We conducted a retrospective study of patients with PCa treated with radical prostatectomy at Sun Yat-sen University Cancer Center (126 cases, named training cohort) from January 2005 to December 2014, or at Sun Yat-sen Memorial Hospital (113 cases, named validation cohort) from January 2009 to December 2013. Oral consent was obtained from each patient in both cohorts. Formalin-fixed, paraffin-embedded tissue blocks were available from all cases and H&E-stained tumor slides were used to confirm the Gleason score, tumor stage and lymph node status by two senior pathologists according to the American Joint Committee on Cancer 2009 tumour, node, metastases classification for PCa. Imaging examination was used to determine distal metastasis. Overall survival (OS) was calculated from prostatectomy to death or the last follow-up, which was December 31, 2019. Cancer-specific survival (CSS) was defined as the time from prostatectomy to death from PCa or the last follow-up.

In line with our previous study,30 briefly, formalin-fixed PCa tissue sections were paraffinized in xylene and hydrated in a diluted alcohol series. Then, antigen retrieval was accomplished with sodium citrate buffer (10 mM, pH 6.0) using a microwave for 10 min. After that, the sections were immersed in H₂O₂ (0.3%) for 15 min to reduce activity of endogenous peroxidase. Following blocking with goat serum, the sections were incubated with corresponding primary antibodies (anti-PD-L1: Cell Signaling Technology (CST) #13684; anti-B7-H3: CST, #14058; anti-HHLA2: clone 566.129; anti-CD8: CST, #85336; anti-Foxp3: CST, #12653) at 4℃ overnight. Then, after washing for three times using phosphate buffered saline (PBS), the sections were subsequently incubated with secondary antibody for 1 hour. Finally, the sections were visualized using a DAKO EnVision Detection System (Dako) and counterstained with hematoxylin.

All sections were assessed by two independent pathologists blinded to the clinical profiles of the patients, and discrepancies were resolved after consensus. PD-L1 positivity in this study was defined as PD-L1 staining in ≥1% of tumor cells.31 Both B7-H3 and HHLA2 immunostaining in tumor cells were semi-quantified using the H-score on account of the percentage of positive staining and the staining intensity. In detail, the staining intensity was divided into four levels as follows: absent (0), weak (1), moderate (2) and strong (3) (online supplemental figure S1A,B). The H-score of each case was determined by multiplying the staining intensity by the corresponding percentage of positive cells. The H-score range was 0 to 300. An H-score >0 was defined as positive expression and the cut-off values for differentiating high and low B7-H3 and HHLA2 expression were 120 and 80 determined by X-tile program, respectively.

The infiltration extent of CD8⁺ and Foxp3⁺ TILs was evaluated according to the mean of counts in five independent high-power fields (400×) within mesenchyme, which represented the densest immune cell infiltrates. The mean value was considered the cut-off point for marker stratifications in this study.25

The distribution of clinicopathological parameters, including the densities of different sets of immune cells over B7-H3 or HHLA2 expression categories, was evaluated using χ² or Fisher’s exact test. The OS and CSS curves were depicted with the Kaplan-Meier method, and the intergroup differences were examined using a log-rank test. Cox proportional hazard regression analysis was performed for univariate and multivariate analyses. All statistical analyses were performed using SPSS V.24.0 (Chicago, Illinois, USA) or GraphPad Prism V.8 (La Jolla, California, USA) software.

The basic clinicopathological characteristics of 239 patients with PCa from two independent cohorts are summed up in online supplemental table 1. The mean age at operation was 66 years (range, 42–86) in the training cohort and 69 years (range, 52–89) in the validation cohort, respectively. At the end of follow-up, 36 (28.6%) patients died, including 25 (19.8%) cases died from tumor progression in the training cohort. The median follow-up time was 88.5 months (range, 4–171). The 5-year and 10- year OS rates were 81.4% and 64.7%, respectively, and the 5-year and 10-year CSS rates were 85.0% and 73.2%, respectively. In the validation cohort, 22 (19.5%) patients died, including 15 (13.3%) cases died from tumor progression, and the median follow-up time was 78.0 months (range, 1–126). The 5-year and 10-year OS rates were 84.2% and 78.5%, respectively, and the 5-year and 10-year CSS rates were 87.8% and 85.4%, respectively.

To evaluate the expression spectrum of PD-L1, B7-H3 and HHLA2 in PCa tissues from the two cohorts, immunohistochemistry (IHC) was performed, and typical IHC photomicrographs of PD-L1, B7-H3 and HHLA2 expression are presented in figure 1A. PD-L1 was positively stained in 10 (8.2%) cases in the training cohort and 6 (5.3%) cases in the validation cohort (figure 1B). However, B7-H3 was widely expressed in 105 (83.3%) PCa cases in the training cohort and 98 (87.6%) cases in the validation cohort (figure 1C). In terms of HHLA2, positive staining of HHLA2 was detectable in 87 (69.0%) and 77 (68.1%) PCa cases in the two cohorts, respectively (figure 1D). These data demonstrate that B7-H3 and HHLA2 are more frequently expressed in PCa tissues, compared with PD-L1 expression.

To evaluate the clinical relevance of B7-H3 and HHLA2, the patients with PCa were segregated into two groups according to the optimal cut-off value for the H-score of B7-H3 (H-score=120) or HHLA2 (H-score=80). As detailed in table 1, high B7-H3 expression (H-score ≥120) was significantly associated with advanced Gleason score (p=0.032 and p=0.006, respectively) and tumor stage (p=0.006 and p=0.001, respectively) in both cohorts. Furthermore, high expression of B7-H3 was also significantly correlated with lymph node metastasis in the training cohort (p=0.001, table 1). Similarly, in both cohorts, increased HHLA2 expression (H-score ≥80) displayed a significant correlation with advanced Gleason score (p=0.001 and p=0.028, respectively), tumor stage (p=0.049 and p=0.031, respectively) and lymph node metastasis (p=0.002 and p=0.046, respectively) (table 1). Of note, significant associations were detected between B7-H3 and HHLA2 expression in both cohorts (p=0.002 and p=0.020, respectively, table 1). Taken together, these data show that high expression of both B7-H3 and HHLA2 might be associated with worse clinicopathological features.

Microphotographs of CD8⁺ and Foxp3⁺ TILs are shown in figure 2A. The density of CD8⁺ TILs was much higher than the density of Foxp3⁺ TILs in both cohorts (p<0.001 in both cohorts, figure 2B). In both cohorts, significantly higher densities of CD8⁺ TILs were detected in tumors with low B7-H3 or HHLA2 expression than in tumors with high B7-H3 or HHLA2 expression (training cohort: B7-H3, p<0.001, HHLA2, p<0.001; validation cohort: B7-H3, p<0.01, HHLA2, p<0.001; figure 2C–D). Conversely, no significant differences in Foxp3⁺ TILs were observed between tumors with high B7-H3 expression and tumors with low B7-H3 in either cohort (figure 2E–F). Similarly, no significant correlation was observed between HHLA2 expression and Foxp3⁺ TILs in the training cohort (figure 2E). However, a high density of Foxp3⁺ TILs was significantly associated with high HHLA2 expression in the validation cohort (p<0.01, figure 2F). These results indicate that both B7-H3 and HHLA2 might account for the immunosuppressive environment within PCa tissues.

Given the prevalence of B7-H3 and HHLA2 overexpression, a univariate analysis of OS and CSS was generated to evaluate the prognostic significance of B7-H3 and HHLA2 expression. Interestingly, high B7-H3 expression conferred a greater than twofold increase in the risk of mortality, especially CSS, in both cohorts (OS: training cohort, HR=2.70, 95% CI=1.33 to 5.50, p=0.006; validation cohort, HR=2.85, 95% CI=1.21 to 6.68, p=0.016; CSS: training cohort, HR=3.01, 95% CI=1.26 to 7.24, p=0.014; validation cohort, HR=2.90, 95% CI=1.03 to 8.17, p=0.043; table 2). Similarly, Kaplan-Meier analyses demonstrated that patients with high B7-H3 expression had shorter OS and CSS than those with low expression of B7-H3 in both cohorts (for the training cohort, OS: p=0.004, CSS: p=0.009; for the validation cohort, OS: p=0.011, CSS: p=0.034; figure 3A–B).

In line with the results for B7-H3, patients with high HHLA2 expression demonstrated significantly lower OS, especially CSS, than patients with low HHLA2 expression in both cohorts (OS: training cohort, HR=2.73, 95% CI=1.31 to 5.69, p=0.007; validation cohort, HR=3.70, 95% CI=1.55 to 8.83, p=0.003; CSS: training cohort, HR=3.44, 95% CI=1.37 to 8.65, p=0.008; validation cohort, HR=3.06, 95% CI=1.09 to 8.62, p=0.034; table 2). Furthermore, high HHLA2 expression was demonstrated to contribute to shorter survival of patients with PCa in both cohorts (for the training cohort, OS: p=0.005, CSS: p=0.005; for the validation cohort, OS: p=0.002, CSS: p=0.025; figure 3C–D).

Multivariate analysis revealed that neither B7-H3 nor HHLA2 were independent risk factors for OS and CSS in the training cohort (table 3). However, HHLA2 was identified as an independent risk factor for OS in the validation cohort (HR=3.15, 95% CI=1.10 to 9.00, p=0.032, table 3).

A previous study demonstrated that extensive infiltration of CD8⁺ TILs can independently predict a favorable prognosis in PCa.32 Considering the close correlation between B7-H3 and HHLA2 and the infiltration of CD8⁺ TILs, it is confusing that neither B7-H3 nor HHLA2 were independent prognostic factors. To further determine the relationship between B7-H3 and/or HHLA2 expression and the prognosis of patients with PCa, we attempted to generate a potential tool for predicting survival, based on the expression of B7-H3 and HHLA2. The patients with PCa were presorted into four groups according to the expression of B7-H3 and HHLA2 as follows: group I (B7-H3^low/HHLA2^low), group II (B7-H3^low/HHLA2^high), group III (B7-H3^high/HHLA2^low) and group IV (B7-H3^high/HHLA2^high). In both cohorts, Kaplan-Meier analysis demonstrated that patients in groups II–IV experienced shorter OS and CSS than patients in group I (online supplemental figure S2A,B). However, no significant difference was detected among patients in groups II–IV in both cohorts (online supplemental figure S2A,B). Thus, we then divided the patients into two groups: low B7 score, B7-H3^low/HHLA2^low; high B7 score, B7-H3^low/HHLA2^high, B7-H3^high/HHLA2^low or B7-H3^high/HHLA2^high. As shown in online supplemental table S2, a high B7 score was significantly associated with advanced Gleason score (p=0.040 and p=0.001, respectively) and tumor stage (p=0.022 and p=0.002, respectively) in the two cohorts. Moreover, a high B7 score was significantly correlated with lymph node metastasis in the training cohort (p=0.010, online supplemental table S2). Kaplan-Meier analysis also demonstrated that high B7 score significantly stratified OS and CSS in both cohorts (for the training cohort, OS: p=0.001, CSS: p=0.004; for the validation cohort, OS: p<0.001, CSS: p=0.002; figure 4A,B). More importantly, in a multivariate analysis, patients with high B7 score were still over four times more likely to die, especially from tumor-related factors, after adjusting for possible confounding variables in the training cohort (OS: HR=4.33, 95% CI=1.29 to 14.56, p=0.018; CSS: HR=4.54, 95% CI=1.03 to 20.02, p=0.046; table 4). In the validation cohort, the B7 score was an independent predictor for OS and tended to be an independent predictor for CSS (OS: HR=10.76, 95% CI=1.41 to 82.50, p=0.022; CSS: HR=6.52, 95% CI=0.82 to 52.09, p=0.077; table 4). Of note, a high B7 score was inversely associated with a high density of CD8⁺ TILs (p<0.001 in both cohorts, figure 4C). However, no correlation was detected between Foxp3⁺ TILs and B7-H3 or HHLA2 expression in either cohort (figure 4C).

To sum up, these data indicate that the combination of B7-H3 and HHLA2 expression might have better prognostic value in patients with PCa.

Based on the above findings, we developed a novel immune classification scheme according to the B7 score and the density of CD8⁺ TILs. Specifically, the patients with PCa were classified into four immune types: immune type I (low B7 score, high CD8⁺ TILs), immune type II (low B7 score, low CD8⁺ TILs), immune type III (high B7 score, high CD8⁺ TILs) and immune type IV (high B7 score, low CD8⁺ TILs). The proportions of patients with PCa with immune types I, II, III and IV were 14.3% (18/126), 17.5% (22/126), 22% (22/126) and 50.8% (64/126), respectively in the training cohort. Of 113 patients with PCa from the validation cohort, 29.2% (n=33) belonged to immune type I, 14.2% (n=16) belonged to immune type II, 18.6% (n=21) belonged to immune type III and 38.1% (n=43) belonged to immune type IV. According to the Kaplan-Meier analysis, immune type IV was more significantly correlated with the poorest OS and CSS among patients with the four immune types in both cohorts (figure 5A–B). Conversely, immune type I was significantly correlated with improved OS and CSS in both cohorts, even no death was observed during the follow-up (figure 5A–B). However, no difference in OS and CSS was found between immune type II or III in either cohort (figure 5A–B). These data demonstrate that this novel immune classification might provide direction for predicting prognosis and immunotherapy results.