You are here:
Use of Network Inference to Elucidate Common and Chemical-specific Effects on Steoidogenesis
Citation:
GARCIA-REYERO, N., T. HABIB, L. ESCALON, DAN VILLENEUVE, G. T. ANKLEY, AND E. J. PERKINS. Use of Network Inference to Elucidate Common and Chemical-specific Effects on Steoidogenesis. Presented at International Conference on Systems Biology, Heilberg/Mannheim, GERMANY, August 28 - September 01, 2011.
Impact/Purpose:
Microarray data is a key source for modeling gene regulatory interactions. Regulatory network models based on multiple datasets are potentially more robust and can provide greater confidence. In this study, we used network modeling on microarray data generated by exposing the fathead minnow (Pimephales promelas) to several chemicals that affect steroidogenesis at different steps. We then analyzed gene expression changes on the fish ovary. To analyze the data, we used two different approaches. First, a “combined” network was reconstructed using all the microarray data, a total of 1,472 samples from 23 different exposures combined together. In the second approach, “individual” networks were reconstructed from a subset of seven exposures. In this subset, fathead minnows were exposed to the chemicals fadrozole, flutamide, prochloraz, trenbolone, vinclozolin, trilostan, or ketoconazole for 8 days, then the chemicals were removed and fish were kept in clean water for eight more days. These data provided an overview of expression responses in the fathead minnow ovary when exposed to chemicals, particularly those that affect the hypothalamic-pituitary-gonadal (HPG) axis. Microarrays were performed using a single-color protocol on an 8x15k Agilent platform. A mutual information (MI) based algorithm was used to infer the network and a z-score cutoff threshold (>8.0) was used to filter the high-scoring MI-derived interactions between the gene pairs. Based on the idea that an edge is more likely to represent a true interaction if it appears in multiple input networks, we identified “high-confidence” sub-networks using interactions that appeared across all seven “individual” networks as well as in the “combined” network. Key HPG genes from individual networks were found to interact with “high-confidence” subnetworks. Genes most highly connected in the subnetworks were EGF-receptor, forkhead box L2 (FOXL2), NEDD4 interacting proteins, extracellular matrix 1 (ECM1), estrogen receptor alpha (ESR1), and androgen receptor (AR). The high confidence interactions obtained from both approaches were found significantly enriched in pathways such as focal adhesion, ECM-receptor signaling and estrogen receptor signaling. Common interactions between the combined and individual networks could offer insights about the common mechanism of the effects of chemicals on the HPG-axis, whereas interactions specific for “individual” networks could provide information on the specific chemical mode of action.
Description:
Microarray data is a key source for modeling gene regulatory interactions. Regulatory network models based on multiple datasets are potentially more robust and can provide greater confidence. In this study, we used network modeling on microarray data generated by exposing the fathead minnow (Pimephales promelas) to several chemicals that affect steroidogenesis at different steps. We then analyzed gene expression changes on the fish ovary. To analyze the data, we used two different approaches. First, a “combined” network was reconstructed using all the microarray data, a total of 1,472 samples from 23 different exposures combined together. In the second approach, “individual” networks were reconstructed from a subset of seven exposures. In this subset, fathead minnows were exposed to the chemicals fadrozole, flutamide, prochloraz, trenbolone, vinclozolin, trilostan, or ketoconazole for 8 days, then the chemicals were removed and fish were kept in clean water for eight more days. These data provided an overview of expression responses in the fathead minnow ovary when exposed to chemicals, particularly those that affect the hypothalamic-pituitary-gonadal (HPG) axis. Microarrays were performed using a single-color protocol on an 8x15k Agilent platform. A mutual information (MI) based algorithm was used to infer the network and a z-score cutoff threshold (>8.0) was used to filter the high-scoring MI-derived interactions between the gene pairs. Based on the idea that an edge is more likely to represent a true interaction if it appears in multiple input networks, we identified “high-confidence” sub-networks using interactions that appeared across all seven “individual” networks as well as in the “combined” network. Key HPG genes from individual networks were found to interact with “high-confidence” subnetworks. Genes most highly connected in the subnetworks were EGF-receptor, forkhead box L2 (FOXL2), NEDD4 interacting proteins, extracellular matrix 1 (ECM1), estrogen receptor alpha (ESR1), and androgen receptor (AR). The high confidence interactions obtained from both approaches were found significantly enriched in pathways such as focal adhesion, ECM-receptor signaling and estrogen receptor signaling. Common interactions between the combined and individual networks could offer insights about the common mechanism of the effects of chemicals on the HPG-axis, whereas interactions specific for “individual” networks could provide information on the specific chemical mode of action.