Subgroup-Independent Mapping of Renal Cell Carcinoma—Machine Learning Reveals Prognostic Mitochondrial Gene Signature Beyond Histopathologic Boundaries

• DOI: 10.3389/fonc.2021.621278
• Publication Date: 2021-03-15T06:09:22Z
• Authors: André Marquardt; Antonio Giovanni ...; Alexander Kerscher; Max Bittrich; Charis Kalogirou; Hubert Kübler; Andreas Rosenwald; Ralf Bargou; Philip Kollmannsb...; Bastian Schilling; Svenja Meierjohann; Markus Krebs
• Keywords: Chromophobe cell; Gene signature

Background: Renal cell carcinoma (RCC) is divided into three major histopathologic groups—clear (ccRCC), papillary (pRCC) and chromophobe RCC (chRCC). We performed a comprehensive re-analysis of publicly available datasets from the TCGA (The Cancer Genome Atlas) database, thereby combining samples all subgroups, for an exploratory transcriptome profiling subgroups. Materials Methods: used FPKM (fragments per kilobase million) files derived ccRCC, pRCC chRCC cohorts representing transcriptomic data 891 patients. Using principal component analysis, we visualized as t-SNE plot cluster detection. Clusters were characterized by machine learning, resulting gene signatures validated correlation analyses in dataset external (ICGC RECA-EU, CPTAC-3-Kidney, GSE157256). Results: Many co-clustered according to histopathology. However, substantial number clustered independently origin ( mixed subgroup )—demonstrating divergence between histopathology data. Further via learning revealed predominant mitochondrial signature—a trait previously known chRCC—across Additionally, ccRCC presented inverse angiogenesis-related genes validation cohorts. Moreover, affiliation was associated with highly significant shorter overall survival patients ccRCC—and longer Conclusions: Pan-RCC clustering RNA-sequencing distinct histology-independent strengthened weakened signatures. went along significantly research could offer therapy stratification specifically addressing metabolism such tumors its microenvironment.