Abstract

Most in vitro chemical safety workflows are performed in ambient air, although many human tissues experience substantially lower physiological oxygen levels. This mismatch can alter redox tone, stress-response signaling, and interpretation of chemical effects. Here, we present R3ACT (Redox & Antioxidant Capacity Tracker), an animal-free, oxygen-controlled live-cell NAM for mechanistic profiling of redox perturbation and antioxidant-response activation. In human HaCaT keratinocytes adapted to 5 kPa O₂, R3ACT combines viability-gated concentration selection, oxygen-controlled plate-reader measurements, high-content microscopy, and deep-learning-assisted single-cell analysis. The workflow integrates cytosolic and mitochondrial HyPer7.2 biosensors for H₂O₂ dynamics with POINTER, an Nrf2 transcriptional reporter, to assess antioxidant-response engagement. As proof of concept, four exposure-relevant materials were profiled in the main screen, with iron formulations assessed as additional pharmaceutical case examples. Zinc oxide produced the strongest reproducible Nrf2 activation across two independent runs, confirmed by single-cell microscopy. Polystyrene nanoplastics likewise showed reproducible, microscopy-confirmed Nrf2 activation, whereas perfluorooctanoic acid showed a borderline response that was not confirmed by microscopy. Salicylic acid induced Nrf2 activation under physiological oxygen and was interpreted as antioxidant-response engagement rather than skin-sensitization classification. HyPer7.2 provided compartment-specific mechanistic context, but often revealed transient fluctuations rather than consistent material-induced H₂O₂ increases. Together, these data position R3ACT as an early-stage, oxygen-controlled NAM that may support animal-free chemical and drug-screening workflows by generating mechanistic redox/Nrf2 response profiles for compound prioritization. Further benchmarking, predefined performance criteria, and inter-laboratory studies are required before regulatory applicability can be assessed.