Analyzing high-throughput assay data to advance the rapid screening of environmental chemicals for human reproductive toxicity

Abstract

While high-throughput (HTP) assays have been proposed as platforms to rapidly assess reproductive toxicity, there is currently a lack of established assays that specifically address germline development/function and fertility. We assessed the applicability domains of yeast (S. cerevisiae) and nematode (C. elegans) HTP assays in toxicity screening of 124 environmental chemicals, determining their agreement in identifying toxicants and their concordance with reproductive toxicity in vivo. We integrated data generated in the two models and compared results using a streamlined, semi-automated benchmark dose (BMD) modeling approach. We then extracted and modeled relevant mammalian in vivo data available for the matching chemicals included in the Toxicological Reference Database (ToxRefDB). We ranked potencies of common compounds using the BMD and evaluated correlation between the datasets using Pearson and Spearman correlation coefficients. We found moderate to good correlation across the three data sets, with r = 0.48 (95% CI: 0.28-1.00, p<0.001) and r_s = 0.40 (p=0.002) for the parametric and rank order correlations between the HTP BMDs; r = 0.95 (95% CI: 0.76-1.00, p=0.0005) and r_s = 0.89 (p=0.006) between the yeast assay and ToxRefDB BMDs; and r = 0.81 (95% CI: 0.28-1.00, p=0.014) and r_s = 0.75 (p=0.033) between the worm assay and ToxRefDB BMDs. Our findings underscore the potential of these HTP assays to identify environmental chemicals that exhibit reproductive toxicity. Integrating these HTP datasets into mammalian in vivo prediction models using machine learning methods could further enhance the predictive value of these assays in future rapid screening efforts.

Keywords: New Approach Methodologies (NAMs); aneuploidy; benchmark dose modeling; benchmark response; chemical assessment; developmental toxicology; germ cell development; meiosis.

Figure 1.

Project workflow for predictive reproductive modeling based on two HTP assays. We modeled dose-response data generated in our two models (S. cerevisiae and C. elegans) and available matched in vivo datasets from ToxRefDB [28]. Benchmark doses (BMDs) were derived to assess the concordance of toxic potencies between the two HTP assays and their correlation with in vivo reproductive toxicity endpoints.

Figure 2.

Overview of BMDS modeling results across three data sets (two HTP assays and ToxRefDB).

Figure 3.

S. cerevisiae data BMD_1SD modeling results (n=82 chemicals with increasing dose-response curves) color-coded by chemical class.

Figure 4.

C. elegans data BMD_1SD modeling results (n=71 chemicals with increasing dose-response curves) color-coded by chemical class.

Figure 5.

Correlation between S. cerevisiae (x-axis) and C. elegans (y-axis) BMD_1SD data color-coded by chemical class (n=51).

Figure 6.

Correlation between the BMD_1SD from ToxRef (y-axis) and (A) yeast data (x-axis) or ToxRef (y-axis) and (B) C. elegans data (x-axis). There were seven overlapping chemicals in A (BPA, DEET, TCEP, Triclosan, Hydroxyurea, DEHP, and Malathion) and seven overlapping chemicals in B (BPA, DEET, TCEP, Triclosan, Hydroxyurea, Pyraclostrobin, and Tricresyl phosphate [TCP]), with five overlapping chemicals across all three data sets.