Genomic and proteomic predictors of sites of metastases in renal cell carcinoma.

Abstract

4538 Background: Among patients with renal cell carcinoma (RCC), the most common sites of metastasis are lung, lymph nodes, and bone. While some sites of metastases are associated with better cancer-specific outcomes than others, the underlying biology of metastatic organ tropism is not well understood. We performed genomic and proteomic analyses to investigate the biological underpinnings of different metastatic sites in RCC. Methods: Institutional cohorts of patients with metastatic RCC from the Dana-Farber Cancer Institute (DFCI) were analyzed using a next-generation tumor somatic mutation assay (n = 633) and with a highly multiplexed plasma proteomics assay (n = 258). Data were clinically annotated for sites of RCC metastasis. Genomic analyses were performed using a two-sided Fisher’s exact test on the cBioPortal platform at DFCI with pairwise comparison of patients with versus without metastases to lung, liver, brain, bone, adrenal, and lymph nodes. The Benjamini-Hochberg method was applied for FDR-adjusted q-values. Exploratory proteomic analyses were performed using logistic regression for each metastatic site with multivariate adjustment for other sites of metastasis. For each metastatic site, the top five associated proteins were selected to build a multivariate model to predict the presence of each metastatic site. Bootstrapping with R = 1,000 was employed for the assessment of model performance. Results: Tumor genomic alterations in SETD2 (q-value = 0.004) and CDKN2A (q-value = 0.04) were associated with lung and lymph node metastases, respectively. Logistic regression analyses of proteomic data were used to identify circulating proteins with the strongest associations with the sites of metastases. For instance, circulating collagen alpha-1(IX) chain (CO91A) and relaxin receptor 1 (RXFP1) were the top circulating proteins associated with bone metastases, while GGT2 and tenascin were associated with liver metastases and matrilysin (MMP-7) was associated with lymph node metastases. Multivariate models using the top five proteins to predict the presence of each metastatic site demonstrated bootstrapped C-statistics from 0.72 to 0.80 for lymph nodes, lung, adrenal, brain, liver, and bone, respectively. Conclusions: We identified genomic and proteomic predictors of organ-tropic metastases in RCC. Next, we will validate these findings in independent external cohorts.