Integrating EMT dynamics in model-based metastasis prediction

Background and Objectives: Metastatic tumors are the primary causes of death for most cancer patients. Therefore, their detection and treatment is crucial for improving life expectancy in these patients. However, that requires medical imaging that incurs large expenses for any healthcare system in terms of money and workforce involved. Prediction of time of detectable metastasis is therefore of utmost importance, both from the patient’s viewpoint and from the healthcare system perspective. Methods: In this work, we have focused on epithelial-to-mesenchymal transition (EMT) as a crucial step in metastasis and transforming growth factor beta (TGF-β) as a critical regulator of this process. We present a novel mathematical modeling approach that leverages TGF-β dynamics and EMT signaling to provide distribution parameters for a model describing the growth of the primary tumor and its metastases under chemotherapy and radiotherapy treatment. Next, a virtual patients cohort is generated, in which patients are differentiated with parameters sampled from that distribution and a simulation of tumor growth and its response to the therapy is run for each patient. Simulation results, in the form of metastasis-free survival and overall survival curves, are subsequently compared to available clinical data. Results and Conclusions: As the modeling results are in concordance with clinical data, it yields two conclusions, one of clinical importance and the other important for development of similar models. It shows that it is the dynamics of how TGF-β level changes that might be more important than its absolute level. This explains why, despite known TGF-β association with metastatic processes, its value as a prognostic marker has so far been arguable. Moreover, it provides a recommendation to replace single measurements with a series of them, thus helping to increase prognosis accuracy without having to resort to expensive imaging techniques. From the modeling perspective, the approach presented here shows how to take into account patient-specific intracellular processes to generate virtual patient population, thus bringing it closer to an actual population.