Temporal dynamics of the immune response to neoadjuvant androgen deprivation therapy suggest a window of opportunity for checkpoint inhibitor therapy in prostate cancer

Abstract

Background

Novel therapies to prevent lethal castration-resistant prostate cancer in response to standard-of-care androgen deprivation therapy (ADT) are required. Unfortunately, most prostate cancers are “immune cold” and fail to respond to checkpoint inhibitors (CPIs). To assess whether ADT induces changes that enable more effective CPI therapy, we examined the tumor immune microenvironment (TiME) following neoadjuvant ADT (nADT).

Methods

Radical prostatectomy specimens from 43 nADT-treated patients were stratified into three duration groups and compared with each other and with matched controls. RNA sequencing and quantitative multispectral immunofluorescence (qmIF) staining were performed to analyze transcriptomic and TiME abundance and cellular spatial relationship differences after nADT.

Results

Immune and inflammatory pathways, particularly of antigen presentation and adaptive immune response, were increased, most notably in tumors receiving 3–5 months nADT. qmIF revealed a complex temporal response in the TiME, with a dramatic influx of cytotoxic T lymphocytes (CTLs) and T-helper cells after 3–5 months of nADT. However, after 6 months nADT, M2-like tumor-associated macrophages (TAMs) and regulatory T cells were strikingly increased while CTLs decreased. Spatially, CTLs and T-helper cells clustered near tumor cells at 3–5 months nADT, but were replaced by M2-TAMs in tumors receiving ≥6 months of nADT.

Conclusions

These data reveal the induction of a bi-phasic response in the TiME: robust CTL activation 3–5 months after nADT is initiated, followed by myeloid immunosuppression in tumors receiving prolonged nADT. This ADT-induced reprogramming of the TiME suggests a critical window of opportunity where short-duration ADT might augment CPI efficacy, converting cold into immunologically responsive tumors.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Immune checkpoint inhibitors (CPIs) have not been effective in treating most human prostate cancers. Patients with advanced and recurrent prostate cancer are treated with androgen deprivation therapy (ADT), and although ADT is known to influence immune signaling, most evidence of ADT effects on prostate tumor derives from preclinical models, or single time point clinical studies that capture static views of the immune landscape. Consequently, how immune–tumor interactions evolve during ADT remains unclear.

WHAT THIS STUDY ADDS

• This study describes the temporal dynamics of the immune response of primary prostate cancers to neoadjuvant ADT (nADT), thus revealing key information to develop a strategic approach to improve the efficacy of CPIs in prostate cancer. After several months of nADT, inflammation and immune-related pathways were activated, accompanied by a robust infiltration of both CD8⁺ and CD4⁺ T cells into prostate tumors, suggesting effector T-cell education and activation. In contrast, 6 or more months of nADT led to an immunosuppressive shift, evidenced by increased M2-like tumor-associated macrophages and regulatory T cells.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our findings suggest a critical window of opportunity following nADT for initiating CPIs. This provides a rationale for the precise sequencing of nADT and CPI regimens to maximize therapeutic benefit.

Introduction

The mainstay therapy for advanced and recurrent prostate cancer (PrCa) is androgen deprivation therapy (ADT). However, extended ADT leads to androgen pathway modulator-resistant PrCa (formerly referred to as castration-resistant PrCa, CRPC), which has a poor prognosis. Trials of checkpoint inhibitor (CPI) therapies alone or in combination with androgen signaling pathway inhibitors in the CRPC setting have not been effective.^1–3 Indeed, fewer than 10% of CRPC tumors express an immune inflamed signature or other biomarkers predictive of CPI response, thus PrCa is classified as immune “cold”.⁴ However, the tumor immune microenvironment (TiME) in PrCa may be more aptly characterized as similar to the “immune excluded” immunophenotype described by Chen and Mellman.⁵ Indeed, we previously reported that cytotoxic T lymphocyte (CTL) density correlates with survival in primary PrCa,⁶ consistent with the possibility that immunosuppressed or partially-activated T cells reside in primary PrCas.

Successful cancer immunotherapy depends on forming intratumoral immune triads of antigen-presenting cells, CD4⁺ helper, and CD8⁺ T cells to educate T effector cells, enabling eradication of tumor.^7–9 Androgens and other steroid hormones regulate the abundance and activity of most effector (T, natural killer (NK), B) and regulatory immune cells (regulatory T cells (Treg), tumor-associated macrophages (TAM), Myeloid-derived suppressor cells (MDSC), both normally and in disease.¹⁰ Because ADT is well-established as the primary therapy for advanced or recurrent PrCa, many investigations have sought to characterize the effects of ADT on the PrCa TiME. Multiple reports demonstrate that neoadjuvant ADT (nADT) induces a complex infiltrate containing each of these immune cell subtypes in PrCa,^11–13 suggesting ADT might have immuno-activating effects. ADT increases major histocompatibility complex (MHC)-related gene expression, stimulating antigen presentation and antitumor immunity.¹⁴ However, this stimulation is transient, and MHC activity is subsequently repressed in CRPC.¹⁴ T-cell activation occurs within a few weeks of nADT initiation in patients with PrCa, as indicated by infiltration of CD4⁺ and CD8⁺ T cells, with increased effector cell response to tumor antigens.^{15 16} Further, within 2 months of initiating nADT for locally advanced PrCa, immune and cytokine pathways, as well as CD8⁺ T cells, were enriched,¹⁷ highlighting the potential for nADT to alter immune modulation. In contrast, there have been multiple reports that ADT is accompanied by increasing Tregs and M2-like macrophages,^18–20 which correlate with poor prognosis for men with PrCa.^{13 21} These reports of both immune activation and suppression in response to ADT lack carefully executed temporal studies. Therefore, better defining ADT-induced changes in the prostate tumor microenvironment (TME) may enable combination therapy strategies to improve PrCa response to CPIs.²²

In this report, we analyzed transcriptomic profiles, imputed immune-cell content, and spatially identified immune-cell phenotypes and temporal changes in surgical specimens from a cohort of men treated with nADT prior to prostatectomy. Several months of nADT induced a transient immune activation that was followed by a compensatory immunosuppression. This suppressive state may render immune oncology therapeutics ineffective, and thereby drive PrCa resistance and recurrence. These findings illustrate the complex interplay between androgen signaling in the PrCa TiME and immune modulation. Importantly, nADT’s immune-activating properties suggest a limited-duration window of opportunity that might be a target for combination CPI therapy before an immunosuppressive TiME emerges.

Materials and methods

Retrospective cohort

We identified patients with PrCa treated with ADT prior to radical prostatectomy performed at the Roswell Park Comprehensive Cancer Center. Many of these patients presented for a second opinion after initiating ADT (via a luteinizing hormone-releasing hormone (LHRH) agonist with or without the androgen receptor antagonist bicalutamide) in the community. After a chart review to determine the duration of nADT exposure, clinical and pathologic data were abstracted from patient records. Data collected included age, race, year of prostatectomy, prostate-specific antigen (PSA) levels, Gleason score, clinical and pathological stage. A cohort of 43 cases of men diagnosed with PrCa who received nADT for up to 1 year prior to prostatectomy was selected. These men underwent prostatectomy between 1995 and 2019 (mean 2006, IQR 6 years). Each patient had serum PSA levels determined prior to initial nADT, prior to prostatectomy, and during long-term follow-up. A control cohort of 43 cases matched 1:1 on age, Gleason grade and stage, and year of prostatectomy who did not undergo nADT prior to prostatectomy was identified for comparative analysis (table 1). Patients in either cohort with a previous history of radiotherapy or chemotherapy were excluded from this study.

Table 1. Clinical features and pathological data for cases/controls.

	nADT cohort (43)	Controls (43)	P value
Race	%	%	0.5698
White	88.4 (38)	90.7 (39)
Black/AA	11.6 (5)	9.3 (4)
Age	Years	Years	0.4185
Range of ages	50–71	48–69
Median age at surgery	58	58
ADT	%	%
2 months or less	41.9 (18)	0
3–5 months	32.5 (14)	0
6+months	25.6 (11)	0
PSA	(ng/dL)	(ng/dL)	0.1685
Pre-surgery PSA average	9.8	7.1
Pre-surgery PSA range	0.16–54.26	0–32.5
Clinical staging	%	%	0.4977
Tx	0 (0)	2.3 (1)
T1c	62.8 (27)	62.8 (27)
T2a-2c	37.2 (16)	32.6 (14)
T3b	0 (0)	2.3 (1)
Pre-surgery Gleason score	%	%	0.2617
Gleason 6	25.6 (11)	25.6 (11)
Gleason 7
3+4=7	37.2 (16)	30.2 (13)
4+3=7	7.0 (3)	14.0 (6)
Gleason 8–10	30.2 (13)	30.2 (13)
Pathological staging	%	%	0.05107
pT1c	0 (0)	2.3 (1)
pT2a-2c	53.5 (23)	72.1 (31)
pT3a	27.9 (12)	16.3 (7)
pT3a, N1	2.3 (1)	0 (0)
pT3b	11.6 (5)	7.0 (3)
pT4	4.7 (2)	2.3 (1)
Post-surgery Gleason score	%	%	0.6497
Gleason 6 and under	16.3 (7)	13.9 (6)
Gleason 7
3+4=7	30.2 (13)	44.1 (19)
4+3=7	25.6 (11)	21.0 (9)
Gleason 8–10	25.6 (11)	21.0 (9)
Unknown	2.3 (1)	0 (0)

P values calculated using paired t-tests.

AA, African American; ADT, androgen deprivation therapy; nADT, neoadjuvant ADT; PSA, prostate-specific antigen.

For each of the 86 cases, serial 5 µm thick sections were obtained from representative formalin-fixed, paraffin-embedded (FFPE) blocks and used for H&E staining, immunohistochemistry (IHC), and for quantitative multiplex immunofluorescence (qmIF) analysis, as well as for RNA and DNA extraction. H&E-stained slides were reviewed by a genitourinary pathologist to estimate tumor fraction, which ranged from ~25% to 75%. Transmembrane serine protease 2 (TMPRSS2) IHC staining (Clone AL20, cat. PT-mAb-AL20, 1:800 dilution) was performed on FFPE sections by the Roswell Park Pathology Network Shared Resource. Slides were digitally scanned and exported using Aperio ScanScope and ImageScope (Aperio Technologies), and exported images analyzed for H-score using ImageJ software.

Genomic mutational analysis

Transcriptomic and targeted mutation analyses were performed using Agilent’s HS2 FFPE RNA reverse transcription and gDNA fragmentation and capture system using the Agilent SureSelect Cancer Comprehensive Genomic Profiling (CGP) Assay. The CGP panel targeted 679 genes for Single Nucleotide Variant (SNV) analysis, 80 of which were also analyzed for gene fusions (including TMPRSS2), 26 for copy number variation (CNV) (including BRCA1/2, MYC, and PTEN) and exons of 12 genes specifically for previously reported chromosomal translocations. Additionally, tumor mutational burden (TMB) and microsatellite instability (MSI) were measured for each sample. After capture, barcoded DNA was amplified, purified, and pooled. The pooled DNA was sequenced on an Illumina NovaSeq S1 flow cell using a standard paired-end 2×150 bp protocol. Sequencing was performed to an average depth of 40 million reads. Demultiplexed sequencing data from each patient was analyzed for germline and somatic mutation using the Alissa Clinical Information platform (Agilent). SNVs were classified as likely pathogenic or not using ClinVar and a minimum 3% variant allele frequency was used to identify potentially significant variants. Since a paired peripheral germline comparison sample was not available for sequencing, 16 mutations per megabase DNA was used as the cut-off for TMB high, and 20% of markers showing instability as the cut-off for MSI high.

RNA sequencing and analysis

Total RNA from the 86 matched specimens was sequenced using Illumina’s NovaSeq 6000. The quality of raw reads was assessed by FastQC V.0.11.5 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were mapped to the human reference genome GRCh38 (release 27) using STAR V.2.6.0a.²³ Gene expression was then quantified using the RSEM package.²⁴ Estimated gene counts from RSEM were filtered and upper quartile normalized using edgeR.²⁵ Differential gene expression analysis was performed using the limma package after voom transformation followed by linear regression.²⁶ The significance threshold was set to false discovery rate (FDR)/adjusted p value of <0.05 with a fold change greater than 2 (FDR <0.05 and |log2 fold change|>1). Transcripts per million of selected genes were analyzed using R packages. The TiME in the nADT and matched-control groups were compared using two independent tools, CIBERSORT²⁷ and ImmuCellAI (http://bioinfo.life.hust.edu.cn/web/ImmuCellAI/).²⁸ Gene Set Enrichment Analysis (GSEA)²⁹ for the hallmark gene sets from the molecular signatures database³⁰ was performed using the fgsea³¹ Bioconductor package in R. GSEA for Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), WikiPathways and Reactome Pathway gene sets was performed using the gseGO, gseKEGG, gseWP, and gsePathway functions of the R package clusterProfiler.³²

Quantitative multiplex immunofluorescence analysis

qmIF staining was performed using the Vectra system (PerkinElmer/Akoya Biosciences) to examine 5 µm sections of prostate tissue that were stained manually using the Opal 6-Plex Manual Detection Kit with tyramide signal amplification (Akoya Biosciences). The multiplex panel consisted of DAPI, CD3, CD8, CD68, HLA-DR, FoxP3, and NKX3.1 (detailed in online supplemental table S1). The multiplex-stained slides were imaged using a Vectra 3 Polaris Automated Quantitative Pathology Imaging System. An initial quality control assessment of antibody expression in the qmIF panel was compared with that in single stained slides on control (tonsil) and prostate tissue by a pathologist. Staining patterns did not differ between the single and multiplex scanned slides. This combination of markers distinguished CTLs, helper and Tregs, tumor and stromal cells, and M1-like and M2-like TAMs. Online supplemental table S2 defines the marker combinations used in this study.

The multispectral library slides were unmixed into eight channels using inForm software V.2.8: DAPI, HLA-DR-OPAL480, FOXP3-OPAL520, CD8-OPAL 570, CD68-OPAL 620, CD4-OPAL 690, NKX3.1-OPAL 780 and Auto Fluorescence and were fused with Phenoimager software V.3.0 to create a single q.tiff file for each sample. To reduce sampling-induced variability on tissue sections with infiltrative tumor, and to capture surrounding tumor-adjacent and normal prostate tissue TiME, regions-of-interest (ROIs) covering the whole tissue section were analyzed. After acquiring the panoramic low-magnification images at 4×, images encompassing the specimen using 1 mm² sized ROIs were acquired, and areas of poor quality and voids were excluded using Phenochart (PerkinElmer) software.³³ In total, 1,744 ROIs from 86 specimens were acquired for analysis. Cell phenotyping was based on biomarker expression within cells, nuclei and membranes using inForm Cell Analysis. Cells were then segmented into either tumor or stromal compartments, with NKX3.1 expression identifying tumor, using the inForm Tissue Finder tool. The density of each cell phenotype (tumor, stroma, and the delineated immune cell types) was then assessed (as cells/mm²).

qmIF spatial analysis

The cellular neighborhood identification was performed using phenoptrReport (R, V.4.1.3), (https://akoyabio.github.io/phenoptr/). Specifically, image datasets, marked with specific cellular markers, and the “Cell Classification” tool were used to label and segment cell types. Online supplemental table S3 defines distances measured in this study. Distance between two cell subtypes was calculated using the x‐coordinates and y‐coordinates from the inForm raw data and nearest neighbor per‐cell distance calculated using the Phenoptr GUI. A detailed spatial analysis was then conducted to examine the interactions between pairs of cell types (M2 to tumor and CTLs to M2, CTL to tumor, T-helper and CTL dyad to tumor (only if the CTL-T-helper cell dyad distance was less than 15 µm⁷) through a mixed radii nearest neighbor approach, capturing dynamic immune interactions and changes by treatment cohort. To identify shifts in immune cell relationships and their proximities over the course of treatment, the mixed radii nearest neighbor tool was used to measure neighboring cell proximities. Tabulated results showing the distribution and frequency of interactions at each radius across different treatment time points were generated using PhenoptrReports.

Statistics and bioinformatics

For qmIF, multiple ROIs per specimen were acquired to capture spatial heterogeneity, then aggregated using the inForm PhenoptReport workflow into a single specimen-level value for each marker/cell population. Statistical analyses were performed on these aggregated values of all ROIs per patient specimen. nADT-treated specimens were compared with their predefined matched control specimen, and comparisons between groups were performed using two-tailed paired Student’s t-tests using Origin pro software. Comparisons of ratios were calculated for each specimen and compared between nADT-treated tumors and matched controls across treatment-duration cohorts using the Wilcoxon signed-rank test after log-transformation.

Cell-to-cell spatial distance analyses were performed using raw distance measurements between defined cell populations. Distances were measured in micrometers (µm), aggregated as specimen-level mean values, and compared between treatment and matched control groups using the Wilcoxon signed-rank test. For outcome-associated analyses, patients were stratified by survival status, and mean spatial distances compared between survival groups. Distance distributions were visualized using kernel density estimation in OriginPro. Smoothed probability density curves from the original distance measurements illustrate shifts in cell-to-cell spacing across cohorts. These were not analyzed as independent biological measurements.

Clinical parameter and biomarker comparisons were carried out using paired t-tests in R. Statistical analyses of inForm defined cell density changes between samples in treated and untreated cohorts were conducted using a paired t-test using Origin Pro software (OriginLab). Correlation of CTL density (dichotomized as the upper quartile vs lower 75%-tile) with clinical outcome was determined by Kaplan-Meier testing with the alive patients censored, and a multivariate logistic regression model was used to examine the effect of CTLs density on patient survival.

Data availability

In accordance with the Health Insurance Portability and Accountability Act, de-identified data presented in this study are available on request from the corresponding authors with institutional approval. However, de-identified RNA-sequencing data are available in GEO, GSE337290, and exome data at NCBI Sequence Read Archive, PRJNA1484982. All R scripts used in this study are available from the authors on request.

Results

Clinical characteristics of cases and matched controls

Prostatectomy specimens from 86 men from 49 to 71 years old were included in this study; 43 of whom received neoadjuvant ADT prior to prostatectomy (table 1). These 43 patients received a LHRH agonist, and 16 of these nADT treated patients also received the non-steroidal anti-androgen bicalutamide. As diagrammed in figure 1A, the nADT cohort was further divided into three subgroups based on treatment duration: 2 months or shorter (n=18), 3–5 months (n=14), and 6 months or longer (n=11). There were no differences in pretreatment PSA levels between nADT duration groups (online supplemental figure S1A, and data not shown). Specimens from men who did not receive nADT, but were treated directly by prostatectomy were matched 1:1 with each of the nADT cases by age, race, stage/grade, and year of prostatectomy as a control cohort (n=43). Both cohorts were comprised principally of white men (77/86). The majority of their PrCa (54/86) were stage T1c, and the men were younger than average for PrCa diagnosis (median age of 58 in both cohorts). Between the nADT and control cohorts, there were no differences in pretreatment PSA, nor in demographics, biopsy Gleason grade or stage, nor were there any differences in the final stage or grade of the patient tumors recovered following prostatectomy (table 1 and online supplemental figure S1A).

Figure 1. Duration of nADT exposure and genetic variant characterization of prostatectomy specimens from treated and matched control cohorts. (A) Swimmer plot illustrating duration of nADT prior to prostatectomy. Specimen Gleason score is indicated by colors, symbols indicate nADT therapy regimen, in the inset box. (B) Table of samples indicating nADT exposure (n=39, red) or control samples (n=43, green); TMB and MSI quantitation (color indicates mutations per megabase, and MSI ≥20 (high)). Pathogenic mutations (yellow); CNV, (gene loss in blue/ gene amplification in pink); and gene fusions (purple) for the indicated genes, among 82 samples indicated across the top of the table. CNV, copy number variation; LHRH, luteinizing hormone-releasing hormone; MSI, microsatellite instability; nADT, neoadjuvant androgen deprivation therapy; TMB, tumor mutational burden.

Time to biochemical recurrence, metastasis, and overall survival

The median PSA follow-up post-prostatectomy was 7.7 years (range 0.1–20.9) for the combined cohorts, 6.7 years for the nADT cohort, 9.7 years for the control cohort. Seven patients were lost to PSA follow-up (two in the nADT cohort and five in the control cohort). 41 patients had biochemical recurrence (BCR), defined as PSA >0.2 ng/mL (20 in the nADT cohort and 21 in the control cohort). The median time to BCR was 16.3 years in the nADT cohort, but was not reached in the control cohort (online supplemental figure S1B). 33 patients (38.4%) developed metastases. The median time to metastasis was 2.3 years (95% CI 2.1 to 3.6), and did not differ between the cohorts. There were 10 deaths specifically from PrCa among these patients, 6 in the nADT cohort and 4 in the control cohort. The median overall survival (OS) was 16.1 years in the nADT cohort and 17.3 years in the untreated control cohort (online supplemental figure S1C).

Genomic characteristics of nADT and control cohorts

Most (82/86) specimens from the nADT cases and controls yielded sufficient RNA and DNA for analysis of PrCa relevant genomic alterations. A total of 679 genes were assayed (see Materials and methods) and 93% of samples yielded >100× coverage for these genes, with an effective genomic coverage of 1598±2.3 Mb each. Clinical genomics analysis using the Agilent interpretive platform revealed frequent (18/82) MYC amplification, and a variety of lower frequency amplifications and deletions in cancer-related genes (figure 1B). The PTEN/PI3K pathway was also frequently affected (14/82), but this is likely an underestimate. Specifically, the CGP assay identified loss of PTEN in only two samples, suggesting technical limitations of the assay, as PTEN deletion has been reported to occur in up to 20% of PrCa.³⁴ DNA repair genes (BRCA2, BRCA1, BRIP1) were disrupted in the germline of five patients, and somatically in an additional seven samples. One sample displayed high MSI; another showed very high TMB and eight additional samples showed modest TMB (16–45 non-synonymous mutations or indels/MB). Terminating and frameshift mutations in PTEN, BRCA1 and BRCA2 were frequent in those samples with high TMB or MSI. TMPRSS2 fusions were identified in 14 samples (11 fusions with ERG, 1 each with ETV1, MRPS18C, and NBEAL1).

TMPRSS2 expression was decreased by nADT

Since TMPRSS2 is androgen-regulated in prostate tumors,^{35 36} we employed it as a biomarker to measure nADT efficacy. TMPRSS2 protein expression was quantitated by IHC staining of the 43 pairs of prostatectomy specimens from the nADT cohort and their matched controls (online supplemental figure S2A). Pairwise comparison of the resulting H-scores indicates TMPRSS2 expression was reduced by nADT (online supplemental figure S2B). This reduction was sustained in the longer duration nADT groups (online supplemental figure S2C). Thus, TMPRSS2 protein expression indicates the nADT was effective in each of these duration groups prior to prostatectomy.

nADT regulates AR, immune and inflammatory gene pathways in prostate tumors

To decipher the effects of nADT on the tumor, we then compared the transcriptomes from the 43 nADT samples to the paired control samples. Consistent with the IHC staining result, the RNA level of TMPRSS2 was strongly reduced and expression levels of other known androgen receptor (AR)-regulated genes, including KLK2, KLK3, KLK4 and AMACR were also decreased (blue in figure 2A). Expression of both chemokines CCL4/14/22 and chemokine receptors CCR4/7/10, important in T-cell trafficking, were increased (red in figure 2A). The transcriptome data were then interrogated using GSEA to identify key molecular signaling or biological pathways altered by nADT. As expected, the ANDROGEN RESPONSE pathway was downregulated in nADT samples, which also showed enrichment of the INFLAMMATORY RESPONSE pathway (figure 2B). Further, Hallmarks of inflammation (TNFA SIGNALING VIA NFKB and IL2 STAT5 SIGNALING) were increased after nADT. To dissect immune-related signaling pathways activated by nADT, we used GSEA of the GO gene sets to identify differences between the nADT-treated and the control tumor. As shown in figure 2C, nADT induced significant enrichment of immune effector GO pathways, particularly for antigen presentation and the adaptive immune response pathways. Additionally, regulation of lymphocyte and myeloid cell activation and migration GO gene sets were enriched by nADT (online supplemental figure S3A,B). Finally, GO analysis confirmed AR signaling was reduced in each of the duration groups of nADT-treated tissues (figure 2D).

Figure 2. Differentially expressed genes and regulation of androgen and immune pathways in nADT cohort. (A) Volcano plot of differentially expressed genes comparing nADT cohort to matched controls. Red indicates upregulated and blue down-regulated genes. (B) GSEA of hallmark gene sets for nADT cohort compared with matched controls. The deep colored (vs light colored) bars represent significantly enriched pathways. (C) Enrichment of GO immune response pathways in nADT cohort compared with matched controls. Red indicates positive and blue negative enrichment of the indicated pathways. (D) Enrichment of GO androgen receptor signaling pathway (GO, “GO:0030521”) in the three nADT duration groups. (E) Enriched GO and Hallmark pathways in the three nADT duration groups. GO, Gene Ontology; GSEA, Gene Set Enrichment Analysis; nADT, neoadjuvant androgen deprivation therapy.

We next tested whether the immune response was duration dependent. Gene Set Enrichment Analysis (GSEA) Hallmark pathway analysis indicated enrichment of the inflammatory response pathway in the ≤2 months nADT group and the interferon (IFN) gamma response pathway in the 3–5 months nADT group, but these were reversed in the ≥6 months nADT group (online supplemental figure S3C–E). Interestingly, a comprehensive GSEA GO pathway analysis indicated immune-related pathways, particularly lymphocyte activation, were enriched in both of the shorter duration nADT groups, while in the ≥6 months treated group, these were replaced by negative regulation of disparate growth-related pathways (figure 2E). Similarly, these same immune-related Hallmark pathways were most expressed in the 3–5 months nADT group pathways vs either the ≤2 months and ≥6 months nADT groups (online supplemental figure S3F,G).

nADT alters the abundance of immune cell infiltrates in tumor microenvironment

To better characterize the functional outcome of nADT based on the altered immune-related pathways identified by the GO and Hallmark analysis, a paired analysis of the transcriptomes of the 43 nADT samples and their matched controls was performed to impute immune cell populations using the ImmuCellAI algorithm. Overall immune infiltration was increased by nADT, including an increase in antigen-presenting dendritic cells (DCs) (online supplemental figure S4). As an alternative to the transcriptomic immune cell subset reconstruction algorithms, we evaluated the expression levels of marker genes characteristic of individual immune cells. Paired analysis indicated the upregulation of CD8A, particularly in the ≤2 months nADT group (figure 3A, online supplemental figure S5A) and of the cytotoxic marker GZMB (figure 3B). CD4 expression was increased only in the 3–5 months nADT group (figure 3C, online supplemental figure S5C). The Treg marker FOXP3 expression was unchanged (figure 3D). In contrast, expression of genes encoding checkpoints programmed cell death protein 1 (PD-1) and T-cell immunoglobulin and mucin-domain containing-3 (TIM-3) were both increased, while those for T cell immunoreceptor with immunoglobulin and ITIM domain (TIGIT) and lymphocyte activation gene 3 protein (LAG3) were unchanged by nADT (figure 3E–H). Expression of the macrophage marker CD68 was elevated after nADT (figure 3I). Expression of genes for markers of immune-suppressive M2-like macrophage CD163, CD204 (MSR1), and CD206 (MRC1) were each increased (figure 3J–L). Expression of HLA-DRA, a marker of activated protumorigenic M1 polarized macrophages, was overall unchanged, but increased specifically in the ≤2 months group and not with longer duration nADT (figure 3M, online supplemental figure S5M). These data suggest an initial immune activation and pro-inflammatory myeloid skewing, followed by an immunosuppressive TME after nADT driven by the M2-like myeloid population.

Figure 3. Immune marker gene expression after nADT. Comparison of TPM of the indicated genes in paired nADT (orange) and matched control tissues (blue) on left (43 pairs). Red lines connecting pairs indicate an increase in expression, while blue lines indicate a decrease after nADT. Paired t-test p values are marked in red if <0.05. ADT, androgen deprivation therapy; LAG3, lymphocyte activation gene 3 protein; nADT, neoadjuvant ADT; TIGIT, T cell immunoreceptor with immunoglobulin and ITIM domain; TIM-3, T-cell immunoglobulin and mucin-domain containing-3 ; TPM, transcripts per million.

To test whether similar changes in the expression of messenger RNAs (mRNAs) corresponding to TiME marker genes were observed with extended exposure to adjuvant ADT, we analyzed publicly available datasets of localized and advanced/metastatic CRPC.^37–39 We found that following the long-term duration ADT that is a characteristic of CRPC treatment protocols, expression of the mRNAs for genes encoding immune checkpoints, including PD-1, LAG-3, TIGIT, and CD160, was increased (online supplemental figure S6A). Applying CIBERSORT analysis to these CRPC datasets, the increased expression of checkpoint genes in CRPC was accompanied by increases in CD8 T cells and Tregs, while estimates of memory and helper T-cell subsets, as well as activated NK cells, were decreased (online supplemental figure S6B). Expression of mRNAs encoding markers of suppressive (M2) macrophages, including IL-10, IL-13, VEGFA, and ARG1, were each increased in CRPC samples (online supplemental figure S6C). This was accompanied by an increase in the CIBERSORT estimate of M2 macrophages and a reciprocal decrease in M1 macrophages (online supplemental figure S6D). Finally, although the CIBERSORT estimate of CD8+T cells increased in CRPC samples, the ratio of CD8+T cells to M2 cells was decreased, whether using CIBERSORT immune cell population reconstruction, or measuring the ratio of mRNAs encoding CD8 and CD68 as specific markers of the corresponding cell populations (online supplemental figure S6E,F, respectively). Thus, overall we observed an increase in transcriptomic markers of immuno-suppressive cell populations in CRPC samples that received long-term ADT, similar to those in ≥6 months nADT treated specimens.

Immune cell populations in the TME dynamically altered by the duration of nADT

We then directly interrogated protein markers to determine the temporal and spatial patterns of immune cell populations in the TME in response to nADT. Specifically, we compared tissues from the nADT and control cohorts using Vectra qmIF staining to determine immune cell proportions and spatial interactions, as diagrammed in online supplemental figure S7. We first examined the entire nADT (and control) cohorts (online supplemental figure S8). Cell type quantitation in prostatectomy specimens confirmed that nADT reduced tumor cell density while increasing stromal cell density consistent with cell death due to ADT. Interestingly, we observed a small but significant increase in CTLs and Tregs in the nADT cohort, as well as M2-like TAMs, while M1-like TAMs were unchanged (online supplemental figure S8B).

Since we observed a mixed immune cell infiltrate of both CTLs and immunosuppressive myeloid cells when we examined the samples in the nADT cohort in aggregate, we next sought to examine the temporal effects of nADT duration on immune cell populations (figure 4). Visual inspection of representative immunofluorescent stained tissue sections (figure 4A) revealed that within 2 months of nADT, there was prominent M1-like TAMs (cyan) infiltration in the tumor stroma relative to matched control tissue, consistent with a pro-inflammatory response to tumor cell death. However, after 3–5 months nADT, CTLs (yellow) and T-helper cells (red) were dramatically more abundant, suggesting a profound immune activation. After 6 months nADT, there were few CTLs or T-helper cells, and a notable increase in both M2-like TAMs (green) and Tregs (white) around the tumor (magenta), suggesting a shift toward an immunosuppressive environment. These progressive changes in immune cell staining between each of the three nADT duration groups, and from their matched controls (figure 4A) were validated by using image analysis software to quantitate all individual cells in the 1,744 ROIs encompassing all of the prostate tissue in each of the 86 resection specimens (figure 4B). Notably, as observed visually, M1-like TAMs were increased in the ≤2 months nADT group, while CTLs and T-helper cells were broadly increased, particularly CTLs in tumors receiving 3–5 months nADT (figure 4B). The dynamic changes in cell proportions are summarized in figure 4C. However, after 6 months nADT, M2-like TAMs and Tregs were dramatically increased while CTLs were decreased (figure 4B,C). Finally, the ratio of CTLs to M2-like TAMs and of CTLs to Tregs in nADT-treated and control samples were both decreased with longer nADT (figure 4D). Overall, these data suggest that nADT treatment initially leads to cell death-related inflammation, characterized by M1-like TAMS, followed by immune activation, with increased CD8+T cell infiltration, but that within 6 months nADT, the immune environment becomes suppressive, with increases in M2-like TAMs and Tregs, reflecting a shift from immune activation to suppression over the course of treatment.

Figure 4. Extended nADT induced an increasingly suppressed TiME. (A) Representative high power fields from 1,744 qmIF images used to compare the distribution of immune cells in tumor between nADT-treated duration groups (≤2 months, 3–5 months, and ≥6 months) and matched control specimens. Marker identified by color indicated in inset at right, scale bars as indicated. Online supplemental figure S9 shows magnified insets, single fluorescent channels, and additional examples. (B) Quantification of changes in cell populations (as a percentage of total DAPI marked cells) in ROIs encompassing each specimen for each duration nADT group (429, 356, 292 ROIs) versus matched control groups (322, 187, 158 ROIs). Data are presented as mean±SEM, with each patient specimen indicated by a filled circle in the box plot. Statistical comparisons were performed using paired two-tailed Student’s t-tests. (C) Stacked bars represent the proportion of each of the cell types (indicated in the color coded key below) for each nADT group and the matched control groups and were used for descriptive visualization only. (D) Violin plots comparing the ratios of CTLs to M2-like TAMs (left) and CTLs to Tregs (right) in nADT-treated and control groups. Each data point represents the ratios from an individual patient sample, with comparisons made using the Wilcoxon signed-ranks test (+,*,**,*** indicates p values of <0.1, <0.05; <0.01; <0.001, respectively). ADT, androgen deprivation therapy; CTL, cytotoxic T lymphocyte; nADT, neoadjuvant ADT; ROI, regions-of-interest; TAM, tumor-associated macrophage; TiME, tumor immune microenvironment; Treg, regulatory T cell.

ADT induced immune cell populations to dynamically re-localize in the TME

Spatial analysis of the prostate TiME, identified by qmIF staining (figure 5), revealed dramatic changes in the distribution of CTLs that correlate with changes in T-helper and M2-like TAMs, as the duration of nADT increased. At ≤2 months nADT, M2-like TAMs were scattered or intraglandular, suggesting a phagocytic role and low immune suppressive activity. There was minimal immune cell infiltration, with CTLs and T-helper cells sparsely distributed and located far from the tumor, suggesting limited immune activation during the early phase of treatment (figure 5A, left column). By 3–5 months nADT, there was a remarkable increase in both CTLs and T-helper cells (figure 5A, middle column), and these cells were found in close proximity to the tumor (figure 5A, middle column inset figures). This clustering resembles the immune cell triad of CD4+ and CD8+ T cells engaging with nearby antigen presenting cells, necessary for licensing cytotoxic T cells to mediate antitumor immunity.^7–9 Later, at ≥6 months nADT, the immune landscape again shifted dramatically: there was a large reduction in the number of CD8+T cells, and while CD4+T cells remained clustered, fewer were proximal to the CTLs (figure 5A, third column). M2-like TAMs dramatically increased and infiltrated tumor glands (figure 5A, top row), and CTL interaction with tumor cells was dramatically reduced, suggesting a transition toward an immunosuppressive environment. Untreated matched control tissues showed no such changes (online supplemental figure S10A). While there was some heterogeneity in the predominant immunophenotype, ROIs from additional cases showed similar patterns of immune cells localization (online supplemental figure S11). More than 75% of the cases in the 3–5 months nADT group showed robust CTL staining, and more than 75% of the ≥6 months nADT group showed robust M2-like staining (online supplemental table S4).

Figure 5. 3–5 months nADT increased T-helper and CTL proximity to tumor cells and dispersed M2-like TAMs. (A) Spatial mapping of immune cell distribution in nADT-treated prostatectomy specimens from each of the duration groups (columns). Example inForm cell mapping images from the samples nearest the mean of the distance distribution for each group illustrating localization of CD68+, HLA-DR− M2-like TAMs and their distance to NKX3.1+ tumor (top row); CD8+CTLs and the distance to nearest M2-like cell (second row); CD4+/CD8+immune cell dyad to nearest tumor cell (third row); CD8+CTLs and the distance to nearest tumor cell (bottom row). (B) Density plots showing temporal changes in cell–cell proximities in the ROIs from the indicated (by color) groups of nADT-treated samples. Matched control examples and distances for each nADT group in online supplemental figure S10. Additional nADT and control examples in online supplemental figure S11. CTL, cytotoxic T lymphocyte; nADT, neoadjuvant androgen deprivation therapy; ROI, regions-of-interest; TAM, tumor-associated macrophage.

Quantitation of the spatial distribution of these cells in the tumor tissues revealed dynamic changes with nADT. The nearest neighboring cell–cell interaction distances, visually depicted by white lines in figure 5A, were quantitated in density plots of cell proximities in the 1,077 ROIs from the nADT cohort resection specimens (figure 5B). M2-like TAM proximity to tumor cells was variable based on duration of nADT. In the ≤2 months nADT group, M2-like TAMs were largely intraglandular and close to tumor cells (maxima 13 µm, pink peak in figure 5B top panel). In contrast, M2-like TAM proximity to tumor cells showed a wide distribution by 3–5 months nADT (centered around 35 µm, blue in figure 5B top panel), suggesting reduced immunosuppressive influence. However, after ≥6 months nADT, M2-like TAMs were again proximal to tumor cells (brown peak at 15 µm in figure 5B top panel). The distance between CTLs and M2-like TAMs in specimens from the ≤2 months nADT group was widely distributed, but by 3–5 months nADT, this separation was further increased (by ~50 µm), suggesting reduced interaction and enhanced CTL activity (figure 5B, pink and blue, second panel). Similar to the M2 to tumor distance, after 6 months nADT, the CTL to M2 distance dramatically decreased (20–50 µm, figure 5B, brown, second panel), suggesting regions of continued immune suppression and CTL exhaustion. The clustering of dyads of T helper cells and CTLs with adjacent tumor cells is a proxy for the activated antitumor immune triads.⁷ These T-helper/CTL dyads were closely associated (≤20 µm) with adjacent tumor cells in specimens from both the ≤2 months nADT and 3–5 months nADT groups, suggesting coordinated immune activation (figure 5B, pink and blue curves, third panel). However, by ≥6 months nADT, the proximity decreased, with bimodal peaks at 20 µm, and a new peak at 50 µm (figure 5B, third panel, brown curve). This progression from tumor-immune interaction dense cell subsets early in the course of nADT to the increased proximity of suppressive M2-like TAMs to tumor cells later is apparent visually (figure 5A, bottom panels) and is quantitatively reflected in the large number of closely interacting CTLs in the TiME of the 3–5 months nADT group, which is greatly reduced by ≥6 months nADT (figure 5B, bottom panel). These alterations in nearest neighbor proximities in the nADT specimens were not evident in ROIs from the matched control tissues (online supplemental figure S10). Importantly, the significance of each of the proximity findings in figure 5B was confirmed by spatial distribution analysis of all 1,744 ROIs representing the 43 individual nADT and 43 individual matched control patient resection specimens (online supplemental figure S10B).

Discussion

While current guidelines do not support nADT prior to radical prostatectomy, the tissue archives at Roswell Park Comprehensive Cancer Center accumulated prostatectomy specimens from patients treated with neoadjuvantly with ADT for 1–12 months (figure 1). Endpoint transcriptomic and spatial quantitative immunofluorescent profiling of these specimens provided an opportunity to investigate the temporal evolution of the PrCa TiME, as might occur in advanced, recurrent, or metastatic PrCa. Validation of nADT efficacy was determined using TMPRSS2 expression as a biomarker of ADT effect in the treated cohort, and found to be reduced for each of the duration groups (online supplemental figure S2). nADT was previously reported to downregulate TMPRSS2 mRNA expression in resections after 7 days⁴⁰ and 1 month.⁴¹ To our knowledge, this is the first demonstration of parallel reduction of both TMPRSS2 mRNA and protein, and of a reduction over longer durations of nADT in patient specimens, suggesting wide utility as a biomarker of ADT. Overall, nADT induced a loss of tumor cells with a corresponding increase in the proportion of stroma, accompanied by increased immune infiltrate (online supplemental figure S8).

In addition to effective AR suppression, transcriptomic pathway analysis revealed inflammatory pathway response (figure 2B–E). Quantitation of the immune cells in the TiME revealed an increase in M1-like TAMs in the ≤2 months nADT group that accompanied a decrease in tumor cells (figure 4B). These observations are consistent with tumor cell death, unequivocally the clinical goal of ADT. The early increase in M1-like TAMs is also consistent with an immunogenic cell death (ICD) model for the earliest immune response to ADT. ICD, induced by pro-apoptotic anticancer therapies, is a type of tumor cell death that elicits an anticancer immune response relying on the activation of damage-associated molecular patterns (DAMPs).⁴² Antigens from dying cancer cells promote the recruitment and activation of antigen-presenting cells (APCs), especially DCs and M1-type TAMs. DAMPs consequently enhance DC maturation and DCs ability to stimulate transient immune effector T cells.⁴³

This initial period of ICD in response to nADT is followed by a window of intense immune activation. This was confirmed by early (≤2 months) upregulation of cytotoxic T-cell markers (eg, CD8A, GZMB; figure 3), followed by a dramatic increase in CTLs by 3–5 months nADT (figure 4). This correlated with GSEA evidence of increased T-cell trafficking and IFNG/tumor necrosis factor (TNF) signaling to sustain T-cell activation (figure 2E), and increased DCs which may mediate antigen presentation (online supplemental figure S4B) following 3–5 months nADT. Similar transcriptomic changes in Hallmark gene sets of androgen signaling, as well as increases in IFNG and TNF signaling have been reported after enzalutamide therapy.^{44 45} The early active immune response was confirmed by spatial qmIF analysis which revealed proximal paired T-helper cells and CTLs near tumor cells (figure 5A, left column). When combined with antigen presenting DCs (online supplemental figure S4B), this cell triad clustering might mediate active licensing of CD8⁺ T cells to effect antitumor immunity.^7–9 These T-helper/CTL dyads moved closer to tumor by 3–5 months nADT, when an immune “storm” of CD8⁺ T cells adjacent to both T-helper cells and tumor was apparent (figure 5A middle column).

However, after ≥6 months nADT, the PrCa TiME becomes immunosuppressive. Both Tregs and M2-like TAMs were increased (figure 4B). The overall proportion of CTLs was reduced, as were the ratios of CTLs to M2-like TAMs and CTLs to reg cells (figure 4D). The CTLs were juxtaposed to the M2-like TAMs, but more distant from tumor cells (figure 5). These M2-like TAMs are anti-inflammatory and protumorigenic,⁴⁶ and predict poor survival post-ADT.⁴⁷ In the specimens from the ≥6 months nADT group, the M2-like TAMs appear to form a protective layer around the tumor glands (figure 5A). These observations in the ≥6 months nADT group are consistent with reports that on progression to CRPC, the TiME contains high levels of interferon and activated but dysfunctional CD8⁺ T cells, and the tumors are resistant to CPIs.^{2 48}

Several other reports demonstrate ADT induces T-cell activation in PrCa tumors. ADT (or radiation) results in autoantibodies in 25–30% patients.¹⁵ During the first month after nADT prior to resection, markers of CD4+ and CD8+ T cells increased, with peak CD3+T cell activity at 3 weeks.¹² Within 1 month of nADT, circulating naïve T cells increase, with an increase in effector cell response.¹⁶ Similar transient activation of immune response has also been observed in mouse models.¹² Combining a granulocyte-macrophage colony-stimulating factor-secreting vaccine with nADT increased CD8+T cell infiltration, but also raised levels of Tregs.¹⁸ This finding underscored the need for strategies targeting Tregs to improve therapeutic efficacy.²² AR signaling activity suppresses hallmark IFN response and suppresses T-cell activity, and ADT enhances response to checkpoint blockade therapy in mouse melanoma and PrCa models,⁴⁸ analogous to the IFN-associated checkpoint responsiveness in human colorectal cancer.⁴⁹ Since ADT reduces immune tolerance to prostate-expressed antigens in mice, allowing T cells to develop effector function in response to vaccination,⁵⁰ our data suggest that ADT might best be employed several months prior to immunotherapy with CPIs.

Independent of CPIs, the immune activation and subsequent suppression was consequential in this nADT cohort. High CTL density was prognostic for lack of biochemical (PSA) recurrence in patients treated with fewer than 6 months of nADT (with a similar trend for OS); however, the effect was reversed in the longer duration nADT group (online supplemental figure S12), in accord with the above temporal characterization. The poor outcomes with high CTL density in patients treated with more than 6 months of nADT were not due to treatment-driven neuroendocrine tumor evolution (online supplemental figure S13), suggesting survival differences are related to nADT duration-dependent immune changes. That CTL density is prognostic in this nADT cohort is in agreement with our previous finding of a correlation of CTL density with PFS and OS in a much larger (n=230) cohort of primary resection specimens.⁶ Indeed, the finding is in accord with a report that the combination of CTL and programmed death-ligand 1 expression predicts biochemical recurrence.⁵¹ Thus, the density of immune cells and their proximity to tumor cells may serve as independent prognostic biomarkers of clinical outcomes.

Limitations of the study include the nADT cohort sample size, particularly of the duration subgroups, the retrospective providential nature of the cohort, and extended period of time required to gather sufficient samples for analysis. Many of the patients presented at a comprehensive cancer center for a second opinion in the mid-2000s, having been previously treated in the community with ADT. These men may have received neoadjuvant therapy in response to perceived risk at presentation. To control for any changes due to evolving practice, each nADT sample was paired with a control sample of identical stage/grade and year of resection. However, there was no difference between the control and nADT duration groups in PSA levels determined pre-ADT, or in survival (online supplemental figure S1). Further, there was no substantive difference in the genetic profile between the nADT treated cohort and the untreated control cohort (figure 1B). The few patients with higher TMB and MSI are not among the examples presented in figure 5, and were more prevalent in the untreated control cohort (figure 1B). Some LHRH-agonist treated men also received bicalutamide (figure 1A); however, these were balanced across duration groups. Nonetheless, potential confounding from bicalutamide versus potential transient androgen flare cannot be excluded. Importantly, the study was cross-sectional, with nADT duration evaluated between matched pairs rather than through serial sampling within the same individuals. Thus, the observed immune states should be interpreted as duration-associated phenotypes rather than definitive within-patient temporal changes. Finally, qmIF analysis using the Vectra system restricted the panel of antibodies to six. This necessarily limited the depth of the TiME cell phenotyping, and future studies might employ more extensive panels to, for example, directly characterize the extent of T-cell exhaustion.

In conclusion, long-term ADT remodels the TiME in PrCa with a transient immune activation and then immune suppression, which leads to the occurrence of CRPC that is also CPI-resistant (online supplemental figure S14A). One approach to prevent the development of CRPC is to improve the initial treatment of patients with advanced PrCa by identifying windows of opportunity to combine CPIs with ADT. The present examination of nADT duration groups enabled dissection of the time-course of changes in the TiME (online supplemental figure S14B). These findings align with an ICD model, with DAMPs released from nADT-induced dying tumor cells activating APCs and M1 macrophages, boosting T-cell responses, enhancing antitumor immunity by promoting CTL priming and sustained immune engagement, supporting the early robust immune activation observed in our study. We hypothesize that in response to the initial immune activation phase, there is a reciprocal induction of an immune suppression phase (see model, online supplemental figure S14), blue line) marked by checkpoint expression, Treg and myeloid cell recruitment resulting in effector T-cell suppression (online supplemental figure S14C). Reprogramming the TiME by modifying production of chemokines and cytokines is an emerging strategy to convert cold tumors into immunologically responsive tumors and improve patient outcomes. It is not just the quality and the quantity of CTLs, but also the spatial positioning of immune cells in tumors that is critical for long-term tumor control.⁷ Indeed, we found dramatic changes in the distribution of CTLs over time of exposure to nADT, suggesting large differences in potential response to immune-directed therapies early or late in the course of ADT. However, these findings are correlative and lack direct functional validation. While these data strongly suggest a therapeutic window of opportunity, future in vivo functional and mechanistic studies in model systems will be necessary to confirm these observations.

Nonetheless, these findings provide crucial insights into the temporal dynamics of ADT-induced immune modulation in PrCa. By identifying a period of high T-cell infiltration after several months of nADT but a subsequent increase in immunosuppressive Tregs and M2 TAMs after 6 months nADT, we highlight a critical window for therapeutic intervention. In advanced and recurrent patients initiating ADT, delivering CPIs in this early phase of ADT-induced immune activation may optimize treatment efficacy. Moreover, understanding the shift toward immune suppression over time opens avenues to counteract M2 or Treg mediated immune suppression that could be sequenced with longer duration ADT,^52–54 potentially improving long-term outcomes in PrCa management.

The funders had no involvement in the study design; in the collection, analysis and interpretation of the data; in the writing of the report; and in the decision to submit the paper for publication.

Data availability statement

Data are available upon reasonable request.