Androgen receptor blockade promotes response to BRAF/MEK-targeted therapy

Treatment with therapy targeting BRAF and MEK (BRAF/MEK) has revolutionized care in melanoma and other cancers; however, therapeutic resistance is common and innovative treatment strategies are needed1,2. Here we studied a group of patients with melanoma who were treated with neoadjuvant BRAF/MEK-targeted therapy (NCT02231775, n = 51) and observed significantly higher rates of major pathological response (MPR; ≤10% viable tumour at resection) and improved recurrence-free survival (RFS) in female versus male patients (MPR, 66% versus 14%, P = 0.001; RFS, 64% versus 32% at 2 years, P = 0.021). The findings were validated in several additional cohorts2–4 of patients with unresectable metastatic melanoma who were treated with BRAF- and/or MEK-targeted therapy (n = 664 patients in total), demonstrating improved progression-free survival and overall survival in female versus male patients in several of these studies. Studies in preclinical models demonstrated significantly impaired anti-tumour activity in male versus female mice after BRAF/MEK-targeted therapy (P = 0.006), with significantly higher expression of the androgen receptor in tumours of male and female BRAF/MEK-treated mice versus the control (P = 0.0006 and P = 0.0025). Pharmacological inhibition of androgen receptor signalling improved responses to BRAF/MEK-targeted therapy in male and female mice (P = 0.018 and P = 0.003), whereas induction of androgen receptor signalling (through testosterone administration) was associated with a significantly impaired response to BRAF/MEK-targeted therapy in male and female patients (P = 0.021 and P < 0.0001). Together, these results have important implications for therapy. Treatment with neoadjuvant BRAF/MEK-targeted therapy results in higher rates of major pathological response in female compared with male patients with melanoma, and pharmacological inhibition of androgen receptor signalling improved the responses of male and female mice to BRAF/MEK-targeted therapy.

Treatment with therapy targeting BRAF and MEK (BRAF/MEK) has revolutionized care in melanoma and other cancers; however, therapeutic resistance is common and innovative treatment strategies are needed 1,2 . Here we studied a group of patients with melanoma who were treated with neoadjuvant BRAF/MEK-targeted therapy (NCT02231775, n = 51) and observed significantly higher rates of major pathological response (MPR; ≤10% viable tumour at resection) and improved recurrence-free survival (RFS) in female versus male patients (MPR, 66% versus 14%, P = 0.001; RFS, 64% versus 32% at 2 years, P = 0.021). The findings were validated in several additional cohorts 2-4 of patients with unresectable metastatic melanoma who were treated with BRAF-and/or MEK-targeted therapy (n = 664 patients in total), demonstrating improved progression-free survival and overall survival in female versus male patients in several of these studies. Studies in preclinical models demonstrated significantly impaired anti-tumour activity in male versus female mice after BRAF/MEK-targeted therapy (P = 0.006), with significantly higher expression of the androgen receptor in tumours of male and female BRAF/MEK-treated mice versus the control (P = 0.0006 and P = 0.0025). Pharmacological inhibition of androgen receptor signalling improved responses to BRAF/MEK-targeted therapy in male and female mice (P = 0.018 and P = 0.003), whereas induction of androgen receptor signalling (through testosterone administration) was associated with a significantly impaired response to BRAF/MEK-targeted therapy in male and female patients (P = 0.021 and P < 0.0001). Together, these results have important implications for therapy.
To date, multiple studies have demonstrated that the male sex is associated with worse outcomes in patients with melanoma, including patients with stage III and IV disease treated with BRAF-targeted therapy 1,2 . However, the underlying biology is incompletely understood. In classical hormonally responsive malignancies such as prostate cancer, reciprocal interactions between androgen receptor (AR) signalling and MAPK and other signalling pathways are evident 5 , although this has been less well studied in melanoma and other cancers. Variations in the immune background 6,7 , tumour microenvironment [8][9][10][11] and tumour cell susceptibility to targeted therapy 6 have also been posited as driving factors of this sexual dimorphism in treatment outcomes. Moreover, innate hormonal differences in the oestrogen and androgen pathways have been implicated as potential mechanisms driving observed disparities in melanoma preclinical models 9,10,12-14 after immune checkpoint blockade 15,16 and targeted therapy 17 . However, further insights from clinical and preclinical studies are needed to help to derive innovative strategies to improve patient survival.

Article
To help address this area of need, we investigated the effect of biological sex on the response to BRAF/MEK-targeted therapy in several independent cohorts of patients with metastatic melanoma. We first studied a cohort of patients with locoregional metastatic melanoma treated with 8-12 weeks of neoadjuvant BRAF/MEK-targeted therapy 18 (n = 51; Fig. 1a). In this study, we observed strong sexual dimorphism in response to treatment, with a significantly higher rate of major pathological response (MPR; defined as <10% viable tumour on pathologic evaluation of the surgically resected tumour at time of operative intervention) in female versus male patients (66% versus 14%, respectively, odds ratio (OR) = 12.0, 95% CI = 2.85-50.59, P = 0.001) (Fig. 1b). Importantly, relapse-free survival after neoadjuvant BRAF/MEK-targeted therapy was also noted to be significantly higher in female versus male patients (62% versus 34% at 2 years, P = 0.021) (Fig. 1c). No other clinical factors were significantly associated with achievement of MPR after logistic regression analysis, including age (P = 0.40), stage IIIC or D versus IIIA or B (P = 0.45), stage IV versus stage IIIA or B (P = 0.78), BRAF V600E versus V600 non-E mutation (P = 0.81), Eastern Cooperative Oncology Group (ECOG) performance status (P = 0.89), body mass index (>30) (P = 0.90), serum lactate dehydrogenase (P = 0.29) or recurrent versus de novo disease (P = 0.24) (Supplementary Table 1). There was no association between MPR and menopausal status in the female subset of the neoadjuvant cohort, in cases in which menopausal status was available (MPR; 9 out of 13 (69%) pre-menopausal women and 11 out of 17 (65%) post-menopausal women (P = 0.79)), suggesting that oestrogen did not have a major role in our observations (Extended Data Fig. 1a). Response Evaluation Criteria in Solid Tumors (RECIST) response rates in the neoadjuvant cohort were not significantly different between female and male patients (Extended Data Fig. 1b; P = 0.54), which is consistent with our previous observation of a poor correlation between RECIST response rates and pathological response in patients on neoadjuvant BRAF/MEK-targeted therapy 18,19 .
Intrigued by these findings, we sought to validate them in additional cohorts of patients. To do this, we next studied several cohorts of patients with unresectable metastatic disease treated with BRAF and/or MEK inhibitors until time of progression. This included a cohort of patients from our own institution who were treated with definitive BRAF/MEK-targeted therapy until time of progression (n = 69) ( Fig. 1d and Supplementary Table 2). Clinical outcomes were assessed by both RECIST 20 and progression-free survival (PFS). In this cohort, a significantly higher rate of clinical benefit (defined as a complete response, partial response or stable disease for 6 months or greater) was noted in female versus male patients on BRAF/MEK-targeted therapy (80% versus 68%; OR = 3.65, 95% CI = 1. 16-11.47, P = 0.022 in 69 evaluable patients) (Fig. 1e). Significantly improved PFS was also noted in female versus male patients (median 12 months versus 7 months, hazard ratio = 0.42, 95% CI = 0.23-0.75, P = 0.003 in 80 evaluable patients) (Fig. 1f). Moreover, female patients exhibited a significantly greater reduction in overall tumour burden compared with male patients in a combined analysis of both cohorts using cross-sectional imaging (44% versus 27%, P = 0.02, n = 116 evaluable patients; Extended Data Fig. 1c). We next validated these findings in several different cohorts of patients treated with BRAF and/or MEK inhibitor therapy for metastatic melanoma (total of n = 664 patients) [2][3][4] . Analysis of the COMBI-D study (NCT01584648) of patients treated with combined BRAF/MEK-targeted therapy (n = 211) demonstrated that female patients had improved PFS/overall survival compared with male patients at 2 years ( Fig. 1g-i; relative risk = 0.81, 95% CI = 0.67-0.98, P = 0.03, and relative risk = 0.73, 95% CI = 0.54-0.99, P = 0.04, respectively). Differences were not as substantial in the setting of treatment with BRAF inhibitor monotherapy (n = 212 patients), with no significant differences observed in PFS/overall survival between male and female patients, although a trend towards improved overall survival was noted in the female patients (Extended Data Fig. 2a-c). Patients with metastatic melanoma treated with single-agent MEK inhibition monotherapy were also studied (METRIC study, NCT01245062, including 206 patients with 112 male and 94 female patients), demonstrating significantly improved PFS and overall survival in female versus male patients (Extended Data Fig. 2d-f; P = 0.043 and P = 0.0021, respectively). Importantly, we also included an additional cohort of patients with locally advanced melanoma treated with neoadjuvant BRAF/MEK inhibition (NCT01972347) 4 . In contrast to our studies, analysis of this cohort did not reveal significant differences in MPR rates or RFS in female versus male patients. This highlights heterogeneity in these clinical observations, although this cohort was substantially smaller than our own cohort, and was not balanced with regard to the number of enrolled female and male patients (n = 35 patients total with the majority being male) (Extended Data Fig. 2g-i).
We next compared AR expression in matched pre-treatment and on-treatment time points in patients receiving neoadjuvant BRAF/ MEK-targeted therapy (Fig. 2a), and observed significantly higher expression of AR in male patients at on-treatment compared with pre-treatment time points ( Fig. 2b; P = 0.01). In female patients, a trend towards increased AR expression was observed from pre-treatment to on-treatment samples, but this did not reach statistical significance ( Fig. 2b; P = 0.21), probably owing to the lower baseline testosterone levels in female patients. Notably, circulating testosterone levels were not significantly affected by treatment with BRAF/MEK-targeted therapy when assessing matched pre-treatment and on-treatment blood samples of patients treated with neoadjuvant BRAF/MEK-targeted therapy (Extended Data Fig. 2k). We specifically examined AR expression levels in pre-treatment and on-treatment tumour samples of patients treated with neoadjuvant BRAF/MEK-targeted therapy, and observed significantly higher AR expression in on-treatment tumours of patients who did not achieve an MPR compared with those who achieved an MPR ( Fig. 2c (P = 0.006) and Supplementary Table 3), suggesting a possible association between AR expression after treatment and response. However, this analysis was confounded by the lower proportion of viable tumour cells in tumours that had a significant response in on-treatment samples (Extended Data Fig. 3 and Supplementary  Table 4). No significant differences in AR expression were observed in pre-treatment samples of patients who achieved an MPR versus those who did not (Extended Data Fig. 2l; P = 0.72), suggesting that baseline AR expression is not predictive of resistance but that resistance is associated with upregulation of AR on treatment with BRAF/ MEK-targeted therapy. AR-signalling genes were assessed 21 in bulk RNA-sequencing (RNA-seq) data from available tumour biopsy samples of treated patients, again demonstrating a significantly higher AR signature score in on-treatment samples of patients who did not achieve an MPR ( Fig. 2d; P = 0.011), with no differences observed in pre-treatment samples (Extended Data Fig. 3e; P = 0.95). Furthermore, no significant differences in AR-related genes were observed when considering on-treatment and surgical samples independently rather than in aggregate (Extended Data Fig. 3e; P = 0.35, P = 0.066 and P = 0.011), although sample size was substantially lower, limiting the power of such analyses.
We next aimed to validate these findings in more carefully controlled preclinical models, with the goal of identifying potential therapeutic strategies to enhance response and survival rates to BRAF/ MEK-targeted therapy. To do this, we studied the effect of biological sex on the response to BRAF/MEK-targeted therapy in three independent preclinical models of BRAF-mutant melanoma (Fig. 2a). We first assayed this in two immunocompetent models (C57BL/6 mice transplanted with two independently generated BRAF-mutant melanoma cell lines), demonstrating significantly impaired tumour control in male versus female mice treated with BRAF/MEK-targeted therapy (Fig. 2e,f (P = 0.031 and P = 0.0006, respectively) and Extended Data Fig. 4a,b (P = 0.039 and P = 0.45)). No difference in tumour outgrowth was noted in female versus male mice treated with vehicle alone (control) (Fig. 2e,f (P = 0.13 and P = 0.22) and Extended Data Fig. 4a,b (P = 0.49 and P = 0.47)) or in male mice treated with vehicle and endocrine modulation through androgen blockade or castration (Extended Data Fig. 4c; P = 0.31 and P = 0.96, respectively) suggesting that the mechanism behind these findings was related to treatment with BRAF/MEK-targeted therapy and not tumour onset in this particular model. We next assayed an immunocompromised model (CD-1 nude mice) and observed similar findings ( Fig. 2g; P = 0.004), suggesting that the dominant mechanism behind this observation is probably not immune mediated, although limitations exist with this particular model as B cells and macrophages are still present, with opportunities to validate this in an NOD scid gamma mouse model. An analysis of the three models confirmed that there was a larger tumour size in male mice across multiple studies ( Fig. 2h; P = 0.034; tumour volume curves for associated experiments are provided in Extended Data Fig. 5).
We next sought to examine the mechanism through which male sex could be contributing to impaired tumour control in the setting of treatment with BRAF/MEK-targeted therapy. Given the observation that AR staining was increased in on-treatment models ( Fig. 2i; P = 0.0006 and P = 0.0025), we hypothesized that increased AR signalling might be at play given our findings in patients and in preclinical models. This is supported by recent literature suggesting that androgen signalling has a role in the initiation and progression of melanoma 14,22,23 . To test whether AR signalling was driving resistance to BRAF/MEK-targeted therapy and whether this was tumour-cell intrinsic, we generated androgen receptor knock-out (AR-KO) melanoma cell lines using CRISPR technology. AR-KO cell lines were implanted into male and female mice and the mice were treated with BRAF/MEK-targeted therapy, demonstrating equivalent and effective tumour control in male and female mice compared with the vehicle control. There was no evidence of sexual dimorphism in the setting of treatment with BRAF/MEK-targeted therapy alone ( Fig. 2j; P = 0.76), or in the setting of modulation of androgen signalling through the administration of supplementary testosterone or AR blockade with enzalutamide (Extended Data Fig. 4d

Article
We next assessed AR staining and gene expression profiling in male and female mice treated with vehicle versus BRAF/MEK-targeted therapy, and in available pre-treatment and on-treatment tumour samples from patients treated with neoadjuvant BRAF/MEK-targeted therapy. In these studies, we observed significantly higher expression of AR in both male and female mice after treatment with BRAF/MEK-targeted therapy versus the vehicle control ( Fig. 2i (P = 0.0006 and P = 0.0025) and Extended Data Fig. 6a,b), suggesting that treatment with BRAF/ MEK-targeted therapy is associated with increased AR staining in both male and female mice. Notably, baseline and on-treatment levels of AR were significantly higher in male compared with female mice, probably related to higher levels of testosterone in male mice (Extended Data Fig. 6c,d), and might be attributable to the known role of testosterone in stabilizing AR protein as demonstrated in other malignancies 24 .

Fig. 2 | Treatment with BRAF/MEK inhibition is associated with increased
AR expression and phenotype is recapitulated in preclinical models. a, Schematic of the clinical analyses and mouse studies. b, AR staining in paired pre-and post-treatment male (blue) and female (pink) patient samples show increased AR expression in male samples (P = 0.01) but not in female patient samples (P = 0.21, two-sided t-test). c, AR staining post-treatment in male (n = 14, blue) and female (n = 9, pink) patients by MPR (P = 0.006, two-sided t-test). d, Androgen signalling score in patients achieving MPR (n = 11) versus <MPR (n = 11, P = 0.011, two-sided t-test). e, The percentage change in tumour volume in C57BL/6 mice that were implanted with YUMMER1.7 cells treated with vehicle or BRAF/MEK inhibition (BRAF/MEKi) (P = 0.031 male versus female BRAF/MEKi). n = 10 mice per group, aged 12-13 weeks. f, The percentage change in tumour volume in C57BL/6 mice implanted with Braf V600E Pten −/− mouse melanoma (BP) cells (P = 0.0006; male versus female BRAF/MEKi). Mice were treated as described in a. n = 10 mice per group, aged 12-13 weeks. The experiment was performed three times (Extended Data Fig. 3a,b). g, The percentage change in tumour volume in CD-1 nude mice that were implanted and treated as described in a (P = 0.004; male versus female BRAF/MEKi). n = 10 mice per group, aged 11 weeks. h, Aggregate end-point tumour volumes (n = 10 mice per study from five independent studies, P = 0.034, two-sided t-test). Individual points represent different studies. i, The percentage of cells with AR + nuclei in YUMMER1.7 tumours in C57BL/6 mice that were treated with vehicle (n = 7 male and n = 7 female) or BRAF/MEKi (n = 7 female (P = 0.0006), n = 5 male (P = 0.0025), two-sided t-test) for 3 days. j, The percentage change in AR-KO BP tumour growth in CD-1 mice that were treated with vehicle or BRAF/ MEKi (P = 0.76, male versus female BRAF/MEKi). n = 10 per group, aged 11 weeks. The tumour growth curves shown in e, f, g and j show mean ± s.e.m. and P values were computed using analysis of variance (ANOVA) with correction for multiple comparisons. The box plots in d and h show the median (centre line), interquartile range (IQR) (box limits) and the most extreme point within 1.5 × IQR (whiskers).
However, there was not a significant change in plasma testosterone when comparing the vehicle and BRAF/MEK-treated groups (Extended Data Fig. 6c; P > 0.99). Loss of AR staining was confirmed in AR-KO tumours (Extended Data Fig. 6e), which was accompanied by loss of sexual dimorphism in response to BRAF/MEK-targeted therapy. This suggests that differences in response to BRAF/MEK-targeted therapy

Article
are driven by tumour intrinsic AR activity. No differences in outcomes were noted in female mice treated with supplemental oestrogen and oophorectomy in combination with BRAF/MEK-targeted therapy compared with BRAF/MEK-targeted therapy alone (Extended Data Fig. 7a; P = 0. 43), suggesting that oestrogen signalling is not a dominant factor underlying the observed sexual dimorphism in this preclinical model, although additional study is needed. Together, these findings suggest that the sexually dimorphic phenotype is not primarily driven through the ovarian production of steroids.
On the basis of these findings, we next conducted studies to test the hypothesis that pharmacological manipulation of AR is associated with differential responses to treatment with BRAF/MEK-targeted therapy. To do this, male and female mice were implanted with melanoma tumours and were treated with either vehicle or BRAF/MEK-targeted therapy, along with pharmacological manipulation of AR activity (through systemic administration of testosterone or AR blockade) either alone or in combination ( Fig. 3a and Extended Data Fig. 6c,d). We observed significantly impaired tumour control in male and female mice treated with testosterone and BRAF/MEK-targeted therapy compared with BRAF/ MEK-targeted therapy alone (  Fig. 7d). This suggests a potential feedforward loop whereby BRAF/MEK inhibition results in an increase in AR and AR-regulated targets. AR expression promotes proliferation of melanoma cells, and this response is magnified in testosterone-rich environments 24,26,27 . Notably, treatment with enzalutamide was not associated with differences in the abundance of nuclear AR (Fig. 3h,i); however, induction of AR activity after exposure to exogenous testosterone together with BRAF/MEK-targeted therapy and enzalutamide increased AR expression compared with BRAF/MEK-targeted therapy and enzalutamide in male and female mice (Fig. 3h,i (P = 0.016 and P = 0.01) and Extended Data Fig. 7f,g). Consistent with these data, castration of tumour-bearing mice significantly improved their response to BRAF/MEK-targeted therapy, yet this benefit was lost after testosterone administration (Extended Data Figs. 7e and 8; P = 0.0004). Notably, resistance to BRAF/MEK therapy was independent of MAPK [28][29][30][31] (Extended Data Fig. 9), ZIP9/YAP1-MAPK 32,33 (Extended Data Fig. 9) or exogenous oestrogen treatment 34,35 (Extended Data Fig. 7a).
Taken together, these data have important clinical implications. First, these data provide evidence that treatment with BRAF/MEK-targeted therapy is associated with an increase in AR expression in tumour cells promoting therapeutic resistance, and that AR blockade by anti-androgens such as enzalutamide may promote response to BRAF/ MEK-targeted therapy in both male and female patients. Moreover, we provide evidence that treatment with testosterone may promote resistance to BRAF/MEK-targeted therapy in male and female patients, which has important therapeutic implications as testosterone is widely used for multiple indications and perhaps its use needs to be carefully considered in patients with melanoma. These data have potential relevance in cancer beyond melanoma, as numerous malignancies are now being treated with BRAF/MEK-targeted therapy and other strategies targeting MAPK signalling pathways, warranting the study of AR signalling in the setting of treatment with targeted therapy across cancer types and in the setting of other agents targeting MAPK and related pathways in cancer. Importantly, AR activity is also relevant in other forms of cancer treatment, as AR signalling has been shown to promote T cell exhaustion, and AR blockade may promote a response to anti-PD-1 immune checkpoint blockade 36 . The findings from these studies have a potentially immediate clinical impact, although nuances exist, as treatment with AR blockade in the absence of hypothalamic suppression can result in increased AR signalling and higher testosterone levels 37,38 , and resistance to enzalutamide can occur through a number of different mechanisms including AR splice variants 39,40 and glucocorticoid receptor expression 41 . Nonetheless, clinical trials investigating combination therapy strategies with AR blockade and BRAF/ MEK-targeted therapy or immune checkpoint blockade are warranted with some trials currently underway (NCT02885649, NCT04926181, NCT01974765, NCT03207529).
Importantly, further insights are needed to better delineate the mechanism behind this, with opportunities to develop strategies to modulate AR signalling in a more targeted manner. This has the potential of abrogating side effects that are commonly experienced with androgen-deprivation therapy. There is also a critical need to better understand the effect of other hormones (oestrogens, glucocorticoids) that are produced by the host or are modulated by gut microbiota in modulating responses to cancer treatment [42][43][44] . Although the effect demonstrated here occurs while on BRAF/MEK-targeted therapy and was not observed in mice that were treated with vehicle, sexual dimorphisms are noted across a variety of cancer therapies. Notably, these results overall have potential relevance beyond their effect on treatment with BRAF/MEK-targeted therapy or immune checkpoint blockade for patients with advanced cancer. AR signalling is implicated across numerous cancer types, and trials are underway incorporating AR blockade with conventional chemotherapy and other strategies in patients with advanced cancer (NCT02684227). Furthermore, AR signalling and targeting are also being studied in the context of carcinogenesis, with evidence that AR signalling may induce carcinogenesis and may promote increased invasiveness and metastasis [45][46][47][48][49] . Additional studies are needed to better understand the mechanism through which AR signalling promotes resistance to BRAF/MEK-targeted therapy and other cancer treatments, and should be examined in additional patient cohorts as well as in preclinical models, as important differences may exist and murine models are not a perfect surrogate to studying this in clinical cohorts. Further research is also needed to better understand the effect of androgens and other hormones on carcinogenesis, therapy response and other disease conditions. The examination of clinical cohorts and studies in preclinical and other models will help us to better understand the relative contribution of sex hormones and related factors on cancer and other disease states, with opportunities to modulate these over a lifetime to promote overall precision health.

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/s41586-022-04833-8.

Clinical cohorts
Patients enrolled in the initial clinical trial cohort were previously described 18 (NCT02231775). In brief, patients aged ≥18 years with histologically proven clinical stage III or oligometastatic stage IV BRAF V600E/K melanoma deemed to be resectable by multidisciplinary consensus and measurable disease by RECIST 1.1 criteria were enrolled and those randomized to the experimental arm received 8 weeks of neoadjuvant dabrafenib (150 mg orally twice daily) plus trametinib (2 mg orally daily) before surgical resection, followed by up to 44 weeks of adjuvant dabrafenib and trametinib (neo-BRAF/MEKi) (n = 12). Additional patients (n = 20) from treated with neoadjuvant BRAF/MEKi along with an additional retrospective cohort of patients treated off protocol for logistical reasons with neoadjuvant BRAF/MEKi (n = 16), dabrafenib (n = 3) and encorafenib plus binimetinib (n = 1) were included. Radiographic responses to neoadjuvant therapy were determined at week 8 before surgery and pathological responses were determined by microscopy examination of the complete surgical specimen by a melanoma pathologist, including SOX10 immunostaining when applicable to confirm the presence or absence of viable melanoma cells. These patients were treated at the University of Texas MD Anderson Cancer Center and had tumour samples collected and analysed under Institutional Review Board (IRB)-approved protocols. Notably, these studies were conducted in accordance with the Declaration of Helsinki and approved by the UT MD Anderson Cancer Center IRB.
The cohort of patients with metastatic disease treated at MD Anderson Cancer Center were collected and described in a retrospective analysis from a prospectively collected database 50 with appropriate IRB approval. These patients were treated according to multidisciplinary consensus standard of care. Their measurable disease by RECIST 1.1 was measured on available computed tomography (CT) scans and the outcomes of progressive disease were recorded. Of the 80 patients, 69 had measurable disease by RECIST 1.1 with 10 patients having had unmeasurable leptomeningeal or miliary lung recurrences, and another having his RECIST-measured lesion resected in line with standard of care.
Patients enrolled in the COMBI-D trial were previously described previously (NCT01584648) 2 . Patients with histologically proven unresectable clinical stage III or stage IV BRAF V600E/K melanoma who met eligibility criteria were enrolled. Eligibility criteria included age of greater than 18, measurable disease by RECIST 1.1 criteria, confirmed BRAF V600 E/K mutation, ECOG performance status of 0 to 1, and no history of previous systemic treatment for advanced or metastatic cancer. Patients were randomized (1:1) to receive either oral dabrafenib (150 mg twice daily) and oral trametinib (2 mg once daily) or oral dabrafenib (150 mg twice daily) and placebo. Disease baseline assessment was performed by CT scan or magnetic resonance imaging of the chest, abdomen and pelvis, and skin lesions were photographed. Follow-up was performed every 8 weeks for 52 weeks and every 12 weeks after until death, progression or withdrawal.
The patients enrolled in the METRIC trial 51 (NCT01245062) had histologically proven BRAF V600 E/K mutation, unresectable clinical stage III or IV cutaneous melanoma, and an ECOG performance status of 0 or 1. The patients were randomized to receive either oral trametinib (2 mg once daily) or intravenous chemotherapy consisting of either dacarbazine or paclitaxel every 3 weeks. For this study, data on only the patients who received trametinib were extrapolated and analysed for sex-specific differences in clinical outcomes.
Patients enrolled in the NeoCombi trial (NCT01972347) were previously described 4 . Adults (aged >18 years) with histologically proven BRAF V600 E/K mutation, resectable, clinical stage IIIB-C melanoma and an ECOG performance status of 0 or 1 were enrolled for the study. Similar to the previous studies, the patients required RECIST 1.1-measurable disease. Patients received oral dabrafenib (150 mg twice daily) and 2 mg of trametinib (once daily) for 12 weeks before surgical resection and 40 weeks after surgery. CT and positron emission tomography scans were performed before resection and pathological responses were measured.
Overall survival and recurrence-free survival within each clinical study were calculated using STATA (v.13.1) and R software (v.4.1.3). The study was conducted in accordance with the Declaration of Helsinki and approved by the medical ethics committee of MD Anderson Cancer Center. All of the participants provided informed consent before their participation in the study.

RNA-seq analysis
Tumour biopsies were obtained as feasible by punch or core biopsy before and during the neoadjuvant treatment period. Fresh-frozen tumour biopsy material was used for RNA-seq library preparation. Total RNA was extracted from snap-frozen tumour specimens using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen) after assessment of tumour content by a pathologist, and macrodissection of the tumour bed if required. RNA quality was assessed on an Agilent 2100 Bioanalyzer using the Agilent RNA 6000 Nano Chip with smear analysis to determine DV200 and original RNA concentration. On the basis of RNA quality, 40-80 ng of total RNA from each sample then underwent library preparation using the Illumina TruSeq RNA Access Library Prep kit according to the manufacturer's protocol. Barcoded libraries were pooled to produce final 10-12 plex pools before sequencing on an Illumina NextSeq 500 sequencer using one high-output run per pool of 76 bp paired-end reads, generating 8 fastq files (4 lanes, paired reads) per sample.

RNA-seq data processing
RNA-seq FASTQ files were first processed using FastQC (v.0.11.5) 52 , a quality control tool to evaluate the quality of sequencing reads at both the base and read levels. Reads with ≥15 contiguous low-quality bases (phred score < 20) were removed from the FASTQ files before STAR 2-pass alignment (v.2.5.3) 53 using the default parameters to generate one BAM file for each sequencing event. RNA-SeQC (v.1.1.8) 54 was next used to generate quality control metrics including read counts, coverage and correlation. A matrix of Spearman correlation coefficients among all of the sequenced samples was subsequently generated by RNA-SeQC and, after careful review, the sequencing data generated from one library pool that showed poor correlation with the other library pools from the same RNA sample were removed before sample-level merging of BAM files.
HTSeq-count (v.0.9.1) 55 was applied to aligned RNA-seq BAM files to count how many aligned reads overlapped with the exons of each gene. The raw read counts generated from HTSeq-count were normalized into fragments per kilobase of transcript per million mapped reads (FPKM) using the RNA-seq quantification approach suggested by the bioinformatics team of NCI Genomic Data Commons (GDC; https://docs.gdc. cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/). In brief, FPKM normalizes read count by dividing it by the gene length and the total number of reads mapped to protein-coding genes using the calculation described below: where RC g is the number of reads mapped to the gene, RC pc is the number of reads mapped to all protein-coding genes, L is the length of the gene in base pairs (calculated as the sum of all exons in a gene). The FPKM values were then log 2 -transformed for further downstream analyses.

Differential gene expression and pathway enrichment analysis
DESeq2 (v.3.6) was used to identify differentially expressed genes between patients who achieved an MPR (defined as Y) and those that did not (defined as N). The Wilcoxon rank-sum test was used to identify significantly different differentially expressed genes between the MPR Y and N groups. A cut-off false-discovery-rate-adjusted q value of <0.05 was applied to select the most significant DEGs. For pathway analysis, the curated AR gene sets 21 were downloaded. Gene set enrichment analysis was applied and pathway scores were calculated for each sample using the fgsea software package 56 . The pathway scores were then compared between the MPR Y and N groups. For preclinical analyses, there were 244 homologues within the 300 gene androgen signature used in analyses of clinical samples. Differential expression was studied between the high-testosterone group comprising the male vehicle, male BRAF/MEKi and female BRAF/MEKi with testosterone groups and the low-testosterone group comprising the female vehicle, female BRAF/ MEKi and male BRAF/MEKi with enzalutamide groups.
AR signature score Here, for sample s, AR s denotes its AR signature score; G is number of AR genes; and EXP g is the FPKM of AR gene g.
A Student's t-test was applied, showing a significant difference in AR signatures between sample groups MPR Y and N.

Animals and xenograft studies
Female or male C57BL/6 mice (0000664, purchased from Jackson Lab), aged between 9 to 14 weeks and weighing approximately 20 to 25 g were used for in vivo studies. Female or male CD-1 nude mice (086, purchased from Charles River Laboratories), aged between 10 to 11 weeks and weighing approximately 25-40 g were used for the immunodeficient model in vivo studies. The specific age of mice used in each experiment is indicated in the corresponding figure legends. Animal health was monitored daily by observation and sentinel blood sample analysis. Animal experiments were conducted in accordance with the Guideline of IACUC at MDACC.
Braf V600E Pten −/− mouse melanoma (BP) cell lines were previously generated by our group. YUMMER1.7 cells were a gift from M. Davis. Both tumour lines were periodically authenticated and tested for contamination by mycoplasma. BP cells were scaled up in DMEM culture medium supplemented with 10% FBS, collected and prepared such that each mouse received 0.8 × 10 6 cells in 0.2 ml PBS. For YUMMER1.7 cells, mice received between 0.5 × 10 6 -2 × 10 6 cells. Cells were implanted subcutaneously into the right flank of each mouse. For some male mice, physical castration was required and performed two weeks before treatment or before cell implantation. For some female mice, physical oophorectomy was required and performed one week before treatment or before cell implantation. Testosterone pellets (5 mg per day, Innovative Research of America) were implanted subcutaneously into the left flank 1 week before treatment with vehicle or BRAF/MEK-targeted therapy. Trametinib (Chemietek) at 1 mg kg −1 and dabrafenib (Chemietek) at 30 mg kg −1 were suspended at concentrations as needed in an aqueous vehicle containing 0.5% hydroxypropylmethylcellulose and 0.2% Tween-80 in distilled water and adjusted to pH 8.80 with diluted NaOH solution. Enzalutamide (Chemietek) at 10 mg kg −1 was formulated in 1% carboxymethyl cellulose (Sigma-Aldrich), 0.1% Tween-80 and 5% DMSO.
BP and YUMMER1.7 tumours were monitored using callipers before randomly sorting and dividing into experimental groups (n = 10 mice per group for efficacy, or n = 5-7 for the acute pharmacodynamic (PD) 3-day study in the case of YUMMER1.7). Tumour dimensions for the in vivo experiments were determined by a dedicated team of technicians. According to animal welfare guidelines, mice were euthanized if tumours grew larger than 4,000 mm 3 . Although these technicians were not blinded to the specific treatment groups, they did not have a working knowledge of the expected outcomes in these studies limiting, but not eliminating, potential bias. Treatment was started from day 14 to 17 after implantation depending on mouse strain and sex. Vehicle controls and compound treatments were given orally using a sterile 1 ml syringe and 18G needle for the times noted for each study. Dosing was 5 h apart between administration of trametinib + dabrafenib and enzalutamide for these specific treatment groups.
Tumour volume was calculated using the following formula: L × (W 2 )/2, where L is the length of the tumour and W is the width of the tumour. Tumour and plasma were collected 4 h after the last dose. Tumours were snap-frozen and the plasma was divided for monitoring drug concentrations and hormone levels.
Tumour volumes are represented as raw volumes, percent change in tumour volume, as well as aggregate tumour volumes from independent studies. Percent change in tumour volume was calculated from study day 0 (Tumour day 14 to 17).

Quant-seq library construction and sequencing
DNase-treated RNA samples (1,000 ng or 500 ng) were converted to cDNA using the QuantSeq 3′ mRNA kit according to the manufacturer's protocol (Lexogen). The libraries were amplified with 12 or 13 PCR Cycles and purified using the provided Lexogen. The purified libraries were quantified using a Kapa library quantification kit (KAPA Biosystems) and loaded onto the NextSeq 500 Sequencer (Illumina) at a final concentration of 2.6 pM to perform cluster generation, followed by 1 × 76 bp sequencing on the NextSeq 500 Sequencer (Illumina).

Generation of AR-KO CRISPR BP cell line
Three guide sequences were designed to target a 100 bp region of exon 1 using HorizonDiscovery's CRISPR Design Tool. Guide 1 (GACTTGGGT AGTCTACATGG AGG) was cloned into pLentiCRISPR.v2 according to addgene lentiCRISPRv2 and lentiGuide oligo cloning protocol for the purpose of pool selection. Guide 2 (GCTTGATACGGGCGTGTGGAT GGG) and guide 3 (CTGGAGAACCCATTGGACTA CGG) were ordered as crRNAs (IDT). The Neon electroporation system was used to transfect ribonucleoprotein (RNP) complex + plasmid into the BP cells. The crRNAs were rehydrated to 200 nM and pooled in equal volume (3 µl each) for annealing with Alt-R CRISPR-Cas9 tracrRNA-ATTO 550 (IDT; 1075928) at 95 °C for 5 min, then slowly cooled to 10 °C at 0.1 °C s −1 . crRNA:tracrRNA (5 µl) was combined with 5 µl of Alt-R S.p. HiFi Cas9 nuclease (IDT; 1081060) at room temperature for 10 min to generate the RNP complex. Electroporation Enhancer (2 µl) (IDT; 1075916) and 2 µg of plasmid were added to the RNP. During the annealing reaction, 1 × 10 6 BP cells were pelleted at 600g for 3 min. The total volume of RNP + plasmid + electroporation enhancer was transferred to an aspirated cell pellet. Next, 95 µl of R buffer was added to the cells and they were gently resuspended to single cells for an immediate 100 µl electroporation reaction. The Neon settings for the BP cells were two 30 ms pulses at 1,150 V. The cells were transferred to a single well of a six-well plate and allowed to recover in growth medium. After 24 h, the medium was replaced and 1.5 µg µl −1 of puromycin was added for a 48 h selection. The cells were allowed to recover from selection for 48 h before single-cell clone selection. After clones were selected and expanded, AR protein was analysed by western blot to confirm AR KO. In brief, cells were lysed for 30 min at 4 °C in 150 µl RIPA buffer with phosphatase and protease inhibitors. Protein (15 µg) was run on 10% SDS-Page (Bio-Rad) and transferred overnight onto PDF membranes using Tris glycine methanol buffer. After blocking for 1 h at room temperature in 5% milk diluted in TSBS-T, anti-AR antibodies (Abcam) were incubated at 1:2,000 dilution for 24 h. GAPDH was used as a loading control. The lentiviral expression vectors pLV-105 were purchased from Genecopeia to express either GFP as a transduction control. Lentivirus was generated using standard protocols and psPAX2 and pMD2.G as the packaging vectors. BP cells were transduced at 90% efficiency with viral supernatant and selected for 48 h with 2 µg ml −1 puromycin.

Generation of the AR-NT-KO BP control cell line
The non-targeting control sequences targeted luciferase and LacZ. The luciferase-specific target sequence (ACAACTTTACCGACCGCGCC) was cloned into pLentiCRISPR_v2. Furthermore, 2 crRNAs were designed and ordered: luciferase, ACAACTTTACCGACCGCGCC; and LacZ, CCCGAATCTCTATCGTGCGG. The transfection and selection of the sgRNA into the BP cells were the same as the KO cells. This quality metric is reported in Extended Data Fig. 10.
Immunofluorescence analysis FFPE blocks were sectioned (5 µm), mounted on charged microscope slides (Leica, 38002092) and dried at 37 °C overnight. Slides were then baked at 60 °C for 1 h in an oven (Biocare, DRY2008US), deparaffinized in three changes of xylene, and then rehydrated in three changes of 100% ethanol followed by a series of 95%, 70% and 50% ethanol and distilled water (5 min each). Antigen retrieval (10 mM sodium citrate, pH 6.0 with 0.05% Tween-20) was performed by heating slides to 95 °C for 15 min in a microwave (Biogenex EZ Retriever System v.3) followed by cooling down for 30 min at room temperature. Slides were then washed with TBST (Thermo Fisher Scientific, TA-999-TT). The area around each section was traced with a PAP pen (Sigma-Aldrich, Z672548), blocked with Background Sniper (Biocare, BS966) for 10 min and then washed with TBST.
Primary antibodies (rabbit anti-androgen receptor (EPR1535(2)), Abcam, ab133273, 1:300, diluted in Dako S3022) were incubated overnight at 4 °C, followed by a TBST wash. Secondary antibodies (goat anti-rabbit Alexa-647 conjugate, Thermo Fisher Scientific, A32733, 1:400, diluted in fluorescence antibody diluent (FAD), Biocare, FAD901L) were added and allowed to incubate for 1 h at room temperature, followed by a TBST wash. The sequence that the antibody was raised against was a human sequence and had been validated previously and used in numerous references [13][14][15][16] . This antibody was validated first in mice by testing both AR-expressing (such as testis, prostate) and non-expressing (such as surrounding tissue, fat) tissues and then in humans using KO cell lines (using western blot and immunofluorescence) as well as in AR-expressing and non-expressing human patient-derived xenograft (PDX) tumours. A sequential incubation with fluorophore-conjugated primary antibodies was performed (rabbit anti-sodium potassium ATPase (EP1845Y), Alexa-488 conjugate, Abcam, ab197713, 1:500, diluted in FAD) for 1 h at room temperature, followed by a TBST wash. Finally, slides were incubated with 4′,6-diamidino-2-phenylindole, dihydrochloride (5 mg ml −1 stock in DMF, Thermo Fisher Scientific, D1306) at 0.25 µg ml −1 (diluted in TBST) for 10 min at room temperature. The slides were then washed with a series of TBST, followed by TBS and then distilled H 2 O. Excess liquid was removed and the slides were mounted with ProLong Diamond Gold (Life Tech P36930) and allowed to harden overnight. For patient samples, a TSA-amplification of AR signal was performed.
In brief, after antigen retrieval/tissue tracing using a PAP pen, performed as described above, slides were incubated with treated with Bloxall to remove endogenous peroxidase (Vector Labs SP-6000) for 10 min at room temperature. The slides were then rinsed in water (30 s) and TBST (30 s). The slides were blocked with 2.5% normal horse serum (Vector Labs MP-7401 Component) for 25 min at room temperature and then block was tapped off. Slides were then blocked with Opal-specific PE/Diluent/Block (AKOYA ARD1001EA) for 10 min at room temperature and then block was tapped off. Primary antibodies (rabbit anti-androgen receptor (EPR1535(2)), Abcam, ab133273, 1:300, diluted in PE diluent, AKOYA, ARD1001EA) were incubated overnight at 4 °C, followed by a TBST wash for 3 min. Anti-rabbit secondary HRP polymer was then added to slides (Impress MP-7401) for 25 min at room temperature after which slides were washed in TBST. Opal signal was generated by adding Opal 570 fluorophore (AKOYA, FP1488001KT, diluted 1:200 in 1× Plus Amplification Diluent, AKOYA, FP1609) for 10 min at room temperature and then washed in TBST. All of the following steps, including staining with the membrane marker (Abcam, ab197713) and DAPI and coverslip mounting, were performed identically to the steps described above.

Immunofluorescence image acquisition and analysis
Slides were imaged on a Vectra 3 or a Vectra Polaris using the A UPlanSApo ×10/0.40 NA air objective first. Images were acquired using all available channels with the Vectra software (v.3.0.5) and the raw data were saved as .qptiff files. Regions of interest were created using PhenoChart (v.1.0.10) and these areas were then imaged again at ×20 magnification on the Vectra. ×20 images were then spectrally unmixed using inForm software and saved as Component TIFFs. The files were opened in QuPath software, channels were split and saved individually (or merged) as TIFFs. An APP was created in Visiopharm to segment cells and assess intensity of AR immunofluorescence. The percentage of positive cells was calculated for each sample. Raw analysed data were exported as a .csv file, and graphing and statistical analysis was performed using GraphPad PRISM 8 . A t-test test was performed to test significance.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
The additional datasets generated during and/or analysed during the current study of clinical trial NCT02231775 are available at the European Genome-Phenome Archive (EGAS00001006196). Other datasets generated during and/or analysed during the current study are available from the corresponding authors on reasonable request.     WWC1  NR2F6  DHCR24  GDF15  SOX4  HIST1H3E  USP11  ENDOD1  C1orf116  KCNN2  GOT1  HOMER2  TNFRSF10B  SMPD4  ENC1  SORD  DTL  CDC25A  MCM3  RFC3  MCM4  C5orf30  FDFT1  CENPN  SHROOM2  ACSL3  MCCC2  NEDD4L  CHKA  SLC39A6  SEMA6A  NCAPD3  FEN1  HMGCR  MPHOSPH9  IDI1   Gender Fig. 4 | Murine model of melanoma validates a sexually dimorphic response and suggests AR activity as a mechanism of resistance. A-B) Percent change in tumour volume for male and female C57BL/6 mice implanted subcutaneously with BP cells that were treated with Vehicle or BRAF/MEKi (n = 10 mice per group; A -mice aged 9 weeks, B -mice aged 12 weeks). Results from the second and third repeats of this experiment are shown in A and B, respectively (p = 0.039 and p = 0.45). C) Percent change in tumour volume in BP injected C57BL/6 male mice treated with vehicle in the presence or absence of BRAF/MEKi with endocrine modulation through androgen blockade with enzalutamide or castration (mice aged 14 weeks).

Extended Data
All tumour growth curves were compared by ANOVA with multiple comparisons (n = 10/group except BRAF/MEKi + castration where n = 9; p = 0.003 BRAF/MEKi vs BRAF/MEKi + Enzalutamide; p = 0.031 BRAF/MEKi vs BRAF/MEKi + Castration). D-E) Percent change in tumour volume for AR-KO BP tumours in female (D) (p = 0.99) and male (E) (p = 0.98) CD-1 mice treated with vehicle or BRAF/MEKi in combination with either testosterone or enzalutamide, respectively (n = 10 mice/group; aged 11 weeks). All tumour growth represented as mean + SEM and p-values were calculated using ANOVA with multiple comparisons.                    Fig. 9 | Effect of BRAF/MEKi on pERK, ZIP9/YAP1 associated transcripts and YAP1 associated transcripts in BP tumours. A) Staining and quantification of phosphor-ERK in BP tumours of female and male mice on treated with vehicles, BRAF/MEK inhibition, or BRAF/MEK inhibition + testosterone. Histogram represent mean + SD. Differences were calculated using one-way ANOVA (n = 5/group). B) Staining and quantification of ZIP9/ YAP1 associated transcripts in BP tumours of female and male mice on treatment with vehicles (n = 6), BRAF/MEK inhibition (n = 4), or BRAF/MEK inhibition + testosterone (n = 8). Histogram represent mean + SD. Differences were calculated using one-way ANOVA. C) Staining and quantification of YAP1 associated transcripts in BP tumours of female and male mice on treatment with vehicles (n = 6 females, 4 males) or BRAF/MEK inhibition (n = 4 females, 3 males. Histogram represent mean + SD. Differences were calculated using one-way ANOVA.