A Human Pluripotent Stem Cell-Based Assay, devTOX quickPredict, Accurately and Reproducibly Predicts the Developmental Toxicity Potential across a Diverse Set of Chemicals
Author Block: J. A. Palmer, A. M. Smith, M. R. Colwell, E. L. Donley, and R. E. Burrier. Stemina Biomarker Discovery Inc., Madison, WI.
Assessing the accuracy, reproducibility, and applicability domain of new approach methods (NAMs) is necessary step for establishing confidence in these methods and enabling their use in a regulatory setting. Over 100 chemicals have been evaluated with the devTOX quickPredict (devTOXqP) assay, which predicts the developmental toxicity potential of a chemical based on changes in human iPS cell metabolism. The assay predicted the developmental toxicity potential across this diverse set of chemicals with 87% accuracy (88% sensitivity, 86% specificity). Within individual chemical use classes (i.e., pharmaceuticals or pesticides), assay accuracy ranged from 81% to 94%, demonstrating the broad applicability of the assay. To further define the assay’s applicability domain, the results were separated into different pharmacological categories and performance was assessed. The assay’s sensitivity in these pharmacological categories ranged from 50% to 100% and provides insight into the assay’s biological applicability domain. For example, the assay had 100% sensitivity for developmental toxicants classified as channel, kinase, and transcription modulators and DNA modifiers. In contrast, receptor modulators were predicted with 50% sensitivity, and were highly dependent upon whether the iPS cells expressed the specific receptor being modulated. The reproducibility of the predictive model was evaluated using independent replicates of three chemical treatments (carbamazepine, n=34; methotrexate, n=34; thalidomide, n=9) conducted by multiple technicians with multiple iPS cell lines, freeze lots and reagents over the course of 5 years. The interpolated developmental toxicity potential (dTP) values (determined using the devTOXqP predictive model) were within two standard deviations of the mean for each of the chemicals, demonstrating that the assay endpoints are reproducible over time. These data demonstrate the importance of understanding a NAM’s biological system, its strengths and its limitations. Taken together, these data demonstrate the accuracy, reproducibility and broad applicability domain of the devTOXqP assay and support its use as an alternative to animal models for developmental toxicity testing.