Prevalence and spectrum of AR ligand-binding domain mutations detected in ctDNA across disease states in men with metastatic castration-resistant prostate cancer

Abstract

Purpose:

Metastatic castration-resistant prostate cancer (mCRPC) is typically treated with agents directly or indirectly targeting the androgen receptor (AR) pathway. However, such treatment is limited by resistance mechanisms, including the development of activating mutations in the AR ligand-binding domain (AR-LBD).

Methods:

This study evaluated a database of over 15,000 advanced prostate cancer patients undergoing comprehensive ctDNA analysis (Guardant360, Redwood City, CA) between 2014 and 2021, with associated clinical information from administrative claims (GuardantINFORM database).

Results:

Of 15,705 prostate cancer patients included, 54% had mCRPC at the time of their blood draw. Of those, 49% had prior treatment with an AR pathway inhibitor (ARPi). AR-LBD mutation prevalence was 15% in mCRPC patients who were untreated with a next-generation ARPi, 22% in those after 1 line of ARPi therapy, and 24% in those after 2 lines of ARPi treatment. Next-generation ARPi treatment yielded an increase in AR L702H and T878A/S mutations post-abiraterone, and an increase in AR L702H and F877L mutations post-enzalutamide. AR-LBD+ patients demonstrated unique biology, including increased co-mutations in the cell-cycle, WNT, HRR, and PI3K pathways (all p <0.0005), and greater low-level (copy number <10) AR amplifications (p = 0.0041). AR-LBD+ patients exhibited worse overall survival relative to a matched cohort of AR-LBD– patients (50.1 vs 60.7 months, unadjusted log-rank p=0.0133).

Conclusions:

This large database analysis demonstrates that AR-LBD mutation prevalence increases following next-generation ARPi use. AR-LBD+ tumors demonstrate unique biology (more oncogenic pathways and low-level AR amplification) and reduced overall survival. These findings inform the development of novel therapies designed to circumvent AR-mediated therapeutic resistance.

INTRODUCTION

The primary therapeutic target for prostate cancer (PC) patients is the androgen receptor (AR). Androgen deprivation therapy via surgical castration has been largely supplanted by medical castration in the United States¹. While first-generation androgen deprivation therapies (ADT) established AR blockade as a therapeutic strategy by achieving castration levels of testosterone (i.e., < 50 ng/dL), they do not completely block the AR signaling pathway². Most recurrent or metastatic PC patients initially respond to androgen deprivation therapy³; however, responses are variable and progression to metastatic castration-resistant prostate cancer (mCRPC) after first-line androgen deprivation therapy usually occurs after 1–3 years^3,4.

Next-generation antiandrogen therapies have increased efficacy and potency and, when combined with standard ADT, represent the standard-of-care for CRPC patients. Several trials have demonstrated suppression of androgen signaling extends survival of mCRPC patients⁵; survival gains are even greater when these agents are used as initial therapy for metastatic hormone-sensitive prostate cancer (mHSPC)⁶. The four next-generation antiandrogens currently approved by the U.S. FDA are abiraterone acetate, enzalutamide, apalutamide, and darolutamide², the latter two being approved only in the mHSPC state and the non-metastatic CRPC state. In real-world clinical practice, these novel hormonal agents are often used as sequential lines of therapy, despite a lack of prospective phase III data supporting this approach—a practice that may lead to cross-resistance and limited clinical benefit or duration of response⁷. NCCN, ESMO, and EAU-EANM-ESTRO-ESUR-ISUP-SIOG guidelines for PC do not specify a preferred sequence or combinations of these agents, leading to uncertainty about optimal sequencing of these therapies^{7,–11 11}. The need for clinical data and expertise in developing new treatment guidelines for the optimal use of next-generation antiandrogens in mCRPC patients is essential¹².

The use of drugs targeting the AR pathway is hampered by the development of resistance, including activating AR ligand-binding domain (AR-LBD) point mutations¹³. Treatment-induced mutations in the AR-LBD are known to convert AR antagonists into agonists¹⁴. For example, the F877L mutation converts enzalutamide, an AR inhibitor, into an agonist, especially when the T878A mutation is present¹⁴. Additionally, some AR-LBD mutations might be activated by non-androgenic steroid hormones and may be found in 10–20% of mCRPC patients progressing on AR pathway inhibitors (ARPi)^15,16.

Analysis of a real-world clinical-genomic database is needed to enhance the characterization of PC-relevant mutations and promote the use of precision oncology in the management of men with advanced disease¹⁷. A robust liquid-biopsy database may complement tissue biopsies to fulfill this need. In a recent genomic analysis of plasma-derived circulating-tumor DNA (ctDNA) from 3,334 mCRPC patients, investigators found genomic profiling of ctDNA had a high level of concordance to tissue biopsies in detecting mutations⁴. In analyses of AR mutations in ctDNA from advanced PC patients, researchers have identified multiple AR mutations with direct pharmacologic relevance to the development of resistance and selection of patients for future clinical trials¹⁸.

In the current study, we used the GuardantINFORM database¹⁹, a platform featuring an extensive liquid-biopsy dataset of advanced cancer patients linked with claims-based clinical outcomes. This real-world clinical-genomic database includes 200,000+ patients (spanning >50 cancer types) with integrated clinical and molecular information, including 15,705 PC patients. Our study’s primary objective was to characterize the prevalence, molecular landscape, and outcomes of mCRPC patients harboring AR-LBD mutations detected by next-generation ctDNA sequencing. The clinical features of patients with AR-LBD mutation-positive mCRPC were further compared to a matched group lacking AR-LBD mutations.

METHODS

Clinical Outcomes

After identifying the AR-LBD mutation-positive cohort, patients were matched in a 1:3 ratio with patients without AR-LBD mutations. Matching was based on age (±5 years), baseline Elixhauser Comorbidity Index (ECI) (±1 SD), smoking status, prior ARPi use, and year of Guardant360 testing using the coarsened exact matching (CEM) method²⁰. We also conducted an exploratory survival analysis after stratifying the AR-LBD–positive group according to several distinct mutations in an unmatched cohort.

Statistical Analyses

Chi-squared/Fishers exact test was used for comparison of categorial variables. Student’s t-test was used for continuous variables. False Discovery Rate (FDR) and its analog, the q-value, were utilized for co-mutation analysis among AR-LBD+ patients compared with AR-LBD– patients.

Among patients with >1 Guardant360 test, gains and losses of AR-LBD mutations were analyzed using the paired McNemar test. For outcomes analysis, overall survival of AR-LBD–positive mCRPC patients was compared with AR-LBD–negative patients using the Kaplan-Meier method; log-rank tests were used to evaluate differences across groups. Unadjusted and adjusted Cox proportional hazards models were performed to obtain pairwise hazard ratios (HRs) and 95% confidence intervals (95% CIs). Age at index, ECI, next-generation ARPi use, ctDNA level, and site of metastasis variables were used for the adjusted cox model. P-values of <0.05 were considered statistically significant. All analyses were conducted using SAS software package 9.4 (SAS Institute, Cary, NC).

RESULTS

Patients and samples

A total of 15,705 PC patients were available in the GuardantINFORM database (Figure 1). Of these, 8,420 (54%) patients were identified to have mCRPC at the time of sample collection. Within the mCRPC cohort, 1,700 (20.2%) patients had AR-LBD mutations; 6,720 (79.8%) had no AR-LBD mutations. Among the mCRPC patients included, 4,122 (49%) had prior treatment with a next-generation ARPi, of which 854 (21%) had ≥1 activating AR-LBD mutation; 3,268 (79%) had no AR-LBD mutations.

Figure 1.

Consort Diagram

Landscape of AR-LBD mutations in mCRPC patients

The prevalence and spectrum of AR-LBD mutations is depicted in Supplemental Table 1. Among patients with AR-LBD mutations, 1,122 (66%) had one AR-LBD mutation and 578 (34%) had ≥1 concurrent AR-LBD mutation. AR L702H (59%), T878A (33%), and H875Y (23%) were the three most common LBD mutations identified, comprising three-quarters of all AR-LBD mutations (Supplemental Table 1).

AR-LBD mutation prevalence was next assessed at distinct time points relative to first-generation and next-generation ARPi treatment (Table 1). For this analysis, there were 387 mCRPC patients for whom a Guardant360 test was done prior to any ARPi treatment (9.3%). There were 747 mCRPC patients for whom a Guardant360 test was done after receiving only a first-generation ARPi agent (flutamide, nilutamide, bicalutamide) (18.1%). AR-LBD mutations were detected at a 15% prevalence prior to any next-generation ARPi use; this prevalence remained at 15% in patients who had previously received only a first-generation ARPi. For patients treated with a prior next-generation ARPi, the prevalence of AR-LBD mutations was 22%. The AR-LBD mutation rate increased marginally, to 24%, following receipt of two next-generation ARPi agents. These results suggest AR-LBD mutation rates increase as a result of next-generation (but not first-generation) ARPi treatment, and are not significantly further enriched following two next-generation ARPi drugs.

Table 1.

Detection Rates of AR-LBD mutations relative to prior ARPi use.

Patient Population	Detection Rate	Prevalence
AR-LBD mutations prior to any next-generation ARPi	59/387	15%
Emergence of AR-LBD mutations as acquired resistance mechanisms to *first-generation ARPi	115/747	15%
AR-LBD mutations post only one next-generation ARPi, any line	353/1628	22%
AR-LBD mutations post next-generation ARPi in two lines	509/2118	24%

^*First generation ARPi agents include flutamide, bicalutamide or nilutamide.

Co-alteration analysis

To assess the biology of AR-LBD–mutated samples, co-occurring mutations were compared between AR-LBD+ vs. AR-LBD– cohorts (Figure 2). In addition to individual genes (e.g. TP53 and RB1), three pathways were examined: WNT (APC and CTNNB1), tHRR (BRCA1/2, ATM and CDK12), and PI3K (PTEN, PIK3CA and AKT1). We found AR-LBD+ mCRPC patients were significantly enriched for HRR pathway mutations (35% vs. 30%, q=0.0005), PI3K pathway mutations (36% vs. 27%, q <0.0001), WNT pathway mutations (28% vs. 18%, q <0.0001), and RB1 mutations (7% vs. 5%, q=0.0008). AR-LBD– mCRPC patients were marginally enriched in TP53 mutations (49% vs 48% q=0.01), although this difference was not clinically significant. Additional significant differences by gene are shown in Supplemental Table 2. Of these upregulated pathways, HRR and PI3K include alterations associated with targeted therapies. Thus, we analyzed survival outcomes for patients with and without these pathway mutations. The log-rank p-value of AR-LBD mutations +/− HRR mutations was not significant (p=0.98), while the p-value of AR-LBD mutations +/− PI3K mutations was significant at (p=0.02), indicating inferior survival for patients with concurrent AR and PI3K pathway mutations (Supplemental Figure 1).

Figure 2.

Detection of Co-occurring Mutations among mCRPC Patients with (N=1700) vs. without (N=6720) AR-LBD Mutations.

AR amplification was also compared between AR-LBD+ and AR-LBD– cohorts (Table 2) to explore potential overlap versus mutual exclusivity of these two forms of androgen receptor activation. AR amplifications were present in 510 (30%) of the AR-LBD+ cohort and 1,828 (27%) of the AR-LBD– cohort (p=0.02, q=0.06). There was a statistically significant increase in AR copy number levels <10 for the AR-LBD+ (p=0.004; q=0.02); however, there were not significant differences between the cohorts for AR copy numbers ≥10.

Table 2.

Detection rates of AR amplifications among AR-LBD+ (n=1700) and AR-LBD− (n=6720) cohorts.

AR Amplification category	AR-LBD+	AR-LBD−	P-value	Q-value
All detected amplifications	510 (30%)	1828 (27%)	0.0230	0.0575
Amplifications w/ CN<10	457 (27%)	1547 (23%)	0.0041	0.0205
Amplifications w/ CN 10–15	31 (2%)	156 (2%)	0.0791	0.1318
Amplifications w/ CN 15–20	8 (0.5%)	48 (0.7%)	0.1921	0.1921
Amplifications w/ CN ≥20	14 (0.8%)	75 (1%)	0.1901	0.1921

Paired Analyses for AR-LBD Mutations

We analyzed patients with paired Guardant360 samples. There were 207 PC patients who received a Guardant360 test both before and after treatment with a next-generation ARPi. Gains and losses of AR-LBD activating mutations were evaluated after treatment with next-generation ARPi agents (Table 3). Among the AR-LBD mutations assessed, none demonstrated a significant decrease following next-generation ARPi treatment. Conversely, AR L702H and T878A/S mutations significantly increased following abiraterone treatment (p=0.0013 and 0.0209, respectively), whereas AR L702H and F877L significantly increased following enzalutamide treatment (P=0.0114 and 0.0253, respectively). Gains and losses of AR-LBD mutations were also assessed following taxane chemotherapy (for mCRPC) in patients with Guardant360 samples collected before and after chemotherapy treatment. There were no statistical gains or losses of AR-LBD mutations following chemotherapy use.

Table 3.

Gains and losses of AR-LBD mutations in paired samples. Frequencies analyzed after next-generation ARPi use [abiraterone (N=180) or enzalutamide (N=131)], or chemotherapy use (N=57), among those with Guardant360 samples collected both before and after such treatment. Mutations highlighted were significantly enriched post-treatment relative to pre-treatment.

Alteration	Pre-therapy (%)	Gained (%)	Lost (%)	Post-therapy (%)	Aggregate Change	P-value
Post-Abiraterone
L702H	3.33	7.78	−0.56	10.56	+7.23	0.0013
V716M	1.11	0.56	−0.56	1.11	0.00	0.9999
W742C/L	3.33	0.56	−2.22	1.67	−1.66	0.1797
H875Y	3.33	2.22	−1.11	4.44	+1.11	0.4142
F877L	1.11	0.56	−1.11	0.56	−0.55	0.5637
T878A/S	1.67	5.00	−1.11	5.56	+3.89	0.0209
Post-Enzalutamide
L702H	5.34	6.87	−0.76	11.45	+6.11	0.0114
V716M	1.53	0.76	−0.76	1.53	0.00	0.3173
W742C/L	2.29	1.52	−1.52	2.29	0.00	0.9999
H875Y	4.58	1.53	−1.53	4.58	0.00	0.9999
F877L	0.00	3.82	0.00	3.82	+3.82	0.0253
T878A/S	7.63	5.34	−3.05	9.92	+2.29	0.3657
M896T/V	0.00	0.76	0.00	0.76	+0.76	0.3173
Post-Chemotherapy
L702H	7.02	8.77	−1.75	14.04	+7.02	0.1025
V716M	0.00	1.75	0.00	1.75	+1.75	0.3173
H875Y	1.75	1.75	0.00	3.51	+1.76	0.3173
F877L	3.51	0.00	−1.75	1.75	−1.76	0.3173
T878A/S	7.01	1.75	0.00	8.76	+1.75	0.3173

AR-LBD+ Patients Demonstrate Reduced Survival versus Matched AR-LBD– Controls

To assess real-world survival outcomes in patients with and without AR-LBD mutations, a matched cohort (1:3) was created, as described in the Methods section and Figure 1. Demographics of the matched cohort demonstrated no significant differences in terms of age, ECI, smoking status, prior ARPi use, or index year (Supplemental Table 3). In this matched cohort, the AR-LBD+ cohort had an increased mean ctDNA burden compared to AR-LBD– (variant allele frequency of 18.94% vs. 10.04%). The AR-LBD– cohort demonstrated increased frequencies of post-1L therapies and a significantly shorter time to ctDNA testing from PC diagnosis. The treatment patterns in the 1L and 2L setting were also similar among the matched cohort (Supplemental Table 4).

Overall survival (OS) was assessed between matched AR-LBD+ versus AR-LBD– cohorts (Figure 3A). AR-LBD+ patients exhibited shorter median OS (relative to their first therapy for mCRPC) compared to AR-LBD– patients (50.1 vs 60.7 months, HR = 1.184, 95% CI 1.035–1.353; log-rank p=0.013) (Figure 3A). Similar rates of patients in this analysis received a 2L therapy (43% (299/695) AR-LBD+, 41% (855/2085) AR-LBD–). Upon adjusting by Cox regression for age at index, ECI, next-generation ARPi use, ctDNA level and site of metastasis (bone, brain, liver and lung), the trend was no longer statistically significant (HR=1.02, 95% CI 0.888–1.172; p=0.776).

Figure 3.

A. Survival outcomes by AR-LBD status following first-line mCRPC treatment. Depicted are Kaplan-Meier plots for overall survival (OS) stratified by AR-LBD+ (red) vs. AR-LBD– (blue) status among matched patient cohorts. B. Survival outcomes by AR-LBD mutation following first-line mCRPC treatment. Depicted are Kaplan-Meier plots for overall survival (OS) stratified by the 3 most common AR-LBD mutations vs. control in unmatched patient cohorts.

Next, the impact of individual AR-LBD mutations on OS were assessed. This analysis used an unmatched cohort and excluded F877L, T878S, V716M, and W742L/C mutations due to small samples size. Figure 3B depicts the survival outcomes of patients with H875Y, L702H, or T878A mutations compared to the control arm. The median OS for the AR-LBD mutation-negative (control) arm was 59.7 months, for AR H875Y was 47.2 months, for AR L702H was 46.2 months, and for AR T878A was 46.8 months. Thus, the AR H875Y, L702H, and T878A mutations all appear to be associated with worse survival as compared to the control group.

DISCUSSION

Targeting the AR signaling axis is a cornerstone of therapy in recurrent and advanced prostate cancer. However, many patients progress following ARPi treatment due to emergent molecular changes that give rise to therapeutic resistance, one of which involves activating AR-LBD mutations. This study using the GuardantINFORM real-world database evaluated a large cohort of mCRPC patients to assess AR-LBD mutation prevalence, AR-LBD mutation dynamics following treatment with ARPi and chemotherapy, and overall survival outcomes.

Prevalence of AR-LBD mutations in this cohort of mCRPC patients was 21% overall. The most common AR-LBD mutations were L702H, T878A, and H875Y; these mutations were also identified as the most prevalent AR mutations in CRPC via tissue based NGS and in a previous study of AR mutations also using the Guardant360 database¹⁶. Of note, the previous AR study using the Guardant360 database focused on all advanced PC (i.e., not specific to mCRPC) and demonstrated increased frequencies of AR amplifications and decreased AR-LBD mutations relative to this study. This could support unique biological differences between advanced PC and mCRPC; however, these studies had multiple differences in sample size, dates of sampling, and overall research hypotheses, making direct comparisons challenging.

In the present study, AR-LBD mutation prevalence increased following next-generation ARPi therapy but not following first-generation antiandrogen use. There were distinct mutations gained following abiraterone versus enzalutamide in the paired-sample analysis. Abiraterone treatment was associated with a significant gain of L702H and T878A/S; enzalutamide treatment was associated with a significant gain of L702H and F877L (Table 3). These emergent mutations are similar to those previously described in the literature²¹.

The co-occurring mutational landscape analysis identified unique biology distinguishing AR-LBD+ from AR-LBD– cohorts. AR-LBD+ patients demonstrated increased ctDNA burden and higher frequencies of co-occurring HRR, PI3K, and WNT pathway mutations (Figure 2). Some prior literature supports additive signaling between the dysregulated HRR and PI3K pathways and upregulated AR signaling²². Recognition of these co-altered pathways is also relevant to developing therapy combinations in mCRPC^22,23. For example, there are now three FDA-approved combinations of a PARPi with a next-generation ARPi in mCRPC patients with HRR mutations^24,25. Preclinical biology and clinical trial data seem to support at least additive effects of PARPi/ARPi combinations. Our novel finding of an increased prevalence of HRR mutation in mCRPC patients with AR-LBD mutations raises the question of whether such patients may be more responsive to combination PARPi/ARPi therapy; this hypothesis remains speculative. We also discovered interesting biology with respect to the frequency of co-occurring AR amplifications. AR-LBD+ patients had an increased frequency of AR amplifications versus AR-LBD– patients, particularly in those with copy numbers <10 (low level AR amplifications). This may suggest greater pressure on the AR in such patients, resulting in multiple AR-related aberrations that converge upon activated AR signaling. However, a positive association was not observed between AR-LBD mutations and high-level AR amplifications, possibly suggesting that the co-occurrence of AR activating mutations and amplifications does not provide the cancer cell with an additional survival advantage.

Our study also evaluated dynamics of AR-LBD mutation prevalence relative to commonly used systemic therapies for mCRPC using a large cohort. Molecular genomics are not required prior to selecting ARPi treatment. However, some patients have ARPi-resistant phenotypes at baseline and may not respond effectively to subsequent ARPi therapy. In our cohort, 15% of patients had AR-LBD mutations prior to next-generation ARPi use. The prevalence for patients treated only with a first-generation ARPi was also 15%, implying that weaker antiandrogens do not lead to significant enrichment of AR-LBD mutations. Within the cohort of patients treated with a next-generation ARPi, the prevalence of AR-LBD mutations after two lines of second-generation ARPi at 24% was not substantially greater than the prevalence after one line of ARPi (22%). This finding suggests the emergence of AR-LBD mutations occurs most frequently after the first line of next-generation ARPi therapy without further increases in mutation prevalence following additional treatment lines. This could inform clinical trial design of new AR-LBD–targeting agents; specifically, in the sequencing of novel therapies within a treatment paradigm.

Robust identification of AR-LBD mutation-positive mCRPC patients may also be relevant as a drug-development strategy for novel hormonal therapeutics targeting the androgen receptor via alternative mechanisms. For example, MK-5684 (formerly ODM-208) is a novel agent blocking CYP11A1, the most proximal steroid synthesis enzyme that converts cholesterol to all downstream adrenal hormones²⁶. Preliminary clinical data suggest MK-5684 may possess greater clinical activity in mCRPC patients with AR-LBD mutations compared to those without. A different investigational agent, bavdegalutamide (formerly ARV-110), acts as a proteolytic-targeting chimera that enhances proteasomal degradation of the AR protein²⁷. That agent has also demonstrated preliminary evidence of enhanced activity amongst mCRPC patients with certain AR-LBD mutations. Thus, our data will help estimate the prevalence and spectrum of AR-LBD mutations across a range of clinically-relevant mCRPC disease states, which may prove useful when designing future clinical trials using the above agents or other drugs that may preferentially target AR-LBD–mutant advanced PC.

There were several limitations to this study, many of which are inherent to real-world evidence studies. Mainly, the data lacks granular clinical information, including performance status of patients, physician-assessed tumor responses or PSA responses, and results of concurrent biomarker testing (including tissue-based NGS), which may inform clinical decision-making and be found in studies using detailed electronic health records. The data also lacks information on sampling criteria (i.e., factors prompting physicians to order ctDNA testing for mCRPC patients during routine clinical practice). We posit patients were tested primarily at baseline (when about to begin a new systemic therapy) and possibly following progression on therapy. However, there is a risk that the data may not be an accurate cross-section of the overall mCRPC population. For example, patients with very low disease burden or minimal PSA levels may not have undergone ctDNA testing due to lower tumor-DNA yields in such scenarios; this type of bias could have overestimated the prevalence of AR-LBD mutations if those selected for ctDNA testing had higher-burden disease. Another limitation is that the samples in this study were taken from a panel-based ctDNA assay, which is validated for a variety of advanced stage solid tumors yet does not include all relevant PC genes. For example, the SPOP gene, which is recurrently mutated in PC, is not included on the panel; thus, correlations between AR-LBD mutation status and SPOP mutations could not be explored. Finally, the Guardant360 assay is not optimal for assessing complex structural rearrangements of the AR gene, nor is it able to identify amplifications of the AR enhancer genomic regions, limiting exploration of further mutations associated with PC outcomes.

CONCLUSIONS

About one-fifth (21%) of mCRPC patients undergoing Guardant360 ctDNA testing harbor activating AR-LBD mutations. AR-LBD mutation rates are lowest for mCRPC patients untreated with next-generation ARPi agents (15%), and increase (up to 24%) following exposure to these agents. Patients with AR-LBD mutations are more likely to harbor other concurrent oncogenic mutations (in the HRR, WNT, and PI3K pathways) and greater low-level AR amplifications. The presence of AR-LBD mutations is associated with worse overall survival in unadjusted analysis, suggesting the need for novel therapeutics for these patients. These results may help design future clinical trials targeting AR-LBD mutation-positive patients with mCRPC.

CONTEXT SUMMARY.

Key Objective

We interrogated the prevalence and spectrum of activating AR gene mutations from ctDNA in metastatic castration-resistant prostate cancer patients, according to the number and type of prior systemic therapies received.

Knowledge Generated

We found that activating AR mutations become more prevalent with increasing lines of prior AR-pathway inhibitor treatment, with a 24% prevalence in patients with 2 prior lines of such treatments. These AR mutations confer a reduced overall survival and are also associated with mutations in other oncogenic pathways including the homologous recombination repair pathway.

Relevance

These findings inform the development of novel therapeutics designed to circumvent AR-mediated resistance or directly targeting activating AR mutations.