Presence of PBAF mutations in TCGA across cancers

To evaluate the prevalence of PBAF complex mutations, we queried the pan-cancer TCGA atlas (n = 10,359) and analyzed all three genes in the complex (PBRM1, ARID2, and BRD7). Overall, 7.7% of all tumors possessed any PBAF complex mutation; incidence among malignancies included in the pan-cancer TCGA cohort was highest in ccRCC (KIRC) particularly for PBRM1 mutations, followed by melanoma (SKCM), cholangiocarcinoma (CHOL), stomach (STAD), uterine (UCEC), and bladder (BLCA) cancers (Fig. 1a, b). Additionally, highly mutated tumors were more likely to possess PBAF complex mutations (Fig. 2). Mutations in PBRM1, ARID2, and BRD7 each represented about 3.8% (52.7% LOF), 3.6% (39% LOF), and 1.1% (34.4% LOF), respectively (Supplementary Fig. 1).

MSK-IMPACT immunotherapy pan-cancer cohort

We first assessed the incidence of PBRM1 and ARID2 mutations (BRD7 not included in IMPACT) among all patients treated with ICB (n = 3700). For the MSK-IMPACT cohort, we restricted our analysis to 189 ccRCC patients and 2936 patients treated with immunotherapy comprising 11 other cancer types that had a minimum of 50 patients and 5 PBRM1 or ARID2 mutants. Clinical characteristics of the included cohort, including age, gender and drug class, as well as TMB and fraction genome altered (FGA) vary substantially across cancer types and are included in Table 1. Available PBAF complex mutations included PBRM1 and ARID2, present at 7.4% and 6.5%, respectively, across the pan-cancer cohort; LOF frequencies are 3.9% and 2.3%, respectively.

Consistent with the TCGA analysis, PBRM1 mutations were most common in ccRCC patients (46.6%), followed by non-melanoma skin cancer (9%) and melanoma (8%), while ARID2 was most common in melanoma (13%), followed by non-melanoma skin cancer and colorectal cancer (11%) (Supplementary Fig. 2). With the exception of ccRCC, several of the tumors harboring PBAF complex mutations were highly mutated cancer types (Fig. 3a, b). This included both LOF mutations (frameshift and nonsense mutations) as well as missense mutations.

PBAF mutation and response to immunotherapy in ccRCC

Given the findings of Miao et al.⁶ and Braun et al.⁹ with respect to PBRM1 LOF mutations and response to ICB in RCC, we analyzed our cohort of ICB-treated metastatic ccRCC patients (n = 189) with more detailed clinical annotations including International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) prognostic score (n = 180; Table 2), treatment details and outcomes with therapy, including time-to-treatment failure (TTF). PBRM1 LOF mutations were present in 61 of these 189 patients (32%), and non-LOF mutations were found in 27 patients (14%). Since ARID2 mutations were only present in six patients (4 of them LOF) and might have a distinct effect on outcome, we further analyzed only PBRM1 mutations. There were 57 deaths and 147 treatment failures among these 189 patients, with a median overall survival (OS) of 68.2 months (95% CI 44.4, NA) and median TTF of 8.9 months (95% CI 6.9, 12.42), but no difference for either outcome when comparing patients with PBRM1 LOF to others (Fig. 4). PBRM1 mutation rate was not significantly different in patients who received first line (n = 97, 30% LOF mutations) or second or higher line (n = 92, 35%) ICB or ICB/VEGF combinations. PBRM1 mutations were not associated with TTF in the entire ccRCC cohort (LOF HR 0.73, p = 0.11; non-LOF HR 1.05, p = 0.84) and not significantly associated with OS (LOF HR = 1.5, p = 0.16; non-LOF HR = 1.05, p = 0.91) (Table 3). When comparing outcomes with first-line ICB therapy in patients with PBRM1 LOF mutations vs. wild type, no significant differences were seen for TTF (HR = 0.6, p = 0.075) or OS (HR = 1.7, p = 0.29); similarly, no differences were seen for those receiving ICB in the second line or higher (TTF HR = 0.87, p = 0.61; OS HR = 1.71.3, p = 0.44) (Fig. 5). In a multivariate model adjusted for TMB and drug class (significant predictors of progression free survival), PBRM1 was not significantly associated with TTF (LOF HR = 0.85, 95% CI 0.57, 1.28, p = 0.44; non-LOF HR = 1.22, 95% CI 0.77, 1.94, p = 0.4). Similarly, in the model for OS adjusted for IMDC risk score and line of therapy (significant predictors of OS), PBRM1 was not significant (LOF HR = 1.24, 95% CI 0.69, 2.25, p = 0.47; non-LOF HR = 0.88, 95% CI 0.36, 2.14, p = 0.78).

PBAF complex mutation and ICB outcomes in other cancer types

To assess the impact of PBAF complex mutations in non-RCC cohorts treated with ICB profiled with MSK-IMPACT, we restricted our analysis to 11 tumor types with at least 50 patients and at least 5 patients with PBRM1 or ARID2 mutations (n = 2936). These included bladder, colorectal, non-small-cell lung, esophagogastric, endometrial, non-melanoma skin, hepatobiliary, head and neck cancers, melanoma, glioma, and cancer of unknown primary. Overall prevalence was 4.9% for PBRM1 (2% of them LOF) and 6.7% for ARID2 (3% of them LOF). PBRM1 mutations were not significantly associated with OS in a cohort of 11 cancer types in a Cox model stratified by cancer type (LOF HR = 0.9, 95% CI 0.6,1.4, p = 0.7; non-LOF HR 1.03, 95% CI 0.73,1.5, p = 0.86), and remained insignificant after adjusting for TMB and total CNA (LOF HR = 1.2, 95% CI 0.8,1.81, p = 0.37; non-LOF HR = 1.32, 95% CI 0.92,1.9, p = 0.13) (Supplementary Table 1). Results were similar when combining PBRM1 and ARID2; LOF HR = 0.85, 95% CI 0.65, 1.1, p = 0.25 unadjusted and HR = 1.1, 95% CI 0.83,1.45, p = 0.52 adjusted. Given the higher frequency of ARID2 mutations in the non-RCC cohorts, we combined PBRM1 and ARID2 LOF and non-LOF mutations for individual subtype analysis, which was significant in non-small-cell lung cancers (Fig. 6a, Supplementary Table 2). When univariately examining LOF mutations in PBRM1 and ARID2 as well as LOF in PBRM1 alone, they remained significantly associated with adverse OS in non-small-cell lung cancer (Fig. 6b, c). In individual cancer types, PBRM1 was correlated with worse OS in non-small-cell lung cancers (n = 983; HR 2.91, p < 0.001) after adjusting for TMB and total CNA (Supplementary Table 3). A significant correlation with adverse OS was also seen in bladder cancer (n = 245; HR 11.85, p < 0.001); however, only three PBRM1 mutants comprised this group. ARID2 was not significant in either cancer type.

PBRM1 mutations and the TME

Previous work by our group and others suggested that PBRM1 loss was associated with further hypoxic signaling and angiogenic expression^{11, 12}. This was further bolstered by the association with improved response of PBRM1 mutated tumors to VEGF blockade therapies^13,14,15. We utilized transcriptomic data from three independent cohorts to analyze the impact of PBRM1 LOF mutations on transcriptional pathway enrichment. These included COMPARZ¹⁶, a phase 3 randomized trial comparing the efficacy and safety of pazopanib and sunitinib as first-line therapy (n = 352 (targeted exome and whole-genome RNA microarray)), McDermott et al.⁴, a randomized phase 2 study of atezolizumab alone or combined with bevacizumab versus sunitinib in treatment-naive metastatic renal cell carcinoma (n = 201 (whole exome + RNASeq)), and Miao et al.⁸, which analyzed a cohort of approximately 100 metastatic ccRCC patients to identify genomic alterations correlating to response to ICB (n = 41 (whole exome + RNASeq)). All three cohorts demonstrated higher hypoxia pathway enrichment in PBRM1 mutated samples with GSEA p value as 0.002, 0.008, and 0.002, respectively. In the COMPARZ and McDermott et al. cohorts, we observed downregulation of interferon alpha and gamma response genes. With respect to interferon gamma response or JAK/STAT signaling, we were able to validate higher expression in the Miao et al. cohort (as previously reported) but we found lower expression in both the COMPARZ and McDermott et al. cohorts. We further performed immune deconvolution using ssGSEA focusing on immune and angiogenic gene signatures. We consistently observed significantly higher angiogenic gene expression in PBRM1 mutated tumors in the COMPARZ and McDermott et al. data sets, p = 0.0004 and 0.005, respectively, and a similar trend in Miao et al. cohort (Fig. 7a). Further, immunohistochemical (IHC) staining from the COMPARZ and McDermott cohort demonstrates significantly higher CD31-positive staining in PBRM1 mutated tumors, implying higher degrees of tumor angiogenesis in PBRM1 mutated tumors (Fig. 7b). IHC studies from the two cohorts also reveal higher PD-L1-negative and lower PD-L1-positive staining tumors in PBRM1 mutated tumors compared to wild type (Fig. 7b) and no difference in CD8 positivity between PBRM1 mutant and wild-type tumors (Supplementary Fig. 3). Immune deconvolution of bulk expression data failed to find any specific immune enrichment patterns across the three cohorts when stratified by PBRM1 mutation status (Supplementary Fig. 4).

a Immune deconvolution using single sample gene set enrichment analysis (GSEA), focusing on immune and angiogenic gene signatures. Significantly higher angiogenic gene expression was observed in PBRM1 mutated tumors in the COMPARZ¹⁶ and McDermott et al.⁴ data sets, p = 0.0004 and 0.005, respectively, and a similar trend in the Miao et al.⁸ cohort. b Immunohistochemistry staining results from COMPARZ¹⁶ and McDermott et al.⁴ data sets demonstrate significantly higher CD31+ staining in PBRM1 mutated tumors and lower PD-L1+ staining in PBRM1 mutated tumors. Box plot: middle line of box indicates median and the bounds indicate quartile 1 and quartile 3. The whiskers reach to the maximum/minimum point within the 1.5 × interquartile range from quartile 3/quartile 1, respectively. P values from COMPARZ¹⁶ bar plots generated by Fisher’s exact test; p values from the GSEA plot derived from a permutation test; p values from immune deconvolution difference plot and box plots derived from Wilcoxon rank-sum test. The Fisher’s Exact test and Wilcoxon rank-sum test p values are two sided. No adjustments made for multiple comparisons; all p values are nominal.