The Shanghai cohort consisted of 809 patients with ccRCC from Zhongshan Hospital Fudan University (Shanghai, China). All these patients received partial or radical nephrectomy between January 7, 2008 and December 23, 2010, and corresponding archived formalin-fixed paraffin-embedded specimen were preserved for immunological staining. None of these 809 patients received neoadjuvant radiotherapy or chemotherapy. Twelve patients with distant metastatic diseases were excluded. Baseline demographic, clinical and laboratory data were collected simultaneously, MRI and CT scans were reassessed in radiology units and all archived diagnostic H&E slides were pathologically central reviewed independently. Patient characteristics are presented in online supplementary table 1.

Sixteen ccRCC fresh tumor specimens (8 BAP1-mutant ccRCC and 8 BAP1-wildtype ccRCC) and five paired peripheral blood samples were collected from Zhongshan Hospital Fudan University. Patients enrolled underwent nephrectomy and were treatment-naïve.

The Cancer Genome Atlas (TCGA) data were downloaded from Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/) in August 2016.14 Five hundred thirty-three patients with ccRCC who underwent renal mass surgery between 1998 and 2013 were enrolled. During survival analysis, 518 patients with intact follow-up data were enrolled. Likewise, the number of patients with recurrence data was 516. RNA expression level was normalized as Z-score or log₂ (expression +1) before analysis.

The primary outcomes of survival analyzes were overall survival (OS) and recurrence-free survival (RFS), representing the time from the date of surgery to the date of death (OS) and recurrence (RFS), or to the date of last follow-up. Further details of the materials and methods used in this study were presented in the online supplementary materials and methods.

The mutation rate of BAP1 in ccRCC was 14%–15%. Genome sequencing in the TCGA cohort identified 40 BAP1-mutant patients with 42 mutations, about 24% (10/42) of which was frameshift mutation, 26% (11/42) of which was nonsense mutation, 12% (5/42) of which splice mutation, 7% (3/42) of which was translation start site mutation. In-frame deletion and missense mutation accounted for 2.4% (1/42) and 28.5% (12/42) cases, respectively. Missense mutations did not affect protein levels. Frameshift mutations led to the degradation of proteins due to translation errors. Nonsense mutations and translation start site mutation led to the failure of protein transcription. Splice site mutation resulted in post-transcriptional splicing errors. Above all, about 70% of these mutations induced a decrease in BAP1 protein expression. Another cohort found that nearly 90% (22/25) BAP1-negative patients with ccRCC measured by immunohistochemistry (IHC) had BAP1 mutation.9 Thus, we use BAP1 mRNA/protein level as proxy of mutation in patients with ccRCC in this study (online supplementary figure 2d). Consistent with previous reports, we found that BAP1 mutation was typically associated with loss of the protein and indicated poor prognosis for patients with renal cell carcinoma9 15 (online supplementary figure 2a-c and online supplementary table 2). To understand the possible mechanism underlying this phenomenon, we analyzed the correlation between BAP1 mRNA expression and various chemokines plus their receptors. We found that the expression of CCR5 and its ligands were anticorrelated with BAP1 (figure 1A). Additionally, patients with higher expression of CCR5 and its ligands were at greater risk of death (figure 1B). Real-time PCR analysis confirmed this in BAP1-mutant patients with ccRCC (figure 1C). IHC analysis further revealed that BAP1 protein was negatively associated with CCR5 and its ligands (figure 1D-E). Next, we stably integrated an RNA interference vector into murine RENCA cells and established a BAP1 knockdown cell line, which was then used to construct a subcutaneous tumor-bearing mouse model. Flow cytometry and ELISA results showed that the proportion of CCR5⁺ cells and the expression levels of CCL2, CCL3 and CCL5 in ccRCC tumors were higher than those in the control group (figure 1F). These results indicated the association existing between BAP1 gene mutation and high expression of CCR5 and its ligands.

The above-mentioned results suggested that CCR5 might be a potential therapeutic target for BAP1-mutant ccRCC. Indeed, we found that in BAP1-low group of patients, increased CCR5 expression was indicative of poor OS and RFS (figure 2A), while no difference was observed in BAP1-high group of patients (online supplementary figure 3a). In BAP1-knockdown subcutaneous tumor-bearing mouse model, tumor volume was significantly reduced after treating with an anti-CCR5 antibody (figure 2B). A survival benefit was observed in the orthotopic mouse model as well (figure 2C). Moreover, we found that the use of anti-CCR5 antibody caused extensive necrosis of tumor cells (figure 2D) and a similar result was observed by flow cytometry analysis. Our previous study16 established a novel in vitro culture system incorporating bulk single cell suspensions from resected ccRCC tumors to simulate the in vivo tumor immune system of patients. And similar models were used to test the therapeutic effect of histone deacetylation inhibitor in lung cancer.17 In our study, resected ccRCC specimens were disaggregated and cultured in vitro, and maraviroc (an FDA-approved anti-CCR5 antagonist) was added for activity testing. The fraction of dead renal tumor cells in BAP1-mutant patients was significantly higher than that in the isotype control (figure 2E). However, anti-CCR5 antibody had no effect in BAP1-wildtype ccRCC (online supplementary figure 3b-c). These findings suggested that blockade of CCR5 could effectively inhibit the progression of BAP1-mutant ccRCC.

After observing a large number of immune cells around the necrotic foci of renal tumor cells treated with the anti-CCR5 antibody, we postulated that blockade of CCR5 could enhance the antitumor immune response. Thus, we investigated the infiltrating level of local immune cells in tumor-bearing mice. We found that the proportions of tumor-infiltrating CD8⁺ T cells and dendritic cells (DCs) were increased in mice treated with anti-CCR5 antibody and the corresponding proportions of interferon (IFN)-γ⁺, GZMB⁺, PRF1⁺, CD80⁺ and CD86⁺ cells were also increased, indicating enhanced cell cytotoxicity and antigen presentation (figure 3A). Meanwhile, the proportions of programmed cell death ligand 1 (PD-L1)⁺ and CTLA-4⁺ cells were significantly decreased (figure 3A). We further digested fresh tissue specimens from BAP1-mutant patients with ccRCC into single cell suspensions and cultured the cells in vitro with maraviroc as the negative allosteric modulator for 4 days. Subsequently, flow cytometry revealed no significant differences in the activation marker CD69 of CD4⁺ T and CD8⁺ T cells; however, the proportion of CD4⁺ T cells secreting the functional effector molecule IFN-γ and CD8⁺ T cells secreting IFN-γ and GZMB or expressing CD107a were significantly increased. The number of DCs increased together with elevated expression of CD80 and CD86; however, the change in HLA-DR was not significant. The proliferation ability of CD8⁺ T cells and DCs were enhanced as well (figure 3B). In addition, interleukin-10 (IL-10), transforming growth factor-β (TGF-β), arginase and inducible nitric oxide synthase expression levels in tumor supernatants were measured. We found that the expression levels of all these inhibitory molecules decreased following maraviroc treatment (figure 3C). However, the effect of CCR5-blocking antibody on the antitumor immune response was not significant in BAP1-wildtype ccRCC mice (online supplementary figure 3d).

We next explored the specific mechanism by which blockade of CCR5 activated the antitumor immunity. We analyzed the fraction of infiltrating immune cells in ccRCC using the CIBERSORT method18 and found a higher fraction of Tregs in BAP1-mutant patients with ccRCC (figure 4A). A previous study indicated that CCR5 was partially expressed on the surface of Tregs and blockade of CCR5 affected the recruitment of Tregs.19 Likewise, we found that the total number of Tregs, especially CCR5⁺ Tregs, was decreased after treatment with the CCR5-blocking antibody compared with the control group, but no obvious change was found in CCR5⁻ Tregs (figure 4B,C). In addition, we found that total CD8⁺ T cells increased and CD4⁺ T cells and macrophages decreased slightly after CCR5 blockade, but there was no significantly difference in total number of CD8⁺ T cells, CD4⁺ T cells and macrophages or corresponding CCR5⁺ or CCR5⁻ immune cell subtypes between the two groups (online supplementary figure 4a). It suggested that CCR5⁺ Tregs might play an important role in this process. We then isolated human peripheral blood-derived CCR5⁺ Tregs and found that CCR5⁺ Tregs were recruited by CCL3 and CCL4, but this chemotaxis was significantly attenuated after maraviroc treatment (figure 4D). Next, CCR5⁺ Tregs and CCR5⁻ Tregs were isolated to investigate their functional differences. Compared with CCR5⁻ Tregs, CCR5⁺ Tregs secreted higher levels of IL-10 and TGF-β (figure 4E), which indicated their stronger immunosuppressive function.

Given that research has shown positive PD-L1 expression in tumor cells20 and that our study revealed a decreased number of tumor-infiltrating PD-L1⁺ cells after treatment with the CCR5-blocking antibody (figure 3A), we analyzed the distribution of PD-L1 in BAP1-mutant tumor tissues and found that tumor cells had a significantly higher proportion of PD-L1⁺ cells than non-tumor cells (figure 5A). The proportion of PD-L1⁺ tumor cells decreased, while the proportion of PD-L1⁺ non-tumor cells did not change substantially after maraviroc treatment (figure 5B).

Interestingly, tumor cells also included a relatively high proportion of CCR5⁺ cells (figure 5C). We stimulated the isolated BAP1-mutant renal tumor cells with CCR5 ligands in vitro with addition of maraviroc. Subsequent analysis revealed CCL5 could significantly enhanced PD-L1 expression by tumor cells, followed by CCL3 and CCL8. However, this effect was reversed after blocking CCR5 (figure 5D). Moreover, blockade of CCR5 caused PD-L1 expression to decrease without addition of exogenous chemokines. We isolated tumor cells from BAP1-wildtype and BAP1-mutant patients with ccRCC and cultured them for 3 days in vitro. The concentration of CCL2-5 was significantly higher in BAP1-mutant ccRCC (figure 5E). A similar result was obtained from the BAP1-knockdown RENCA cell line constructed with BAP1 short hairpin RNA (figure 5F).