(332c) Inputs Drive Variability of Regulatory Network Phenotypes

Current high-throughput technologies have allowed us to examine single cell behavior extensively at the molecular level. The ability to develop such comprehensive data sets provides the opportunity to explore and identify system-wide gene expression behavior associated with individual cellular states. Accordingly, the ability to generate causal input-output gene regulatory relationships will provide significant insight into how these molecular mechanisms drive cellular state and function. We aim to develop a network inference technique that models causal input-output gene relationships from single cell transcriptomic data sets. Originally, analyzing such data sets was complicated by the extensive transcriptional variability observed across cells within a particular phenotype. However, correlated gene expression observed in heterogeneous transcriptional behavior across single cells provides clues to underlying regulatory network structures. Moreover, single cell transcriptomic data sets provide a rich experimental sampling of transcriptional profiles that can be used to infer regulatory networks that drive these correlated relationships. 

We obtained a high-throughput in vivo transcriptomic measures taken from single brainstem neurons of rats undergoing a physiologically induced hypertensive challenge. Our analysis showed that correlated gene expression patterns underlie a distribution of neuronal subtypes¹. In the present study we examined this single-neuron transcriptional data set to infer regulatory networks corresponding to these neuronal subtypes. For this purpose, we applied a constrained fuzzy logic (cFL) approach² to systematically compare gene interaction relationships within a proposed regulatory network against our neuronal in vivo transcriptional data set. Because the measured transcriptional profiles were in response to an overall physiological perturbation, as opposed to direct neuronal stimulation or perturbation, we have modified the original cFL approach to consider graded gene expression values, as observed in vivo, that stimulate or perturb the gene regulatory network. Subsequently we trained a proposed a priori network composed of causal gene interactions (obtained from literature) involved in the angiotensin II type 1 receptor-mediated pathway, a target of several hypertensive treatments. Using this approach we investigated the regulatory networks contributing to the observed distribution of neuronal phenotypes.

Our analysis revealed distinct regulatory network model structures corresponding to neuronal subtypes, distinguished by transcription factor regulation and feedback inhibition of neuronal functional genes. Using these distinct network models, we examined how a distribution of neuronal subtypes may arise by simulating individual network models over a range of gene expression level inputs and evaluating their in silico transcriptional outputs using multidimensional scaling. In silico network simulations produced a range of neuronal subtypes similar to what was experimentally observed and showed that distinct gene expression level inputs to distinct networks yield similar transcriptional phenotypes despite their differing regulatory relationships. This stands in contrast to the previous expectation of distinct networks producing distinct transcriptional phenotypes given identical inputs. These results suggest that regulatory network structure and network response to inputs contribute to a range of neuronal subtypes. Consequently, we can now use our models to explore how changes to regulatory network structures may alter the balance of neuronal subtypes within a phenotypic population and contribute to the dysregulation of blood pressure as observed in hypertension. 

1. Park J, Brureau A, Kernan K, Starks A, Gulati S, Ogunnaike B, Schwaber J, Vadigepalli R. 2014. Inputs drive cell phenotype variability. Genome Res.
2. M. K. Morris, J. Saez-Rodriguez, D. C. Clarke, P. K. Sorger, and D. a Lauffenburger, “Training signaling pathway maps to biochemical data with constrained fuzzy logic: quantitative analysis of liver cell responses to inflammatory stimuli.,” PLoS Comput. Biol., vol. 7, no. 3, p. e1001099, Mar. 2011.

Research Support:  NIH (NIGMS R01 GM083108; NHLBI R01HL111621; NIAAA 5 T32 AA007463-26)