Prostate cancer

Integrative molecular analyses define correlates of high B7-H3 expression in metastatic castrate-resistant prostate cancer

Localized prostate cancer (PC) is curable, but options are limited for recurrent or metastatic tumors developing resistance to androgen-deprivation therapy (ADT) or AR targeted therapy (ART), known as metastatic castration-resistant prostate cancer (mCRPC). Overexpressed tumor antigens, such as PSMA, are targets of novel PET imaging approaches1 as well as precision therapeutics (177Lu-PSMA-617) in mCRPC ( Identifying additional mCRPC tumor antigens contributes to new strategies to develop precision antibody-drug conjugates that permit immuno- or cellular therapies2,3.

B7-H3 is a transmembrane glycoprotein in the B7 immune checkpoint superfamily4. Other well-known members, such as PD-L1 and CTLA-4, are targets in various hematologic and solid tumors5,6,7. B7-H3 is overexpressed in several cancers including prostate cancer, with minimal expression in normal prostatic tissue8,9,10,11,12. Higher expression of B7-H3 correlates with poor cancer prognosis8. B7-H3 has implications for cancer cell transformation and metastasis, and is thought to have a significant effect on the tumor microenvironment and immune suppression11,13. New strategies have been developed to target B7-H3 through antibody-dependent cell-mediated cytotoxicity14, antibodydrug conjugates15 and linking with immunotherapy such as chimeric antigen receptor-T cell16 or NK cell therapies17. However, little is known about the molecular features and regulatory mechanisms of B7-H3 in mCRPC, which prevents the optimal design of such targeted interventions and precludes rational patient selection. To address these barriers, we characterized the genomic, transcriptomic and epigenomic features of B7-H3 expression in mCRPC.

We first evaluated transcript expression profiles of B7-H3 (CD276) and other immune-regulatory genes in mCRPC. We conducted bioinformatic interrogations on whole-exome (WES) and whole-transcriptome sequencing (WTS) data from the datasets including MSKCC 2010 (n = 131, primary PC; n = 19, CRPC)18, SU2C/PCF (n = 208, mCRPC)19, SUWC (n = 101, mCRPC)20, and GTEx (n = 245, benign prostate tissue) datasets21. The MSKCC samples included both primary and metastatic tumors processed through the same platform. mRNA expression of B7-H3 was significantly increased in metastatic PC compared to primary PC (p = 0.004) (Fig. 1a). In other mCRPC datasets, expression of B7-H3 was significantly elevated in both mCRPC datasets compared to benign prostate tissues (median TPM 115, 87 vs. 60, p < 0.0001) (Fig. 1b). We also evaluated the association of the expression of B7-H3 mRNA with protein levels and found significant association in 369 cancer cell lines (p = 1.03E-72) (Supplementary Fig. 1a). In 10 patient-derived xenograft (PDX) models of castration-resistant prostate cancer (LuCaP PDX series)22, we detected B7-H3 expression in each PDX tumor pair and found a positive trend between mRNA and protein expression (n = 10, r = 0.52, p = 0.06) (Supplementary Figs. 1b–c and 4). Based on TPM, other B7 family members including PD-L1 (CD274), PD-L2 (PDCD1LG2), and CTLA-4, exhibited reduced expression in both mCRPC datasets and had limited overall transcript abundance compared to B7-H3 (Fig. 1c–e). Other immunological markers exhibited low abundance or were not overexpressed in mCRPC (Supplementary Fig. 2). The expression of B7-H3 in mCRPC was independent of that of PSMA, and exhibited robust expression even in PSMA-low tumors (Fig. 1f–g). This suggests that targeting B7-H3 could be an attractive alternative for PSMA-negative/low mCRPC patients.

Fig. 1: B7-H3 is selectively overexpressed in mCRPC.
figure 1

a mRNA expression of B7-H3 in primary and metastatic prostate cancer (PC). Whole-transcriptome sequencing (WTS) data were obtained and analyzed from MSKCC 2010 (primary PC in gray, n = 131; metastatic PC in red, n = 19). All data are median with 95% CI. Statistical significance was using student t-test. **P < 0.01. be mRNA expression of B7-H3, PD-L1, PD-L2, and CTLA-4, respectively, in mCRPC (SU2C, green; SUWC, orange) and normal prostate tissue (GTEX, blue). Whole-transcriptome sequencing (WTS) data were obtained and analyzed from SU2C/PCF (n = 208, mCRPC), SUWC (n = 101, mCRPC), and GTEx (n = 245, benign prostate tissue). Data of other B7 family genes are shown in Supplementary Fig. 2. All data are median with 95% CI. Statistical significance was assessed using one-way ANOVA for multiple comparisons. ****P < 0.0001. f, g Lack of correlation of mRNA expression of B7-H3 and PSMA in mCRPC datasets SU2C/PCF (n = 208, r = 0.0003) and SUWC (n = 101, r = −0.024), respectively. Associations were determined by Pearson correlations. h Percentage of cells expressing B7-H3 (B7-H3 positive, green; B7-H3 negative, gray) before (17.9%) and after (38.5%) enzalutamide treatment. ScRNA-Seq analysis was performed on cells from patient before (n = 112) and after enzalutamide (n = 83) treatment. i GSEA of B7-H3 with functional oncogenic pathways. WTS data of mCRPC datasets SU2C/PCF (n = 208) and SUWC (n = 101) mCRPC were combined for GSEA. NES cutoff value 1.4. NES normalized enrichment score. FDR false discovery rate.

To examine how mCRPC tumor cells regulate B7-H3 expression in response to the ART, enzalutamide, we analyzed single-cell mRNA sequencing (scRNA-seq) data of paired biopsy samples from one patient (pre- and post-enzalutamide). We found an increased proportion of B7-H3-expressing tumor cells post-enzalutamide (38.5%) relative to pre-enzalutamide (17.9%) (Fig. 1h). Based on the genomic alterations in mCRPC with high B7-H3 expression, B7-H3 was associated with several known resistance markers including PTEN inactivation and AR-V7 detection19,20,23 (Supplementary Fig. 3). We conducted Gene Set Enrichment Analysis (GSEA)24 on mCRPC datasets (SU2C/PCF, n = 208 and SUWC, n = 101) and identified B7-H3 was enriched of TGF-beta, WNT, and Epithelial-to-Mesenchymal Transition (EMT) signaling pathways (Fig. 1i); each has been associated with resistance to enzalutamide25,26,27. Altogether, we found robust B7-H3 expression in mCRPC patients with existing molecular or signaling features that promote resistance to ADT and/or ART.

To enhance our mechanistic understanding of B7-H3 expression in mCRPC, we developed a machine-learning algorithm that quantitatively measures the degree of all gene-to-gene interactions to construct gene networks for all detectable genes. We used this algorithm to compare the degree of gene-network interactions between B7-H3 and all other detectable gene networks in the 208 mCRPC patients from the SU2C/PCF study. The overarching degree of gene-network association was visualized on UMAP, depicted through distances on an x-y plane. Remarkably, B7-H3 networks were closely clustered with those of AR, as well as with FOXA1, HOXB13, SPOP, MYC, and ERG (Fig. 2a). CTLA-4, PD-L1, PD-L2, and other immune markers were in distinct clusters (Fig. 2a), which indicated a lack of association with AR-signaling genes. We further examined the similarities of gene networks of B7-H3 and key regulators of AR signaling on a violin plot, in which the degree of overlap represents similarity. We observed that the B7-H3 gene network overlapped with those of AR, HOXB13, and FOXA1, and to a lesser degree with SPOP, but exhibited no intersection with PD-L1 (Fig. 2b). Altogether, these analyses suggest a robust convergence between B7-H3 and multiple genes with known functions in AR signaling.

Fig. 2: B7-H3 is significantly associated with and regulated by AR-signaling pathways.
figure 2

a Machine-learning (ML)-based UMAP analysis of the association between B7-H3 and key PC pathways. Each dot in UMAP represents one gene. The spatial distance between two genes represents the similarity of their gene networks. Key PC pathways are visualized including AR signaling (green), Cell cycle (blue), Kinases (yellow), and Immune markers (purple) along with B7-H3 (pink). b ML-based analysis of the gene-network association between B7-H3 and key AR-signaling pathway genes in mCRPC patients (SU2C/PCF, n = 208). Data are shown in violin plots, in which red lines represent median and blue lines represent first quartile (lower) and third quartile (upper). The boundary of the violin represents the range of all data points. The degree of overlap of the plots represents the similarity of the networks they are associated with. PD-L1 was used as a negative control. c Comparison of H3K27ac enhancement at B7-H3 promoter in mCRPC and primary PC from representative patient-derived xenografts (PDXs)22. Pdiff indicates the FDR-adjusted P-value for comparison between primary prostate cancer and mCRPC using DESeq2. d Enhanced interaction between B7-H3 distal enhancers (green box) and its promoter (red box) in mCRPC as compared to primary PC. Top track indicates H3K27ac HiChIP data from LNCaP, which reflects long range chromosomal interactions in LNCaP cells. e Binding of AR, FOXA1, and HOXB13 to multiple putative B7-H3 enhancer sequences (green box) in mCRPC. f Increased AR binding in mCRPC at one of the putative CD276 enhancers indicated in d, labeled with green *. Six representative AR ChIP-seq profiles of primary prostate cancer and mCRPC are shown32.

B7-H3 is regulated epigenetically in nasopharyngeal carcinoma28 and glioblastoma29, via histone acetylation and DNA methylation at the promotor, respectively. The convergence between B7-H3 and AR signaling that we identified agreed with prior studies30. We thus sought to interrogate this mechanism of regulation through CHIP-seq data from both primary prostate cancer and mCRPC xenograft samples31,32. Remarkably, we observed enhanced histone-3-lysine-27 acetylation (H3K27ac) marks at the B7-H3 promoter and at putative B7-H3 distal enhancers in mCRPC as compared to primary prostate cancer (Fig. 2c, d), which reflected molecular mechanisms that increased transcription of B7-H3 in mCRPC. Further, we found that AR (and its co-regulators HOXB13 and FOXA1) were directly bound to B7-H3 enhancers (Fig. 2e). Notably, we found that AR exhibited selective binding to one of the B7-H3 putative enhancers in mCRPC as opposed to primary tumors (Fig. 2f). Although AR signaling is active at all stages of prostate cancer, our ChIP-seq analysis illustrated differential epigenetic regulation of B7-H3 transcripts in mCRPC compared to primary prostate cancer.

Our findings provide support that B7-H3-targeting therapies can fulfill an unmet medical need for ADT/ART-resistant mCRPC patients. Strategies to target B7-H3 with checkpoint inhibitors (NCT03729596), monoclonal antibodies (NCT02923180), antibody-drug conjugates (NCT03729596, NCT04145622), or tri-specific killer engager (TriKE) agents, are currently under investigation33,34. These therapeutics could be rationally designed for mCRPC patients that harbor ADT/ART-resistant biomarkers (e.g., PTEN loss, AR-V7 or ERG fusion) or other oncogenic signaling pathways (WNT, EMT, TGF-Beta). Further, targeting B7-H3 may be relevant in mCRPC patients with limited expression of PSMA, although our analysis did not address the status of B7-H3 in neuroendocrine/ small-cell prostate cancers. Our findings established a mechanistic connection between B7-H3 expression and AR-related signaling in mCRPC. This may also hold true in high-risk localized prostate tumors, since B7-H3 immunostaining is reduced after intense neoadjuvant ADT given before radical prostatectomy30. Finally, the epigenetic modifications we found may act as surrogates to measure B7-H3 levels from noninvasive liquid biopsies that include circulating-tumor DNA from mCRPC patients27.