In Silico Populations Optimized on Optogenetic Recordings Predict Drug Effects in Human-Induced Pluripotent Stem Cell Derived Cardiomyocytes

Michelangelo Paci1, Elisa Passini2, Aleksandra Klimas3, Stefano Severi4, Jari Hyttinen1, Blanca Rodriguez2, Emilia Entcheva3
1Tampere University of Technology, BioMediTech, 2University of Oxford, 3The George Washington University, 4University of Bologna


Optical approaches offer contact-free high-resolution measurements of key electromechanical parameters in cardiomyocytes, e.g. action potentials (AP), Ca2+ transients (CaTr), or contraction. Recently, all-optical high-throughput systems allowed simultaneous AP and CaTr measurements from cardiomyocytes within multicellular context, offering means to speed up in-vitro drug tests. We aim to demonstrate the utility of optically-obtained data for in-silico modeling of human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) and their response to drugs, by developing hiPSC-CM populations, which predict drug effects based on optically-measured cell-level data. The experimental dataset contains three plates of hiPSC-CM syncytia at room temperature, used for simultaneous AP and CaTr recording under non-paced conditions. Four incremental doses of Diltiazem (ICaL blocker), Cisapride and Astemizole (IKr blockers) were administered; 0.1% DMSO served as negative control and 0.5nM Dofetilide as positive control. We challenged the latest Paci hiPSC-CM model simulating hiPSC-CM responses to drugs in these conditions. Firstly, we modified the model kinetics to adapt our simulations to room temperature. Then, for each plate we generated a model population using multi-objective genetic algorithms (MoGAs, max 200 models, 30-40 generations), constrained by AP and CaTr experimental biomarkers from control hiPSC-CMs only. Finally, we administered to each in-silico population Dofetilide and its plate-specific drug (based on IC50s for IKr, ICaL and INa). MoGAs produced three control populations (130, 200, 200 models) fully in agreement with the control biomarkers. For all the tested drugs, we obtained agreement with experiments: Diltiazem shortened APs (mean values, simulations:-52% vs experiments:-19%, at dose 3) while Astemizole and Cisapride prolonged APs (+163% vs +40% and +137% vs+84%, at dose 4, respectively). In conclusion, we showed that control all-optical experimental data can be used to optimize in-silico hiPSC-CM populations without individual ion channel characterization, and that these populations can then predict hiPSC-CM responses to drugs, not used for the parameter optimization.