Identification of a novel prognostic gene signature in a clear cell renal cell carcinoma (ccRCC) population using an integrated multi-study single-cell RNAseq dataset.

Abstract

465 Background: Clear cell renal cell carcinoma (ccRCC) is a highly heterogeneous disease with varying prognoses and treatment responses. Understanding the underlying molecular determinants of this diversity is key to tailoring effective treatment strategies for each patient (pt). Here, we leveraged single–cell RNA sequencing (scRNA–seq) data to assess intra–tumoral diversity and its impact on pt prognosis. Methods: scRNA–seq raw data from five published studies were processed with Seurat’s standard workflow and integrated to remove biases using Harmony. We then re–analyzed the malignant cell cluster, defined by CA9, NDUFA4L2, and IGFBP3 expression, at a higher resolution. The resulting tumor cell sub–clusters were subjected to differentially expressed gene (DEG) and gene ontology (GO) enrichment analyses. A signature based on all DEGs with a fold change > 1.25 was applied on TCGA–KIRC cohort using single–sample gene set enrichment analysis and then correlated with relapse–free survival (RFS) and overall survival (OS) by Kaplan–Meier and multivariate Cox analyses. Results: We integrated scRNA–seq data from 50 samples from 44 pt’s (40% T3–4, mean age 75, 10% females). A total of 288K cells were classified into 19 clusters. The tumor cluster was re–analyzed and three biologically distinct tumor cell sub–clusters (labeled MC1, MC2, and MC3) were identified, each with unique molecular markers. GO analysis showed enrichment in genes associated with iron sequestration, oxidative phosphorylation, and apoptotic signaling, respectively. In the KIRC cohort, a 23–DEG signature from MC2 strongly correlated with RFS (HR 0.49; 95% CI 0.35−0.67 p < 0.001; 5-RFS: 74% vs. 54%) and OS (HR 0.44; 95% CI 0.33−0.60 p < 0.001; 5-OS: 75% vs. 51%) and was independent of other clinical variables in the multivariate analysis (Table 1). Moreover, this signature identified a subset of T1–T2 tumors (47.8%) with low risk of relapse (HR 0.38; 95% CI 0.19−0.78, p = 0.008; 5-RFS: 92% vs. 80%) and longer OS (HR 0.35; 95% CI 0.20−0.62 p < 0.001; 5-RFS: 90% vs. 68%). Conclusions: We have identified three distinct tumor cell sub–populations in an integrated scRNA–seq database, each characterized by unique transcriptomic profiles. A gene expression signature based on the MC2 sub–cluster was prognostic in the TCGA dataset and may help in identifying patients with a higher risk of relapse and candidates to adjuvant therapy.[Table: see text]