Identification of Potential Cancer Therapy Target Genes through Synthetic Interaction Analyses

Human cancer cell lines are important models to study molecular mechanisms associated with tumor growth, for the identification of novel prognostic biomarkers, and the development of new therapeutics. Genome-scale metabolic models (GSMs) have been employed for studying human metabolism in a systematic manner. Double gene deletion analysis can be carried out in silico in GSMs to identify synthetic interactions in cancer models, which can in turn be compared to synthetic interactions in normal tissues. This strategy, in principle allows identification of unique sets of binary target genes that their concurrent pairwise inhibition can differentially affect the growth of cancer cells and not the healthy ones. The latest genome-scale reconstruction of human metabolic network, Recon2, describes 7,440 reactions involving 1,789 associated genes in 8 cellular compartment, and provides information on 3,168 drug interactions for 308 genes in the model.  Here, we used 16 tissue specific Recon2 GSMs that support generation of biomass to explore the effect of genic perturbations using the COBRA toolset.  In addition, gene expression data from the human carcinoma cell lines, HeLa and HCT, were generated in house through RNAseq and integrated with the Recon2 model to develop cancer-specific models. For each cell-specific model, we predicted the consequences of all possible pairwise gene deletions (nearly 2,400,000 combinations) and identified the combinations that demonstrate genetic interactions with respect to biomass reduction. Furthermore, we determined the shared synthetic interaction gene pairs between the models and those that are unique for each model. We were able to identify a large number of synthetic sick interactions in each cell-specific model ranging from 651 in smooth muscle tissue, to 6,330 in pancreatic exocrine glandular cells; however, very few lethal interactions could be identified (2 and 24 in gall bladder glandular cells and lateral ventricle neuronal cells, respectively).  Comparison of these interactions allowed identification of unique target pairs between any given pairs of cell-specific models. These identified interacting gene-pairs can be used to guide development of chemotherapeutic strategies. We plan to validate the obtained predictions through wet-bench experiments, using the available gene-drug interaction information applicable to HeLa and HCT cell lines. Further, we are identifying additional gene targets through correlated reaction set analyses to complement those identified through synthetic interaction predictions. This work was supported by New York University Abu Dhabi Research Enhancement Fund RE500 and NYU Abu Dhabi Faculty Research Fund AD060.