You are here:
Transcriptomics and cell morphology screens complement each other when grouping chemicals with similar bioactivity
Citation:
Rogers, J., J. Bundy, J. Harrill, R. Judson, L. Everett, AND K. Friedman. Transcriptomics and cell morphology screens complement each other when grouping chemicals with similar bioactivity. Society of Toxicology (SOT) 63rd Annual Meeting and ToxExpo, Salt Lake City, UT, March 10 - 14, 2024. https://doi.org/10.23645/epacomptox.25521205
Impact/Purpose:
Presentation to the Society of Toxicology (SOT) 63rd Annual Meeting and ToxExpo March 2024
Description:
Background and Purpose: A tiered assessment strategy for chemical hazard with integration of multiple data streams is needed for next generation risk assessment with new approach methods (NAMs). Combining information from multiple high content assays such as high-throughput transcriptomics (HTTr) and high-throughput phenotypic profiling (HTPP) may improve confidence when assessing key hazards by providing comprehensive evidence of chemical-induced gene expression and morphological changes. However, further investigation is needed to determine whether HTTr and HTPP data streams capture distinct mechanisms-of-action, and if integrating readouts from both data streams can capture additional effects not observed in either assay alone. Here, unsupervised computational approaches are applied to paired HTTr and HTPP chemical screening data to determine if unique patterns in bioactivity are observed between data streams. Methods: Data from a screen of 1,201 chemicals using both HTTr and HTPP were used for all analyses, in which U-2 OS cell lines were exposed to chemicals in 8-point concentration response, and gene expression or morphological feature measurements were extracted using published bioinformatic pipelines. Concentration-response modeling software was used to estimate Benchmark doses (BMDs) and efficacy for each chemical across 9265 genes and 520 morphological features. Signed, scaled area-under-the-curve (ssAUC) values were then derived as estimates of total bioactivity for each curve. To identify and compare patterns in bioactivity profiles between data streams, cosine similarity values between all chemical pairs were calculated using ssAUC profiles across features from either HTTr, HTPP, or both data streams combined, and density-based unsupervised clustering of chemicals was repeated for the three resulting similarity matrices. To assess whether patterns in bioactivity were indicative of targeted molecular effects or structural features of clustered chemicals, cluster membership based on HTTr, HTPP, or both data streams were tested for enrichment with expert-curated annotations derived from the ClassyFire structural ontology or previously observed target-specific data from US EPA’s ToxCast program. Results: 493 of 1,201 chemicals met criteria for bioactivity in both platforms (at least 10 concentration-responsive genes and morphological features) whereas 419 chemicals met these criteria in HTTr or HTPP only (144 and 275 chemicals respectively). From the 493 chemicals with sufficient bioactivity in both platforms, 5 clusters representing 84 chemicals were identified from HTTr profiles alone and 8 clusters representing 88 chemicals were detected from HTPP profiles alone, while the remaining chemicals did not belong to any cluster. Two clusters demonstrated high concordance between HTTr and HTPP with 67-80% of chemicals overlapping, and these clusters were significantly associated with retinoid and corticosteroid structure-based annotations, respectively (p ≤ 0.05 via Fisher’s exact tests). Chemicals in the corticosteroid-associated cluster demonstrated significant association with positive bioactivity hitcalls in multiple glucocorticoid and androgen receptor-related endpoints from ToxCast, suggesting that HTTr and HTPP bioactivity profiles for corticosteroids are concordant with effects from targeted assays.An additional 5 clusters were detected based on HTTr or HTPP profiles alone that were not found in the opposing data stream. Several of these clusters were significantly associated with structure-based annotations such as histamine H1 receptor antagonist (HTTr only), triazole fungicide (HTPP only), and chlorinated cyclodiene insecticide classes (HTPP only), further indicating that unique patterns detected from either platform may be associated with chemical structure classes......
URLs/Downloads:
DOI: Transcriptomics and cell morphology screens complement each other when grouping chemicals with similar bioactivity
SOT_2024_POSTER_CONFSUBMISSION_20240301_FINAL.PDF (PDF, NA pp, 1626.686 KB, about PDF)