Fathead minnow steroidogenesis: in silico analyses reveals tradeoffs between nominal target efficacy and robustness to cross-talk

Background: Interpreting proteomic and genomic data is a major challenge in predictive ecotoxicology that can be addressed by a systems biology approach. Mathematical modeling provides an organizational platform to consolidate protein dynamics with possible genomic regulation. Here, a model of ovarian steroidogenesis in the fathead minnow, Pimephales promelas, (FHM) is developed to evaluate possible transcriptional regulation of steroid production observed in microarray studies. Results: The model was developed from literature sources, integrating key signaling components (G-protein and PKA activation) with their ensuing effect on steroid production. The model properly predicted trajectory behavior of estradiol and testosterone when fish were exposed to fadrozole, a specific aromatase inhibitor, but failed to predict the steroid hormone behavior occurring one week post-exposure as well as the increase in steroid levels when the stressor was removed. In vivo microarray data implicated three modes of regulation which may account for over-production of steroids during a depuration phase (when the stressor is removed): P450 enzyme up-regulation, inhibin down-regulation, and luteinizing hormone receptor up-regulation. Simulation studies and sensitivity analysis were used to evaluate each case as possible source of compensation to endocrine stress. Conclusions: Simulation studies of the testosterone and estradiol response to regulation observed in microarray data supported the hypothesis that the FHM steroidogenesis network compensated for endocrine stress by modulating the sensitivity of the ovarian network to global cues coming from the hypothalamus and pituitary. Model predictions of luteinizing hormone receptor regulation were consistent with depuration and in vitro data. These results challenge the traditional approach to network elucidation in systems biology. Generally, the most sensitive interactions in a network are targeted for further elucidation but microarray evidence shows that homeostatic regulation of the steroidogenic network is likely maintained by a mildly sensitive interaction. We hypothesize that effective network elucidation must consider both the sensitivity of the target as well as the target's robustness to biological noise (in this case, to cross-talk) when identifying possible points of regulation.