Proteogenomic characterization of hepatocellular carcinoma

Abstract

We performed a proteogenomic analysis of hepatocellular carcinomas (HCCs) across clinical stages and etiologies. We identified pathways differentially regulated on the genomic, transcriptomic, proteomic and phosphoproteomic levels. These pathways are involved in the organization of cellular components, cell cycle control, signaling pathways, transcriptional and translational control and metabolism. Analyses of CNA-mRNA and mRNA-protein correlations identified candidate driver genes involved in epithelial-to-mesenchymal transition, the Wnt-β-catenin pathway, transcriptional control, cholesterol biosynthesis and sphingolipid metabolism. The activity of targetable kinases aurora kinase A and CDKs was upregulated. We found that CTNNB1 mutations are associated with altered phosphorylation of proteins involved in actin filament organization, whereas TP53 mutations are associated with elevated CDK1/2/5 activity and altered phosphorylation of proteins involved in lipid and mRNA metabolism. Integrative clustering identified HCC subgroups with distinct regulation of biological processes, metabolic reprogramming and kinase activation. Our analysis provides insights into the molecular processes underlying HCCs.

Multi-omics subtyping of hepatocellular carcinoma patients using a Bayesian network mixture model

“…The BIC and AIC scores indicated = 3 as the optimal number of clusters (Fig 4A). = 3 clusters were also found as optimal for the same data in [26] and in another HCC study applying a network-based method to the TCGA HCC dataset [41]. Similar to the clusters discovered in [26] Survival analysis).…”

“…We model a cancer subtype as a Bayesian network, whose nodes represent different omics measurements of the same set of genes. The HCC dataset [26] includes five omics types, namely mutations and copy number changes (both genome, denoted and ), transcriptome ( ), proteome ( ), and phosphoproteome ( ). The edges in the network represent statistical dependencies among all observations across all omics types.…”

“…We analyzed the HCC multi-omics dataset [26] comprising 50 biopsies from 48 patients and including five omics types, namely mutations and CNAs (both genome), transcriptome, proteome, and phosphoproteome. In order to apply bnClustOmics, we first performed feature selection as follows.…”

“…In order to apply bnClustOmics, we first performed feature selection as follows. To select features, we used a list of significantly mutated genes in the analyzed cohort identified by Ng et al [26]. In addition, we included possible drivers of HCC identified in other studies [37][38][39].…”

“…Furthermore, Indovina et al [49] mention Cdk inhibitors as possible therapies which can prevent RB1 phosphorylation. Indeed, Ng et al [26] found an association of overactive CDK1/CDK2/CDK5 kinases and the phenotype associated with mutations in TP53. The central role of phosphorylation of RB1 in all networks suggests that inhibition of Cdk can be beneficial for patients in all clusters.…”