Proteomics-based system biology analyses unravel a functional structure with prognostic value

Abstract

Background: Urothelial cancer has been traditionally classified based on histology features. Recently, some works have proposed a molecular classification of muscle-invasive urothelial carcinoma (MIUC) into basal and luminal subtypes.We aimed to define molecular subtypes of MIUC and evaluate the status of several biological processes in the tumor tissue and address its clinical value. Methods: Tissue samples were obtained from 57 pts who underwent curative surgical resection at "Universitary Hospital 12 Octubre" between 2006/12.We analyzed the proteome applying a high-throughput proteomics approach to routinely archive FFPE tumor tissue. Tryptic digests were analyzed by mass spectrometry for protein identification using a Q-Exactive mass spectrometer. Subgroups were defined by hierarchical clustering and random forest. Functional structure was developed using probabilistic graphical models with local minimum Bayesian Information Criterion and Gene Ontology Analysis. Data analysis was done using MeV, BRBArray Tools, R and Cytoscape software suites and Uniprot (http://www.uniprot.org/) and DAVID (http://david.abcc.ncifcrf.gov) webtools. Results:We identified two different molecular subgroups with differential prognosis. Systems biology analyses showed that wide protein expression assessment allows building a functional structure where several nodes with defined biological activity were defined. Activity measurement for each node showed differences between two subtypes in metabolism, focal adhesion, RNA and splicing nodes. Subtypes defined by protein expression are comparable to basal and luminal subtypes defined by gene expression. Moreover, the focal adhesion node has prognostic value in the whole population, and this prognostic information is independent from a predefined prognostic signature (submitted Abstract: Proteomics profile profiling predicts poor prognosis in patients with muscle invasive urothelial carcinoma). Conclusions: Protein data analysis using random forest showed subgroups matching with basal and luminal subtypes obtained by hierarchical cluster analysis. Importantly, we were able to establish different nodes according to biological functions, with diagnostic and prognostic value.