Prevalence and Landscape of Actionable Genomic Alterations in Renal Cell Carcinoma

A recent surge in the understanding of the molecular and genetic underpinnings of several malignancies has led to the ability to precisely target mutations and aberrant pathways in patients with cancer, thus harkening the era of “precision oncology.” Prospective clinical sequencing has etched its way into the treatment algorithms and guidelines of many cancer types, as the benefits of such sequencing efforts have significantly improved outcomes in genetically defined populations (1). The gradually increasing shift toward personalized medicine likewise highlights the disparate benefits achieved by molecular profiling of specific cancers. Although the benefits of tumor genotyping are clear in certain tumors, they are less apparent and sometimes absent in others (2–6).

Renal cell carcinoma (RCC) represents one such cancer with seemingly fewer benefits reaped from genomic innovations. Despite therapeutic advances stemming from insight into the fundamental dysregulation of the HIF1a pathway, long-term survival is not achieved for the majority of patients with metastatic disease (7). Thus, ongoing efforts to exact and refine prognostic models and treatment paradigms in RCC have progressively integrated molecular and genotypic information (8–13). Although the advent of tyrosine kinase inhibitors (TKI) and immune-checkpoint blockade (ICB) therapy represent significant therapeutic advances in RCC, there are currently no known biomarkers of response to therapy in this disease (14).

We report our institutional experience in a cohort of 753 RCCs profiled with a targeted next-generation sequencing (NGS) panel, summarizing the frequency and evidence of actionability for all identified somatic alterations that have been reported to be associated with therapy response in any cancer type. Given that the therapies in question are mostly relevant in the metastatic setting, we expand our analyses to explore different cohorts and look for associations between prevalence of these alterations and disease stage. We further leverage data from other landmark sequencing efforts to validate our findings and explore the role of these alterations during tumor evolution, with emphasis on metastatic disease.

All research-related activities were preapproved by the institutional ethics review board at Memorial Sloan Kettering Cancer Center. All the individuals who participated provided written informed consent to have their tissue profiled and their clinical data shared in a nonidentifiable manner. All activities related to this study were conducted in accordance with the principles of Good Clinical Practice and the Declaration of Helsinki. Individuals with a diagnosis of more than one cancer type were excluded from the analysis. In addition, in the RCC cohort, patients with multifocal RCCs or benign renal tumors (i.e., angiomyolipoma or oncocytoma) were excluded from the analysis.

Sequencing was performed at a Clinical Laboratory Improvement Amendments (CLIA)–certified laboratory using MSK-IMPACT, a targeted NGS panel approved by the FDA for the study of solid tumors (15). The panel captures the exons of 341 to 468 cancer genes (varies by version, Supplementary Table S1), a set of known single-nucleotide polymorphisms (SNP) tiled across the genome used for copy-number estimation, as well as the degree of microsatellite instability (MSI; via MSIsensor; ref. 16). Different thresholds have been proposed to determine the degree of MSI in tumors (e.g., >3; refs. 16, 17). To minimize false positives, we used a more restrictive threshold that has been reported in previous colorectal cancer studies. Individuals with matched-normal samples available whose tumors had MSIsensor scores ≥10 were considered to show MSI. The types of samples included in the analysis consisted of both primary and metastatic tumor specimens. For individuals with more than one specimen available, all the unique somatic events identified across all samples were considered. Individuals included were categorized as having localized (stage I–III) or metastatic disease (stage IV) according to the American Joint Committee on Cancer 8th edition. Raw data were processed and analyzed using a previously validated bioinformatics pipeline (15) that includes mutation and fusion detection as well as conventional copy-number estimation (via GISTIC 2.0; ref. 18).

Only somatic aberrations were considered in the analysis. Individual gene-level events were categorized according to their therapeutic implications using the publicly available Precision Oncology Knowledge Base (OncoKB) through its Python implementation (https://github.com/oncokb/oncokb-annotator/). This clinical support tool distills information from the published literature, projecting drug actionability based on the available clinical evidence. It is updated on a continuous basis, including emerging biomarker data for FDA-approved regimens but also those still under clinical investigation.

The OncoKB workflow has been previously described in detail (Supplementary Fig. S1; ref. 19). Briefly, alterations are first identified from public databases. Clinical therapeutic implications of alterations are then curated from several public resources; finally, OncoKB annotations are vetted by a committee of clinicians and cancer biologists across a number of disease management teams (based on current guidelines) before being made publicly available (http://oncokb.org). A level of evidence classification system was developed, and potentially actionable alterations are assigned to one of four levels based on the strength of evidence that the alteration is an FDA-recognized or standard care biomarker predictive of response to an FDA-approved drug (level 1), a standard care biomarker recommended by expert panels predictive of a response to an FDA-approved drug (level 2), has compelling clinical evidence supporting the biomarker being predictive of response to a drug (level 3A), or alterations that are considered predictive of response based on promising clinical data to FDA-approved or investigational agents being tested in clinical trials in another indication (level 3B). Only alterations with therapeutic implications corresponding to OncoKB levels of evidence 1 to 3B were considered “actionable” and included in this analysis. The OncoKB database query was performed on December 28th, 2020.

Two additional open-source knowledge databases were queried for comparison purposes (Supplementary Fig. S1C), these included the Cancer Genome Interpreter database (20) and the Clinical Interpretation of Variants in Cancer (CIViC) database (21). The data from both platforms were queried up to January 2021. All the mutations identified in RCC samples from the MSK-IMPACT cohort were used for this analysis (n variants = 1,808). We could confirm that OncoKB reliably captured all the actionable variants listed in the other databases. The only five variants not detected by OncoKB likely represent false positives; these consisted of truncating mutations in oncogenes that CIViC captured due to the nomenclature being nonspecific. In this particular case, the mutations in those genes were logged as potentially oncogenic but OncoKB does not consider truncating variants in oncogenes as such.

For a subset of tumors with available allele-specific copy-number data an in-depth genomic analysis was performed with the goal of showing potential biallelic inactivation events (Supplementary Fig. S2, “RCC copy-number cohort”). Mutations and copy-number alterations across the four genes approved by the FDA to determine mismatch repair (MMR) deficiency (i.e., MLH1, MSH2, MSH6, and PMS2) were evaluated. Samples included in this analysis had a tumor purity ≥20%, which allows for allele-specific resolution during copy-number estimation. The rest of the samples not included in the study had purity values between 10% and 20% that is suboptimal for these purposes, but still allows assessment of imbalanced events such as chromosomal gains. Allele-specific estimates allow for the integration of mutation and CN data to better detect “double-hits” in tumor suppressors. Tumors displaying heterozygous (i.e., single copy) losses in MMR genes and no evidence mutations in the remaining allele were not considered. These same genes were also investigated when replicating the analysis in the validation cohorts. The analysis included mutation and allele-specific copy-number estimation using FACETS v0.5.6 (22) through a publicly available R package (https://github.com/mskcc/facets/).

Tumor mutational burden (TMB) was defined as the number of nonsynonymous somatic mutations per DNA megabase (mut/Mb). Tumors with ≥10 mut/Mb were considered to have high TMB (TMBhi); this was based on the recent FDA approval for pembrolizumab in solid tumors (23). Because the samples included in the study were profiled using a variety of sequencing assays, the results were normalized by the total length of the panel (in Mb) for comparison purposes. The total length of the IMPACT sequencing panels used in the analysis were 0.9 Mb (IMPACT341), 1.1 (IMPACT 410), and 1.3 Mb (IMPACT468). Tumor samples profiled via exome-wide sequencing (see Materials and Methods, result validation) were normalized considering a panel length of 36.8 Mb (24).

Sequencing data from The Cancer Genome Atlas (TCGA) pan-kidney cohort (KIPAN) were used for validation. This comprises a total of 695 individuals with clear cell (n = 356), papillary (n = 274), and chromophobe RCC (n = 65) whose primary tumors were profiled using exome sequencing. Data obtained from the publicly available NCI Pancancer Atlas website (https://gdc.cancer.gov/about-data/publications/pancanatlas) were used for analysis. Although the original TCGA publications included a total of 860 individuals, only a subset of the samples was later found to have high-quality data and thus included in the Pancancer Atlas studies. Only mutation and conventional CN data were available for the TCGA. This cohort consisted exclusively of primary samples; however, 10% of the individuals had stage IV (metastatic) disease at the time of diagnosis/profiling.

In addition, sequencing data from several randomized clinical trials in patients with metastatic RCC were obtained from the original investigators and used for validation purposes; only somatic mutation data were available for these cohorts. Although all individuals had clinical evidence of metastasis, data on the sample type profiled (primary vs. metastases) were only available for some. RECORD-3 was a phase 2 clinical trial comparing the efficacy of sequential therapy with everolimus and sunitinib in metastatic RCC (25). Tumor samples from these individuals were sequenced using the IMPACT341 panel and the majority (94%) consisted of primary tumors (12). Similarly, data from COMPARZ (26), a phase 3 randomized clinical trial comparing the efficacy of sunitinib versus pazopanib in metastatic clear cell RCC were analyzed; only primary tumor samples from the individuals enrolled in the study were profiled using the IMPACT410 panel (8). Finally, we included data from IMmotion150, a more recent phase 2 trial of sunitinib versus atezolizumab versus the combination of atezolizumab plus bevacizumab in metastatic RCC (27). The tumor samples from this study were profiled using exome sequencing, but data on the type of sample used were not available for review (9).

Multiregional sequencing data from the TRACERx RENAL cohort (28) were used to assess the clonality of the most frequent targetable alterations identified. In this landmark study, multiple tumor regions from 101 patients were profiled with a targeted NGS panel that captures approximately 110 well-known RCC drivers. Only somatic mutations were considered in this analysis. The same clonality definitions from the original publication were used [i.e., mutations had to be present in every region at a cancer-cell fraction (CCF) ≥50% to be considered clonal]. CCF is a measure of relative abundance that represents the proportion of cancer cells in the specimen bearing a specific variant. Only nonsynonymous somatic mutations at or above 5% CCF were reported in the TRACERx studies and therefore considered in the analysis.

All statistical analyses were performed using the R platform v4.0.0. Results were reported as point estimates along with their 95% confidence intervals. Hypothesis tests were always two-sided and a P value below 0.05 was used to define statistical significance. Differences in proportions were evaluated using the prop.test() function that considers the χ² distribution with k−1 degrees of freedom. Fisher's exact tests and odds ratios were estimated using the fisher.test() function. Both functions are contained in the R package “stats.” Odds ratios were estimated considering the hypergeometric distribution from a given 2 × 2 table while assuming fixed marginals (i.e., conditional maximum likelihood estimate). Multivariable logistic regression analysis was performed to evaluate the factors associated with increased odds of encountering a targetable alteration or biomarker of ICB response in RCC, results were presented in the form of forest plots using the R package “forestmodel.”

After excluding individuals with multiple cancer types, the pan-cancer MSK-IMPACT cohort consisted of 37,978 patients (Supplementary Fig. S2). The final RCC cohort consisted of 753 patients, of which 15.3% (95% CI, 12.7%–17.8%) harbored at least one targetable alteration in their tumor specimen (exclusive of ICB-associated features). On the other hand, tumors such as melanoma (67.2%), breast (66.3%), and bladder carcinomas (62.9%) showed high a prevalence of actionable alterations (Fig. 1). These high actionability rates are largely driven by alterations in single genes that are recurrent in specific cancer types (e.g., BRCA2 and BRAF alterations in breast cancer and melanoma, respectively). None of the targetable alterations identified in RCC consisted of FDA-recognized biomarkers (level 1), and most of the evidence (83.5%, level 3B) originated from studies in other tumor types. The majority of individuals in the RCC cohort had tumors of clear cell histology (59.7%), followed by papillary (8.6%), and chromophobe (5.7%). Remaining subtypes, grouped as “other,” included fumarate hydratase (FH)–deficient, succinate dehydrogenase (SDH)–deficient, collecting duct, translocation-associated, and unclassified RCC. Notably, this cohort consisted mostly of patients with stage IV disease (62.8%), and 36.7% of samples originated from metastatic sites (Supplementary Fig. S3).

The prevalence of targetable alterations was compared among histologic subtypes (Fig. 2A). Although chromophobe tumors appeared to have a lower prevalence of actionable alterations, no significant differences were observed in the rate of actionability among the different subtypes (clear cell = 16.3%; 95% CI, 12.9–19.7), papillary = 15.4% (95% CI, 6.6–24.2), chromophobe = 7.0% (95% CI, 0–14.6), other = 14.7% (95% CI, 9.8–19.7), χ² d.f. = 3, P = 0.4). Papillary tumors exhibited the largest proportion of alterations with level 2 OncoKB evidence (12.3%), a result that was driven by a higher frequency of gains/amplifications involving the MET locus (Ch 7q31). In clear cell tumors, mutations were more common and the prevalence of targetable alterations were level 2 (0.9%), level 3A (1.3%), and level 3B (14.1%). Thus, not only was the overall frequency of actionable genomic alterations low across RCC but the evidence supporting their use was rather weak, with complete absence of level 1 targets (not considering alterations associated with response to ICB therapy). Regarding the genes involved (Fig. 2B), these included TSC1 (23.5%), PIK3CA (21.7%), ATM (14.8%), and MET (12.2%). Supplementary Fig. S4 and Supplementary Table S2 present a summary of the gene-level actionable alterations identified.

The prevalence of somatic alterations predictive of response to ICB therapy in other solid-tumor malignancies was assessed in the MSK-IMPACT “RCC copy-number” cohort (Fig. 3). The alterations evaluated included TMB, MSI, as well as genomic alterations in MMR genes. The median TMB was 3.1 mut/Mb and only 1.1% of individuals showed TMBhi tumors (95% CI, 0.3–1.9). Notably, only one individual with RCC showed microsatellite-instability and consisted of a chromophobe RCC tumor. It is worth noting that other studies have used more lenient definitions for tumors with MSI (16, 17). When applying these parameters to our data, we observed that an MSI score threshold of 3.0 increases the number of MSI tumors from 1 to 18 (2.4%). The prevalence of somatic MMR gene alterations was found to be 4% (95% CI, 2.5–5.5%). Of these, the majority (80.8%) consisted of biallelic inactivation events, which included homozygous deletions or a combination of mutation and heterozygous loss, whereas the rest showed only mutations. The majority of mismatch repair deficiency (dMMR) alterations were found in clear cell RCC tumors. The combined prevalence of all biomarkers of response to ICB therapy (i.e., MSI or TMBhi or dMMR) in the MSK-IMPACT cohort was 5.0% (95% CI, 3.3–6.6). Of the tumors that were found to bear a biomarker of response to ICB, the majority (75%) were due to alterations in MMR genes, and only 21.9% of patients had TMBhi tumors.

The TCGA KIPAN cohort used for validation purposes included 695 individuals with clear cell (KIRC), papillary (KIRP), and chromophobe (KICH) RCC tumors. Of note, all of these samples were obtained from primary tumors and the vast majority of the individuals (89.9%) had not developed metastatic disease at the time the TCGA analysis was reported. Notably, this cohort contained a higher frequency of papillary tumors compared with the MSK-IMPACT cohort (39.4 vs. 8.6, difference = 30.8%; 95% CI, 26.5–35.1; P < 0.001). Which could be due to differences in the criteria used to define papillary histology in the two studies (i.e., TCGA includes both papillary RCC type 1 and 2). The frequency of targetable alterations in the TCGA KIPAN cohort was found to be 9.1% (95% CI, 4.0%–7.1%), significantly lower than the MSK-IMPACT cohort (difference = 4.5%; 95% CI, 1.1–7.9; χ² d.f. = 1; P = 0.009; excluding fusions). This difference was mainly driven by clear cell RCC tumors (difference = 7.2%; 95% CI, 2.5–11.9; χ² d.f. = 1; P = 0.003). The most commonly altered genes were very similar in both studies, with the top four targetable genes being ATM, TSC1, MET, and PIK3CA, suggesting an overrepresentation of mTOR pathway alterations. We observed that the proportion of actionable alterations in this pathway (i.e., considering only TSC1, TSC2, PIK3CA or MTOR) was higher in MSK-IMPACT compared with the TCGA (difference = 20.7%; 95% CI, 4.6–36.8; χ² d.f. = 1; P = 0.01). Finally, for the TCGA cohort the frequency of alterations associated with response to ICB therapy in other cancers (i.e., dMMR/TMBhi) was 5.2% (95% CI, 3.5–6.8), nearly identical to the MSK-IMPACT cohort (difference = 0.4%; 95% CI, 2.1–2.9; χ² d.f. = 1; P = 0.9). Only 1.9% of tumors were found to be TMBhi in the TCGA cohort.

Next, we explored the association between the presence of actionable alterations and metastasis. The odds of finding at least one actionable alteration were found to be significantly higher in patients with metastatic disease (OR, 2.50; 95% CI, 1.16–6.16; Fisher's; P = 0.01). Notably, this difference was found to be due to disease stage and did not appear to be confounded by the type of sample profiled (Fig. 5A). On the contrary, no differences were observed in the prevalence of ICB-associated biomarkers (i.e., TMBhi or MSI or dMMR) between individuals with localized and metastatic disease (OR, 1.35; 95% CI, 0.46; 5.40; P = 0.8); with primary tumors and metastatic samples showing similar results (OR, 0.97; 95% CI, 0.44–2.21; P = 0.99). Mutation data from several clinical trials as well as from the TCGA and MSK-IMPACT cohorts were aggregated and a total of 1434 individuals with metastatic RCC were analyzed. Of these, 10.9% (95% CI, 9.29–12.5) showed at least one actionable alteration and 3.3% (95% CI, 2.4–4.2) were either dMMR/TMBhi. Although between-cohort differences were observed in the prevalence of actionable alterations, no differences were seen in the prevalence of dMMR/TMBhi tumors (Fig. 5B).

Multivariable logistic regression analysis was used to evaluate the association between disease stage (i.e., metastatic/stage IV vs. localized/stages I–III) and presence of actionable alterations while adjusting for histologic subtype (Supplementary Fig. S5). In the MSK-IMPACT cohort, stage IV disease was found to be significantly associated with higher odds of finding an actionable alteration (OR, 1.79; 95% CI, 1.08–3.15; P = 0.03) but not a biomarker of response to ICB therapy (OR, 1.20; 95% CI, 0.45–4.16; P = 0.7). When attempting to reproduce these analyses in the TCGA cohort, no significant associations were observed with either targetable alterations (OR, 1.42; 95% CI, 0.6–3.01; P = 0.4) nor biomarkers of ICB response (OR, 1.89; 95% CI, 0.68–4.49; P = 0.2).

Finally, we assessed the relative timing of the most frequent actionable mutations during tumor evolution using multiregional sequencing data from the TRACERx RENAL cohort. The final cohort included 96 clear cell patients with RCC, with a total of 1,146 regions profiled (Fig. 6A). Individuals with multifocal tumors were excluded from the analysis (n = 5). The frequency of targetable alterations in this cohort was found to be 18.8% (95% CI, 10.9–26.6), a result comparable with the MSK-IMPACT cohort (−4.2%; 95% CI, 13.3–4.9; χ²d.f. = 1; P = 0.4). However, in only 4.2% (95% CI, 0.2–8.2) of the tumors these mutations were found to be clonal and thus present throughout the entire tumor (Fig. 6B). All the subclonal events identified were observed in specific regions of the primary tumor and none was private to metastatic lesions.

Using a large prospective clinical sequencing database of more than 38,000 patients with any cancer type, we demonstrate a 15.3% prevalence of actionable alterations in RCC. Our analysis of the TCGA RCC cohort confirms this low frequency of targetable alterations in a separate set of 695 patients and shows a near-identical, low prevalence of biomarkers previously proposed to associate with response to ICB therapy across solid-tumor malignancies. We show that actionable alterations tend to be enriched in advanced RCC and, by using multiregional sequencing data, we demonstrate that many of these mutations tend to be subclonal.

Although current treatment paradigms for patients with RCC use targeted therapies, they are not directed against specific actionable genomic alterations. Several large-scale sequencing efforts by the TCGA and others have characterized the genomic landscape of RCC (29–31). However, other than insights in molecular diagnostics, few practical clinical applications of comprehensive genetic profiling can be extracted from these analyses and a paucity of data exists on the clinical actionability of these alterations. An earlier report of our experience with MSK-IMPACT by Carlo and colleagues (32) detailed the frequency of clinically actionable mutations across a cohort of 115 patients with non-clear cell RCC, showing 15.3% of tumors harboring clinically actionable somatic alterations. Here, we expand on the spectrum of alterations in RCC, with the goal of providing new insights into the role of genomic actionability in this context.

Results are introduced at the pan-cancer level for comparative purposes, but formal testing on the differences between cancer types was purposely not performed due to the large number of potential confounders (e.g., distribution of histologic subtypes, disease stage, disease frequency, therapeutic context, etc.). For example, it is expected that tumors whose targetable alterations correspond to cancer initiation events will show higher rates of actionability (e.g., mutations in BRCA1/2 in ovarian/breast cancer or BRAF in melanoma). Further analyses on cancer types other than RCC are outside the scope of this study. By exploring the association between actionable alterations and other clinical variables we made important observations, such as an obvious enrichment of MET amplifications in papillary RCC, which have been shown to be associated with response to crizotinib (33, 34). We also noted a higher prevalence of targetable alterations in patients with metastatic disease. However, the association with advanced stage could not be replicated in the TCGA cohort (stage I–III vs. stage IV; difference 2.9%; 95% CI, 5.6–11.5; P = 0.6) and it is unclear whether factors such as study-specific tumor selection factors, might have influenced the results (e.g., clear cell papillary tumors, a distinct and indolent molecular entity, are known to be included in the KIRC cohort; ref. 35).

The majority of the targetable alterations identified were found in four key components of the PI3K/mTOR pathway (i.e., MTOR, PIK3CA, TSC1, TSC2). These accounted for 55.6% of all targetable alterations in the MSK-IMPACT renal cohort and 34.9% in the TCGA KIPAN cohort. Randomized clinical trials showing clinical benefit of the mTOR inhibitors temsirolimus and everolimus in metastatic RCC ultimately led to regulatory agency approval (36–38). Although earlier outlier studies from our group and others identified mTOR and TSC1 as candidate biomarkers in a limited number of cases with exceptional response to rapalog therapy, these findings could not be confirmed in an unselected population (39–41). Interestingly, a significant difference was observed in the rate of actionable targetable alterations in mTOR pathway genes between the MSK-IMPACT and TCGA cohorts. These results could be due at least in part to the differences in the distribution of disease stage and histologic subtypes in the two studies, such as the overrepresentation of metastatic clear cell RCC in the MSK-IMPACT cohort.

Biomarkers of response to ICB therapy were explored separately throughout the study. Somatic mutations in the MMR genes MLH1, MSH2, MSH6, and PMS2 lead to dMMR, genetic hypermutability, and a resultant MSI phenotype (42). MSI tumors have been shown to be susceptible to the antiprogrammed death 1 immune checkpoint inhibitor pembrolizumab (43), resulting in the FDA-approval in 2017 of pembrolizumab for the treatment of MSI or dMMR solid tumors in pretreated patients (14). As further indication of a paradigm shift toward personalized medicine, this particular approval was unique in that it was the first site-agnostic anticancer therapy approved for use based on a biomarker, and it was followed by an additional approval for TMBhi tumors (23). Cohort studies ranging from 51 to 339 patients with predominantly localized RCC report a 0.7% to 5.9% prevalence of MSI tumors (44–46). These newly recognized biomarkers might represent an emerging change in paradigm for guiding therapeutic decisions in RCC, as the identification of specific genomic changes in tumors may render patients eligible for basket trials regardless of tumor type or location. To our knowledge, this analysis represents the largest MSI analysis in RCC to date, which has particular implications for immunotherapy-eligible patients in the metastatic setting.

Among the MSK-IMPACT RCC cohort, only one patient out of 753, or 0.13%, harbored an MSI tumor. Although this is a rather low value, it falls within the confidence interval of previously published estimates, which often include a 0% rate. On the other hand, we did use a strict MSI cutoff, which might be a contributing factor. Because of the rarity of this feature in RCC, there is a paucity of data regarding optimal definitions and further research is warranted to optimize these definitions for kidney cancer. In terms of the rate of TMBhi tumors, we found that only 1.1% of RCC tumors showed 10 muts/Mb, which is consistent with prior pan-cancer reports that place RCC in the lower end of the TMB spectrum (47, 48). It is important to note that the cohorts analyzed in this study were profiled using different sequencing platforms and bioinformatics pipelines. Nevertheless, a strong correlation between TMB values obtained with targeted panels and exome sequencing has previously been reported (49, 50) and TMB measurements from these assays have successfully been used to predict outcomes in the ICB therapy setting (48).

With regards to the prevalence of MMR gene alterations observed, approximately 4% of tumors ultimately harbored mutations in MLH1, MSH2, MSH6, or PMS2. However, it should be noted that IHC would be needed to confirm loss of one or more of these proteins and label these tumors as dMMR according to the recent FDA-approval. Although an attempt was made to determine the number of tumors that had “double hits” in these genes, results should be interpreted with caution as dMMR can arise from other molecular mechanisms that were not considered in the study, such as epigenetic silencing (51). When aggregating the data, we observed a prevalence of dMMR–TMBhi–MSI tumors of 5% and no differences among any of the cohorts evaluated were observed. It is important to note that only mutations were considered in the validation sets, which will likely result in an underestimation of the true prevalence at the population level. A recent retrospective analysis of DNA damage repair (DDR) gene mutations (which include MMR genes) in RCC revealed a 19% prevalence rate in 229 patients with metastatic clear cell RCC, which was associated with longer overall survival in patients receiving immune-oncology and not TKI therapies (52). However, the larger set of genes and the different DNA repair processes considered complicate the interpretation of these findings. Although further research is needed to explore the role of these alterations in ICB response prediction, our data suggest that they could be valuable in a small subset of individuals.

The rarity of targetable alterations and the unknown degree to which identified alterations respond to genomically targeted therapies likely do not justify the widespread adoption of clinical sequencing in RCC currently. Nevertheless, we and others have shown that mutations in RCC can have prognostic value even though they are limited in guiding therapy (13, 53), which could lead to new efforts in the adoption of this technology in clinical practice. As Tannock and colleagues posit, factors predicated around the cost of large-scale genomic analyses, fostering collaborative efforts to limit waste of resources, and tumor-specific factors such as intratumoral heterogeneity, must be brought to the forefront of any efforts seeking to integrate costly tests into routine care (54).

Intratumoral heterogeneity in RCC is significant, and may explain the modest responses observed with targeted antiangiogenic therapies despite near-ubiquitous VHL gene inactivation and resultant angiogenic dysregulation in clear cell RCC (31, 55). Although detailing the clinical responses to genomically targeted therapies was not the goal of the current study, we did assess the clonality of druggable alterations in the TRACERx RENAL cohort (28). On the rare instances when a targetable alteration is identified, these alterations were found to be mostly subclonal (i.e., present only in a subset of tumor cells), suggesting that single-region profiling is inadequate in capturing the full spectrum of these alterations. Notably, all the subclonal alterations identified were private to the primary tumors—a finding that if validated could have important implications in the treatment of metastatic disease. MMR genes were not sequenced and MSI data were not available on this research endeavor, limiting our ability to assess ICB-associated alterations in this cohort.

We recognize several limitations to our study. Data from MSK-IMPACT are based on the genetic profile of a single region of tumor that may potentially limit the ability to detect subclonal targetable alterations (56, 57). As a result, the true prevalence of targetable alterations in RCC might have been underestimated. This is particularly important in the metastatic disease setting, where the genomic composition of primary tumors might differ significantly from that of metastatic deposits (58). Notably, the absence of metastatic sample profiling in the TCGA cohort, together with the abundance of such specimens in the MSK-IMPACT cohort, might have affected our results. For example, a potential undersampling of targetable events in the TCGA study (which included only primaries) could help explain the differences observed in the prevalence of targetable alterations by disease stage in these two cohorts.

Nevertheless, analysis of a single tumor sample reflects current clinical sequencing efforts, and the results reported in our study are likely a reflection of the rates of targetable alterations that will be encountered in clinical practice. In addition, bioinformatic tools have inherent limitations and biases, which inevitably result in batch effects and between-cohort differences that are difficult to account for. Given the differences that exist in sequencing platforms, bioinformatic processing and patient selection between the different cohorts studied, we did not perform formal statistical testing to compare the prevalence of actionable alterations or biomarkers of ICB response in our study.

Although several precision oncology knowledge databases exist and other investigators have reported a higher targetable variant detection rate by using more than one tool to annotate variants (59), we used OncoKB in this study for various reasons. Compared with other tools that rely on an online community to distill clinical actionability data, this platform is curated by a panel of experts from a world-leading institution in cancer genomics research. Furthermore, its frequent updates and strict adherence to expert panel guidelines to categorize clinical data make it extremely useful for practical purposes. Many other knowledge databases are in early phases of development or lack sufficient information to support treatment decisions across distinct tumor types. Importantly, OncoKB is freely accessible with standardized, interpretable formatting intended for clinicians of all knowledge levels; in fact, it is currently being considered for FDA approval for use as a clinical decision-support tool. We must emphasize that the definitions of actionability of most of the alterations described in our cohort (level 3B) are based on clinical evidence extracted from other cancer types. Although the targets in question have not been specifically studied in the RCC population, these results are included for descriptive purposes and are meant to be hypothesis-generating, not to be interpreted as definitive proof of response prediction in RCC.

Finally, it is important to acknowledge that the prevalence of actionable alterations will almost certainly change over time with the recognition of new biomarkers of response and the development of novel targeted therapeutics, thus the need for an up-to-date resource that is vetted with the highest clinical standards. For example, targeting truncal molecular alterations in RCC (i.e., occurring early during tumor evolution) has been proposed as a logical strategy to treat to clear cell tumors, which are known to be in their vast majority VHL deficient (29). Exploitation of the consequential upregulation of HIF2a for therapeutic purposes is a topic of active research. In fact, belzutifan (MK-6482), a small-molecule HIF2a inhibitor, has demonstrated promising antitumor activity and safety in a phase 1/2 trial of previously treated patients with VHL-associated RCC (NCT03401788) and a randomized phase 3 trial comparing belzutifan to everolimus is currently underway (NCT04195750). Significant advances in these areas could represent a paradigm shift in the degree of actionability of RCC. However, additional research would also be needed to determine whether VHL deficiency in itself could constitute a reliable biomarker of response to these agents.

The present study provides important new insights into the landscape and clinical significance of targetable alterations in RCC, representing a first step toward the optimization of personalized cancer therapy while hopefully encouraging further research into biomarkers of response in this context. We demonstrate a substantially lower prevalence of actionable alterations in RCC. Druggable alterations identified were nearly universally subclonal, suggesting that routine clinical sequencing outside of concerted, academic efforts to be assays of low clinical yield. Although some cancers have reaped demonstrable benefits from the current genomic revolution, the same benefits have not been yet observed in kidney cancer, necessitating further efforts to identify the precise clinical role of genomic tumor profiling in this setting.

C.H. Lee reports grants and personal fees from Bristol Myers Squibb; grants and nonfinancial support from Calithera; grants, personal fees, and nonfinancial support from Eisai; grants from Eli Lilly; grants and personal fees from Exelixis, Merck, and Pfizer; personal fees from Amgen, AiCME, Intellisphere, and Research to Practice outside the submitted work. J.C. Durack reports other support from Adient Medical, Verix Medical, and Serpex Medical outside the submitted work. T.A. Chan reports other support from Gritstone Bio and grants from Illumina, Nysnobio, and Pfizer outside the submitted work. R.J. Motzer reports grants and personal fees from Pfizer, Novartis, Exelixis, Genentech; personal fees from Incyte, grants and personal fees from Roche; grants from Bristol Myers Squibb; and personal fees from AstraZeneca, EMD Serome, and Aveo outside the submitted work. M.H. Voss reports grants and personal fees from Pfizer and personal fees from Corvus, Exelixis, Eisai, Calitehrap, Merck, Aveo, and Chengdu outside the submitted work. No disclosures were reported by the other authors.

K. Attalla: Conceptualization, data curation, formal analysis, investigation, writing–original draft. R.G. DiNatale: Conceptualization, data curation, software, formal analysis, investigation, visualization, writing–review and editing. P.M. Rappold: Supervision, writing–review and editing. C.J. Fong: Data curation, software, formal analysis, methodology. F. Sanchez-Vega: Data curation, software, formal analysis, investigation. A.W. Silagy: Resources, data curation, formal analysis. S. Weng: Data curation, formal analysis. J. Coleman: Supervision, project administration. C.-H. Lee: Data curation, formal analysis, supervision. M.I. Carlo: Data curation, supervision, writing–review and editing. J.C. Durack: Resources, data curation. S.B. Solomon: Resources, data curation. V.E. Reuter: Resources, data curation. P. Russo: Resources, data curation, supervision. T.A. Chan: Supervision. R.J. Motzer: Supervision, writing–review and editing. N.D. Schultz: Supervision, writing–review and editing. E. Reznik: Software, formal analysis, methodology. M.H. Voss: Resources, supervision, writing–review and editing. A.A. Hakimi: Resources, supervision, funding acquisition, writing–review and editing.

We thank Charles Swanton, Samra Turajlic, Nicholas McGranahan, Kevin Litchfield, and all the members of the TRACERx consortium for facilitating the mutation data used in the study. We thank the Reznik and Chan Laboratory members for helpful discussions. We thank the staff and physicians of the MSK Department of Medicine Kidney Program and the Urology Service for helpful suggestions. We acknowledge the use of the Integrated Genomics Operation Core, funded by the NIH/NCI Cancer Center Support grant (CCSG, P30 CA008748), Cycle for Survival, and the Marie-Josée and Henry R. Kravis Center for Molecular Oncology. This work was in part supported by grants NIH R35 CA232097 and DOD KC180165 (to T.A. Chan), as well as the DOD Kidney Cancer Research Program W81XWH-18–1-0318 and the Kidney Cancer Association Young Investigator Award (to E. Reznik). This work was also supported by the Mellnikoff Fund (to T.A. Chan) and the Weiss Family Fund (to A.A. Hakimi and R.J. Motzer) as well as the Ruth L. Kirschstein National Research Service Award T32CA082088 (to K. Attalla).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.