Predictive Analysis Using Chemical-Gene Interaction Networks Consistent with Observed Endocrine Activity and Mutagenicity of U.S. Streams

Abstract

In a recent U.S. Geological Survey/U.S. Environmental Protection Agency study assessing more than 700 organic compounds in 38 streams, in vitro assays indicated generally low estrogen, androgen, and glucocorticoid receptor activities, with 13 surface waters with 17β-estradiol-equivalent (E2Eq) activities greater than a 1-ng/L estimated effects-based trigger value for estrogenic effects in male fish. Among the 36 samples assayed for mutagenicity in the Salmonella bioassay (reported here), 25% had low mutagenic activity and 75% were not mutagenic. Endocrine and mutagenic activities of the water samples were well correlated with each other and with the total number and cumulative concentrations of detected chemical contaminants. To test the predictive utility of knowledge-base-leveraging approaches, site-specific predicted chemical-gene (pCGA) and predicted analogous pathway-linked (pPLA) association networks identified in the Comparative Toxicogenomics Database were compared with observed endocrine/mutagenic bioactivities. We evaluated pCGA/pPLA patterns among sites by cluster analysis and principal component analysis and grouped the pPLA into broad mode-of-action classes. Measured E2eq and mutagenic activities correlated well with predicted pathways. The pPLA analysis also revealed correlations with signaling, metabolic, and regulatory groups, suggesting that other effects pathways may be associated with chemical contaminants in these waters and indicating the need for broader bioassay coverage to assess potential adverse impacts.

Figure 1.

Site-specific pCGA counts for the USGS chemical mixtures study. Sites were clustered according to similarity in count among individual genes (black – high; red – mid-high; yellow – mid; green – mid-low; blue – low). Diamonds indicate total number of detected chemicals at each site.

Figure 2.

The interaction counts (solid bars, left axis) and concentrations (hatched bars, right axis) of selected individual genes at example sites within the five clusters identified in the total interaction count analysis. The listed genes were key to cluster separation and represent common neurotransmitters, hormone regulators, and metabolic signaling genes.

Figure 3.

Comparison of cumulative net (agonist – antagonist) concentrations of CTD-predicted (solid bars) gene-active chemicals or cumulative concentrations of known (hatched bars; ER/ESR1 and AR only) gene-active chemicals with responses (▲) for the estrogen receptor (ER/ESR1 gene), androgen receptor (AR), and glucocorticoid receptor (GR/NR3C1 gene) in vitro bioassays reported in Conley et al. Open symbols indicate bioassay activity less than 1 ng/L E2Eq (ER plots only) or not significantly different from control (AR, GR plots).