A pan-cancer analysis of PBAF complex mutations and their association with immunotherapy response

There is conflicting data regarding the role of PBAF complex mutations and response to immune checkpoint blockade (ICB) therapy in clear cell renal cell carcinoma (ccRCC) and other solid tumors. We assess the prevalence of PBAF complex mutations from two large cohorts including the pan-cancer TCGA project (n = 10,359) and the MSK-IMPACT pan-cancer immunotherapy cohort (n = 3700). Across both cohorts, PBAF complex mutations, predominantly PBRM1 mutations, are most common in ccRCC. In multivariate models of ccRCC patients treated with ICB (n = 189), loss-of-function (LOF) mutations in PBRM1 are not associated with overall survival (OS) (HR = 1.24, p = 0.47) or time to treatment failure (HR = 0.85, p = 0.44). In a series of 11 solid tumors (n = 2936), LOF mutations are not associated with improved OS in a stratified multivariate model (HR = 0.9, p = 0.7). In a current series of solid tumors treated with ICB, we are unable to demonstrate favorable response to ICB in patients with PBAF complex mutations.

MINOR POINTS 1. Authors should discuss the concept that any novel putative prognostic and/or predictive molecular biomarker (the latter being definitely more important) should prove to be superior to the best available clinical prognostic/predictive factor, before entering clinical practice. This is even more relevant if the proposed molecular biomarker is much more complex, costly and difficult to be reproduced, as compared to the clinical one. If the above is true, it is clear that we are still in the infancy of our quest for a really useful molecular biomarker and this should be clearly highlighted.

-We appreciate the reviewers' comments and certainly agree with the notion that novel prognostic and/or predictive molecular biomarkers should prove to be superior to the best available clinical prognostic/predictive factors before entering clinical practice. In particular, two recent studies have specifically examined the applicability of genomic signatures to improve the prognostic performance of established prognostic models; de Velasco et al demonstrated a model of genomic signatures which improved the prognostic performance of the IMDC and MSKCC classification [de Velasco et al, Oncologist, 2017]. Similarly, a study from our institution reported an integrated genomic and transcriptomic analysis of patients with clear cell renal cell carcinoma and identified four distinct molecular subgroups associated with response and survival, which ultimately has broad implications for optimizing precision treatment of renal cell carcinoma [Hakimi et al, Cancer Discov, 2019]. These relevant points have been added to the discussion section of the manuscript (changes highlighted in yellow).
2. Theoretically, the most important use of a molecular/genetic biomarker would be to aid us in selecting the treatment most likely efficacious for a given patient with a given tumor, or else to avoid potentially unuseful and/or harmful treatments. A number of preliminary reports have been used in the past few years to claim for a precision approach to cancer treatment. Given the conflicting results available, and the lack of useful correlations emerging from important studies such the present, despite intriguing and promising backgrounds, Authors should briefly discuss the present limits of precision Medicine. In this sense, commenting the words of wisdom from Tannock and Hickman (N Engl J Med 2016;375:1289-94) would be appropriate.
-We agree with the reviewers' point in the importance of recognizing the current limitations of precision medicine, and as such, these limitations are included and expounded upon in the discussion section of the manuscript.
Reviewer #2 (Remarks to the Author): Hakimi et al. assessed the frequency of PBAF complex mutations across two large pan-cancer cohorts (TCGA and MSKCC), they evaluated the prognostic significance of PBRM1 mutations in clear cell RCC patients, and examined the association of PBRM1 mutations with clinical response to ICB therapies across a series of solid tumor cohorts from MSKCC. The authors leveraged their access to the large genomic and clinical trial datasets to address an important question. While this work is of interest, there are however a number of issues (e.g superficial analysis, lack of mechanistic insights) that limit the overall impact of the study.
Specific points: 1. Only three (two for the ICB studies) genes of the PBAF complex were studied, while other potentially functional elements of this complex were not included. -

Of the three genes, we studied PBRM1 and ARID2 in our MSKCC pan-cancer and RCC cohort; BRD7 is not included in MSK-IMPACT, and was therefore unable to be studied (this is mentioned in paragraph two of the results section).
2. Loss of function of PBRM1 was solely defined based on the single platform mutation data (whether there was frameshift or nonsense mutations in this gene. It is unclear whether the authors included the splicing site mutations as LOF or classified them as non-LOF mutations; Were the missense mutations treated uniformly as non-LOF without further functional evaluation? The clonal status of the mutations were not analyzed. It is unclear whether the author stratified their analysis by copy number alterations (such as 3p loss which is very frequent in some cancers); Importantly, other alterations such as RNA editing, splicing variants, or epigenetic dysregulation may also lead to loss of function of PBAF complex, but was not examined in this study. -

Missense mutations were treated uniformly as non-LOF, but certainly agree that some missense mutations may lead to LOF. To address the point that the clonal status of mutations were not analyzed, several high-impact sequencing studies including TRACERx (Turajlic et al, Cell, 2018) and TCGA (Creighton et al, Nature, 2013) have shown that 3p loss (which encompasses four commonly mutated genes: VHL, PBRM1, SETD2, BAP1) is a ubiquitous, pathognomonic event in clear cell RCC, occurring in upwards of 90% of tumors. This point has been included in the methods section (highlighted in yellow).
3. It is unclear about the functional impact of the genomic mutations, as LOF of PBAF complex was not assessed at protein level. -

The reviewer brings to light an important point regarding the functional impact of LOF mutations of the PBAF complex. While we did not test protein expression levels, immunohistochemical data from others suggest that protein level loss is similar to PBRM1 mutation rate in clear cell RCC, approximately 43% (Ho et al, Urol Oncol, 2015).
4. The tumors that were wt or non-lof for PBRM1 may carry other genomic or epigenomic alterations that result in PBRM1 loss. -

We agree with the possibility that wild-type or non-lof for PBRM1 may carry other genomic or epigenomic alterations that result in PBRM1 loss.
5. For the survival analysis, the authors should discuss the potential confounding factors (clinical, histopathological, or genomic that are likely to also confound the analysis), and if needed, stratify their analysis by the confounding factors.

We agree that there may be potential confounding factors in our analysis, and as such, we have stratified our analysis by the potential confounding factors found in the regression models in
6. For the correlation analysis with ICB response, the authors should include other known factors that influence patient response such as the levels of immune cell infiltration, CD8 expression, cytotoxic T cell proportion, PD-1/PD-L1/CTLA-4 expression, etc. and correlate these known factors with PBAF complex status to better understand their function in immune regulation and predictive/prognostic values on ICB response.

Data from others, including secondary analyses of the Checkmate 025 clinical trial [Motzer et al, N Engl J Med, 2015], has demonstrated that PD1/PDL1 status in RCC has not been shown to correlate with response to ICB. We unfortunately do not have RNA data for our cohort of patients, as RNA data is not included in MSK-IMPACT. We correlated many of the factors that the reviewer mentions with PBRM1 status, which is found in Figure 6. We have now added to our results the immunohistochemical (IHC) staining from the COMPARZ and McDermott cohorts which demonstrate significantly higher CD31 positive staining in PBRM1 mutated tumors, implying higher degrees of tumor angiogenesis in PBRM1 mutated tumors (Figure 6B). IHC studies from the two cohorts also reveal higher PDL1 negative and lower PDL1 positive staining tumors in PBRM1 mutated tumors compared to wild-type (Figure 6B) and no difference in CD8 positivity between PBRM1 mutant and wild-type tumors (Supplementary
7. The patients may have a heterogeneous background (if ICB was not the first line therapy) which may also contribute to the highly variable response to ICB. In addition, the agents may be different (single or combo), and number of cycles, whether received steroid therapy, whether tumors are MSI or had deficiency in the DNA repair pathways, tumor stage, etc. however, there factors were not carefully evaluated in their analysis. Even modest difference was reported in non-small cell lung cancer, it is, however, unknown whether it was a likely "true" correlation or just co-incidence. And, the mechanistic insights are missing.
-To address the comment with respect to line of therapy which the reviewer raises, overall survival and time to treatment failure was stratified by line of therapy ( Figure  4). Further, our multivariate model in Table 3

is adjusted for line of therapy. Steroid therapy is rarely administered, and even if given for ICB related toxicity, has not been shown to impact response to ICB. Data recently presented by our group at the 2020 GU ASCO meeting showed that the prevalence of MSI high tumors in RCC is exceedingly rare, with only one case out of 953 patients in MSK-IMPACT. Further, this case was of chromophobe histologic subtype, with no cases of MSI high tumors in clear cell RCC. We are currently examining the prevalence of DNA repair pathway deficiencies, which also appears to be rare in RCC. As pointed out in the discussion section of the manuscript regarding the effect of different features on outcome, we attempted to correct for this in our overall analysis survival (by correcting for IMDC status and line of therapy, knowing that the vast majority of therapy is given as 1 st line) but agree that heterogeneity is a relevant factor.
8. Immune deconvolution using bulk analysis is not convincing, given the heterogenous and dynamic feature of the tumor cells and its immune microenvironment and the complex interplay between them.
- Figure 6B). The manuscript would benefit from several additions: 1. LOF mutations of PBRM1 were only 3.9% (line 73) in the MSKCC cohort, therefore potentially only 7-8 patients total. Would the authors add to the baseline characteristics Table 2 the numbers of patients who had LOF mutations in PBRM1 and ARID2, and whether they were treated with single agent IO, IO-IO combo, or other?

We appreciate the limitations of utilizing bulk transcriptomic data to infer immune and stromal populations, however, we and other have validated the approaches used in this study (ssGSEA) through various orthogonal methods such as immunofluorescence and flow cytometry (Şenbabaoğlu et al, Genome Biology, 2016). Additionally, we have added IHC validation of CD31 for angiogenesis signature, and PDL1 and CD8 for the immune deconvolution component (
-We appreciate the reviewers' comments and suggestions. We believe that the 3.9% of LOF mutations of PBRM1 that the reviewer is referring to is in the pan-cancer cohort (n=2,936), not in the RCC cohort. We have also added to