Scoring and ranking of metabolic trees to computationally prioritize chemicals
Scoring and ranking of metabolic trees to computationally prioritize chemicals for testing using fit-for-purpose in vitro estrogen receptor assay
Oak Ridge Institute for Science and Education, Oak Ridge, TN
National Exposure Research Laboratory, U.S. EPA, Durham, NC
Increasing awareness about endocrine disrupting chemicals (EDCs) in the environment has driven concern about their potential impact on human health and wildlife. Tens of thousands of natural and synthetic xenobiotics are presently in commerce with little to no toxicity data and therefore uncertainty about their impact on estrogen receptor (ER) signaling pathways and other toxicity endpoints. As such, there is a need for strategies that make use of available data to prioritize chemicals for testing. One of the major achievements within the EPA’s Endocrine Disruptor Screening Program (EDSP), was the network model combining 18 ER in vitro assays from ToxCast to predict in vivo estrogenic activity. This model overcomes the limitations of single in vitro assays at different steps of the ER pathway. However, it lacks many relevant features required to estimate safe exposure levels and the composite assays do not consider the complex metabolic processes that might produce bioactive entities in a living system. This problem is typically addressed using in vivo assays. The aim of this work is to design an in silico and in vitro approach to prioritize compounds and perform a quantitative safety assessment. To this end, we pursue a tiered approach taking into account bioactivity and bioavailability of chemicals and their metabolites using a human uterine epithelial cell (Ishikawa)-based assay. This biologically relevant fit-for-purpose assay was designed to quantitatively recapitulate in vivo human response and establish a margin of safety. In order to overcome the overwhelming number of metabolites to test, a prioritization workflow was developed based on ToxCast chemicals (1677) and their predicted metabolites (15,406). A scoring function was used to rank the metabolic trees of the considered chemicals combining in vitro data from ToxCast and the literature in addition to in silico data from the Collaborative Estrogen Receptor Activity Prediction Project (CERAPP) consensus model and five of its single QSAR models. The bioavailability of the parent chemicals as well as the metabolites and their structures were predicted using ChemAxon metabolizer software. The designed workflow categorized the metabolic trees into true positives, true negatives, false positives and false negatives. The final output was a top priority list of 345 ranked chemicals and related metabolites from the ToxCast library as well as an additional list of 593 purchasable chemicals with known CASRNs. We are currently moving forward to test the highest-priority metabolic trees in the Ishikawa assay and are using a liver bioreactor to confirm important metabolites.
Disclaimer: The views expressed in this presentation are those of the authors and do not necessarily represent the views or policies of the U.S. EPA.