Validation, Acceptance, and Extension of a Predictive Model of Rodent Reproductive Toxicity Using Toxcast High-Throughput Screening.

Abstract

The EPA ToxCast research program uses a high-throughput screening (HTS) approach for predicting the toxicity of large numbers of chemicals. Phase-I tested 309 well-characterized chemicals (mostly pesticides) in over 500 assays of different molecular targets, cellular responses and cell-states. Phase-II is testing another roughly 700 chemicals and, in addition to the original assay set, will be testing the full chemical library in new assay technologies. Using ToxCast Phase-I data, a predictive model of reproductive toxicity was generated capable of externally predicting rodent reproductive toxicity, as observed in rat multigeneration toxicity tests, with over 75% accuracy. New ToxCast Phase-II chemicals and assays are being used to further validate the model, identify strengths and weaknesses, and, where appropriate, update the model with additional assays covering existing molecular targets in the model or new targets associated with reproductive toxicity. Of the chemicals in Phase II, roughly 100, have been tested in high quality reproductive tests deriving the validation chemical set, which includes pesticides, industrial chemicals, food additives, and pharmaceuticals. Using preliminary ToxCast Phase II assay data on the external validation chemical set, the model accuracy remains above 75%. An example of model extension is the incorporation of new assay data including a cell proliferation assay testing for estrogenicity. The final resulting model is a classification model sufficient for large-scale chemical testing prioritization and begins to infer modes of action, e.g., estrogenicity, anti-androgenecity, or altering steroid metabolism. The model also has transparent inputs and outputs that are biologically translatable improving the chances for the model’s acceptance and use in chemical testing prioritization and targeted testing applications. Extension of the model toward risk assessment applications requires incorporation of toxicokinetic and cell-level information where quantitative and dynamic outputs of dose and time can be placed into a systems modeling framework. This abstract does not necessarily reflect U.S. EPA policy. (poster)