Marcella Brescia, James Gallant, Andrea Chatrian, Paul Keselman, Elsa Sorman Paulsson, Mervyn P.H. Mol, Rickard Sjögren, Karine Raymond, Valeria V. Orlova, Kalpana Barnes, Richard Wales, Jonas Austerjost, Michael W. Olszowy, C. Mummery, Berend J. van Meer, Richard P. Davis
Cardiotoxicity is a significant challenge in cancer therapies, particularly with doxorubicin, a widely used anthracycline. More predictive in vitro models are needed to understand doxorubicin-induced cardiac damage and patient-specific responses. Here, human pluripotent stem cell (hPSC)-derived cardiomyocytes (hPSC-CMs), cardiac fibroblasts (hPSC-cFBs), and endothelial cells (hPSC-ECs) were cultured in mono- or multi-cell-type formats and repeatedly treated with doxorubicin to mimic cumulative clinical exposure. A machine learning-based tool enabled continuous quantification of the early toxicity marker caspase-3/7 and accurately identified hPSC-CMs within mixed cultures. Notably, hPSC-ECs were more sensitive to doxorubicin than hPSC-CMs or hPSC-cFBs, with nitric oxide signaling contributing to the elevated cardiomyocyte toxicity observed in co-culture. These results question the conventional in vitro focus on cardiomyocytes regarding drug-induced cardiac damage, highlighting the interplay among different cardiac cell types in mediating the toxic effects of doxorubicin. Furthermore, the work demonstrates the potential of AI-based tools to provide scalable strategies for assessing drug-induced cardiotoxicity.