A Machine Vision-Guided Microphysiological Platform With Automated Microfluidics Enables Longitudinal Biomarker Monitoring and Emulation of Translationally Relevant Exposure Scenarios

Low success rates in clinical drug development can be largely attributed to the poor predictive power of existing preclinical models. Microphysiological systems (MPS) have greatly advanced in vitro modeling; however, current platforms do not adequately support long-term sampling and often fail to recapitulate nutrient and drug exposure dynamics. To address these limitations, we established a machine vision-guided MPS with real-time fluidic control that enables fully automated periodic sampling with high temporal resolution, media replenishment, and programmable dosing, allowing for the simulation of dynamic nutritional or pharmacological exposure scenarios. We showcase the system's capability by emulating physiological insulin profiles and repeated-dose pharmacokinetic exposures over multiple weeks. Furthermore, pharmacokinetically accurate acetaminophen exposure in 3D primary human liver spheroids mimicking an acute overdose rapidly induced liver toxicity, as evidenced by aminotransferase release, cytokine secretion and a drop in cellular ATP. In contrast, dose-equivalent constant exposure patterns did not elicit detectable hepatotoxicity. Mechanistically, targeted proteomics of sampled supernatants and Cell Painting revealed that toxicity was paralleled by disrupted lipid homeostasis, loss of tight junctions and extracellular matrix remodeling. These results demonstrate the robustness and versatility of the machine vision-guided automated microphysiological platform and underscore the importance of incorporating drug exposure dynamics for mechanistic toxicology.