Improving patient survival by direct targeting of chimeric protein-protein interaction networks

Abstract

Protein fusions produced by the "slippage" of two genes or by chromosomal translocations are essential diagnostic biomarkers of cancer. Fusions produce novel protein-protein interactions, which eliminate other interactions by changing protein domains in such fusions. The impact of these changes disseminates along protein-protein interaction networks, thereby altering cancer-promoting activity and creating cancer phenotypes. Currently, for most patients, the determination of appropriate drugs is totally empirical. As such, a personalized therapy approach based on unique patient genomic markers is needed. In this study, we considered 672 aliquot IDs containing 3091 fusions from The Cancer Genome Atlas (TCGA), accounting for 25 cancer sub-types assigned as leukemias, lymphomas, sarcomas, melanoma, glioblastoma and carcinomas. Protein-protein interaction maps showed distinct patterns according to cancer sub-type, reflected as different phenotypic traits. We induced site-directed percolations, i.e., critical transitions, by selective knockouts of genes encoding proteins in a given interaction network so as to identify breakdown points. The number of genes that needed to be knocked out in a site-directed manner before inducing a breakdown ranged from 1-7 based on whether the fusion protein was part of a network hub. Quantitatively, in leukemia, lymphoma, melanoma and glioblastoma, breakdown was achieved in 218 fusion networks when only higher degree hubs were percolated, and in 280 networks when a combination of higher and lower degree hubs was targeted, with an FDR < 0.05. These were subsequently addressed in survival studies. We found that patient survival may be improved by considering breakdown points as drug targets.