIC dataset was used⁶⁶. Microarray expression data and clinical metadata were downloaded from cBioPortal (accession date: 2024-09-26). ER+ patients were selected, obtaining a total of 1505 samples to be included in the analysis. 113 of the total 122 genes of the SARM signature were found in the METABRIC dataset, and from the resulting signature scores, high and low score groups were created using a 0.5 quantile as a threshold, and survival was compared with Kapla–Meier plots to assess prognosis. Kaplan–Meier plots were done using the survival (v3.7.0) and survminer (v0.4.9) R packages⁶⁷. SigCheck⁶⁸ version 2.34.0 was used to evaluate the predictive performance of the SARM-response signature as compared to the signature from AR-Hickey²¹, as well as known signatures like MammaPrint, Oncotype, Endopredict, PAM50⁵⁵, or Hallmarks. To that end, the value of the first principal component of the expression values in the signature (PCA1) was used as the default scoring method for separating validation samples into high and low groups (using a 0.5 percentile threshold).

Statistical analysis

Statistical tests were performed using GraphPad Prism 10 version 10.5.0 (GraphPad Software). Details of sample size, replicates, mean, and error bars are provided in figure legends. Normal distribution was assessed with the Shapiro–Wilk test in all comparative studies and correlations to assume or not a Gaussian distribution. For two-group comparisons, an unpaired or paired two-tailed test was used if the data were normal; otherwise non-parametric test was used: Mann–Whitney test for unpaired samples, the Wilcoxon matched-pairs signed rank test for paired samples. For more than two-group comparisons, one-way analysis of variance (ANOVA) corrected with Dunnett’s multiple comparison test was used. Two-way ANOVA corrected with Sidak’s multiple comparison test was used for two-group comparisons across time. The coefficient of correlation was computed using the Pearson normality test. All analyses were two-tailed unless otherwise stated. Fisher’s exact test was performed to determine the association between the two variables. A generalized linear model was used where “proportion of positive cells” was the response variable and the experimental condition and sensitivity (2 levels), were fixed factor, while the random factor was the PDX model. The lmer() function from the package lme4⁶⁹ was used with R v4.3.2⁷⁰.

Author contributions

L.M. contributed to the collection and/or assembly of data, data analysis and interpretation, and manuscript writing. J.G.-G., C.M., A.O., R.M.-P., C.Y., F.P., M.G., O.R., and F.B.-M. contributed to the collection and/or assembly of data, data analysis, and interpretation. C.V. contributed to the collection and/or assembly of clinical samples data. E.G.-G. and R.D. contributed to statistical analysis. J.P., T.Y.D., I.A.-B., G.F., and L.N. contributed to bioinformatic analysis. V.S., M.B., and G.P.V. contributed to conception and design of the study, financial support, collection and/or assembly of data, data analysis and interpretation, and manuscript writing. A.P., S.C., C.S., and M.O. contributed to data analysis and interpretation. All approved final versions of the manuscript.

Data availability

Data supporting the findings of this study is available upon request to the corresponding author. Annotated DNA sequencing data from PDX models and associated clinical data from breast cancer patients treated at Vall d'Hebron Hospital are included. RNA-sequencing data generated in this study are available from the Gene Expression Omnibus under accession number GSE300130. Raw DNA sequencing data are available for academic research from the corresponding author upon reasonable request and subject to completion of a Data Transfer Agreement with Memorial Sloan Kettering Cancer Center (New York, USA).

Competing interests

F.B-M. reports part-time employment from Reveal Genomics, and has patents filed: PCT/EP2022/086493, PCT/EP2023/060810, EP23382703, and EP23383369. S.C. has received institutional grant/funding from Daiichi-Sankyo and AstraZeneca, Shares in Totus Medicines, and consultation/Ad board/Honoraria from Daiichi-Sankyo, AstraZeneca, Lilly, and Casdin Capital. R.D. declares advisory role for Roche, Foundation Medicine, AstraZeneca, Pfizer, received a speaker’s fee from Roche, Foundation Medicine, GuardantHealth, Ipsen, Amgen, Servier, Sanofi, Libbs, Merck Sharp & Dohme, Lilly, AstraZeneca, Johnson and Johnson, Takeda, Bristol Myers Squibb, GlaxoSmithKline, Gilead, Pfizer, research grants from Merck, Novartis, Daiichi-Sankyo, GlaxoSmithKline, Pfizer and AstraZeneca, and is investor in Trialing Health, S.L. A.P. reports advisory and consulting fees from AstraZeneca, Roche, Pfizer, Novartis, Daiichi Sankyo, and Peptomyc, lecture fees from AstraZeneca, Roche, Novartis, and Daiichi Sankyo, institutional financial interests from AstraZeneca, Novartis, Roche, and Daiichi Sankyo; stockholder and employee of Reveal Genomics; patents filed PCT/EP2016/080056, PCT/EP2022/086493, PCT/EP2023/060810, EP23382703 and EP23383369. M.O. received research grants from AstraZeneca, Ayala Pharmaceuticals, Boehringer-Ingelheim, Genentech, Gilead, GSK, Immutep, Roche, Seagen, and Zenith Epigenetics. M.O. reports advisory and consulting roles from AstraZeneca, Cureos Science, Daiichi-Sankyo/AstraZeneca, Gilead, iTEOS, Lilly, MSD, Relay Therapeutics, Roche, and Seagen. M.O. received honoraria from AstraZeneca, Eisai, Gilead, Libbs, Lilly, MSD, Novartis, Pfizer, Roche, and Seagen. M.O. received travel grants from AstraZeneca, Eisai, Gilead, and Pierre-Fabre. M.O. serves as the President of the SOLTI Breast Cancer Group. C.S. reports consulting fees or honoraria for advisory boards or travel grants from AstraZeneca, Daiichi Sankyo, Eisai, Exeter Pharmaceuticals, Gilead, Lilly, MediTech, Menarini, MSD, Novartis, Pfizer, Philips Healthcare, Pharmalex, Pierre Fabre, Puma Biotechnology, Roche, Seagen, Synthon, and Zymeworks. V.S. reports research grants from AstraZeneca and personal fees and/or non-financial support from AstraZeneca and GSK. The remaining authors declare that they have no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41523-026-00993-2.