A Phase II Clinical Trial of Pembrolizumab and Enobosarm in Patients with Androgen Receptor‐Positive Metastatic Triple‐Negative Breast Cancer

Abstract

Lessons Learned

• The combination of enobosarm and pembrolizumab was well tolerated and showed a modest clinical benefit rate of 25% at 16 weeks.

• Future trials investigating androgen receptor‐targeted therapy in combination with immune checkpoint inhibitors are warranted.

Background

Luminal androgen receptor is a distinct molecular subtype of triple‐negative breast cancer (TNBC) defined by overexpression of androgen receptor (AR). AR‐targeted therapy has shown modest activity in AR‐positive (AR+) TNBC. Enobosarm (GTx‐024) is a nonsteroidal selective androgen receptor modulator (SARM) that demonstrates preclinical and clinical activity in AR+ breast cancer. The current study was designed to explore the safety and efficacy of the combination of enobosarm and pembrolizumab in patients with AR+ metastatic TNBC (mTNBC).

Methods

This study was an open‐label phase II study for AR+ (≥10%, 1+ by immunohistochemistry [IHC]) mTNBC. Eligible patients received pembrolizumab 200 mg intravenous (IV) every 3 weeks and enobosarm 18 mg oral daily. The primary objective was to evaluate the safety of enobosarm plus pembrolizumab and determine the response rate. Peripheral blood, tumor biopsies, and stool samples were collected for correlative analysis.

Results

The trial was stopped early because of the withdrawal of GTx‐024 drug supply. Eighteen patients were enrolled, and 16 were evaluable for responses. Median age was 64 (range 36–81) years. The combination was well tolerated, with only a few grade 3 adverse events: one dry skin, one diarrhea, and one musculoskeletal ache. The responses were 1 of 16 (6%) complete response (CR), 1 of 16 (6%) partial response (PR), 2 of 16 (13%) stable disease (SD), and 12 of 16 (75%) progressive disease (PD). Response rate (RR) was 2 of 16 (13%). Clinical benefit rate (CBR) at 16 weeks was 4 of 16 (25%). Median follow‐up was 24.9 months (95% confidence interval [CI], 17.5–30.9). Progression‐free survival (PFS) was 2.6 months (95% CI, 1.9–3.1) and overall survival (OS) was 25.5 months (95% CI, 10.4–not reached [NR]).

Conclusion

The combination of enobosarm and pembrolizumab was well tolerated, with a modest clinical benefit rate of 25% at 16 weeks in heavily pretreated AR+ TNBC without preselected programmed death ligand‐1 (PD‐L1). Future clinical trials combining AR‐targeted therapy with immune checkpoint inhibitor (ICI) for AR+ TNBC warrant investigation.

Discussion

There has been strong interest in targeting AR in breast cancer; however, single‐agent androgen receptor antagonists have shown modest benefit to date. Single‐agent bicalutamide showed a CBR of 19% at 6 months and no PR. Single‐agent enzalutamide trial reported a CBR of 35% at 4 months with a median PFS of 14 weeks and a RR of 5.9% [1]. Early literature supports synergy of AR‐targeted therapy with ICI, and concurrent expression of PD‐L1 and AR in TNBC has the potential to be a molecular biomarker.

Enobosarm (GTx‐024) is an oral nonsteroidal, tissue‐SARM that was initially developed for anabolic activity with minimal androgenic activity [2,3,4]. Preclinical study demonstrated that enobosarm inhibited proliferation and tumor growth of AR+ TNBC [5]. In this clinical trial of combined enobosarm and pembrolizumab in heavily pretreated patients with AR+ TNBC without preselected PD‐L1, we observed an RR of 12% and CBR of 25% at 16 weeks, with one patient showing complete response (Fig. 1). The modest response rate is comparable to the single‐agent pembrolizumab RR of 12% observed in KEYNOTE‐12 but was higher than the RR of 5% observed in KEYNOTE‐86 [6,7]. There were no grade ≥ 4 adverse events (AEs) observed on the study and minimal grade 3 AEs: one (6%) musculoskeletal ache, one (6%) dry skin, and one (6%) diarrhea.

Median follow‐up was 24.9 months (95% CI, 17.5–30.9). PFS was 2.6 months (95% CI, 1.9–3.1), and overall survival was 25.5 months (95% CI, 10.4–NR). The trial was stopped early because of withdrawal of the enobosarm drug supply. Our results may reflect modest efficacy of enobosarm as an AR‐targeted agent. Future clinical trials investigating AR‐targeted therapy in combination with ICI for AR+ TNBC are warranted.

Trial Information

Drug Information

Patient Characteristics

Primary Assessment Method

Adverse Events

Assessment, Analysis, and Discussion

Triple‐negative breast cancer (TNBC) is defined by lack of hormone receptor (HR) expression and human epidermal growth factor receptor 2 (HER2) overexpression and accounts for 15% of all breast cancer. This is an aggressive type of breast cancer with poor prognosis in the metastatic setting. TNBC is molecularly heterogeneous, with at least four distinct TNBC subtypes identified through genomic profiling, including luminal androgen receptor (LAR), mesenchymal, basal‐like immunosuppressed, and basal‐like immune‐activated [9, 10]. Recent U.S. Food and Drug Administration (FDA)–approved targeted therapies make precision medicine feasible in the treatment of TNBC. These include PARP inhibitors for BRCA mutated tumors [11, 12], immune checkpoint inhibitors for programmed death ligand‐1 (PD‐L1)–positive TNBC [13], and antibody drug conjugate targeting Trop‐2 receptor [14]. Despite these advances, treatment choice for metastatic androgen receptor (AR) + TNBC remain limited. The LAR subtype accounts for approximately 10%–15% of TNBC, and LAR cell lines have shown sensitivity to AR antagonist [10, 15]. There is strong interest in the use of AR as a potential therapeutic target.

Expression of AR has been observed in 70%–80% of all breast cancer [16–18], and HR‐positive (+) breast cancer exhibits significantly higher AR expression than HR‐negative (−) breast cancer [17, 18]. Moreover, AR expression seems to be associated with favorable clinical outcome in estrogen receptor (ER) + breast cancer [19, 20]. In TNBC, 10%–40% of tumors express AR across several studies, with immunohistochemistry (IHC) cutoff values ranging between 1% and 10% [18, 21, 22]. The prognostic value and precise mechanism of AR in TNBC remain controversial [21–23]. In preclinical studies, androgen induces migration and invasiveness in AR+ TNBC cell lines, whereas the AR antagonists bicalutamide and enzalutamide inhibit growth of TNBC cell lines in vitro [24, 25]. Gucalp et al. reported the efficacy of bicalutamide in patients with AR+ metastatic TNBC (mTNBC) [1]. The clinical benefit rate (CBR) of 19% observed with bicalutamide shows proof of concept for the efficacy of androgen blockade in AR+ (≥10% IHC) TNBC. The efficacy of enzalutamide has been studied in patients with AR+ mTNBC with a CBR 25% or 33% in AR with ≥1% or ≥ 10% by IHC, respectively, at 16 weeks [26]. Similarly, the 17α‐Hydroxylase/C17,20‐lyase (CYP17) inhibitor abiraterone showed modest clinical activity in AR+ (≥10% IHC) mTNBC with a response rate (RR) of 6.7%, CBR of 20%, and progression‐free survival (PFS) of 2.8 months [27]. The regimen was well tolerated, with grade 1–2 fatigue, hypertension, hypokalemia, and nausea being the most common adverse events (AE). Despite these encouraging results, efficacy of these single‐agent AR‐targeted therapies appears to be modest, and more effective combination therapies are needed.

Enobosarm (GTx‐024) is an oral nonsteroidal, tissue‐selective, AR modulator that was initially developed for anabolic activity with minimal androgenic activity [2–4]. Preclinical study demonstrated that enobosarm inhibited proliferation and tumor growth of AR+ TNBC [5]. Enobosarm provides a novel targeted approach to exploit the therapeutic benefits of AR modulation without virilization or estrogenic effects. In a proof‐of‐concept phase II study evaluating the safety and efficacy of 9 mg daily dosing of enobosarm in patients with AR+ HR+ metastatic breast cancer, the drug was well tolerated, and 8 of 22 (36%) patients had stable disease (SD). Pembrolizumab is a highly selective humanized monoclonal antibody of programmed death receptor (PD‐1) and has shown promising antitumor effects in solid tumors. Single‐agent pembrolizumab showed an RR of 18.5% and a median response of 17.9 weeks in KEYNOTE‐12, which enrolled 27 patients with PD‐L1+ mTNBC [6]. In PD‐L1 nonselected heavily pretreated patients with mTNBC, the RR was only 5.3% in the KEYNOTE‐86 cohort A (n = 170) [7]. Preclinical evidence has suggested a potential therapeutic synergy by combining AR modulation and PD‐1 blockade. Androgens are immunosuppressive and target both innate and adaptive immune systems to dampen the immune response [28–34]. Inhibition of AR sensitizes cancer cells to immune‐mediated killing. AR suppresses PD‐L1 expression by binding to the PD‐L1 promotor and directly attenuating PD‐L1 gene transcription [29, 30]. In addition, AR signaling inhibits thymocyte production [31], whereas AR blockade has been shown to increase naive T cell generation resulting in a larger diversity of T cell repertoire that could potentially respond to PD‐1 blockade [32, 33]. Moreover, AR blockade yielded increased survival benefits to vaccination in a murine prostate cancer model [34].

The complementary modes of action and low potential of overlapping toxicities of these two modalities make the combination therapy of AR‐targeted agent and immune checkpoint inhibitor an attractive option for AR+ TNBC. A proof‐of‐concept KEYNOTE‐199 trial showed that pembrolizumab plus enzalutamide demonstrated a RR of 12% and disease control rate of 51% for patients with metastatic castration‐resistant prostate cancer. We hypothesized that enobosarm could enhance pembrolizumab‐induced immune response while targeting AR. The current study was designed to assess the safety and efficacy of enobosarm plus pembrolizumab in patients with AR+ mTNBC.

A total of 18 patients were enrolled from April 2017 to April 2019. In April 2019, GTx Inc. notified the study team of withdrawal of enobosarm drug supply, and further enrollment was halted. Two patients were ineligible because of undiagnosed brain metastasis. Patient characteristics and treatment variables are described in Table 1 (n = 16). Median age was 64 (range, 36–81), with 50% non‐Hispanic white, 31% Hispanic, 13% Asian, and 6% African American. A total of 56% received 0–1 previous lines of therapy, and 44% received ≥ 2 prior lines of therapy.

The combination of pembrolizumab and enobosarm was well tolerated. Median dose delay and dose modification were both 0 (range, 0–2). Five of 16 (31%) had more than one dose delay, and 3 of 16 (19%) had more than one dose reduction. Causes of dose delay and/or dose reduction were grade 2 elevated liver function test (LFT); hypoglycemia; fever, nausea, or diarrhea; elective tissue expander removal; and patient's choice.

There were no grade ≥ 4 AEs observed in the study. Grade 1–3 AEs are shown in Table 2. Grade 3 AEs were one musculoskeletal ache (pain), one dry skin, and one diarrhea. AEs associated with potential pembrolizumab‐induced immune side effects were consistent with known side effects per FDA package insert.

Among 16 patients evaluable for response, 1 of 16 (6%) achieved complete response (CR), 1 of 16 (6%) achieved partial response (PR), 2 of 16 (13%) had SD, and 12 of 16 (75%) had progressive disease. Response rate (RR) was 2 of 16 (13%) (Fig. 1). CBR at 16 weeks was 4 of 16 (25%). Durable CR was found in one patient who continues to be progression‐free for over 32 months (Fig. 2). The median follow‐up was 24.9 months (95% confidence interval [CI], 17.5–30.9). PFS was 2.6 months (95% CI, 1.9–3.1), and overall survival (OS) was 25.5 months (95% CI, 10.4–NR; Fig. 3).

Stromal tumor infiltrating lymphocytes (TILs), tumor mutation burden (TMB), and PD‐L1 expression were evaluated in pretreatment tumor biopsies as potential biomarkers of response. Stromal tumor infiltration levels ranged from 2% to 12.5% and showed no correlation with PFS or response (Fig. 4A). TMB was low in all patients, ranging from 0–4.6 mutations per megabase (m/MB) (Fig. 4B).

Gene expression patterns in tumor tissues were assessed by whole transcriptome analysis (n = 16) and showed clustering of the top 20 variants for responders versus nonresponders (Fig. 5A). Although principle component analysis identified two distinct subgroups of patients, no noticeable differences in response to therapy or time point in therapy were found (Fig. 5B).

We next quantified potential differences in the activity of the androgen‐related pathway. The gene expression and signature score of the androgen‐related pathway were calculated and compared with a clinically relevant AR signature [39]. No significant differences in the AR signature score between responders and nonresponders were identified (Fig. 6).

CIBERSORTx with merged LM22 signatures were used to deconvolute RNA expression profiles to overall immune infiltration levels and the relative abundances of 18 immune cell subtypes. No significant difference in overall immune infiltration levels (absolute immune score) were found (p = .10) (Fig. 7A). Of note, the absolute abundance of naive B cells (p = .028) were observed to be significantly higher in responding patient tumors compared with nonresponding patient tumors (Fig. 7B).

Peripheral blood mononuclear cells were collected from patients at pretreatment, immediately prior to cycle 2, and immediately prior to cycle 4 for evaluation of immune phenotypes by flow cytometry. At pretreatment, there were no differences between responding and nonresponding patients in memory T cell composition in either CD4+ T cells or CD8+ T cells (Fig. 8). Similarly, we found no differences at pretreatment between responding and nonresponding patients in CD4+ T cell or CD8+ T cell expression of PD‐1, TIGIT, TIM‐3, KLRG1, or CD137 or frequencies of regulatory T cells or follicular helper T cells. We also found no differences in CD8+ T cell expression of PD‐1, TIGIT, TIM‐3, KLRG1, and CD137. To assess the potential impact of treatment on T cell phenotypes, we calculated fold change in expression percentages relative to pretreatment for cycle 2 and cycle 4 collected samples. No significant differences between responder and nonresponder patients were found over the course of therapy in CD4+ T cell or CD8+ T cell subsets. We do note that the patient with CR was observed to have a relatively high expression of TIGIT on CD4+ T cells at pretreatment that increased over the course of therapy. Similarly, the patient with CR was observed to have relatively high fold change in increased CD137 expression on both CD4+ and CD8+ T cells. Together, these data suggest that both pretreatment and on‐therapy peripheral T cell activation markers may be useful in monitoring patient responses to immune checkpoint inhibitors (ICIs).

Genomic alterations in responders versus nonresponders were analyzed, and the gene alterations were reported (Fig. 9). A total of 9 of 13 (69%) patients had a TP53 mutation, and 2 of 13 (15%) had EGFR, NF1, PIK3R1, and/or PTEN alterations. Results from this analysis were consistent with published breast cancer profiles [9, 10].

Our final exploratory analysis evaluated the microbiome. Patient samples classified by MetaPlAn2 were examined [35] by Bray‐Curtis hierarchical clustering (Fig. 10A). One cluster (green) consists of samples from a patient 11 and was notable for a marked lack of diversity. This patient's microbiome is dominated by Escherichia coli, and patient had PFS of 1 month. Two clusters (gold and black) contained stable disease and responders (patient 4 had CR and patient 8 had PR) with high levels of some Bacteroides spp. and Alistipes spp., indicating a potentially healthier gut microbiome. Alpha diversity by patient is relatively stable for most (Fig. 10B). Patient 9 had high diversity at baseline that decreased at cycle 5 (patient had PFS of 2 months). In contrast, patient 11 had increased diversity at cycle 5. Significant organisms determined by PFS (< 6month vs. >6 month) is shown in Figure 10C. There was no clear correlation between diversity and response or PFS.

The trial was stopped early because of GTx's withdraw of drug supply, with a total of 16 evaluable patients included in the analysis.

In the current study, we observed an RR of 12% and CBR of 25% with pembrolizumab plus enobosarm in heavily pretreated patients without PD‐L1 preselection. One patient showed complete response and remains in remission at the time of this manuscript submission. This study was originally designed to test the hypothesis that enobosarm plus pembrolizumab would lead to an RR of 29% with 29 patients. With the early termination of the accrual, we did not meet this aim. Although from a strict statistical point of view, these data do not allow a rejection of the null hypothesis, from a clinical perspective, our results remain meaningful. The modest response rate is comparable with the single‐agent pembrolizumab RR of 12% observed in KEYNOTE‐12 [6] yet was numerically higher than RR of 5% observed in KEYNOTE‐86 [7]. Within the bicalutamide study, a CBR of 19% at 6 months and no PR was reported. In the enzalutamide trial, CBR at 4 months was 35% and median PFS was 14 weeks. A total of 2 of 118 (1.7%) CR and 5 of 118 (4.2%) PR were observed, which translates to 5.9% RR. Early literature supported synergy of AR‐targeted therapy with ICI. Concurrent expression of PD‐L1 and AR in TNBC showed potential to apply combination treatment of AR inhibitor and ICIs [36]. In addition, preclinical studies demonstrated that blockade of AR stimulates thymic volume and maturation of naive T cell clones, which can facilitate antitumor immune responses and induce synergic effect with ICI [32, 37]. Currently, a phase II trial studying dual immune checkpoint blockade (nivolumab plus ipilimumab) plus bicalutamide is ongoing in patients with metastatic AR+ HER2− breast cancer (BC) and TNBC [33]. Our modest result could also reflect modest efficacy of enobosarm as an AR‐targeted agent.

AR is a highly expressed steroid receptor in breast cancer, with 75–95% of ER+ and 50% of ER− BCs expressing AR [38]. AR is an intracellular steroid hormonal receptor of the nuclear receptor family along with ER, PR, glucocorticoid, and mineralocorticoid receptors. It links a transcription factor that regulates specific genes involved in cellular processes with a diverse range of actions such as stimulating both cell proliferation or apoptosis, depending on the concurrent activated signaling pathways [39, 40]. The role of AR in breast cancer is complicated and varies in different subtypes. It was observed that ligand‐bound AR binds to an estrogen‐related element in the nucleus leading to cell apoptosis in ER+ AR+ cell lines. In contrast, AR binds to an androgen‐related element in the nucleus, leading to cell proliferation in ER negative AR+ cell lines [41, 42]. In TNBC, the AR pathway shows cross‐talk with several other key signaling pathways, including the PI3K/AKT/mTOR pathway [43–45]. AR status, by itself, may not be a reliable prognostic marker and it varies across different race and ethnicity [46].

Despite the enthusiasm for targeting AR for LAR TNBC, single‐agent AR‐targeted therapies have been associated with modest response [26, 27]. Combination of AR‐targeted therapy with other cancer driver pathway inhibitors have been investigated. LAR tumors have focal gains twice as frequently on 11q13.3 (CCND1, fibroblast growth factor family) and 14q21.3 (MDGA2) [9]. Lehmann et al. reported that PIK3CA mutations were highly clonal and more frequent in AR + versus AR− TNBC (40% vs. 4%) and often associated with concurrent amplification of the PIK3CA locus. The Pan‐PI3K inhibitor GDC‐0941 or the dual PI3K/mTOR inhibitor GDC‐0980 had an additive growth inhibitory effect when combined with genetic or pharmacological AR targeting with bicalutamide in AR + TNBCs in vitro. A phase II trial combining the α‐specific PIK3CA inhibitor alpelisib with enzalutamide in patients with AR+/PTEN loss metastatic breast cancer is ongoing (NCT03207529). In addition, phase II trials combining CDK 4/6 inhibitors ribociclib (NCT03090165) or palbociclib (NCT02605486) with bicalutamide are actively being conducted [47, 48] in AR+ metastatic breast cancer, and results are eagerly awaited.

Immune biomarkers to predict responses have been extensively studied for personalized treatment. We analyzed TILs [49], TMB [50], PD‐L1 expression(IHC 22C3 pharmDx (Dako, Inc.) at QualTek Molecular Laboratory), and subsets of T cells in the study. Whereas breast cancer has low TMB compared with other immunogenic cancers, such as melanoma and lung cancer that have shown good responses to PD‐1 inhibitors [51], TNBC has relatively higher TMB than other types of breast cancer [52]. In our study, TMB was generally low in all patients, ranging from 0–4.6 m/MB, whereas the patient with CR had the highest TMB. TILs were not associated with treatment responses, and PD‐L1 positivity was observed in two patients, including one who achieved CR. There were no significant differences in CD4+ T cell nor CD8+ T cell subsets between responder and nonresponder patients.

In our exploratory CIBERSORTx analysis [53, 54, 55], M2 macrophages and mast cells were more enriched in the tumors of nonresponders, which is consistent with the immune suppressive roles of M2 macrophages and mast cells in cancer. M2 macrophages inhibit inflammation and promote tissue remodeling and angiogenesis [56]. In preclinical studies, mast cell‐secreted angiogenic cytokines directly facilitate tumor vascularization and thereby stimulate other inflammatory cells of the tumor microenvironment to release other angiogenic mediators. Hence, increased numbers of mast cells have been associated with angiogenesis in breast cancer [57].

The relation between the human microbiomes and cancer has been explored recently. Although many questions still remain unanswered, emerging evidence from various studies show that the presence of specific microbiomes play a role in breast cancer tumorigenesis [58]. In TNBC, one study that analyzed stool microbiomes using 16S rRNA amplicon sequencing identified that microbial signatures (Bacteroides and Ruminococcaceae) were associated with treatment response to chemotherapy [59]. Several other studies showed that microbiome may affect the response to ICI in solid tumors [60, 61]. In metastatic melanoma, baseline gut microbiomes were associated with clinical responses to CTLA‐4 inhibitor or PD‐1 inhibitor [62, 63]. We analyzed gut microbiomes of patients using whole metagenome sequencing and classification by MetaPhlAn2. One patient with a lack of diversity on the microbiomes had a very short PFS, but there was no clear correlation between diversity and response when compared responders with nonresponders. Alpha diversity by patient was relatively stable for most patients. Several organisms of potential interest associated with patients with increased PFS were identified by LEfSe. Further study is warranted.

Acknowledgments

Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA provided drug, and financial support for the study. We thank GTx, Inc., for providing the study drug enobosarm. We also thank the STOP Cancer Foundation (principal investigator [PI] Yuan Yuan), phase I Foundation (PI Yuan Yuan), NCI K‐12 Career Development Award (K12CA001727, PI Joanne Mortimer), and the National Institutes of Health (P30CA033572). We thank the patients who participated this study. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute.

Disclosures

Yuan Yuan: Puma, Pfizer, Immunomedics (C/A), Merck, Eisai, Novartis, Puma, Genentech, Celgene and Pfizer (RF), Eisai, Novartis, Genentech, AstraZeneca, Daiichi Sankyo, Immunomedics (SAB); Mina Sedrak: Eli Lilly, Seattle Genetics, Novartis, Pfizer (RF‐institutional). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board.

Figures and Tables

References