Expression of PD-L1 in Hormone-naïve and Treated Prostate Cancer Patients Receiving Neoadjuvant Abiraterone Acetate plus Prednisone and Leuprolide

It is estimated that over 26,120 patients died of prostate cancer in 2016 (1). Nearly a third of patients with Gleason score 3+4 = 7 and higher (Grade Group 2–5 using the 2014 grading system) will develop disease recurrence after radical prostatectomy (RP; ref. 2). While neoadjuvant androgen deprivation therapy (ADT) and/or cytotoxic therapy prior to RP can decrease tumor volume, these treatments rarely result in complete responses and have not yet been shown to reduce disease relapse (3). In our recent phase II clinical trial, 58 patients with localized, intermediate- to high-risk prostate cancer received neoadjuvant treatment combining abiraterone acetate and prednisone with leuprolide (Neo-AAPL) prior to RP (4). Following analysis of RP specimens, four patients were found to have achieved a pathologic complete response while an additional subset of patients showed minimal residual disease. However, despite these successes, a significant proportion of patients exhibited extensive residual tumor. Overall, these observations highlight an urgent need to develop more effective neoadjuvant/adjuvant therapies for men with intermediate- to high-risk prostate cancer.

In recent years, programmed cell death ligand-1 (PD-L1)/programmed cell death-1 (PD-1) blockade therapies have achieved significant and durable responses in a number of advanced malignancies (5–7). However, these same studies showed disappointing results in castration-resistant prostate cancer (CRPC). PD-L1 expression correlates positively with response to PD-L1/PD-1 blockade in non–small cell lung carcinoma (NSCLC) and melanoma (8). The failure of PD-L1/PD-1 blockade in CRPC has been hypothesized to be due to the poor immunogenicity of prostate cancer, as PD-L1 expression is thought to be in responsive to antitumor immunity. Early studies appeared to support this hypothesis, as both primary prostate cancer and CRPC specimens were shown to express little to no PD-L1 (7, 9, 10). However, recently there have been a number of reports describing significant PD-L1 positivity in both primary prostate cancer (up to 92%) and CRPC (up to 19%; refs. 11–13). One possible explanation for these conflicting reports is the use of different PD-L1 antibody clones for IHC. Those IHC studies reporting rare staining for PD-L1 in prostate cancer were performed with the earlier clone 5H1, while those with much higher positivity were performed with newer clones [EPR1161(2), 015, and E1L3N]. Importantly, the type of tissue studied would affect the results significantly. Ilie and colleagues (14) recently reported significant discordance in PD-L1 expression between preoperative biopsy and the matched resection specimens in 160 NSCLC (the overall discordance rate = 48% and κ value = 0.218). Those studies utilizing tissue microarrays (TMA) rather than whole sections could underestimate PD-L1 tumor positivity due to intratumoral heterogeneity, as levels as low as 1% are associated with responses. Finally, it is important to note that while many studies focus on PD-L1 expression by tumor cells, several studies have suggested that PD-L1 expression on tumor-infiltrating immune cells is more predictive of response to therapy (15–17), and this remains an understudied phenotype in prostate cancer. Overall, the frequency of PD-L1 expression in primary prostate cancer remains unclear.

Recently, a small clinical trial reported a significant reduction of serum PSA levels in 30% of patients with enzalutamide-resistant prostate cancer treated with anti-PD-1 pembrolizumab (18). In agreement with these results, increased PD-L1 expression has been reported in the circulating dendritic cells of patients with enzalutamide-resistant prostate cancer and in preclinical models of enzalutamide-resistant prostate cancer (19). To our knowledge, there are no studies that have addressed the effects of abiraterone acetate plus prednisone on prostate cancer PD-L1 expression. In this study, we evaluated tumor PD-L1 expression in prostate cancer after Neo-AAPL treatment (n = 44; ref. 4) and compared that with PD-L1 expression in matched untreated controls. Surprisingly, Neo-AAPL–treated tumors exhibited a trend toward reduction of PD-L1 expression. Furthermore, treated tumors were associated with significantly less CD8⁺ tumor-infiltrating lymphocytes (TIL). In addition to these observations in Neo-AAPL–treated tumors, we conducted an expanded analysis using a cohort of 130 hormone-naïve primary prostate cancer. We detected PD-L1 expression in 14% primary prostate cancer and identified clinicopathologic parameters including African American ethnicity that predict PD-L1 positivity.

In accordance with the U.S. Common Rule and after institutional review board (IRB) approval, tissue was collected from the following resources: (i) the Neo-AAPL cohort including 44 patients with intermediate- to high-risk prostate cancer who underwent RP after Neo-AAPL treatment in a multicenter phase II clinical trial COU-AA-201 (NCT00924469; ref. 4); and (ii) the hormone-naïve cohort including 130 patients who underwent RPs between 2010 and 2016 at Beth Israel Deaconess Medical Center (BIDMC, Boston, MA) and deidentified in accordance with BIDMC IRB protocol #2010-P-000254. Among these 130 hormone-naïve patients, 44 patients were identified as matched controls for the Neo-AAPL cases. Tumor was graded using both new Grade Group (GG) system and Gleason Score (GS) as recommended by the 2014 International Urological Pathology (ISUP) consensus (2). Considering a consecutive series of RPs would result in over-representation of GS7 cases, we selected approximately 30 cases from each Grade Group, GG1 (GS3+3 = 6), GG2 (GS3+4 = 7), GG3 (GS4+3 = 7), and GG4–5 (GS8–10), to comprise the 130 hormone-naïve cohort. Clinicopathologic characteristics of the two cohorts are listed in Tables 1 and 2. Prostate tissue was fixed in formalin, processed, and embedded in paraffin using standard methods. For each case, one tissue block with the largest dimensions of the dominant tumor was selected for IHC.

Three commercially available monoclonal anti-PD-L1 antibody clones were tested and optimized, including E1L3N (rabbit, Cell Signaling Technology; 1:200), SP-142 (rabbit, Spring Biosciences; 1:100), and 405.9A11 (mouse, Cell Signaling Technology; 1:100). Optimization was performed using human placenta tissue and a multitumor “sausage” test block containing benign prostate, prostate cancer, colon medullary carcinoma, and additional tissue and tumor types. Five-micron sections were baked at 60°C for an hour, followed by deparaffinization, rehydration, and epitope retrieval using the Dako PT Link platform. IHC staining was carried out on the Dako Link 48 autostainer, with antibody incubation for 60 minutes, amplification using Envision FLEX rabbit or mouse linkers, and visualization using the Envision Flex High-sensitivity visualization system (Dako). Assay quality was evaluated on a subset of Neo-AAPL cohort and matched control cases using an FDA-approved Ventana SP263 PD-L1 assay [for durvalumab (AstraZeneca)] at Boston University Medical Center (Boston, MA) as the reference standard. For all cases, sections were also stained using anti-AR (N20, Santa Cruz Biotechnology; 1:1,000), anti-ERG (EPR3864, Epitomics; 1:100), and anti-PTEN (D4.3, Cell Signaling Technology; 1:50) antibodies. For 44 trial cases and 44 controls, sections were stained anti-CD8 (C8/144B, Dako; 1:50) antibody. For a subset of cases with 40%–50% tumor cellularity (N = 14), sections were additionally stained with anti-CD3 (F7.2.38, Dako; 1:200) and anti-PD-1 (EPR4877, Abcam; 1:500) antibodies. For cases with ≥ 1% PD-L1 positivity (N = 21), sections were stained with anti-MSH2 (Biocare Medical, 1:25), anti-MSH6 (BD Biosciences, 1:100), anti-MLH1 (Biocare Medical, 1:20); and anti-PMS2 (BD Biosciences, 1:50). Dual immunostains for PD-L1 (E1L3N, Cell Signaling Technology; 1:100) and CK18 (DC10, Dako; 1:4) were stained using Dako EnVision G|2 Doublestain System following manufacturer's protocol.

PD-L1 positivity was defined by moderate to strong membranous staining. Cytoplasmic staining was not considered. PD-L1 percentage scoring was performed semiquantitatively as follows: 0 (negative or < 1%), 1 (1%–4%), 2 (5%–24%), 3 (25%–49%), and 4 (≥ 50%). AR immunointensity score was scored as negative (0), weak (1), moderate (2), and strong (3), based on the most predominant intensity pattern. AR percentage score was scored on the basis of percentage of tumor cells demonstrating the most predominant intensity pattern or stronger as: 0 (negative), 1 (1%–9 %), 2 (10%–49%), and 3 (≥ 50%). AR score (0–9) was based on the immunointensity score multiplied by the percentage score. ERG was scored as negative (0) and positive (1). PTEN was scored as 0 (no loss or weak) and 1 (complete loss in ≥ 5% tumor cells). MSH2, MSH6, MLH1, and PMS2 were scored as 0 (no loss) and 1 (complete loss in all tumor cells). All immunostains were evaluated by an experienced pathologist (H. Ye) and a trained MD investigator (C. Calagua). In addition, SP263 immunostains of treated tumors and matched untreated controls were evaluated by a second pathologist (Y. Sun).

As many treated tumors had significantly reduced tumor cellularity, density of tumor-infiltrating CD8⁺ cells was normalized by the number of tumor cells in 44 treated tumors and 44 matched untreated tumors using the following method. First, microscopic fields containing the highest density of tumor-infiltrating CD8⁺ cells in each case were identified on CD8 immunostains, consisting of 1 high power field (HPF) in untreated tumors and 1 or more HPF in treated tumors to examine comparable numbers of tumor cells; second, tumor-infiltrating lymphocytes (TIL) were defined as lymphocytes 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); third, the total numbers of tumor cells and tumor-infiltrating CD8⁺ lymphocytes in the identified fields were visually counted independently by two pathologists (H. Ye and R. Lis; median numbers of tumor cells examined: 516 in untreated tumors and 360 in treated tumors); Finally, density of tumor-infiltrating CD8⁺ cells was normalized as the number of CD8⁺ cells per 100 tumor cells.

Densities of CD3⁺ and PD-1⁺ TILs were measured in 7 PD-L1–positive tumors (all 3 treated tumors that were PD-L1–positive, and all 4 untreated control tumors with PD-L1 staining in ≥ 5% of tumor cells) and 7 PD-L1-negative tumors with matched tumor cellularity (3 treated and 4 untreated). Densities of CD3⁺and PD-1⁺ TILs were counted using the method described above, on corresponding immunostains performed on consecutive sections. CD3⁺ and PD-1⁺ TILs were counted in the same microscopic fields where CD8⁺ TILs were counted.

Stata 12.1 (StataCorp) was used for statistical analysis. Characteristics between groups were compared using Pearson χ² and Fisher exact tests for categorical variables, and Wilcoxon rank-sum test for nonparametric continuous and ordinal variables. Logistic regression was used to analyze candidate predictors for binary outcomes. P < 0.05 was considered statistically significant.

Using human placenta as positive control tissue and a multi-tumor “sausage” as test block, staining protocols were optimized for three commercially available anti-PD-L1 antibodies, including E1L3N, SP142, and 405.9A11. Antibody performance was examined using a FDA-approved Ventana SP263 assay as the reference standard. Antibody 405.9A11 showed the lowest sensitivity and was determined suboptimal for this study (Supplementary Fig. S1). Antibodies E1L3N and SP142 demonstrated predominantly membranous staining pattern in placenta and colon medullary carcinoma, with comparable specificity and sensitivity compared to Ventana SP263 (Supplementary Figs. S1 and S2). Staining performance of E1L3N and SP142 in study cases was first examined in a pilot panel. E1L3N and SP142 demonstrated concordant staining results (Supplementary Figs. S3 and S4), while both exhibited strengths and weaknesses. E1L3N consistently stained nerve fibers, while SP142 (and Ventana SP263) did not (Supplementary Figs. S3A–S3C, blue arrow). In contrast, SP142 showed some nonspecific staining in luminal secretions and in the cytoplasm of epithelium from both benign and malignant glands, which was not present in those sections stained with E1L3N. We proceeded to stain all study cases with both antibodies.

Overall, E1L3N and SP142 immunostains demonstrated an excellent concordance, with a correlation coefficient r = 0.90 (N = 174). Given that E1L3N only showed nonspecific staining in nerve fibers (a structure that is morphologically easy to identify and exclude), while SP142 showed sporadic nonspecific cytoplasmic staining that could lead to a false positive classification, we chose to perform statistical analysis using the results from the E1L3N staining.

We focused on analysis of PD-L1 expression on tumor cells. PD-L1 percentage scoring was performed semi-quantitatively as follows: 0 (negative or < 1%), 1 (1%–4%), 2 (5%–24%), 3 (25%–49%), and 4 (≥ 50%). PD-L1 immunostains of three cases that were positive in 5%, 50%, and 5–24% of tumor cells are represented in Figure 1. In all tumors scored 1 (1%–4%), PD-L1 IHC unequivocally stained tumor cells, as shown in PD-L1 IHC in Supplementary Fig. S5 and PD-L1/CK18 dual IHC in Supplementary Fig. S6A. In lung cancer, a range of PD-L1 positivity (1% or 50% of cells) on IHC has been used as a cutoff to classify a tumor as PD-L1-positive, depending on antibody, platform, drug, and first-line versus second-line treatment (20, 21). In the setting of prostate cancer, the % cutoff for PD-L1 positivity in relation to PD-L1/PD-1 blockade treatment outcomes has yet to be established. We presented our results using a 1% cutoff when a dichotomization was needed.

We examined PD-L1 immunostains in 44 treated tumors from the Neo-AAPL trial (4) and compared them to 44 untreated control tumors matched for pre-RP characteristics. As shown in Table 1, the two groups demonstrated similar pre-RP characteristics, including ethnicity, serum PSA level, biopsy tumor grade, number of positive cores, and maximum cancer percentage on positive cores. On RP specimens, PTEN and ERG status were comparable between the two groups. Neo-AAPL RP specimens showed significantly lower prostatic weight, lower pT stage, less margin involvement, and a lower AR expression level, consistent with the androgen depleting effects of Neo-AAPL treatment.

To our surprise, Neo-AAPL tumors showed a trend toward reduced PD-L1 expression compared with untreated tumors, with 7% of Neo-AAPL tumors compared with 21% of untreated tumors showing PD-L1 staining in ≥ 1% of tumor cells (P = 0.062) and 5% of Neo-AAPL tumors compared with 9% of untreated tumors showing PD-L1 staining in ≥ 5% of tumor cells (Table 1).

Next, we evaluated PD-L1 expression in an expanded cohort of hormone-naïve prostate cancer (n = 130). This cohort was composed of roughly 30 cases from each Grade Group, GG1 (GS3+3 = 6), GG2 (GS3+4 = 7), GG3 (GS4+3 = 7), and GG4–5 (GS8–10). Eighteen (14%) of hormone-naïve tumors were PD-L1–positive (Table 2). 9%, 4%, and 1% showed PD-L1 positivity in ≥ 5%, ≥ 25%, and ≥ 50% of tumor cells, respectively.

Examination of PD-L1 immunostains revealed three different distribution patterns of PD-L1⁺ tumor cells (Supplementary Table S1). The most common pattern was “scattered” pattern, in which positive tumor cells were distributed in scattered small clusters (Supplementary Fig. S5A–S5C). This pattern was seen in 10 cases (56%). The second most common pattern was “nodular” pattern, in which distinct regions or nodules of tumor were extensively positive for PD-L1 (PD-L1^high component), while the remaining regions were negative or focally positive for PD-L1 (PD-L1^low component). The positive components are potentially microdissectable. This pattern was observed in 7 cases (39%), with representative pictures shown in Fig. 1B and Supplementary Fig. S5D–S5F. In those 7 cases, the PD-L1^high and PD-L1^low components showed consistently concordant ERG status but variable PTEN staining patterns (an example shown in Supplementary Fig. S6B–S6D). The least common pattern was “interface” pattern, seen in a single case (5%). In this case, PD-L1⁺ tumor cells were enriched at the tumor–stroma interface bordering the leading edge of a tumor nodule (Fig. 1C), where the richest lymphocytic infiltrates were identified. We also examined the PD-L1 expression pattern of 3 Neo-AAPL–treated tumors that were PD-L1–positive. All three cases showed “scattered” pattern of distribution of PD-L1⁺ tumor cells (Fig. 1A; Supplementary Fig. S4).

PD-L1 expression level was increased in high-grade tumors. Using 1%, 5%, 25%, and 50% cutoff for PD-L1 positivity, 27%, 18%, 9%, and 3% of GG4-5 tumors (n = 34) were PD-L1–positive, in comparison with 9%, 6%, 2%, and 0% of GG1-3 tumors (n = 96; P = 0.013, 0.049, 0.079, and 0.092, respectively). Next, we dichotomized PD-L1 staining results and sought to examine the clinicopathologic characteristics of PD-L1–positive hormone-naïve prostate cancer. As shown in Table 2, patients with PD-L1–positive tumors had a higher serum PSA level, a higher tumor grade and stage, and a higher rate of margin positivity on RP specimens. In addition, PD-L1–positive prostate cancer was found in 26% of African American (AA) patients and 12% of non-AA patients (13% in Whites). Furthermore, PD-L1–positive tumors demonstrated a trend toward a lower rate of ERG immunopositivity and higher AR expression. PD-L1 positivity was identified in 19% of ERG-negative tumors and in 8% of ERG-positive tumors. Interestingly, five tumors that demonstrated PD-L1 staining in ≥ 25% tumor cells were all ERG-negative (Supplementary Table S2).

To extend this analysis, we examined which clinicopathologic parameters were independently correlated with tumor PD-L1 expression in a multivariate model (Table 3). Among the parameters assessed, race (AA), high serum PSA level, and low prostate weight remained as statistically significant independent predictors for PD-L1 positivity.

The interaction of the tumor with the immune microenvironment is the basis for tumor PD-L1 expression and immune evasion. We therefore examined the immune environment of Neo-AAPL–treated tumors and their matched controls. The distribution of lymphocytes was first examined on hematoxylin and eosin stains. Scattered lymphoid cells and lymphoid aggregates were present in association with both malignant and benign glands. The amount of TILs varied to a great extent among different patient specimens. By visual inspection, there was no obvious enrichment of TILs in tumors from the Neo-AAPL cohort compared with untreated tumors. On IHC, the distribution of CD8⁺ and PD-1⁺ TILs appeared to be enriched in PD-L1⁺ areas (Fig. 2; Supplementary Fig. S7).

Next, we performed quantitative analysis of CD8⁺ TILs. The commonly used metric, the number of CD8⁺ TILs per mm² of tumor, was not suitable for this study as most treated tumors had significantly fewer tumor cells than untreated tumors in a 1 mm² area due to treatment effect. Instead, we measured “normalized maximum density of CD8⁺ TILs,” which was designated as the number of CD8⁺ TILs per 100 tumor cells in regions with the maximum CD8⁺ TILs. As shown in Fig. 3A, our results demonstrated that PD-L1–positive tumors were associated with significantly more CD8⁺ TILs in comparison with PD-L1–negative tumors (P < 0.001), consistent with PD-L1 expression being indicative of an antitumor immune response. Neo-AAPL–treated tumors had significantly less CD8⁺ TILs compared with matched controls (P = 0.034).

To examine the correlation of different phenotypes of TILs with PD-L1 expression, we examined the density of CD3⁺ and PD-1⁺ TILs in 3 treated and 4 untreated control tumors that were PD-L1–positive. Their CD3⁺ and PD-1⁺ TIL densities were compared with that of PD-L1–negative cases with matched tumor cellularity, including 3 treated and 4 untreated cases. The densities of CD3⁺ and CD8⁺ TILs each showed a strong positive correlation with the density of PD-1⁺ TILs, suggesting these cells may exhibit both markers (Fig. 3B). The strong correlation among CD3⁺, CD8⁺, and PD-1⁺ TILs was well described in different tumor types including prostate cancer (22–24), therefore, was likely to be representative for the entire cohort. Importantly, PD-L1–positive tumors demonstrated significantly increased densities of CD3⁺, CD8⁺, and PD-1⁺ TILs, when compared with PD-L1–negative tumors (Supplementary Fig. S8). Taken together, our analysis demonstrated that (i) tumor PD-L1 expression was associated with increased CD3⁺, CD8⁺, and PD-1⁺ TILs; (ii) Neo-AAPL treatment was associated with reduced CD8⁺ TILs.

Inflamed “hot” tumor phenotype with increased numbers of TILs can be associated with increased tumor mutation burdens and DNA repair gene dysfunction. We hypothesized that PD-L1–positive prostate cancer might be due in part to microsatellite instability-high (MSI-H) subtype, similar to what is observed in lymphocyte-rich medullary colorectal cancer. To address this question, we used IHC to examine expression of DNA mismatch repair proteins, including MSH2, MSH6, MLH1, and PMS2 in PD-L1–positive tumors. Of 21 PD-L1–positive tumors (18 untreated tumors and 3 treated), one (5%) showed complete loss of MSH2 and concordant loss of MSH6 across the entire tumor, suggesting a truncal genetic alteration (Fig. 3C). The patient, UN127, was a 68-year-old Caucasian man with a 156-gram prostate that harbored a GG5 (Gleason score 5+4 = 9) pT3bN1 prostate cancer without prior neoadjuvant treatment. Interestingly, this tumor displayed an “interface pattern” of PD-L1 immunopositivity and TILs, which is commonly seen in MSI-H colorectal carcinoma. The remaining 20 PD-L1–positive tumors showed no evidence of MMR protein loss.

To our knowledge, this is the first study to assess the effects of combined neoadjuvant abiraterone acetate plus prednisone + leuprolide treatment (Neo-AAPL) on PD-L1 expression in clinical samples of prostate cancer. Initially our hypothesis was that residual tumors that survived Neo-AAPL therapy would express higher levels of PD-L1 compared with untreated controls. This hypothesis was based on several lines of evidence. Androgen ablation is thought to upregulate adaptive immunity in prostate through increasing naïve T-cell expansion with continued thymic output and reducing T-cell activation threshold (25–29). In addition, androgen deprivation through castration has been shown to increase the immune response in mice and patients (27–29). Finally, enzalutamide resistance has been associated with the upregulation of PD-L1 expression, both in the circulating immune cells of patients as well as in tumor cells of preclinical models (19). To our surprise, instead of exhibiting increased PD-L1 expression, Neo-AAPL–treated tumors actually showed a trend toward decreased PD-L1 expression compared with untreated controls. In agreement with this finding, our assessment of TILs, which positively correlated with PD-L1 expression levels in several tumor types, revealed significantly less CD8⁺ TILs in Neo-AAPL–treated tumors. These results conflict with our previous perception that ADT upregulates adaptive immunity in prostate cancer. It is possible that the trend toward decreased PD-L1 expression we observed in Neo-AAPL–treated tumors reflects a true decrease in adaptive immunity following 6 months of Neo-AAPL treatment, and several lines of evidence support this notion. Animal model experiments have demonstrated a dual effect of ADT on immune response in prostate cancer, with castration-induced anti-tumor CD8⁺ T-cell response increasing in the acute phase of ADT, while the effect was countered by a subsequent increase in regulatory T cells (Tregs; ref. 30). In genetically modified mice, treatment with the AR antagonists flutamide and enzalutamide leads to immunosuppression, while orchiectomy, leuprolide, and abiraterone showed no inhibitory effects on T-cell response, suggesting a modality-dependent effect of ADT on immunity (31). Another possible explanation may lie in the design of the Neo-AAPL trial, in which patients received low-doses of the immunosuppressive prednisone (5 mg/day) as glucocorticoid replacement therapy. Alternatively, immunogenic prostate cancers might be the excellent responders to Neo-AAPL treatment with no or minimal residual tumors detected in RP specimens, while the tumors we evaluated on RP specimens were poor responders that might be non-immunogenic. Importantly, prior studies observed PD-L1 upregulation in enzalutamide-resistant tumors (19). It is not clear whether the residual tumors in our study were AAPL-resistant or still hormonal sensitive. Six-month Neo-AAPL treatment is unlikely to transform most tumors to AAPL-resistant. Long-term effect of AAPL treatment on tumor PD-L1 expression remains unknown. In summary, the immune environment in a patient receiving both abiraterone acetate + leuprolide (presumably immune stimulatory, but perhaps inhibitory) and prednisone (immune inhibitory) is likely to be a complicated one, and may be time- and dose- dependent. The trend and mechanisms of decreased PD-L1 expression following Neo-AAPL treatment in prostate cancer will need to be addressed by future studies.

One weakness of our study is lack of a longitudinal comparison before and after Neo-AAPL treatment in the same patients. The decision to use matched untreated cases as controls rather than pretreatment biopsies from the same patient was due to several factors. First, baseline core biopsy tissue was not available in more than half of the Neo-AAPL trial cases, many of which were diagnosed at outside hospitals. Second, PD-L1–positive cases showed focal or patchy staining for PD-L1, suggesting that PD-L1 expression in biopsy cores would not accurately reflect PD-L1 status in the entire tumor. Therefore, we chose to stain regular tissue sections of the largest tumor blocks from both Neo-AAPL–treated cases and matched untreated controls to give a more accurate representation of PD-L1 expression.

With regard to our analysis on 130 hormone-naïve prostate cancer, our results differ from the earlier reports of low PD-L1 expression levels in prostate cancer and likely reflect improvements in antibody, staining platform, and study material. For simplicity of data analysis, we defined PD-L1 positivity as ≥ 1% of tumor cells stained positive for PD-L1. This threshold was determined empirically as a cutoff of 1% and has previously been used for both E1L3N and SP142 antibodies in NSCLC (20). It should be kept in mind that this threshold may not accurately predict treatment response in prostate cancer, as supporting data from large anti-PD-L1/PD-1 clinical trials in the treatment of prostate cancer are lacking. Nonetheless, our results add to the growing body of evidence that PD-L1 is indeed expressed in a significant fraction of high-grade primary prostate cancer.

In our study, PD-L1 expression level was increased in high grade tumors. Using 1%, 5%, 25%, and 50% cutoff for PD-L1 positivity, 27%, 18%, 9% and 3% of GG4-5 tumors (n = 34) were PD-L1–positive, in comparison with 9%, 6%, 2%, and 0% of GG1-3 tumors (n = 96; P = 0.013, 0.049, 0.079, and 0.092, respectively). Results from univariate and multivariate analysis suggest that increased PD-L1 expression is associated with a more aggressive prostate cancer phenotype, in agreement with a recently published study by Gevensleben and colleagues (12), in which they reported similar statistically significant associations between PD-L1–high expressing prostate cancer and higher serum prostate-specific antigen (PSA) levels, higher tumor grade, and margin positivity. However, the unusually high rate of PD-L1 (61.7%) expression in the Gevensleben study raises a question of staining specificity, as their PD-L1 IHC figures appeared to be overstained (diffuse staining in tonsil and positive prostate cancer tissue, in contrast to focal to patchy staining in our study and others). At the time of preparation of this article, the PD-L1 antibody EPR1161(2) used in the Gevensleben study is no longer available at Abcam due to issues on compatibility with human tissue. In summary, our results indicate that a subset of patients with high-grade prostate cancer may potentially benefit from PD-L1/PD-1 blockade monotherapy.

There are two mechanisms mediating PD-L1 expression on tumor cells: (i) “adaptive immune resistance” through dynamic induction of PD-L1 expression in tumor cells in response to cytokines (i.e., IFNγ) secreted by activated T cells in tumors of “T-cell inflamed phenotype” and (ii) “intrinsic immune resistance” through constitutive expression of PD-L1 by tumor cells due to genetic alterations or constitutive activation of particular signaling pathways. CD8⁺ immune cells as well as presumptive PD-1⁺ immune cells were significantly increased in PD-L1–positive prostate cancer, supporting an adaptive immune resistance mechanism. In our cohort, 1 of 21 PD-L1–positive tumors (5%) showed loss of expression of DNA mismatch repair proteins MSH2 and MSH6. Therefore, genomic instability may lead to adaptive immune resistance in a subset of patients with prostate cancer. The “interface pattern” in the MSH2-deficient tumor closely resembles MSI-H colorectal carcinoma, likely reflecting their similar molecular mechanisms. In the 333-case TCGA cohort of primary prostate cancer, deletion or mutations in MLH1, MSH2, and MSH6 genes in 0.3%, 0.3%, and 1.5% of cases, respectively (32). Overall DNA repair gene mutations were identified in 19% in GS8-10 tumors in The Cancer Genome Atlas (TCGA) cohort. Other genes involved include BRCA1, BRCA2, CDK12, ATM, FANCD2, and RAD51C. We expect that genetic analysis of PD-L1-positive tumors, particularly those with the “nodular” and “interface” patterns, will reveal genomic instability leading to adaptive immune resistance.

Possible intrinsic immune resistance mechanisms include constitutive PD-L1 overexpression due to genetic alterations including EGFR mutations, ALK rearrangements, KRAS mutations, and loss of tumor suppressor PTEN in other types of malignancy (33, 34). In this study, PD-L1 status is independent of PTEN status making an intrinsic immune resistance mediated through PTEN–PI3K signaling cascade less likely. Our study revealed an association between increased PD-L1 positivity and AA ethnicity, and a trend toward tumors with higher AR expression and negative for ERG fusion protein. To our knowledge, this is the first study revealing the link between PD-L1-positive prostate cancer and AA ethnicity. It is possible that PD-L1 expression serves as an indicator of prostate cancer aggressiveness, as AA patients tend to have higher grade prostate cancer and a more aggressive disease course (35). Similarly, PD-L1 positivity may be increased in AA men because they tend to have ERG-negative prostate cancer (36). Nevertheless, race remained an independent predictor for PD-L1 positivity in multivariate analysis when all other risk factors, including PSA, tumor grade, stage, margin, and ERG status were held constant (Table 3). This strong and independent association suggests that the immunogenicity of prostate cancer in AA men may be intrinsically distinct from that in non-AA men.

In summary, we show that PD-L1 positivity is found in approximately 14% of primary prostate cancer in our cohort and PD-L1 positivity is associated with clinicopathologic characteristics found in aggressive prostate cancer. With multivariate analysis, AA race, high serum PSA levels, and small prostate independently predict tumor PD-L1 positivity. To our knowledge, these results show for the first time indicate that Neo-AAPL treatment is associated with reduced CD8⁺ TILs and there is no evidence of upregulation of PD-L1 in treated tumors. These results suggest that combining the current Neo-AAPL treatment with PD-L1/PD-1 blockade may not to be synergistic in treating patients with high-risk primary prostate cancer.

M.-E. Taplin is an employee of, is a consultant/advisory board member for, and reports receiving commercial research grants from Janssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Calagua, J. Russo, G.J. Bubley, M.-E. Taplin, H. Ye

Development of methodology: C. Calagua, J. Russo, K. Mahoney, H. Ye

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Sun, R. Schaefer, Z. Zhang, G.J. Bubley, M. Loda, M.-E. Taplin, H. Ye

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Russo, Y. Sun, R. Lis, K. Mahoney, M.-E. Taplin, S.P. Balk, H. Ye

Writing, review, and/or revision of the manuscript: C. Calagua, J. Russo, R. Lis, K. Mahoney, G.J. Bubley, M. Loda, M.-E. Taplin, S.P. Balk, H. Ye

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Calagua, J. Russo, R. Lis, Z. Zhang, H. Ye

Study supervision: J. Russo, M.-E. Taplin, H. Ye

We thank Dr. Paul VanderLaan for critical review of the manuscript, Dr. Douglas Lin for providing the multi-tumor blocks for antibody optimization, and Ms. Wenyang Mao for her guidance on statistical analysis. The original COU-AA-201 study was supported by Janssen Research & Development.

Career Development Award from the Prostate and Breast Cancer Research Program at the Beth Israel Deaconess Medical Center (H. Ye), Young Investigator Award from the Prostate Cancer Foundation (H. Ye), Developmental Research Award (P50CA090381-15, H. Ye), 2016 Challenge Award from the Prostate Cancer Foundation (to S.P. Balk/M.-E. Taplin), NIH R01 grant (to M. Loda; RO1CA131945), Department of Defense (PC130716; to M. Loda), the Prostate Cancer Foundation (to M. Loda), and DF/HCC-Prostate Cancer SPORE (P50CA090381; to S.P. Balk/M. Loda).

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.