Artificial Intelligence Augmented Multiplexed Cardiomyopathy and Proarrhythmia in vitro Assay of Stem Cell Cardiomyocytes for Early Drug Development

Cardiotoxicity, which includes both drug-induced arrhythmias and primary structural toxicity of the cell, is a leading cause of clinical failures in drug development. Cardiotoxicity is dose-limiting for many anti-cancer therapeutics, so improving safety would improve efficacy. The number of cancer survivors may exceed 22.1 million by 2030 and their risk of cardiovascular death may soon exceed that of the cancer itself (K. Miller, et al., 2019). A recent FDA-authored review reported that there are no effective in vitro tools to predict myopathies (X. Yang, T. Papoain, 2018).  Animal studies, which are low throughput and costly, have historically detected only ~70% of human-relevant toxicities (H. Olson, et al., 2000). There is thus an unmet need for development of in vitro cardiotoxicity assays to aid selection of uHTS hits and inform lead development.

In vitro screens of hiPSC-CMs are commonly based on whole-well or whole-field averaged data, but averaging can both obscure and illicit artifacts of important waveforms like triangulation and afterdepolarizations. By measuring each CM individually, errors produced by, e.g., propagation and asynchronous contractions are largely eliminated to improve sensitivity. Our single-cell, kinetic image cytometry (KIC), approach was previously shown to predict arrhythmias of known drugs with about 90% accuracy in two studies of 60 and 125 compounds (H. Lu, et al., 2015; E. Pfeiffer, et al., 2016).

Here we report on progress developing a cardiotoxicity assay combining readouts of mitochondrial and endoplasmic reticulum (ER) health with kinetics of voltage, Ca²⁺ transients, and contraction, all quantified in a single-cell fashion on human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes (CMs). Accuracy of detection of arrhythmic single-cell and proarrhythmic single-transient waveforms, including early and delayed afterdepolarizations (EADs and DADs), ventricular tachycardia (VT), and nonsustained VT (NSVT), will also be improved by artificial intelligence. In addition to the increased efficiency of multiplexing myopathic and proarrhythmia assays, the underlying mechanisms may overlap (e.g., tyrosine kinase inhibitor cancer therapeutics can exhibit one or both proarrhythmia and myopathy risks, A. Sharma, et al., 2017). By multiplexing readouts of multiple subcellular systems, this approach may also guide experiments to locate the specific target(s) more quickly. This high throughput cardiotoxicity assay thus has the potential to better predict agents that may cause myopathy and arrhythmia and may also predict primary multiple subcellular system in which the target resides.