(393e) Modeling Signaling Interactions Controlling Heterogeneity and Fate Choice of Human Embryonic Stem Cells

Motivation: Human embryonic stem cells (hESCs) are an attractive raw material for regenerative medicine due to their potential to deliver a variety of clinically important mature lineages. Methods to induce and monitor self-renewal and differentiation of hESCs are now well established. However generation of pure and well-defined lineages is still a major challenge due to the heterogeneity of stem cell cultures. Recent research from diverse stem cell systems has shown that heterogeneity results from a complex interaction between variations in protein expression levels, signaling reaction rate/mechanism and the cell cycle stage [1,2]. To control heterogeneity, it is necessary to understand how above features influence network level signal propagation. In this project, we employed a systems level approach to evaluate network dynamics and variability in differentiating hESCs by integrating computational models with experimental techniques (measuring single cell and population dynamics). The system of interest is the TGF-β/SMAD2,3 pathway and its crosstalk with PI3K/AKT pathway existing during regular endoderm differentiation of hESCs [3].

Methods:

Experimental: For self-renewal, H1 hESCs were maintained on matrigel-coated plates in mTeSR1. Endoderm differentiation was then performed using 100 ng/ml Activin A (to activate TGF-β/SMAD2,3) with or without modulation of PI3K/AKT pathway using PI3K inhibitor, Wortmannin. Phosphorylation dynamics of participating signaling molecules at the cell population level were measured using MagPix® xMAP technology. The initial selection of key molecules was based on the study by Singh et al. [4]. These mainly included SMAD2,3,4,7 molecules, TGF-β Receptors, AKT and ERK. In order to facilitate measurement of protein dynamics at the single cell level, we generated hESC cell lines expressing GFP-fused SMAD proteins. These transformed cells facilitated the measurement of nucleo-cytoplasmic shuttling kinetics of molecules from single cells using Fluorescence Recovery after Photo-bleaching (FRAP) analysis. Mathematical and computational tools described in the next section were used to guide generation of data by in-house experiments.

Mathematical Analysis: Dynamic Bayesian Network Analysis (DBN) was used to identify key sensitive molecules, hypothesize interactions and identify most informative time points. Detailed mechanistic Ordinary Differential Equation (ODE) model for the TGF-β/SMAD2,3 pathway with crosstalk interactions with PI3K/AKT was developed for a systems level analysis. The model was calibrated using Replica Exchange Ensemble Modeling (EM) and sensitive reactions were identified using computationally efficient Global Sensitivity Approach (GSA) called Random Sampling High Dimensional Model Representation. Compartmental ODE models in combination with quantitative image analysis were used to simulate photo-bleaching experiments and estimate the nucleo-cytoplasmic shuttling rates of SMAD molecules.

Results and Discussion:

We have developed a detailed ODE model representing signaling events in the PI3K/AKT pathway under self-renewal status of hESCs. Here we applied GSA to identify sensitive nodes from the pathway that modulated a critical self-renewal molecule, p-AKT. The computational cost of GSA was significantly reduced by application of a meta-model framework to capture behavior of the ODE model. Using this approach, we recently showed that a negative feedback loop in the pathway controlled the levels of p-AKT [2]. This feedback loop constrained the cell-to-cell variability of p-AKT within the population as well as between consecutive in vitro passages.

In the current project, we are expanding this approach to model the dominant signaling events regulating initial endoderm differentiation of hESCs. First, to identify possible crosstalk interactions between the signaling molecules, DBN analysis was performed on a detailed time series from the average population data. DBN generated for high and low PI3K conditions showed that multiple crosstalk points exist during endoderm differentiation and temporal inhibition of PI3K at the early phase of signaling was sufficient to remove crosstalk. Further, the receptor RII levels influenced the downstream molecules during the entire phase of the signaling dynamics. HESCs further showed divergence in the dynamics of regulatory SMADs, a characteristic not observed in other reported cell systems. The reason for this divergence was evaluated by analyzing parametric ensembles in the high dimensional parameter space. The nucleo-cytoplasmic shuttling kinetics of SMAD molecules with and without crosstalk with AKT at the single cell level was used to inform the ODE model. The resulting mixture of single cell ODE models was integrated and compared to the mean population dynamics for validation as well as constraining free parameters. This integration of single cell and population model enabled us to evaluate the specific reactions and network motifs that contributed most to the variation in nuclear SMAD levels, which are the determinants of endoderm differentiation. We are in the process of relating the variations in nuclear SMADs to downstream transcriptional heterogeneity.

Conclusions:

Our systems analysis results show that during signal transduction in the self-renewal state and during differentiation, network motifs have an important role in controlling the levels and heterogeneity of stem cells. Through this work, we have demonstrated that application of efficient computational tools and workflows that explore single cell and population dynamics will enable extraction of mechanistic information for systems with high heterogeneity. This will offer new avenues to control heterogeneity and improve signal propagation by targeted modulation of current culture conditions.

References:

[1] Singh, AM (2015) Stem Cells International, Article ID 219514. doi:10.1155/2015/219514

[2] Mathew et al. (2014) Bioinformatics, 30(16): 2334-2342

[3] Mathew et al. (2015) MDPI Processes, 3(2): 286-308

[4] Singh et al. (2012) Cell Stem Cell, 10(3), 312-326