Immune Modulation with RANKL Blockade through Denosumab Treatment in Patients with Cancer

Denosumab is FDA approved for the prevention of skeletal-related events (SRE) in patients with bone metastases from solid malignancies as well as in patients with nonmetastatic prostate cancer who are receiving androgen deprivation therapy and patients with breast cancer who are receiving adjuvant aromatase inhibitor therapy. Osteoblasts and osteoclasts, respectively responsible for bone formation and bone resorption, are critical in maintaining the skeletal structure. To maintain a balance between bone formation and resorption, osteoblasts express receptor activator of NFκB ligand (RANKL) and osteoclast precursors express RANK receptor. This RANK–RANKL system drives osteoclast stimulation and function. The RANK–RANKL interaction can be inhibited by the endogenous decoy receptor osteoprotegerin (1) or therapeutically by denosumab. Denosumab is a fully human mAb that interferes with RANK–RANKL interactions and inhibits osteoclast maturation, recruitment, and activity (2).

The RANK–RANKL pathway also plays critical roles in the development and regulation of the immune system, especially with regard to the regulation of dendritic cell (DC) activity. RANKL is primarily expressed on CD4⁺ and CD8⁺ T cells following stimulation and interfaces with RANK, which is highly expressed on DCs. RANKL has been shown to enhance DC survival (3), potentially via induction of the antiapoptotic protein B-cell lymphoma-extra large (Bcl-xL; ref. 4) and the antiapoptotic serine/threonine kinase AKT/PKB (5). RANKL also induces DC expression of multiple activating cytokines, including IL1, IL6, IL12, and IL15. While RANKL–RANK signaling has been shown to be important in immune activation, this interaction may also have a role in immunosuppression: RANKL has been shown to induce T-cell tolerance via expansion of CD4⁺CD25⁺ regulatory T cells (Treg; ref. 6). RANK–RANKL signaling also plays a key biological role in immune system development. RANKL is required for proper lymph node development and the RANK–RANKL pathway regulates proliferation of medullary thymic epithelial cells (mTEC) and thymic medulla formation, which ultimately contribute to establishment of self-tolerance (7).

Preclinical observations have revealed that the RANK–RANKL axis is involved in the modulation of DCs and the negative selection of developing thymic produced naïve T cells (3–7). The effects of denosumab on immune activation are not well understood, leading to uncertainty as to whether this drug might enhance or inhibit immunotherapies including immune checkpoint inhibitors. Therefore, we initiated a prospective, nonrandomized, biobanking protocol with patients that presented with bone metastases, solid malignancies—other than multiple myeloma—and received standard-of-care denosumab treatment.

This study was a prospective, nonrandomized, biobanking protocol at University of California, San Francisco (UCSF) to obtain serial blood samples from patients starting standard-of-care denosumab (120 mg every 4 weeks). All patients or their legally accepted representatives provided written informed consent to participate after being informed of the nature, purpose, participation conditions, and standard risks of phlebotomy. Key eligibility criteria for this study included patients had to be ages 18 or older with a histologically confirmed diagnosis of solid tumor malignancy (other than multiple myeloma) with bone metastasis and qualify to receive treatment with denosumab for prevention of SREs. Patients who received new systemic treatment (either intravenous or oral) within 4 weeks of the start of denosumab initiation were excluded from the study.

Patients had blood drawn for peripheral blood mononuclear cell (PBMC) isolation and cryopreservation at the following time points: screening visit (pre-denosumab, TP1), prior to second denosumab injection (TP2), prior to third denosumab injection (TP3), prior to sixth denosumab injection (TP4). To assess any potential interactions with immunotherapies, additional blood could be drawn pretreatment and posttreatment when a patient was transitioning to an immune-based therapy, including sipuleucel-T and ipilimumab.

A total of 26 patients with advanced disease enrolled on this protocol, 23 patients with prostate cancer and 3 patients with renal cell carcinoma. Characteristics of trial participants are summarized in Supplementary Table S1. This collection was approved by the UCSF Institutional Review Board and conducted as a single institution trial (protocol CC#13982). This study was conducted in accordance with Declaration of Helsinki, Belmont Report, Nuremberg Code, and National Bioethics Advisory Commission.

High-dimensional cytometry by time of flight (CyTOF) Banked cryopreserved PBMCs from the study were thawed, barcoded, and stained with metal-labeled antibodies to create an immune-monitoring panel for PBMCs and T-cell subsets (for details of the antibodies, see Supplementary Table S2). Mass tag barcoding of samples was performed by incubating cells with distinct combinations of isotopically-purified palladium ions chelated by isothiocyanobenzyl-EDTA (Fluidigm, catalog no. 201060). Samples were washed and data collected through mass cytometry (Helios, Fluidigm). Data were then prepared for analysis using the R Premessa 0.2.0 package (https://github.com/ParkerICI/premessa/). Barcodes were run through CytoNorm (ref. 8; https://github.com/saeyslab/CytoNorm) to correct for small batch variances using several samples that were run redundantly to serve as validation samples between different barcodes, clustering samples to normalize the data between files. PBMCs were extracted from FCS files and divided into 200 unsupervised clusters using statistical scaffold analysis and clusters were entered into a force-directed graph centered around predetermined landmark nodes made by manual gating on Cytobank (9) as shown in the supplementary gating strategy (Supplementary Fig. S1). Calculations for clustering [by both Uniform Manifold Approximation and Projection (UMAP) and Scaffold] were performed based upon the relative staining of CD3, CD4, CD8a, CD11b, CD11c, CD14, CD19, CD25, CD31, CD33, CD45RA, CD15, CD123, TCRgd, CD56, Tbet, FceR1a, CCR7, FoxP3, HLA-DR, CD61_CD235ab, CD19, CD45, and CD16. Cluster frequency and Boolean expressions for markers in each cluster were passed through Significance Across Microarrays algorithm and the results were formulated into Scaffold maps for visualization (https://github.com/spitzerlab/statisticalscaffold). Scaffold heat maps, fold changes, significances, cell counts, and nearest landmark nodes were extracted and visualized using custom script to generate heat maps using R pheatmap package (https://CRAN.R-project.org/package=pheatmap). UMAP was downsampled to 1,000 cells from each of the 74 samples and generated by using the uwot package (https://github.com/jlmelville/uwot) for dimensionality reduction and Rphenograph (ref. 10; https://github.com/JinmiaoChenLab/Rphenograph) for clustering.

Cryopreserved PBMCs were thawed and 250,000 cells were extracted for RNA using the Invitrogen RNAqueous-Micro Total RNA Isolation Kit (Thermo Fisher Scientific #AM1931). A total of 10.4 ng of RNA was used for reverse transcription (RT) and cDNA generation, following the Smart-seq2 protocol with modifications as described here (11). The oligo-dT primer was biotinylated at the 5′ end. We added a barcode, unique molecular modifiers, and biotin to the 5′ end of the TSO primer (Integrated DNA Technologies, Primer sequences in Supplementary Table S3). Maxima H Minus Reverse Transcriptase (Thermo Fisher Scientific #EP0753) was used for RT. Barcoded RT product was then pooled together and purified using a DNA Clean & Concentrator-5 Kit (Zymo Research, #D4013). A total of 16 cycles were used for cDNA amplification. cDNA was then enriched for alpha and beta T-cell receptors (TCR) separately and used for Illumina library generation by PCR (Integrated DNA Technologies, Primer sequences in Supplementary Table S3). Libraries were sequenced on a NovaSeq 6000 and run across two SP 300 lanes with run parameters 22×6×0×297 (R1×I1×I2×R2).

Sample barcodes were used to separate each sample from the library pool and the data were preprocessed with MiXCR version 3.0.13. Among 81 samples with valid TCR sequencing, we had 69 evaluable samples. Clonality (12) and TCR convergent frequency (13) were used to measure the global TCR metrics. TCR relative clonality (RCL, i.e., ratio of clonality at later time point vs. baseline clonality) and TCR clonality were used to measure the treatment effect. The comparison between posttreatment time points and baseline was done by Wilcoxon signed-rank test. The comparison between the two groups was done by the Wilcoxon rank-sum test. No adjustments for multiple testing were done.

Cryopreserved PBMCs were thawed and 300,000 to 400,000 cells per patient time point were extracted for DNA using a QIAamp DNA Blood Mini Kit (Qiagen, #51104) following the manufacturer's protocol. A total of 10–100 ng of DNA was used for quantification of T-cell receptor excision circles (TREC) by RT-PCR with a TRECSProbe (Applied Biosystems #45002), Actin Probe (5′- 6FAM ATG CCC TCC CCC ATG CCA TCC TGC GT TAMRA - 3′), and TaqMan (Applied Biosystems #4326708). Plates were prepared per Applied Biosystems recommendations for the StepOnePlus Real-Time PCR System (Thermo Fisher Scientific #4376598) and amplified for 40 total cycles. Standard curves of tRNA were prepared on every reaction plate in triplicate for a detectable range for 100,000 to 12 TREC copies and β-actin (Supplementary Fig. S2A).

The RANKL qPCR was carried out with RANKL TaqMan Gene Expression Assay (FAM, Hs00243522_m1 catalog no. 4331182) and internal control actin-B, ACTB TaqMan Gene Expression Assay (VIC, Hs01060665_g1 catalog no. 4448489, diluted 1:8 in dH2O) using TaqMan Gene Expression Master Mix Kit Life Technologies (catalog no. 4369016) on cDNA generated with High-Capacity RNA-to-cDNA Kit Life Technologies (catalog no. 4387406), from 5 ng of RNA extracted from 250,000 PBMC. Delta C_t (cycle threshold) is C_t of RANKL minus C_t of ACTB in each sample. Fold change is 2^{−delta delta Ct}. Delta delta C_t is delta C_t of time point 2 or time point 3 minus delta C_t of pretreatment time point 1.

Human TNFSF11 (RANKL) ELISA Kit was used to quantitate soluble RANKL protein in human plasma (Abcam, ab213841). Previously frozen plasma from patients’ samples were thawed at room temperature, spun at 1,500 × g for 15 minutes and 100 μL of supernatant was loaded into duplicate wells of the ELISA plate. RANKL was quantitated according to the manufacturer's instructions.

In general, continuous outcomes were summarized by median and interquartile range, compared between time points by Wilcoxon signed-rank test, and compared between groups by Wilcoxon rank-sum test. Statistical significance was declaimed on the basis of P value < 0.05 unless otherwise noted. Major statistical analysis was done by using the statistical computing software R unless noted.

Mass cytometry data will be shared publicly on FlowRepository while R-based analysis codes are available at https://github.com/fonglab/scaffold-heatmap-pub. The TCR sequencing data generated in this publication have been deposited in NCBI's Gene Expression Omnibus (GEO) and are accessible through GEO Series accession number GSE248774. All other data generated in this study are available within the article and its Supplementary Data or upon request from the corresponding author.

Relevant Institutional Review Boards at UCSF approved the study protocol (CC#13982), which allowed for patient biospecimen collection on patients receiving standard-of-care denosumab for metastatic cancer. Each patient provided signed informed consent prior to enrollment.

All authors have provided their consent to the publication of this work.

We obtained written informed consent and collected PBMC samples from patients with cancer with bone metastasis receiving standard-of-care denosumab treatment (Fig. 1A). High-dimensional CyTOF was used to assess the effects of denosumab in the circulating immune compartment (Fig. 1B). Because RANKL is expressed on myeloid cells, we initially assessed for changes in the different myeloid-cell populations between the pretreatment baseline time point (TP1) and the on-treatment time point 3 (TP3). We did not see any significant changes in frequencies of myeloid cells (no blue or red nodes on scaffold map) of the different myeloid-cell populations (Fig. 1C). Heat maps were then used to assess for changes in phenotypic marker expression across the myeloid-cell populations. Although there were minimal changes on circulating myeloid cells overall, we did observe a transient increase in CD137 expression on some populations of conventional DCs, classical monocytes, and intermediate monocytes in TP2, but these were lost in subsequent time points. We also observed transient increases in PD-L1 and TIM3 expression on the classical and intermediate monocyte cell populations at TP3 (Fig. 1D).

Next, we interrogated the lymphocyte compartment to assess whether RANKL inhibition can lead to downstream effects on effector populations. We again did not observe a statistically significant difference in the frequency of various lymphocyte populations after denosumab treatment (Fig. 2A). Similar to the myeloid-cell populations, we observed modest changes in activation markers overall and a transient increase in CD137 on naïve CD4⁺ T cells, and Tregs, and a rare population of natural killer T (NKT) cells at TP2. We also observed transient increases in PD-L1 on naïve CD8⁺ T cells and NKT cells. We saw transient increases at TP3 in CD40 on naïve and effector CD4⁺ T cells, and Ki-67 on effector CD8⁺ T cells (Fig. 2B). With additional doses of denosumab at TP4, we no longer observed these changes. We also observed expression of the immune checkpoint markers PD-L1, TIM3, and CTLA-4, irrespective of denosumab and concomitant therapies, prior to denosumab treatment at TP1 and posttreatment (Supplementary Fig. S3A and S3B). Thus, changes observed in our time point comparisons of functional markers are relative to a baseline level of expression.

To examine whether RANKL expression was affected in circulating immune cells, we quantified mRNA expression pretreatment and posttreatment. Relative to pretreatment expression levels of RANKL, changes in RANKL expression were minimal. However, we observed a trend toward a decrease in the expression of RANKL in circulating immune cells at TP3 (Supplementary Fig. S2B; P = 0.055). We next quantified the amount of circulating RANKL from plasma samples at TP1 and TP3. In contrast with the trend of decreased RANKL expression, overall, the amount of soluble RANKL was low and we did not observe statistically significant changes after denosumab treatment (Supplementary Fig. S2C).

Next, we interrogated whether RANKL inhibition could effect changes in the T-cell repertoire. We performed TCR β chain sequencing on serial PBMCs (Fig. 3A). We found that clonality at TP3—after the two treatments of denosumab—trended higher compared with the clonality at baseline and after the first injection of denosumab (Fig. 3B). We also examined TCR convergent frequency across the course of treatment but only saw modest changes in convergence between time points (Fig. 3C).

We examined whether denosumab could alter the frequency of recent thymic emigrants. We quantified TRECs across patient treatment time points. The level of TRECs was similar across the time points and there was no statistically significant change after denosumab treatment (Fig. 4A). Because these patients were receiving denosumab for metastatic cancer as standard of care, some of the patients concurrently received chemotherapy, steroids, or both, as part of their treatment. Although not statistically significant, we observed that treatment with chemotherapy, steroids, or both, reduced the amount of TRECs (Fig. 4B). Furthermore, there was no difference in RANKL mRNA expression between the two groups (Supplementary Fig. S2D).

To determine whether there were any broad effects of chemotherapy or steroids, or both, during the course of denosumab treatment, we performed statistical scaffold analysis comparing TP1 with on-treatment time points separating the patients by whether or not they were receiving these other treatments. The patients who received concurrent chemotherapy or steroids, or both, showed broad changes with denosumab treatment whereas patients who received denosumab without these concomitant treatments did not (Fig. 5A). In patients receiving these concomitant treatments, the changes with treatment included significantly higher levels of CTLA-4 and TIM3 after treatment across a broad range of lymphoid- and myeloid-cell populations. When comparing these patient groups at pretreatment TP1, patients who received chemotherapy, steroids, or both, before denosumab showed greater immune modulation across the myeloid- and lymphoid-cell populations (Supplementary Fig. S4). When we removed patients who received steroids and compared patients who only received chemotherapy with patients who did not receive concurrent treatment prior to denosumab, we observed similar changes (Supplementary Fig. S5A). We further separated the concurrent treatments patients received into three groups and compared them at TP2 and TP3: patients who received chemotherapy or chemotherapy and steroids, patients who received only steroids, and patients who did not receive chemotherapy or steroids. Similar to our comparison of concomitant treatments, we observed broad changes in activation markers between the groups and modest differences in the frequency of immune cell populations in the lymphoid- and myeloid-cell compartments (Supplementary Fig. S5A–S5C). We also evaluated the effects of chemotherapy and steroids on the T-cell repertoire. To interrogate changes in the TCR sequences, we compared the clonality between the patients who only received denosumab and those who also received denosumab with concomitant chemotherapy, steroids, or both at all time points (Fig. 5B). We found a trend toward increased TCR clonality in patients who received concomitant therapies across all time points. We also found that the relative clonality from baseline to TP3 was higher in the patients who received chemotherapy, steroids, or both, though it was not statistically significant (Fig. 5C; 1.17 vs. 1.04, P = 0.073).

Denosumab is used to prevent SREs and bone metastasis in multiple cancer types. However, the RANK–RANKL axis has been shown to play a role in inducing DC expression of stimulatory cytokines (14) and in regulating the homeostasis of Tregs (15). Because of the effects of the RANK–RANKL axis on the immune system, patients receiving denosumab have been excluded from many immunotherapy trials. However, RANK–RANKL interactions are also critical for the development of mTECs, which express autoimmune regulator (Aire) and are critical in mediating the negative selection of autoreactive T cells (7). Indeed, preclinical models have shown that RANKL blockade can enhance antitumor immune responses as well as enhance the efficacy of immune checkpoint inhibition (16–19). One study in a murine model found that treatment with RANKL blockade increases the frequency of CD4⁺ single positive T cells (17). While another showed that RANKL and CTLA-4 blockade in combination increases the frequency of tumor-infiltrating lymphocytes (20). An additional study reported increased CD8⁺ T-cell frequency in blood and tissue of murine models and in paired blood and tumor tissue from patients with breast cancer after denosumab treatment (21). In contrast, our study focused on interrogating the circulating immune cells of patients with metastatic cancer receiving denosumab through longitudinal sampling of PBMCs. Our data comparing PBMC samples at pretreatment with posttreatment time points demonstrated no significant changes in circulating immune cell frequencies. We found only modest changes in the phenotype of circulating myeloid- and lymphoid-cell populations. Additional studies should be conducted to validate the effects of denosumab on immune cell frequencies.

Because of preclinical models showing modulation of T cells, we further interrogated how denosumab affected the T-cell repertoire by analyzing TCR sequences from patients. Although not statistically significant, we observed a transient increase in clonality immediately after treatment, suggesting a narrowing of the repertoire. Yet, we did not notice any other significant changes or trends in clonality and TCR convergent frequency. RANKL upregulation has been shown in the contexts of T-cell response to tolerogenic signal (22) and peripheral CD8⁺ T cells undergoing deletional tolerance (23). Thus, blocking the RANK–RANKL pathway with denosumab may play a role in the transient clonality increase we observed through tolerogenic mechanisms. We also investigated whether denosumab would lead to the generation of more thymic naïve T cells due to less stringent negative selection. We hypothesized that by disrupting the development of mTECs, denosumab would alter the self-tolerance in developing T cells (24). An increase in TRECs would suggest increased T-cell diversity (17), but we did not observe any significant differences in the amount of TRECs after denosumab treatment. The thymus involutes with age and is estimated to decline over time in thymic T-cell production with a half-life of about 15.7 years (25–27). Because the median age of our study patients is 67.5 years, the lack of observed change in TRECs could reflect the age-related decline in thymic function (28, 29). Perhaps if denosumab is used in a younger population, the outcomes of thymic emigrants may differ. Overall, these data suggest that denosumab minimally changes the diversity of both the peripheral T-cell population as well as recent thymic emigrants.

Stratifying our data by patients who received conventional cancer treatments, such as chemotherapy and corticosteroids, in combination with denosumab revealed greater immune modulation in patients who received these other therapies. In patients who received chemotherapy and/or steroids, we observed an increase in TIM3 expression across DC and monocyte populations and an increase in CTLA-4 in the T-cell compartment. Despite TIM3 facilitating DC antigen cross-presentation (30), other studies have shown that TIM3-expressing DCs inhibit activation of nucleic acid sensing Toll-like receptors (31) and weaken CD8⁺ T-cell cytotoxicity indirectly (32). Thus, the implications of our observed increase in TIM3 expression are unclear. Patients that did not receive chemotherapy or steroids showed no changes in functional markers after receiving denosumab consistent with modest effects of denosumab on circulating immune cells. However, when we stratified patients at denosumab pretreatment TP1, the changes in activation markers were significantly more robust. Considering this, the changes observed in patients who received concomitant treatment could be due to pre-existing differences before denosumab treatment began. Further segregating the concurrent treatments also revealed greater immune modulation in patients who received chemotherapy or chemotherapy and steroids. Because steroids are commonly given together with chemotherapy as an antiemetic treatment, our interrogation of concurrent treatments was limited, as there was only 1 patient who received only chemotherapy. Furthermore, only 1 patient received only steroids prior to denosumab treatment, consequently limiting our pretreatment analysis to comparing chemotherapy or chemotherapy and steroids with no chemotherapy or steroids. Dividing the treatment groups also reduced our statistical power; 2 patients received only steroids at TP2 and 3 patients received only steroids at TP3. A more abundant sample size of patients across the various concomitant treatment groups is required to validate our observations. Our analysis of clonality in the T-cell repertoire revealed that clonality and relative clonality in patients who received chemotherapy, steroids, or both were not significantly different. This aligns with a previous finding that chemotherapy depletes T-cell subsets, but clonality does not significantly change (33).

The design of our study as a prospective, nonrandomized, biobanking protocol has limitations. Because of scope of the study, progression-free survival and overall survival were not captured to assess clinical outcomes. In addition, because patients received denosumab as standard of care, we were not able to obtain serial biopsies due to the attendant risks so analysis of changes in the tumor microenvironment could not be performed. Although we did not have a control cohort, we were still able to evaluate differences in the peripheral immune landscape longitudinally, because the first time point was obtained without denosumab. While our study provides insights into the immune modulation of patients with cancer receiving denosumab, this dataset is derived from a small cohort where only prostate and renal cell carcinoma are represented.

In summary, our data suggest that administering denosumab alone results in very modest changes with respect to immunomodulation, thymic T-cell generation, and peripheral T-cell repertoire. In contrast, administering denosumab with conventional cancer therapies, such as chemotherapy and corticosteroids, results in broad changes in phenotypic markers that include the induction of multiple immune checkpoints across multiple lymphoid- and myeloid-cell populations. Larger cohorts will be needed to separate the effects of steroids and chemotherapy, and to tease apart the mechanisms by which these treatments can sensitize these different immune populations to denosumab. Clinical data suggest that patients receiving corticosteroids either prior or during immune checkpoint inhibitor therapy differ in neither overall survival nor progression-free survival (34). Despite the limited effects of denosumab on the immune landscape, future studies should further explore the relationship of denosumab with concomitant cancer therapies including immune checkpoint inhibitors.

H. Chang reports grants from Amgen during the conduct of the study. M.L. Cheng reports personal fees from Pfizer, Mirati Therapeutics, Janssen, Boehringer Ingelheim, and AstraZeneca, and grants from Palleon Pharmaceuticals outside the submitted work. E.D. Chow is a founder of Survey Genomics. R.R. Aggarwal reports grants from Amgen and Janssen, grants and personal fees from AstraZeneca and Merck, and personal fees from Curio, Boxer Capital, PCCTC, Novartis, Lumanity, Targeted Oncology, Scitaris, and KeyQuest outside the submitted work. T.W. Friedlander reports grants and personal fees from Seagen, personal fees from Astellas, Aadi Biosciences, Merck, and Gilead, and grants from Roche Genentech outside the submitted work. E.J. Small reports personal fees and other support from Fortis, other support from Harpoon, and Teon, personal fees from J&J, Janssen, and Deciphera during the conduct of the study, and personal fees from J&J, Janssen, and Deciphera outside the submitted work. M. Anderson reports other support from Merck, Inc outside the submitted work. L. Fong reports grants and personal fees from Amgen during the conduct of the study, grants and personal fees from Abbvie, BMS, Dendreon, and Roche/Genentech, grants from, Bavarian Nordic, Janssen, Merck, and Nektar; nonfinancial support from Actym, Atreca, Bolt, Immunogenesis, Nutcracker, Scribe, and Senti; personal fees from AstraZeneca, Crescendo, Daiichi Sankyo, and Innovent, and personal fees and nonfinancial support from Bioatla, NGMBio, RAPT, and Sutro outside the submitted work. No disclosures were reported by the other authors.

H. Chang: Data curation, formal analysis, writing–original draft, writing–review and editing. J. Marquez: Data curation, formal analysis. B.K. Chen: Visualization. D.M. Kim: Investigation. M.L. Cheng: Conceptualization. E.V. Liu: Formal analysis, visualization. H. Yang: Software, formal analysis, visualization. L. Zhang: Data curation, formal analysis. M. Sinha: Formal analysis, supervision. A. Cheung: Data curation, supervision. S.S. Kwek: Data curation, formal analysis, investigation. E.D. Chow: Software. M. Bridge: Data curation. R.R. Aggarwal: Data curation. T.W. Friedlander: Data curation. E.J. Small: Data curation. M. Anderson: Conceptualization. L. Fong: Conceptualization, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing.

The clinical trial was supported in part by Amgen. L. Fong was supported by a Prostate Cancer Foundation Challenge Grant and NCI R35CA253175.