Randomized Phase II Study Evaluating the Addition of Pembrolizumab to Radium-223 in Metastatic Castration-resistant Prostate Cancer

This first-in-human, randomized trial demonstrates that combining radium-223 with pembrolizumab does not significantly impact the tumor microenvironment in patients with metastatic castration-resistant prostate cancer. Data highlight that upregulation of immune checkpoint molecules could impede the combination’s clinical efficacy.

Abstract

The checkpoint immunotherapeutic pembrolizumab induces responses in a small minority of patients with metastatic castration-resistant prostate cancer (mCRPC). Radium-223 (R223) may increase immunogenicity of bone metastases and increase pembrolizumab (P) activity. In a randomized phase II study, we assessed the effect of R223+P compared with R223 on tumor immune infiltration, safety, and clinical outcomes in patients with mCRPC. The primary endpoint was differences in CD4⁺ and CD8⁺ T-cell infiltrate in 8-week versus baseline bone metastasis biopsies; secondary endpoints were safety, radiographic progression-free survival (rPFS), and overall survival (OS). Of the 42 treated patients (29 R223+P, 13 R223), 18 R223+P and 8 R223 patients had evaluable paired tumor biopsies. Median fold-change of CD4⁺ T cells was −0.7 (range: −9.3 to 4.7) with R223+P and 0.1 (−11.1 to 3.7) with R223 (P = 0.66); for CD8⁺ T cells, median fold-change was −0.6 (−7.4 to 5.3) with R223+P and −1.3 (−3.1 to 4.8) with R223 (P = 0.66). Median rPFS and OS was 6.1 (95% confidence interval: 2.7–11.0) and 16.9 months [12.7–not reached (NR)], respectively, with R223+P and 5.7 (2.6–NR) and 16.0 (9.0–NR), respectively, with R223. Although R223+P was well tolerated with no unexpected toxicity, the combination did not improve efficacy. High-dimensional flow cytometry demonstrated minimal immune modulation with R223, whereas R223+P induced CTLA-4 expression on circulating CD4⁺ T cells. Clinical responders possessed lower circulating frequencies of Ki67⁺ T and myeloid cells at baseline and higher circulating frequencies of TIM-3⁺ T and myeloid cells by week 9. Although R223+P did not induce T-cell infiltration into the tumor microenvironment, exhaustion of induced peripheral T-cell immune responses may dampen the combination's clinical activity.

Introduction

The anti-programmed death-1 (PD-1) checkpoint inhibitor (CPI) pembrolizumab is FDA approved for unresectable or metastatic microsatellite instability-high (MSI-H), mismatch repair deficient (dMMR), or tumor mutational burden-high (TMB-H) solid tumors, including prostate cancer, that have progressed following prior treatment. Clinical activity is seen in the approximately 2%–3% of metastatic castration-resistant prostate cancer (mCRPC) cases with the MSI-H/dMMR phenotype (1), but it is not yet clear whether TMB-H status is an independent predictive biomarker for CPI response in prostate cancer (2, 3). Clinical benefit with nivolumab (4), pembrolizumab (5), and atezolizumab (6) in biomarker-unselected cases has also been limited.

One proposed strategy for increasing the activity of CPIs in prostate cancer and other solid tumors is to combine these agents with radiotherapy (7, 8). The observation of an abscopal effect in metastatic cancers, where radiation to a single site of disease leads to responses outside of the irradiated lesion, demonstrates that radiotherapy can stimulate immune responses in patients (9). The proposed mechanisms for immune stimulation by radiotherapy include increasing tumor antigen presentation, inducing immunogenic cell death, increasing costimulatory signaling, upregulating PD-L1 expression, and altering the tumor microenvironment (TME) by increasing lymphocyte recruitment and infiltration (10, 11). Many clinical trials combining radiotherapy and immunotherapy have been reported or are in-progress, with the overarching aim to enhance antitumor immune responses and improve response rates and patient outcomes (12).

Given limited clinical evidence of meaningful abscopal effects, it has been argued that the likelihood of benefit from combining single-site radiation with checkpoint immunotherapy is low (13), and that multi-site radiotherapy is preferred (14). Multi-site radiation can be achieved through external beam radiation to multiple lesions or through systemic administration of radiopharmaceuticals. The radiopharmaceutical radium-223 is a calcium mimetic that accumulates at sites of bone metastases due to their high rate of bone turnover. It binds to new bone stroma and emits alpha particles, which induce double-stranded DNA breaks that result in tumor cell death (15). Alpha particle radiation can induce immunogenic cell death (16), and is thus of particular interest for combination with immunotherapeutics (17). Tumor cell killing with radium-223 would be expected to release tumor-associated antigens, which, when presented by antigen-presenting cells, would induce T cells to mature into tumor-specific effector cells that proliferate and infiltrate the TME (18). These effector cells can then secrete IFNγ, which would in turn induce adaptive upregulation of PD-L1 in the tumor (19). The addition of an anti-PD1 CPI would be expected to antagonize the effect of PD-L1 upregulation on cancer cells and reverse any potential anergy of PD-1⁺ infiltrating antitumor immune cells. Tumor heterogeneity and the dynamic nature of PD-L1 expression on tumor cells, as well as PD-1 and PD-L1 expression on immune cells in the TME, supports the potential for synergistic activity of radium-223 with pembrolizumab, even in the absence of high PD-L1 expression prior to treatment initiation.

To test the hypothesis that the combination of pembrolizumab with radium-223 induces an immune response that would enhance clinical activity compared with radium-223 alone, we conducted a randomized phase II trial of radium-223 with or without pembrolizumab in patients with mCRPC to bone to evaluate changes in immune cell milieu of the bone metastases and TME, as well as to assess the safety and preliminary efficacy of the combination.

Methods and Materials

Patient selection

Eligible patients were 18 years of age or older, had histologically confirmed prostate adenocarcinoma, Eastern Cooperative Oncology Group (ECOG) performance status ≤1, radiographic evidence of metastatic disease to bone by conventional imaging, serum testosterone levels <50 ng/dL, and evidence of disease progression based on rising PSA or bone scan progression per Prostate Cancer Working Group 2 (PCWG2) criteria (20). There was no limit on the number of prior lines of therapy, although patients who previously received radium-223 or PD-1, PD-L1, or PD-L2 blocking therapy were excluded. Patients must have been willing to undergo a biopsy of a bone metastasis prior to study drug initiation and after approximately 8 weeks on study. After observing a high number of nondiagnostic baseline biopsies in the first 21 patients, the protocol was amended to require presence of tumor in the biopsy specimen for eligibility. If the initial biopsy specimen was nondiagnostic, patients were required to undergo a second biopsy to enroll. However, if the second biopsy was also nondiagnostic, patients were permitted to enroll on the study.

Small cell histology was excluded, as were patients with visceral metastases, local recurrence in the prostate bed, non-nodal soft-tissue lesions (including soft-tissue lesions in the bone) ≥1 cm. Nodal disease was permitted if <1.5 cm in the short axis or if 1.5 to 3 cm and no evidence of growth by >5 mm over the prior 6 months and if not causing symptoms or obstruction. Patients with immunodeficiency, receiving systemic steroid therapy prednisone >10 mg daily or equivalent, active autoimmune disease that has required systemic treatment in the past 2 years, or any history of noninfectious pneumonitis were excluded. The Dana-Farber Cancer Institute (DFCI) Institutional Review Board approved the protocol. Informed written consent was obtained from each subject. The study was conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonization of Good Clinical Practice Guidelines (NCT03093428).

Study design

The study schema is depicted in Supplementary Fig. S1. After enrollment, study participants underwent bone biopsy and were randomized 2:1 to receive radium-223 at 55 kBq/kg every 4 weeks + pembrolizumab at 200 mg every 3 weeks (Arm A, R223+P) or radium-223 at 55 kBq/kg every 4 weeks alone (Arm B, R223). Standard-of-care radium-223 was acquired from commercial supply, and pembrolizumab was provided by Merck. Stratification factors were alkaline phosphatase ≥220 versus <220 U/L [as in the phase III ALSYMPCA trial of radium-223 (21)] and high versus low volume bony metastases by CHAARTED criteria (22). Participants then underwent repeat biopsy after 8 weeks of therapy (+3-week window) or after two doses of radium-223 if delays occurred. PSA was measured at baseline, week 1 day 1 (per treating investigator's discretion), every 4–6 weeks on therapy, and at study end. CT imaging of the chest, abdomen, and pelvis, as well as bone scintigraphy were performed every 12 weeks. If restaging after three doses of radium-223 showed at least stable disease (SD) by RECIST 1.1 and PCWG2 criteria [i.e., without progressive disease (PD) in lymph nodes or soft tissue, or unconfirmed PD in bone], R223+P patients suspended R223 dosing and continued pembrolizumab alone until radiographic PD. Upon PD, radium-223 was resumed if no new visceral metastases were seen and if felt appropriate by the treating investigator. Participants with SD at initial restaging were permitted to continue radium-223 without interruption per discretion of the treating physician. Patients continued treatment until clinical/radiologic progression, unacceptable toxicity, or completion of six radium-223 doses and up to 24 months (or 35 doses) of pembrolizumab. The primary endpoint was differences in CD4⁺ and CD8⁺ T-cell infiltrates after 8 weeks of treatment versus baseline biopsy as detailed below. Key secondary endpoints were safety/tolerability, radiographic progression-free survival (rPFS), and overall survival (OS). Exploratory endpoints included rate of symptomatic skeletal events (SSE) and PSA response.

Hematoxylin and eosin, IHC, and immunofluorescence staining

Biopsies for biomarker studies were performed by an interventional radiologist under CT guidance using an 18-gauge needle or larger with plan to obtain four to six specimens. The first core was placed in a specimen container with 10 mL 10% neutral-buffered formalin (EMS #15742-10) at ambient temperature and embedded in paraffin per institutional standards. For the second core, touch preparation was performed by transferring the biopsy core on a prelabeled slide and rolling against the slides gently to leave blood smear on the slide, then fixed immediately in methanol. The second core was then rapidly frozen in prechilled optimal cutting temperature (OCT) medium (Tissue-Tek # 4583) in a metal base mold (Sakura/VWR #4163/#25608-828) in dry ice, then after OCT was solidified, a razor blade was used to trim excess OCT from the sides of the block to fit into mega cassettes (Tissue-Tek #4173). The procedures for formalin fixation/paraffin embedding and for touch preparation/OCT freezing were alternated for subsequent biopsy specimens. The specimens fixed in formalin and embedded in paraffin were stored at 4°C and the frozen specimens were stored at −80°C in the mega cassettes until time of shipment. Frozen OCT biopsies were provided to the Center for Cancer Precision Medicine team at DFCI for genomic analysis, and formalin-fixed paraffin-embedded (FFPE) biopsies were shipped ambient to University of California San Francisco (UCSF) in sturdy packaging. The baseline biopsy was obtained up to 12 weeks (84 days) prior to initiation of treatment on day 1. Bone biopsy can have been done prior to screening; archival specimens of bone metastasis were permitted if done for another purpose and available. The on-treatment biopsy was performed after approximately 8 weeks of study therapy or after two doses of radium-223 if delays had occurred such that the patient did not receive two doses of radium-223 prior to the 8-week timepoint.

The frequency of CD4⁺ or CD8⁺ cells was assessed with multiplex immunofluorescence (mIF) and quantitated with digital image analysis. Tissue staining was conducted in the UCSF Histology and Biomarker Core. FFPE sections were collected at 4-μm thickness and mounted on center of positively charged glass slides (Superfrost Plus Slides Fisher part no. 12-550-15) following lab standard procedures for the sectioning of paraffin blocks. Slides were air-dried overnight for 8 to 16 hours then baked and deparaffinized. For hematoxylin and eosin (H&E) staining (Thermo Fisher Scientific part no. 6765015), slides were baked at 60°C for 15 minutes prior to staining; for IHC staining, slides were baked overnight at 60°C prior to staining. H&E staining was performed on the Leica XL Autostainer using hematoxylin (Thermo Fisher Scientific part no. 6765015) for 7 minutes and alcoholic eosin (Thermo Fisher Scientific part no. 6765040) for 20 seconds. IHC was done on the Roche Ventana Discovery Ultra autostainer (PD-L1, Cell Signaling Technology part no. 13684, 1:100; Ki67, Ventana part no. 790-4286, RTU; CD8, Ventana part no. 790-4460, RTU; CD4, Ventana part no. 790-4423, RTU; FoxP3, Abcam part no. 999963, 1:25). For the mIF, the Zeiss Axio scanner was employed for digital slide scanning, with brightfield and immunofluorescent images collected at 20× magnification using Zeiss Zen software. Image analysis was performed using StrataQuest Tissue Analysis Solutions Software v.5.0 (TissueGnostics Medical and Biotech Solutions). Cells were considered CD4⁺ or CD8⁺ with the presence of high-intensity fluorescence (cell shade mean intensity of 20 and 3, respectively). The proportion of each cell population was calculated as (cell count +1)/total cells to prevent division by 0 for fold-change calculations.

Tissue staining, image analysis, and correlation with blood frequencies

RNAscope (Advanced Cell Diagnostics, ACD Bio) ISH was performed on 4 μm FFPE sections from bone marrow metastasis specimens, with existing mass cytometry and immunofluorescent data. Tissues were pretreated with target retrieval reagents and protease to improve target recovery based on guidelines provided in the RNAscope Multiplex Fluorescent Reagent Kit v2 Assay protocol. Probes for human CD68, PD-L1, Tim3, and TIGIT mRNA (ACD Bio) were incubated at 1:700 dilution for 2 hours at 40°C. The probes were then hybridized with Opal 7-Color Manual IHC Kit (PerkinElmer) for detection of CD68, PD-L1, Tim3, and TIGIT using Opal 520, Opal 620, Opal 570, and Opal 690, respectively. Tissues were counterstained with DAPI (ACD Bio, part no. 323100, RTU). Slides were imaged at 40X magnification using a Zeiss Axio Scan.Z1 (Zeiss Group). No staining was seen with negative control probes (ACD Bio, part no. 321831, RTU) on prostate tissue or healthy tonsil tissue. Three regions of interest (ROI) from each specimen slide were selected at random for analysis.

The ARK Python library v0.6.3 (23) was used to perform cell segmentation based on the DAPI (nuclear) stain, extract single-cell marker counts, and generate a CSV file of cell size-normalized and arcsinh-transformed single-cell marker counts. Microsoft Excel was used to produce histograms and biaxial scatterplots based on the arcsinh-transformed counts that were then used to determine appropriate marker cutoffs (CD68: 0.0, PDL1: 6.0, TIM3: 4.0) used to gate our populations of interest: PDL1⁺CD68⁺ cells (pop1) and TIM3⁺CD68⁻ cells (pop2). Population frequencies from each sample were averaged (n = 3 ROIs) and compared with frequencies of comparable PDL1⁺CD68⁺ myeloid cell (pop1) and TIM3⁺CD3⁺ T-cell (pop2) populations in the corresponding blood samples (details below) taken at baseline from each patient (n = 10). The Pearson correlation statistic was used to determine an r value of −0.544 and −0.079 for pop1 and pop2, respectively.

Mass cytometry staining and analysis

Cryopreserved whole blood peripheral blood mononuclear cells (PBMC) were obtained at week 1 (predose), week 5, week 9, week 13, week 21, and time of progression for biomarker analysis. A total of 40 mL of blood at each timepoint was collected into five CPT sodium heparin tubes and processed to PBMCs at the DFCI Center for Immuno-Oncology Immune Assessment Laboratory within 2 hours of collection, aliquoted into four cryovials, then batch shipped to the UCSF Cancer Immunotherapy Lab for long-term storage in liquid nitrogen. Available cryopreserved PBMCs were thawed, barcoded, and stained with heavy metal-labeled antibodies (Supplementary Table S1) for analysis by mass cytometry by time-of-flight (CyTOF) as described previously (24). The samples were then washed and acquired on a mass cytometer (Helios, Fluidigm). Contour plots were generated for CD3⁺ T cells with Cytobank (25). We used flowCore v2.0.1 (https://rdrr.io/bioc/flowCore/) to load the data into R 4.0.2, uwot (https://arxiv.org/abs/1802.03426) to make the uniform manifold approximation and projection (UMAP) projection, Rphenograph (26) for clustering, and ggplot2 (https://ggplot2.tidyverse.org) for plotting. The calculation of both the UMAP projection and the clustering were based on the relative staining of CD3, CD4, CD8a, CD11b, CD11c, CD14, CD16, CD19, CD25, CD31, CD33, CD45RA, CD56, CD66, CD117, CD123, CD127, CD235ab/CD61, BDCA3, CCR7, FceRIa, FoxP3, γδTCR, HLA-DR, T-bet, TCR Va24-Ja18, and VISTA. Rphenograph clustering with a k value of 200 yielded eight clusters, which were annotated on the basis of the relative staining of CD4, CD8, FoxP3, CD25, CD127, CCR7, CD45RA, T-bet, and HLA-DR. Cytobank was used to perform manual gating and create landmark nodes for statistical scaffold analysis (27). Live T cells (Singlets, Intercalator⁺ Cisplatin⁻ CD45⁺, CD61⁻, CD235ab⁻, CD3⁺, CD19⁻) were extracted from the fcs files and divided up into 30 unsupervised clusters using statistical scaffold. Clusters were assigned vectors associated with the average median value of markers and edges, which were defined as similarity between vectors to produce graphs, which show the relationships between different clusters. Cluster frequencies and Boolean expression for functional markers for each cluster were passed through the Significance Across Microarrays (28) algorithm and results were formulated into the Scaffold maps for visualization (github.com/nolanlab/scaffold). For Scaffold heat maps, fold-change, significance, cell count, and nearest landmark node, data were extracted from the outputs of the Scaffold analysis using a custom script (https://github.com/fonglab/scaffold-heatmap-pub). The heat map was created in R using pheatmap (https://CRAN.R-project.org/package=pheatmap), with log₂ fold-change capped at 2.

Whole-exome sequencing

Slides were cut from FFPE blocks and macrodissected for tumor-enriched tissue. Paraffin was removed. DNA was extracted using the QIAGEN AllPrep FFPE DNA/RNA Extraction kit (Qiagen catalog no. 80234). Germline DNA was obtained from PBMCs using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen catalog no. 80224). After shearing was performed, libraries were then prepared using KAPA HyperPrep Kit with Library Amplification (Roche Sequencing Solutions product KK8504). Samples underwent exome hybrid capture using a custom bait set (Twist Biosciences) then sequenced using the Illumina NovaSeq S4 platform. Tumor samples were included for analysis only if they successfully underwent whole-exome sequencing (WES) and had a matched germline sample. Samples with tumor purity <20%, as determined by ABSOLUTE, were excluded (29). FFPE specimens for 14 of the 42 treated patients had no/insufficient tumor, and of the 28 that underwent DNA/RNA extraction, 10 DNA specimens reached >50% target coverage for WES, one of which had a mismatched germline specimen, leaving nine for analysis. Reads were aligned using BWA-MEM (30). Somatic mutations were called using a customized version of the Getz Lab CGA WES Characterization pipeline (gcr.io/broad-getzlab-workflows/cga_production_pipeline:v0.1) developed at the Broad Institute. In brief, we used MuTect1 (31) to call single-nucleotide variants and Strelka (32) to call short indels. MuTect2.1 (33) was used to confirm Strelka indel calls. Variant calls were filtered for FFPE (34) and 8-oxoG (35) artifacts. DeTiN was used to rescue true somatic variants that were removed because of tumor-in-normal contamination (36). Variants were filtered through a panel of normal samples to remove artifacts from miscalled germline alterations and other rare error modes. Variants were annotated using VEP and Oncotator. Nonsense mutations and indels were considered deleterious, as were missense mutations categorized as significant by COSMIC Cancer Mutation Census (https://cancer.sanger.ac.uk/cmc/home). Allelic copy number, tumor purity, and tumor ploidy were determined using ABSOLUTE (29). TMB was calculated as the number of nonsynonymous protein-coding mutations divided by megabase of exome sequenced.

Plasma cell-free DNA-targeted panel sequencing

Plasma collection for circulating cell-free DNA (cfDNA) analysis was introduced in a protocol amendment, with plasma specimens available for 21 of the 42 treated patients. A total of 10 mL of blood were collected in ethylenediamine tetraacetic acid blood tubes (BD366643) at week 1 (predose), week 5, week 13, and time of progression. A total of 0.6 mL whole blood from the predose specimen was transferred to a cryovial for germline DNA and stored at −80°C. The remaining whole blood was immediately centrifuged within 1 hour of collection at 1,500 (± 150) × g for 10 minutes, with supernatant centrifuged again at 3,000 (± 150) × g for 10 minutes, then plasma supernatant was collected stored in cryovials at −80°C. Plasma samples were thawed at room temperature and subject to high-speed centrifugation. The QIAsymphony DSP Circulating DNA Kit (Qiagen) was used to extract cfDNA from the plasma samples. Germline DNA was obtained from PBMCs using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen catalog no. 80224) and sheared as described above. Libraries for duplex sequencing were prepared using KAPA HyperPrep Kit with Library Amplification (product KK8504) and duplex unique molecular index adapters. Unique 8-base dual index sequences embedded within the p5 and p7 primers (IDT) were added during PCR. For ultra-low pass whole-genome sequencing, up to 95 libraries were pooled and run on one lane using paired 151 bp runs with cluster amplification using Exclusion Amplification cluster chemistry (Illumina) and NovaSeq SP flowcells. Tumor fraction (TF) was estimated using ichorCNA (37), and for 16 of the 21 participants with a cfDNA specimen with TF > 0, the same library underwent hybrid capture for 319 prostate cancer-relevant genes. Hybridization and capture are performed using the relevant components of IDT's XGen hybridization and wash kit and following the manufacturer's suggested protocol, with the exception of using custom panels synthesized by TWIST biosciences instead of the recommended xGen Lockdown Panels and Probe Pools. A set of 12-plex prehybridization pools were created. These prehybridization pools were created by equivolume pooling of the normalized libraries, IDT XGen blocking oligos and human Cot-1 included in the IDT xGen Hybridization and Wash Kit (IDTDNA catalog no 1080584). The prehybridization pools underwent lyophilization using the Biotage SPE-DRY concentrator. After lyophilization, custom exome bait (TWIST Biosciences, Supplementary Data S1–S3), along with hybridization mastermix, was added to the lyophilized pool prior to resuspension. Samples are incubated overnight at 65°C using an Eppendorf Mastercycler x50. Library normalization and hybridization setup were performed on a Hamilton Starlet liquid handling platform, whereas target capture was performed on the Agilent Bravo automated platform. After capture, a PCR was performed to amplify the capture material using custom primers developed by IDT (Dual Index Universal Primer Forward: AATGATACGGCGACCACCGAGATCTCAC; Dual Index Universal Primer Reverse: CAAGCAGAAGACGGCATACGAGAT). After capture enrichment, library pools were quantified using qPCR (automated assay on the Agilent Bravo), using the KAPA Library Quantification kit (Illumina catalog no. 07960204001) with probes specific to the ends of the adapters. On the basis of qPCR quantification, pools were normalized using a Hamilton Starlet to 2 nmol/L and sequenced using Illumina sequencing technology. Cluster amplification of library pools was performed according to the manufacturer's protocol (Illumina) using Exclusion Amplification cluster chemistry and HiSeq X flowcells. Flowcells were sequenced on v2 Sequencing-by-Synthesis chemistry for HiSeq X flowcells. The flowcells are then analyzed using RTA v.2.7.3 or later. Each pool of libraries was run on paired 151 bp runs, reading the dual-indexed sequences to identify molecular indices, and sequenced across the number of lanes needed to meet coverage for all libraries in the pool. Reads were aligned using BWA-MEM. Consensus duplex sequences were generated from reads using fgbio CallDuplexConsensusReads (http://fulcrumgenomics.github.io/fgbio/). Next, we used Mutect2 to identify single-nucleotide variants and short indels (33). Variants were annotated using Funcotator (https://gatk.broadinstitute.org/hc/en-us/articles/360037224432-Funcotator). Nonsense mutations and indels were considered deleterious, as were missense mutations categorized as significant, by COSMIC Cancer Mutation Census (https://cancer.sanger.ac.uk/cmc/home). Given the overall expected low tumor fraction of plasma cfDNA in prostate cancer, copy-number alterations were not called. Similar to WES, TMB was calculated as the number of nonsynonymous protein-coding mutations per megabase.

Germline variant calling

Germline variants were determined using DeepVariant (38). Variants were considered deleterious if listed as “pathogenic” or “likely pathogenic” in ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/).

Statistical analysis

The primary endpoint was the comparison of differences in CD4⁺ or CD8⁺ T-cell infiltration measured in the bone biopsy specimens from baseline and 8 weeks between the treatment arms. Assuming all enrolled patients had paired biopsies at baseline and 8 weeks, we targeted a sample size of 45 (30 in R223+P and 15 in R223) to detect a standardized effect size of 0.92 in the change of CD4⁺ and CD8⁺ T cells between the two treatment arms. This would achieve a power of 80% at a two-sided alpha level of 0.05 using Wilcoxon rank-sum test. A two-sided P value of less than 0.05 was considered statistically significant.

Toxicities were graded per Common Terminology Criteria for Adverse Events (CTCAE) version 4.0 (39). Adverse events (AE) of special interest were liver function test abnormalities, dyspnea/pneumonitis, pruritus/rash, and fractures. SSEs were defined as the need for external beam radiation to relieve bone pain, new/symptomatic pathologic fracture, spinal cord compression, or orthopedic surgical intervention related to bone metastasis. rPFS was defined as the time from treatment initiation to the earlier of the first documentation of definitive disease progression by RECIST v 1.1 criteria for soft-tissue disease and PCWG2 guidelines for bone disease or death due to any cause or censored at last imaging date. Given the number of patients who discontinued trial therapy for clinical progression or physician/patient decision and thus censored for rPFS, we also assessed time to treatment failure (TTF), defined as the time from treatment initiation to any progression (clinical, symptomatic, or radiographic) during the treatment or follow-up, initiation of new anticancer therapy, or death due to any cause, whichever came first or censored at last imaging date. OS was defined as time from treatment initiation to death due to any cause or censored at date last known alive. PSA progression was defined using PCWG2 criteria (20); time to PSA progression was defined from first date of PSA measured postbaseline or prestudy PSA, depending on the data availability, to PSA progression according to PCWG2 criteria or censored at date of last PSA collected. The distributions for rPFS, TTF, OS, and time to PSA progression were estimated using the Kaplan-‐Meier method. PSA response rates were summarized descriptively.

We additionally explored the association of T-cell infiltration, both at baseline and changes with treatment, with clinical endpoints. Participants in both arms were combined for this analysis because only 8 R223 patients had evaluable paired biopsies. Formal hypothesis testing was not performed because of the small sample size.

Data availability

The data generated in this study are available within the article and its Supplementary Data. Genomic data from whole exome sequencing from tumor tissue and targeted panel sequencing of cfDNA are available in dbGaP through accession number phs001988.v2.p1. Additional data are available upon request from the corresponding author.

Results

All analyses were performed using data obtained as of the cut-off date of August 11, 2020. Between July 05, 2017, and August 22, 2019, 63 patients at DFCI and Beth Israel Deaconess Medical Center were screened for enrollment, and 45 were randomized. Of these, 2 patients randomized to Arm A (R223+P) and 1 patient to Arm B (R223) did not initiate trial therapy (Supplementary Fig. S2), whereas 42 patients received at least one dose of study treatment (29 R223+P, 13 R223) and were analyzed per protocol (Table 1). Among 42 treated patients, 41 underwent biopsies at baseline and 32 underwent biopsy after approximately 8 weeks on therapy. Of these, 26 paired biopsy specimens were evaluable for the primary endpoint (Supplementary Table S2).

Table 1.

Baseline characteristics in overall cohort and by treatment arm.

	Overall	Radium-223 + Pembrolizumab	Radium-223	P-valueb
N	42	29	13
Age, years, median [IQR], (range)	70 [64–72], (58–81)	69 [64–72], (58–79)	71 [65–73], (60–81)	0.45
ECOG performance status				0.76
0	33 (78.6%)	22 (75.9%)	11 (84.6%)
1	9 (21.4%)	7 (24.1%)	2 (15.4%)
Gleason score				0.47
6	1 (2.4%)	0 (−)	1 (7.7%)
7	14 (33.3%)	10 (34.5%)	4 (30.8%)
8–10	20 (47.6%)	15 (51.7%)	5 (38.5%)
Missing	7 (16.7%)	4 (13.8%)	3 (23.1%)
Laboratory data
PSA, ng/mL, median [IQR], (range)	22.0 [7.6–119.7], (0.8–1990.0)	26.6 [7.5–94.4], (2.9–1990.0)	18.1 [8.1–185.9], (0.8–311.8)	0.69
Absolute neutrophil count, K/μL, median [IQR], (range)	3.9 [3.1–4.7], (2.4–41.4)	3.9 [3.3–4.5], (2.4–41.4)	3.8 [3.0–4.7], (2.4–6.7)	0.60
Absolute lymphocyte count, K/μL, median [IQR], (range)	1.2 [1.0–1.5], (0.3–1.9)	1.2 [1.0–1.5], (0.3–1.9)	1.0 [1.0–1.2], (0.8–1.8)	0.41
Alkaline phosphatase, U/L, median [IQR], (range)	110 [74–216], (47–596)	107 [74–233], (47–596)	123 [79–157], (61–562)	0.73
<220 U/L	32 (76.2%)	21 (72.4%)	11 (84.6%)
≥220 U/L	10 (23.8%)	8 (27.6%)	2 (15.4%)
Hemoglobin, g/dL, median [IQR], (range)	12.8 [11.5–13.7], (9.1–15.7)	12.8 [11.5–13.7], (9.1–15.3)	12.9 [11.5–13.7], (10.9–15.7)	0.69
Volume of bone metastasesa				0.99
Low	5 (11.9%)	4 (13.8%)	1 (7.7%)
High	37 (88.1%)	25 (86.2%)	12 (92.3%)

Note: Values are n (%) or median [IQR], (range).

^aAccording to the ECOG 3805 CHAARTED criteria, a high volume of metastases was defined as ≥4 bone lesions with at least one beyond the vertebral bodies and pelvis.

^b P values were calculated using Kruskal–Wallis test for continuous variables and Fisher exact test for categorical variables to compare patients in two treatment arms.

T-cell infiltration

CD4⁺ and CD8⁺ T-cell infiltration measured in bone biopsy specimens were compared at baseline and 8 weeks of treatment for 26 of 42 (62%) treated patients: n = 18 R223+P and n = 8 R223. Three patients elected not to pursue the 8-week biopsy because their baseline biopsy was negative; all other treated patients who did not undergo the 8-week biopsy were off-treatment (one due to patient withdrawal after 2 weeks on study and the others for progression, patient/physician decision, or death). Fold-change in CD4⁺ and CD8⁺ T cells (as a proportion of the total number of cells) between baseline and 8-week biopsy timepoints in the 26 evaluable pairs is depicted in Fig. 1. Median fold-change of the CD4⁺ T-cell proportion was -0.7 (range: −9.3 to 4.7) for R223+P-treated patients and 0.1 (−11.1 to 3.7) for R223-treated patients (P = 0.66); for CD8⁺ T cells, −0.6 (−7.4 to 5.3) was the median fold-change for R223+P-treated patients and −1.3 (−3.1 to 4.8) for R223-treated patients (P = 0.66). For participants in both arms of the trial, there was an increase in CD4⁺ and/or CD8⁺ T-cell infiltration detected in some paired biopsy specimens, whereas unchanged or decreased frequencies were observed in others (Fig. 1). We also examined changes in CD4⁺FoxP3⁻ effector cells and in the ratio of CD8⁺ T cells to CD4⁺FoxP3⁺ T cells [including naïve regulatory T cells (Treg), effector Tregs, and non-Treg populations (40)], which also showed no consistent change with treatment in either arm (Supplementary Fig. S3).

Figure 1.

Changes in CD4⁺ and CD8⁺ T-cell infiltration in bone biopsy specimens with radium-223 plus pembrolizumab and radium-223 alone. The frequency of CD4⁺ or CD8⁺ cells in baseline and 8-week bone biopsy specimens was assessed with mIF and quantitated with digital image analysis. The total cell count and CD4⁺ T-cell and CD8⁺ T-cell counts were obtained from three ROIs from each specimen slide at random, and the proportion of each cell population was calculated as (cell count +1)/total cells to prevent division by 0 for fold-change calculations. Waterfall plots of log₂ fold-change for proportions of A, CD4⁺ T cells and B, CD8⁺ T cells from baseline to 8 weeks on treatment are shown here. Each bar represents an individual patient, with blue and red bars indicating the assigned treatment [blue: radium-223+pembrolizumab (R223+P), N = 18; red: R223 alone, N = 8]. A two-sided P value of less than 0.05, determined through the Wilcoxon rank-sum test for the standardized effect size, indicated statistical significance in the change of CD4⁺ T-cell and CD8⁺ T-cell infiltration levels between the two treatment arms. Representative images of mIF for CD4⁺ (green) and CD8⁺ (red) T cells with DAPI counterstain (blue) for patients with an increase (C) and decrease (D) in T-cell infiltrate on treatment. Scale bar = 50 μm.

Circulating immune cell modulation

High-dimensional mass cytometry (CyTOF) was utilized to more broadly evaluate the effects of treatment on circulating immune cells (Fig. 2A). We applied statistical scaffold analysis to compare the frequency of immune cells at baseline (week 1) versus week 5 (Fig. 2B). In participants who received R223+P, we observed increased frequencies in central memory CD4⁺ T cells and CD8⁺ effector memory T cells from baseline to week 5 of treatment (Fig. 2B, clusters #2 and #9 depicted in the red populations), whereas no populations were significantly changed in participants who received R223 alone. To assess the changes in the activation state of the different T-cell subtypes, we applied statistical scaffold analysis for the percentage of T cells positive for various functional markers. Whereas essentially no changes from baseline to week 5 were seen with R223 alone (Fig. 2C, “Arm B”), R223+P resulted in decreased PD-1 staining across multiple cell types, as would be expected because the patients were treated with a PD-1 antibody (Fig. 2C, “Arm A”). R223+P also resulted in increased CTLA-4 on CD4⁺ effector and memory T cells. We also observed increased CD25 on specific natural killer (NK) cell and myeloid populations.

Figure 2.

High-dimensional flow cytometry assessment of circulating immune cells. Available cryopreserved PBMCs from baseline and week 5 timepoints were thawed, barcoded, and stained with heavy metal-labeled antibodies for analysis by CyTOF. The calculation of both the UMAP and the clustering were based on the relative staining of CD3, CD4, CD8a, CD11b, CD11c, CD14, CD16, CD19, CD25, CD31, CD33, CD45RA, CD56, CD66, CD117, CD123, CD127, CD235ab/CD61, BDCA3, CCR7, FceRIa, FoxP3, γδTCR, HLA-DR, T-bet, TCR Va24-Ja18, and VISTA. Contour plots were generated for CD3⁺ T cells with Cytobank, which was also used to perform manual gating and create landmark nodes for statistical scaffold analysis. A, A UMAP plot of immune cells from a representative PBMC sample. Clusters corresponding to canonical immune subsets are represented using distinct colors. NK: natural killer cells; cDC: conventional dendritic cells; pDC: plasmacytoid DCs; Treg: regulatory T cells. B, Scaffold map of significant changes in the R223+P Arm (Arm A; radium-223+pembrolizumab) between baseline (week 1) and week 5 (L = lymphoid cell types, M = myeloid cell types). Black-colored nodes represent landmark nodes that are canonical T-cell types identified by traditional manual cellular gating. The remaining nodes represent 30 unsupervised T-cell clusters created by scaffold analysis, which are arranged around the landmark nodes, and each other, based on similarity and connected by edges (the length of which is dependent on cluster similarity). The size of the clusters is proportional to cellular abundance. The color-coding represents clusters that exhibited significant differences (q<0.05; red = increase, blue = decrease) in percent marker-positive clusters when comparing the two groups. The absence of blue color coding indicates that no clusters are significantly decreased. C, Heat maps summarizing log₂ fold-changes resulting from statistical scaffold analysis of functional markers 4-1BB, CD44, CTLA-4, GITR, HLA-DR, ICOS, Ki-67, OX40, PD-1, PD-L1, PD-L2, TIGIT, TIM-3, and VISTA when comparing the pretreatment timepoint with week 5 in arms A and B. Each box represents a T-cell cluster that is labeled according to the nearest connected landmark node and ordered by cell count abundance. The color-coding represents clusters that showed a significant difference (q<0.05) in the log₂ fold-change, with red being significantly higher and blue being significantly lower at the week 5 timepoint. The intensity of the color code is proportional to the log₂ fold-change and is capped at 2 and −2.

Safety

A summary of AEs experienced by study participants is depicted in Table 2, and a complete listing of AEs is detailed in Supplementary Table S3. Pembrolizumab was administered for a median of 8 doses (range: 1–32) over a median of 4.8 months (range: <1–23.3). Six and 4 patients experienced dose delay and dose interruption in pembrolizumab, respectively, and 2 patients discontinued treatment due to toxicity. The median number of radium-223 doses was 4 (range: 1–6) administered over a median of 3.0 months (range: <1–18.2) in the combination arm, and a median of 6 doses (range: 1–6) administered over a median of 4.7 months (<1–5.1) in the R223 arm, respectively. One patient each in the R223+P and R223 arms had a dose delay in radium-223. Three patients who received R223+P experienced grade 3 non-hematologic AEs reported as related to treatment with pembrolizumab by the investigator: diarrhea, aspartate aminotransferase (AST) elevation, and pneumonitis (one each). Grade 4 sepsis and grade 5 stroke were experienced by 1 patient each treated with R223+P and were deemed not related to study drugs per the treating investigator. There was no evidence for excess SSEs for R223+P compared with R223 (Supplementary Table S4).

Table 2.

Summary of AEs regardless of attribution.

	Radium-223 + Pembrolizumab, n = 29		Radium-223, n = 13
CTCAE v. 4.0	Any grade	Grade ≥3	Any grade	Grade ≥3
AEs of special interest
Fracture	4 (14%)	1 (3%)	1 (8%)	0 (−)
ALT increased	2 (7%)	0 (−)	0 (−)	0 (−)
Alk Phos increased	7 (24%)	0 (−)	1 (8%)	1 (8%)
AST increased	7 (24%)	1 (3%)	1 (8%)	0 (−)
Total bilirubin increased	2 (7%)	0 (−)	0 (−)	0 (−)
Dyspnea	8 (28%)	1 (3%)	0 (−)	0 (−)
Pneumonitis	4 (14%)	2 (7%)a	0 (−)	0 (−)
Pruritus	6 (21%)	0 (−)	0 (−)	0 (−)
Rash	13 (45%)	0 (−)	0 (−)	0 (−)
AEs in ≥20% of participants
Fatigue	14 (48%)	0 (−)	3 (23%)	0 (−)
Anemia	12 (41%)	5 (17%)	4 (31%)	1 (8%)
Diarrhea	10 (34%)	2 (7%)	5 (38%)	0 (−)
Nausea	10 (34%)	0 (−)	3 (23%)	0 (−)
Pain	6 (21%)	0 (−)	5 (38%)	0 (−)
Back pain	9 (31%)	2 (7%)	1 (8%)	0 (−)
Hypothyroidism	8 (28%)	0 (−)	2 (15%)	0 (−)
Thrombocytopenia	5 (17%)	2 (7%)	3 (23%)	0 (−)
Other grade ≥3 AEs
Neutrophil count decreased	4 (14%)	1 (3%)	1 (8%)	1 (8%)
Pain in extremity	3 (10%)	1 (3%)	1 (8%)	0 (−)
Leukocyte count decreased	2 (7%)	2 (7%)	1 (8%)	0 (−)
Urinary tract infection	3 (10%)	2 (7%)	0 (−)	0 (−)
Dehydration	3 (10%)	1 (3%)	0 (−)	0 (−)
Adrenal insufficiency	2 (7%)	2 (7%)	0 (−)	0 (−)
Lung infection	2 (7%)	1 (3%)	0 (−)	0 (−)
Hypoxia	1 (3%)	1 (3%)	0 (−)	0 (−)
Periodontal disease	1 (3%)	1 (3%)	0 (−)	0 (−)
Stroke	1 (3%)	1 (3%)	0 (−)	0 (−)
Sepsis	1 (3%)	1 (3%)	0 (−)	0 (−)
Acute coronary syndrome	0 (−)	0 (−)	1 (8%)	1 (8%)
Renal calculi	1 (3%)	1 (3%)	0 (−)	0 (−)

Abbreviation: ALT, alanine aminotransferase.

^aOne grade 3 pneumonitis event occurred 6 months after completion of treatment.

Clinical outcomes

Among the 29 patients randomized to receive R223+P, 24 received at least three doses of R223. Of these, 5 patients discontinued R223 after three doses for PCWG2 radiographic progression at initial restaging, 3 continued after three doses without radiographic progression, 11 took a break from R223 per protocol while continuing pembrolizumab monotherapy, and 5 elected to continue R223 without a break (Supplementary Fig. S4). Twelve of 29 patients in the R223+P arm and 8 of 13 patients in the R223 arm received six R223 doses. Maximal PSA decline in study participants is depicted in the waterfall plot in Fig. 3. Three patients experienced a ≥50% PSA decline (PSA50 response) on study, all of whom received R223+P. Median rPFS was 6.1 months [95% confidence interval (CI): 2.7–11.0] for R223+P-treated patients (n = 20 radiographic progressions/29 patients) versus 5.7 months [2.6–not reached (NR)] for the R223 group (n = 4/13; Fig. 4A). Median TTF was 5.1 months (95% CI: 2.7–7.6; n = 29 events/29 patients) for R223+P-treated patients compared with 4.8 months (95% CI: 1.4–5.7; n = 11/13) for R223-treated patients (Fig. 4B). Median OS was 16.9 months (95% CI: 12.7–NR; n = 15/29) for patients treated with R223+P versus 16.0 months (95% CI: 9.0–NR; n = 7/13) for patients treated with R223 (Fig. 4C). Median time to PSA progression was 3.3 months (95% CI: 3.1–4.6; n = 25/29) and 3.3 months (95% CI: 2.8–3.6; n = 9/13) for R223+P- and R223-treated patients, respectively (Fig. 4D).

Figure 3.

Maximal PSA decline with radium-223 + pembrolizumab and radium-223 treatment. Waterfall plot of maximal % PSA decline from baseline to 8 weeks in the radium-223 plus pembrolizumab (R223+P) and radium-223 (R223) treatment arms. Each bar represents an individual patient, with blue and red bars indicating the assigned treatment (blue = R223+P, N = 29; red = R223, N = 13). Baseline is the date of prestudy measurement. The dashed line represents 50% PSA reduction. No formal hypothesis testing was performed for the assessment of PSA response. Prestudy PSAs were collected a median of 14 days (range: 4–42 days) prior to initiation of study treatment.

Figure 4.

Clinical outcomes with radium-223 + pembrolizumab and radium-223 treatment. The distributions for rPFS (A), TTF (B), OS (C), and time to PSA progression (D) in the radium-223 plus pembrolizumab (R223+P, N = 29) and radium-223 (R223, N = 13) treatment arms were estimated using the Kaplan‐-Meier method, with definitions provided in the “Statistical analysis” section. A two-sided P value of less than 0.05 was considered statistically significant.

No clinical or genomic feature clearly distinguished PSA responders versus nonresponders. All PSA responders (3/3, 100%) had pretreatment alkaline phosphatase <220 U/L (compared with 74.4% of nonresponders; Supplementary Table S5). Genomic profiling was successfully performed in 2 of the 3 responders, neither of whom had evidence of dMMR or elevated TMB. One patient randomized to receive R223+P had a pathogenic somatic alteration in MSH6 detected in tumor and circulating cell-free DNA sequencing, but did not experience a PSA response or disease stabilization (Supplementary Table S6).

Among the 26 patients with evaluable paired biopsies (n = 18, R223+P; n = 8, R223), median TTF was similar in patients who had a baseline CD4⁺FoxP3⁻ T-cell proportion, CD8⁺ T-cell proportion, and CD8⁺/CD4⁺FoxP3⁺ T-cell ratio above, versus below, median (Supplementary Fig. S5; Supplementary Table S7). Median TTF was also similar in patients with an increase, versus decrease, in CD4⁺FoxP3⁻ T cells and CD8⁺ T cells with treatment (Supplementary Fig. S6; Supplementary Table S7); median TTF was 6.7 months (95% CI: 2.7–11) in patients with an increase in CD8⁺/CD4⁺FoxP3⁺ T-cell ratio and was 4.2 months (95% CI: 2.8–6.1) in those with a decrease.

Circulating T-cell states differ in clinical responders versus nonresponders

To determine whether there were any immune state differences associated with clinical response in the R233+P arm, we performed statistical scaffold analysis comparing PSA50 responders to nonresponders at the baseline (week 1), week 5, and week 9 timepoints (Fig. 5A). Prior to treatment (baseline), responders showed significantly lower expression of Ki67⁺ on effector memory CD8 T cells, NK cells, and classical monocytes than nonresponders (Fig. 5A, left). Responders showed a significantly lower expression of PD-L1 on myeloid cells than nonresponders, which could contribute to systemic immunosuppression. By week 9, responders showed a significantly higher expression of TIM-3⁺ on B, T, and myeloid cells than nonresponders (Fig. 5A, right). Statistical scaffold plots of TIM-3 at week 9 confirmed the induction of TIM-3 across multiple cell populations at this timepoint (Fig. 5B). Taken together, these exploratory data suggest that clinical responses defined by PSA declines associated with the late induction of the immune checkpoint TIM-3 on circulating immune cells, which may serve to restrict the persistence of induced antitumor immune responses. We did not find a correlation between frequencies in the blood versus tissue for PD-L1–expressing myeloid cells and TIM-3–expressing cells (Supplementary Fig. S7), so the changes in the circulating cell populations did not necessarily reflect changes in the tumor microenvironment detectable in individual biopsy specimens (for biological reasons, or potentially related to technical or sampling issues in the latter). If we used rPFS instead of PSA50 to stratify for clinical response, we observed a similar pattern of higher Ki67⁺ immune cells at baseline and higher TIM-3 expression at week 9 in responders (Supplementary Fig. S8).

Figure 5.

Immune cell functional states differ in clinical responders compared with nonresponders. CyTOF and Statistical Scaffold analysis were performed as in Fig. 2, but now comparing responders versus nonresponders (as defined by PSA responses) at the baseline (week 1), week 5, and week 9 timepoints. A, Heat maps summarizing log₂ fold-changes resulting from statistical scaffold analysis of functional markers 4-1BB, CD44, CTLA-4, GITR, HLA-DR, ICOS, Ki-67, OX40, PD-1, PD-L1, PD-L2, TIGIT, TIM-3, and VISTA when comparing responders with nonresponders for baseline (week 1), week 5, and week 9. Each box represents a T-cell cluster that is labeled according to the nearest landmark node and ordered by cell count abundance. The color-coding represents clusters showing a significant difference (q<0.05) in log₂ fold-change, with red being significantly higher and blue being significantly lower in the nonresponders. The intensity of the color code is proportional to the log₂ fold-change and is capped at 2 and −2. B, Scaffold map indicating significant changes in TIM-3 when comparing nonresponders to responders at week 9. Node color coding is the same as discussed for Fig. 2B, and the absence of blue color coding indicates no significant decrease.

Discussion

Here, we report results of a first-in-human phase II trial of radium-223 with pembrolizumab in a biomarker-unselected population of patients with mCRPC to bone. There were no significant improvements in clinical outcomes with the addition of pembrolizumab to radium-223. Furthermore, there were no consistent therapy effects on immune infiltration in tumor biopsy specimens or convincing evidence that changes in T-cell infiltration on treatment correlated with clinical outcomes. A suggestion that participants who experienced an increase in CD8⁺/CD4⁺FoxP3⁺ T-cell ratio on treatment had slightly longer median TTF compared with those with a decrease was not statistically significant, given the limited sample size, and needs to be validated in a larger cohort. With regards to the circulating immune responses, combination treatment induced inhibitory CTLA-4 expression on CD4⁺ T cells, which could represent a compensatory induction of another immunologic checkpoint. Moreover, in clinical responders, induction of TIM-3 at later timepoints on circulating immune cells could serve to dampen treatment-induced immune responses.

Despite preclinical data suggesting benefit with the combination of radium-223 and CPI compared with radium-223 alone in a syngeneic prostate cancer mouse model (41), our study, along with a similar study of radium-223 with the anti-PD-L1 agent atezolizumab (42), observed no evidence of clinical benefit with this combination. In the radium-223 plus atezolizumab study, patients were randomized into two different staggered dosing schedules (atezolizumab or radium-223 introduced a full cycle before the other), whereas in our study, the schedule of radium-223 was modified so that participants would receive an initial three doses in combination with pembrolizumab followed by a radium break. This strategy was designed to induce an immune response, leading to T-cell trafficking to the TME through priming doses of radium-223, and to avoid the potential for continuous treatment with radium-223 to dampen evolving immune responses through ablation of T cells during the effector phase. Ultimately, none of these dosing schedules appear to lead to preferential clinical benefit from adding PD-1 blockade to radium-223.

Limitations of our study include the small sample size for both the analysis of T-cell infiltration and clinical endpoints. Evaluation of the study's primary endpoint was constrained by the well-characterized limitation of biopsy yield from bone metastases, as well as the number of patients who did not undergo the on-treatment biopsy. Despite amending the protocol after 21 patients had enrolled to require evaluable tumor tissue in pretreatment biopsy specimens, only 26 of 42 treated patients had paired biopsy tissue specimens amenable to analysis. Although underpowered and potentially biased because of exclusion of several patients with early disease progression, there was no consistent effect in the evaluable paired biopsies of radium-223 with or without pembrolizumab on CD4⁺ or CD8⁺ T-cell immune infiltration in the bone TME. Similarly, although the study was not powered for clinical endpoints, the similarity in median rPFS, TTP, and OS across both treatment arms suggested no difference in clinical efficacy.

The lack of an increase in immune cell infiltration in most R223+P participants was surprising and is in contrast with previous findings from our group using similar assays in the context of neoadjuvant treatment with the autologous cellular therapeutic sipuleucel-T prior to prostatectomy (43). In our prior studies, sipuleucel-T led to a greater than 3-fold increase in infiltrating CD4⁺FoxP3⁻ effector, and CD8⁺, T cells in radical prostatectomy tissues compared with the pretreatment biopsy, which was not seen in a control group consisting of 12 concurrent patients who did not receive any neoadjuvant treatment prior to prostatectomy (43). These discrepant results could be related to the different mechanism of action of radium-223 plus pembrolizumab compared with sipuleucel-T, sampling issues related to core bone biopsies versus resection of the entire prostate gland, different prior therapies and disease states (mCRPC vs. primary hormone-sensitive disease), and differences in tumor content and the TME between the bone and prostate gland. Sipuleucel-T is generated through stimulation of a patient's dendritic cells to target prostatic acid phosphatase, whereas pembrolizumab does not involve specific stimulation against a prostate cancer antigen. Clinical trials of neoadjuvant pembrolizumab (NCT04009967, NCT03753243) have not yet reported if this agent leads to increase in infiltrating T cells in the prostate.

The bone has also been reported to represent an immunosuppressive microenvironment that is mediated through modulation of multiple signaling pathways (44), including downregulation of Wnt (45) and JAK/STAT (46) signaling and upregulation of TGFβ signaling (47). Single-cell analyses from bone metastases in patients with mCRPC (48) reveals the appearance of prominent populations of myeloid cells, including tumor inflammatory monocytes and tumor-associated macrophages (TAM) with high expression of the chemokine CCL20, which contribute to immunosuppression through increasing infiltration by activated Tregs and exhausted CTLs in the TME. Bone biopsy specimens in patients treated with the combination of anti-CTLA-4 (tremelimumab) and anti-PD-L1 (durvalumab) reveal upregulation in myeloid subset transcriptional signatures and myeloid cell markers, suggesting this cell type as a potential mediator of resistance to this regimen (49). There was not sufficient tissue to quantify myeloid cell subsets in the biopsy specimens from our study, but these subsets would be important to analyze in future studies of immunotherapeutics in bone-predominant mCRPC.

Indeed, responses to pembrolizumab have been less frequent in patients with predominantly bone metastases in clinical trials. For example, in KEYNOTE-199 (5), only 1 of 59 patients in cohort 3 (bone-predominant disease) had a PSA50 response compared with 13 of 184 patients across cohorts 1 (PD-L1⁺) and 2 (PD-L1-negative) with measurable soft-tissue disease. PSA50 responses were also less common in patients with nonmeasurable (generally bone-predominant) disease compared with those with measurable soft-tissue disease in studies of pembrolizumab plus olaparib (50) and pembrolizumab plus enzalutamide (51, 52). In our study, 3 of 29 patients who received radium-223 with pembrolizumab had PSA50 response (compared with 0 of 13 with radium-223 alone), but it is not known whether this represents monotherapy activity of one of the two agents versus additive or synergistic activity in this small subset. Despite PSA responses only seen in the R223+P arm, clinical endpoints of rPFS, TTF, and OS were similar in the two arms, suggesting that any clinical benefit was short-lived, and indeed, we observed that the duration of PSA response (from time of initial ≥50% PSA reduction to PSA progression by PCWG2 criteria) was relatively short with 2.5, 6.0, and 6.4 months in the 3 responders.

Our findings are also in consonance with emerging data from other studies suggesting limited evidence for synergistic activity of radiotherapy with checkpoint inhibitors. In addition to effects that may stimulate an immune response such as an abscopal effect as detailed in the introduction, radiation may also have immunosuppressive effects, such as through induction of CSF1R (53) and STING/type I IFN pathways (54) for recruitment of immunosuppressive tumor-infiltrating myeloid cells. Despite encouraging results from an early phase II trial in non–small cell lung cancer (55), in subsequent trials, responses in nonirradiated lesions to single-site radiation in combination with checkpoint immunotherapy in head and neck squamous cell carcinoma (56), non–small cell lung carcinoma (57), and renal cell carcinoma (58, 59) are similar in frequency to what would be expected with CPI alone. Some promising results beyond what would be expected with CPI alone have been reported with this approach in prostate cancer (60). A phase III trial of the CTLA-4 antibody ipilimumab in post-docetaxel mCRPC after a single dose of radiotherapy to one or more bone metastases did not demonstrate statistically significant improvement in OS with ipilimumab compared with placebo (61), but OS rates at 3, 4, and 5 years were higher with ipilimumab (62). These findings suggest that some patients may benefit from CPI after bone radiation, but the degree to which the bone radiation contributes to clinical outcomes is unclear, and the optimal therapeutic strategy and appropriate patient selection have not yet been determined.

Novel immunotherapeutic approaches to prostate cancer are under investigation, but optimal strategies remain elusive (63, 64). A number of trials combining radiotherapy with immunotherapeutics in prostate cancer are in progress (7, 8), many of these testing the combination with the addition of a vaccine or another pharmacologic agent. It is certainly possible that greater clinical benefit from radium-223 with pembrolizumab could be achieved by adding another immunomodulatory agent to increase immune stimulation or decrease immune suppression related to the bone microenvironment. Although radium-223 targets bony metastases as a calcium mimetic, it is not a molecularly targeted therapy. Combination with molecularly targeted agents that exploit specific therapeutic vulnerabilities such as the DNA damage repair response (41), or counteract specific molecular mechanisms of immunosuppression, may improve clinical efficacy. The latter could include antagonism of pre-existing mechanisms of immunosuppression within an individual patient's tumor (65), or of immune checkpoints that are upregulated in response to treatment.

Another therapeutic strategy under study is combining pembrolizumab with the prostate-specific membrane antigen (PSMA)-targeting radiopharmaceuticals ¹⁷⁷Lu-PSMA-617 (NCT03805594, NCT03658447) and ²²⁵Ac-J591 (NCT04946370), as these agents lead to more frequent PSA and radiographic responses than radium-223, and the greater tumoricidal activity may further enhance immune stimulation. ¹⁷⁷Lu is a beta particle-emitting radiopharmaceutical, whereas ²²⁵Ac is an alpha-emitter. Although alpha particle radiation can induce immunogenic cell death (16) and has higher energy and shorter path length compared with beta particle radiation, relative effects on the TME of mCRPC bone metastases in patients to predict differential clinical activity in combination with CPI are not known. Which of these strategies is most likely to lead to an antitumor response in an individual patient is an active area of investigation. Genetic and molecular biomarkers, as well as markers of immune cell activation both in the tumor milieu and in the circulation, may help identify patients most likely to respond to these combinatorial strategies.

Supplementary Material

Supplemental Figure 1

Radium-223 was administered as 55 kBq/kg q4weeks and pembrolizumab as 200 mg q3weeks. Biopsies were performed pre-treatment and after 8 weeks of treatment.

Supplemental Figure 2

Consort diagram for status of trial. Progression (PD) includes protocol defined progression and clinical/symptomatic progression per physician.

Supplemental Figure 3

Swimmer’s plot of treatments received. Number of radium-223 and pembrolizumab doses received are listed to the left and right of the bars, respectively. The brackets “[” and “]” represent the planned break period from radium-223 dosing with “[” representing cycle 3 day 28 and “]” representing cycle 4 day 1 when the interval between these was greater than 28 days. Patients could resume radium treatment at radiographic progression by PCWG2 criteria (“△”) if there were no new visceral metastases, but two patients (bars 7 and 13 below) resumed without confirmed radiographic progression.

Supplemental Figure 4

Changes in A. CD4+FoxP3- effector cells and B. ratio of CD8+ to CD4+/FoxP3+ (Treg) cells from baseline to 8 weeks on treatment

Supplemental Figure 5

Kaplan-Meier estimates of time to treatment failure (TTF), radiographic progression-free survival (rPFS), and overall survival (OS) by A. baseline CD4+FoxP3- (effector) proportion above (high) vs. below (low) median. B. baseline CD8+ proportion above vs. below median. C. baseline CD8+ / CD4+FoxP3+ (Treg) ratio above vs. below median. N.B. Patients were event-free for 2 months as they need to have on treatment biopsy at week 8 of treatment.

Supplemental Figure 6

Kaplan-Meier estimates of time to treatment failure (TTF), radiographic progression-free survival (rPFS), and overall survival (OS) by A. CD4+FoxP3- (effector) proportion increase vs. decrease from baseline to 8-week biopsy. B. CD8+ increase vs. decrease. C. CD8+ / CD4+FoxP3+ (Treg) ratio increase vs. decrease. N.B. Patients were event-free for 2 months as they need to have on treatment biopsy at week 8 of treatment.

Supplemental Figure 7

Immunofluorescence staining of PD-L1+ CD68 myeloid cells in the bone biopsies. A. RNAscope staining of DAPI (cyan), CD68 (orange), PD-L1 (magenta), Tim3 (green), TIGIT (white) and overlay of bone marrow metastasis. CD68+ cells that co-express PD-L1 are indicated with arrow. Tim3+ cells are indicated with pointer. Scale bar, 20 μm. B. Scatterplot of PDL1+ CD68+ cell frequencies of tissue imaging data and PDL1+ CD14+ CD3-CD19- myeloid cell frequencies of CyTOF data from blood samples taken at the same timepoint (baseline, n=10 patients). R2 value is based on Pearson correlation. C. Scatterplot of TIM3+ CD68- cell frequencies on tissue imaging and TIM3+ CD3+ T cell frequencies of CyTOF data from blood samples taken at the same timepoint (baseline, n=10 patients). R2 value is based on Pearson correlation.

Supplemental Figure 8

Immune cell functional states differ in clinical responders compared to non-responders. (A) Heat maps summarizing log2 fold changes resulting from statistical scaffold analysis of functional markers 4-1BB, CD44, CTLA-4, GITR, HLA-DR, ICOS, Ki-67, OX40, PD-1, PD-L1, PD-L2, TIGIT, TIM3 and VISTA are shown for Week 1, Week 5, and Week 9, when comparing responders to non-responders as defined by rPFS.

Supplemental Table 1

Summary of biopsy yield

Supplemental Table 2

Maximum grade adverse events, regardless of attribution. AEs are summarized as n (%) for each maximum grade.

Supplemental Table 3

Proportion of patients who experienced any or given category of symptomatic skeletal events by the treatment arms

Supplemental Table 4

Clinical characteristics at time of randomization of PSA responders vs. non-responders

Supplemental Table 5

Summary of genetic findings from targeted sequencing of ctDNA and whole exome sequencing (WES) from bone biopsy specimen.

Supplemental Table 6

Time to event (months) clinical endpoints by baseline and change in CD4+/FoxP3- (effector) proportion, CD8+ proportion, CD8+ / CD4+FoxP3- (Treg) ratio

Supplementary File 1

Supplementary File 1

Supplementary File 2

Supplementary File 2

Supplementary File 3

Supplementary File 3

Authors' Disclosures

A.D. Choudhury reports grants from Bayer and Merck during the conduct of the study; grants and personal fees from Bayer, Pfizer, Eli Lilly; personal fees from AstraZeneca, Astellas, Blue Earth, Janssen, Sanofi Aventis, Tolmar, and Lantheus outside the submitted work. A. Tripathi reports personal fees from Foundation medicine, Pfizer, Genzyme, Exelixis, Deka, Seattle Genetics; grants from Aravive Inc and Aadi Biosciences; grants and personal fees from EMD Serono outside the submitted work. B. Maynard reports personal fees from Deciphera Pharmaceuticals outside the submitted work. A.F. Pace reports personal fees from Astellas and SeaGen outside the submitted work. E.M. Van Allen reports personal fees from Tango Therapeutics, Genome Medical, Genomic Life, Monte Rosa Therapeutics, Manifold Bio, Illumina, Enara Bio, Foaley & Hoag, Riva Therapeutics, and Serinus Bio; grants and personal fees from Novartis and Janssen; grants from BMS and Sanofi outside the submitted work; in addition, E.M. Van Allen has a patent for Institutional patents filed on chromatin mutations and immunotherapy response, and methods for clinical interpretation pending and issued. B.A. McGregor reports grants and personal fees from BMS, Exelixis, Pfizer, SeaGen, and Gliead; personal fees from Eisai outside the submitted work. R.S. Bhatt reports other support from Bristol Myers Squibb outside the submitted work. C.J. Sweeney reports grants and personal fees from Bayer and MSD during the conduct of the study. H.A. Jacene reports other support from Blue Earth Diagnostics, Monrol, ITM, and Spectrum Dynamics outside the submitted work. L.C. Harshman reports grants from Merck and Bayer during the conduct of the study; and employment and stock options at Surface Oncology. L. Fong reports grants and personal fees from Merck during the conduct of the study; grants and personal fees from Abbvie, BMS, and Roche/Genentech; grants from Amgen, Janssen, Nektar outside the submitted work. No disclosures were reported by the other authors.

Authors' Contributions

A.D. Choudhury: Resources, data curation, formal analysis, supervision, investigation, visualization, writing–original draft, writing–review and editing. L. Kwak: Formal analysis, visualization, methodology, writing–review and editing. A. Cheung: Formal analysis, investigation, visualization, writing–review and editing. K.M. Allaire: Formal analysis, investigation, visualization, writing–review and editing. J. Marquez Garcia: Formal analysis, investigation, visualization, writing–review and editing. D.D. Yang: Formal analysis, investigation, writing–review and editing. A. Tripathi: Data curation, formal analysis, investigation, project administration, writing–review and editing. J.M. Kilar: Data curation, investigation, project administration, writing–review and editing. M. Flynn: Data curation, project administration, writing–review and editing. B. Maynard: Investigation, project administration, writing–review and editing. R. Reichel: Formal analysis, investigation, writing–review and editing. A.F. Pace: Data curation, investigation, project administration, writing–review and editing. B.K. Chen: Formal analysis, investigation, writing–review and editing. E.M. Van Allen: Investigation, writing–review and editing. K. Kilbridge: Investigation, writing–review and editing. X.X. Wei: Investigation, writing–review and editing. B.A. McGregor: Investigation, writing–review and editing. M.M. Pomerantz: Investigation, writing–review and editing. R.S. Bhatt: Investigation, writing–review and editing. C.J. Sweeney: Investigation, writing–review and editing. G.J. Bubley: Investigation, writing–review and editing. H.A. Jacene: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. M.-E. Taplin: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. F.W. Huang: Formal analysis, investigation, methodology, writing–review and editing. L.C. Harshman: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. L. Fong: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.