Mechanistically resolved prediction of compound hepatotoxicity using primary human liver spheroids-Application to...

Drug-induced liver injury remains a major cause of clinical trial attrition and postmarketing drug withdrawals, reflecting persistent translational gaps in predictive preclinical models of compound hepatotoxicity. Here, we applied a well-characterized 3-dimensional primary human liver spheroid platform to interrogate diverse mechanisms of hepatotoxicity and benchmark performance against recent real-world clinical cases. After multicenter technology transfer, spheroids maintained stable hepatic functionality, including expression of metabolic enzymes and transporters, albumin secretion, and long-term CYP activity. Importantly, repeated-dose exposure to established hepatotoxins demonstrated robust, concentration-dependent toxicity with excellent interexperimental (R² = 0.98) and interdonor (R² = 0.89) reproducibility. To resolve toxicity mechanisms, we implemented Seahorse extracellular flux analysis in liver spheroids, which enabled sensitive detection of early mitochondrial dysfunction, allowing mechanistic discrimination between mitochondrial and nonmitochondrial toxins. Cholestatic liability was identified using bile acid coexposure with accurate classification of chlorpromazine and bosentan and confirmation of bile salt export pump downregulation. Importantly, the model also recapitulated hepatotoxicity signals observed in recent clinical development, including for Bruton's tyrosine kinase inhibitors (tolebrutinib and evobrutinib), oral glucagon-like peptide-1 receptor agonists (danuglipron vs orforglipron), synergistic toxicity on azelaprag-tirzepatide coexposure and enhanced sensitivity to the IL-17A inhibitor LY3509754 on coculture with nonparenchymal liver cells. Collectively, we present new assays for the mechanistic resolution of hepatotoxicity and provide further evidence that 3-dimensional human liver spheroids enable reproducible, mechanistically informative, and clinically aligned assessment of drug-induced liver injury across multiple modalities, which supports their application for preclinical de-risking of drug candidates. SIGNIFICANCE STATEMENT: This study shows that primary human liver spheroids provide a reproducible and mechanistically informative platform for preclinical assessments of drug metabolism and liver toxicity. The model maintains stable CYP expression and metabolic activity, captures the impact of nonparenchymal cells on hepatic clearance, and detects mitochondrial, cholestatic, and interaction-driven toxicity, including recent real-world clinical cases. These findings support 3-dimensional human liver spheroids as translational new approach methodologies for integrating metabolism and toxicity in drug development.