You are here:
Profiling the ToxCast library with a pluripotent human (H9) embryonic stem cell assay
Knudsen, Thomas B., T. Zurlinden, Katerine S. Saili, J. Palmer, AND N. Baker. Profiling the ToxCast library with a pluripotent human (H9) embryonic stem cell assay. Society of Toxicology, San Antonio,TX, March 11 - 15, 2018.
The Stemina devTOX quickPredict platform (STM) is a human pluripotent H9 stem cell-based assay that predicts developmental toxicants. Using the STM model, we screened 1065 ToxCast chemicals and entered the data into the ToxCast data analysis pipeline. Presented at Society of Toxicology Meeting Mar 2018
The Stemina devTOX quickPredict platform (STM) is a human pluripotent H9 stem cell-based assay that predicts developmental toxicants. Using the STM model, we screened 1065 ToxCast chemicals and entered the data into the ToxCast data analysis pipeline. Model performance was 83.3% accuracy (sensitivity 0.78, specificity 0.92) based on 30 common reference compounds in the dataset. Of the 1065 screened chemicals, the STM model predicted 181 (17% tested) as putative developmental toxicants. STM predictivity versus animal studies was examined by a concordance model using prenatal rat and rabbit developmental toxicity studies for 146 chemicals testing positive (dLEL ≤125 mg/kg/day) or negative (no dLEL ≥ 1000 mg/kg/day) in ToxRefDB. The cell-based STM model had 76.7% accuracy (sensitivity 0.34, specificity 0.84) when compared with the in vivo results. Compound classes detected by the assay included phthalates, developmental vascular toxicants, and developmental neurotoxicants; misses included glycol ethers, perfluorinated compounds, and endothelin antagonists. The lower sensitivity of the STM model suggests certain pathways/targets may fall outside the responses of a pluripotent H9 stem cell response. To address this question, a logistic regression model was built via machine-learning to mine the strongest positive and negative correlates to 331 enzymatic and receptor signaling assays in the ToxCast NovaScreen dataset (NVS). An aggregate model using the more potent NVS-defined AC50s defined 47 positive and 38 negative correlations. Top sensitive pathways in the STM-NVS model were kinase signaling, some neuroactive GPCRs, and corticotrophs. In contrast, top negative correlations were observed with other neuroactive GPCRs (dopamine, serotonin, endothelins), estrogen signaling, and RAR antagonism. These findings point to molecular processes that potentially account for the sensitivity of the STM model and guide a battery of fit-for-purpose ToxCast assays covering potential in vivo outcomes. [This abstract may not reflect US EPA policy].
URLs/Downloads:KNUDSEN_SOT_2018_FINAL.PDF (PDF,NA pp, 108.471 KB, about PDF)
SOT_2018__V3.PDF (PDF,NA pp, 1664.276 KB, about PDF)