Fig 1.
(A) Four-stage revision process to derive a hybrid context-specific model with an automated validation module as a core, and context-specific data and signaling network interaction graph as inputs. (B) Schematic of context-specific global sensitivity analysis (GSA). Two-step Morris-Sobol global sensitivity analysis identifies key reactions of model validity in each context and reduces the model dimension for calibration of main parameters with semi-quantitative data.
Fig 2.
Estimating default parameters enhances model prediction accuracy in an expanded data compendium covering four validation categories.
(A) Classification of myocyte-specific data in four classes based on the function of measured and perturbed signaling nodes in the network. (B) Comparison of the original [4] and expanded datasets in terms of frequency of each data class. (C) Data distribution in the original and expanded datasets regarding the context (hypertrophic agonist). (D) Comparison of the model prediction accuracy in all data classes before and after estimation of default parameters. (E, F) Variations of model validation percent in terms of four default parameters of the model (n, WR, Wi, We). In each figure, the non-variant parameters were fixed to optimum values. The white star shows the optimum parameter set.
Fig 3.
Classified qualitative validation reveals incomplete parts of the ISO-specific network.
Comparison of model predictions with 75 experimental data points in three classes of data including input-output (7/7 correct prediction), input-intermediate (27/30 correct prediction), and intermediate-inhibition (22/38 correct prediction) after isoproterenol stimulation. The red, blue, and gray boxes illustrate increase, decrease, and no change, respectively. In the model, changes in the measured node activity greater than +1% or smaller than -1% have been considered as an increase or decrease, respectively. In experimental data, statistically significant changes in comparison with the control have been considered as an increase or decrease based on the direction of changes.
Fig 4.
Single-reaction deletion identifies and categorizes context-specific and shared key reactions.
(A) Changes in model validation percent after one-by-one removal of each 174 intermediate reactions in the hypertrophy network for ISO (cross), PE (circle), AngII (triangle), and stretch (square) contexts. (B) The key reactions in ISO-specific and (C) AngII-specific contexts have been visualized on the hypertrophy network map. The thicker and bolder red, or blue arrows illustrate more decrease, or increase in validation percent, respectively, after removing each reaction.
Fig 5.
Two-step Global sensitivity analysis identifies key single and pair reactions in an ISO-specific context.
(A) Morris sensitivity analysis result including Morris index (μ*) and standard deviation (σ) illustrates important and non-important reactions in the ISO-specific context. Greater Morris index (μ*) demonstrates more influence on model validation percent. Larger σ to μ* ratio (above μ* = σ diagonal line) demonstrates a more nonlinear effect on model prediction accuracy which implies on more interactions with other network reactions. The orange dots and blue stars illustrate monotonic and non-monotonic effects of each reaction on model validation percent, respectively. The vertical red line indicates the significance level for identifying non-important reactions. (B) Visualization of important reactions in ISO-specific context based on their ranks from Morris sensitivity analysis. The thicker red arrows illustrate the larger Morris index (μ*). (C) Sobol sensitivity analysis reveals importance of nonlinear interactions between network signaling reactions in ISO-specific context. The Total (checked red bars) and Main (hashed blue bars) Sobol indexes and pair synergic effects are illustrated for important reactions (35/174) in ISO-specific context obtained from Morris sensitivity analysis. The error bars represent 95% confidence interval of a mean. More difference between the Total and Main indexes demonstrates a more nonlinear effect. The pair synergic effect displays the ratio of the second-order effect of each reaction pair to the sum of higher-order effects of its reactions.
Fig 6.
Exploring possible interactions nominates CaMKII as the key signaling node for missing crosstalks in ISO-specific network.
(A) Model predictions for crosstalks between CaMKII and network reactions through the “AND” gate. Each arrow represents a reaction that by adding an “AND” gate from CaMK to that reaction, we observed a significant improvement in model accuracy. (B) Model predictions for crosstalks between CaMKII and network nodes through the “OR” gate. The thicker red arrows illustrate reactions with higher positive effect on the validation percent in ISO-specific model.
Fig 7.
Inhibition of CaMKII-Gβγ or CaN-Gβγ suppresses ISO-induced ERK1/2 activity.
(A) Effects of pretreatment for 30 min with five inhibitors including JAK-i (AG490), CaMKII-i (KN93), CaN-i (CsA), Gβγ-i (Gallein), and MEK-i (U0126) were examined on basal and ISO-induced ERK1/2 activity. (B) Upper panel shows western blotting of ERK1/2 phosphorylation after pretreatment with inhibitors in comparison with control (DMSO). Lower panel shows ERK1/2 phosphorylation after ISO stimulation with pretreatment with same inhibitors. *represents statistically significant change (p<0.05) in comparison with control (DMSO) for the upper panel, or with ISO in the lower panel. # represents significant change (p<0.05) compared with control (DMSO) in the lower panel. Data were collected from 3 independent experiments (Mean±SEM). (C) Immunofluorescence images show ISO-induced ERK1/2 phosphorylation and its suppression by MEK inhibitor in myocytes (alpha-actinin positive) and fibroblasts (alpha-actinin negative). The scale bar is 50 μm (D) Immunofluorescence imaging of ERK1/2 phosphorylation shows increase of ERK1/2 phosphorylation after CaMKII inhibition with KN93 and its suppression after pretreatment with Gβγ+CaMKII and Gβγ+CaN combined inhibitions. The scale bar is 200 μm.
Fig 8.
Revised model of β-adrenergic cardiac hypertrophy.
The solid red arrows exhibit predicted and validated interactions in ISO-specific context. The dashed red arrows illustrate proposed interactions predicted by model based on the new experimental data on the effects of CaMKII inhibition on ISO-induced ERK1/2 activity.