The androgen receptor is a tumor suppressor in estrogen receptor–positive breast cancer

The role of the androgen receptor (AR) in estrogen receptor (ER)-α-positive breast cancer is controversial, constraining implementation of AR-directed therapies. Using a diverse, clinically relevant panel of cell-line and patient-derived models, we demonstrate that AR activation, not suppression, exerts potent antitumor activity in multiple disease contexts, including resistance to standard-of-care ER and CDK4/6 inhibitors. Notably, AR agonists combined with standard-of-care agents enhanced therapeutic responses. Mechanistically, agonist activation of AR altered the genomic distribution of ER and essential co-activators (p300, SRC-3), resulting in repression of ER-regulated cell cycle genes and upregulation of AR target genes, including known tumor suppressors. A gene signature of AR activity positively predicted disease survival in multiple clinical ER-positive breast cancer cohorts. These findings provide unambiguous evidence that AR has a tumor suppressor role in ER-positive breast cancer and support AR agonism as the optimal AR-directed treatment strategy, revealing a rational therapeutic opportunity. Functional interplay of sex hormones in estrogen receptor–positive breast cancer unveils the therapeutic potential of androgen receptor agonists.

I n mammary gland development, estrogen stimulates and androgen inhibits post-pubertal growth 1 . ER is required for breast development and unequivocally drives the majority of breast malignancies (~80%), but the role of AR is unresolved 2 . While androgens were historically used to treat breast cancer 3,4 , knowledge of hormone receptors in breast tissue was rudimentary at the time and efficacy was thought to be via suppression of the hypothalamic-pituitary-gonadal axis. Androgen therapy was ultimately discontinued due to virilizing side effects and the advent of drugs that directly target ER 5 , which are collectively referred to as endocrine therapy. ER inhibitors are standard of care for ER + breast cancer, but resistance to these agents (endocrine resistance) is the major cause of breast cancer mortality 6 . The need for alternative strategies has renewed interest in androgen therapy 7 , especially as nearly all ER + breast cancers express AR 8 and nonvirilizing selective AR modulators (SARMs) are now clinically available 9 . However, preclinical studies, predominantly performed with breast cancer cell lines, have produced conflicting evidence as to the role of AR in ER + breast cancer, especially in the context of endocrine-resistant disease 2,10-13 . These equivocal findings have resulted in clinical trials investigating both AR agonists and AR antagonists for women with ER + breast cancer, highlighting an urgent need for clarification.

AR is a favorable prognostic marker of ER + breast cancer survival.
Immunoreactive AR was assessed in the large, single-institution Nottingham Tenovus Primary Breast Cancer Series used to establish the Nottingham Prognostic Index (NPI) 14 . AR was positively associated with breast cancer-specific survival in ER + cases, including high-risk ER + disease treated with endocrine therapy (Extended Data Fig. 1). Multivariate analyses confirmed that AR was a favorable prognostic factor, independent of other variables, including all NPI components and established (for example, progesterone receptor (PR), HER2) or recently incorporated (for example, FOXA1, GATA3) factors (Supplementary Table 1a,b). The data also indicate that AR is a positive predictive biomarker of response to endocrine therapy. Analysis of one of the largest breast cancer cohorts to have gene expression, gene mutation and patient outcome data (METABRIC 15 ) further supports AR as an independent good prognostic factor and biomarker of endocrine therapy response, even when controlling for all prognostic factors and contemporary genomic parameters that influence disease outcome (for example, commonly mutated genes) (Supplementary Table 1c). These clinical associations implicate AR as having a tumor suppressor function in multiple disease contexts of ER + breast cancer.
Androgen antagonizes estrogen activity in patient-derived breast tissue explants. To test whether the association between AR and breast cancer survival is due to its innate ability to restrain ER-driven growth, we interrogated human breast tissues using an ex vivo patient-derived explant (PDE) model (Fig. 1a) that sustains tissue complexity, histological features of the original tissue and sex The androgen receptor is a tumor suppressor in estrogen receptor-positive breast cancer hormone responsiveness 16,17 . ER + tumor PDEs treated with estrogen (E2) alone had a higher proliferative index (Ki67 positivity) compared to patient-matched explants treated with E2 plus the potent natural androgen, 5α-dihydrotestosterone (DHT) (Fig. 1b and Supplementary  Table 1d). Antagonistic sex hormone activity was also evident in normal and malignant breast epithelial cells when assessing proliferation with BrdU, a marker of newly synthesized DNA ( Fig. 1c and Extended Data Fig. 2a,b). Transcriptional consequences of AR and ER activation were interrogated via RNA sequencing (RNA-seq) using additional cases of normal or ER + malignant breast PDEs treated with E2 or E2 + DHT for 24 h (Supplementary Table 1e). Consistent with the cytological proliferation markers, E2 positively and E2 + DHT negatively correlated with gene signatures of cell cycle regulation in normal and malignant breast PDEs (Fig. 1d,e and Extended Data Fig. 2c-e). Androgen exposure also enriched for a hallmark androgen response gene set. Collectively, these data provide clear evidence of the antiproliferative effect of androgen in clinical breast tissues exposed to estrogen and suggest that AR signaling events that constrain normal breast growth are sustained or can be reactivated by AR agonism in ER + breast cancers to inhibit tumor growth.  PDes of eR + breast cancer (n = 17 cases) exposed to estrogen (e2; 10 nM) or estrogen plus androgen (e2 + DHT; 10 nM each). Data are represented as mean ± s.e.m. Data were analyzed using a two-tailed Wilcoxon matched-pairs signed-rank test (W = −153.0, r = 0.4165, ****P < 0.0001). c, BrdU immunohistochemistry in e2 and e2 + DHT-treated PDes. Images are representative of n = 3 independent cases of normal breast PDe and n = 9 independent cases of eR + breast cancer. Scale bars, 50 µm. d, Normalized enrichment scores (NeS) for signature gene sets representing RNA-seq data from PDes of normal breast (n = 17) or eR + breast cancer (n = 8). GO, gene ontology. e, Gene set enrichment plots associated with data presented in d. f, Dual-label immunofluorescence for AR and eR in an eR + IDC. Arrows indicate AR + eR + cells. g, AR and eR colocalization in a cohort consisting of histologically normal (nonmalignant; NM) breast tissues (n = 24), eR + ductal carcinoma in situ (DCIS; n = 63), invasive ductal carcinoma (IDC; n = 67) and lymph node metastases (LN mets; n = 23).
( Supplementary Fig. 2), another estrogen-dependent breast cancer model that endogenously expresses both receptors and is growth inhibited by androgen in vitro 21 . These data indicate that activated AR reproducibly associates with and redistributes the ER cistrome in breast cancer cells.
Whereas ERBSs lost upon AR activation included loci associated with established ER target genes (for example PGR, MYB, BCL2), gained ERBSs included loci associated with genes regulated by AR signaling in ER + breast cancer cells (for example SEC14L2, EAF2, ZBTB16) 18,22 (Fig. 2b-d and Supplementary Fig. 2c,d). DNA motif analysis revealed that loss of ER occupancy occurred at estrogen response elements (EREs) and gain of ER occupancy occurred at androgen response elements (AREs), implying relocation of ER to ARBSs (Extended Data Fig. 3a). AR knockdown reversed DHT-induced changes in ER binding at representative loci (Extended Data Fig. 3b-d), confirming that AR is the key mediator of these events. AR-mediated loss of ER binding was associated with downregulation of E2-induced genes ( Fig. 2c and Extended Data Fig. 3e,f) and gain of ER binding was associated with enrichment of AR at DHT-induced genes ( Fig. 2d and Extended Data Fig. 3g,h). Assessment of H3K27ac, a mark of transcriptional activation 23 , was concordant with DHT-induced repression of ER target genes and induction of AR target genes (Extended Data Fig. 3e,g). Hence, AR-mediated rearrangement of the ER cistrome had transcriptional consequences.
Androgen-induced loss of ER binding associated with downregulation of an ER target gene is exemplified by a locus within intron 1 of the MYB gene ( Fig. 2c and Extended Data Fig. 4a,b), shown to be critical for ER-induced expression 24 . MYB encodes a transcription factor that stimulates cell cycle progression 25 via pathways upregulated by E2 and downregulated by E2 + DHT in normal and malignant PDEs (Fig. 1d,e), indicating that antagonistic sex hormone regulation of MYB may represent an early regulatory event that impacts proliferative capacity of breast epithelial cells. The ability of activated AR to displace ER from chromatin required an agonist ligand because treatment with an antagonist ligand (bicalutamide) did not reduce ER binding at ER target genes, nor inhibit E2-stimulated proliferation (Extended Data Fig. 4c-e). AR binding to DNA was also required because induction of a constitutively active AR into MCF7 breast cancer cells, which endogenously express very low levels of AR, inhibited ER binding and transcriptional activity at cell cycle genes, including MYB, but a mutant AR unable to bind DNA did not (Extended Data Fig. 4f-h). Accordingly, only of AR redistributes ER and p300 on chromatin to repress cell cycle genes. a, Overlap of consensus AR and eR cistromes in ZR-75-1 breast cancer cells treated in vitro with e2 (10 nM) or e2 + DHT (10 nM each). b, Two-factor log-ratio (M) plot displaying DHT-induced changes in eR and AR enrichment at consensus eRBSs presented in a. Point color denotes eR consensus peak occupancy; e2 unique (yellow-orange), e2 + DHT unique (orange-red) and e2/e2 + DHT shared (rust, plotted at rear). Point co-ordinates are derived from the average enrichment score of three ChIP-seq replicates for each consensus binding site. example binding sites near eR target genes and AR target genes are highlighted in pink and purple, respectively. c, Genome browser images showing averaged eR and AR ChIP-seq signals at binding sites associated with eR target genes (MYB, PGR) in three replicates of ZR-75-1 cells (left). RNA-seq heat map of e2-induced genes antagonized by co-treatment with DHT in ZR-75-1 cells (right). Veh, vehicle; eD, e2 + DHT. d, Genome browser images showing averaged AR and eR ChIP-seq signals at binding sites associated with AR target genes (EAF2, SEC14L2) in three replicates of ZR-75-1 cells (left). RNA-seq heat map of DHT-induced genes in ZR-75-1 cells (right). e, Two-factor log-ratio (M) plot showing DHT-induced changes in e2-stimulated p300 and H3K27ac enrichment at consensus p300 binding sites in ZR-75-1 cells. Point color denotes p300 consensus peak occupancy; hormone-responsive e2 unique (pink), e2 + DHT unique (purple) and basal (black; plotted at rear). Point co-ordinates are derived from the average enrichment score of three ChIP-seq replicates for each consensus binding site. example p300 binding sites associated with eR (yellow-orange) and AR (orange-red) target genes are highlighted. f, Motif analysis of hormone-responsive p300 binding sites as shown in e. PRe, progesterone response element; GRe, glucocorticoid response element. g, p300 ChIP-qPCR at enhancers of eR-regulated cell cycle genes in ZR-75-1 cells treated in vitro under designated hormone conditions. Data were analyzed by a two-way analysis of variance (ANOVA) (F = 56.98, 10.99 and 1.851 for hormone treatment (P < 0.0001), test site (P < 0.0001) and their interaction (P = 0.1061), respectively; d.f. = 30). Post hoc analyses were performed using Tukey's multiple comparisons test, where MYB, P = 0.0006 and P = 0.0107 for Veh versus e2 and e2 versus e2 + DHT, respectively; MYC, P = 0.0034 and P = 0.0228 for Veh versus e2 and e2 versus e2 + DHT, respectively; CCND1, P < 0.0001, P = 0.0452 and P = 0.0523 for Veh versus e2, Veh versus e2 + DHT, and e2 versus e2 + DHT, respectively; BCL2, P = 0.003 and P = 0.0984 for Veh versus e2 and e2 versus e2 + DHT, respectively; FOXM1, P < 0.0001, P = 0.0001 and P = 0.0463 for Veh versus e2, Veh versus e2 + DHT and e2 versus e2 + DHT, respectively. Data are represented as mean ± s.e.m. of three replicates representing independent passages of cells. Asterisks denote statistical significance; *P < 0.05; **P < 0.01; *** P < 0.001; ****P < 0.0001. a transcriptionally functional AR inhibited E2-stimulated proliferation of this cell line (Extended Data Fig. 4i). Hence, classic AR transcription factor activity is essential for antagonism of ER signaling in breast cancer cells. This AR activity clearly impacts the ability of ER to transcriptionally upregulate cell cycle genes and likely represents a key means by which AR inhibits proliferation.  In addition, induction of AR target genes with purported tumor suppressor activity (for example EAF2, SEC14L2, ZBTB16) [26][27][28][29] may also contribute to growth inhibition of ER + breast cancer.
The enzyme p300 is a rate-limiting factor required for H3K27 acetylation at transcriptionally active enhancers 23,30 , including those regulated by ER in breast cancer cells 31 and AR in prostate cancer cells 32 . Therefore, ER and AR potentially compete for p300 in breast cancer cells. AR activation caused a major redistribution of E2-stimulated p300 binding sites that were concordant with changes to the H3K27ac signal in ZR-75-1 cells (Fig. 2e). Over half (53%) of the E2-stimulated p300 binding sites were lost upon treatment with E2 + DHT (Extended Data Fig. 5a-c), and these loci were highly enriched for EREs (Fig. 2f). p300 binding was reduced at E2-regulated genes antagonized by co-treatment with DHT (Extended Data Fig. 5d,e), including loci associated with ER-regulated cell cycle genes ( Fig. 2g and Extended Data Fig. 5f,g). Therefore, AR-mediated displacement of p300 at transcriptionally active enhancers associated with ER-regulated cell cycle genes is another means of preventing ER from stimulating a proliferative transcriptome. This p300 displacement coincided with relocation to and activation of AR target genes (Fig. 2e,f and Extended Data Fig. 5d,e). We next assessed SRC-3, a coregulatory protein required for ER to recruit p300 (ref. 33 ). AR-mediated changes to the SRC-3 cistrome closely tracked with changes to the p300 cistrome (Extended Data Fig. 6), suggesting that SRC-3 and p300 may be repositioned as a complex. As AR does not require SRC-3 to recruit p300 (ref. 34 ), it may have a competitive advantage over ER in the recruitment of this critical co-activator. Collectively, these data indicate that AR-mediated antagonism of ER signaling involves sequestration of p300 and SRC-3 in breast cancer cells.

AR agonism, not antagonism, durably inhibits in vivo growth of ER + breast cancer.
To test the influence of clinically relevant AR-targeting drugs in vivo, ZR-75-1 xenografts were established in mice. Tumor growth was stimulated with E2 and the effect of AR activation (via two distinct agonists, DHT and enobosarm) or AR inhibition (via a receptor antagonist, enzalutamide) was assessed. Enobosarm lacks virilizing activity in women and has AR agonist activity in breast cancer cells 9,35 . Enzalutamide is used to treat prostate cancer 36 . Both drugs have been investigated in clinical trials for women with ER + breast cancer. While enzalutamide reduced ZR-75-1 xenograft growth initially, this was not sustained; the proliferative index of enzalutamide-treated tumors was similar to controls by 5 d of treatment (Extended Data Fig. 7a-d). In contrast, both AR agonists (DHT, enobosarm) durably inhibited tumor growth throughout the course of 90 d, with a marked reduction in the proliferative index after 5 d of treatment (Extended Data Fig. 7a-d). Accordingly, AR agonists potently antagonized expression of E2-regulated genes and induced expression of AR target genes, but enzalutamide did not (Extended Data Fig. 7e,f). AR and ER ChIP-seq was performed in ZR-75-1 xenografts from vehicle and AR-agonist treatment arms (Supplementary Table 2d). A substantial loss of E2-stimulated ERBSs occurred in tumors from animals treated with AR agonists (DHT: 97%; enobosarm: 67%), including key ERBSs in the MYB and CCND1 genes (Extended Data Fig. 8a-d), consistent with downregulation of hallmark cell cycle gene sets (Extended Data Fig. 8e). These in vivo data corroborate the in vitro data, showing that AR-mediated antiproliferative activity requires an agonist ligand, involves genomic antagonism of ER signaling and can be induced by a clinically relevant SARM.
To test the efficacy of an AR agonist strategy in the context of endocrine-resistant ER + breast cancer, we employed patient-derived xenograft (PDX) models ( Fig. 3a) derived from the metastases of four patients who had disease progression following one or more lines of endocrine therapy. Consistent with the endocrine-resistant phenotype, three models harbor aberrations in the ESR1 gene, which encodes ER (mutations: UCD4, HCI-005; amplification: UCD65) 37,38 and one model (GAR15-13) has amplification of CCND1 (data not shown), all implicated as drivers of therapy resistance 39,40 . The PDX models express AR and ER (described elsewhere 38 and Extended Data Fig. 9a), but vary in their requirement for exogenous E2 supplementation. Initial experiments with UCD4 HCI-005, n = 7 tumors) via AR activation with DHT (UCD4, n = 15; HCI-005, n = 7) or enobosarm (SARM) (UCD4, n = 14; HCI-005, n = 7). Data were analyzed using a two-tailed, unpaired Student's t-test. Test details are UCD4, t = 4.439 and 3.764, d.f. = 27 and 26 and P = 0.0001 and 0.0009 for e2 + DHT and e2 + SARM treatment groups, respectively; HCI-005, t = 6.996 and 6.176, for e2 versus e2 + DHT and e2 versus e2 + SARM treatment groups, respectively, with d.f. = 12 and P < 0.0001 for each test. c, Two-factor log-ratio (M) plot showing SARM-induced changes in eR and AR enrichment at consensus eRBSs in PDX HCI-005 tumors collected 5 d after treatment. Point color denotes eR consensus peak occupancy; e2 unique (yellow-orange), e2 + SARM unique (orange-red) and e2/e2 + SARM shared (rust; plotted at rear). Point co-ordinates are derived from the average enrichment score of three (AR e2) or four (AR e2 + SARM, eR e2, eR e2 + SARM) replicate tumors for each consensus binding site. example eRBSs associated with eR and AR target genes are highlighted in pink and purple, respectively. d, Genome browser images showing averaged ChIP-seq signals for AR and eR at binding sites associated with eR target genes (MYB, CCND1; left) and AR target genes (MTSS1, ZBTB16; right). Data represent an average of three to four replicate tumors per treatment as described above. e, HCI-005 PDX growth curves. Tumors treated in vivo with vehicle (e2, n = 5), AR agonist (e2 + DHT, n = 5), AR antagonist (e2 + enzalutamide (enz), n = 6) or a combination of AR agonist and antagonist (Combo, n = 4) (left  All data are represented as mean ± s.e.m. and were analyzed using a two-tailed, unpaired Student's t-test. Asterisks denote statistical significance, where *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. and UCD65 revealed that DHT potently inhibited E2-stimulated tumor growth (Extended Data Fig. 9b). Indeed, activation of AR with DHT or enobosarm inhibited tumor growth in all three PDX models of endocrine resistance that have ESR1 genomic aberrations ( Fig. 3b and Extended Data Fig. 9b,c). Consistent with ZR-75-1 xenograft data, AR agonists significantly reduced the proliferative index of HCI-005 tumors within 5 d of treatment (Extended Data Fig. 9d) and potently reduced ER chromatin binding genome-wide, including cell cycle gene loci (Extended Data Fig. 9e,f). As expected, SARM treatment induced AR binding at loci associated with AR-regulated tumor suppressor genes (SEC14L2, ZBTB16, EAF2), but a coincident gain of ER binding was not always evident ( Fig. 3c,d). Rather, a SARM-induced gain in ER binding was stronger at other loci, including a locus near the MTSS1 gene (Fig. 3c,d), which encodes a suppressor of breast cancer metastasis 41 . Hence, in endocrine-resistant breast cancer with genomic aberrations of ESR1, activation of AR redistributed ER chromatin binding and reduced growth. To directly compare an AR agonist with an AR antagonist approach, additional in vivo experiments were performed with the HCI-005 PDX model and in vitro experiments      Fig. 5 | Clinical significance of AR activity in primary ER + breast cancers. a, Kaplan-Meier survival plot showing the ability of an AR-associated gene signature (n = 142 genes) derived from three in vivo xenograft models (ZR-75-1; HCI-005; GAR15-13) to stratify eR + cases in the MeTABRIC breast cancer cohort into good and poor outcome groups. Data were analyzed using the log-rank test; χ 2 = 58.9, d.f. = 1. P values are indicated within the graph. b, Graphs of nonmalignant (n = 144), good prognosis (n = 839) and poor prognosis (n = 389) eR + breast cancers in a, depicting expression levels of genes in the AR signature that were downregulated (left) or upregulated (right) by AR activation. Box plots illustrate the interquartile range (IQR) of the data distribution; Q1 (25th percentile) Q2 (median) and Q3 (75th percentile). The minima and maxima, representing 1.5× IQR, are denoted by whiskers. Individual data points denote outliers. Data were analyzed using the Wilcoxon signed-rank test. P values are shown within the graph in gray. c, Average read density plots for eR (left) and AR (right) chromatin binding proximal (<100 kb) to AR signature genes in ZR- 75 with a cell line derived from the UCD4 PDX model 42 . While the AR antagonist had no effect on E2-stimulated growth, it curtailed the growth inhibitory effect of DHT ( Fig. 3e and Extended Data Fig. 9g). Notably, an AR agonist had greater antitumor efficacy than an ER antagonist (tamoxifen) in the HCI-005 PDX model (Fig. 3e). The GAR15-13 PDX model is distinct from the others as it does not require exogenous E2 for robust basal growth in vivo 10 . Nonetheless, treatment with AR agonists potently reduced growth of GAR15-13 tumors, whereas treatment with the AR antagonist had no effect on short-(Extended Data Fig. 9h-i) or long-term (Fig. 3f) measures of tumor growth. These observations, and ability of the AR antagonist to curtail activity of the AR agonist, were independently confirmed in the GAR15-13 PDX model (Extended Data Fig. 9j,k). Collectively, these therapeutic studies involving the most clinically relevant in vivo model systems provide strong support for an AR agonist strategy in multiple scenarios of advanced ER + breast cancer.
As AR has been specifically implicated in the promotion of tamoxifen resistance 10,11,13 , we established mammary intraductal (MIND) xenografts 43 (Fig. 3g) using a derivative of the ER + MCF7 breast cancer cell line rendered resistant to tamoxifen (Tam R ) ( Supplementary  Fig. 3a,b). AR activation reduced in vitro colony formation and induced cell cycle arrest in the parental MCF7 and Tam R cells ( Supplementary Fig. 3c,d). Within mouse mammary ducts, Tam R cells formed highly proliferative, invasive tumors by 3 months and these remained positive for ER and AR expression ( Supplementary  Fig. 3e). Mice were injected with Tam R cells and after 30 d were randomly allocated to vehicle or DHT treatment arms. DHT significantly reduced tumor weight ( Supplementary Fig. 3f) and the proliferative index within in situ and invasive lesions (Fig. 3h). The MIND xenograft data show that AR sustains a tumor-suppressor role in the specific context of tamoxifen resistance.
Inhibitors of cyclin-dependent kinase 4/6 (CDK4/6i; for example palbociclib) in combination with endocrine therapy are becoming standard of care for women with advanced breast cancer 44 . The combination of an AR agonist and palbociclib exerted the greatest growth inhibition in Tam R and UCD4 cell lines in vitro ( Fig. 4a and Supplementary Fig. 4). Combination therapy was also the most potent inhibitor of Tam R colony formation (Fig. 4b). These observations were recapitulated in vivo using the GAR15-13 PDX model, in which the superior growth inhibitory effect of the combination therapy over palbociclib treatment alone was evident in the proliferative index of tumors collected after 5 d of treatment (Fig. 4c,d). While CDK4/6 inhibitors are effective in advanced breast cancer, the development of resistance is inevitable 45,46 . Therefore, we created an MCF7-derivative cell line that was resistant to palbociclib (Palb R ) (Fig. 4e). Palb R cells had a compensatory increase in CDK2 expression, indicative of resistance to CDK4/6i 40 , but maintained expression of AR and ER comparable to the parental line (Fig. 4f). AR agonists reduced proliferation of Palb R cells, but growth inhibition was enhanced in the presence of palbociclib, suggesting that AR agonism may have partially restored sensitivity to the CDK4/6i (Fig. 4g). Hence, an AR agonist strategy is a viable option for combination therapies involving a CDK4/6 inhibitor, even in cases of acquired resistance to CDK4/6 inhibition.
A gene signature of AR activity is prognostic in ER + breast cancer. A gene signature of AR activity was generated by integrating RNA-seq transcriptomic data from multiple in vivo models (ZR-75-1, HCI-005, GAR15-13). The resulting 142 AR gene signature (59 upregulated and 83 downregulated genes; Supplementary Table 3a-d) stratified ER + cases in the METABRIC breast cancer cohort 15 into good and poor outcome groups (Fig. 5a). Downregulated genes were expressed at lower levels in tumors with a better survival outcome (Fig. 5b) and these genes were enriched for hallmark G2/M checkpoint (for example MKI67, AURKB, CCNB2) and estrogen response (for example MYB, CXCL12, IGFBP4) pathways (Extended Data Fig. 10a). In ZR-75-1 xenograft and HCI-005 PDX models, ER chromatin binding associated with downregulated genes was markedly reduced by AR activation (Fig. 5c). Upregulated genes in the AR signature were expressed at higher levels in tumors from patients with a better survival outcome (Fig. 5b), were enriched for the hallmark androgen response gene set (for example SEC14L2, EAF2, ZBTB16) (Extended Data Fig. 10a) and were associated with increased AR chromatin binding in ZR-75-1 xenograft and HCI-005 PDX tumors treated with AR agonists (Fig. 5c). These data indicate that signature downregulated genes are largely ER target genes and upregulated genes are AR target genes, consistent with changes in the H3K27ac signal (Extended Data  Fig. 10b,c) and the concept that activation of AR target genes results in trans-repression of ER target genes. The data also accord with transcriptional data generated in PDEs (Fig. 1). The AR gene signature had highly significant prognostic performance in the METABRIC 15 and ROCK 47 cohorts as well as in an ER + tamoxifen-treated dataset 48 (Fig. 5d). In an independent assessment, the AR signature had greater prognostic power than cancer-associated signatures (MSigDB) and gene sets derived or annotated from the MammaPrint gene signature (Fig. 5d), as well as other commercially available genomic tests (OncotypeDx, EndoPredict and PAM50) used for breast cancer prognostication (Supplementary Table 3e). Thus, endogenous AR activity is evident in primary ER + breast cancers, is positively associated with longer disease-free survival and has significant prognostic power. This signature may be useful in breast cancer prognostication or for patient selection and assessment of treatment response in future trials testing an AR agonist target therapy.

Discussion
Herein, we provide extensive evidence that AR is a tumor suppressor in multiple contexts of ER + breast cancer, including endocrine resistance, which strongly supports an AR agonist strategy for treatment. Activated AR inhibited ER-driven growth of breast cancers by displacing ER and critical transcriptional co-activators (p300, SRC-3) from chromatin at ER-regulated cell cycle genes, leading to transcriptional downregulation. It also upregulated AR target genes, sequestering p300 and SRC-3 away from ER target genes in the process (Fig. 5e). AR target genes in ER + breast cancers included tumor suppressors that may directly inhibit tumor growth; regulation of these genes may occur in co-operation with or independent of ER signaling (Fig. 5e). Such mechanisms likely underpin the sex hormone antagonism evident in normal breast tissues, as treatment with androgen opposed estrogen-stimulated proliferation and induction of cell cycle genes in patient-derived explants of normal and ER + malignant breast tissues, while simultaneously upregulating AR target genes (Fig. 1).
Trans-repression of one transcription factor by another via squelching of a critical cofactor is emerging as an important biological regulatory mechanism with implications for tissue development, homeostasis and pathology 49 . Indeed, squelching of p300 is a mechanism by which ER indirectly represses gene transcription, transiently binding at repressed genes to hijack p300 to activated genes 50 . The relatively weak AR binding we observed at ER target genes that were repressed upon AR activation suggests a similar mode of action. Previous evidence generated in silico or in the context of nonchromatinized DNA indicate that AR can bind to an ERE 18,35 , which may play a role in the ability of AR to displace ER from chromatin at loci associated with repressed ER target genes. While p300 is clearly a pivotal factor, with a direct and powerful influence on the activation of H3K27, our data indicate that other cofactors common to AR and ER, including SRC-3 (ref. 51 ), likely play a role in crosstalk between these receptors and warrant further investigation, especially as a new study shows that AR and ER differentially interact with p300 and SRC-3 when bound to DNA 34 .
The PR is also able to modulate the interaction of ER with chromatin to reduce proliferation of breast cancer cells 17,38 , but the mechanism involved is different to that exerted by AR. Activated PR directly interacts with ER, resulting in a widespread gain of ER binding events in vitro and in vivo. In contrast, a direct interaction between AR and ER has not been detected to date 18 , and activation of AR caused a more prevalent loss of ER-binding events, especially in vivo. The differences between AR and PR interaction with ER make biological sense considering that androgen inhibits pubertal breast development, whereas progesterone stimulates functional differentiation (for example lactation). While stimulation of PR has therapeutic potential in primary ER + breast cancers 52 , loss of PR expression is a common event 17 . Conversely, AR expression is largely maintained in primary and metastatic lesions 2 (Fig. 1) and therefore represents a more prevalent therapeutic target, especially in the endocrine-resistant setting.
This work has immediate implications for women with metastatic ER + breast cancer. It provides compelling evidence that exploitation of an endogenous, ligand-activated AR pathway to inhibit ER-mediated transcriptional activity constitutes an attractive, implementable therapeutic opportunity to treat ER + breast cancers, even those resistant to current forms of endocrine therapy and those with genomic aberrations of ESR1 or CCND1. Moreover, we provide new evidence that AR agonists can be more effective than existing (for example tamoxifen) or new (for example palbociclib) standard-of-care treatments and, in the case of the latter, can be combined to enhance growth inhibition. The new insights from this study should dispel widespread confusion over the role of AR in ER-driven breast cancer, an issue that currently hinders progress in leveraging modern AR-targeted therapies, including available SARMs that lack the undesirable side effects of androgens 9 , for clinical benefit. Moreover, SARMs such as enobosarm can confer health benefits in women, including promotion of bone, muscle and mental health 9 . Given the efficacy of AR agonism at multiple stages of ER + disease, this treatment strategy has potential to become an alternative endocrine therapy for breast cancer.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41591-020-01168-7. Tissues were cultured ex vivo as previously described 16,17 . In brief, the tissue specimen from a patient was cut into small pieces, randomly allocated to a gelatin sponge (3-4 per sponge) pre-soaked in culture medium and incubated at 37 °C with 5% CO 2 for ~36 h. Subsequent hormone treatments included E2 (10 nM) alone or in combination with androgen (E2 + DHT; 10 nM each). Tissue explants were treated with hormones for 24 h to extract RNA or 48 h to assess proliferation markers (Ki67, BrdU). In all experiments, representative pieces of tissue were processed for histological analysis by a pathologist (S.J.) to quantify the presence and proportion of normal or malignant breast epithelial cells within the explants. Those with at least 60% epithelial content were utilized for downstream analyses (IHC and RNA-seq). RNAi. RNAiMAX (Invitrogen) was used at 0.5 µl cm −2 to reverse-transfect cells with siRNAs at time of seeding according to standard protocol; siRNA targeting AR (siAR(1), Ambion S1539, 5 nM; siAR (2) For colony-formation assays, cells were seeded into 12-well plates the day before treatment. Indicated treatments were refreshed every 3 d. Colonies were visualized with a crystal violet staining solution (0.05% w/v crystal violet, 1% formaldehyde and 1% methanol). Images of stained wells were captured using the Perfection V800 Scanner (Epson) and area occupancy was measured using the ColonyArea FIJI plug-in 54 .
Fluorescence-activated cell sorting. Standard techniques were employed to sort propidium iodide-stained cells on the BD FACS Canto II (Becton Dickinson). Cell cycle phase analyses were performed on ModFit LT (Verity) without gating, where subdiploid populations were excluded from further analyses. Source data were visualized using Flowing Software (Cell Imaging Core, Turku Centre for Biotechnology).
Immunoblotting. In vitro experiments with protein end points were collected by scraping into radioimmunoprecipitation assay buffer at time points as indicated in figure legends. Protein concentration was quantified by BCA protein assay (Thermo Fisher Scientific) before immunoblotting. Standard techniques were employed using primary antibodies from Abcam (AR, Ab108341; To determine kinase activity, immunoprecipitates of CDK2 were collected with protein A-sepharose beads (Zymed, Invitrogen) from 100 μg of cell lysate and the activity of each toward histone H1 substrates was determined in kinase reactions with [γ-32 P]ATP followed by polyacrylamide gel electrophoresis and exposure of dried gels to X-ray film as described 55  When tumors became palpable, they were measured repeatedly twice weekly using electronic calipers to monitor growth kinetics and volume was calculated using the formula (length × width 2 ) / 2. Tumor-bearing mice were randomly assigned into relevant treatment groups when tumor volume reached ~200 mm 3 (n = 4-15 mice per group for therapeutic studies, n = 3-5 mice per group for biochemical studies; exact numbers are specified in figure legends). The SARM, enobosarm (10 mg kg −1 d −1 ) and the AR antagonist, enzalutamide (20 mg kg −1 d −1 ), were resuspended in 25% Tween 80 and delivered daily by oral gavage. Palbociclib (25 mg kg −1 d −1 ) was dissolved in water and delivered daily by oral gavage as described 57 . DHT and tamoxifen pellets were surgically implanted subcutaneously. Upon reaching ethical or predefined experimental end points, mice were killed and the primary tumor collected. After weighing, the tumor was cut into pieces that were allocated to be snap frozen, fixed overnight at 4 °C in 10% neutral-buffered formalin or embedded in cryoprotective optimal cutting temperature compound before being snap frozen. Frozen samples were kept at −80 °C. Formalin-fixed samples were sent to the Garvan Institute Histology Core Facility for paraffin embedding. MIND xenografts were established as described 38 by injecting Tam R cells (1 × 10 5 cells 10 µl −1 ) into the fourth inguinal mammary ducts of NSG mice (Australian BioResources). Tam R cells formed DCIS and IDC lesions within the injected glands by 3 months after injection ( Supplementary Fig. 3). To examine the effect of DHT, mice were injected with Tam R cells and after 1 month of incubation, were randomly allocated to vehicle (n = 7) and DHT (n = 7) treatment arms. Injected mammary glands were collected 2 months later for histochemical analyses.  Fig. 4 was performed using Andy's algorithm DAB + IHC pipeline for FIJI 58 , calculating the average percentage positivity for each tumor based on ten representative images collected at ×20 magnification. All other Ki67 and BrdU staining was quantified by manual counting of at least 1,000 cells or all visible epithelial cells, whichever threshold was reached first, over multiple representative areas.
FastQ files were quality checked (FastQC Galaxy v.0.70) 60 , concatenated and then trimmed (Trimmomatic Galaxy v.0.35) 61 to remove poor-quality reads and adapters. Trimmed reads were aligned to the hg19 genome assembly using Bowtie2 (ref. 62 ). Poorly mapped (MAPQ < 10) and duplicate reads were removed using SAMtools 63 . Peaks were called using MACS2 (default settings) 64,65 against a pooled input sample. A modified MACS2 cutoff (P < 0.005) was used to call peaks in data associated with GSE141582 (Extended Data Fig. 4c,d). For cell line ChIP-seq data, consensus cistromes were created using BEDTools 66 and included peaks called in at least two of the three independent replicates for a given factor and experimental condition. For xenograft ChIP-seq data, FASTQ files were aligned to hg19 using bwa (v.0.7.15; default parameters 67 ) and mapped reads with a minimum MAPQ < 15 were removed using SAMtools. Peaks were called using MACS2 (default settings) and consensus cistromes were created using DiffBind as previously described 68 and included peaks called in at least two of the total replicates representing a given factor and treatment condition. Occupancy analyses were performed using BEDTools, whereby 'lost' , 'gained' and 'shared' binding sites were identified from consensus cistromes corresponding to different treatment conditions for a specific factor; 'unique' and 'shared' binding sites were identified from consensus cistromes corresponding to different factors under the same treatment condition. ChIP-seq data were visualized on the genome using the Integrative Genomics Viewer 69 ; whereby an average enrichment signal from all replicates was generated by merging mapped reads for a given condition, followed by subsequent conversion and normalization to bigwig format using deepTools 70 or HOMER 71 . Heat maps were generated using deepTools, whereby the average enrichment signal from all replicates for a particular factor and treatment condition was assessed at specific sites determined by a selected consensus cistrome of interest. Two-factor log-ratio (M) enrichment plots were generated after occupancy analyses using deepTools multiBigwigSummary, whereby the average ChIP-seq enrichment scores for all replicates representing a factor and treatment condition of interest were calculated at specific sites determined by selected consensus cistromes. The resultant matrix was annotated using CisGenome 72 and the log 2 enrichment ratio between treatments for each of two factors of interest was calculated and plotted in R using ggplot2, where point color corresponds to the occupancy analysis results. In the two-factor log 2 ratio plots, a limited number of points may appear out of context (that is sites identified as 'lost' that have an increased log 2 ratio value). This occurs when one replicate had stronger enrichment than other replicates for a factor and treatment condition but was not called as a peak in at least two replicates. These points tend to represent weak binding sites. Tag density plots were generated using HOMER or deepTools at consensus binding sites of interest, which are further identified in figures and/ or figure legends. Down-and upregulated genes associated with ER/AR binding events were identified using BETA 73 or by annotating peaks within 100 kb of a transcription start site (TSS) with CisGenome followed by data integration using BEDTools. Differential de novo motif analysis was determined for each consensus cistrome (for example ER 'lost' or ER 'gained' binding sites), using the opposing dataset as background, in the HOMER findmotifsgenome.pl function (-bg flag; hg19 genome build; 200-bp window flanking the center of each peak).
ChIP-PCR reactions were prepared using iQ SYBR Green Supermix (BIO-RAD) and primers as listed in Supplementary Table 4a. PCR was performed with the CFX384 Real Time PCR Detection System (BIO-RAD) and standard cycling conditions. ChIP-PCR data were analyzed by the percentage input method and further analyzed as fold enrichment over IgG or negative control, where appropriate.
RNA integrity was assessed using the Experion RNA StdSens Analysis kit For RNA-seq in ZR-75-1 cells grown in vitro and PDEs: hg19 alignment was performed using CLC Genomics Workbench 11 (QIAGEN) using the following mapping parameters: mismatch cost 2; insertion cost 3; deletion cost 3; length fraction 0.8; and similarity fraction 0.8. For in vitro ZR-75-1 cells, multiple mapped reads from a given sample were merged after mapping. Differential gene expression was performed using a generalized linear model with false discovery rate (FDR) correction based on negative binomial distribution. For PDEs; treatment comparisons were performed on samples individually due to tissue heterogeneity. Genes common between both Kahl's z test 74 (FDR < 0.01) and Baggerley's test 75 (FDR < 0.01) were determined to be significantly differentially regulated. For both in vitro ZR-75-1 and PDEs: expression was measured based on the total counts of mapped short reads and normalized based on RPKM (CLC Genomics Workbench).
For gene set enrichment analysis 85 of RNA-seq data, genes were ranked in E2 versus E2 + DHT or E2 versus E2 + SARM conditions according to expression using the Signal2Noise metric and gene set enrichment analysis was implemented using the Broad Institute's public GenePattern server, with default parameters. To focus on pathways most relevant to the study, we specifically selected pathways relevant to the cell cycle, estrogen or androgen response and mammary gland biology.
For RNA-seq validation, quantitative PCR with reverse transcription was performed. RNA was extracted from consecutive passages of cell lines (MCF7 n = 3; ZR-75-1 and T-47D n = 4) or tumors collected 5 d after treatment in vivo (vehicle, enzalutamide, enobosarm n = 4; DHT n = 5) and quantified as described above. RNA was DNase-treated using the TURBO DNase kit (Invitrogen) and reverse transcribed using the iScript cDNA Synthesis kit (BIO-RAD). PCR with reverse transcription was performed as per ChIP-qPCR but using primers outlined in Supplementary Table 4b. Gene expression was calculated by the 2 -ΔΔCt method and normalized to the expression of GAPDH (cell lines) or IPO8 and PUM1 (tumors), using the CFX Manager Software (BIO-RAD).
Generation of an AR-associated gene signature. Identification of an AR-associated gene signature was derived from DHT-induced genes in ZR-75-1, HCI-005 and GAR15-13 xenograft models as described 86 . This methodology involves a series of filtering steps including (1) gene matching, (2) identification of cancer-specific genes and (3) estimation of the prognostic value of each gene. The Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset 15 , which constitutes 1,230 cases of patients with clinically annotated ER + luminal breast cancer with long-term survival information, was used to investigate the prognostication potential of the DHT-induced gene set.

Statistical analyses.
Statistical calculations were performed using either STATISTICA (Stat Soft) SPSS (v.17) and GraphPad Prism (GraphPad Software). Normality was assumed for all statistical tests unless otherwise stated. Multiple comparisons were adjusted for as indicated in the statistical tests below.
The Nottingham Tenovus Primary Breast Cancer Series was analyzed using Pearson's chi-squared, Student's t-test and ANOVA tests, where appropriate. Cumulative survival probabilities and 10-year breast cancer-specific survival (BCSS) and disease-free survival (DFS) were estimated using the univariate Cox proportional hazards models and the Kaplan-Meier plot method where appropriate and differences between survival rates were tested for significance using the log-rank test. Multivariable analysis for survival was performed using the Cox proportional hazard model. The proportional hazards assumption was tested using standard log-log plots. Hazard ratios and 95% confidence intervals were estimated for each variable. All tests were two-sided with a 95% confidence interval and a P value <0.05 was indicative of statistical significance.
In vitro proliferation experiments were analyzed either by two-tailed Student's t-test ( Supplementary Fig. 3c) or by one-way (Fig. 4a,b,g) or two-way (Fig. 4e) ANOVA, followed by Tukey's multiple comparisons test. Further details are in figure legends.
Tumor growth curve data was analyzed at ethical end point using a two-tailed, unpaired Student's t-test. End-point tumor mass was analyzed using a two-tailed, unpaired Student's t-test. Further details are in figure legends. IHC data were analyzed by a two-tailed, unpaired Student's t-test ( Fig. 3h and Extended Data Figs. 7c and 9d,h) or a one-way ANOVA and Tukey's multiple comparisons test (Fig. 4d). PDE experiments (Fig. 1b and Extended Data Fig. 2a) were analyzed using a two-tailed Wilcoxon matched-pairs signed-rank test and a two-tailed Pearson's correlation test was used to analyze cases of Ki67-and BrdU-positive explants. Further details are in figure legends.
ChIP-PCR experiments for p300 and SRC-3 in ZR-75-1 or T-47D cells and all ChIP-PCR experiments involving MCF7 AR and AR mut cells were analyzed using a two-way ANOVA, followed by Tukey's multiple comparisons test. ChIP-seq validation experiments, including ER and AR ChIP-PCR in ZR-75-1 cells, were analyzed by an unpaired, one-tailed Student's t-test. Further details are in figure legends. AR signature analyses were performed in R using survival and SigCheck. Statistical analyses were performed by Kaplan-Meier log-rank, Kruskal-Wallis (pairwise comparisons) and Wilcoxon tests where appropriate.
Life Sciences Reporting Summary. Further information on experimental design is available in the Life Sciences Reporting Summary.

Data availability
All ChIP-seq and RNA-seq data are deposited in the Gene Expression Omnibus under accession number GSE123770. Source data are provided with this paper. Fig. 1 | AR protein levels are prognostic in ER-positive breast cancer. Immunogenic AR protein levels predict breast cancer-specific survival in the Nottingham Tenovus Primary Breast Cancer Series 14 . Kaplan-Meier plots depicting the association between immunogenic AR positivity (dichotomized by high ≥ 78% and low < 78% nuclear staining based on an optimal cut-point criterion 8 ) and breast cancer-specific survival. a, All cases; b, eR-positive cases (defined by immunogenic eR positivity of >1% as per current ASCO guidelines); c, High-risk (defined by Nottingham Prognostic Index ≥ 3.4) eR-positive cases treated with adjuvant endocrine therapy (Tamoxifen; Tam); d, eR-positive cases not treated with adjuvant Tamoxifen. Survival probabilities were estimated using a univariate Cox proportional hazards model. Differences between survival rates were tested for significance using the log-rank test. Hazard ratios (HR), 95% confidence intervals (95% CI), and p values are provided within each graph. All tests were two-sided. Fig. 2 | Anti-proliferative effects of androgen in patient-derived explants (PDEs) of breast tissue exposed to estrogen. a, Quantification of BrdU immunohistochemistry in eR + malignant patient-derived explants (PDes; n = 9) treated with estrogen (e2, 10 nM) or e2 plus androgen (DHT, 10 nM). Data represented as mean ± SeM. Data was analyzed using a two-tailed Wilcoxon matched-pairs signed rank test (W = -45.0, r = 0.1149, **p = 0.0039). b, Correlation between immunohistochemical quantification of both BrdU-and Ki67-positive cells shown in (a) and (Fig. 1b). At least 1,000 cells, or all visible epithelial cells, whichever threshold was reached first, were examined in n = 9 independent cases of eR + malignant PDes, treated with either e2 (left panel) or e2 + DHT (right panel). Data was analyzed by linear regression (trend line) and a two-tailed Pearson's correlation test (e2 treatment; r = 0.697 [0.061, 0.931]; e2 + DHT treatment; r = 0.912 [0.627, 0.982]. P values are provided within each graph. c,d, RNA-seq data derived from PDes of (c) normal breast tissue (n = 17 cases) and (d) eR + invasive breast cancer (n = 8 cases) treated with e2 (10 nM) or e2 + DHT (10 nM each). Heat maps depict the top 200 differentially expressed genes. e, Gene set enrichment plots for the Hallmark G2M Checkpoint gene set derived from RNA-seq data representing normal breast tissues (left panel) or eR + breast cancers (right panel). Fig. 3 | AR-mediated changes in ER binding at ER and AR target genes. a, Motif analysis of e2-stimulated eR binding sites associated with Fig. 2a,b, which are altered in the presence of androgen (DHT) in ZR-75-1 cells. GRHL (grainy-head-like motif); eRe (estrogen response element); ARe (androgen response element). b, Cropped immunoblot showing effective siRNA-mediated knock-down of AR (left panel) and eR (right panel) in ZR-75-1 cells treated with estrogen (e; 10 nM) alone or in combination with androgen (e + D; 10 nM each) for 4 hours. β-Actin was used as a loading control. Blots are representative of three experiments corresponding to independent passages of cells. c, Left panels: Genome browser images showing eR and AR ChIP-seq signals (associated with Fig. 2b) at e2-stimulated eR binding sites (BCL2, PGR) that are reduced in the presence of androgen (DHT, 10 nM). Right panels: Corresponding eR ChIP-PCR for BCL2 (upper panel) and PGR (lower panel) following AR knockdown and subsequent treatment with estrogen (e2, 10 nM) alone or in combination with androgen (e2 + DHT, 10 nM each) for 4 hours in ZR-75-1 cells. Data was analyzed using a one-sided unpaired t-test. Upper panel test details: t = 5.320, 0.2325, and 0.2274, for siCon (p = 0.0030), siAR (1) (p = 0.4138), and siAR (2) (p = 0.4915), respectively. Lower panel test details: t = 2.074, 0.2437, and 0.3767 for siCon (p = 0.0534), siAR (1) (p = 0.4097), and siAR (2) (p = 0.3628), respectively; df = 4 for each test. d, Left panels: Genome browser images showing AR and eR ChIP-seq signals (associated with Fig. 2b) at DHT-stimulated AR binding sites (ZBTB16, SEC14L2). Right panels: Corresponding AR ChIP-qPCR for ZBTB16 (upper panel) and SEC14L2 (lower panel) following eR knockdown and subsequent treatment with estrogen (e2, 10 nM) alone or in combination with androgen (e2 + DHT, 10 nM each) for 4 hours in ZR-75-1 cells. each genome track in (c) and (d) depicts the average read density of all ChIP-seq replicates for the designated receptor and treatment condition. Data was analyzed using a one-sided unpaired t-test.  Fig. 4 | AR-mediated modification of the ER cistrome requires an agonist ligand and AR DNA binding capacity. a, Genome browser image of eR and AR chromatin binding events after treatment for 4 hours with either estrogen (e2, 10 nM) or estrogen plus androgen (e2 + DHT, 10 nM each) in ZR-75-1 cells at the MYB gene locus based on ChIP-seq data from Fig. 2. each genome track depicts the average read density of all ChIP-seq replicates for the designated receptor and treatment condition. enh., enhancer. Tracks are scaled to facilitate visualization of the Intron 1 locus. b, ZR-75-1 in vitro ChIP-qPCR validation data for eR, H3K27ac and AR at three loci associated with the MYB gene locus as depicted in (a). Bar graphs depict mean ± SeM of 3 independent passages of cells. Data was analyzed using a one-sided unpaired t-test. eR ChIP; t = 1.345, 1.17, and 0.5662 for enh. 1 (p = 0.1249), Intron 1 (p < 0.0001), and enh. 2 (p = 0.3101), respectively. H3K27ac ChIP; t = 0.3320, 2.310, and 0.5664 for enh. 1 (p = 0.3783), Intron 1 (p = 0.0410), and enh. 2 (p = 0.3007), respectively. AR ChIP; t = 3.555, 1.480, and 3.993 for for enh. 1 (p = 0.0118), Intron 1 (p = 0.1064), and enh. 2 (p = 0.0085), respectively; df = 4 for each test. Grey asterisks denote statistical significance. c, Average read density plot of eR ChIP-seq enrichment at eR binding sites proximal (<100 kb) to genes down-regulated in e2 + DHT versus e2 treatment arms in ZR-75-1 cells, as per extended Data Fig. 3e. Data shows eR enrichment in ZR-75-1 cells following in vitro treatment with Vehicle (Veh), estrogen (e2, 1 nM), an AR agonist (DHT, 1 nM), an AR antagonist (Bicalutamide; Bic 1 µM), alone or with indicated combinations. The data independently recapitulates the DHT-induced reduction in the e2-stimulated eR signal observed in extended Data Fig. 3c,e at 10 nM doses of hormones, and using an independent eR antibody; Millipore 06-935. Treatment with an AR antagonist does not induce loss of eR at these loci. d, Genome browser images of eR binding sites in ZR-75-1 cells at eR target genes (BCL2, PGR), showing reduced binding following treatment with an AR agonist (DHT) but not an AR antagonist (Bicalutamide). each genome track depicts the average read density of all ChIP-seq replicates for the designated receptor and treatment condition. e, Growth curve showing that estrogen (e2, 1 nM)-stimulated in vitro growth of ZR-75-1 cells is inhibited by an AR agonist (DHT, 1 nM) but not an AR antagonist (Bicalutamide, 1 µM). Data represents mean ± SD of three replicate cell culture wells per condition, and is a representative of two independent experiments. f-i, experiments in MCF7 breast cancer cells transformed to overexpress a constitutively active, truncated AR (1-707 aa; AR) or a mutant AR that is unable to bind DNA (AR mut ). AR expression is induced upon treatment with doxycycline (Dox). f Cropped immunoblot showing Dox-induced expression of truncated AR in cells treated with Vehicle or estrogen (e2, 1 nM), and resultant eR expression. β-Actin was used as a loading control. Blots are representative of three experiments corresponding to independent passages of cells. g Heatmap of RT-qPCR data for two eR target genes (MYB, CCND1) and two AR target genes (SEC14L2; ZBTB16) in cells treated with Vehicle (Veh) or estrogen (e2, 1 nM). Data represents the average normalized gene expression from three independent passages of cells replicates. h eR ChIP-qPCR at loci associated with eR target genes (MYB, CCND1) (top panels) and an AR target gene (ZBTB16). AR ChIP-qPCR is also shown for the AR target gene. Data was analyzed by a two-way ANOVA followed by Tukey's multiple comparisons test. Upper-left panel; F = 125.9, 11.75, and 3.143 for presence of e2 (p < 0.0001), AR status (p = 0.0015), and their interaction (p = 0.0799), respectively. Asterisks denote statistical significance, where **p = 0.0060, NS p = 0.9998. Upper-right panel; F = 81.79, 13.85, and 1.565 for presence of e2 (p < 0.0001), AR status (p = 0.0008), and their interaction (p = 0.2489), respectively; where **p = 0.0094, NS p > 0.9999. Lower-left panel; F = 140.8, 284.6, and 94.59 for presence of e2 (p < 0.0001), AR status (p < 0.0001), and their interaction (p < 0.0001), respectively; where ****p < 0.0001. Lower-right panel; F = 0.1975, 147.1, and 0.09898 for presence of e2 (p = 0.6647), AR status (p < 0.0001), and their interaction (p = 0.9065), respectively; where ****p < 0.0001. df = 4 for each test. Bar graphs depict mean ± SeM of 3 independent passages of cells. NS = Not significant; **p < 0.01; ****p < 0.0001. i, Growth curve showing that induction of constitutively-active AR inhibits e2-stimulated growth but induction of the constitutively-active AR mut does not. Data represents mean ± SD of three replicate cell culture wells per condition, and is a representative of two independent experiments. Fig. 5 | Activation of AR relocates p300 from ER to AR target genes. a, Replicate data for p300 ChIP-seq in vitro experiments in ZR-75-1 cells associated with Fig. 2e. Venn diagrams show the overlap of three independent experiments representing consecutive passages of cells treated with vehicle (Veh), estrogen (e2; 10 nM) or estrogen plus androgen (e2 + DHT; 10 nM each). Peaks present in at least 2 of 3 replicates were used to generate a consensus cistrome, indicated below the Venn diagrams, for further comparative analyses. b, Overlap of consensus p300 cistromes under e2 or e2 + DHT hormone treatments, after subtracting peaks present under basal (Veh) conditions to generate as set of hormone-regulated peaks. c, Consensus p300 ChIP-seq data from (b), associated with Fig. 2e, showing average read density plots (top panels) and heatmaps (bottom panels), illustrating changes in hormone-regulated p300 chromatin binding sites following treatment with e2 + DHT. d, Average read density plots for p300 binding at eR binding sites (eRBS) proximal (<100 kb) to genes down-regulated by androgen under estrogenic conditions (left panel), and AR binding sites (ARBS) proximal (<100 kb) to genes up-regulated under the same conditions (right panel). e, example genome browser images showing averaged p300 ChIP-seq signals at binding sites associated with an eR target gene (PGR; left panel) and AR target gene (SEC14L2; right panel) in ZR-75-1 cells. Data represents the average signal of three replicates. f, p300 ChIP-qPCR at enhancers of eR-regulated cell cycle genes (as per Fig. 2g) in T-47D cells treated in vitro under designated hormone conditions. Data was analyzed by a two-way ANOVA (F = 93.45, 80.19, and 14.78 for hormone treatment (p < 0.0001), test site (p < 0.0001), and their interaction (p < 0.0001), respectively; df = 30). Data represented as mean ± SeM of 3 independent passages of cells. Post-hoc analyzes were performed using Tukey's multiple comparisons test, where asterisks denote statistical significance; *p = 0.0275; *** p = 0.0040; ****p < 0.0001. g, Heatmap of RT-qPCR data for genes associated with Fig. 2g assessed in ZR-75-1 and T-47D cells treated in vitro with estrogen (e2, 10 nM) alone or in the presence of androgen (e2 + DHT, 10 nM each). Data represents the average normalized gene expression of four experiments conducted on independent passages of cells. Fig. 8 | ZR-75-1 xenograft sequencing data representing tumors harvested 5 days post-AR agonist treatment in vivo. a, Read density plots (upper panel) and heatmaps (lower panel) of eR ChIP-seq data from ZR-75-1 xenograft tumors, illustrating the loss of eR binding upon activation of AR by agonists (SARM, DHT). Data is presented as an average of 4 (e2 + Veh) or 5 (e2 + SARM, e2 + DHT) replicate tumors. b, Two-factor log-ratio (M) plot showing SARM-induced changes in eR and AR enrichment at consensus eR (yellow-orange), AR (blue), or Shared (that is AR and eR co-occupied; grey-brown; plotted at rear) binding sites seen across the Vehicle, SARM, and DHT treatment arms. example binding sites near eR target genes and AR target genes are highlighted in pink and purple, respectively. enrichment scores were calculated from an average of 4 (eR, e2) or 5 (eR, e2 + SARM, AR e2, e2 + SARM) tumors. c, Read density plots (upper panel) and heatmap (lower panels) depictions of eR ChIP-seq presented in (a,b). d, Genome browser images of eR and AR binding in ZR-75-1 xenograft tumors treated with e2 or e2 + SARM at loci associated with eR target genes (MYB, CCND1; left panels) and AR target genes (SEC14L2, ZBTB16; right panels). Data represents an average of 4-5 replicates as described in (b). e, Normalized enrichment scores for signature gene sets correlated with e2 or e2 + SARM treatments, derived using RNA-seq data associated with extended Data Fig. 7e.