Single-cell profiling of peripheral blood mononuclear cells in diverse men with prostate cancer reveals a fatal disease-associated immune signature

Abstract

Alterations in peripheral immunity may contribute to fatal prostate cancer and the excessive disease burden among African American (AA) men. Thus, we investigated peripheral immunity in prostate cancer patients in association with disease outcome and AA descent. We performed single-cell RNA and T cell receptor (TCR) sequencing using peripheral blood mononuclear cells (PBMC) from a cohort of 43 prostate cancer patients with survival follow-up and 16 healthy controls. Analyzing 273,229 high-quality PBMC, we observed that men who developed fatal prostate cancer (n=20) distinctively expressed the LAG3 and TIGIT exhaustion markers in their CD8⁺ T effector memory cells (CD8 TEM). These men also showed a loss of type-2 conventional dendritic cells (cDC2). Further analyses revealed additional transcriptome-inferred functional differences in cDC2, CD16⁺ monocytes, and CD8 TEM comparing men with and without fatal disease. Those describe cDC2 as immunosuppressed, CD16⁺ monocytes as pro-inflammatory but less cytotoxic, and CD8 TEM as chronically activated in patients with fatal disease. Comparing AA with European American patients, we observed upregulated LAG3 and TIGIT in AA PBMC. For cDC2 and CD16⁺ monocytes, fatal disease and AA patients showed shared marker expression. Furthermore, TCR profiling discovered over-representation of large, hyperexpanded clonotypes and a lower TCR diversity among fatal disease and AA patients. Lastly, the CD8 TEM functional status and LAG3 expression in PBMC emerged as a candidate predictor of disease fatality. We conclude that peripheral immunity may distinctly vary in association with disease outcome and AA descent of prostate cancer patients.

Introduction

We and others have reported that tumor immunobiology differs between African American (AA) and European American (EA) prostate cancer patients (1–4). Yet, the immunological differences between AA and EA patients may not be restricted to tumors. Accordingly, we described a serum proteome–defined suppression of tumor immunity signature that associated with metastatic and fatal prostate cancer (5). These findings suggest that both tumor and peripheral host immunity and a peripheral deficiency in immune surveillance against tumors may contribute to disparities and fatal prostate cancer. Thus, we aimed to further investigate peripheral immunity in prostate cancer patients in association with disease fatality and AA descent with an in-depth analysis of PBMC.

PBMC are key mediators of systemic immunity, contributing to both immune regulation and immunosurveillance of primary tumors and cancer metastasis (6). Given their accessibility, they serve as a minimally invasive but informative resource for characterizing immune function in cancer patients. To define their composition in association with fatal prostate cancer and AA descent, we performed combined single-cell RNA (scRNA) and T cell receptor (TCR) sequencing using PBMC collected from a diverse cohort of 43 prostate cancer patients with long-term survival follow-up and 16 healthy controls. This integrative single-cell sequencing approach enabled high-resolution mapping of immune cell composition, transcriptional states, and clonal TCR architecture. It has also generated a high-quality single-cell data resource for others to investigate peripheral immune profiles. Using this resource and a methodology that inferred biology based on the single-cell transcriptome, we discovered that circulating immune cell populations in prostate cancer patients show differences in composition and predicted function in association with disease fatality and self-identified patients’ race, being either AA or EA men.

Materials and Methods

Study population.

Recruitment of cases and controls and blood collection were conducted as part of the NCI-Maryland Prostate Cancer Case-Control study, as described (7,8). Between 2005 and 2015, prostate cancer patients were recruited at either the Baltimore Veterans Affairs Medical Center or the University of Maryland Medical Center through arrangements with physicians. The study enrolled 976 prostate cancer patients diagnosed with the disease within the last 2 years. All had prostate cancer at the time of recruitment. Controls were identified through the Maryland Department of Motor Vehicle Administration database. The control group included 1,034 men without a history of cancer other than non-melanoma skin cancer, with a residential working phone number, born in the United States, and fluent in English well enough to be interviewed. Exclusion criteria included prior history of radiotherapy or chemotherapy, severe illness, or residing in an institution. Participants self-reported their race and signed a written informed consent for their participation prior to interview. Survival and causes of death were identified through annual searches of the National Death Index (NDI) database (last follow-up: 12/31/2022). The 20 patients with fatal prostate cancer had an average survival time of 1,363 days (range: 277 to 4,381), whereas those without fatal prostate cancer on follow-up (n = 23) had an average survival time of 3,319 days (range: 1,035 to 4,467). The high prevalence of fatal disease was by design, not involving randomization, to allow for investigations in relation to survival. As selection criteria, these men were enrolled between 2011 and 2015 and had a collection of viable PBMCs. See Supplementary Table S1 for more information about these patients. Study materials and procedures were approved by the NCI (protocol # 05-C-N021) and University of Maryland (protocol #0298229) Institutional Review Boards. Research followed the ethical guidelines set by the Declaration of Helsinki. Study design and analysis followed REMARK criteria.

Medication use.

Aspirin use information was obtained from the study survey. Use of androgen deprivation therapy (ADT) or immunomodulatory drugs was obtained from medical records. Use of immunomodulatory drugs included the use of high-dose corticosteroids, alkylating agents, antimetabolites, transplant-related immunosuppressive drugs, TNF blockers, and medication used to treat auto-immune diseases such as multiple sclerosis, Crohn’s disease, ulcerative colitis, rheumatoid arthritis, psoriasis, and others. We searched for 51 commonly prescribed immunomodulatory drugs.

National Comprehensive Cancer Network (NCCN) risk score for prostate cancer.

We assigned risk groups based on the patient’s TNM stage, Gleason score, Gleason pattern, and PSA level at diagnosis following the 2019 NCCN guideline for prostate cancer (9). Risk groups were defined as localized, regional, and metastatic prostate cancer based on clinical parameters at the time of diagnosis. Localized cases were further classified as low, intermediate, high, and very high risk based on the likelihood of their disease progressing to lethal disease according to the NCCN guidelines. Cases with lymph node involvement but no distant metastasis at diagnosis were classified as regional disease whereas those with distant metastasis at the time of diagnosis were classified as metastatic disease. Four a priori defined risk groups were used for categorization: low, intermediate, high/very high, and regional/metastatic, as previously described (5,10).

Collection of viable PBMC.

Whole blood was collected in heparin green top BD Vacutainer tubes (VWR, BD367874). Lymphocyte separation medium (VWR, 45000-728) was added, and samples were centrifuged at 1,400 rpm for 30 min. The lymphocyte layer was transferred to a new tube, 1X phosphate-buffered saline (PBS) was added, and then the samples were centrifuged at 1,200 rpm for 8 min. The supernatant was discarded. The remaining sample was resuspended in PBS and centrifuged at 800 rpm for 8 min. Supernatant was again discarded. Samples were then resuspended in freezing media: RPMI (Thermo Fisher Scientific, 11875093) + 20% human AB serum (GeminiBio, 100-512). Cells in the samples were counted and frozen in 900 μl of freezing media and 100 μl of DMSO per cryovial. Vials were stored in liquid nitrogen.

Single-cell suspension preparation.

Frozen vials of PBMC were rapidly thawed in RPMI medium supplemented with 10% fetal bovine serum (HyClone Laboratories, SH30071). Thawed PBMC were washed twice with 4 mL of pre-warmed (37°C) PBS (Thermo Fisher Scientific) containing 0.04% bovine serum albumin (BSA). At the NCI Single Cell Analysis Facility, cell concentration and viability were assessed using the Luna-FX7 fluorescent cell counter. Samples were diluted, if necessary, to obtain 6,000 viable cells suspended in PBS/0.04% BSA. Each sample was loaded into a single capture lane on a Chromium Next GEM Chip K (10X Genomics), targeting recovery of up to 6,000 cells per lane, according to the manufacturer’s instructions in the Chromium Single Cell 5’ v2 User Guide (https://cdn.10xgenomics.com/image/upload/v1722286172/support-documents/CG000331_Chromium_Next_GEM_Single_Cell_5_v2_UserGuide_RevF.pdf). Single-cell suspension was prepared to target 10,000 cells/sample and loaded with barcoded gel beads and partitioning oil to generate Gel Beads-in-Emulsion (GEM). PBMC for each patient were processed once unless the single-cell preparation or subsequent sequencing failed the quality control assessment.

Library construction for RNA and TCR sequencing.

Library construction and sequencing were carried out at the NCI Single Cell Analysis Facility, an NCI core service. Cells were encapsulated into GEMs by combining barcoded gel beads, a master mix containing cells, and partitioning oil. After GEM generation with uniform consistency, reverse transcription was performed overnight to produce barcoded full-length cDNA. Amplified cDNA was used to construct gene expression (GEX) and TCR libraries. V(D)J regions were enriched using the Human T-cell Enrichment Kits (10X Genomics; PN-1000005). Library preparation was performed using Chromium Next GEM Single Cell 5’ Reagent Kits v2 (10X Genomics; PN-1000263, PN-1000265, PN-1000190, PN-1000252, PN-1000253), following the manufacturer’s protocol. All quality control steps were performed as recommended. Final libraries were sequenced on the Illumina NovaSeq platform using a paired-end 28/91 bp configuration for both 5’ GEX and TCR libraries.

Quality control assessment of single cell data and exclusions.

Sequencing of barcoded cDNA yielded data from 322,897 single cells. Data integration was performed using the Merge() function in Seurat (v5.2.1) (RRID:SCR_016341), followed by SCTransform normalization and scaling. Cells were filtered based on standard quality control criteria for exclusion from further analysis: fewer than 200 detected genes, fewer than 500 or more than 30,000 total counts, or greater than 10% mitochondrial content. This filtering step resulted in high-quality data from 300,859 cells. Next, cell type annotation was carried out via multimodal reference mapping using the pbmc_multimodal_2023.rds reference dataset of 161,764 cells. Batch correction and sample integration were performed with RunHarmony() v1.2.4 (RRID: SCR_022206).

Unsupervised clustering was conducted using the top 30 principal components at a resolution of 0.8. Cell type annotations were manually refined based on canonical marker expression. Low-abundance and less relevant cell populations, including platelets, erythrocytes, doublets, innate lymphoid cells, and hematopoietic stem and progenitor cells, were excluded from further analysis. Thus, the final dataset included sequencing data from 273,229 high-confidence cells for downstream analyses.

Cell-type annotation using Seurat multimodal reference mapping.

Cell-type annotations were obtained by multimodal reference mapping against the Satija Lab PBMC CITE-seq reference object, pbmc_multimodal_2023.rds, which is a Seurat v4 PBMC CITE-seq reference. Mapping does not use a fixed gene list. The Seurat reference-mapping workflow annotates each query cell based on reference-defined cell states and projects query cells into the reference uniform manifold approximation and projection (UMAP). In the PBMC example, anchors are identified using the reference spca reduction with MapQuery transfers using celltype.l1 and celltype.l2 labels from the reference to the query.

Pseudobulk analysis with covariate adjustment.

This analysis was performed by aggregating single-cell sequencing profiles into sample-level expression matrices. The approach facilitates differential gene expression analysis across key clinical and demographic contrasts, including cases vs. controls, AA vs. EA men, or fatal vs. non-fatal prostate cancer. Differential expression analysis was conducted using DESeq2 (RRID:SCR_000154), incorporating relevant covariates to adjust for potential confounders, and applying an adjusted P <.05 for significance. Cases vs. controls comparisons were adjusted for age, BMI, and self-identified race; AA vs. EA patient comparisons were adjusted for age, BMI, and NCCN risk score; fatal vs. non-fatal comparisons were adjusted for age, BMI, self-reported race, and NCCN risk score. Sensitivity analyses added additional covariates.

Gene set enrichment and pathway activity analyses.

To assess the biological relevance of transcriptional changes, we performed gene set enrichment analysis (GSEA) (11) using the fgsea R package (v1.32.4; RRID:SCR_020938) and covariate-adjusted pseudobulk-derived ranked gene lists. Reference gene sets comprised the Hallmark collection from the Molecular Signatures Database (MSigDB; RRID:SCR_016863) and curated custom gene sets representing interferon signaling and T-cell exhaustion pathways. Enrichment scores were ranked as normalized enrichment score (NES), and the statistical significance was determined using the Benjamini-Hochberg method to adjust for multiple hypothesis testing and applied the false discovery rate (FDR)-adjusted significance threshold of P < .05. Pathway activity at the single-cell level was quantified using Hallmark gene sets from MSigDB (h.all.v2023.2.Hs), imported with gmtPathways. For each pathway, only genes detected in the Seurat object (SCT assay) were retained. SCT-normalized expression values were extracted, and expression for each gene was z-scaled across all cells. Pathway/signature activity scores were then computed as the mean z-score across all genes within the pathway/signature for each cell, generating a continuous per-cell metric.

TCR repertoire analyses.

High-quality TCR sequencing data were available for 37 prostate cancer patients and 16 healthy men and were processed using the 10x Genomics Cell Ranger VDJ pipeline (v6.1; RRID:SCR_017344). Reads were aligned to the human GRCh38 human reference genome. Only high-confidence, full-length, productive contigs were retained. Cells were filtered to include only those with paired, productive chains. The repertoire analysis was conducted in R (v4.4.3; RRID:SCR_001905) using the scRepertoire package (v2.2.1; RRID:SCR_025691). A valid clonotype required at least one TCR A and B locus chain. In situations of multiple productive chains, the dominant TCR A/B pair was used to define the clonotype. Analysis also included a Shannon entropy index assessment to characterize diversity. TCR data were integrated with scRNA-seq data using the combineTCR and combineExpression functions.

Statistical analysis.

Analysis was conducted using the R software (v4.4.3; RRID:SCR_001905). Statistical significance is defined as two-sided P <.05. We assessed unadjusted differences between two groups with the Wilcoxon rank-sum test. Logistic regression models were used to examine significance after adjustments. Lethal disease signatures: To investigate the association of immune signatures with fatal disease, we interrogated the single-cell sequencing data for gene signatures predictive of prostate cancer survival. For each immune cell type, we searched for signature genes that were differently expressed between fatal and non-fatal cases after adjusting for age, BMI, race and the NCCN risk score (P < .05). LASSO regression was applied for further gene selection, using two approaches: (i) Cox proportional hazards modeling for survival time, and (ii) logistic regression for fatal vs. non-fatal status. When a survival predictive signature was identified, a signature score was computed per individual as weighted sums of expression values for the selected genes. The score was then evaluated as a continuous exposure in multivariable Cox regression models to assess the association with all-cause and disease-specific mortality, adjusting for potential confounders selected a priori, including age (continuous), BMI (continuous), self-reported race (dichotomous), NCCN risk score (categorical), and ADT treatment (dichotomous). Because 26 cell types were analyzed, a gene signature score was considered as being associated with survival at the adjusted significance threshold of P ≤ .0019. Signature scores were additionally dichotomized at the median and adjusted marginal survivor curves were computed, and statistical significance was assessed via Wald tests. The same multivariable Cox regression modeling was applied to test a priori exposures of interest, including abundance of circulating CD8 TEM cells (percentage of total PBMC) and normalized LAG3 gene expression in CD8 TEM cells and all PBMC. Additional sensitivity analyses were conducted which included patients’ diabetes status, aspirin use, and use of immunomodulatory drugs as covariates. In all analyses, proportional hazards assumptions were confirmed.

Research Resource Identifiers (RRIDs)

We used 10x Genomics Chromium as our single cell analysis platform: RRID:SCR_024537.

Results

Donor characteristics and single cell profiling of PBMC.

We performed scRNA-seq of PBMC from 59 men. Those included 26 AA and 17 EA men with prostate cancer (n=43) and 8 AA and 8 EA healthy men without a history of the disease (n=16) who were recruited as population-based controls into the NCI-Maryland Prostate Cancer Case-Control Study (Table 1). More cases than controls were either AA men or had diabetes. AA and EA patients were similar in age and body mass at time of PBMC collection, but more AA patients presented with advanced/metastatic disease at time of recruitment (38% versus 23%). Only two patients took immunomodulatory drugs, per medical record, whereas aspirin use was prevalent (44%). Twenty-five cases experienced an all-cause mortality on follow-up. Per study design, we enriched the patient population to have an adequate representation of men with fatal prostate cancer on follow-up (n = 20; 47%). Among them, 14 were AA and 6 were EA patients and 13 had received ADT at time of recruitment, whereas only one of the patients without a disease fatality (n = 23) received ADT.

Table 1.

Characteristics of the prostate cancer study population (n=59)

	African American Men		European American Men
Characteristics	Controls (n=8)	Cases (n=26)	Controls (n=8)	Cases (n=17)
Age
Mean in years ± SD	65.0 ± 2.0	66.1 ± 8.8	65.3 ± 2.0	63.3 ± 7.8
BMI
Mean in kg/m2 ± SD	26.5 ± 2.2	27.5 ± 4.6	31.5 ± 6.4	28.4 ± 4.8
Diabetes1
Yes	0	10 (38%)	2 (25%)	4 (23%)
No	8 (100%)	16 (62%)	6 (75%)	13 (77%)
Aspirin Use
Yes	0	12 (46%)	7 (88%)	7 (41%)
No	8 (100%)	14 (54%)	1 (12%)	10 (59%)
Immune modulatory drug use2
Yes3		2 (8%)		0
No		21 (80%)		17 (100%)
Unknown		3 (12%)		0
Androgen ablation therapy use
Yes		9 (35%)		5 (29%)
No		17 (65%)		12 (71%)
NCCN Risk Score4
Low		3 (12%)		2 (12%)
Intermediate		5 (19%)		8 (47%)
High or very high		8 (31%)		3 (18%)
Regional/distant metastasis		10 (38%)		4 (23%)
Survival5
All-cause death		17 (65%)		8 (47%)
Prostate cancer mortality		14 (54%)		6 (35%)

Abbreviations: SD = Standard deviation, NCCN = National Comprehensive Cancer Network

¹Diabetes status: obtained from both study survey and medical records

²Obtained from medical records and not available for controls

³Methotrexate, hydroxychloroquine

⁴Low, intermediate, and high-or very high-risk NCCN categories apply to localized prostate cancers

⁵National Death Index Censor Date 12/15/2022

The experimental design of our study is shown in Supplementary Figure S1. Applying quality filtering and data integration using Harmony, we retained 273,229 high-confidence cells for downstream analyses. For these cells, the scRNA-seq data quality was high, as shown by the total RNA counts (range: 500–30,000 per cell), number of genes (>200 per cell), and mitochondrial gene percentage for each cell (≤10%), which are standard quality control measures (Supplementary Figure S2A). Unsupervised clustering combined with reference mapping distinguished 5 main cell types representing B cells, T cells, natural killer (NK) cells, monocytes, and dendritic cells (DC) (Supplementary Figure S2B). These cells showed abundance variation across donors (Supplementary Figure S2C). More in-depth multimodal reference mapping further identified 26 distinct immune cell subtypes, encompassing B, T, NK, and DC cells, and monocytes (Supplementary Figure S3A). These 26 immune cell populations showed large abundance differences within the collected PBMC (Supplementary Figure S3B). Our cell type annotation was validated across the 26 cell populations using selected canonical markers and manual curation (Supplementary Table S2), verifying that the identified immune cell subsets express the expected marker (Supplementary Figure S4A-C). Moreover, assigning cytotoxicity scores to T-cell subpopulations correctly identified those known for being cytotoxic (Supplementary Figure S4D).

Peripheral immune cell composition and differences in function by case-control status.

Although not the main objective of our study, we reviewed the immune cell composition among cases and controls and did not find abundance differences between them when focusing on the 5 major immune cell types, B cells, T cells, NK cells, monocytes, and DC (Supplementary Figure S5). However, an exploratory analysis expanding into the 26 immune cell subtypes revealed that peripheral CD4⁺ naïve T cells were proportionally more abundant in the control population whereas CD8 TEM were more abundant in the patient population, after applying a Bonferroni-adjusted significance threshold of P ≤ .0019 (Figure 1A). CD8 TEM in these patients showed expression of a set of marker genes: CCL5, GZMH, CD8A, TRAC, KLRD1, NKG7, GZMK, CST7, CD8B, TRGC2, DUSP2, IL32, CMC1, CD160, TIGIT, KLRG1, ITGB1, HOPX, GNLY, GZMB, among others. The use of a covariate-adjusted (age, BMI, and self-reported race) pseudobulk differential gene expression analysis (Figure 1B), followed by a GSEA with the Hallmark gene sets (11) (Figure 1C), revealed additional transcriptional variance consistent with functional differences for these two cell types comparing cases with controls at a FDR-adjusted P < .05. For CD4⁺ naïve T-cell function, GSEA indicated increased oxidative phosphorylation in the mitochondria and a decrease in developmental and cytokine signaling and mesenchymal transition, together with a reduced inflammatory response, interferon signaling, and allograft rejection, among cases. CD8 TEM cells exhibited differing changes among cases, with up-regulation of the TNFα and other cytokine-related signaling pathways, inflammatory response, heme metabolism, proliferation-related pathways, and apoptosis (Figure 1C). Adding diabetes status and aspirin use for cases and controls as additional covariates to the pseudobulk expression analysis further supported these conclusions, showing that these immunomodulatory exposures did not confound our GSEA results (Supplementary Figure S6). Together, these observations inferred from the single-cell transcriptome indicate a suppressed immune response and a shift towards tolerance for circulating CD4⁺ naïve T cells but increased effector function and cytotoxicity, and aging, for CD8 TEM cells in men with prostate cancer. Lastly, prostate cancer patients presented with a reduced CD4/CD8 ratio (Figure 1D), an indicator of both immunosenescence and increased morbidity/mortality risk in HIV and cancer patients (12–15). This association persisted after adjustment for age, BMI, self-reported race, diabetes status and aspirin use in a logistic regression model (P < .002).

Figure 1. Immune cell composition and functional status of PBMC in men with prostate cancer.

(A) Relative abundance of CD4⁺ naïve T cells and CD8⁺ effector memory (CD8 TEM) T cells among cases (n=43) and controls (n=16). Wilcoxon rank sum test for significance testing. (B) Volcano plots illustrating differential gene expression in CD4⁺ naïve T cells and CD8 TEM for contrast cases versus controls. Analysis was adjusted for age, BMI, and self-identified race. (C) Gene Set Enrichment Analysis (GSEA) showing enrichment in Hallmark gene sets for two immune cell types, CD4⁺ naïve T cells and CD8 TEM, comparing cases versus controls (reference). Dot plot display of normalized enrichment scores. GSEA was performed using DESeq2-ranked, covariate-adjusted genes and evaluated against the Hallmark gene sets with FDR correction (Padj < .05 for inclusion). (D) Decreased CD4/CD8 ratio in prostate cancer patients. Wilcoxon rank sum test for significance testing.

Peripheral immune cell composition and differences in function by survival status.

Next, we explored if immune cell profiles or their function may vary by patient survival status. We initially evaluated the expression of immune exhaustion markers in PBMC. The analysis revealed that patients who developed fatal prostate cancer on follow-up had a PBMC population with elevated LAG3 and TIGIT exhaustion markers (Figure 2A). This upregulation was at least partly driven by elevated LAG3 and TIGIT in CD8 TEM cells of fatal disease patients (P =.017 and .019 for elevated LAG3 and TIGIT, respectively). To ensure that this finding was not confounded by underlying ADT differences, we performed a logistic regression analysis with ADT adjustment and confirmed that upregulation of LAG3 and TIGIT in CD8 TEM was largely independent of obtained ADT (P = .016 each). Applying an additional sensitivity analysis that added patients’ diabetes status, aspirin use, and use of immunomodulatory drugs to the logistic regression model - besides ADT - further confirmed that LAG3 and TIGIT are upregulated in CD8 TEM independent of these candidate confounders (P = .013 for TIGIT and P = .039 for LAG3). Two other exhaustion markers, CTLA4 and PDCD1, were also elevated in fatal disease patients, yet not as distinctively as LAG3 and TIGIT (Figure 2A). Despite immune exhaustion markers being elevated in association with disease fatality, the cytotoxicity scores for the various T-cell subpopulations did not separate fatal from non-fatal disease cases (Figure 2B). Extending our investigations into the 5 major immune cell types, we noticed that patients with fatal disease had lower numbers of circulating DCs (Figure 3A) with a distinct loss of type 2 conventional DCs (cDC2) (Figure 3B). In contrast, the abundance of CD4⁺ regulatory T cells, an immune cell type of high interest as a therapeutic target (16), did not show variation in association with disease outcome in this patient cohort (Figure 3B). Low cDC2 counts may weaken antitumor immunity (17). Most of the annotated cDC2 expressed the CD1C functional marker with another subset co-expressing CD14 and CD5 (Figure 3C). Additional confounder-adjusted pseudobulk gene expression (Supplementary Figure S7) and GSEA-based enrichment analyses (Figure 3D) revealed further transcriptome-inferred functional differences in cDC2, CD16⁺ monocytes, and CD8 TEM cells comparing men with and without fatal disease (adj. P <.05). In cDC2, GSEA pointed to the downregulation of anabolism and the inflammatory, the interferon, and the androgen responses, and TNFα-signaling-via-NFκB, in association with disease fatality, consistent with a more tolerogenic and immune function–suppressed cell status (Figure 3D). In the CD16⁺ monocytes, GSEA indicated an increase in the inflammatory response and TNFα-signaling-via-NFκB, and a down-regulation of oxidative phosphorylation, interferon response, and allograft rejection, in association with fatality, characterizing these cells as pro-inflammatory monocytes with TNFα overproduction but reduced cytotoxicity. Lastly, the transcriptome-inferred functional differences in CD8 TEM cells comparing fatal versus non-fatal cases (Figure 3D) were reminiscent of the contrast for cases versus controls (Figure 1C), indicating an increased effector function and cytotoxicity for CD8 TEM cells among patients with a future prostate cancer fatality. Adding ADT, diabetes status, aspirin use, and use of immunomodulatory drugs as additional covariates to the pseudobulk expression analysis, besides age, BMI, self-reported race, and NCCN risk score, did not affect these GSEA-derived conclusions (Supplementary Figure S8).

Figure 2. Exhaustion status of PBMC and T cells in men with fatal prostate cancer.

(A) Seurat dot plot highlighting exhaustion marker gene expression in PBMC and CD8 TEM according to the disease fatality status. Patients who developed fatal prostate cancer on follow-up exhibit upregulated expression of LAG3 and TIGIT in both PBMC and CD8 TEM. (B) Cytotoxicity scores for T-cell subsets comparing fatal with non-fatal disease cases. Cytotoxicity scores for the various T-cell subpopulations did not separate fatal from non-fatal cases.

Figure 3. Immune cell composition and functional status of PBMC according to patients’ survival status.

(A) Relative abundance of the 5 major immune cell types, B cells, T cells, NK cells, monocytes, and dendritic cells (DC), and (B) two DC subsets in patients with and without fatal disease outcome. Patients with fatal disease showed a reduced proportion of DC, with the most marked decrease in type 2 conventional dendritic cell (cDC2), indicating a potential impairment in antigen-presenting capacity associated with lethal progression. In contrast, there was no difference in CD4 regulatory T cells. (C) CD14, CD5, and CD1C marker expression in cDC2 cells and their association with patients’ survival status. Most cDC2 cells expressed the CD1C functional marker with fewer co-expressing CD14 and CD5. Expression of these markers tended to be lower in non-fatal cases (Padj <.05 for CD5). (D) GSEA showing enrichment in Hallmark gene sets for contrast fatal versus non-fatal disease (reference) covering cDC2, CD16⁺ monocytes, and CD8 TEM. Dot plot display of normalized enrichment scores. GSEA was performed using DESeq2-ranked, covariate-adjusted genes and evaluated against the Hallmark gene sets with FDR correction (Padj <.05 for inclusion).

Association of self-identified race with immune cell composition and transcriptional programs among cases.

Although we did not observe differences in immune cell abundance between the 26 AA and 17 EA patients when we divided the PBMC into the 5 major immune cell types, further stratification into the 26 immune cell subtypes suggested increased numbers of circulating CD4⁺ cytotoxic T lymphocytes (CD4 CTL) and type 1 conventional DCs (cDC1), and fewer NK CD56^bright cells, among AA patients (Supplementary Figure S9). However, the observed fraction of circulating cDC1 in our patient cohort was low (Supplementary Figure S3B). An additional evaluation of immune exhaustion marker expression among all PBMC showed enrichment of LAG3 and TIGIT expression in AA patients, with reduced CTLA4 (Figure 4A). Further focusing on cDC2 and CD16⁺ monocytes, two cell types that associated with fatal prostate cancer, we uncovered transcriptional profiles consistent with potential functional differences between AA and EA patients (Figure 4B) and an expression pattern for key markers in these two cell types that showed an overlap between patients who were either AA men or developed fatal prostate cancer on follow-up (Figure 4C). Expression of CD1C, FCER1A, ENHO, and CIITA, which regulates MHC-II expression, was increased in cDC2 of both AA and fatal disease patients. This observation suggests that the CD1C⁺ and high-affinity IgE receptor–positive subset of DCs, which contributes to allergic diseases by enhancing T-cell immunity and inflammation, is more common in these patient groups. For CD16⁺ monocytes, FCGR3A, MS4A7, LST1, and IFITM3 were up regulated in both the AA and fatal disease patient groups. IFITM3 is an interferon signaling response gene with its highest reported expression in CD16⁺ monocytes. LST1 is a common immune cell marker whereas FCGR3A encodes CD16 itself, suggesting an elevated CD16 expression in CD16⁺ monocytes of AA and fatal disease patients. CD16⁺ monocytes with increased expression of the canonical markers FCGR3A and MS4A7 may contribute to hyperinflammation as MS4A7 is known to activate the NLRP3 inflammasome (18). MS4A7 is also required for NFκB activation and antitumor immunity of cDC1 (19). Adding potential confounders including survival status (prostate cancer death yes/no), diabetes, aspirin use, and immunomodulatory medication as covariates to pseudobulk/GSEA for the AA vs. EA contrast did not change the pathway enrichment pattern and the GSEA-derived conclusions (Supplementary Figure S10).

Figure 4. Differences in immune cell function between AA and EA patients.

(A) Seurat dot plot showing elevated expression of LAG3 and TIGIT exhaustion markers in PBMC of AA patients, compared to EA patients. (B) GSEA showing enrichment in Hallmark gene set for contrast AA versus EA patents (reference) covering cDC2 and CD16⁺ monocytes. Dot plot display of normalized enrichment scores. GSEA was performed using DESeq2-ranked, covariate-adjusted genes and evaluated against the Hallmark gene sets with FDR correction (Padj < .05 for inclusion). (C) Seurat dot plots showing expression of canonical marker genes for cDC2 and CD16⁺ monocytes comparing the EA versus AA patient contrast with the non-fatal (NF) versus fatal (F) disease contrast.

TCR clonal expansion and repertoire remodeling in association with fatal disease.

We obtained high-quality single-cell TCR sequencing data for 37 cancer patients and 16 healthy men. The analysis of the combined cohort revealed substantial alterations in the clonal architecture in prostate cancer patients, particularly among those with fatal disease and AA patients. The TCR repertoire profiling pointed to an over-representation of large, hyperexpanded clonotypes (Figure 5A) and a lower TCR diversity (Figure 5B) among fatal disease and AA patients. Furthermore, using paired TCR sequencing, we could assign TCR sequence–defined clonal expansion to the various T-cell subtypes in our patient population (Figure 5C). This approach uncovered that hyperexpanded TCR clones (clonal frequency >1%) were predominantly observed in CD8 TEM and CD4 CTL cells from AA patients with fatal disease. The top 20 TCR clonotypes, including TRAV19-TRAJ34/TRBV12-5-TRBJ1-3, were enriched in these subgroups (Supplementary Table S3). Many of these top clones were unique to AA patients and not detected in the EA patients.

Figure 5. T cell receptor (TCR) clonotype expansion and diversity in association with disease fatality and AA identity.

(A) Stacked bar plot showing the distribution of TCR clonotypes by relative abundance across samples. Clonotypes were grouped into five categories—rare, small, medium, large, and hyperexpanded—based on their relative frequency. Six samples, all from AA patients with fatal disease, exhibit prominent hyperexpanded clones (yellow), suggesting robust T-cell clonal expansion likely driven by antigen exposure in advanced disease. Hyperexpanded clonotypes are TRBV12-5, TRBV19, TRBV27, TRBV4-1, TRBV6-5, and TRBV9. (B) TCR diversity is decreased in both fatal disease and AA patients. Diversity was measured by the Shannon entropy index. Wilcoxon rank sum test for significance testing. (C) T-cell clonotype counts across subpopulations, stratified by self-reported race and fatal disease status. Notably, all hyperexpanded clones were observed in AA patients with fatal disease, specifically within CD8 TEM and CD4 CTL subsets.

CD8 TEM transcriptional program and prostate cancer survival.

We explored the prognostic relevance of the 26 circulating immune cell populations and used LASSO regression to construct a multigene prognostic signature for each of them. This approach led to the discovery of a CD8 TEM–related survival predictive 25-gene signature (see Supplementary Table S4 for more details) that stood out in its relationship to prostate cancer–specific survival at the P ≤ .0019 significance threshold (Table 2, Supplementary Figure S11). The genes enriched in this signature showed a relationship to cytosolic DNA–sensing pathway and antiviral response (Supplementary Figure S12). Using multivariable Cox regression confirmed that the signature remained a prognostic marker of both all-cause and prostate cancer–specific mortality (Table 2; HR: 4.1; 95%CI: 2.1 to 8.0; P = 3.46 x 10⁵ for prostate cancer–specific survival). In contrast, the number of CD8 TEM in the circulation (% PBMC) did not significantly associate with survival. Yet, increased LAG3 expression in these cells also predicted fatal prostate cancer (Table 2, Supplementary Figure S11), indicating a significance of the CD8 TEM functional status in prostate cancer survival. In addition, LAG3 expression across all PMBC was also associated with an all-cause mortality (P = .008) but did not reach significance for prostate cancer–specific survival (P = .075; Table 2). Applying a sensitivity analysis with additional inclusion of diabetes status, aspirin use, and the use of immunomodulatory medication as covariates substantiated the findings (Supplementary Table S5). In this model, however, LAG3 expression across all PMBC significantly associated with prostate cancer–specific survival (P = .03) whereas LAG3 expression in CD8 TEM did not (P = .09).

Table 2:

Association of immune markers with all-cause and prostate cancer-specific mortality

	All-Cause Mortality
Exposure (as continuous variable)	N	N events	HR	95% LCI	95% UCI	P Value
CD8 TEM Signature Score	43	25	2.84	1.78	4.53	1.16E-05
CD8 TEM (%)	43	25	1.79	0.89	3.57	0.100
LAG3 in CD8 TEM	43	25	2.14	1.16	3.98	0.016
LAG3 in PBMC	43	25	2.67	1.29	5.51	0.008
	Prostate Cancer-Specific Mortality
	N	N events	HR	95% LCI	95% UCI	P Value
CD8 TEM Signature Score	43	20	4.10	2.10	8.00	3.46E-05
CD8 TEM (%)	43	20	1.59	0.74	3.39	0.235
LAG3 in CD8 TEM	43	20	2.03	1.01	4.09	0.047
LAG3 in PBMC	43	20	1.97	0.94	4.17	0.075

Cox Proportional Hazards Model adjusted for age at diagnosis (continuous), body mass index (continuous), self-reported race (AA/EA), NCCN risk category (categorical), and ADT (yes/no)

Discussion

Circulating monocytes and peripheral immunity may promote prostate cancer metastasis (5,20). Thus, their functional status should be investigated. Applying an exploratory approach and single-cell transcriptomic and TCR profiling to PBMC, we uncovered that peripheral immunity may vary distinctly in association with prostate cancer, disease fatality, and patients’ descent. Key observations were as follows: 1. In prostate cancer patients, there was a shift towards tolerogenic circulating CD4⁺ naïve T cells and an increased effector function and aging in CD8 TEM; 2. There were predicted functional differences in cDC2, CD16⁺ monocytes, and CD8 TEM cells comparing patients with and without fatal disease. These transcriptome-inferred differences describe cDC2 as immunosuppressed, CD16⁺ monocytes as pro-inflammatory but less cytotoxic, and CD8 TEM cells as chronically activated effector cells in patients with a fatal disease outcome; 3. Fatal disease and AA patients showed elevated expression of exhaustion markers and a lower TCR diversity, but with hyperexpanded TCR clones, in their PBMC. 4. The functional status of CD8 TEM and LAG3 expression in PBMC are candidate predictors of disease fatality.

To our knowledge, few studies have investigated the single-cell transcriptome of PBMC in cancer patients. One immune-profiling study described five major immunotypes based on an investigation of PBMC from 442 cancer patients, but with only 18 of them having prostate cancer (6). The authors described an over-representation of terminally differentiated CD8 TEM and loss of both CD4⁺ and CD8⁺ naïve T cells in cancer patients, when compared to healthy controls, which is consistent with our findings for prostate cancer. In a scRNA sequencing study of 124,966 PBMC covering 68 cancer patients, but none with prostate cancer, and 14 healthy controls, the authors noticed a lower proportion of naïve T cells in cancer patients, particularly CD8⁺ T cells, when compared to controls, but no differences for B cells, NK cells, or myeloid cells (21). Moreover, T-cell activation tended to be higher in cancer patients, which is noteworthy because chronic T-cell activation leads to exhaustion and apoptosis as these cells cannot revert to quiescent memory cells (22). Accordingly, GSEA indicated increased apoptosis in activated CD8 TEM of prostate cancer patients in our study. Another investigation explored the single-cell transcriptome of PBMC in 4 patients with prostate cancer and 3 patients with benign prostate hyperplasia (23). In that report of PBMC single-cell sequencing for prostate cancer patients, the authors observed differences with elevated CD14⁺ monocytes, NK and γδ T cells in cancer patients and differential expression of genes, together with pathway enrichment patterns, for various immune cells. Our observations are in variance with these findings. Lastly, the single-cell transcriptome of PBMC from 10 breast cancer patients with metastatic disease was compared to 15 publicly available single-cell PBMC transcriptomes from healthy controls and differences were reported (24). Thus, PBMC single-cell transcriptome data from cancer patients remain sparse although their analysis may reveal therapy response predictors, as previously proposed (6), or predictors of disease fatality, as suggested by our study.

TCR profiling in our study pointed to over-representation of large, hyperexpanded clonotypes and a lower TCR diversity among fatal disease and AA patients. Furthermore, hyperexpanded TCR clones were restricted to mostly CD8 TEM but also included the CD4 CTL population. It has been shown that the TCR gene rearrangement diversity changes with age and upon exposure to chronic infections and diseases like cancer. These changes may include a loss of TCR diversity and clonal expansion of memory T cells like CD8 TEM (25–28). Furthermore, these alterations to T-cell homeostasis are also hallmarks of increased T-cell immunosenescence (25,29). Others have reported that cancer patients with a low TCR diversity tend to show signs of an increased inflammatory response (26), which is consistent with our findings where the CD8 TEM population appears to be both a key source of hyperexpanded clonotypes and proinflammatory signaling. It is assumed that a low TCR diversity may reflect a decreased probability of tumor-antigen recognition and thereby decreases the probability of cancer control. Accordingly, a low peripheral TCR diversity is a poor prognosis marker and associates with cancer progression, reduced response to immune blockade therapies, and decreased survival of renal cell carcinoma, non-small cell lung cancer, and bladder cancer patients (26,30,31).

Immune exhaustion markers are targets of cancer immunotherapy (29). They include cell-surface proteins involved in immune checkpoint blockade, such as PD-1, CTLA-4, Tim-3, TIGIT, and LAG-3. Although immune checkpoint blockade therapies have not been successful in the treatment of prostate cancer, efforts to include them continue (32). We observed that LAG3 and TIGIT were upregulated in PBMC and CD8 TEM from fatal disease patients, suggesting increased peripheral exhaustion affecting these patients. There are some reports on the effect of TIGIT expression in prostate cancer patients. One found that the presence of TIGIT on NK cells enhances checkpoint blockade immunity against castration-resistant prostate cancer (33). LAG3 is more widely expressed by PBMC than TIGIT and is an immunotherapy target in various clinical trials (34). Anti-LAG3 therapy for prostate cancer is currently at the stage of preclinical evaluation, whereas the anti-LAG3 Relatlimab has been FDA approved for treatment of melanoma patients (35). Like with TIGIT, there are few reports linking LAG3 expression to prostate cancer. LAG3 was reported as being up-regulated in prostate tumors of patients at high risk of a biochemical recurrence (41) and may show population heterogeneity in its expression (36).

Testosterone affects immune function. Testosterone has a direct effect on T cells and reduces the activation of CD4⁺ T cells via androgen receptor signaling (37). Vice versa, ADT leads to an expansion of peripheral CD4⁺ naïve T cells in prostate cancer patients, whereas overall numbers of T and B cells are not altered (38). ADT enhanced the response to immunotherapy in a model of prostate cancer (39). More recently, it was recognized that inhibition of androgen receptor signaling prevents T-cell exhaustion and improves T-cell responses and tumor control, and the response to immune blockade therapy (40,41). Together, the findings establish the androgen signaling pathway as a driver of T-cell dysfunction in prostate cancer that can be targeted by ADT. We explored the influence that ADT may have on our findings related to the peripheral immune status in patients with fatal prostate cancer. More of these than other patients were on ADT at recruitment. However, adding ADT as a covariate in our analysis showed that our observations were largely independent of ADT.

Our study has several strengths. It provides an assessment of peripheral immunity in men with prostate cancer applying single-cell analysis of PBMC that included an investigation of both the transcriptome and TCR repertoire. We used a well-characterized historical cohort (5,42) that includes AA patients, a high-risk patient group, which is a strength. All patients in this cohort had a NCCN-defined risk assessment and long-term survival follow-up. We controlled for potential confounders and multiple comparisons in our analysis, which represents another strength. Yet, our study has limitations. It is an observational study and does not allow for causal inference. Yet, the observations in association with fatal disease are based on PMBC collections that preceded patients’ death by an average of more than 3 years. Thus, these observations may capture a future mortality risk. The study was geographically restricted to Baltimore and surrounding areas in Maryland and had only 43 patients in our analysis, which limits power and generalizability. Still, when observations could be compared with the existing literature, we could validate them.

Conclusion

Our observations suggest common alterations to peripheral immunity in patients who are at an increased risk of prostate cancer morality. These observations describe a survival predictive 25-gene signature in CD8 TEM and survival predictive LAG3 expression in these cells and PBMC; both observations may support the future stratification of patients into a group with immune dysfunction and high mortality risk. Moreover, the CD8 TEM functional status in association with fatal disease constitutes a candidate therapeutic target where reprogramming of these cells with an adjustment of their metabolic fitness or reversing exhaustion via checkpoint inhibition may improve patient survival.

Synopsis:

Immunologic factors influencing outcome for prostate cancer patients may be intratumoral or peripheral. Through single-cell transcriptomic and TCR analysis of PBMCs, the authors identify differences in peripheral immunity that may vary in association with disease outcome.

Abbreviations:

PBMC

peripheral blood mononuclear cells

AA

African American

EA

European American

TCR

T cell receptor

DC

dendritic cells

CD8 TEM

effector memory CD8⁺ T cells

CD4 CTL

CD4⁺ cytotoxic T lymphocytes

NCCN

National Comprehensive Cancer Network

GSEA

Gene Set Enrichment Analysis

ADT

androgen deprivation therapy

Data Availability Statement.

The single-cell RNA sequencing data generated in this study are publicly available in Gene Expression Omnibus (GEO; RRID:SCR_005012) under accession number GSE270387. The data can be linked to the patient data in Supplementary Table S1 through the listed accession numbers.

Code Availability.

The source codes, analysis scripts, and computational workflows supporting the cancer immunology analyses presented in this study have been deposited in a Code Ocean capsule to facilitate transparency, reproducibility, and reuse. The capsule contains the computational resources required to reproduce the principal bioinformatic analyses, including immune-cell annotation, single-cell transcriptomic profiling, marker-gene analysis, module-score calculation, and visualization of immune-cell states. The Code Ocean capsule can be viewed using the following link: https://codeocean.com/capsule/9637556/tree/v1