A Subset of Localized Prostate Cancer Displays an Immunogenic Phenotype Associated with Losses of Key Tumor Suppressor Genes

This article is featured in Highlights of This Issue, p. 4665

Immune checkpoint inhibitors (ICI) that target the programmed cell death ligand-1 (PD-L1)/programmed cell death-1 (PD-1) pathway have demonstrated antitumor activity in a growing number of malignancies and are approved by the FDA for treatment of microsatellite instability–high (MSI-H)/mismatch repair–deficient (dMMR) or high tumor mutational burden (TMB) solid tumors. Broadly, tumor features associated with clinical response to anti–PD-1/PD-L1 therapy include T-cell infiltration, PD-L1 expression in tumor and/or immune cells, increased TMB, and IFNγ-derived T-cell gene expression profiles (T-GEP; refs. 1, 2).

In advanced prostate cancer, the effect of anti–PD-1 ICI has been limited aside from the small subset of tumors that exhibit MSI-H/dMMR (3) or CDK12 biallelic inactivation (4). Overall response rate (ORR) to pembrolizumab monotherapy has been estimated at 5%–17% in unselected metastatic castration-resistant prostate cancer (mCRPC; refs. 5–7). Preliminary results of nivolumab plus ipilimumab (anti–CTLA-4) in a mCRPC phase II trial (CHECKMATE-650) showed an ORR of 10% and 26% in patients with and without previous taxane-based chemotherapy, respectively (8). The impact of anti–PD-1 ICI in patients with nonmetastatic disease remains unknown.

PD-L1 expression in prostate cancer (≥5% of tumor cells staining positively) has been described from 15%–52% in the primary setting (9–11) and from 4%–50% in castration-resistant disease (12, 13). Tumor heterogeneity, antibody selection, IHC protocols, and scoring are responsible for the variability of expression. To account for these differences, our group previously used two antibodies to evaluate PD-L1 expression in untreated radical prostatectomy (RP) specimens and matched RP cases that had been treated on trial with neoadjuvant hormonal therapy (medical castration plus abiraterone and prednisone) and had residual CRPC foci (14). This approach identified PD-L1 expression in 13.8% of localized prostate cancer overall, and in 26.5% of Grade Group (GG) 4–5 tumors, with no increase in neoadjuvant-treated tumors. High numbers of CD8⁺ tumor-infiltrating lymphocytes (TIL) positively correlated with PD-L1 expression. Interestingly, PD-L1 expression was more frequent in tumors from Black patients, consistent with data suggesting uniquely immunogenic phenotypes in this population (15–18).

These data suggest that a subset of localized prostate cancer tumors, possibly enriched among high-grade cases although not consistently (19), are eliciting immune responses, which may be predictive of responsiveness to ICI. Both the expression of PD-L1 and increased number of TILs are reflective of interferon gamma exposure (20, 21), and spatial distribution of TILs has identified three distinct cancer immune phenotypes: immune-excluded, immune-desert, and immune-inflamed (20). However, an inflamed microenvironment does not ensure response to ICI. Chronic inflammation, which is frequent in cancer, can result in an overactive immune response that ultimately leads to decreased cytotoxicity of CD8⁺ T cells (22–24). However, a subset of exhausted T cells can be reinvigorated by ICI, leading to proliferation and activation of antitumor responses (25, 26). Recently, transcription factor TCF1 (encoded by TCF7) has been linked to T cells with a “stem-like” phenotype, which have exhausted progenitor features and are associated with patient response to ICI (25–28). Other immune checkpoints associated with T-cell exhaustion, like TIM3, LAG3, TIGIT, and VISTA, may also play a role in regulating antitumor T-cell response.

Here we selected a series of localized prostate cancer cases based primarily on PD-L1 expression, with several additional cases having extensive TILs but low or absent PD-L1 expression, for genomic and microenvironment analyses. In contrast to the PD-1–negative antigen-naïve phenotype of sparse CD8⁺ T cells in unselected cases, the majority of CD8⁺ T cells in these selected cases had an effector phenotype, with a subset expressing TCF1, indicative of exhausted progenitor cells that may respond to ICI. These were found within antigen-presenting cell (APC) niches. Genomic analysis showed significant enrichment for alterations in RB1 and BRCA2, and deep deletions in CHD1 were also prevalent. These results indicate that a subset of localized prostate cancer (which we term immunogenic) are eliciting immune responses, that these tumors may have genomic features distinct from previously described drivers of immunogenicity, and may respond to ICI.

In accordance with the U.S. Common Rule and after Institutional Review Board (IRB) approval, RP tissue (or biopsy cores alone in three cases) was collected retrospectively between 2007 and 2019 at Beth Israel Deaconess Medical Center (BIDMC, Boston, MA) and deidentified in accordance with BIDMC IRB protocol #2010-P-000254. Given previous findings of enriched PD-L1 expression and TILs among high-grade cases, we examined 115 total cases that were GG 4–5 (50 cases), or had other National Comprehensive Cancer Network (NCCN) high-risk features (extracapsular extension or serum PSA >20 ng/dL; 54 cases). We identified 29 total cases (25% of screened cases) with tumor PD-L1 expression ≥5% by IHC; we included an additional three RP cases and three cases with only core biopsies available, all PD-L1–negative but with high TILs [≥20 lymphocytes per high-power field (hpf)], forming a cohort of 35 total cases (30% of screened cases). These cases comprised the “immunogenic” cohort. Additional MSI-H cases [also included in T-cell receptor (TCR) sequencing analysis] were analyzed with multiplex panel #3 described below. Clinical and pathologic details are provided in Supplementary Table S1.

Prostate tissue was fixed in formalin, processed, and embedded in paraffin using standard methods. For each case, the tissue block with the largest dimensions of the dominant tumor plus two or three additional blocks were selected for PD-L1 and CD3 IHC. Additional details are presented in Supplementary Methods.

Immunogenicity was defined as PD-L1 moderate to strong membranous staining in ≥5% of tumor cells; cytoplasmic staining was not considered. High density of TILs, defined as ≥20 lymphocytes per hpf either present within the tumor cell nests or glands or immediately adjacent to tumor cells (i.e., distance between the lymphocyte and its nearest tumor cell being less than the diameter of an average tumor cell), was used to include six PD-L1–negative cases. All immunostains were evaluated by an experienced pathologist (H. Ye) and a trained MD investigator (C. Calagua).

We selected 11 immunogenic cases for DNA sequencing. We used hematoxylin and eosin (H&E) and IHC (PD-L1, CD3) staining to annotate areas as immunogenic (PD-L1 ≥5% and extensive TILs), nonimmunogenic (PD-L1–negative and no TILs), and normal (benign glands and/or stroma nonadjacent to tumor cells). We created tissue microarrays (TMA) using the EZ Manual Microarray Kit (IHC WORLD). Each core measured 3 mm in diameter and TMAs included one or more immunogenic areas, one nonimmunogenic area, and one benign area.

We extracted DNA, generated libraries, and performed hybrid capture with a custom bait panel targeting exons for 754 genes including those in commercial platforms, additional genes altered in prostate cancer (29), and genes related to immune function. Full details are outlined in Supplementary Methods. We then performed targeted sequencing, with median on-bait coverage of 252× (range: 101× to 427×). We identified somatic and germline mutations, somatic copy-number alterations, TMB, and percent genome altered as outlined in the Supplement. Data are available in the database of Genotypes and Phenotypes (dbGaP, accession number: phs002398.v1.p1).

GISTIC2-summarized copy-number values were downloaded from cBioPortal for 489 prostate tumors from The Cancer Genome Atlas (TCGA). Amongst cases where both RB1 and BRCA2 deletions were identified, the SEG file from each tumor was manually inspected using Integrative Genomics Viewer (IGV) to determine if the deletions were part of a single focal event (i.e., co-deleted) or occurred individually. A two-sided Fisher exact test was used to compare the frequency of focal cooccurrence events between the The Cancer Genome Atlas (TCGA) cohort and our immunogenic cohort. P values were adjusted [false discovery rate (FDR)] using the method of Benjamini and Hochberg.

To investigate T-cell clonality, we extracted genomic DNA as above and sequenced the VDJ region of the TCR β chain using the immunoSEQ multiplex PCR-based method (Adaptive Biotechnologies). We used Adaptive Analyzer to analyze output.

We used multiplex immunofluorescence (IF) panel 1 (see Supplementary Methods and Supplementary Table S2) to evaluate a set of unselected RP cases, spanning all GGs. This seven-plex panel included antibodies to CD3, CD8, PD-1, FoxP3, CD68, CD163, along with DAPI counterstaining. Regions of interest (ROI) included both immune-cell–rich and nonrich areas and included both tumor and benign areas. For analysis of CD8⁺ T-cell density, we quantified lymphocytes present in gland plus stroma. For analysis of CD8 and PD-1 coexpression, we analyzed lymphocytes present in gland only (restricted to epithelial cells) since the ROIs analyzed in comparison immunogenic cases were nearly devoid of stroma.

For multiplex IF panel 2 applied to immunogenic cases and to renal cell carcinoma (RCC) cases, an in-house six-plex assay was optimized, including antibodies to CD8, PD-1, TIM3, LAG3, and TCF1, along with DAPI counterstain (Supplementary Table S3). Details of the image acquisition and image analysis workflow were reported previously (30).

Multiplex IF panel 3 consisted of a previously described panel of antibodies to CD8, PD-1, TCF1, and MHC II, and DAPI counterstaining (ref. 31; Supplementary Table S4).

We performed comparisons of multiplex IF datasets with the Mann–Whitney test using GraphPad Prism 9 (GraphPad Software). For multiple comparisons within multiplex IF datasets, we adjusted alpha using the Bonferroni correction; otherwise, alpha was set at 0.05.

While immune infiltrates are generally uncommon in primary prostate cancer, we observed previously that multiple tumor foci in 14%–27% of localized prostate cancer, enriched in high-grade cases, had PD-L1 expression and associated T-cell infiltrates (Fig. 1A–F). To determine whether these cases may reflect a distinct immunogenic subset, we first examined randomly selected ROIs in unselected, primarily high-risk RP cases, with multiplex IF panel 1 for CD8 and PD-1 expression. In these cases, as expected, CD8⁺ infiltrate was quite low, with a median density of 54 cells/mm² (Fig. 1G). In addition, a very small percentage of the already rare CD8⁺ cells coexpressed PD-1, indicating they were primarily antigen-naïve (Fig. 1H; Supplementary Table S5).

We next examined a series of primary prostate cancer cases selected primarily on the basis of PD-L1 expression, which we termed “immunogenic.” We characterized TILs from these cases in lymphocyte-rich ROIs using multiplex IF panel that included CD8 and PD-1 (IF panel 2). Tumor foci in these cases were highly enriched in CD8⁺ TILs compared with unselected prostate cancers (P = 0.0002, Fig. 1G). The median CD8⁺ cell density within lymphocyte-rich ROIs was 253 cells/mm² (Supplementary Table S6). Furthermore, PD-1 expression indicated that these TILs were much more antigen exposed: a median of 42% of CD8⁺ cells coexpressed PD-1 in this prostatectomy cohort, compared with a median of 0% in unselected prostatectomy cases (P < 0.0001; Fig. 1H). Of note, the unselected cases were examined with the same antibody but a differently optimized IF panel and applied to random tumor foci, so it may underestimate total CD8⁺ T-cell content if unselected cases had rare T-cell infiltrates not detected by H&E and CD3 staining. However, these findings are consistent with other literature showing rare lymphocyte infiltration in localized (as well as metastatic) prostate cancer (32). Thus, these data indicate tumor foci with extensive CD8⁺ T-cell infiltration are found in approximately 30% of aggressive localized prostate cancer, and that these CD8⁺ T cells are qualitatively distinct in having evidence of antigen exposure.

We further investigated CD8⁺ lymphocyte phenotypes in these TIL-infiltrated cases by multiplex IF with a panel against CD8, PD-1, LAG3, TIM3, and TCF1 (IF panel 2; Fig. 2A). In contrast to the substantial fraction of CD8⁺ T cells expressing PD-1 (median 42%), smaller fractions expressed TCF1 (19%), TIM3 (9%), or LAG3 (2%; Fig. 2B; Supplementary Table S6). The large majority (median 85%) of CD8⁺PD-1⁺ cells were negative for TIM3 and LAG3, indicating that they were not terminally exhausted, whereas 15% (median) coexpressed either TIM3 or LAG3, suggestive of terminal exhaustion (Fig. 2C). Within the CD8⁺PD-1⁺ TIM3(-) LAG3(-) subset we found that a median of 34% coexpressed TCF1, suggestive of an exhausted progenitor phenotype, whereas a median of 66% did not coexpress TCF1, TIM3, or LAG3, indicating possible differentiated effector status. As expected, TCF1⁺ cells rarely coexpressed TIM3 or LAG3 (Supplementary Table S7). Overall, PD-1 was the most expressed exhaustion marker, followed by TIM3.

To benchmark our results, we compared them to results obtained with the same multiplex IF panel and same ROI selection in metastatic clear-cell renal cell carcinoma (mccRCC) nephrectomy and metastatectomy specimens from the CheckMate-025 trial, a phase III randomized trial comparing nivolumab with everolimus in patients with RCC previously treated with anti-angiogenic therapy. We recently characterized mccRCC immune infiltrates in this trial (30). Median CD8⁺ density was nearly identical between immunogenic prostate cancer and mccRCC (P = 1.00, Fig. 2D; Supplementary Table S8). CD8⁺ TILs coexpressed PD-1 (and coexpressed PD-1 without TIM3 or LAG3) at comparable rates (P = 0.22 and P = 0.24, respectively). Interestingly, exhausted progenitors [CD8⁺PD-1⁺TCF1⁺ TIM3(-) LAG3(-)] were the one cell subtype seen at higher rates in immunogenic prostate cancer compared with mccRCC (P = 0.0009; Fig. 2E), which remained statistically significant with correction for multiple-comparison testing.

In murine models of chronic viral infection, the T-cell response is known to be maintained by a progenitor-type cell that expresses the transcription factor TCF1 and resides predominantly in the lymphoid tissue (25). Interestingly, a number of studies reported the identification of analogous CD8⁺TCF1⁺ exhausted progenitor T lymphocytes directly in tumor tissue (25, 28, 33), and we previously observed these cells in close proximity to MHC-II⁺ APCs. These aggregated TCF1⁺ T cells and MHC-II⁺ APCs form intratumoral APC niches (31), which closely resemble the T-cell zone of lymphoid tissue, suggesting that these niches may support TCF1⁺ stem-like cell survival by analogous mechanisms to those in the lymphoid tissue. Previous studies examined these TCF1⁺ T cells and this pattern of immune organization in predominantly classically immunogenic tumor types, with a small subset of analysis in unselected prostate cancer.

Here, we identified intratumoral APC niches in immunogenic prostate cancer (Fig. 3A–D). As in prior work, APC niches are defined as areas within tumor tissue with MHC-II⁺ cells and CD8⁺TCF1⁺ T cells identified in the same 100 × 100 μmol/L area. These niches were present in the described immunogenic prostate cancer cohorts, as well as in two MSI-H prostate cancer cases (one RP specimen, one biopsy from a patient with mCRPC at time of resistance to PD-1 blockade after initial response), and the percentage of tissue with these APC niches was not significantly different when comparing the immunogenic prostate cancer cohorts to MSI-H cases (P = 0.15; Fig. 3E). These findings are consistent with the conclusion that these tumors are eliciting immune responses, although the extent to which this is predictive of responsiveness to ICI remains to be determined. The percentage of tumor tissue occupied by APC niches loosely correlated with amount of infiltrating MHC-II⁺ cells (R² = 0.18, P = 0.065; Fig. 3F) and more tightly correlated with the number of infiltrating CD8⁺ and TCF1⁺ TILs (R² = 0.57, P = 0.0001 for both comparisons; Fig. 3G and H), further supporting the role of these APC niches in maintaining the tumor T-cell response.

We sequenced the TCR β VDJ region in a pilot study of five randomly selected immunogenic cases, including a separate positive control case with MMR deficiency as demonstrated by loss of MSH2 and MSH6 by IHC (PCA25-MMR). Total productive rearrangements ranged from 2,480 to 13,645 (Supplementary Fig. S1). Productive clonality ranged from 0.05 to 0.11. Maximum productive frequency ranged from 0.79% to 5.81%. The clones with productive rearrangements were more expanded, suggesting antigen-driven expansion. The top productive clones made up approximately 1%–5% of total clones, and the top 10 productive clones made up approximately 5%–15% of total, indicating antigen-driven oligoclonal expansion. The dMMR case was comparable to the other cases along all parameters.

We next carried out microdissection of matched immunogenic and nonimmunogenic tumor foci and exome sequencing in our first 11 immunogenic cases to evaluate genomic alterations. Of note, there was heterogeneity with respect to immune infiltrate in many cases, with some tumor foci near immunogenic foci having few immune cells and low or absent PD-L1 expression. Therefore, where possible we microdissected two or more immunogenic foci and a nonimmunogenic area from each case, as well as an area without tumor for germline DNA.

The most frequent alterations observed were in BRCA2 and RB1, which are colocalized on chromosome 13 (Fig. 4; Supplementary Fig. S2). Shallow RB1 deletions were found in immunogenic tumor foci from eight cases, and deep deletion was observed in one case (case 3, where a nonsense mutation was also found in one focus). In one additional case (Case 10) we found a shallow RB1 deletion in the nonimmunogenic focus, but not the immunogenic foci analyzed. This frequency of RB1 alterations was higher (10 of 11 cases) than in localized prostate cancer included in TCGA Firehose Legacy (218 out of 489; P = 0.003, P_adj = 0.017; Fig. 5A), which remained significant after excluding GG 1 (P = 0.004, P_adj = 0.011) or GG 1–2 (P = 0.003, P_adj = 0.006) tumors. On the basis of these observations, we carried out IHC for RB1 protein in a case where shallow RB1 loss was found in immunogenic foci, but not in a nonimmunogenic focus. This showed intact expression in the latter nonimmunogenic focus and heterogeneous staining in the immunogenic foci (Fig. 5B), suggesting that a subpopulation of tumor cells may have two-copy losses.

Five of these cases with RB1 loss also had corresponding losses of BRCA2, which is located adjacent to RB1 on chromosome 13. Of note, two of these cases with BRCA2 loss also had germline frameshifts in BRCA2 (Cases 1 and 3). The frequency of BRCA2 loss (6 of 11) was higher than in TCGA (126 of 489; P = 0.042, P_adj = 0.053; Fig. 5A), which remained significant after excluding GG 1 tumors (P = 0.037, P_adj = 0.057) but not GG 1–2 (P = 0.077, P_adj = 0.078). Focal BRCA2/RB1 codeletion has recently been described as a marker of aggressiveness and possible sensitivity to PARP inhibitors (34). In the immunogenic cases, the frequency of focal clonal co-deletion (3 of 11) was significantly enriched compared with prostate TCGA cases (23 of 489; P = 0.016, P_adj = 0.039; Fig. 5A).

Among other genes recurrently altered and involved in DNA damage response (DDR), we also found deep deletions in CHD1 in four cases, which is greater than the approximate 5%–10% frequency of deep deletions found in unselected locally advanced or metastatic prostate cancer (although statistical comparison with TCGA data for any CHD1 loss did not reach statistical significance, likely due to small numbers and limited power; Fig. 5A). However, two-copy loss compared with wild-type (WT) was significantly enriched compared with TCGA (P = 0.006), unchanged if excluding GG 1 (P = 0.012) or GG 1–2 (P = 0.010). We then performed single-sample gene set enrichment analysis (ssGSEA), comparing TCGA tumors with deep CHD1 loss versus WT. The top differentially expressed gene set was BIOCARTA_DC_PATHWAY (P < 0.001, P_Adj = 0.003; Fig. 5C). This appeared to be driven predominantly by increased expression of the dendritic cell marker ITGAX (CD11c) in CHD1-loss tumors (P = 0.003; Fig. 5D). Interestingly in all cases there were corresponding losses in RASA1, a suppressor of RAS activity. While this may reflect proximity on chromosome 5, the codeletion of CHD1 and RASA1 is rare in unselected cases. Pathogenic SPOP mutations were also found in three of the cases with CHD1 loss, consistent with a previously reported association between CHD1 loss and SPOP mutations (35).

An additional altered DDR gene was ERCC5, which had deep deletions in two cases, although these were found in the nonimmunogenic foci, but not the corresponding immunogenic foci. Further potential tumor suppressor genes with mutations or losses in three or more cases were DIS3, FMN1, LHFP, and SLCO4C1 (which may be linked to CHD1). A missense mutation was found in TP53 in the immunogenic and nonimmunogenic foci from one case. Interestingly, we did not observe deep losses in PTEN, which has been linked with immunosuppressive tumor immune microenvironment (36), which would be expected in a substantial minority of aggressive prostate cancer.

The nonimmunogenic foci analyzed in all cases had some shared alterations with the corresponding immunogenic foci, indicating they were clonal. However, in some cases there were additional alterations in the immunogenic foci. Several cases with shallow RB1 and BRCA2 losses in the immunogenic foci appeared to be WT in the corresponding nonimmunogenic foci. Notably, three of four cases with deep CHD1 loss in immunogenic foci were WT in the corresponding nonimmunogenic foci. FMN1, with a role in adherens junction formation, had mutations in immunogenic foci from two cases that were not detected in the corresponding nonimmunogenic foci. Full results are displayed in Supplementary Figs. S3–S5.

Percent of genome altered (PGA), defined as the percentage of the genome affected by copy-number variation, has been described as a prognostic biomarker in patients undergoing RP. Patients with tumors having PGA greater than the median of 5.4% had a significantly higher risk of relapse (37). PGA in immunogenic and nonimmunogenic tumor foci from immunogenic cases was at or above 5.4% in all but three cases (PCA4, PCA5, and PCA8; Fig. 6A and C). The overall median PGA was much higher in immunogenic foci (11.2%) compared with all foci (1.2%), although within individual cases, the correlation between immunogenic and nonimmunogenic foci PGA was variable. This could also be due to contamination of the “nonimmunogenic” foci by neighboring immunogenic foci.

TMB, the number of non-synonymous mutations present per megabase (mut/Mb), is a potential biomarker for response to immunotherapy. Median TMB in these immunogenic cases was 1.73 mut/Mb in the TIL(-) PD-L1(-) nonimmunogenic foci and 3.88 mut/Mb in TIL-high PD-L1⁺ immunogenic foci (Fig. 6B). The latter is higher than the 1.36 mut/Mb and 2.93 mut/Mb previously described in primary and metastatic prostate cancer, respectively (29), but lower than the threshold of 10 mut/Mb recently used for a tumor-agnostic FDA approval of PD-1 inhibition. TMB was not consistently higher in immunogenic foci compared with the corresponding nonimmunogenic focus (Fig. 6B). However, four cases did exhibit TMB of 10 mut/Mb or greater; notably, TMB was heterogeneous even across immunogenic foci within these cases, meaning that low TMB in a core biopsy of one area of the prostate would not rule out high TMB in another area (Fig. 6D). A small subset of advanced prostate cancer may harbor mismatch repair deficiency leading to microsatellite instability and high TMB. In the immunogenic cases, we did not find mismatch repair alterations by next-generation sequencing (NGS) or IHC for the four mismatch repair proteins, nor did we identify any POLE mutations.

Prostate cancer has been considered to be a poorly immunogenic tumor, with low expression of PD-L1, sparse T-cell infiltration, and little clinical benefit in response to ICI (5, 38). In contrast to this general characterization, we demonstrate that a substantial subset of localized prostate cancer, approximately one quarter of the screened high-grade or high-risk cases, demonstrates tumor PD-L1 expression and substantial CD8⁺ T-cell infiltration. The CD8⁺ infiltrates in these cases include exhausted progenitor (expressing TCF1) and effector phenotypes, and are comparable with those found in metastatic RCC, which is among the most immune-infiltrated cancer types.

TCF1 is necessary to support a progenitor CD8⁺ T-cell population by repressing proexhaustion factors and inducing BCL6, a promoter of progenitor phenotype (39). It also appears to be essential for response to both vaccination and combined CTLA-4/PD-1 blockade in mouse models (25, 28) and has been characterized as a predictor of response to ICI in melanoma (33). Exhausted progenitor cells are found in close proximity to APCs in localized prostate, kidney, and bladder specimens, and they give rise to clonally related terminally differentiated T cells via epigenetic derepression of key genes involved in differentiation (31). Compared with mRCC, immunogenic prostate cancer appears to contain a higher density of exhausted progenitors. In Checkmate-025, a trial comparing PD-1 blockade with everolimus in mRCC, we recently found that density of exhausted progenitors had a significant interaction with treatment arm with respect to progression-free survival (PFS; ref. 30). Taken together, in immunogenic prostate cancer cases, the findings of exhausted progenitors in APC niches, differentiated effector T cells, and oligoclonal expansion imply an active and tumor-specific immune response.

Terminally differentiated T cells are the primary target of ICIs in clinical use and in development. Here, we find that PD-1 is the primary exhaustion marker expressed in immunogenic localized prostate cancer, with some coexpression of TIM3. Similarly, in CheckMate-010, a study of PD-1 blockade in advanced RCC, we found that PD-1⁺ TIM3(-) LAG3(-) comprised approximately half of CD8⁺ TILs, with TIM3⁺ TILs present at lower levels and only rare LAG3⁺ TILs (40). Percentage of CD8⁺PD-1⁺ TIM3(-) LAG3(-) was associated with improved clinical outcomes, potentially due to a subpopulation of TCF1⁺ cells, although this was not evaluated. Moreover, the combination of two biomarkers—high tumor PD-L1 expression and high percentage of CD8⁺PD-1⁺ TIM3(-) LAG3(-) TILs—identified a population of patients that had particularly good outcomes with PD-1 blockade. Conversely, absence of these two biomarkers was associated with a 0% response rate. We recently independently validated these findings in RCC in data from Checkmate-025 (30). Thus, our finding of a CD8⁺PD-1⁺ population without TIM3 or LAG3 in immunogenic, PD-L1–expressing prostate cancer mirrors findings in the most PD-1 blockade-responsive subset of kidney cancer.

TIM3 is inversely correlated with TCF1 (25), making these two markers useful to assess the balance between progenitors and terminally differentiated cells. TIM3 is an important marker that in combination with PD-1 is associated with a particularly exhausted (but importantly, not senescent) T-cell phenotype, and it can be co-targeted along with PD-1 for potentially synergistic effects (41, 42). In localized kidney cancer, the percentage of CD8⁺PD-1⁺TIM3⁺ TILs is the one CD8⁺ T-cell subset that negatively correlates with clinical outcomes and has impaired function after stimulation (43). Similar findings were seen in ovarian high-grade serous carcinoma (44). TIM3 represents a particularly attractive therapeutic target given its more restricted expression when compared with checkpoints such as CTLA-4, and thus potentially a more favorable toxicity profile.

We were able to identify a number of genomic alterations present in these immunogenic cases, and to differentiate genomic alterations between immunogenic and nonimmunogenic foci within these cases. One key finding is the prevalence of alterations in DNA-damage repair genes, including BRCA2 and CHD1, loss of which leads to decreased error-free homologous recombination (HR) repair of double-strand DNA breaks and therefore increased error-prone non-homologous-end-joining (NHEJ) repair (45). Another key finding is the marked prevalence of loss of RB1, primarily as shallow deletions, but with occasional deep deletions and one truncating mutation. Loss of RB1 is associated with aggressive and castration-resistant prostate cancers (CRPC), and it results in replicative stress. In particular, the co-alteration of BRCA2 and RB1, even as single-copy loss, has been recently identified as associated with a uniquely aggressive subtype of localized prostate cancer (34). Pending confirmation in a larger cohort, the prevalence of these findings raises the hypothesis that these alterations result in replicative stress and more error-prone NHEJ repair. This may result in increased (but potentially subclonal) mutational burden, or in disrupted mitosis and cytoplasmic DNA, leading to Stimulator of Interferon Genes (STING) pathway activation.

Our analysis has so far primarily focused on infiltrating T cells, but the immune microenvironment clearly contains other elements, especially in advanced disease. Single-cell RNA sequencing (RNA-seq) of advanced prostate cancer bone metastases has revealed depletion of B cells and an increase in M2-polarized tumor-associated macrophages (TAM; ref. 46). This is associated with increased exhaustion and decreased cytotoxicity programs in cytotoxic T lymphocytes in a CCR6-CCL20–dependent manner. Similarly, myeloid-derived suppressor cells (MDSC) emerge in the context of castration resistance (47) and could be a mechanism of resistance to ICIs in advanced CRPC. Intriguingly, CHD1 loss has been recently linked via IL6 signaling with reduced MDSCs and increased CD8⁺ T cells in prostate cancer models (48), suggesting one mechanism through which CHD1 loss may drive immunogenicity. In our immunogenic tumors with deep CHD1 deletions, SPOP inactivating mutations were commonly present. This association has been noted previously and linked to enhanced DNA damage, but in an immune context may also reflect SPOP activity as a ubiquitin ligase for PD-L1. Notably, deep CHD1 deletions were highly associated with increased dendritic cell–marker expression in TCGA, supporting an increase in APC activity in these tumors. Thus, CHD1 loss, alone or in combinations with SPOP mutation, may drive immunogenicity by several mechanisms, and be a biomarker of response to ICI.

These data provide a rationale for testing ICI in earlier prostate cancer spaces and with biomarker selection. We are currently accruing patients to a trial of PD-1 blockade in high-risk biochemically recurrent prostate cancer, over-sampling for PD-L1–expressing primary tumors (NCT03637543), and will test for correlations between the features identified above and clinical response. Another trial is testing neoadjuvant PD-1 blockade with or without IL6 blockade, and it will be interesting to see if CHD1 status has any effect on the IL6 blockade (NCT03821246). A third trial is testing PD-1 blockade in combination with chemotherapy for newly metastatic castration-sensitive prostate cancer, with genomic and proteomic biomarker selection (NCT04126070). Interestingly, while most patients with de novo metastatic prostate cancer appear to have nonimmunogenic tumors, there is a high-TIL subset, similarly correlated with high-grade histology (49), as in our localized tumors. Finally, further biomarker work may allow for the identification of immunogenic tumors prior to RP and subsequent development of biomarker-driven neoadjuvant immunotherapy trials that exploit intratumoral progenitor T-cell populations.

D.B. Skyes is a cofounder and holds equity in Clear Creek Bio, is a consultant and holds equity in SAFI Biosolutions, and is a consultant for Keros Therapeutics. H. Ye reports personal fees from Janssen Pharmaceuticals outside the submitted work. S. Signoretti reports grants from Bristol Myers Squibb and Exelixis, grants and personal fees from AstraZeneca, and personal fees from Merck, CRISPR Therapeutics, NCI, and AACR outside the submitted work; in addition, S. Signoretti has a patent for Biogenex with royalties paid. D.J. Einstein reports grants from Prostate Cancer Foundation and Department of Defense during the conduct of the study, as well as grants from Bristol Myers Squibb and other support from Foundation Medicine outside the submitted work. No disclosures were reported by the other authors.

C. Calagua: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft. M. Ficial: Resources, data curation, software, investigation, writing–review and editing. C.S. Jansen: Resources, data curation, formal analysis, investigation, visualization, methodology, writing–review and editing. T. Hirz: Resources, data curation, investigation, writing–review and editing. L. del Balzo: Data curation, software, investigation, methodology. S. Wilkinson: Resources, investigation. R. Lake: Resources, investigation. A.T. Ku: Data curation, formal analysis. O. Voznesensky: Investigation, methodology, project administration. D.B. Sykes: Resources, writing–review and editing. P.J. Saylor: Resources, writing–review and editing. H. Ye: Conceptualization, supervision, writing–review and editing. S. Signoretti: Conceptualization, resources, supervision, investigation, methodology. H. Kissick: Conceptualization, resources, formal analysis, supervision, investigation, writing–review and editing. A.G. Sowalsky: Data curation, software, formal analysis, supervision, methodology, writing–original draft, writing–review and editing. S.P. Balk: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing. D.J. Einstein: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing.

The authors gratefully acknowledge the patients and the families of patients who contributed to this study. We wish to acknowledge Ying Huang, PhD and the Molecular Pathology Core Lab at Dana-Farber Cancer Institute for providing multiplex IF on unselected prostatectomy samples, Yue Sun, MD at Beth Israel Deaconess Medical Center for pathology support, and Kai Wucherpfennig, MD, PhD at Dana-Farber Cancer Institute for scientific advice and review of data. Portions of this work utilized the computational resources of the NIH HPC Biowulf cluster. This work was supported by the Intramural Research Program of the NIH, NCI (to A.G. Sowalsky); a Prostate Cancer Foundation Challenge Award 18CHAL09 (to S.P. Balk, D.J. Einstein, H. Ye); NCI grants 1-F30-CA-243250 (to C.S. Jansen), P50 CA090381 (to S.P. Balk), and Developmental Research Project P20 CA233255 (to S.P. Balk and D.J. Einstein); and Department of Defense Awards W81XWH-17–1-0350 (to D.J. Einstein) and W81XWH-16–1-0431 (to A.G. Sowalsky and S.P. Balk).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.