We present a novel approach to structure and interpret long-term environmental monitoring data, by combining the Adverse Outcome Pathway (AOP) concept with probabilistic causal modeling. We applied this framework to a monitoring program following the 2015 tailings dam collapse in southeastern Brazil. Six years of data (2018-2024) spanning river-to-sea gradients included dissolved metal(loid)s, fish bioaccumulation, and sub-individual biomarkers. Metal(loid)s in water were first integrated into an index representing spatial variation in toxic pressure. Within the AOP framework, biomarkers of antioxidant defense, oxidative damage, and tissue injury were linked to a growth-related outcome using piecewise structural equation mixed-effects models, accounting for species-level and temporal variability. Toxic pressure was consistently higher in the river and decreased towards the sea. Biomarker responses reflected these gradients: in the river, metallothionein, lipid peroxidation, and histopathological lesions were strongly associated with toxic pressure and specific metals, whereas in marine environments, responses were weaker and more variable. Across models, species and temporal variability accounted for a substantial share of explained variance, highlighting the importance of mixed-effects approaches to unravel biological heterogeneity. Lesions were consistently linked to oxidative stress and, in the freshwater, negatively associated with fish condition index, suggesting that tissue-level damage can scale up to growth impairment. Overall, responses revealed both expected and unexpected patterns, highlighting the context-dependent nature of defense mechanisms under mixed-metal exposure and the value of integrated approaches to environmental monitoring.
