Deciphering the intra-tumor and inter-patient signaling heterogeneity in cancer for the rational design of patient specific drug cocktails

We developed a computational, thermodynamic-based strategy to accurately design anti-cancer personalized drug combinations or to predict cells response to drug treatments. While conventional approaches classify cancers based on characteristic biomarkers, we identify central protein nodes representing a patient-specific signaling signature (PaSSS), consisting of several distinct subnetworks, or unbalanced processes (i.e. a set of altered co-expressed, protein-protein subnetworks deviating from the steady state in each tumor).

Our in–vivo results in melanoma, breast, lung and oral cancers indeed show that PaSSSs successfully dictate the design of patient-specific drug cocktails. We demonstrate that simultaneous inhibition of central protein targets from the entire set of distinct unbalanced processes blocks the patient-specific altered signaling flux and is more effective than the treatments used in clinics.
We also extended the approach to single cells. This development allows quantifying a set of unbalanced processes in each individual cell and then mapping distinct subpopulations in every tumor without a need for large datasets, usually including multiple tissues from different patients. A high number of measured cells (>500,000) from one tumor are compared instead of many patients, thereby providing statistically significant information about different unbalanced processes within each sample. This strategy was validated in Triple Negative Breast Cancer (TNBC) wich was able to be sensitize to radiotherapy (RT). Using mouse models and patient-derived TNBC tumors we show that two subpopulations expanded in response to RT. We demonstrate that simultaneous targeting of central protein nodes representing those subpopulations, Her2 and cMet, was essential in order to sensitize TNBC to RT and stop its growth.