ABSTRACT: Integrated Approaches to Testing and Assessment (IATA) are emerging as a means to strategically assess and test chemicals for toxicity, using existing information, and helping to identify potential data gaps. Adverse Outcome Pathways (AOPs) have recently been shown to be a good starting point for IATA development. Our interest in IATAs and AOP Networks (AOPNs) focuses on developing efficient testing and risk assessment strategies for new materials, new chemical entities, and setting environmental cleanup goals. We combine the maths logic of causality, specifically the notion of sufficiency to infer an adverse outcome, with AOPNs to develop Minimally Sufficient Sets of Key Events (MinSSKEs). These MinSSKEs represent a set of key events (can just be one key event) that taken together allow assessors to predict if an adverse outcome is likely or not. For our case study we use the military explosive trinitrotoluene (TNT). TNT is a relatively data rich chemical where the US Environmental Protection Agency has set a reference dose of 0.0005mg/kg-day, based on a point of departure for liver toxicity in dogs of 0.5mg/kg-day (a lowest observed adverse effect level; LOAEL) with uncertainty of 1,000X. Our IATA measures gene expression for genes in involved in our AOPN for energy metabolism, steatosis and oxidative stress in human induced pluripotent stem cells differentiated to hepatocytes exposed to TNT. We used a Bayesian statistical approach to identify differentially expressed genes, and a nonlinear data-driven dose-response modeling approach (Good Risk Assessment Values for Environmental Exposures; GRAVEE) to estimate points of departure for reference dose estimation. We obtained an RfD of 0.12mg/kg-day, with a 90% credible interval spanning from 0.02-0.19mg/kg-day using our IATA approach. Our RfD is based on a POD of 1.22mg/kg-day with a 90% credible interval spanning from 0.24-1.94mg/kg-day, and uncertainty of 10X after applying the IRIS uncertainty factors. Our PODs are highly comparable, with a difference of only 0.72mg/kg-day (or 2.44x). Our RfDs differ by 240x, owing to the fact that we have only a 10x uncertainty factor for human variability, whereas the IRIS assessment uses 1000x uncertainty for 10x animal-to-human extrapolation, 10x LOAEL-to-NOAEL extrapolation, and 10x for subchronic-to-chronic extrapolation. In this case study, our IATA resulted in a highly similar POD compared to the IRIS assessment, and an overall decrease in the uncertainty factors that needed to be applied. Our IATA approach worked well in this instance, and demonstrated several advantages by focusing on using a human-relevant in vitro model, data-driven POD modeling, and a net decrease in the overall risk assessment uncertainty factors, while maintaining the uncertainty associated with the data.
