Effects on survival of BAP1 and PBRM1 mutations in sporadic clear-cell renal cell carcinoma: a retrospective analysis with independent validation

SUMMARY

Introduction

Over 60,000 new cases and 13,000 deaths from tumors of the kidney and renal pelvis were expected in the US in 2012.(1) Approximately 70% of renal cell carcinomas (RCC) present with localized disease and ~30% of patients who undergo surgery with curative intent may experience a recurrence.(2) Several variables have emerged that influence outcome including TNM stage, tumor size, Fuhrman grading and necrosis.(3-6) In the metastatic setting, patients can be stratified based on clinical and laboratory parameters.(7) More recently, Heng et al. stratified patients with metastatic RCC based on time from diagnosis to treatment, Karnofsky performance status, hemoglobin, corrected calcium, neutrophils and platelets into favorable (median OS, not reached), intermediate (median OS, 27 months), and poor (median OS, 8.8 months) risk groups.(8) However, what determinants in the tumor account for the different behaviors is poorly understood.

Different behaviors may be driven by different mutations. Clear cell renal cell carcinoma (ccRCC), which accounts for 70-80% of all RCC,(2) is characterized by inactivation of the von Hippel-Lindau gene (VHL).(9) In addition, truncating mutations in PBRM1 (polybromo 1), a gene encoding a SWI/SNF chromatin-remodeling complex component, are found in 41% of ccRCC.(10) While mutated at a substantially lower frequency, other genes implicated in ccRCC are SETD2, KDM6A (UTX), and KDM5C (JARID1C).(10-12) However, whether these mutations affect outcomes is unknown.

Recently, we reported that the gene BAP1 (BRCA1 associated protein-1) was mutated in approximately 15% of ccRCC.(13) Interestingly, BAP1 and PBRM1 mutations in tumors are largely mutually exclusive.(13) In addition, whereas tumors with BAP1 mutations are typically of high grade, tumors exclusively mutated for PBRM1 tend to be of lower grade.(13) These results led us to hypothesize that BAP1-mutated tumors may be associated with worse outcomes than those of patients with PBRM1-mutated tumors.

Methods

Results

The median age was 62 years (IQR=54, 70) and there was a male predominance (Table 1). The mean tumor size was 5.7 cm (IQR=4.1, 8.7). Fifty-two (76 of 145) percent of the tumors were of high grade (Fuhrman grade 3 and 4). Eleven percent (7 of 63) of patients who underwent regional lymph node dissection had nodal metastases and 9% (13 of 145) had distant metastases at the time of surgery.

We investigated a second, completely independent, cohort of ccRCC cases from TCGA. At the time of our analyses, mutation data was available for 327 patients, which represent the cohort analyzed. Demographics and other characteristics are summarized in table 1. The mean age was 61 years (IQR=52,71), with a male predominance. Fifty-eight percent (187 of 325) of the tumors were of high grade. Eight percent (12 of 159) of patients who underwent regional lymph node resection had nodal metastases and 16% (52 of 327) had distant metastases at the time of surgery. Overall, the TCGA and UTSW cohorts were similar, although more patients presented with higher grade and metastases in the TCGA cohort.

Among the patients in the UTSW cohort, there were 21 with BAP1-mutated tumors, including 3 with mutations in both BAP1 and PBRM1. The number of patients with tumors exclusively mutated for PBRM1 was 78. VHL mutations were present in 71% (15 of 21) of BAP1-mutant and 87% (68 of 78) of PBRM1-mutant tumors, respectively. A comparison of patients with BAP1- versus PBRM1-mutated tumors showed that patients with BAP1-mutated tumors were more likely to present with aggressive features including higher grade, sarcomatoid and rhabdoid histology, coagulative tumor necrosis, and mTORC1 activation (p<0.05 for all) (table 2).

In the TCGA cohort, there were 20 BAP1-mutant tumors (including 4 tumors with mutations in both BAP1 and PBRM1) and 74 tumors with mutations exclusively in PBRM1. Consistent with the results in the UTSW cohort, BAP1-mutated tumors showed a trend towards higher grade (p=0.095) and BAP1 mutation was associated with necrosis (p=0.038). In addition, BAP1-mutated tumors were more likely to have advanced pT and clinical stage (p<0.05 for both) (table 2). While the latter associations were not observed in the UTSW cohort, in every instance, BAP1-mutant tumors were associated with indicators of poor outcome.

In the UTSW cohort, patients with BAP1-mutant tumors had a median OS of 4.6 years (95%CI, 2.1-7.2 years), which was substantially shorter than that of patients with PBRM1-mutated tumors, whose median OS was 10.6 years (95%CI, 9.8-11.5 years). The differences in OS corresponded to a HR of 2.7 (95%CI, 0.99-7.6, p=0.044) (figure 1). As in the UTSW cohort, in the TGCA cohort, patients with BAP1-mutated tumors had a significantly higher probability of death (HR, 2.8; 95%CI, 1.4-5.9; p=0.004) (figure 1). While the median OS values differed between the TCGA and UTSW cohorts, possibly reflecting differences in the patient population and the availability of targeted therapies at the time of presentation (supplementary figure 1), the HR in both cohorts were almost identical (2.7 vs. 2.8). These data show that BAP1-mutated tumors are associated with significantly worse OS than PBRM1-mutated tumors.

To assess how representative the cohorts of patients with BAP1- and PBRM1-mutated tumors were, we performed univariate Cox regression analyses. As expected, in both the UTSW and TCGA cohorts, pN, M, stage, grade and necrosis were all associated with OS (supplementary table 1). Race, on the other hand, was not associated with OS in either cohort (supplementary table 1).

Multivariate Cox regression analyses were performed with all variables that reached 0.20 significance in univariate analyses in each cohort respectively (supplementary table 1) except for pN, which had missing data in half of the patients. A backwards elimination process to identify the best fit model found that M and grade were independently associated with OS in the UTSW cohort (supplementary table 2). Other known predictors of outcome were not recovered, possibly owing to the small sample size. Interestingly, in addition to M and grade, mutations in BAP1 and PBRM1 were independently associated with OS in the TCGA cohort (HR, 2.3; 95% CI, 1.03-5.1; p=0.041) (table 3).

Our previous studies with a small number of samples suggested that BAP1- and PBRM1-mutated tumors have different gene expression signatures.(13) To evaluate this notion further, we analyzed gene expression signatures from RNASeq data of the TCGA (available for 308 of the 327 samples with mutation information). We asked whether BAP1-mutant tumors could be distinguished from the rest. A comparison of gene expression between BAP1-mutated tumors (n=20) and the rest identified 3,250 genes that were deregulated in the BAP1-mutant group after a false-discovery rate (FDR) correction (q<0.05) (table 4). To ascertain the significance of these deregulated genes, we asked how many genes would distinguish a group of the same size chosen randomly. When 20 tumors were chosen at random, only 115 genes distinguished this group from the rest. This was repeated twice, and the numbers that distinguished these arbitrary groups were 63 and 120 respectively (table 4). The differences in the number of genes identified between the BAP1-mutant group (3,250 genes) and the groups of random tumors was highly statistically significant (p=<0.0001). These data show that BAP1-mutated tumors are associated with a characteristic gene expression signature.

A comparison of PBRM1-mutant tumors (n=66) to the rest revealed 2,235 genes that distinguished these tumors at an FDR q<0.05 (table 4). When compared to three groups of 66 tumors selected at random, the difference in the number of distinguishing genes (2,235 vs. 0, 0, and 3) was highly significant (p=<0.0001) (table 4). Thus, PBRM1 mutated tumors are also associated with a characteristic gene expression signature.

The number of genes in common between the BAP1 and PBRM1 signatures was 369 (figure 2). However, the overlap expected at random was 381. Thus, the signatures were non-overlapping and this reflected aberrations in different pathways (figure 2). Supplementary figure 2 lists genes in the BAP1 and PBRM1 signatures that most clearly distinguished these groups. BAP1-mutant tumors were characterized by changes in the expression of genes implicated in growth factor signaling, whereas PBRM1-mutant tumors exhibited expression changes in genes implicated in cytoskeleton and tissue architecture.

Finally, while the number of double mutant tumors was very small (3 and 4 in the UTSW and TCGA cohorts respectively), pathological studies of the UTSW cohort had suggested that these tumors were associated with rhabdoid features and may be particularly aggressive.(13) In keeping with these results, Kaplan-Meier estimates of the UTSW cohort showed that double mutant tumors were associated with the worst outcomes (HR 5.3; 95% CI, 1.2-22.9; p=0.012) (supplementary figure 3A). This was also the case for the TCGA cohort (HR, 10; 95%CI, 3.2-33.6, p=<0.0001) (figure 3). We also included in our analyses tumors for which mutations in BAP1 or PBRM1 were not identified (figure 3 and supplementary figure 3B). However, this group is likely to be heterogeneous and made up by more than one molecular genetic subtype.

Discussion

Our findings show that ccRCC can be subclassified into at least two biologically and clinically distinct entities: BAP1-mutant and PBRM1-mutant tumors. These tumors are associated with distinct gene expression signatures, and therefore different biology, and BAP1-mutant tumors exhibited pathological features suggestive of aggressive disease. BAP1-mutant tumors are associated with significantly worse overall survival than PBRM1-mutant tumors (median OS in UTSW of 4.6 vs. 10.6 years; p=0.044). This difference corresponded to a HR of 2.7 (95%CI, 0.99-7.6) and an almost identical HR was observed in the TCGA cohort (HR, 2.8; 95%CI, 1.4-5.9; p=0.004). Thus, this study establishes the foundation for the first molecular genetic classification of sporadic ccRCC.

Why BAP1-mutant tumors are associated with worse survival is not understood. However, in both cohorts, when compared to PBRM1-mutated tumors, BAP1 mutation in tumors was associated with coagulative necrosis, an independent predictor of outcome.(18) In addition, BAP1-mutated tumors exhibited higher Fuhrman grade (p=<0.0001 and p=0.095 in UTSW and TCGA, respectively). The main determinant of Fuhrman grading in everyday practice is nucleolar prominence,(19) which by itself is associated with survival.(20) The nucleolus is the site within the cell where ribosomes are synthesized. Ribosomes are necessary for mRNA translation and both ribosome biogenesis as well as mRNA translation are regulated by mTORC1.(21) Furthermore, mTORC1 activity has been linked to nucleolar size.(22, 23) In RCC, a correlation was found between Fuhrman grading and S6 phosphorylation,(24) a marker of mTORC1 activation. Interestingly, BAP1 mutation is linked to mTORC1 activation (although this association is likely to be indirect)(13) and this may contribute to explain the connection between BAP1 mutation high Fuhrman grade and outcome. Interestingly, in uveal melanoma, where BAP1 is also found to be somatically mutated, BAP1 mutations were present in the majority of metastasizing but in only a minority of non-metastasizing tumors.(25)

ccRCC is characterized by VHL mutations, but VHL inactivation alone is insufficient for tumor initiation.(26, 27) Both BAP1 and PBRM1 are two-hit tumor suppressor genes and they are located on chromosome 3p (where VHL is found), in a region that is deleted in the vast majority of sporadic ccRCC.(14, 28) We speculate that, in many instances, ccRCC development is initiated by a focal mutation in VHL, followed by a 3p deletion. 3p loss would leave cells without VHL gene function and with just one copy of BAP1 and PBRM1. Mutation of the remaining BAP1 or PBRM1 allele may initiate tumorigenesis resulting in tumors, depending upon which gene is mutated, of different aggressiveness. Thus, tumor aggressiveness may be established early on during the process of tumorigenesis.

We hypothesize that BAP1 and PBRM1, when mutated, represent truncal events and that the pathways deregulated by their loss are ideal targets for drug development. Likely accounting for macro- and microscopic differences within tumors, significant genetic heterogeneity has been reported in primary ccRCC.(29) According to their prevalence, mutations can be classified into ubiquitous, shared and private. Ubiquitous mutations are shared across all tumor cells and include initiating events. We postulate that mutations in BAP1 and PBRM1 (and VHL) represent ubiquitous, truncal, drivers of tumor development. Thus, the discovery of BAP1 and PBRM1 mutations in ccRCC may pave the way for the next generation of targeted therapies. Given the particularly poor outcomes of BAP1-mutant tumors, identifying vulnerabilities resulting from BAP1 loss is particularly important. Once candidate drugs are found, their evaluation may be facilitated by the availability of tumorgraft models reproducing the molecular genetics and treatment responsiveness of RCC in humans.(30)

How mutations in BAP1 and PBRM1 drive renal carcinogenesis is not understood. We convincingly show that BAP1- and PBRM1-mutant tumors exhibit highly specific gene expression signatures that distinguish these tumors from the rest. The BAP1 and PBRM1 gene expression signatures are quite distinct and this is in keeping with differences in pathological features and patient outcomes. Interestingly, most genes that make up the BAP1 signature were downregulated in BAP1-mutated tumors. These data raise the possibility that BAP1, which is a nuclear deubiquitinase, may act by deubiquitinating transcription factors, which in the absence of BAP1 are ubiquitinated and targeted for proteosomal-mediated degradation. Support for such a model is provided by a recent report.(31) Similarly, most genes that made up the PBRM1 signature were downregulated in PBRM1-mutant tumors. PBRM1 is the chromatin targeting subunit of a nucleosome remodeling complex and we speculate that when PBRM1 is mutated, increased levels of closed chromatin impair transcription reducing thereby gene expression.

Other genes have been implicated in ccRCC development including SETD2, KDM5C and genes of the MLL family.(10-12, 32) How mutations in these genes relate to mutations in BAP1 and PBRM1 remains to be explored. Similarly, it is presently unknown whether these mutations define other molecular subtypes with different biology and outcomes.

There are several limitations to this study. First, our study evaluated mostly Caucasian patients and the distribution of mutations in different patient populations remains to be determined. Second, the sample size and follow-up are modest. However, for a study assessing the molecular genetics of renal cancer, two completely independent cohorts of 145 and 327 patients respectively is not insubstantial, and in the TCGA, median follow-up was 35 months. In addition, mutation analyses are less susceptible to subjective calls than the more conventional IHC studies. More importantly, these features did not preclude the identification of meaningful and statistically significant differences in OS associated with BAP1- and PBRM1-mutated tumors and the HR were almost identical in both cohorts (2.7 and 2.8). The nearly identical HR and statistically significant p values in two representative and independent patient cohorts strongly supports the conclusion that BAP1- and PBRM1-mutated tumors are associated with distinct survival outcomes. Third, the median OS for patients with either BAP1 or PBRM1 mutant tumors was shorter in the TCGA cohort. However, regardless of mutation status, the OS for all patients was shorter for the TCGA than the UTSW cohort (10.3 vs 5.4 years). Factors that may explain this difference include differences in the patient population with higher rates of patients with metastatic disease [16% (52 of 327) vs 9% (13 of 145)] and high grade [58% (187 of 327) vs 52% (76 of 145)] in the TCGA. Another factor is the availability of targeted therapies. In fact, 37% (43 of 115) of patients in the TCGA cohort died prior to 2006 (when molecularly targeted therapies became available) and only 11% (4 of 35) in the UTSW cohort (see supplementary figure 1).

It remains to be determined whether BAP1 and PBRM1 are independent predictors of outcome. Multivariate analysis of the TCGA cohort showed that mutations in BAP1 and PBRM1 predicted outcome independently of other variables. While this analysis found other known predictors of outcome (grade and M status) several established predictors failed to surface. This is likely due to the small sample size. Consistent with this, prognostic factors in existing nomograms were identified in significantly larger patient cohorts.(3-6) This may also explain why mutation status was not found as an independent predictor in the UTSW cohort. Our efforts continue in the development of robust immunohistochemistry assays that accurately report on BAP1 and PBRM1 and which could facilitate their definitive evaluation as independent predictors of patient survival in larger patient cohorts.

Importantly, however, this is not simply a biomarker study. Biomarkers refer to indicators of disease state that provide prognostic or predictive information. Biomarkers do not necessarily inform on the biology of the tumor and their value is typically predicated upon how much information they add onto existing nomograms. Furthermore, biomarkers may represent epiphenomenological variables with poorly understood links to tumor biology. In contrast, we linked two genetic drivers of ccRCC to disparate outcomes. The classification we propose is based on mutations in driver genes which are associated with distinct gene expression patterns and a different biology. The study is also not simply a classification based on gene expression, which has been reported previously.(33) Rather, the novelty of the study is in establishing the foundation for the first molecular genetic classification of sporadic renal cancer based on distinct biological subtypes that are associated with different outcomes. Furthermore, these different subtypes may have different responses to treatment and only with their recognition can drug effectiveness be properly assessed. For example, a drug could be very active against BAP1-deficient tumors, but since these tumors account for just 15% of all ccRCC, the effect would be masked in an unselected population.

Finally, while the number of patients with BAP1 and PBRM1 double mutant tumors was very small in both cohorts and these tumors are infrequent (1-2% [3 of 145 and 4 of 327] of all sporadic ccRCC), they tend to be associated with both pathological and outcome measures suggestive of greatest aggressiveness.

In conclusion, our results provide the basis for a biologically meaningful and clinically relevant molecular genetic classification of ccRCC that may influence strategies for improved targeted therapies and personalized care.

Supplementary Material

Acknowledgments

We thank Drs. Jonathan Dowel and Joan Schiller for critically reviewing the manuscript. We would like to acknowledge the TCGA for their efforts and providing data that was instrumental for the independent validation of our results.

Funding: Supported by Cancer Prevention and Research Institution of Texas (CPRIT) RP101075 and by 1P30CA142543. The funding sources had no role in the design or conduct of the study.

Footnotes

References

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.