After obtaining institutional review board approval, we retrospectively identified 91 patients with mRCC who had undergone comprehensive genomic profiling as part of routine clinical care at the City of Hope Comprehensive Cancer Center (Duarte, California, USA) between July 2018 and September 2019. Figure 1 depicts the flow diagram of patient inclusion for cohort analysis. Patients’ demographics and clinical variables including IMDC risk score criteria, treatment type, treatment response and survival outcomes were obtained by reviewing the patients’ electronic medical records and abstracted into a coded database deficient of any patient health identifiers to ensure privacy. Patients who received either an immunotherapy or a VEGF-TKI regimen for treatment of metastatic disease and had received genomic profiling of their tumor were deemed eligible. Treatment response was evaluated via RECIST 1.1 (Response Evaluation Criteria In Solid Tumors) criteria. Patients with a best response of complete response (CR) or partial response (PR) regardless of duration or patients with stable disease (SD) for a duration of at least 6 months were considered to have experienced CB. Patients with progressive disease as the best response to treatment were considered to have no clinical benefit (NCB). We only included patients whose genomic profiling was performed prior to initiation of a systemic treatment, had a follow-up duration of more than 6 months, and treatment responses were evaluable at the time of data cut-off.

Genomic profiling was performed via the GEM ExTra assay in a College of American Pathologists (CAP)-accredited, Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory through Ashion Analytics (Phoenix, Arizona, USA) and encompassed comprehensive DNA (paired tumor-normal whole exome sequencing (WES)) and RNA (tumor whole transcriptome sequencing) next-generation sequencing. Tumor whole exome and transcriptome sequencing was performed on tumor samples obtained as part of routine diagnostic clinical care. Paired normal WES was performed using peripheral blood monocytes. Tumor-normal WES data were available for 58 patients in the analysis cohort, of which 49 patients also had matched tumor whole transcriptome data available. The mean target coverage for exome sequencing was 402X for tumor and 258X for normal. RNA sequencing averaged 187 million aligned reads.

Briefly, following analyte extraction with the Qiagen AllPrep kit and DNA sheering, exome libraries were generated with the KAPA HyperPrep Library Kit (Roche) and captured with a custom IDT xGen probe set targeting 19,396 genes and intronic regions containing breakpoints for 712 known translocation events as well as providing coverage for hotspot mutations within the TERT promoter. RNA libraries were prepared following ribosomal RNA depletion and whole transcriptome random primed reverse transcription (KAPA RNA HyperPrep with Riboerase kit, Roche). Libraries were sequenced using Illumina NovaSeq 6000 reagents and instruments.

Following BCL to FASTQ conversion with bcl2fastq, data were aligned to build 37 of the human reference genomes using BWA-MEM followed by duplicate removal and sorting. Somatic single-nucleotide variants and small indels were identified using Freebayes (Garrison, Erik & Marth, Gabor 2012 arXiv 1207), filtering for variants detected on both strands, a tumor allele frequency 5% or greater, a normal allele frequency less than or equal to 3%, and a tumor:normal allele frequency ratio of 10 or greater. A 2×2 contingency table of reference and alternate alleles in tumor and normal samples was constructed for the remaining variants, and Fisher’s exact test and the Benjamin and Hochberg procedure were used to call somatic variants, targeting a false discovery rate of <1%. Structural variants were detected using Manta, and amplifications and deletions were called using a custom in-house algorithm calculated on a per gene basis based on read count ratios in tumor versus normal samples (amplified: log2 ratio >1; deleted: log2 ratio <−1.35).12 Tumor mutation burden was calculated using a custom in-house algorithm, counting all somatic, protein coding variants predicted to change the amino sequence of the impacted protein in the SnpEff-annotated Freebayes VCF file.

For whole transcriptome analysis, FASTQ files were aligned using STAR. Ashion analysis included fusion detection using STAR-Fusion and FusionInspector, and detection of clinically actionable alternative transcripts (ie, EGFRvIII, ARv7) using a custom algorithm based on detection of unique 32 nucleotide sequences to identify these alternative transcripts. Global transcript quantification and differential expression analysis were performed in the research setting. Briefly, Salmon (V.0.14.1) was run in the quasi-mapping mode using the GRCh38 build to estimate transcript abundance, quantified in transcripts per kilobase million reads values and leveraging transcript IDs to gene names to produce gene level abundance estimates.13 Gene level count files were processed using DESeq2 (Bioconductor).

A Student’s t-test was used to evaluate the association between tumor mutational burden (TMB) and clinical benefit. A two-tailed Fisher’s exact test was used to assess enrichment of genomic alterations (categorical value, gene mutated or non-mutated) and clinical benefit (CB, NCB). P values <0.05 were considered significant. Analysis was performed using GraphPad Prism V.8.3.0.

Figure 1 outlines the inclusion criteria for the analysis cohorts. Of the 91 patients with mRCC that received genomic profiling as part of standard clinical care, 58 patients were included in the analysis. A total of 17 patients received both immunotherapy and VEGF-TKI therapy sequentially. This resulted in 43 patients in the VEGF-TKI therapy and 32 patients in the immunotherapy-treated cohorts, respectively.

Patient characteristics based on treatment are presented in table 1. Median age was 63.2 (range 31.6–84.0), and 43 (74%) patients were male. Clear cell histology was observed in 47 (81%) patients, while 11 (19%) patients had non-clear cell RCC. Non-clear cell RCC subtypes included six patients with papillary RCC, three patients with chromophobe RCC and two patients with sarcomatoid RCC. The majority (77%) of patients in the VEGF-TKI therapy cohort received treatment in the first-line setting. The most commonly used VEGF-TKI therapies were sunitinib (n=17, 40%), cabozantinib (n=9, 21%) and the combination of lenvatinib/everolimus (n=6, 14%). One-third (n=11, 34%) of the patients included in the immunotherapy cohort received nivolumab/ipilimumab combination in the first-line setting. The remaining patients in the immunotherapy cohort received nivolumab in the second-line or third-line setting.

In the VEGF-TKI therapy cohort, 2 (5%) patients achieved a CR, 7 (16%) achieved a PR and 28 (65%) patients achieved SD, of which 25 (58%) maintained SD for over 6 months. In the immunotherapy cohort, the best response was CR in 1 (3%) patient, PR in 6 (19%) patients, and SD in 12 (37%) patients, of which 10 (31%) maintained SD for over 6 months. CB rate was 79% in the VEGF-TKI therapy cohort and 53% in the immunotherapy cohort. After a median follow-up of 14.2 months (95% CI 9.5 to 18.9), median progression-free survival was 14.2 (95% CI 9.0 to 18.5) months in the VEGF-TKI therapy cohort and 15.6 (95% CI NR to NR) months in the immunotherapy cohort.

Clinical tumor-normal WES of the overall cohort yielded frequent genomic alterations in the VHL (64%), PBRM1 (38%), SETD2 (24%), KDM5C (17%), MTOR (12%) and TERT (12%) genes. Figure 2 represents the genomic landscape of the study population along with treatment information and CB outcomes. Notably, TERT promoter mutations were found to be mutually exclusive of PBRM1 mutations.

Median TMB of the overall cohort was 1.2 mutations/Mb (range 0.03–4.0). Box plots demonstrating the comparison of TMB between CB and NCB patients in immunotherapy and VEGF-TKI therapy cohorts can be appreciated in figure 3. No statistical significance was observed between CB and NCB patients in either cohort.

In the immunotherapy cohort, PBRM1 loss-of-function mutations were more frequent in patients with CB, whereas VHL, SETD2, KDM5C, CACNA1D, and TP53 alterations were more frequent in patients with NCB (figure 4A). However, the differences did not reach statistical significance (p values>0.05). TERT promoter mutations were found to be enriched in patients with NCB in the immunotherapy-treated cohort (p=0.038). Moreover, none of the patients with TERT promoter mutated tumors obtained CB from immunotherapies (figures 2 and 4A). In the VEGF-TKI therapy-treated cohort, no single genes were found to be associated with CB (figure 4B).

Transcriptional analysis was then carried out in the IO cohort to determine if there were any significant gene expression signature changes in the TERT promoter-mutated samples that were correlated with NCB compared with the TERT promoter wild-type samples. Of the 32 patient samples who were initially included in the IO cohort for the above-mentioned WES analysis, 28 had RNA-seq data that met the quality control measures of the GEM Extra assay. Of these samples, four (14%) were from TERT promoter mutated with NCB and the remainder were TERT promoter wild type (n=24, 86%). The top differentially expressed genes with higher levels of enrichment in the TERT promoter-mutated group included SST, CYSLTR2, WNK2, and PTGES (online supplementary table S1). Overall, 135 genes were found to be significantly more enriched in the TERT-mutated samples compared with the wild type. Downstream gene set enrichment and pathway analysis with ENRICHR and X2kweb analysis further identified MYC and KAT2A as enriched transcription factor pathways within TERT promoter mutant tumors (online supplementary figure S1). The X2Kweb pathway analysis found that the top enriched kinase pathways were CSNK2A1, CDK1, MAPK14, ATM, and CDK4 (online supplementary figure S2). The final step of the X2K network analysis for the inferred upstream regulatory network integrating results from the transcription factor and kinase pathway analysis with expansion of further predicted targets found that the key kinase nodes were GSK3beta, ATM, CDK4, MAPK14, DNAAPk, and CK2alpha (figure 5).