Responsive genes for fish embryos have been identified so far for some endocrine pathways but not for androgens. Using transcriptome analysis and multiple concentration-response modeling, we identified putative androgen-responsive genes in zebrafish embryos exposed to 0.05-5000 nM 11-ketotestosterone for 24 h. Four selected genes with sigmoidal concentration-dependent expression profiles (EC50 = 6.5-30.0 nM) were characterized in detail. The expression of cyp2k22 and slco1f4 was demonstrated in the pronephros; lipca was detected in the liver, and sult2st3 was found in the olfactory organs and choroid plexus. Their expression domains, the function of human orthologs, and a pathway analysis suggested a role of these genes in the metabolism of hormones. Hence, it was hypothesized that they were induced to compensate for elevated hormone levels. The induction of sult2st3 and cyp2k22 by 11-ketotestosterone was repressed by co-exposure to the androgen receptor antagonist nilutamide supporting a potential androgen receptor mediated regulation. Sensitivity (expressed as EC50 values) of sult2st3 and cyp2k22 gene expression induction after exposure to other steroidal hormones (11-ketotestosterone ∼ testosterone > progesterone > cortisol > ethinylestradiol) correlated with their known binding affinities to zebrafish androgen receptor. Hence, these genes might represent potential markers for screening of androgenic compounds in the zebrafish embryo.

Omics-based methods are increasingly used in current ecotoxicology. Therefore, a large number of observations for various toxic substances and organisms are available and may be used for identifying modes of action, adverse outcome pathways, or novel biomarkers. For these purposes, good statistical analysis of toxicogenomic data is vital. In contrast to established ecotoxicological techniques, concentration-response modeling is rarely used for large datasets. Instead, statistical hypothesis testing is prevalent, which provides only a limited scope for inference. The present study therefore applied automated concentration-response modeling for 3 different ecotoxicotranscriptomic and ecotoxicometabolomic datasets. The modeling process was performed by simultaneously applying 9 different regression models, representing distinct mechanistic, toxicological, and statistical ideas that result in different curve shapes. The best-fitting models were selected by using Akaike's information criterion. The linear and exponential models represented the best data description for more than 50% of responses. Models generating U-shaped curves were frequently selected for transcriptomic signals (30%), and sigmoid models were identified as best fit for many metabolomic signals (21%). Thus, selecting the models from an array of different types seems appropriate, because concentration-response functions may vary because of the observed response type, and they also depend on the compound, the organism, and the investigated concentration and exposure duration range. The application of concentration-response models can help to further tap the potential of omics data and is a necessary step for quantitative mixture effect assessment at the molecular response level.