(497g) Decoding Robustness in Integrated Signaling and Gene Regulatory Networks Using a Combination of Global Sensitivity Analysis and Decision Trees | AIChE

(497g) Decoding Robustness in Integrated Signaling and Gene Regulatory Networks Using a Combination of Global Sensitivity Analysis and Decision Trees


Makadia, H. - Presenter, Thomas Jefferson University
Schwaber, J. S., Thomas Jefferson University
Vadigepalli, R., Thomas Jefferson University

The primary objective of our study is to identify the mechanisms through with the robustness is encoded in a complex network of signaling pathways and gene regulatory networks connected through feed-forward and feedback interactions. The complexity of the regulatory networks containing combinatorial/nonlinear interactions, cross-talk, and feedback circuitry requires investigation beyond topological aspects by explicitly analyzing the network dynamics. We present a novel, systematic, and mathematical approach combining global sensitivity analysis and decision trees, which elucidates such nonlinear relationships and identifies the robust as well as sensitive control points in the network. We focused on the robustness of activation for immediate early transcription factors in response to variations in the dynamic profiles of upstream signaling pathways. The variations in signaling kinase activities represents differential function in individual cells and cell phenotypes. We demonstrate our approach on a biological system of key relevance to neuronal adaptive processes dysregulated in the development of hypertension. Our results indicate several aspects of the network robustness that are not intuitable from static topological considerations alone.

We considered the angiotensin II type 1 receptor (AT1R) signaling and gene regulatory network in the brainstem [1]. AT1R network mediates a wide range of autonomic functions in brainstem, one of which is the control of hypertension by regulation of Tyrosine hydroxylase (TH) production. The gene regulatory network activated by AT1R was previously studied for robustness of transcription factor (TF) activation dynamics to variations in the kinetic parameters[1]. This mechanistic model was implemented as a set of ordinary differential equations, and bridges three signaling kinases viz. FRK, ERK and JNK with activation of AP-1 TF family. In the present study, we seek to identify the robustness of AP-1 and TH activation to variations in the dynamics of kinases FRK, ERK and JNK, to understand the decoding of the combinatorial signaling activity at the level of transcription factor dynamics and the consequences on target gene expression.

The kinase temporal profiles were considered as a set of nonlinear exponential functions with six parameters describing their delay and kinetics of activation, peak retention and deactivation. Potential kinase profiles were randomly sampled in two-fold range to explore network dynamics in biologically meaningful range of signaling activities. We followed a global sensitivity analysis approach using Sobol sensitivity indices for a total of eighteen kinase parameters as they affect AP-1 and TH dynamics[2]. We categorized the results based on the presence/absence of delay in activation of three kinases and compared the sensitivities across these groups. We further analyzed the dependencies between the key kinase profile parameters that were significantly affecting AP-1 or TH dynamics using a decision tree approach. This method organizes the key kinase parameter ranges in a hierarchy with different paths in the tree leading to distinct AP-1 or TH dynamical behavior.

We identified key kinase profile parameter that control the AP-1 family of TFs (c-fos:c-jun hetrodimer and c-jun:c-jun homodimer). Although topology of the mechanistic model suggest an influence of FRK during initial phase, global sensitivity analysis shows a surprising result that delay in activation of JNK and ERK as the most influential parameters in affecting downstream AP-1 activity and target TH gene expression dynamics. The peak retention parameter of the JNK shows a biphasic influence with reduced sensitivity (i.e., robustness) at the transition phase in which the composition of AP-1 is switching form a heterodimer to a homodimer. A higher order interaction between JNK peak retention parameter, FRK deactivation parameter and its time constant controlled the transition phase as demonstrated through their second order sensitivities. The transition period, when the the c-jun:c-jun dimer activity dominates the c-fos:c-jun dimer activity, was a robust phase during which the level of AP-1 or TH was least sensitive to all of the kinase parameters. We interpret this as a network-level constraint that differentiates the state of the network that is dominated by a hetero- vs homo-dimer of AP-1. During the last phase in which the homo-dimer dominated the regulation, the activity of AP-1 and TH gene expression was significantly sensitive to all the parameters of JNK except the delay in activation and corresponding time constant.

The decision tree analysis organized these parameters into a hierarchy, with subsets of kinase profiles leading to distinct activation profiles of AP-1 and TH. These indicate that individual cells or cell phenotypes representing such subsets of kinase activities could lead to distinct downstream network response and contribute towards a heterogeneous tissue scale behavior. The robustness of AP-1 gene regulatory network to a majority of the kinase activity parameters, and differential sensitivity to a select subset, reveal the control points in this system that when dysregulated are likely to lead to an aberrant network response underlying maladaptive autonomic control such as in essential hypertension.


[1] Gregory M Miller, Babatunde A Ogunnaike, James S Schwaber, and Rajanikanth Vadigepalli. Robust dynamic balance of AP-1 transcription factors in a neuronal gene regulatory network. BMC Systems Biology, 4(1):171, 2010.

[2] Andrea Saltelli. Global Sensitivity Analysis: The Primer. John Wiley, March 2008.