Sensitivity analysis of intracellular signaling pathway kinetics predicts targets for stem cell fate control

Abstract

Directing stem cell fate requires knowledge of how signaling networks integrate temporally and spatially segregated stimuli. We developed and validated a computational model of signal transducer and activator of transcription-3 (Stat3) pathway kinetics, a signaling network involved in embryonic stem cell (ESC) self-renewal. Our analysis identified novel pathway responses; for example, overexpression of the receptor glycoprotein-130 results in reduced pathway activation and increased ESC differentiation. We used a systematic in silico screen to identify novel targets and protein interactions involved in Stat3 activation. Our analysis demonstrates that signaling activation and desensitization (the inability to respond to ligand restimulation) is regulated by balancing the activation state of a distributed set of parameters including nuclear export of Stat3, nuclear phosphatase activity, inhibition by suppressor of cytokine signaling, and receptor trafficking. This knowledge was used to devise a temporally modulated ligand delivery strategy that maximizes signaling activation and leads to enhanced ESC self-renewal.

Figure 1. Network Structure of the Jak/Stat3 Pathway Shows Scheme for LIF-induced Activation of Stat3

Closed arrows indicate biochemical reactions, open arrows show transport such as transport into nuclear compartment, and double arrows indicate reversibility. All surface receptors and complexes are internalized similarly, internalized complexes are degraded, and new receptors are constitutively produced. Color-coded dots next to kinetic parameters correspond to grouping of parameters from Figure 5 as explained in text.

Figure 2. Computational Simulation of Jak/Stat3 Pathway Activation at Different LIF Concentrations Shows Transient Kinetics

LIF was added in increasing concentrations of 10, 50, 500, and 1,000 pM in the direction of arrow. LIFRC is the complex of LIFR with LIF, LIFRGP130CP is the phosphorylated heterodimeric complex of LIFR and GP130, STAT3DPN is nuclear Tyr705 phosphorylated Stat3, STAT3DN is nuclear Stat3, SOCS3MRNAN and SOCS3MRNA are the nuclear and cytoplasmic levels of SOCS3 mRNA, and BOUND-SOCS3 is the total number of receptor complexes which are bound to SOCS3. This representative simulation was run for 200 min, and the number of activated proteins per cell are plotted as a function of time.

Figure 3. Experimental Validation of Model Output

(A) Representative images of mouse ECS for quantitative single cell fluorescence microscopy. Hoechst nuclear stain is used to determine location of nuclei, a nuclear mask is drawn, and average fluorescence levels in the mask are used to determined levels of phosphorylated Stat3 and Oct4. (B–D) Show agreement between model-predicted trends and experimental results of activation profiles of Jak phosphorylation, nuclear retention of Stat3, and SOCS3 induction, respectively. For Jak phosphorylation, nuclear accumulation of Stat3, and SOCS3 induction, both experimental and simulation results were normalized based on the maximum of the activation profiles for each of these proteins. For nuclear accumulation of phosphorylated Stat3 the results were normalized based on a steady state level in 500 pM LIF. (E) The model-predicted activation profile of nuclear retention of Tyr-705 phosphorylated Stat3 is shown in solid line and is in agreement with experimental data points.

Figure 4. GP130 Overexpressing Cell Lines Show Non-Monotonic Response of Stat3 Activation to Receptor Overexpression

(A) The model-predicted surface of Stat3 activation is shown as a function of time and change in production of GP130 receptors. Successive increase in GP130 production as shown by black, green, and red arrows in that order shows a non-monotonic response of Stat3 activation and the presence of a local maximum. This is illustrated by the circle in the contour map on the x–y plane. (B) GP130 overexpressing cell lines RG1–RG5 were developed to express the receptor at varying levels compared with R1 cells, and these cell lines retain Oct4 expression in the presence of LIF (C). (D) Model-predicted trends of GP130 overexpression corresponding to (A) shows normal profile of Stat3 activation (black line), Stat3 activation in slight GP130 overexpression (green line), and significant GP130 overexpression (red line). (E) Experimental results show a consistent trend in LIF-induced Stat3 activation profile using RG1 and RG5 cell lines in comparison with model results in (D). (F) IL-6 stimulation of GP130 overexpressing cell lines shows a dose-dependent increase in Stat3 activation as a function of GP130 overexpression.

Figure 5. Global Sensitivity Analysis Based on Stat3 Activation

(A) Individual rate constants were changed 5-fold, and a Euclidean distance measurement was used to determine the resultant change in the Stat3 activation profile. Normalized Euclidean distance measurements are plotted to show the sensitivity of Stat3 activation to each parameter. (B) Global sensitivity analysis results show changes in Stat3 activation (measured by Euclidean distance) due to 5-fold changes in combinations of parameters pairs, shown along the x- and y-axis. The sensitivity is represented using color code, with corresponding Euclidean distance values shown in the color bar. Hierarchical clustering (indicated on the left) is based on correlations between rows of rate constants to demonstrate groups of parameters which interact similarly. Clustering was perfomed using Matlab software by Mathworks. Rate constants, which are grouped together, are color-coded using circular dots, and the color code is included in Figure 1. Sensitivity analysis predictions for the Stat3 nuclear export parameter (marked by square), and receptor phosphorylation (marked by triangle), and their interaction (marked by circle) are shown in (C). Sensitivity trends which were predicted in (C) are experimentally verified (D), demonstrating that predicted trends are relevant. (E) Sensitivity analysis trends of interaction of SOCS3 transcription and receptor production (marked by hexagon) and SOCS3 translation and receptor production (marked by cross) are in agreement with experimental results shown in (F).

Figure 6. Desensitization of ESCs to LIF Stimulation Can Be Investigated Computationally and Used to Control Self-Renewal

(A) Model-predicted trends of desensitization of Stat3 activation to LIF stimulation. Color bar inset shows history of 500 pM LIF stimulations. (B) Experimental results are in agreement with model-predicted desensitization kinetics in (A) and show a gradual loss of desensitization in absence of ligand. (C) Global sensitivity analysis shows parameter interactions which result in right-shift (positive minutes) or left-shift (negative minutes) in restimulation profile of Stat3. Shift in the profile was determined from the maximum of cross-correlation of two consecutive Stat3 activation profiles. (D) Sensitivity analysis clustergram shows Euclidean distance measurement between two consecutive Stat3 activation profiles and distinguishes parameter interactions which influence desensitization by changing the restimulation profile. (E) Experimental results of dose response of Stat3 activation to LIF show the steady state response as well as the peak values of Stat3 activation during transient LIF stimulation. (F) Dose response of Oct4 expression after 72 h at different LIF concentrations. (G) Stat3 activation profiles for 10 pM LIF stimulation with 1-h and 6-h periods in red and blue, respectively. During 6-h cycles, LIF is removed for 3 h during which time a lower Stat3 activation level is observed in comparison with the 1-h period. (H) Histograms of Oct4 expression show that 1-hr period of LIF stimulation (red) maintains a higher percentage of Oct4+ cells than 6-h period of LIF stimulation after 72 h. (I) Oct4 expression of cells maintained for 72 h in different conditions shows that the percentage of Oct4+ cells in the 1-h period is higher than the 6-h period and comparable to 500 pM constant LIF levels.