Neoadjuvant Fc-enhanced anti-CTLA-4 targets Tregs to augment androgen deprivation in high-risk prostate cancer: A randomized phase I trial

Summary

Despite high rates of post-surgical recurrence in men with high-risk localized prostate cancer (PCa), there is currently no role for neoadjuvant therapy. Tumor infiltrating regulatory T cells (TI-Tregs) limit the antitumor effects of presurgical androgen deprivation therapy (ADT). We present a neoadjuvant clinical trial testing whether an afucosylated anti-CTLA-4 antibody (BMS-986218) with ADT is safe, feasible, and reduces TI-Treg frequencies. This single-center, two-arm, open-label study randomizes 24 men with high-risk localized PCa to ADT with or without BMS-986218 prior to radical prostatectomy. Treatment is well tolerated and feasible. Mechanistic studies reveal reductions in TI-Treg frequencies correlate with CD16a/FCGR3A on tumor macrophages, dendritic cell (DC) modulation, and augmented T cell priming following BMS-986218 treatment. Depth of Treg inhibition and increased DC frequencies are associated with improved clinical outcomes. Overall, this study supports the feasibility and biological activity of neoadjuvant ADT + Fc-enhanced anti-CTLA-4 in high-risk PCa. Trial is registered at clinicaltrials.gov (NCT04301414).

Graphical abstract

Highlights

• •Neoadjuvant ADT + Fc-enhanced αCTLA-4 is well tolerated in localized prostate cancer
• •Fc-enhanced αCTLA-4 exposure associates with reduced tumor Treg frequencies
• •Depth of Treg frequency reduction associates with response to Fc-enhanced αCTLA-4
• •DC activation and T cell priming are implicated in response to Fc-enhanced αCTLA-4

Ager et al. investigate immunological effects of a next-generation Fc-enhanced αCTLA-4 antibody with hormonal therapy in a phase I neoadjuvant clinical trial in high-risk localized prostate cancer, leveraging single-cell multi-omics approaches. Their findings establish clinical feasibility, effects of Fc-enhanced αCTLA-4 on regulatory T cells, and putative response biomarkers.

Introduction

To date, immune checkpoint blockade immunotherapy (ICB) has proven largely ineffective for patients with prostate cancer (PCa).^1,2,3 Agents targeting the PD-1/PD-L1 pathway alone or in combination with androgen receptor inhibitors,^4,5,6 chemotherapy,⁷ or PARP inhibition⁸ have failed in phase III trials.⁹ Similarly, CTLA-4 blockade in a chemo-naïve population¹⁰ or post-docetaxel in combination with bone-targeted radiation¹¹ did not meet pre-specified endpoints. However, long-term follow-up indicates a potential survival advantage of ipilimumab versus placebo.¹¹ Furthermore, a phase II study of combination ipilimumab and nivolumab in metastatic patients pre- or post-chemotherapy showed favorable survival compared to historical controls.¹² These findings suggest that CTLA-4 blockade may possess clinical activity in PCa that warrants further investigation, particularly if concerns regarding dose-dependent toxicity can be addressed.

One hypothesis explaining the relatively low efficacy of ICB in PCa is that baseline T cell infiltration is insufficient for ICB to elicit robust T-cell-mediated antitumor immunity.¹ In addition, most immunotherapy agents have first been tested in advanced disease settings where the PCa tumor microenvironment (TME) is generally more immunosuppressive, thereby potentially limiting therapeutic efficacy. Interestingly, in localized disease, standard-of-care androgen deprivation therapy (ADT) increases CD8 T cell infiltration into primary PCa. This finding has been replicated across multiple patient cohorts and murine models.^13,14,15 However, a counter-regulatory influx of immunosuppressive CD4⁺FoxP3⁺ regulatory T cells (Tregs) inhibits the inflammatory effects of ADT, implicating adaptive Treg resistance as a critical inhibitory mechanism.^13,14 As such, depletion of tumor-infiltrating Tregs (TI-Tregs) by anti-CTLA-4 antibodies with enhanced antibody-dependent cellular cytotoxicity or phagocytosis (ADCC/P) extends depth and duration of response to ADT in murine models, whereas non-depleting anti-CTLA-4 antibodies or PD-1 blockade in the same setting is ineffective.¹³ These Fc-enhanced agents have also been hypothesized to have an improved therapeutic index, enabling dosing strategies that deliver improved activity with reduced toxicity.¹⁶ Therefore, we reasoned that inhibiting TI-Treg number and/or function in combination with ADT may effectively potentiate ICB efficacy in PCa.

Methods for depleting TI-Tregs have been studied thoroughly in pre-clinical models but have not been consistently validated in the clinic. Antibodies targeting cell surface proteins enriched on TI-Tregs such as CTLA-4, CD25, CCR4, TIGIT, and CCR8 have either failed to show Treg depletion activity and/or efficacy in clinical trials or remain under investigation.^17,18,19,20 We thus sought to further investigate anti-CTLA-4 in this context, given its potential clinical activity in PCa¹¹ and the fact that the first-generation anti-CTLA-4 antibodies ipilimumab and tremelimumab were not optimized for Treg depletion. Tremelimumab is a human immunoglobulin G2 (IgG2) antibody with relatively low affinity for the FcγRs such as CD16a (FCGR3A) that mediate ADCC/P, while ipilimumab is a human IgG1 isotype that can elicit ADCC/P. However, existing clinical data suggest ipilimumab is a weak depleter, with evidence of Treg depletion restricted to patients with a high-affinity germline polymorphic variant of FCGR3A (V158F).^20,21 To overcome this limitation, antibodies targeting CTLA-4 bearing non-fucosylated Fc regions have been developed to increase binding affinity for CD16a/FCGRIIIA, resulting in enhanced ADCP/ADCC.²² However, it remains unclear whether non-fucosylated anti-CTLA-4 antibodies are more effective in depleting TI-Tregs in patients. Furthermore, whether these next-generation Fc-enhanced antibodies can overcome adaptive Treg resistance following ADT is unknown.

To address this challenge, we designed a randomized neoadjuvant clinical trial to evaluate the feasibility and immunological activity of non-fucosylated ipilimumab (BMS-986218; hereafter anti-CTLA4-NF) with ADT versus ADT alone prior to surgery in men with high-risk localized prostate cancer (NeoRED-P; NCT04301414). This window-of-opportunity trial enabled exploratory interrogation of the immunological mechanisms of response to anti-CTLA4-NF through orthogonal immunoprofiling by single-cell RNA sequencing (scRNA-seq), mass cytometry (CyTOF), and spatial proteomics (multiplex immunofluorescence). We also leveraged the RNA-seq-based pipeline PISCES to perform robust protein activity level assessment of tumor microenvironment subpopulations and of their drug-mediated modulation, thus bridging the transcriptomic and proteomic datasets. Pre-clinical validation studies in the syngeneic MycCaP PCa model were conducted to independently validate our exploratory correlative analyses. Herein, we report on the feasibility of ADT plus anti-CTLA4-NF in the neoadjuvant setting and provide a first-in-human examination of mechanisms underlying the biological activity and clinical response to Fc-engineered non-fucosylated anti-CTLA4 antibody in high-risk localized PCa.

Results

Safety and feasibility of neoadjuvant ADT + anti-CTLA4-NF

Between February 26, 2020, and November 15, 2022, 24 patients were enrolled at Columbia University Irving Medical Center. For correlative studies, samples from a cohort of 12 stage- and grade-matched treatment-naïve patients were collected as untreated controls. The first four trial patients received ADT with degarelix acetate + anti-CTLA4-NF as a safety lead-in, after which 20 patients were randomized 1:1 to ADT-only or ADT + anti-CTLA4-NF arms (Figure 1A). One patient in the ADT + anti-CTLA4-NF arm did not receive surgery due to operating room closure during the COVID-19 pandemic, and one patient in the ADT + anti-CTLA4-NF arm received additional adjuvant treatment roughly 3 months post-prostatectomy, so was excluded from post-surgical secondary outcomes analyses (Figure 1B). Baseline characteristics of the trial patients are summarized in Table 1. The arms were well balanced with respect to age, race and ethnicity, pre-treatment serum PSA or testosterone, rate of nodal involvement, and rate of surgical margin involvement. Patients in the ADT + anti-CTLA4-NF arm trended toward a higher frequency of Gleason Grade Group 5 disease (standardized mean differences for all baseline characteristics is provided in Table S1). Tumor exome and mutational profiling of all patient-derived samples showed 12 patients with AR-V7 splice variants (38%), 8 patients exhibiting TMRPSS2:ERG fusion events (25%), 4 patients bearing mutated TP53 (13%), 3 patients with copy-number loss or mutations in PTEN (9%), and 2 patients each with alterations in APC and CTNNB1 (6%; Figure S1A). Of note, one patient in the ADT + anti-CTLA4-NF arm was found to exhibit high tumor mutational burden (TMB-H; 26 mutations/Mb), associated with a germline mutation in MSH2.

Figure 1.

NeoRED-P trial design and consort diagram

(A) Schematic of NeoRED-P clinical trial design.

(B) NeoRED-P trial consort diagram.

Table 1.

Patient baseline demographics and disease characteristics

Characteristic	ADT N = 10a	ADT + αCTLA-4-NF N = 14a
Age
Median (Range)	66.0 (52.0, 72.0)	65.5 (52.0, 77.0)
Race, n(%)
White	6 (60%)	9 (64%)
Black	2 (20%)	2 (14%)
Other	1 (10%)	3 (21%)
Asian	1 (10%)	0 (0%)
Ethnicity, n (%)
Not Hispanic	7 (70%)	12 (86%)
Hispanic	3 (30%)	2 (14%)
PSAb
Median (Range)	18.44 (2.37, 94.40)	9.08 (4.57, 364.70)
Testosterone
Median (Range)	420.30 (153.90, 553.30)	399.50 (273.80, 783.30)
Grade group, n (%)
3	5 (50%)	2 (14%)
4	3 (30%)	5 (36%)
5	2 (20%)	7 (50%)
T stage, n (%)
T2	5 (50%)	4 (29%)
T3a	2 (20%)	5 (36%)
T3b	3 (30%)	5 (36%)
Nodal involvement, n (%)	2 (20%)	3 (21%)
Surgical margin involvement, n (%)	3 (30%)	3 (21%)
2-year recurrence-free probabilityc
Median (Range)	69.50 (8.00, 88.00)	55.00 (4.00, 82.00)
VHR, n (%)d	8 (80%)	12 (86%)

^aMedian (Range) for continuous variables; N (%) for categorical variables.

^bProstate-specific antigen (PSA), ng/mL.

^cCalculated using Memorial Sloan Kettering Cancer Center’s (MSKCC) post-radical prostatectomy nomogram.

^dVery high-risk (VHR) group defined by National Comprehensive Cancer Network (NCCN) criteria as Gleason ≥8 or PSA >20 ng/mL or clinical T3 stage or higher.

Treatment-related adverse events (TRAEs) of any grade were observed in 80% of ADT-only patients and 71% of ADT + anti-CTLA4-NF patients and were generally attributed to ADT, with injection site reactions being the most common events in both arms (Table 2). Three grade 1–2 gastrointestinal events were observed in patients receiving anti-CTLA4-NF. Grade ≥3 TRAEs were rare; a single patient in the ADT + anti-CTLA4-NF arm exhibited a grade 3 asymptomatic serum lipase elevation that resolved without intervention and did not result in treatment discontinuation. No pre-specified interim safety boundaries were crossed, and no unexpected surgical complications were observed. Overall, the study met its primary endpoint of safety and feasibility, supporting a tolerable safety profile for the addition of anti-CTLA4-NF to ADT prior to radical prostatectomy in patients with high-risk localized PCa.

Table 2.

Treatment-related adverse events

	ADT N = 10		ADT + αCTLA-4-NF N = 14
Treatment-related adverse events (TRAEs)a	All grades	Grade ≥3	All grades	Grade ≥3
Any treatment-related adverse event (TRAE)	8 (80)	–	10 (71)	1 (7.1)
Gastrointestinal disorders	–	–	3 (21)	–
Colitis	–	–	1 (7.1)	–
Diarrhea	–	–	1 (7.1)	–
Elevated transaminases	–	–	1 (7.1)	–
General disorders and administration site conditions	7 (70)	–	8 (57)	–
Chills	1 (10)	–	–	–
Fatigue	–	–	4 (29)	–
Hot flashes	5 (50)	–	6 (43)	–
Injection site reaction	6 (60)	–	7 (50)	–
Investigations	–	–	–	1 (7.1)
Lipase increased	–	–	–	1 (7.1)
Musculoskeletal and connective tissue disorders	–	–	1 (7.1)	–
Bone pain	–	–	1 (7.1)	–
Nervous system disorders	–	–	1 (7.1)	–
Headache	–	–	1 (7.1)	–
Reproductive system disorders	2 (20)	–	–	–
Erectile dysfunction	2 (20)	–	–	–
Skin and subcutaneous tissue disorders	–	–	1 (7.1)	–
Pruritus	–	–	1 (7.1)	–

^aTreatment-related adverse events (TRAEs) graded according to NCI CTCAE v5.0.

Anti-CTLA4-NF counteracts ADT-induced TI-Treg expansion

This study was powered to address its key secondary biological endpoint—to test the hypothesis that anti-CTLA4-NF reduces TI-Treg density in ADT-treated PCa tumors. For this, we employed immunofluorescence staining to quantify the frequency of tumor-infiltrating CD4⁺FoxP3⁺ Tregs among all nucleated cells in whole-tissue FFPE slides obtained at prostatectomy from all evaluable study participants and matched untreated controls (see representative staining images in Figure 2A). For this analysis, we censored one patient with MSI^hi status, given the confounding association between MSI^hi status, high TMB, and increased overall tumor T cell infiltration relative to MSI^WT patients,²³ and we censored all patients who received ADT + anti-CTLA4-NF without randomization during the safety lead-in. In line with our previous data, we found ADT significantly increases TI-Treg frequencies in the PCa tumor parenchyma (p = 0.002; Figure 2B). In patients treated concurrently with ADT and anti-CTLA4-NF, TI-Treg density was significantly reduced as compared to ADT alone (p = 0.031; Figure 2B). Thus, the trial met its key secondary biological endpoint of TI-Treg density reduction, supporting the hypothesis that anti-CTLA4-NF may deplete TI-Tregs in PCa patients.

Figure 2.

Secondary biological and clinical endpoint measures of the NeoRED-P trial

(A) Representative images from whole-slide immunofluorescence data for CD4 and FoxP3 with DAPI nuclear staining. Inset gray box shows a representative CD4⁺FoxP3⁺ cell at 2.5× magnification relative to the original image. Scale bars, 40 μm.

(B) Bar plot representing Treg frequency as measured by percentage of DAPI⁺CD4⁺FoxP3⁺ Tregs of all DAPI⁺ cells in tumor regions iterated by treatment group in the immunofluorescence dataset. Data are represented as mean ± 95% confidence intervals. Untreated, n = 12; ADT, n = 7; ADT + anti-CTLA4-NF, n = 8.

(C) Kaplan-Meier curve of PSA recurrence-free survival in patients treated with neoadjuvant ADT; 95% confidence intervals shown in the shaded area. There were two recurrence events in this arm.

(D) Kaplan-Meier curve of PSA recurrence-free survival in patients treated with neoadjuvant ADT + anti-CTLA4-NF; 95% confidence intervals shown in the shaded area. There were four recurrence events in this arm.

(E) Percent change in serum PSA from pre-treatment to time of surgery in ADT arm patients. Dotted line denotes 50% decline in PSA (PSA50).

(F) Percent change in serum PSA from pre-treatment to time of surgery in ADT + anti-CTLA4-NF arm patients.

This study was not powered to compare efficacy between treatment arms. However, exploratory secondary clinical endpoints of the trial included pathologic complete response rate, PSA response rate at time of surgery, and rate of undetectable PSA at 12 months. No complete pathologic responses were observed in either arm. In the ADT-only arm, 80% of patients had an undetectable PSA at 12 months, with a median recurrence-free survival of 21.7 months (95% confidence interval [CI]: 2.2–24.9 months) (Table 3). In the ADT + anti-CTLA4-NF arm, 75% had an undetectable PSA at 12 months, with a median recurrence-free survival of 23.3 months (95% CI: 7.0–26.6 months) (Table 3). The overall recurrence rate in patients treated with ADT + anti-CTLA4-NF was 33% (4 of 12 evaluable patients) and 20% (2 of 10 patients) in the ADT-alone arm (Table 3; Figures 2C and 2D). Patients in the ADT-only arm had a PSA50 response rate prior to surgery of 33%, while patients in the ADT + anti-CTLA4-NF arm had a PSA50 response rate of 54% (Table 3; Figures 2E and 2F). All patients recovered testosterone to baseline levels within 6 months of prostatectomy (Figure S2). In total, the trial met its key primary and secondary study endpoints of safety, feasibility, and biological activity of anti-CTLA4-NF. This supports further clinical investigation of Fc-engineered anti-CTLA-4 with ADT prior to surgery in high-risk localized PCa.

Table 3.

Secondary outcomes

Outcome	ADT	95% CI	ADT + αCTLA-4-NF	95% CI	p Value
PSA response rate, n (%)1	3 (33%)	9.0%–69%	7 (54%)	26%–80%	0.415
Undetectable PSA at 12 months, n (%)2	8 (80%)	44%–96%	9 (75%)	43%–93%	>0.999
Time to PSA recurrence (months), median (range)3	21.7 (2.2, 24.9)	–	23.3 (7.0, 26.6)	–	0.497
PSA recurrence, n (%)3	2 (20%)	3.5%–56%	4 (33%)	11%–65%	0.646

Abbreviation: CI, confidence interval.

¹PSA response rate defined as ≥50% decrease in PSA prior to prostatectomy.

²Undetectable PSA defined as <0.1 ng/mL at 12 months after prostatectomy.

³PSA recurrence defined as PSA ≥0.2 ng/mL with confirmatory rise ≥2 weeks apart.

Anti-CTLA4-NF targets TI-Tregs through myeloid FcγR engagement

As an exploratory analysis leveraging tissues from this trial, we sought to investigate mechanisms underlying TI-Treg frequency reduction and putative biomarkers of clinical response to anti-CTLA4-NF, employing a single-cell multi-omic immune profiling approach via paired scRNA-seq (using the 10× Genomics droplet-based platform) and CyTOF. These studies utilized independent, freshly processed prostatectomy specimens from trial patients treated with ADT + anti-CTLA4-NF, ADT alone, as well as an additional contemporaneously collected untreated control cohort of stage- and grade-matched patients undergoing radical prostatectomy without neoadjuvant therapy (Figure 3A). Of the 36 patients assessed, 26 had scRNA-seq data of sufficient quality and 25 had CyTOF data of sufficient quality to be included in further analyses (see Figure S3 for a biospecimen CONSORT diagram; Figure S4A and Methods for QC criteria; Figure 3B for a representative schematic). For scRNA-seq data analysis, we utilized the PISCES pipeline incorporating unbiased Louvain-based clustering and metaVIPER (virtual inference of protein activity by enriched regulon analysis²⁴) to quantitatively infer activity of approximately 6,500 regulatory proteins per cell.²⁵ Based on CITE-seq analysis and experimental validation, this approach has been shown to outperform traditional gene-expression-based methods, wherein RNA dropout limits detection of key lineage- and function-determining transcripts and reduces cluster resolution (see Methods for more detail and prior studies^25,26,27 for pipeline validation). Integrating cells from all patients, we stratified cell lineage metaclusters by PISCES, cross-validated cluster identity at the gene expression level by parallel SingleR²⁸ and supervised annotation, and verified tumor cell cluster identity by aberrant copy-number analysis using inferCNV^29,30 (Figures 3C, 3D, and S4B). Within a T/NK lineage metacluster, we clearly resolved a Treg cluster as validated by enrichment of established Treg transcriptional signature scores, inferred activity of Treg lineage marker proteins by VIPER, and SingleR annotation (Figures S5A–S5C). On the CyTOF dataset, we performed semi-supervised FlowSOM³¹ clustering followed by supervised UMAP embedding on all live CD45⁺ events and clearly resolved multiple immune populations expressing known lineage-defining protein markers, including a CD4⁺FoxP3⁺CD25⁺CD127^lo/− Treg cluster (Figures 3E, 3F, S5D, and S5E). Confirming the fidelity of our orthogonal cell clustering and annotation approaches for both datasets, we observed correlation between immune lineage frequencies on a per-patient basis by scRNA-seq versus CyTOF and similar trends in immune lineage proportions between treatment groups between platforms (Figures 3G, 3H, and S5F–S5H). Natural killer (NK) cells were a notable exception that were not significantly correlated across platforms. Though correlated, neutrophils were underrepresented in the scRNA-seq data relative to CyTOF, as it is common in droplet-based scRNA-seq methodologies. Both single-cell platforms mirrored a reduction in TI-Treg frequency in patients receiving ADT + anti-CTLA4-NF as compared to ADT alone; however, neither measure reached statistical significance, as the failure to successfully capture scRNA-seq and CyTOF data from each patient limited the statistical power for this comparison (Figures S5I and S5J). We also observed CTLA-4 expression at the protein level was exquisitely confined to the Treg compartment, further supporting the notion that putative antibody-dependent depletion due to anti-CTLA4-NF would be restricted to Tregs (Figure S5K).

Figure 3.

Single-cell multi-omic interrogation of mechanisms underlying TI-Treg modulation and clinical response to anti-CTLA4-NF

(A) Schematic of immune correlates performed.

(B) Summary of data collection and QC passing at a per-patient level for CyTOF, scRNA-seq, and immunofluorescence datasets. Patients are additionally annotated according to presence or absence of PSA recurrence within 2 years post-surgery.

(C) UMAP plot of single-cell RNA-seq data from all patients, clustered using the PICSES analytical pipeline.

(D) Bubble plot of representative gene expression for each cluster identified in (C).

(E) Supervised UMAP plot of CyTOF data from all patients, clustered using FlowSOM.

(F) Heatmap of lineage-defining protein expression across all clusters. Data represent geometric MFI for each marker normalized to the minimum and maximum values across all clusters.

(G) Heatmap representing Log₂ fold change of scRNA-seq-defined cluster frequencies in indicated treatment group relative to the untreated group.

(H) Heatmap representing Log₂ fold change of CyTOF-defined immune cluster frequencies in indicated treatment group relative to the untreated group.

(I) UMAP plot representing subclustering of myeloid cells in the scRNA-seq dataset utilizing the PISCES pipeline.

(J) Visualization of representative myeloid-lineage-defining gene expression overlayed on the parent NeoRED tumor myeloid scRNA-seq UMAP plot to validate cluster annotations.

(K) Bubble plot validating expression of additional myeloid cluster defining genes.

(L) Visualization of VIPER-inferred protein activity of select Fcγ receptors across myeloid clusters in UMAP space.

(M) Violin plot of FCGR3A/CD16a inferred protein activity on a per cell level stratified by myeloid cluster.

(N) Correlation between Treg frequency and average inferred protein activity of FCGR3A/CD16a across all myeloid cells on a per patient basis, stratified by treatment group. ADT, n = 8; ADT + anti-CTLA4-NF, n = 11.

(O) Correlation between Treg frequency and average inferred protein activity of FCGR3A/CD16a across TREM2⁺ TAM cluster cells on a per patient basis, stratified by treatment group. Pearson correlations were utilized to evaluate statistical significance of correlations.

(P) Bubble plot representing relative expression of phenotypic markers in Tregs from ADT + anti-CTLA4-NF group versus Tregs in the ADT only group, as calculated by log₂ transformed ratio of average geometric MI of each marker on Tregs from ADT + anti-CTLA4-NF versus ADT groups. Bubble size represents relative gMI, while color intensity represents percent expression of indicated markers across all Tregs in the dataset. Red color indicates higher relative gMI in Tregs from patients treated with ADT + anti-CTLA4-NF, while blue color indicates higher relative gMI in patients treated with ADT only.

(Q) Violin plot representing frequency of Tregs at time of surgery in the scRNA-seq dataset stratified by 2-year PSA recurrence status. No recurrence, n = 7; recurrence, n = 4. Two-tailed Welch’s t test was performed to evaluate statistical significance. Solid lines denote group median, and dashed lines denote quartiles.

(R) Kaplan-Meier curve representing time-to-PSA recurrence in ADT + anti-CTLA4-NF-treated patients stratified by Treg frequency in the scRNA-seq data; 95% confidence intervals shown in the shaded areas. There were three recurrence events in Treg-high patients and one recurrence event in Treg-low patients. Log rank test was performed to evaluate statistical significance. Unless otherwise stated, two-tailed Welch’s t test was used to assess statistical significance.

In pre-clinical models, antibody-mediated Treg depletion by anti-CTLA4 requires interactions between the anti-CTLA4 Fc region and activating FcγRs on tumor-associated macrophages (TAMs).³² To determine whether macrophage FcγR expression was associated with Treg frequency reduction in patients treated with ADT + anti-CTLA4-NF, we first annotated macrophage populations in the scRNA-seq dataset by both supervised manual annotation and through systematic testing of established PCa macrophage gene expression signatures³³ (Figures 3I–3K). We identified three putative macrophage subsets: a recently described prostate-specific MT⁺ macrophage subset³³ and two subsets of TAMs defined by TREM2 and CD206 (MRC1), respectively. Of these, TREM2⁺ and CD206⁺ TAMs exhibited the highest predicted protein activity for multiple FcγRs (Figure 3L), and TREM2⁺ TAMs showed the highest activity of the activating FcγR CD16a/FCGR3A (Figure 3M). To investigate whether antibody-dependent depletion may be a viable mechanism for the reduction in TI-Tregs observed in patients receiving ADT + anti-CTLA4-NF, we interrogated correlations between TI-Treg frequency and CD16a/FCGR3A protein activity on a per-patient basis. We found predicted CD16a/FCGR3A protein activity across all myeloid subsets to be significantly correlated with reduced Treg frequencies in ADT + anti-CTLA4-NF patients (p = 0.033; Figure 3N). In contrast, we found no significant correlation between myeloid CD16a/FCGR3A protein activity and TI-Treg frequencies in patients who received ADT alone (p = 0.223; Figure 3N), supporting the notion that CD16a/FCGR3A may be sufficient to trigger myeloid-directed TI-Treg ADCP upon treatment with anti-CTLA4-NF. We also found similar correlations between TI-Treg frequencies and CD16a/FCGR3A activity within the TREM2⁺ TAMs specifically (Figure 3O). In total, these exploratory analyses indicate TI-Treg frequency reduction in patients receiving ADT + anti-CTLA4-NF may be associated with expression of the activating FcγR CD16a/FCGR3A on tumor myeloid cells, warranting future studies to definitively validate TI-Treg depletion as a mechanism of action of anti-CTLA4-NF in humans.

Reduction in phenotypically activated Tregs is associated with improved disease-free survival

We sought to further investigate the biological activity of anti-CTLA4-NF by interrogating the molecular phenotype of residual Tregs in post-treatment tumors. We utilized the CyTOF data for this analysis, as we captured an insufficient number of Tregs by scRNA-seq to permit rigorous subclustering of the TI-Treg compartment. Semi-supervised FlowSOM subclustering and PaCMAP embedding of TI-Tregs in the CyTOF dataset identified nine phenotypically distinct Treg subpopulations. These were primarily resolved by differential expression of CD45RO, Ki67, granzyme B, CD39, CCR8, HLA-DR, and/or 4-1BB (Figures S6A–S6C). This analysis revealed a potential phenotypic shift in residual Tregs for patients treated with anti-CTLA4-NF marked by a loss of CD39⁻ TI-Tregs and enrichment in 4-1BB⁺ Tregs (Figures S6D and S6E). To validate these observations, we measured the fold change in geometric mean intensity (gMI) of all phenotypic markers in our CyTOF panel between Tregs in patients treated with ADT versus patients treated with anti-CTLA4-NF + ADT. We found Tregs in tumors after anti-CTLA4-NF exposure were enriched in expression of activation markers 4-1BB, CD39, CCR8, CTLA-4, and CD25. Together, these data support the notion that TI-Tregs remaining after anti-CTLA4-NF may possess an activated phenotype and warrant further investigation into this phenomenon as a potential mechanism of resistance to anti-CTLA4-NF (Figure 3P).

Given this observation, we next examined whether the depth of TI-Treg frequency reduction observed in treated patients associated with post-surgical recurrence rates. For this, we stratified patients who received ADT + anti-CTLA4-NF by recurrence (n = 4) versus non-recurrence (n = 7) within 2 years following surgery. We limited this exploratory analysis to the scRNA-seq dataset, as we did not capture CyTOF data for all patients who recurred. We found that TI-Treg frequencies at time of surgery were significantly higher in patients who went on to recur, despite the low numbers of patients per strata (p = 0.03; Figure 3Q). In contrast, no other lymphocyte subset was significantly associated with clinical outcome (Figure S6F). In addition, we stratified patients who received ADT + anti-CTLA4-NF by TI-Treg frequency and found that patients with high TI-Treg frequencies exhibited significantly shorter time to recurrence as compared to patients with low TI-Treg frequencies, again despite the relatively low rate of recurrence events in these strata (p = 0.0008; Figure 3R). There were insufficient recurrence events in the ADT-alone arm to perform a statistical analysis of time to recurrence in ADT-treated patients stratified by TI-Treg frequency, though no trend was observed (Figure S6G). In agreement with these data, prior studies show no association between Treg frequency and recurrence risk in localized PCa.³⁴ In sum, these exploratory analyses indicate reduced TI-Treg frequencies in patients receiving ADT + anti-CTLA-4 are significantly associated with greater post-surgical disease-free survival, warranting future studies to interrogate TI-Treg frequency as a biomarker of response to Fc-enhanced anti-CTLA-4 agents and alternatively, phenotypic activation of residual TI-Tregs as a putative mechanism of resistance.

Putative role of tumor DCs in response to ADT + anti-CTLA4-NF

Given the putative role of myeloid-expressed CD16/FCGR3A in modulating TI-Treg frequencies and response to ADT + anti-CTLA4-NF in these patients, we next sought to investigate associations between myeloid population frequencies and clinical outcomes. After stratification of ADT + anti-CTLA4-NF patients by recurrence status, we observed intratumoral dendritic cell (DC) frequencies trended higher in patients who did not recur (p = 0.06; Figure 4A). We then stratified patients by DC frequency at the time of surgery and found those with fewer tumor-infiltrating DCs exhibited a significantly shorter time to recurrence in this exploratory comparison (p = 0.043; Figure 4B). To further investigate this potential association between DC frequencies and anti-CTLA4-NF response, we returned to the MycCaP pre-clinical model of castration-sensitive PCa previously used to assess ADCC-competent versus ADCC-incompetent anti-CTLA4 antibodies in combination with ADT¹³ (Figure 4C). We implanted 8- to 10-week-old male FVB mice with syngeneic MycCaP tumors and performed chemical castration with degarelix acetate (ADT) when tumors averaged ∼200 mm³ in volume. Mice received a single dose of ADT alone or in combination with three doses of anti-CTLA-4 (clone 9D9; 5 mg/kg) administered intraperitoneally (i.p.) at 3-day intervals. To confirm Treg depletion was mediated through FcγR-dependent ADCC, mice received either Fc-competent antibody (mIgG2a; anti-CTLA4 [D]) or the equivalent clone and isotype bearing a mutation that nullifies FcγR interactions³⁵ (mIgG2a-LALAPG; anti-CTLA4 [ND]). These data were consistent with our prior findings that anti-CTLA4 (D), but not anti-CTLA4 (ND), augments response to ADT in the MycCaP model (Figures 4D and 4E). Following treatment, tumors and draining lymph nodes were harvested and profiled by 45-parameter spectral flow cytometry, and data were analyzed by semi-supervised FlowSOM clustering and supervised UMAP or unsupervised PaCMAP embedding for visualization (Figures 4F, S7A, and S7B). ADT + anti-CTLA4 (D) significantly reduced Treg frequencies compared to the ADT and ADT + anti-CTLA4 (ND) groups, indicating robust antibody-dependent Treg depletion (p = 0.0031; Figure 4G). We further found that animals treated with ADT + anti-CTLA4 (D) exhibited significantly greater frequencies of tumor-infiltrating CD11c⁺ MHC-II^hi DCs compared to those receiving ADT or ADT + anti-CTLA4 (ND) (p = 0.0009; Figure 4H). Systematic subclustering of the tumor DC compartment revealed a striking phenotypic shift in DCs in response to anti-CTLA4 (D) (Figures 4I, S7C, and S7D). While DC phenotypes in mice treated with ADT or ADT + anti-CTLA4 (ND) were enriched for canonical cDC1 and cDC2 markers CD103 or CD11b, respectively, nearly all DCs in mice treated with anti-CTLA-4 (D) expressed PD-L2, a marker of the mRegDC (mature DC enriched in immunoregulatory molecules) functional state (Figures 4J and S7E). The mRegDC phenotype is induced upon tumor antigen uptake, and DCs expressing the mRegDC program are thought to possess a unique role in shaping antitumor immune responses.³⁶ In line with the mRegDC phenotype, these cells also upregulated CD40 (Figure 4K). Taken together, these data indicate that ADCC/P-competent antibodies targeting CTLA-4 modulate tumor-infiltrating DCs and warrant further investigation into our exploratory finding that increased DC frequency following treatment with anti-CTLA4-NF is associated with improved recurrence-free survival.

Figure 4.

Effect of ADT + Fc-enhanced anti-CTLA-4 on DCs

(A) Violin plot of myeloid cluster frequencies as percentage of myeloid cells stratified by patient’s 2-year PSA recurrence status. No recurrence, n = 7; recurrence, n = 4. Two-tailed Student’s t test was utilized to test statistical significance. For this and all violin plots to follow, solid lines denote group medians, while dashed lines denote quartiles.

(B) Kaplan-Meier curve representing time-to-PSA recurrence in patients stratified by dendritic cell frequency; 95% confidence intervals shown in the shaded areas. There was one recurrence event in DC-high patients and three recurrence events in DC-low patients. Log rank test was performed to evaluate statistical significance.

(C) Schematic of pre-clinical validation experiment in the MycCaP model.

(D) Percent change in tumor volume relative to baseline volume prior to ADT stratified by treatment group. Average values are represented by solid lines and standard deviations are represented by shaded areas.

(E) Kaplan-Meier curve representing survival of mice shown in (D). Log rank (Mantel-Cox) test was used to evaluate statistical significance.

(F) Left: supervised UMAP of live CD45⁺ cells from MycCaP tumors harvested following therapy as denoted in (D and E). Semi-supervised clustering was performed using FlowSOM and clusters were manually annotated according to expression of lineage-defining markers. Right: heatmap of lineage-defining markers across all clusters. Data represent geometric MFI for each marker normalized to the minimum and maximum values across all clusters.

(G and H) (G) Frequency of TI-Tregs and (H) frequency of cDCs as percentage of all live CD45⁺ cells, stratified by treatment group.

(I) Left: PaCMAP representing FlowSOM-derived cDC clusters. Clusters were manually annotated according to expression of canonical marker proteins. Right: pseudocolor representation of cDC phenotypes in PaCMAP space stratified by treatment group.

(J) Violin plot representing frequency of PD-L2⁺ DCs as percentage of all DCs stratified by treatment group.

(K) Violin plot representing frequency of CD40⁺ DCs as percentage of all DCs stratified by treatment group. Welch’s t test was used to assess statistical significance. Murine data shown are n = 8 mice per group and representative of two independent experiments each for survival and immune profiling studies.

ADT + anti-CTLA-4-NF may enhance antitumor T cell priming

Given the observed effects of anti-CTLA4-NF therapy on tumor Treg frequencies and myeloid phenotypes, we next sought to determine the effect of ADT + anti-CTLA4-NF on CD8 T cell responses in the tumor. For enhanced resolution of CD8 phenotypes in the scRNA-seq data, we first refined the previously identified CD8-1 and CD8-2 clusters, removing a contaminating cell population presenting low Cd8a expression, moderate Cd4 expression, and an otherwise naive gene signature (Figures S8A and S8B). Subcluster analysis of the curated CD8 T cells resulted in an optimal three-cluster solution (Figure S8C). Utilizing established pan-cancer CD8 T cell gene signatures from Zheng et al.³⁷ and prostate CD8 T cell signatures from Tuong et al.,³³ we annotated these clusters as central memory (IL7R⁺ memory signature enrichment, prostate CD8 memory T cell signature enrichment, expression of CCR7), effector memory (enrichment of GZMK⁺ early TEM signature), and effector/resident memory CD8 T cells (enrichment of prostate CD8 effector T cell signature, expression of CD103, GzmB, TOX2; Figure S8D). These subsets were not significantly altered by ADT or ADT + anti-CTLA4-NF (Figure S8E). To further interrogate tumor-infiltrating CD8 T cell phenotypes at the protein level by CyTOF, we used semi-supervised FlowSOM clustering to define eight CD8 T cell subclusters (Figure 5A). While these clusters did not map precisely onto the three cluster solution in the scRNA-seq data, we did observe clusters that most closely matched the effector memory (Tc.4; CD45RA⁻CD45RO⁺CD27⁺HLA-DR⁺PD-1^mid) phenotype and cells that matched the effector/resident memory (Tc.6; CD45RO⁻HLA-DR⁺CTLA-4⁺TIGIT⁺GzmB⁺) phenotype and mirrored non-significant trends in the scRNA-seq data after anti-CTLA4-NF (Figures S8F and S8G). Notably, we observed a small cluster with clear enrichment in a subset of patients receiving ADT + anti-CTLA4-NF defined by 4-1BB and CD39 (Figures 5B–5D), markers that are associated with recently primed and tumor-specific T cells.^38,39 We found the dual 4-1BB⁺CD39⁺ phenotype to also define a distinct subset of CD4⁺FoxP3⁻ conventional CD4 T cells (Tconv) that were uniquely present in some patients receiving ADT + anti-CTLA4-NF (Figures 5E–5H, S8H, and S8I). While this phenotype was observed in both recurring and non-recurring patients and thus was not significantly associated with clinical outcomes in this trial, these data suggest anti-CTLA4-NF may modulate priming of tumor-specific T cell populations, either in concert with or independent of Treg depletion. To test whether this effect requires FcγR interactions, we returned to our murine model where we profiled effects of anti-CTLA4 (D) versus anti-CTLA4 (ND) in combination with ADT. In addition to significant changes in the frequencies of Ly6C effector memory-like CD8 T cells, proliferative PD-1+ effectors, and CD69⁺CD103⁺ resident memory CD8s (Figure S7G), we observed a strikingly similar cluster of 4-1BB⁺CD39⁺ CD8 T cells in Myc-CaP tumors that were uniquely induced by ADT + anti-CTLA4 (D) and were absent in the ADT-only and ADT + anti-CTLA4 (ND) groups (Figures 5I–5L and S7F–S7H), indicating FcγR engagement is required for induction of the 4-1BB⁺CD39⁺ CD8 T cell phenotype following anti-CTLA4 therapy. Supporting our hypothesis that these cells represent a recently primed population, we found a significant increase in CD8 and CD4 Tconv cells bearing phenotypic markers of recent priming (CD44⁺Ki67⁺) in tumor-draining lymph nodes (tdLNs) in mice receiving anti-CTLA4 (D) relative to anti-CTLA4 (ND) or ADT alone (Figures 5M–5O and Figure S7I). This was not associated with Treg depletion in the tdLNs, but rather an increase in Treg frequency in the tdLNs, as has been previously shown in the context of anti-CTLA4 therapeutics⁴⁰ (p < 0.0001; Figure S7J). These exploratory findings warrant future definitive validation of this putative mechanism of action of Fc-engineered anti-CTLA4-NF antibodies, potentially distinct from Treg depletion, involving enhanced priming of tumor-specific CD8 and/or non-Treg CD4 Tconv cells.

Figure 5.

Phenotypic modulation of T cells and enhanced priming by anti-CTLA4-NF

(A) PaCMAP plot of NeoRED-P patient tumor-infiltrating CD8 T cells by CyTOF. Clusters were derived from FlowSOM.

(B) Pseudocolor density plots of CD8 T cells in PaCMAP space stratified by treatment group.

(C) Expression of CD39 and 4-1BB by geometric MI on CD8 T cells as represented by color mapping on PaCMAP plot.

(D) Violin plot representing frequency of manually gated CD39⁺4-1BB⁺ CD8 T cells as a percentage of all CD8 T cells stratified by treatment group. Untreated, n = 7; ADT, n = 8; ADT + anti-CTLA4-NF, n = 8. Single patient with MSI^hi status called out in plot. For this and all violin plots to follow, solid lines denote group medians, while dashed lines denote quartiles.

(E) PaCMAP plot of NeoRED-P patient tumor-infiltrating CD4⁺ FoxP3^- Tconv cells by CyTOF. Clusters were derived from FlowSOM.

(F) Pseudocolor density plots of CD4 Tconv cells in PaCMAP space stratified by treatment group.

(G) Expression of CD39 and 4-1BB by geometric MI on CD4 Tconv as represented by color mapping on PaCMAP plot.

(H) Violin plot representing frequency of manually gated CD39⁺4-1BB⁺ CD4 Tconv cells as a percentage of all CD8 T cells stratified by treatment group. Untreated, n = 7; ADT, n = 8; ADT + anti-CTLA4-NF, n = 8. Single patient with MSI^hi status called out in plot.

(I) PaCMAP plot of MycCaP-infiltrating CD8 T cells by 45-parameter flow cytometry in response to ADT, ADT + anti-CTLA4 (ND), or ADT + anti-CTLA4 (D). Clusters were derived from FlowSOM.

(J) Pseudocolor density plots of CD8 T cells in PaCMAP space stratified by treatment group.

(K) Expression of CD39 and 4-1BB by geometric MFI on CD8 T cells as represented by color mapping on PaCMAP plot.

(L) Violin plot representing frequency of manually gated CD39⁺4-1BB⁺ CD8 T cells as a percentage of all CD8 T cells stratified by treatment group.

(M) Biaxial plots representing expression of CD44 and Ki67 on CD8 T cells in tumor-draining lymph nodes of mice shown in (I–L).

(N) Biaxial plots representing expression of CD44 and Ki67 on CD4⁺FoxP3⁻ Tconv cells in tumor-draining lymph nodes of mice shown in (L–O).

(O) Violin plots representing frequencies of CD44⁺Ki67⁺ CD8 and CD4 Tconv cells as a percentage of parent populations stratified by treatment group. Two-tailed Welch’s t test was used to assess statistical significance. All murine data shown are n = 7 per group and representative of two independent experiments each for survival and immune profiling studies.

Discussion

This study evaluated a next-generation non-fucosylated anti-CTLA4 antibody (BMS-986218) engineered to enhance ADCC in combination with ADT versus ADT alone prior to surgery in men with high-risk localized PCa. While most early-phase clinical studies test investigational agents in highly advanced, late-stage disease, our study evaluated anti-CTLA4-NF in patients with localized PCa based on prior evidence suggesting this setting would be more amenable to the distinct immunological effects of combined ADT + anti-CTLA4-NF. Our study also included an independent contemporaneously collected untreated PCa control cohort to compare immunological effects of ADT + anti-CTLA4-NF to ADT alone or no treatment. Treatment with ADT + anti-CTLA4-NF was feasible with a tolerable safety profile in the pre-surgical setting, meeting our pre-specified primary study endpoint. We demonstrated that anti-CTLA4-NF has promising biological activity in patients, consistent with its proposed mechanism of action of TI-Treg depletion. Our exploratory single-cell multi-omic analyses and associated pre-clinical studies yielded actionable biomarkers of response and resistance to ADT + anti-CTLA4-NF, warranting further investigation and validation in follow-up studies.

These exploratory analyses revealed both expected and unanticipated associations with treatment response. We showed that patients treated with a glycoengineered anti-CTLA-4 antibody exhibit reduced TI-Treg frequencies, this being associated both with expression of the activating FcγR CD16a/FCGR3A on tumor macrophages and with improved rates of disease-free survival. Our findings, though exploratory in nature, do not contradict the proposed mechanism of TI-Treg depletion by Fc-enhanced anti-CTLA-4 agents established in pre-clinical models, but for which limited clinical evidence currently exists.^20,32 If true, these findings would also support the hypothesis that first-generation anti-CTLA4 antibodies ipilimumab and tremelimumab may fail to deplete TI-Tregs in humans in part due to inefficient engagement of CD16a on TAMs, which has been proposed based on data from humanized murine models.⁴¹ Indeed, non-engineered anti-CTLA4 IgG1 antibodies preferentially engage the inhibitory FcγR FCGR2B/CD32b on myeloid cells, preventing TI-Treg-directed ADCC/P.⁴¹ Complicating this mechanism is our observation that inhibitory FCGR2B and activating FCGR3A receptors are highly co-expressed on TAMs. Additional studies are warranted to understand how co-expression patterns of FCGR3A and FCGR2B on TAMs influence efficiency of Treg depletion by Fc-engineered antibodies targeting CTLA-4. An alternative mechanism supported by our data is that tumors with low baseline frequencies of FCGR3A/CD16a⁺ TAMs—or physical segregation of these macrophages from TI-Tregs—may fail to respond to anti-CTLA4-NF. A positive role for FCGR3A/CD16a⁺ TAMs in the context of anti-CTLA4-NF is notable, as FCGR3A/CD16a⁺ TAMs are generally enriched for the immunosuppressive, M2-like phenotype (e.g., TREM2⁺ and CD206⁺ TAMs in our study) that plays a negative prognostic role in many cancers.²⁶ However, the role of these TAMs may vary in distinct therapeutic contexts. Recent data suggest CD206⁺ M2-like TAMs may in fact be necessary to organize antitumor immune responses,⁴² which together with our findings challenges the narrative that M2-like TAMs are universally antagonistic to cancer immunotherapies. In either case, pre-clinical studies suggest anti-CTLA4-mediated FcγR engagement on tumor macrophages can elicit pro-inflammatory macrophage skewing.⁴³ Therefore, Fc-engineered anti-CTLA4 antibodies may deliver a broadly applicable, and possibly synergistic, dual benefit of TI-Treg depletion and M2-to-M1 macrophage skewing that may be effective in overcoming the immunosuppressive microenvironment of many tumor types that do not respond to ICB.

Although patients treated with anti-CTLA4-NF exhibited lower average TI-Treg frequencies as compared to controls, residual TI-Tregs did remain in all tumors. Interestingly, these TI-Tregs appeared phenotypically activated. While our current studies could not confirm whether these Tregs were functionally immunosuppressive, these findings indicate a potential limitation of TI-Treg-targeting therapies that utilize a single coinhibitory receptor such as CTLA-4. We hypothesize that TI-Tregs undergoing blockade of CTLA-4 without experiencing ADCC/P (whether through engagement of FCGR2B on proximal TAMs or spatial segregation from ADCC/P effectors) become activated as a result of reduced coinhibitory receptor activity, as has been observed in mice receiving anti-CTLA4 or in context of anti-PD1 blockade.^40,44 Notably, we found these Tregs to be enriched in other potentially targetable surface proteins, such as CCR8, CD25, CD39, and 4-1BB. Fc competent and glycoengineered antibodies targeting CCR8 are currently in clinical development.^45,46 Thus, co-delivery of glycoengineered antibodies targeting CTLA-4 and CCR8 may be complementary, given the potential capacity for anti-CCR8 to deplete TI-Tregs remaining following anti-CTLA4 and that anti-CTLA4 elicits enhanced T cell priming and myeloid activation. Recent studies also suggest CD4 Tconv in tumors depleted of TI-Tregs upregulate CCR8 and become functionally immunosuppressive.⁴⁷ These studies warrant additional basic and clinical studies of glycoengineered antibodies targeting CTLA4 and CCR8 in prostate and other cancers.

In addition to Treg frequency reduction, we found the glycoengineered anti-CTLA4 antibody may enhance T cell priming, potentially by increasing interactions between CTLA4+ effector T cells and FcγR⁺ dendritic cells. High tumor DC frequencies in our data were significantly associated with lack of disease recurrence in patients receiving ADT + anti-CTLA4-NF. In addition, only patients treated with anti-CTLA4-NF elicited a distinct CD39⁺ 41BB⁺ phenotype in tumor-infiltrating CD8 and CD4 T cells, mirroring established phenotypes for tumor-specific T cells (CD39⁺CD103⁺³⁸) and recently primed tumor-specific cells (41BB⁺). Our pre-clinical studies in the murine syngeneic MycCaP model were in striking agreement with these clinical data, and prior in vitro work with human T cells and DCs in context of bacterial superantigen-mediated stimulation suggest FcγR engagement on APCs is required for enhanced priming activity of antibodies targeting CTLA-4, independent of Treg depletion.⁴⁸ Our pre-clinical findings that FcγR-engaging—but not FcγR-silenced—anti-CTLA-4 antibodies elicit a significant phenotypic shift in tumor DCs suggests a bi-directional effect on both T cells and DCs, reminiscent of observed inflammatory polarization of macrophages post-anti-CTLA-4 in other murine models.⁴³ Nevertheless, increasing evidence indicates spatial co-localization of effector CD8 and CD4 T cells with DCs in tumors (termed immune “triads,” “hubs,” or “networks”^49,50) is critical for productive antitumor immune responses. Further investigation is warranted to determine whether Fc-enhanced antibodies targeting CTLA-4 or other immune checkpoints can facilitate accumulation of these specialized immune clusters in human tumors and whether this is an essential underlying mechanism of action of these therapeutics.

In conclusion, we demonstrate the safety and feasibility of anti-CTLA4-NF (BMS-986218) in combination with ADT prior to surgery in men with high-risk localized prostate cancer. Our robust single-cell immune profiling studies provide first-in-human insight into putative mechanisms underlying reduced TI-Treg frequencies in patients treated with a glycoengineered antibody targeting CTLA-4. In combination with pre-clinical modeling, these findings support both canonical and exploratory mechanisms underlying activity of Fc-enhanced anti-CTLA4 for further investigation. Our study identifies immunological correlates of clinical response to ADT + anti-CTLA4-NF to inform future studies. Given the limited efficacy and significant toxicity associated with neoadjuvant ipilimumab in prior studies involving PCa patients⁵¹ (also see NCT02020070), our findings provide an avenue to explore Fc-enhanced CTLA-4 targeted immunotherapy in patients with PCa.

Limitations of the study

Our study had several limitations. One key limitation was the small sample size, making interpretation of the clinical activity of ADT + anti-CTLA4-NF challenging. Indeed, our study was not designed with sufficient power to compare outcome differences between the treatment arms. This limitation also applies to our exploratory correlative analyses, which also included data from four non-randomized patients receiving ADT + anti-CTLA4-NF during a safety lead-in, thus requiring further validation in larger follow-up studies. Though our analyses benefited from an additional untreated control cohort, a direct comparison of the effects of anti-CTLA4-NF to existing agents such as ipilimumab or tremelimumab was not possible in this study. While the relatively short neoadjuvant treatment interval (3–4 weeks) was designed to study immediate effects of anti-CTLA4-NF on TI-Treg frequencies, our study does not provide insight into the time dependence of our immunological correlative findings. Similarly, our orthogonal multi-omic measures of TI-Treg depletion relied on post-treatment frequency or density comparisons between treatment groups, as opposed to matched pre- and post-treatment tissue, as freshly processed pre-treatment samples were not available. It remains possible that patients in the ADT + anti-CTLA4-NF arm possessed higher baseline TI-Treg densities than untreated or ADT-treated patients, given their relative enrichment for Gleason Grade Group 5 disease. Thus, the observed reduction in TI-Treg frequencies in our dataset could be underestimated. However, comparisons between pre-treatment biopsies and surgical resection tissue may be confounded when assessing the immune TME due to inflammatory effects of biopsy collection or due to spatial intratumoral heterogeneity, as has been seen in colorectal cancer.⁵² Clinically, it is notable that development of BMS-986218 has been discontinued (NCT03110107). Nevertheless, several related second-generation “depleting” anti-CTLA4 antibody agents (botensilimab/AGEN1181, ONC-392, ADG216) are in clinical development. Of note, recent findings from Chand et al. indicate botensilimab, an Fc-enhanced DLE-mutated anti-CTLA4-hIgG1 antibody, may deplete TI-Tregs in an FcγR-dependent fashion across cancer types and potentially modulates DCs, closely mirroring our pre-clinical and clinical findings with BMS-986218 in PCa.¹⁶

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Dr. Matthew Dallos (dallosm@mskcc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All patient-derived sequencing data and CyTOF data are publicly available. See key resources table in STAR Methods for repository accession details.
• •No software or algorithm was generated in this study. The code for reproducing major figures is available on GitHub (https://github.com/Aleksobrad/neoredp_analysis).
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This research was supported by Bristol Myers Squibb, Ferring Pharmaceuticals, National Institutes of Health grants 1P50CA58236-15 and P30CA006973, the Prostate Cancer Foundation Challenge Grant, T.J. Martell Foundation funds, and CUMC institutional funds to C.G.D.; the NCI’s Center for Cancer Systems Therapeutics (CaST) award U54CA274506, the NCI Outstanding Investigator award R35 CA197745, and the NIH Shared Instrumentation Grants S10 OD012351, S10 OD021764 and S10OD032433, to A.C.; a Prostate Cancer Foundation Young Investigator Award to M.D.; NIH grants UL1TR001873 and TL1TR001875, ABRC grant RFGA2023-008-25, and institutional funds from Mayo Clinic to C.R.A.; NIH grant F30CA260765-01 to A.O.; and NIH/NIDDK grant T35DK093430 in support of A.L.E.W. We would like to acknowledge the Columbia University Human Immune Monitoring Core (HIMC) for assistance with single-cell RNA sequencing and CyTOF data acquisition and the Mayo Clinic Immune Monitoring Core for assistance with CyTOF data acquisition. We would also like to acknowledge our collaborators at Bristol Myers Squibb for their help in conducting the study. Most of all, we thank the patients and their families for participating in this study.

Author contributions

Conceptualization, C.G.D., M.C.D., A.C., C.R.A., and A.O.; methodology, C.G.D., M.C.D., A.C., C.R.A., and A.O.; software: A.C.; A.O., P.McC., and A.L.E.W.; validation, C.R.A, A.O., and M.C.D.; formal analysis, C.R.A., A.O., P.McC., and A.L.E.W.; investigation, C.R.A., A.O., P.McC., M.C., A.L.E.W., N.S., S.S.P., R.K.V., C.J.L., J.C.H., C.J., G.D., and N.B.; resources, C.R.A.; A.O.; P.S., G.J.D., C.B.A., J.M.McK., C.S.S., M.N.S., K.R., C.G.D., and M.C.D.; data curation, C.R.A., A.O., P.McC., M.C., K.R., and M.C.D.; writing—original draft, C.R.A., A.O., P.McC., and M.C.D.; writing—review & editing, C.R.A., A.O., K.R.C., P.N., H.I.S., C.B.A., C.G.D., A.C., and M.C.D.; visualization, C.R.A., A.O., P.McC., and A.L.E.W.; supervision, C.R.A., A.O., C.G.D., A.C., and M.C.D.; project administration, C.R.A., A.O., C.G.D., A.C., and M.C.D.; funding acquisition, C.R.A., C.G.D., A.C., and M.C.D.

Declaration of interests

C.G.D. is a co-inventor on patents licensed from Johns Hopkins University to BMS and Janssen. He is currently a paid employee of JnJ Innovative Medicine. A.C. is founder, equity holder, and consultant of DarwinHealth Inc., which has licensed IP related to these algorithms from Columbia University. Columbia University is an equity holder in DarwinHealth Inc.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-Human CD45 (30-F11) 89Y	Standard BioTools	Cat# 3089005B
Viability Stain 103Rh	Standard BioTools	Cat# 201103A
Anti-Human CD3 (UCHT1) 141PR	Standard BioTools	Cat# 3141019B
Anti-Human CD123 (6H6) 142ND	BioLegend	Cat# 306027; RRID: AB_2562823
Anti-Human CD45RA (HI100) 143ND	Standard BioTools	Cat# 3143006B
Anti-Human CD4 (RPA-T4) 145ND	Standard BioTools	Cat# 3145001B; RRID: AB_3661845
Anti-Human Helios (22F6) 147Sm	BioLegend	Cat# 137202; RRID: AB_10900638
Anti-Human CD16 (3G8) 148ND	Standard BioTools	Cat# 3148004B; RRID: AB_10900638
Anti-Human CCR4 (L291H4) 149Sm	Standard BioTools	Cat# 3149029A
Anti-Human OX40 (ACT35) 150ND	Standard BioTools	Cat# 3150023B
Anti-Human HLA-DR (G46-6) 151Eu	Standard BioTools	Cat# 3151023B
Anti-Human CD66b (80H3) 152Sm	Standard BioTools	Cat# 3152011B
Anti-Human TIGIT (MBSA43) 153Eu	Standard BioTools	Cat# 3153019B
Anti-Human Tim-3 (F38-2E2) 154Sm	Standard BioTools	Cat# 3154010B
Anti-Human CD27 (O323) 155Gd	BioLegend	Cat# 302839; RRID: AB_2562817
Anti-Human LAP (S20006A) 156Gd	BioLegend	Cat# 300002; RRID: AB_2892340
Anti-Human 4-1BB (4B4-1) 158Gd	Standard BioTools	Cat# 3158013B
Anti-Human GITR (621) 159Tb	Standard BioTools	Cat# 3159020B
Anti-Human CD39 (A1) 160Gd	Standard BioTools	Cat# 3160004B
Anti-Human CD163 (GHI/61) 161Dy	BioLegend	Cat# 333602; RRID: AB_1088991
Anti-Human FoxP3 (PCH101) 162Dy	Standard BioTools	Cat# 3162011A
Anti-Human CD56 (NCAM16.2) 163Dy	Standard BioTools	Cat# 3163007B
Anti-Human CD45RO (UCHL1) 165Ho	Standard BioTools	Cat# 3165011B
Anti-Human CCR8 (SA214G2) 166Er	BioLegend	Cat# 150341; RRID: AB_3105975
Anti-Human CD33 (WM53) 167Er	BioLegend	Cat# 303437; RRID: AB_2814130
Anti-Human CD8 (SK1) 168Er	BioLegend	Cat# 344727; RRID: AB_2563762
Anti-Human CD25 (2A3) 169Tm	Standard BioTools	Cat# 3169003B; RRID: AB_2938861
Anti-Human CTLA-4 (14D3) 170Er	Standard BioTools	Cat# 3170005B
Anti-Human GzmB (GB11) 171Yb	Standard BioTools	Cat# 3171002B
Anti-Human Ki67 (B56) 172Yb	Standard BioTools	Cat# 3172024B
Anti-Human CD141 (1A4) 173Yb	Standard BioTools	Cat# 3173002B
Anti-Human PD-1 (EH12.2H7) 174Yb	Standard BioTools	Cat# 3174020B
Anti-Human CD14 (M5E2) 175Lu	Standard BioTools	Cat# 3175015B
Anti-Human CD127 (A019D5) 176Yb	Standard BioTools	Cat# 3176004B; RRID: AB_3665122
DNA stain 191Ir	Standard BioTools	Cat# 201192B
DNA stain 193Ir	Standard BioTools	Cat# 201192B
Anti-Human CD11b (ICRF44) 209Bi	Standard BioTools	Cat# 3209003B
Anti-Human TruStain FcX	BioLegend	Cat# 422302; RRID: AB_2818986
Anti-Mouse CD45 (YW62.3) StarBright UV400	BioRad	Cat# MCA1031SBUV400; RRID: AB_3099860
Anti-Mouse CD103 (M290) BUV496	BD	Cat# 741083; RRID: AB_2870687
Anti-Mouse NKp46 (29A1.4) BUV563	BD	Cat# 741435; RRID: AB_2870912
Anti-Mouse CD4 (GK1.5) BUV615	BD	Cat# 613006; RRID: AB_2870274
Anti-Mouse PD-1 (29F.1A12) BUV661	BD	Cat# 568603
Anti-Mouse CD62L (MEL-14) BUV737	Thermo	Cat# 367-0621-80; RRID: AB_2895976
Anti-Mouse CD39 (24DMS1) BUV805	Thermo	Cat# 368-0391-80; RRID: AB_2896107
Anti-Mouse CD64 (X54-5/7.1) BV421	BioLegend	Cat# 139309; RRID: AB_2562694
Anti-Mouse CD19 (6D5) StarBright V440	BioRad	Cat# MCA1439SBV440; RRID: AB_3100284
Anti-Mouse Ki67 (11F6) Pacific Blue	BioLegend	Cat# 151223; RRID: AB_2922472
Anti-Mouse TCRgd (GL3) BV480	BD	Cat# 746343; RRID: AB_2743663
Anti-Mouse SIRPa (P84) BV510	BioLegend	Cat# 144032; RRID: AB_2810411
Anti-Mouse IgD (11-26c.2a) Spark Violet 538	BioLegend	Custom
Anti-Mouse CX3CR1 (SA011F11) BV570	BioLegend	Custom
Anti-Mouse PD-L1 (10F.9G2) BV605	BioLegend	Cat# 124321; RRID: AB_2563635
Anti-Mouse CD80 (16-10A1) BV650	BioLegend	Cat# 104731; RRID: AB_11147759
Anti-Mouse CD8a (53–6.7) BV711	BioLegend	Cat# 100747; RRID: AB_11219594
Anti-Mouse PD-L2 (MIH37) BV750	BD	Cat# 752598; RRID: AB_2917585
Anti-Mouse CD3 (17A2) BV785	BioLegend	Cat# 100231; RRID: AB_11218805
Anti-Mouse Tim-3 (5D12) BB515	BD	Cat# 567810; RRID: AB_2916742
Anti-Mouse IA/IE (M5/114.15.2) Spark Blue 550	BioLegend	Cat# 107662; RRID: AB_2860616
Anti-Mouse ICOS (C398.4A) Spark Blue 574	BioLegend	Custom
Anti-Mouse CD127 (A7R34) BB630	BD	Custom
Anti-Mouse CD40L (MR1) BB660	BD	Custom
Anti-Mouse CD44 (IM7) PerCP	BioLegend	Cat# 103035; RRID: AB_10639933
Anti-Mouse Ly6C (AL-21) RB705	BD	Cat# 570265; RRID: AB_3685626
Anti-Mouse CD206 (MR6F3) PerCPeF710	Thermo	Cat# 46-2069-42; RRID: AB_10853034
Anti-Mouse KLRG1 (2F1) BB755	BD	Custom
Anti-Mouse B220 (RA3-6B2) RB780	BD	Cat# 569208; RRID:AB_3684869
Anti-Mouse CD11b (M1/70) PerCP/Fire 806	BioLegend	Cat# 101293; RRID: AB_2941371
Anti-Mouse CCR8 (CDw198) PE	Miltenyi	Cat# 130-115-377; RRID: AB_2751259
Anti-Mouse Ly6G (1A8) Spark YG 593	BioLegend	Cat# 127668
Anti-Mouse Helios (22F6) PE/Dazzle 594	BioLegend	Cat# 137232; RRID: AB_2565797
Anti-Mouse CTLA-4 (UC10-4B9) PE/Fire 640	BioLegend	Cat# 106333; RRID: AB_2936682
Anti-Mouse CD40 (3/23) PE/Cy5	BioLegend	Cat# 124618; RRID: AB_2075922
Anti-Mouse 4-1BB (17B5) PE/Fire 700	BioLegend	Custom
Anti-Mouse TCF-1 (C63D9) PE/Cy7	CST	Cat# 90511; RRID: AB_3086656
Anti-Mouse CD11c (QA18A72) PE/Fire 810	BioLegend	Cat# 161105; RRID:AB_2904307
Anti-Mouse iNOS (REA982) APC	Miltenyi	Cat# 130-116-423; RRID:AB_2727527
Anti-Mouse Siglec F (S17007L) Alexa 647	BioLegend	Cat# 155519; RRID:AB_2890711
Anti-Mouse CD25 (PC61) Spark NIR 685	BioLegend	Cat# 102069; RRID: AB_2888823
Anti-Mouse FoxP3 (FJK-16S) Alexa 700	Thermo	Cat# 56-5773-82; RRID: AB_1210557
Live/Dead Zombie NIR	BioLegend	Cat# 423106
Anti-Mouse CD69 (H1.2F3) APC/Fire 750	BioLegend	Cat# 104549; RRID: AB_2800561
Anti-Mouse F4/80 (BM8) APC/Fire 810	BioLegend	Cat# 123166
Anti-Mouse TruStain FcX PLUS (S17011E)	BioLegend	Cat# 156604; RRID: AB_2783138
Anti-Human CD4 (EPR6855) Purified	Abcam	Cat# ab133616; RRID: AB_2750883
Anti-Human FoxP3 (236A/E7) Purified	Thermo	Cat# 14-4777-82; RRID: AB_467556
Horse anti-Mouse IgG Biotinylated	Vector	Cat# BA-2000-1.5
Goat Anti-Rabbit IgG Biotinylated	Vector	Cat# BP-9100-50; RRID: AB_3665891
Streptavidin Alexa 594	Thermo	Cat# S11227
Streptavidin Alexa 488	Thermo	Cat# S11223
Vectashield Antifade Mounting Media with DAPI	Vector	Cat# H-1800
Biological samples
Human prostatectomy specimens	This paper	N/A
Chemicals, peptides, and recombinant proteins
Degarelix Acetate	Cayman Chemical	Cat# 24069
RecombiMab anti-Mouse CTLA-4 9D9-mIgG2a	BioXCell	9D9-CP007
RecombiMab anti-Mouse CTLA-4 9D9-mIgG2a-LALAPG	BioXCell	9D9-CP008
Critical commercial assays
Common Hereditary Cancers Germline Panel	Invitae	Cat# 01102
Multi-Cancer Germline Panel	Invitae	Cat# 01101
Columbia Combined Cancer NGS Panel	Columbia University	N/A
Human Tumor Dissociation Kit	Miltenyi Biotec	Cat# 130-095-929
Chromium 3′ v2 Library and Gel Bead Kit	10X Genomics	Cat# 120237
FoxP3 Fixation/Permeabilization Kit	Thermo	Cat# 00-5523-00
Deposited data
Analyzed NeoRED scRNAseq data with code	This paper	https://github.com/Aleksobrad/neoredp_analysis
NeoRED scRNAseq raw data	This paper; Mendeley Data: https://doi.org/10.17632/nzjscnbgf5.1	https://data.mendeley.com/datasets/nzjscnbgf5/1
NeoRED CyTOF data	This paper; Zenodo: https://doi.org/10.5281/zenodo.17526117	https://zenodo.org/records/17526118
Experimental models: Cell lines
Mouse: MycCaP cells	Laboratory of Charles Sawyers	Watson53
Experimental models: Organisms/strains
Mouse: FVB/NJ	The Jackson Laboratory	JAX: #001800
Software and algorithms
Cellranger v7.0	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger/downloads
Seurat v4	Satija Lab	https://satijalab.org/seurat/articles/archive
SingleR	Aran et al.28	https://bioconductor.org/packages/release/bioc/html/SingleR.html
InferCNV	Patel et al.29	https://github.com/broadinstitute/inferCNV
PISCES	https://doi.org/10.1101/2021.05.20.445002	https://github.com/califano-lab/PISCES
QuPath v0.5.1	Bankhead et al.54	https://qupath.github.io/
WSInfer	Kaczmarzyk et al.55	https://wsinfer.readthedocs.io/en/latest/
StarDist v0.5.0	Stevens56	https://stardist.net/
FlowJo v10.10.0	BD	https://www.flowjo.com/
Gtsummary v1.7.2	https://doi.org/10.32614/RJ-2021-053	https://www.danieldsjoberg.com/gtsummary/
R v4.3.1	CRAN	https://cran.r-project.org/bin/windows/base/old/
Rstudio	Posit	https://posit.co/download/rstudio-desktop/
LandScape v1.1.1	Oviz-Bio	https://oviz.org
Srvminer v0.4.9		https://rpkgs.datanovia.com/survminer/
GraphPad Prism (v10.2)	GraphPad	https://www.graphpad.com/features
Other
GentleMACS Octodissociator	Miltenyi Biotec	N/A
10X Chromium Controller	10X Genomics	N/A
PhenoImager HT	Akoya Biosciences	N/A
CyTOF Helios	Standard BioTools	N/A
Cytek Aurora (5L; UV/V/B/YG/R)	Cytek	N/A

Experimental model and study participant details

Patients

A total of 24 trial-enrolled and 12 untreated control male subjects were recruited at Columbia University Medical Center. See Clinical Trial Design and Statistical Considerations section below for sample size justification. Key eligibility criteria included histologically confirmed adenocarcinoma of the prostate (clinical stage T1c-T3b, N0, M0) with no detectable metastatic involvement of lymph nodes, bone, or visceral organs and high-risk disease as defined by central pathologic review of at least two biopsy cores with Gleason sums of ≥4 + 3. Patients were required to demonstrate adequate bone marrow, hepatic, and renal function. Prior exposure to radiation therapy, chemotherapy, immunotherapy, biologic therapy, or concomitant treatment with hormonal therapy and/or systemic corticosteroids (with the exception of short courses <5 days 7 days or more prior to treatment initiation) was not allowed. Participant demographics and characteristics (e.g., age, race, ethnicity, etc.) are provided in Table 1. This trial was approved and all amendments reviewed by the Columbia University Institutional Review Board (IRB) under the protocol number AAAS3560. All patients provided informed consent prior to enrollment in the study.

Mice

Male FVB mice were purchased from Jackson Laboratory (Bar Harbor, ME). Mice were 8–10 weeks old at time of use, and were acclimated for at least a week to the Mayo Clinic vivarium prior to use in experiments. Mice were housed in a maximum of 5 animals per cage with 12-h light cycles and food and water provided ad libitum. All animals were housed in strict accordance with NIH and American Association of Laboratory Animal Care regulations. All experiments and procedures for this study were approved by the Mayo Clinic Institutional Animal Care and Use Committee (IACUC).

Cell lines

MycCaP cells were provided as a kind gift from Dr. Charles Drake and were verified as free of pathogen and cell line contamination using a 19-marker mouse STR panel through IDEXX BioAnalytics (Columbia, MO). MycCaP cells were passaged in RPMI medium (Corning; Corning, NY) supplemented with 10% FBS (Corning; Corning, NY), 100 U/mL penicillin, and 100 mg/mL streptomycin (Gibco; Gaithersburg, MD).

Method details

Clinical trial design and statistical considerations

Trial Design | This was a single-center, randomized, two-arm, open label pilot study conducted at Columbia University Irving Medical Center to evaluate the safety, feasibility, and immunogenicity of neoadjuvant degarelix (ADT) or BMS-986218 (anti-CTLA4-NF) plus ADT prior to radical prostatectomy in men with high-risk localized prostate cancer. 24 patients were enrolled in the study. The first 4 patients were assigned to receive ADT + anti-CTLA4-NF as a safety lead-in. Thereafter, 10 patients each were randomized by the study statistician to the two treatment arms. In the ADT-only arm, men received 240 mg degarelix acetate subcutaneously two weeks prior to surgery. In the ADT + anti-CTLA4-NF arm, men received 20 mg BMS-986218 intravenously on day 1 and day 15, and 240 mg degarelix acetate subcutaneously on day 8, starting 3 weeks prior to prostatectomy (Figure 1A).

Primary and secondary objectives and outcome variables

The primary objective was to characterize the safety, tolerability, and feasibility of ADT with or without anti-CTLA4-NF in the neoadjuvant setting. The secondary objectives included: (i) evaluation an immune response consistent with the proposed mechanism of action of anti-CTLA4-NF, specifically the effect on intratumoral Treg density and CD8:Treg ratio; (ii) estimation of the pathologic complete response rate following neoadjuvant anti-CTLA4-NF plus ADT as compared to ADT alone; (iii) estimation of the PSA50 response rate at time of surgery following neoadjuvant anti-CTLA4-NF plus ADT as compared to ADT alone; (iv) rate of undetectable PSA at 12 months following neoadjuvant anti-CTLA4-NF plus ADT as compared to ADT alone; and (v) estimation of the rate of post-surgical PSA recurrence within two years of completing therapy between patients receiving ADT alone versus ADT plus anti-CTLA4-NF.

Serum PSA and testosterone levels were measured on all patients prior to therapy initiation and at time of surgery. Thereafter, serum PSA and testosterone was measured on all patients every 3 (±1) months during the first post-operative year and every 6 (±2) months during year 2 and 3 of follow up. Adverse events were monitored during the course of therapy and until at least two years post-operatively and registered according to CTCAE version 5.0. Adverse events were deemed associated or not associated with ADT and/or anti-CTLA4-NF according to the treating physician. Tumor samples from prostatectomy specimens were collected at time of surgery; fresh tissue was collected and processed as described below, and archival formal-fixed paraffin-embedded (FFPE) tissues were obtained for indicated correlative studies. In time-to-event reporting, patients were censored at the date of last assessment if any of the following criteria were met: (i) patient lost to follow-up; (ii) patient receiving adjuvant therapy; (iii) patient death unrelated to treatment; or (iv) patient remained PSA recurrence-free at 24 months follow-up.

Randomization method

The first four patients enrolled in the trial received ADT + anti-CTLA4-NF as a mandatory safety lead-in. All subsequent patients were randomized 1:1 to each arm by the trial statistician. An erroneous reference to stratified randomization in the statistical section of the trial protocol was not followed; instead, all patients were randomized according to the “Randomization and Blinding” section of the protocol where patients were randomized 1:1 using a random number generator. This trial was open-label so patients and investigators were not blinded to treatment allocation.

Sample size justification

The clinical trial sample size was driven by detecting a pharmacodynamic effect of the agent based on the key secondary endpoint of Treg infiltration. A biologically meaningful treatment effect was deemed to be a 50% decrease in Treg infiltration with BMS-986218 plus degarelix versus degarelix alone. Previous studies show that Treg density after degarelix is 59 cells/μm² (95% CI: 34–85). A sample size of 10 evaluable subjects per group achieves greater than 80% power to detect a minimum difference of 29 cells/μm² based on two-sided two-sample t test with a significance level (alpha) of 0.05 assuming a standard deviation of 25 cells/μm².

Data analysis plan

A detailed pre-specified analysis plan for key endpoints can be found in the Supplementary Trial Protocol file. As the trial was powered to compare post-treatment Treg frequency in surgical specimens, the Treg frequency endpoint is confirmatory while all other clinical secondary endpoints are exploratory in nature. For time-to-event analyses, we note that in this setting of low event rates, medians and hazard-based estimates are unstable so our findings are interpreted as descriptive and supportive rather than confirmatory. Event rates for all clinical trial time-to-event plots can be found in the associated figure legend. No global multiplicity plan was pre-specified for immune correlative analyses, thus all multi-omic analyses are also exploratory in nature. See relevant statistical subsections below for details on statistical considerations for data from each correlative platform.

Analysis sets

There were three patients that were enrolled in the trial but not included in all endpoint assessments: (1) one patient in the ADT + anti-CTLA4-NF arm with surgery cancellation due to COVID-19 that was not included in any secondary endpoint analysis; (2) one patient in the ADT-only arm that lacked a PSA test at time of prostatectomy so did not have a PSA response rate assessment but was included in all other secondary endpoint analyses; (3) one patient in the ADT + anti-CTLA4-NF arm that received adjuvant therapy roughly 3 months post-prostatectomy so was excluded from the post-surgical endpoint analyses due to receiving additional non-trial-related therapy. A single patient exhibiting an MSI^Hi genotype and resulting TMB^Hi signature (see Figure S1) and all patients receiving anti-CTLA-4NF during the non-randomized safety lead-in phase of the trial were censored from confirmatory Treg frequency comparison (Figure 2B). These patients were allowed to be included in exploratory correlative analyses given the hypothesis-generating nature of these data. Patients with evaluable data for each correlative data platform are shown in Figure 3B. We note that the number of evaluable samples across correlative analyses is smaller than planned in the original power analyses, thus resulting inferences are based on a modest pilot cohort and are exploratory in nature even when statistically significant.

Tumor sample collection & processing

Following radical prostatectomy, surgical tissues were grossed and sectioned by pathologists at Columbia University Irving Medical Center. Up to 1 gram of tumor tissue was collected in Miltenyi MACS Tissue Storage Solution on wet ice, massed, washed with PBS, then diced with microdissection scissors to 1-2mm pieces in a 5mL Eppendorf tube. Tissue pieces were transferred into a Miltenyi C tube (0.5g per tube) in serum free RPMI, then enzymes H, R, and A were added according to the Miltenyi Human Tumor Dissociation Kit protocol. Tumors were digested on a gentleMACS Octodissociator under program 37_multi_A_01. Once finished, RPMI supplemented with 5% FBS (R5 media) was added to quench digest enzymes, then suspensions were filtered (70μm) into a 15mL conical tube and spun at 400xg for 5 min at 4° C. Pellets were RBC lysed by resuspension in Miltenyi 1x ACK buffer for 5 min at RT, followed by centrifugation at 300xg for 5 min at 4° C. Pellets were washed with R5, filtered, and pelleted. Cell numbers were then quantified on a Countess II automated cell counter, then samples were split for downstream analysis by CyTOF and single cell RNA sequencing.

Tumor exome sequencing

Baseline germline mutation testing was performed at trial screening using the Invitae Common Hereditary Cancers Panel (48 genes) or the Invitae Multi-Cancer Panel (70 genes). Archival formalin-fixed, paraffin-embedded (FFPE) tumor specimens with >30–40% tumor content were sectioned at 5uM by the Herbert Irving Comprehensive Cancer Center’s (HICCC) Molecular Pathology Shared Resource (MPSR). Hematoxylin and eosin (H&E) slides were reviewed and tumor area marked by a board-certified pathologist. Tissue sections were submitted for mutational profiling to Columbia University Irving Medical Center’s Laboratory of Personalized Genomic Medicine, a CAP-accredited and CLIA-certified clinical genomics laboratory. Tissue was prepared for sequencing as previously described.⁵⁷

Targeted next-generation sequencing of tumor DNA and cDNA was performed using the Columbia Combined Cancer Panel (CCCP), a New York State Department of Health-certified panel, covering 467 cancer-related genes Raw sequencing data were mapped to the human reference genome (GRCh37, hg19), and analyzed for single nucleotide variants, small insertions and deletions, copy number alterations, gene rearrangements and fusions, tumor mutation burden, and microsatellite instability using validated clinical pipelines. Identified genomic alterations were reviewed and interpreted for pathogenicity and clinical significance by a board-certified molecular pathologist.

Mass cytometry

Cell pellets were resuspended in R5 supplemented with Rh103 viability stain (1:500) in a 1.5mL protein lo-bind tube and incubated 15 min at 37° C. Cells were pelleted, washed once with R5 media, then resuspended in 49ul Cell Staining Buffer (CSB; Fluidigm) followed by addition of 1ul Fc Block (TruStain FcX; BioLegend). Samples were mixed and incubated for 5–10 min on ice. Separate surface and intracellular antibody cocktails were pre-mixed and stored at −80° C (see Table S2); during the Fc Block incubation cocktails were thawed on ice, the surface mix was spun through a PVDF centrifuge filter for 3 min at 1200rpm, then 50ul surface mix was added to the cell suspension for a final volume of 100ul. Surface staining was conducted on ice for 30 min, then cells were washed twice with 1mL CSB. Cells were fixed by resuspending in 500ul FoxP3 Fixation/Permeabilization buffer (eBioscience) and incubating at RT for 30 min. Fixative was washed out with two successive washing steps with 1mL 1X Perm Wash. For intracellular staining, cells were resuspended in 49ul 1X PW + 1ul heparin on ice. Intracellular cocktails were diluted to 50ul with 1X PW and were spun through a PVDF membrane prior to addition to the cell solution for a final staining volume of 100ul. Cell staining was conducted on ice for 30 min, followed by two successive washes with 1mL 1X PW. Fully stained cells were fixed in 400ul FixIR solution (2.4% PFA +0.25 nM Ir-191/193) at room temperature for 30 min. Samples were washed with 600ul CSB and spun for 4 min at 600 xg, and washed again with 1mL CSB. Samples were either resuspended in CSB and held at 4° C for a maximum of 72 h before being acquired on a CyTOF Helios at the Columbia University Human Immune Monitoring Core (HIMC), or frozen at −80° C in FBS +10% DMSO before being thawed and acquired on a CyTOF Helios at either the Columbia University HIMC or the Mayo Clinic Immune Monitoring Core (IMC). All samples were acquired with, and data was normalized using spiked in four-element EQ Calibration Beads (Fluidigm/Standard BioTools) for longitudinal and cross-platform standardization.

Single cell RNA sequencing and analysis

Quality control and statistical consideration

Cells were processed for scRNAseq using the 10X Chromium 3′ v2 Library and Gel Bead Kit (10x Genomics) according to manufacturer’s instructions at the Columbia University HIMC and as described previously.²⁶ ScRNAseq data were processed with Cell Ranger software at the Columbia University Single Cell Analysis Core. Illumina base call files were converted to FASTQ files with the command “cellranger mkfastq.” Expression data were processed with “cellranger count” on the pre-built human reference set of 30,727 genes. Cell Ranger performed default filtering for quality control. The Seurat package was used for quality control filtering of cells. Cells were filtered for quality control and doublet removal to cells with <25% mitochondrial RNA transcripts, UMI count >1,000 and UMI count <15,000. Pooled distribution across all samples of UMI counts, unique gene counts, and percentage of mitochondrial DNA after QC-filtering is shown in Figure S3A. Filtered cells were scaled by SCTransform, regressing out UMI count and mitochondrial read fraction, with subsequent batch integration performed by Seurat AnchorIntegration algorithm on the top 5000 most variable features. Differential protein activity was assessed by t test with multiple-testing correction for total number of proteins by Bonferroni’s method. Statistical significance was defined as adjusted p-value <0.05. Differential cluster abundance between treatment conditions was assessed by Wilcox test, with multiple testing correction for total number of clusters, also by Bonferroni’s method. All code to reproduce the analysis is publicly available at: https://github.com/Aleksobrad/neoredp_analysis.

Gene expression clustering and initial cell type inference

Combined and normalized gene expression matrix was visualized by UMAP projection and unsupervised clustering was performed using the Louvain algorithm applied to top50 principal components of the full data matrix, with optimal clustering resolution determined by bootstrapped silhouette score optimization, as described in prior work.²⁶ Broad cell types were inferred by SingleR with blueprint-ENCODE reference of sorted cell types, with each cluster consensus-labelled by majority-inferred cell type.²⁸ Further re-labelling of tumor cell clusters (labeled as epithelial by SingleR) was performed by InferCNV analysis, demonstrating consistent regions of chromosomal aberration above background rate seen in other cell types.³⁰

Regulatory Network Inference | For each gene expression cluster, a gene regulatory network was inferred on log-normalized counts by the ARACNe algorithm⁵⁸ to identify downstream transcriptional targets of regulatory proteins. ARACNe was run with 100 bootstrap iterations using 1785 transcription factors (genes annotated in gene ontology molecular function database as GO:0003700, “transcription factor activity,” or as GO:0003677, “DNA binding” and GO:0030528, “transcription regulator activity,” or as GO:0003677 and GO:0045449, “regulation of transcription”), 668 transcriptional cofactors (a manually curated list, not overlapping with the transcription factor list, built upon genes annotated as GO:0003712, “transcription cofactor activity,” or GO:0030528 or GO:0045449), 3455 signaling pathway related genes (annotated in GO biological process database as GO:0007165, “signal transduction” and in GO cellular component database as GO:0005622, “intracellular” or GO:0005886, “plasma membrane”), and 3620 surface markers (annotated as GO:0005886 or as GO:0009986, “cell surface”). ARACNe is only run on these gene sets so as to limit protein activity inference to proteins with biologically meaningful downstream regulatory targets, and we do not apply ARACNe to infer regulatory networks for proteins with no known signaling or transcriptional activity for which protein activity may be difficult to biologically interpret. Parameters were set to zero DPI (Data Processing Inequality) tolerance and MI (Mutual Information) p value threshold of 10−8, computed by permuting the original dataset as a null model.

Protein activity inference and clustering

Inference of protein activity was performed from scaled integrated gene expression matrix and the set of all ARACNe-inferred gene regulatory networks using the VIPER algorithm.^24,26 Normalized Enrichment Scores of protein activity were successfully inferred on a per-cell basis for 3,437 regulatory proteins. This protein activity matrix was re-clustered in Seurat using the Louvain algorithm on top50 principal components and clustering resolution parameter optimized by silhouette score, as above, with two-dimensional data projection visualized by UMAP. Differential protein activity between clusters was computed by Wilcoxon rank-sum test with multiple testing correction. Differential abundance of each cluster per patient by treatment group was also assessed by Wilcoxon rank-sum test, as well as differential abundance of each cluster within the combination-treatment group between patients with and without disease recurrence. Cell clusters inferred by SingleR as lymphoid lineage, myeloid lineage, or tumor cells, respectively, were each isolated for further sub-clustering analyses.

Single-cell RNA-seq T/NK subclustering

Subsetting lymphoid cells from initial analysis resulted in a VIPER matrix of 14,208 cells, projected into their first 50 principal components using Seurat’s RunPCA function, followed by using Seurat’s RunUMAP function with method umap-learn. Unsupervised clustering of lymphoid cells was performed by the same silhouette score optimized Louvain approach as above. In addition to SingleR cell type labeling, we applied GSEA on cell-by-cell basis using an expression-based T-reg marker gene set previously published from Obradovic et al. Cell 2021,²⁶ to identify the T-reg cluster. VIPER activity of manually selected relevant markers was also visualized for lymphoid cell subtype identification of all other clusters. Differential abundance of each cluster as percent of lymphoid cells was assessed by treatment group and by recurrence vs. non-recurrence within the combination-treatment group, by Wilcox rank-sum test. The ADT-only treatment group was not stratified due to insufficient recurrences for statistical power and recurrence data were not available in the untreated group. Further testing in the combination treatment group of association between Treg abundance and time-to-recurrence was performed by Kaplan-Meier analysis, with optimal cut-point of high-Treg vs. low-Treg samples determined by log rank maximization, and p-value assessed by Cox regression.

Single-cell RNA-seq myeloid cell subclustering

Isolation of myeloid cells (clusters identified by SingleR as predominantly Monocyte or Macrophage) yielded 5771 Myeloid cells, subclustered in the same manner as lymphoid cells above. RunPCA and RunUMAP from Seurat were used to compute the first 50 principal components and UMAP respectively. Upon sub-clustering, a subpopulation identified as fibroblasts (n = 183) by SingleR with the BlueprintEncode reference dataset from Celldex was removed from downstream analysis. A 7-cluster unsupervised clustering solution was selected as optimal by maximization of silhouette score. Clusters were then annotated by cell type with a combination of visualizing manually selected phenotypic markers and assessing Gene Set Enrichment of differentially active proteins by cluster.

Correlation between FCGR3A activity and Treg frequency

In order to assess for relationship between FCGR3A protein activity among myeloid cells and Treg frequency, we performed correlative analysis between Treg frequency as a percent of lymphoid cells for each patient against average FCGR3A protein activity for myeloid cells as a whole in the same patient, as well as for the specific subset of TREM2+ TAMs. Stratifying these analyses by treatment group shows treatment-dependent association between FCGR3A activity and Treg abundance, such that in combination-treatment alone higher myeloid FCGR3A activity was found to associate with lower Treg abundance.

Single-cell RNA-seq CD8⁺ T cell subclustering

Upon sub-clustering lymphoid cells, we further isolated the two identified clusters of CD8 T-cells (CD8-1 and CD8-2) for deeper sub-phenotyping. This yielded a VIPER protein activity matrix for 7,475 cells. Re-clustering on the top50 principal components of this matrix yielded a four-cluster solution such that one small subset of n = 187 cells appeared to be naive CD8s or potential contamination and were removed. This conclusion was supported by GSEA on the gene expression signatures of these cells with the Naive CD8 gene set from Zheng et al. 2021.³⁷ The remaining true CD8⁺ cells were re-scaled and re-clustered yielding a three-cluster solution. These clusters were labeled as “Central Memory”, “Effector” and “Effector/Resident Memory” by Gene Set Enrichment Analysis. “Memory vs. Effector” and “Effector vs. Memory” signatures were generated using a Mann-Whitney U-Test between the gene expression profiles of CD8⁺ Trm and CD8⁺ cytotoxic populations reported by Tuong et al. 2021.³³ The top genes from each of these signatures and in the signatures from Zheng et al. 2021³⁷ were supplied as gene sets to GSEA to compute enrichment on the gene expression profiles of our CD8⁺ cells. Relevant manually curated marker genes were further visualized by VIPER activity per cluster. Cluster frequencies by treatment group were then calculated to identify changes in these CD8⁺ subtypes in response to treatment.

Immunofluorescence

Archival FFPE prostate tumor specimens were selected by a board-certified pathologist. Immunoblank slides and representative adjacent H&E slides were sectioned at 5uM by the Herbert Irving Comprehensive Cancer Center’s Molecular Pathology Shared Research (HICCC MPSR). Tissue sections were deparaffinized using a standard series of xylene and ethanol washes and rinsed with distilled water. Antigen retrieval was performed by heating the slides in citrate buffer at pH of 6.0 for 20 min in a pressure cooker. Slides were incubated overnight at 4°C with primary antibodies against CD4 (1:200, Abcam) and FOXP3 (1:100, Invitrogen). Subsequently, the slides were incubated for 30 min at room temperature with horse anti-mouse antibody (1:200, Vector Laboratories), and then for 30 min with Streptavidin, Alexa Fluor 594 conjugate (1:1000, Invitrogen). Next, the slides were incubated for 30 min at room temperature with goat anti-rabbit antibody (1:200, Vector Laboratories) and then for an additional 30 min with Streptavidin, Alexa Fluor 488 conjugate (1:300, Invitrogen). Slide sections were mounted at room temperature with 25uL of VECTASHIELD Vibrance Antifade Mounting Medium with DAPI (Vector Laboratories) and coverslipped.

Fluorescent images were acquired by the PhenoImager HT Instrument (Akoya Biosciences). Onboard spectral unmixing was performed using the PhenoImager HT 2.0 software (Akoya Biosciences). The images were analyzed with QuPath version 0.5.1.⁵⁴ Regions of interest were generated referencing tumor-marked H&E slides from adjacent sections. To further refine tumor region selection, ROI annotation was guided by predictions from the prostate-tumor-resnet34.tcga-prad model in the WSInfer QuPath extension— a deep learning model trained on TCGA prostate whole slide images to differentiate tumor from benign tissue.⁵⁵ Nuclear and cell segmentation were performed using the QuPath extension StarDist 0.5.0 with the pretrained model dsb2018_heavy_augment.pb.⁵⁹ Following cellular segmentation, object classifiers were trained on each image to quantify Tregs (CD4+/FOXP3+) per mm² or as a percentage of all DAPI+ cells. We obtained adequate tissue for staining and analysis from 12 untreated patients, 7 patients in the ADT arm, and 10 patients in the ADT + anti-CTLA4-NF arm.

Animal studies

For tumor implantation, Myc-CaP cells at 70–90% confluence were harvested with 0.05% trypsin (Corning; Corning, NY), washed with PBS, counted, and resuspended at 10 × 10⁶ cells/mL in ice-cold PBS. On day 0, cohorts of 8–10-week-old FVB mice were implanted on the right flank with 1 × 10⁶ MycCaP cells in 100 μL PBS. Tumor measurements were recorded in X, Y, and Z dimensions in a non-blinded fashion (largest diameter, smallest diameter, and largest height, respectively) every 2–3 days by digital caliper and tumor volume was calculated by multiplying X∗Y∗Z. Using the final pre-treatment tumor measurements, mice were manually grouped into cohorts of n = 15 animals to normalize the mean tumor volumes and distribution of tumor volumes between the groups. These manually curated cohorts were randomized in an unsupervised fashion to treatment groups using a random number generator. On day 24 when tumors were approximately 200mm³ on average, mice received a single subcutaneous injection in the neck scruff of 625 μg degarelix acetate (ADT; Cayman Chemical; Ann Arbor, MI) resuspended in 100 μL sterile water per mouse. Appropriate treatment groups additionally received anti-CTLA4 (D) (9D9-mIgG2a) or anti-CTLA4 (ND) (9D9-mIgG2a-LALAPG) as indicated by intraperitoneal (IP) injection at 100 μg per mouse in 100 μL sterile PBS on days 24, 27, and 30. Antibodies were purchased from BioXCell (Lebanon, NH). For flow cytometry analysis, cohorts of n = 7 animals were selected at random by random number generator from each group for euthanasia on day 32 for tissue collection and processing for flow cytometry (see below). The remaining n = 8 mice in each group were followed for survival and were deemed eligible for euthanasia when tumor volume exceeded 1,000 mm³ or tumor ulceration exceeded 5 mm in diameter. Independent replicate experiments utilized n = 5 animals per treatment group for each readout. Sample sizes chosen were based upon extensive prior experience assessing response to anti-CTLA-4 immunotherapy in the MycCaP model[ref]. Replicate experiments utilized n = 5 mice per group based upon effect sizes observed in initial experiments. See Statistical Analyses section and figure legends for details on statistical testing. Data from all mice were included in survival analyses. See Analysis of Flow Cytometry Data section below for details on inclusion and exclusion criteria for immune profiling data derived from these animals.

Spectral flow cytometry

Tissue harvest and dissociation

Following mouse euthanasia, tumor-draining inguinal lymph nodes were dissected and placed in 48-well plates containing 150 μL R10 media on ice. Tumors were harvested, massed, and up to 50 mg tumor was diced and placed in X-Vivo 15 media (Lonza; Basel, Switzerland) in 15mL Eppendorf tubes on ice. DNase (40 μL of 20 mg/mL solution; Roche; Basel, Switzerland) and Collagenase D (125 μL of 40 mg/mL solution; Roche; Basel, Switzerland) were added to tumor samples prior to incubation on a shaker at 37° C for 30 min. Digest reactions were quenched by adding 5mL R10 media. Tumor digests were filtered through 70 μm filters (Miltenyi; Bergisch Gladbach, GE), remaining undigested tissue was mashed with the handle end of a 3mL syringe plunger, then filters were washed with 5mL R10 prior to centrifugation. Lymph nodes were physically disaggregated with a 1mL syringe plunger in the 48-well plate. All samples were then transferred in R10 into U-bottom 96-well plates for staining.

Staining protocol

Samples were washed with PBS. Dead cells were stained by resuspension in 100 μL PBS + Live/Dead Fixable Blue dye (1:500; Invitrogen; Waltham, MA). This and all further staining or fixation steps were performed for 30 min at room temperature on a plate shaker protected from light, unless specified otherwise. Samples were washed 2x with PBS, then were resuspended in FACS (PBS +3% FBS + 1mM EDTA +10mM HEPES) supplemented with TruStain FcX (1:50; BioLegend; San Diego, CA) and TruStain Monocyte Blocker (1:20; BioLegend; San Diego, CA) and placed on ice. Surface antibodies were prepared at optimal dilutions (see Table S2) in FACS supplemented with Brilliant Stain Plus buffer (BD; Franklin Lakes, NJ), then were added to samples in blocking solution. After staining, samples were washed 2x with FACS buffer and fixed in 100 μL of FoxP3 Fixation/Permeabilization Kit buffer (eBioscience; San Diego, CA). Samples were washed twice with 1X Permeabilization buffer (1X PW) then were stained in 1X PW plus intracellular antibodies. Plate(s) were sealed and intracellular staining was performed overnight at 4° C on an orbital shaker. Samples were then washed twice in FACS then resuspended in 200 μL FACS and were immediately acquired on a Cytek Aurora 5-laser (UV/V/B/YG/R) cytometer. Simultaneously stained splenocyte samples or SpectraComp beads (Slingshot Biosciences; Emeryville, CA) were utilized for single stain controls as indicated in Table S1.

Analysis of mass cytometry and flow cytometry data

All analysis was performed using FlowJo (BD; San Diego, CA). Live, single cell, CD45⁺ events from all files were identified by manual gating using conventional gating strategies for CyTOF and flow cytometry data as shown in Figures S5 and S6. Samples with fewer than 1,000 live CD45⁺ single cell events were excluded from further analysis. All live CD45⁺ single cell events in the CyTOF data were concatenated for high dimensional analyses, while up to 50,000 live CD45⁺ single cell events from the flow cytometry dataset were downsampled and concatenated for downstream analysis. Canonical immune lineages in both datasets were defined in a semi-supervised manner utilizing FlowSOM.³¹ Data were purposely overclustered to 30 FlowSOM-derived clusters based on expression of lineage-defining markers in each panel. These clusters were then manually collated based on lineage marker expression to obtain the minimal possible number of canonical immune lineages. Clusters were embedded in two-dimensional space for visualization using the supervised UMAP function in FlowJo, which performs dimensionality reduction based on selected lineage-defining marker expression with incorporation of FlowSOM-derived clusters as a weighted parameter. For discovery-based subclustering of individual lineages (e.g., CD4 Tregs), an additional round of semi-supervised FlowSOM was performed in an analogous manner as described above; involving FlowSOM-derived overclustering based on phenotypic markers differentially expressed by cells within the lineage of interest, followed by manual collation of clusters into minimal number of phenotypically distinct subclusters. For visualization of discovery-based subclustering within lineages, we utilized PaCMAP⁶⁰ dimensionality reduction to embed data in two dimensions. Absolute cell numbers and mean geometric fluorescent intensity (gMFI) data was exported for each cluster for downstream analysis. Phenotypic marker data was scaled based on the minimum and maximum gMFI values of each given marker across all live CD45⁺ cells or within the lineage of interest, as indicated in the figure legends.

Quantification and statistical analysis

Baseline patient characteristics and study outcomes were computed and visualized utilizing the gtsummary package (v 2.4.0)⁶¹, in Rstudio (R version 4.3.1, 2023-06-16). Standardized mean differences were calculated using the smd package (version 0.8.0).⁶² Mutational profiles were visualized using Oviz-Bio’s LandScape (v1.1.1).⁶³ Stratification for Kaplain Meier analyses was performed using log rank maximization algorithm as implemented in the survminer package. CyTOF and spectral flow cytometry statistical testing was conducted in GraphPad Prism (v10.2). All other quantitative and statistical analyses were performed using the R computational environment and packages described above. Differential gene expression was assessed at the single-cell level by the MAST single-cell statistical framework as implemented in Seurat v4,⁶⁴ and differential VIPER activity was assessed by Wilcoxon rank-sum test, each with Benjamini-Hochberg multiple-testing correction. Unless otherwise noted, measurements were taken from distinct samples and comparisons of cell frequencies by treatment group were performed by two-tailed Welch’s t test or non-parametric Wilcox rank-sum test, and survival analyses were performed by log rank test. In all cases, statistical significance was defined as an adjusted p value less than 0.05. Details of all statistical tests used, sample size information, and descriptions of precisions measurements can be found in the corresponding figure legends.

Additional resources

Clinical trial registration

The trial was registered with ClinicalTrials.gov on March 6, 2020, with the ClinicalTrials.gov ID NCT04301414.

Published: February 26, 2026