Assessment of a 3D Neural Spheroid Model to Detect Pharmaceutical-Induced Neurotoxicity

Abstract

Drug-induced neurotoxicity is a leading cause of safety-related attrition for therapeutics in clinical trials, often driven by poor predictivity of preclinical in vitro and in vivo models of neurotoxicity. Over a dozen different iPSC-derived 3D spheroids have been described in recent years, but their ability to predict neurotoxicity in patients has not been evaluated nor compared with the predictive power of nonclinical species. To assess the predictive capabilities of human iPSCderived neural spheroids (microBrains), we used 84 structurally diverse pharmaceuticals with robust clinical and preclinical datasets with varying degrees of seizurogenic and neurodegenerative liability. Drug-induced changes in neural viability and phenotypic calcium bursts were assessed using 7 endpoints based on calcium oscillation profiles and cellular ATP levels. These endpoints, normalized by therapeutic exposure, were used to build logistic regression models to establish endpoint cutoffs and evaluate probability for clinical neurotoxicity. The neurotoxicity score calculated from the logistic regression model could distinguish neurotoxic from non-neurotoxic clinical molecules with a specificity as high as 93.33% and a sensitivity of 53.49%, demonstrating a very low false positive rate for the prediction of seizures, convulsions, and neurodegeneration. In contrast, nonclinical species showed a higher sensitivity (75%) but much lower specificity (30.4%). The neural spheroids demonstrated higher likelihood ratio positive and inverse likelihood ratio negative values compared with nonclinical safety studies. This assay has the potential to be used as a predictive assay to detect neurotoxicity in early drug discovery, aiding in the early identification of compounds that eventually may fail due to neurotoxicity.